VIBRATION-BASED STRUCTURAL HEALTH MONITORING OF WIND TURBINES Gustavo

VIBRATION-BASED STRUCTURAL
HEALTH MONITORING OF WIND
TURBINES
Gustavo Miguel Cameira da Silva Oliveira
A dissertation presented to the Faculty of Engineering of the University of Porto

for the degree of Doctor of Civil Engineering
Supervisors: Álvaro Cunha (Full professor); Filipe Magalhães (Assistant professor);

Elsa Caetano (Associate professor)
April 2016
Vibration-Based Structural Health Monitoring of Wind Turbines
To my parents
4
9
ABSTRACT
Wind power is one of the most attractive and reliable options for renewable and clean energy. In the
last years, the installed capacity for exploitation of wind energy has consistently increased worldwide.
Alongside, the technological evolutions of this sector have been based on the development of multi-
MW systems with increasingly larger structures elements, such as support structures and rotor blades.
These wind turbines present considerable advantages considering the power production; however,
they also represent several structural challenges: these wind turbines have become increasingly flexible
structures, prone to resonance phenomena and to rapid wear.
In that sense, the structural integrity and proper operation of large wind turbines is a concern for
owners and operators, specially in turbines installed offshore. Monitoring systems capable of
evaluating the condition of structural elements such as foundation, tower and blades, are seen as a
suitable response to address this problem. Among the several techniques available, vibration-based
systems reveal a great potential for structural monitoring.
Therefore, this thesis is focused on the development and implementation of a vibration-based dynamic
monitoring system for wind turbines. This system is based on Operational Modal Analysis tools,
capable of accurately estimate the modal properties of wind turbine structures from their dynamic
response. The efficiency of these tools in the modal identification of wind turbines was validated with
the analysis of vibrational data collected during a short period from onshore (Izar Bonus 1.3MW/62)
and offshore (Vestas V90-3.0MW) models.
Afterwards, a vibration-based monitoring system was implemented on a 2.0 MW Senvion MM82 wind
turbine allowing the collection of data during more than one year. Throughout this program, the
system was defined with two main purposes: detection of structural anomalies (based on the
continuous estimation and statistical analysis of the variations of the modal properties) and fatigue
assessment (based on a reduced number of installed sensors). The program was complemented with
the analysis of alternative solutions for optimization of the number of installed sensors. The usefulness
of the system was tested with different damage scenarios. It was concluded that the monitoring system
is capable of detecting small damages in the support structure of both onshore and offshore wind
turbines. The tools developed for fatigue evaluation using the estimated modal responses showed
promising results but a complete validation of their accuracy need further research.
9
RESUMO
Energia eólica é uma das mais atrativas e fiáveis opções para energia renovável e limpa. Nos últimos
anos, a capacidade instalada para aproveitamento do recurso eólico tem crescido de forma consistente
em todo o mundo. Ao mesmo tempo, a evolução tecnológica do sector tem sido baseada no
desenvolvimento de sistemas com múltiplos MW, recorrendo a elementos estruturais de dimensões
cada vez maiores, tais como estruturas de suporte e pás. Estas torres eólicas apresentam vantagens
consideráveis do ponto de vista da produção; no entanto, representam também diversos desafios
estruturais: estes aerogeradores tornaram-se estruturas cada vez mais flexíveis, propensos a fenómenos
de ressonância e a um desgaste rápido.
Nesse sentido, a integridade estrutural e a correta operação de modernas torres eólicas é uma
preocupação para donos e operadores, especialmente em turbinas instaladas em offshore. Sistemas de
monitorização capazes de avaliar a condição de elementos estruturais, tais como fundação, torre e pás,
são vistos como uma resposta adequada para lidar com este problema. De entre as várias técnicas
disponíveis, sistemas baseados na análise de vibrações revelam um grande potencial para
monitorização estrutural.
Assim, esta tese foca-se no desenvolvimento e implementação de um sistema de monitorização
dinâmico baseado na análise de vibrações para torres eólicas. Este sistema é baseado em ferramentas de
Análise Modal Operacional, capazes de estimar com precisão as propriedades modais das estruturas de
aerogeradores a partir da sua resposta dinâmica. A eficácia destas ferramentas na identificação modal
de torres eólicas foi validada com a análise de dados de vibração colhidos durante um curto período de
tempo para uma torre instalada em terra (Izar Bonus 1.3MW/62) e uma outra em offshore (Vestas
V90-3.0MW).
Posteriormente, um sistema de monitorização foi implementado numa torre eólica Senvion MM82,
permitindo a recolha de dados duranta mais de um ano. Ao longo deste programa, o sistema foi
definido com dois objetivos principais: deteção de anomalias estruturais (baseado na estimativa
contínua e análise estatística das variações das propriedades modais) e avaliação da condição de fadiga
(com base num reduzido número de sensores instalados). O programa foi complementado com a
análise de soluções alternativas para otimização do número de sensores instalados. A utilidade do
sistema foi testada com diferentes cenários de dano. Foi concluído que o sistema de monitorização é
capaz de detetar danos reduzidos na estrutura de suporte de torres instaladas em terra e em offshore.
As ferramentas desenvolvidas para avaliação de fadiga, através de estimativas de respostas modais,
mostraram resultados promissores, no entanto uma validação completa da sua precisão requere
desenvolvimentos adicionais.
19
ACKNOWLEDGEMENTS
This thesis would not be possible without the help and support of several people. Here I present my
sincere gratitude:
o to Prof. Álvaro Cunha, my advisor, for challenging me with such an interesting topic, for all
the support given during this period and for his care to ensure all the conditions and means
for the proper development of this thesis;
o to Prof. Filipe Magalhães, my co-advisor, for all the support, for the endless hours he received
me in his office and for the patience and sympathy to clarify unscheduled problems;
o to Prof. Elsa Caetano, my co-advisor, for her support whenever needed and for the
opportunity to cooperate on interesting projects;
o to Eng. José Carlos Matos, Eng. Miguel Marques and Eng. Silvina Guimarães from INEGI, for
the opportunity to work within the Rotor Blade Extension project;
o to Prof. Christof Devriendt and Doctor Wout Weijtjens from OWI-Lab, for the interesting
conversations about modal identification and for the opportunity to use the data of the Vestas
V90-3.0MW wind turbine;
o to Eng. José Saraiva from Senvion and to Eng. Carlos Cardoso and Eng. Frederick Saborano
from Cavalum, for all the support given within the study of the Senvion MM82 wind turbine;
o to Prof. Álvaro Rodrigues and Prof. Fernando Marques da Silva, for the interesting suggestions
and conversations that helped improve this thesis;
o to my work colleagues and friends Nuno, Joana(s), Celeste, Berawi, Sandro, Fernando, Aires,
Ana, André, Ricardo and Sérgio, for the fun that was working in our H304;
o to my parents and brother, to whom I owe who I am;
o to Paulina, for putting up with my bad mood and, even thus, always cheer me up;
o to Fundação para Ciência e Tecnologia, for the financial support through the PhD fellowship
SFRH/BD/79328/2011.
9
NOMENCLATURE
‫ܣ‬ Area of cross section
‫ܣ‬, ‫ܤ‬, ‫ܥ‬, ‫ܦ‬ State-space model matrices
‫ܥ‬ଵ Damping matrix
‫ܥ‬஽ Drag coefficient
‫ܥ‬௅ Lift coefficient
‫ܥ‬௉ Power coefficient
‫்ܥ‬ Thrust coefficient
‫ܦ‬ Diameter, Damage
‫ܧ‬ Young modulus
‫ܨ‬ Excitation force
‫ܩ‬ Next-state output correlation matrix, Gyroscopic matrix
‫ܪ‬ Frequency response function
‫ܪ‬௦ Significant wave height
‫ܫ‬ Identity matrix, Second moment of inertia of the cross section
‫ܭ‬ Stiffness matrix, Kalman gain
‫ܭ‬஼ Geometric stiffness matrix
‫ܭ‬௠ Modal Kalman gain
‫ܯ‬ Mass matrix, Bending moment
ܰ஻ Number of blades
ܱ Extended observability matrix
ܲ Power
ܳ, ܴ, ܵ Noise covariance matrices
ܴ Correlation matrix, Radius, Rotation matrix
ܵ Spectrum matrix, Scour depth
ܶ Torque, Period of time

Peak period
Mean zero-crossing period
wind speed
Modal output matrix
Matrix with predictor variables
Axial induction factor
Angular induction factor
Airfoil chord
Innovation sequence
Natural frequency
Number of outputs sensors
Mass flow rate
Number of reference outputs
Pressure
Modal coordinates
Position of the blade under consideration
Input vector
Vectors containing noise due to sensor inaccuracy
Vector containing noise due to disturbances and modelling inaccuracies
State vector (which contains the displacement and velocity vectors of the system)
, Modal state vector at instant
̅ Mean value of the distribution
Output vector, displacement (in time domain)
State vector
, Modal state vector
Γ Reversed extended stochastic controllability matrix
Λ Modal state matrix
Ψ Matrix with eigenvectors
Ω Rotational frequency of the rotor

Angle of attack
Vector with weight parameters of the predictor variables
̂ Residual error
Pitch angle
Tip speed ratio, system poles
Damping ratio
Mass density
Standard deviation
′ Local solidity
Matrix with the mode shape configurations at the measured degrees of freedom, angle of inflow
Circular frequency, angular velocity of the wake
∗ Diagonal elements of a matrix
∗ Moore-Penrose pseudo-inverse
∗ Transpose operator
⊗ Kronecker product operator
∗ Absolute value
∗ Covariance
∗ Expectation value
∗ Real part
∗ Trace of a matrix
CAPEX Capital Expenditure
CM Condition Monitoring
FA Fore-Aft
FEM Finite Element Model
LCL Lower Control Limit
MAC Modal Assurance Criterion
MBC Multi-Blade Coordinate
OMA Operational Modal Analysis
OPEX Operational Expenditure
PDF Probability Density Function
RBE Rotor Blade Extension

RMS Root Mean Square
SCADA Supervisory Control And Data Acquisition
SHM Structural Health Monitoring
SS Side-Side
SSI-COV COVariance-driven Stochastic Subspace Identification
SSI-DATA DATA-driven Stochastic Subspace Identification
UCL Upper Control Limit
p-LSCF poly-reference Least Squares Complex Frequency Domain

CONTENTS
1. INTRODUCTION ............................................................................................................................21
1.1 Research Context .............................................................................................................................. 21
1.2 Objectives and Main Contributions ............................................................................................... 24
1.3 Organization of the Thesis ............................................................................................................... 26
2. FROM WINDMILLS TO THE MODERN WIND TURBINES ......................................................27

2.1 Technological Development ............................................................................................................ 27
2.1.1 The Onshore Development ......................................................................................................... 27
2.1.1.1 The First Attempts ............................................................................................................... 27
2.1.1.2 The First Large Machines.................................................................................................... 30
2.1.1.3 The Modern Wind Turbine Concept ................................................................................ 34
2.1.1.4 Current situation and future trends .................................................................................. 43
2.1.2 The Offshore Development ......................................................................................................... 44
2.1.2.1 The first offshore projects ................................................................................................... 44
2.1.2.2 Current situation and future trends .................................................................................. 45
2.2 Brief Market Analysis ....................................................................................................................... 49
2.2.1 The Start of the Danish and American Market ........................................................................ 49
2.2.2 Onshore Wind Turbine Market.................................................................................................. 49
2.2.3 Offshore Wind Turbine Market ................................................................................................. 53
2.3 Final Considerations ......................................................................................................................... 55
3. BASICS OF WIND TURBINES........................................................................................................57

3.1 General Layout .................................................................................................................................. 57
3.1.1 Wind Turbine Typology .............................................................................................................. 57
3.1.1.1 Rotor ...................................................................................................................................... 59
3.1.1.2 Nacelle ................................................................................................................................... 61
21
3.1.1.3 Tower .....................................................................................................................................62

3.1.1.4 Transition Piece ....................................................................................................................64
3.1.1.5 Foundation ............................................................................................................................64
3.1.1.6 Monitoring and Control System ........................................................................................72
3.1.2 Costs Breakthrough ......................................................................................................................72
3.1.3 Wind Turbine Components Reliability .....................................................................................76
3.2 Wind Turbine Performance.............................................................................................................80
3.2.1 Rotor Power Characteristics (Physics) .......................................................................................80
3.2.2 Axial Momentum Theory (The Actuator Disc) ........................................................................80
3.2.3 Angular Momentum Theory .......................................................................................................85
3.2.4 Blade Element Theory ..................................................................................................................88
3.2.5 Blade Element Momentum ..........................................................................................................90
3.2.6 Control Strategies of Wind Turbines .........................................................................................92
3.2.6.1 Stall Control ..........................................................................................................................92
3.2.6.2 Pitch Control.........................................................................................................................93
3.2.6.3 Rotor Speed Control ............................................................................................................94
3.3 Final Considerations .........................................................................................................................96
4. STRUCTURAL BEHAVIOUR OF WIND TURBINES .................................................................. 97

4.1 Structural Dynamics .........................................................................................................................97
4.1.1 Dynamic Characterization of the Tower ...................................................................................98
4.1.2 Dynamic Characterization of the Rotor.....................................................................................99
4.2 Foundation Soil Stiffness ................................................................................................................105
4.3 Tower-Rotor Dynamic Coupling ..................................................................................................110
4.4 Definition of Tower Stiffness .........................................................................................................111
4.5 Damping ...........................................................................................................................................113
4.5.1 Structural Damping ....................................................................................................................114
4.5.2 Aerodynamic Damping ..............................................................................................................114
4.5.3 Tower Dampers ...........................................................................................................................116
4.5.4 Soil Damping ...............................................................................................................................116
4.5.5 Hydrodynamic Damping ...........................................................................................................117
4.6 Cyclic Vibration Loads ...................................................................................................................118
4.7 Numerical Model of a Wind Turbine ...........................................................................................121
4.8 Final Considerations .......................................................................................................................125
22
5. OPERATIONAL MODAL ANALYSIS OF WIND TURBINES ....................................................127

5.1 Introduction to Operational Modal Analysis .............................................................................. 127
5.2 Application of OMA to Wind Turbines ...................................................................................... 129
5.3 Particularities of OMA Application to Wind Turbines ............................................................. 133
5.4 Output-only Stochastic Identification Methods ......................................................................... 136
5.4.1 Algorithms Based on Identification of State-Space Models ................................................. 136
5.4.1.1 State-Space model .............................................................................................................. 136
5.4.1.2 SSI-COV algorithm ........................................................................................................... 137
5.4.1.3 Kalman filter ....................................................................................................................... 143
5.4.1.4 SSI-DATA algorithm ......................................................................................................... 143
5.4.2 p-LSCF ......................................................................................................................................... 148
5.4.2.1 Half-Spectrum .................................................................................................................... 148
5.4.2.2 Right Matrix-Fraction Description.................................................................................. 149
5.4.2.3 p-LSCF algorithm .............................................................................................................. 151
5.5 Modal Response Estimation .......................................................................................................... 157
5.5.1 Forward Innovation Model ....................................................................................................... 157
5.5.2 Application to the Stochastic Identification Algorithms...................................................... 158
5.5.2.1 Application to the SSI-DATA algorithm ........................................................................ 158
5.5.2.2 Application to the SSI-COV algorithm ........................................................................... 160
5.5.2.3 Application to the p-LSCF algorithm.............................................................................. 160
5.5.3 Quantification of the Modal Contributions ............................................................................ 164
5.6 Auxiliary Tools to Overcome OMA Violations .......................................................................... 167
5.6.1 Coordinate Transformation - Nacelle Rotation ..................................................................... 167
5.6.2 Coordinate Transformation - Rotor Rotation ........................................................................ 168
5.6.3 Periodic Nature of the Excitation Forces................................................................................. 170
5.6.3.1 Kurtosis-based Methods ................................................................................................... 171
5.6.3.2 Cepstral Method ................................................................................................................. 172
5.6.3.3 Time Synchronous Averaging .......................................................................................... 173
5.7 Automated Operational Modal Analysis ..................................................................................... 175
5.7.1 Implemented Hierarchical Cluster Analysis ........................................................................... 176
5.8 Case Study – Izar Bonus 1.3MW/62 ............................................................................................. 179
5.8.1 Introduction ................................................................................................................................ 179
5.8.2 Description of the Wind Turbine ............................................................................................. 179
5.8.3 Measurement System ................................................................................................................. 181
23
5.8.3.1 Measurement System Based on Fibre Bragg Grating Sensors ......................................181

5.8.3.2 Measurement System Based on Accelerometers ............................................................181
5.8.4 Analysis before the installation of the RBE .............................................................................184
5.8.4.1 Analysis of fibre Bragg grating sensors measurements .................................................184
5.8.4.2 Analysis of accelerometers measurements......................................................................189
5.8.5 Analysis after the installation of the RBE ................................................................................198
5.8.5.1 Analysis of the fibre Bragg grating sensors measurements ..........................................198
5.8.6 Conclusions .................................................................................................................................201
5.9 Case Study – Vestas V90-3.0MW..................................................................................................202
5.9.1 Introduction.................................................................................................................................202
5.9.2 Wind Turbine Description ........................................................................................................202
5.9.3 Monitoring System Description................................................................................................204
5.9.4 Acceleration Data Sets ................................................................................................................204
5.9.5 Modal Results ..............................................................................................................................208
5.9.6 Conclusions .................................................................................................................................216
6. FATIGUE ASSESSMENT OF WIND TURBINES ........................................................................ 217

6.1 Introduction to Fatigue Problems in Wind Turbines ................................................................217
6.2 Fatigue Design Loads ......................................................................................................................218
6.3 Application to Wind Turbine Support Structures ......................................................................222
6.3.1 Main Structural Details ..............................................................................................................222
6.3.2 Fatigue Analysis According to Standards/ Guidelines ...........................................................223
6.4 Fatigue Assessment Using Accelerometers..................................................................................225
6.4.1 Proposed Procedure for Fatigue Assessment ..........................................................................228
6.4.1.1 Estimation of the Quasi-Static and Dynamic Stresses Contributions to Fatigue ......233
6.4.2 Virtual Sensors ............................................................................................................................240
6.5 Final Considerations .......................................................................................................................243
7. STRUCTURAL MONITORING OF WIND TURBINES ............................................................. 245

7.1 Introduction to Structural Monitoring of Wind Turbines ........................................................245
7.1.1 Condition Monitoring of Wind Turbines ...............................................................................248
7.1.2 SHM of Wind Turbines .............................................................................................................249
7.2 Vibration-Based SHM System .......................................................................................................252
7.2.1 Damage Detection ......................................................................................................................252
7.2.1.1 Removing of Environmental and Operational Effects on Natural Frequencies ........252
24
7.2.1.2 Damage detection .............................................................................................................. 257

7.2.2 Fatigue Monitoring..................................................................................................................... 259
7.2.2.1 Estimation of the Remaining Lifetime ............................................................................ 261
7.2.3 Overview of the Dynamic Monitoring System Main Steps ................................................... 262
7.3 Case Study – Senvion MM82 ......................................................................................................... 265
7.3.1 Introduction ................................................................................................................................ 265
7.3.2 Wind Turbine Description and Numerical Models ............................................................... 265
7.3.3 Operating Conditions ................................................................................................................ 270
7.3.4 Monitoring System Description ............................................................................................... 273
7.3.5 Preliminary Results..................................................................................................................... 275
7.3.6 Continuous Characterization of the Dynamic Properties..................................................... 281
7.3.6.1 Dynamic Characterization of the Wind turbine ............................................................ 281
7.3.6.2 Automated Identification of the Modal Parameters ..................................................... 288
7.3.6.3 Damping Estimation from Free Decay Responses ........................................................ 298
7.3.6.4 Participation of the Vibration Modes/ Harmonics to the Measured Dynamic
Response .............................................................................................................................................. 303
7.3.6.5 Removal of Operational and Environmental Effects .................................................... 307
7.3.6.6 Damage Detection ............................................................................................................. 316
7.3.7 Fatigue Assessment Results ....................................................................................................... 322
7.3.8 Optimization of the Monitoring System ................................................................................. 332
7.3.8.1 Layout 1 (+74.988 m level) ............................................................................................... 333
7.3.8.2 Layout 2 (+48.392 m level) ............................................................................................... 340
7.3.8.3 Layout 3 (+74.988 m and +48.392 m levels) ................................................................... 345
7.3.9 Conclusions ................................................................................................................................. 350
8. CONCLUSIONS AND FUTURE RESEARCH ..............................................................................353

8.1 Conclusions...................................................................................................................................... 353
8.2 Future Research ............................................................................................................................... 356
Appendix A - MODEL RESPONSE ESTIMATION - VALIDATION ..............................................357

A.1. Introduction ..................................................................................................................................... 357
A.2. Numerical Simulation .................................................................................................................... 358
Appendix B - IZAR BONUS 1.3MW/62 WIND TURBINE – MODE SHAPES ...............................363

B.1. Introduction ..................................................................................................................................... 363
B.2. Analysis of Accelerometers Measurements ................................................................................. 363
25
B.3. Analysis of Fibre Bragg Grating Sensors Measurements (After RBE) .....................................368
REFERENCES..................................................................................................................................... 371
26
Chapter 1
1
INTRODUCTION
1.1 RESEARCH CONTEXT

The electricity sector has undergone profound changes in the last years. The awareness for the
environmental conservation, in particular to the greenhouse effect, alongside with the desire of higher
energy independence by importing countries on fossil fuels, has led to the investment in the
exploitation of renewable energy sources. In fact, renewable sources are in the focus of current energy
policies. Almost all major world economies have developed studies and programs to introduce more
clean energy into their energy mix. In particular, The European Union has been implementing a
program with challenging objectives for 2020, known as “20-20-20” targets (Commission of the
European Communities, 2008). Among them, it was defined a 20 % share of renewable energies in
overall European Union energy consumption by 2020 (European Commission, 2013), including a
target of 34 % of the electricity demand supplied by renewable sources (European Commission,
2015c). Subsequent programs for 2030 (European Commission, 2015a) and 2050 (European
Commission, 2015b) are already under development.
In Portugal, until a few decades ago, the hydroelectric production represented almost the whole
renewable contribution to the energy mix. From the 1990s, wind energy began to arouse some interest
and the initial wind farms were installed. In the next decade, large investments in the implementation
of wind turbines were made, which increased the importance of this source in the national electricity
mix. Successively, wind power has gained importance in the national panorama, recently achieving
power production numbers close (or even higher) to the hydropower generation. Currently, Portugal
has been presenting very good results in the introduction of energy from renewable sources in its
electrical system. The results obtained in 2015 are very good indicators: 50 % of the electrical energy
demand was supplied from renewable sources, with wind power accounting for 22 % (Portuguese
Renewable Energy Association (APREN), 2015). This value confirms the importance of this source in
the national electric system.
Apart from these aspects, two additional perspectives may promote (once again) the installation of
new wind turbines in Portugal (Government of Portugal, 2015). The first one is related to the future
prospects from the national government of achieving a share of 40 % of the gross final consumption of
energy from renewable sources by 2030, including a strong participation of wind power for electric
power demand. The second perspective is related to the future possibility of establishment of European
electric power interconnections. This possibility is related to the export of energy produced by
renewable sources to countries with difficulties to meet the targets agreed with the European
21
Introduction
Commission. This interconnection, which could achieve 15 % of the total installed capacity by 2030,
would lead to a considerable boost of implementation of new wind power plants. Alongside, an
important portion of the installed wind power capacity in Portugal is achieving its design lifetime, with
repowering operations being expected in the next years.
Given this policy background, it is consensually accepted that wind power is a key element in the
present and future electric energy strategy. New wind turbines are expected to be installed in the next
few years with power outputs considerably higher than the nowadays common values, mainly due to
cost aspects. In addition, offshore locations are becoming very attractive to install turbines due to their
higher production yields. Three significant difficulties will be faced by this new wind turbine models:
o The increase of the wind turbines dimension, particularly the increase of the height of tower
and blades length, will turn the wind turbine into more flexible structural systems.
Consequently, the study of the dynamic characteristics of the structure will be of major
importance in order to avoid resonance problems and rapid wear of the components;
o The installation of wind turbines in offshore locations is still shrouded in some uncertainty
concerning the real wear suffered by the structures. On the other hand, the costs associated
with offshore wind turbines are still high, which demand for optimization in their design;
o Apart from reducing the cost of installation of the wind power plants (CAPEX), the reduction
of the costs over the period of its operation (OPEX) also represents an important vector for
competitiveness. A crucial aspect to lower this cost is the optimization of the maintenance
strategy. Considering the high power outputs of new models, the downtime of a single wind
turbine may represent a significant income loss. In that sense, several monitoring system are
being developed to remotely assess the condition of several components of the wind turbine,
allowing to optimize the maintenance interventions according to expected loss of production
during downtime, availability of replacing components and meteorological conditions (in the
case of offshore wind farms or onshore wind farms in mountains).
These facts address the new challenges that, from a structural point of view, will concern the owners of
wind farms. In that sense, a real-time monitoring system capable of assessing the structural integrity of
the support structure of wind turbines (including tower and foundation) can be a very useful tool to
aid decision-making. With a monitoring system, the wind farm owner can analyse the structural health
of the wind turbines in remote locations and adapt the maintenance strategy according to the received
data. In short, in possession of this information, the owner could maximize the investment.
In order to be effective, a monitoring system installed at a wind turbine support structure should be
able to answer two important questions:
o Is the structure damaged?
o How is the fatigue condition of the structure? Is the structure able to fulfil its period of life? Or
should the structure be disassembled before (or after) this period?
To meet these two aspects, different modules need to be integrated into one vibration-based structural
health monitoring (SHM) system: a structural anomalies detection monitoring system and a fatigue
monitoring system.
The structural anomalies detection monitoring system is a tool with the aim of detecting damage. It
can be based on the continuous tracking of the modal properties of the structure. For this purpose,
22
Chapter 1
tools already tested on civil engineering structures can be used in wind turbines in order to extract the
monitored features. These tools, from a branch of the experimental modal analysis field named as
Operational Modal Analysis (OMA), permit to identify the modal characteristics of structures during
the various operating conditions of the turbines. Thus, considering that damage is usually associated
with stiffness reductions, the natural frequencies of the various vibration modes can be used as damage
indicators. The study of the evolution and variation of these indicators indirectly evidence the
existence of damage in the structure.
The second module comprises the fatigue analysis of the structure. Fatigue assessment is a key point in
wind turbines, since these structures are mainly subjected to dynamic loads. Thereby, this module
permits to assess the evolution of the wear of the wind turbine support structure over time. This
information can then be compared with the estimated lifetime of the structure in order to validate the
suitability of this period.
23
Introduction
1.2 OBJECTIVES AND MAIN CONTRIBUTIONS

The main goal of this thesis consists on the development of a complete vibration-based monitoring
system capable of delivering useful results to assess the integrity of the wind turbine support structure.
To achieve this goal, this thesis addresses the following objectives:
o Implementation of output-only modal identification algorithms in the context of wind
turbines and development of strategies for continuous identification of the modal properties of
these structures;
o Development of a monitoring system to identify structural changes in the support structure of
wind turbines based on the identified modal properties;
o Development of a methodology to estimate the fatigue damage at the support structure of
wind turbines based on a small number of sensors;
o Integration of the two components (damage identification and estimation of the fatigue wear)
of the vibration-based dynamic monitoring system in a single software package,
demonstrating with full-scale applications the usefulness of the system in the management of
wind farms.
It is expected that the outcomes presented in this thesis may represent a step forward in the
development of global systems to monitor the most important components of wind turbines. In that
sense, the main contributions of this thesis are the following:
o Synthesis of the major technical progresses of wind turbines and current state of evolution
(Chapter 2);
o Description of the main components of a wind turbine, with a main focus on the different
types of support structure, including costs breakthrough and reliability (Chapter 3);
o Description of the main characteristics related to the dynamic behaviour of wind turbine
structures (Chapter 4);
o Analysis of the implications of the application of output-only modal identification algorithms
to vibration signals from wind turbines (sections 5.3 and 5.7);
o Adaptation of the post-processing tool originally developed to the SSI-DATA algorithm to
decompose the vibration signals into modal responses to be used together with the SSI-COV
and p-LSCF algorithms (section 5.5.2);
o Application of the developed algorithms to data sets recorded on the blades and tower of a
Izar Bonus 1.3MW/62 wind turbine using two different measurement systems: based on fibre
Bragg grating sensors and accelerometers (section 5.8);
o Application of the developed algorithms to data sets collected during a short period on the
support structure of an offshore Vestas V90-3.0MW wind turbine;
o Development of a strategy to estimate the fatigue state along the support structure of a wind
turbine based on a reduced number of sensors (section 6.4.1);
o Creation of a vibration-based dynamic monitoring system, specially designed to meet the
requirements of a wind turbine, able to perform, in a first module, the online automatic
24
Chapter 1
identification of the modal parameters under all operating conditions, to remove the influence
of operational and environmental effects and to compute indices permitting to detect the
presence of small structural anomalies; and, in the second module, estimate the fatigue
conditions along the support structure of a wind turbine (section 7.2);
o Conception, assembly and operation of a monitoring system on a Senvion MM82 wind
turbine for a period of one year, where the developed routines were tested, demonstrating the
ability of the system to detect the occurrence of small damages (section 7.3);
o Optimization analysis concerning the ability (and loss of accuracy) of the dynamic monitoring
system to operate with a reduced number of installed sensors (section 7.3.8).
25
Introduction
1.3 ORGANIZATION OF THE THESIS

The present thesis is divided in 9 chapters. It was the intention of the author to start from a global
perspective of the wind energy industry to adequately frame the problem of wind turbine monitoring
based on modal analysis. The text is then progressively oriented towards more detailed aspects of the
implementation of the vibration-based monitoring system. A description of the contents addressed in
each chapter is presented in the following paragraphs.
Chapter 1 introduces the present work. A small contextualization of the work is presented followed by
a brief description of the organization of the text throughout the chapters.
Chapter 2 describes the evolution of the wind turbines from the initial models to the current offshore
solutions. In addition, some important aspects of the wind energy market are introduced and some
potential future trends are also analysed.
Chapter 3 is essentially dedicated to the description of the components of current wind turbine
models. The operation of a wind turbine is explained according to the Blade Element Theory, serving
as a mean for a better understanding of some operational aspects of these power plants.
Chapter 4 introduces important aspects concerning the dynamic characteristics of wind turbines. The
typical dynamic behaviour of the tower and rotor blade elements is described, as well as some aspects
regarding the different contributions to the damping of the structure.
Chapter 5 addresses the topic of the Operational Modal Analysis, which is a core aspect of the
dynamic monitoring system. Some output-only algorithms are introduced, namely the SSI-COV, the
SSI-DATA and the p-LSCF methods, and the main challenges associated with their application to wind
turbines are addressed. In addition, some capabilities of these algorithms are introduced, namely post-
processing tools to decompose the dynamic response into modal and harmonic responses. Then, the
algorithms previously introduced are used to analyse vibrational data from two case studies: the Izar
Bonus 1.3MW/62 and the Vestas V90-3.0MW wind turbines.
Chapter 6 begins with a brief description of the main dynamic loads acting on a wind turbine and the
particularities of the fatigue design of these structures. The chapter continues with the description of a
methodology to estimate the fatigue condition of the support structure based on a reduced number of
sensors and with the presentation of a virtual sensors technique to estimate the acceleration at
unmeasured locations.
Chapter 7 presents the bridge between the algorithms and methodologies described in Chapter 5 and 6
and the developed dynamic monitoring system. The methodologies implemented for removing the
environmental effects on the damage indicators are introduced, as well as the statistical procedure to
identify structural damage. Alongside, the estimation of the remaining fatigue lifetime based on a
reduced monitoring time is also addressed. Then, a detailed description of the monitoring program of
the Senvion MM82 wind turbine is introduced. The main results obtained with the routines developed
in the scope of the monitoring system are presented
Finally, Chapter 8 summarizes the main conclusions of this thesis and highlights some potential future
research topics.
26
Chapter 2
2
FROM WINDMILLS TO THE MODERN
WIND TURBINES
2.1 TECHNOLOGICAL DEVELOPMENT

2.1.1 THE ONSHORE DEVELOPMENT
Wind power has been exploited for several centuries, from mechanical to, more recently, electrical
purposes. The first wind turbines came almost directly from wind mill concepts. From these models,
several technological progresses were made, which led to current multi-MW turbines. Thus, in this
section, an overview of the most important steps in the technological development of wind turbines is
presented.
2.1.1.1 The First Attempts

The first references to the development of electricity generator windmills are shared between Charles
F. Brush in 1888 (Shepherd, 1990) and Poul LaCour in 1891 (Gasch and Twele, 2012). It is interesting
to note that, even in the beginning of its technological development, different strategies from a
conceptual point of view were followed. The Brush wind turbine (Figure 2.1), installed in Ohio,
Cleveland, in 1888, was an adaptation of the American windmill. It presented a power output of 12 kW
based on a low-speed, multi-blade design. On the other hand, the LaCour wind turbine, erected in
1891 in Askov, Denmark, was designed following the principles of the new engineering science of
aerodynamics (Figure 2.2). In that sense, it stepped away from the low tip-speed ratio design (ratio
between the blade tip speed and the wind speed) of the Brush model since it was not suitable for
electric generation (Gasch and Twele, 2012). The LaCour model was based on the European windmill
and presented some aerodynamic refinements, such as low solidity rotor (ratio between the blade area
and the swept area), leading-edge camber and a low drag coefficient (Shepherd, 1990). This design was
later adapted for commercial purposes, being characterized by its reliability and efficiency (Hau, 2006).
27
From Windmills to the Modern Wind
W Turbines
Figure 2.1 - The Brush win

nd turbine (Anon, 1890) Figu
ure 2.2 - LaCour wind turbine (Hau, 2006)
After the Fiirst World War

W (1914-19 918), some iimportant deevelopments took place. In the theorretical
field, Albertt Betz presennted in 19200 the actuatoor disc theorry to the win
nd turbine (BBetz, 1920). In his
paper (laterr edited in a book (Betz, 1926)), Betzz presents hiss considerations about thhe ideal maximum
factor of kinnetic energyy extractable from the fluux of energy from wind (also called ““Betz Limit”” – see
Chapter 3).
Some new m models weree presented during
d this pperiod. Fromm those, the one
o developeed by the bro others
Marcellus aand Joseph Jaacobs, whosee developmeent started in n 1922 in thee USA, standds out. The model
m
was designeed with a th hree-bladed rotor
r (the noowadays stan ndard design n) with a 4 m diameter rotor
(Figure 2.3)). One of thee most intereesting aspectss in to the development
d of this windd turbine is related
r
to the severre vibration problems detected
d wheen an initial rotor solution based onn two bladees was
tested. It waas noted thatt, during yawwing operatiions of the tu urbine, abnoormal vibratiions were detected
due to chan nges in gyrosscopic inertiaa (Shepherd, 1990). This is one of thee first descripptions of dyn namic
problems reelated to winnd turbines an nd led to a ddifferent desiggn solution, based on a thhree-bladed rotor.
This wind tu urbine was a great comm mercial succeess, with mod dels with powwer output raanging from 1.8 to
3.0 kW (Haau, 2006).
28
Chapter 2
Figure 2.3 - Jacobs wind turbine

t (Hau, 2006)
2
Two innovative concepts

c of wind
w energyy exploitation
n were preseented duringg this period. The first iss
referrred to the Saavonius rotorr, introducedd in 1925 (Saavonius, 193 31). It was chharacterized by a verticall
axis tturbine, whicch S-shaped cross section ns working byb lift (Figurre 2.4). This type of turbine achieved
d
somee popularity due d to its sim
mple construuction and itss ability to op
perate with wwind from an ny direction..
Howeever its low w efficiency prevented
p itts commerciial success (Shepherd, 11990). Neverrtheless, thiss
conceept is still seeen in some sm
mall-scale wiind exploitattion devices.
The ssecond concept was pressented by Geeorges Jean Darrieus
D in 1926 (Darrieeus, 1931). The
T Darrieuss
Rotorr had two orr three bladess connected at the top annd at the botttom to the ccentral shaft (Figure 2.5)..
This concept presented somee advantagess, such as thee placement of the poweer train nearr the ground d
level (where main ntenance is easier) and thhe exemption
n of a yaw sysstem. Furtheermore, the curved-shape
c e
of thee blades wass designed too minimize thhe bending stress
s along its length (SShepherd, 1990). For thiss
reasoon, the bladess could be ch
heaper. This cconcept was later employyed in an utillity-scale mo
odel.
Figu
ure 2.4 - Savonius Rotor Figure 2.5
2 - Darrieus Rootor (Darrieus, 1931)
29
9
W Turbines
In the decadde of 1930s, thet typical wind

w turbine was compossed by a horizzontal upwinnd rotor (thee rotor
is placed in front of thee tower, orien ntated to thee wind – seee Chapter 3) with two orr three blades with
low solidityy. The rotor blades
b operaated at a variiable speed with
w a fixed pitch
p (angle oof blades alo
ong its
longitudinaal axis). Its ellectrical systeem was com mposed by a 121 or 32 V DCD generatorr. This typicaal unit
worked unaattended and d was said to be b very reliabble and longg-lived (Sheph
herd, 1990).
The Great DDepression in n the 1930s alongside wiith the exten
nsion of the Rural
R Electriffication Act in the
USA (a program in whiich loans werre provided tto the installlation of elecctrical distribbutions systeems in
rural areas), led to the demise of th
he small turbbines. Furtheermore, the small turbinnes up to thiis date
were not coost effective to
t compete with
w the invvestments in central pow wer plants. At this point in the
history of w
wind turbines, the investmments wouldd be turned over to largeer wind turbbines conneccted to
the electric grid in ord der to reducce the cost of wind en nergy and to o prevent fuuture problem ms of
worldwide eenergy shortages (Shepheerd, 1990).
2.1.1.2 Th
he First Largge Machines
One of thee first large projects waas the 100 kkW, 30 m diameterd Ballaclava windd turbine in 1931
(Sectorov, 11973) (Figuree 2.6). It had
d a 30 m heigght steel lattice tower an
nd was locateed at Balaclavva, on
Crimea, near the Blackk Sea. It had d a three-blaaded rotor with
w omatic yaw system and pitch
an auto
control (adjjustment oveer the angle of
o the blade ffacing the wiind – see Chaapter 3). Althhough it had
d some
primitive soolutions (succh as gears made
m of woodd and simple blade surfacce treatment)), it achievedd good
results. Thiis machine was
w connected to a steam m power statio on and is con
nsidered the first wind tu
urbine
to be intercoonnected to an AC utilityy system (Sheepherd, 19900).
Figure 2.66 - Balaclava win

nd turbine (She
epherd, 1990) Figure
e 2.7 – Smith-Putnam wind tuurbine (Hau, 2006)
In late 19399, the Ameriican Engineeer Palmer C.. Putnam approach the S. S Morgan Sm mith Compaany of
York with a new wind tu urbine with the latest tecchnology (Vooaden, 1943). The constrruction of thiis new
project startted in 1941 on
o the top off a mountain n called “Gran
ndpa’s Knobb” (near Vermmont, USA) and it
is considereed a landmaark in the wiind energy hhistory. Thiss model refleects the demmand for red ducing
production costs with laarge wind turrbines with hhigh power outputs.
o
30
Chapter 2
Technically, the Smith-Putnam wind turbine was a remarkable and giant project for that time. It had a
53.3 m diameter, 2-bladed upwind rotor that generated a power output of 1.25 MW with a
synchronous (AC) generator (Shepherd, 1990) and a steel lattice tower with a height of 33.5 m
(Voaden, 1943) (Figure 2.7). It included some technological innovations such as (Shepherd, 1990):
o Full span active control of the blade pitch (the blades were made of stainless-steel);
o The use of individual flapping hinges on the blades to reduce gyroscopic loads on the shaft (a
similar problem previously faced in the development of the Jacobs turbine);
o Active yaw control by mean of a servomotor turning a pinion meshing with a large bull gear
between the machinery house (or nacelle) and the tower.
The generator was connected to the electrical system at wind speeds higher than 8 m/s, reached its
rated power at 13.4 m/s (Voaden, 1943) and it ran until the maximum wind speed of 32 m/s (Johnson,
2001). The Smith-Putnam wind turbine is considered the first large wind turbine with two blades
(Johnson, 2001) and it was the largest wind turbine for almost 40 years. Notwithstanding, it faced
some important drawbacks. From the technical point view, it was concluded that the blades were
undersigned, presenting several cracks after some operating time. From the economic point of view, it
demonstrated that some improvements would have to be taken in order to be become cost effective
with respect to the conventional power generation processes (Gasch and Twele, 2012).
At this time, the increase of the capacity of the wind turbines was already a confirmed trend. Although
the Smith-Putnam wind turbine had been a one-off project, some previous and future projects
confirmed this trend, some of them with utopic designs (like the design of Hermann Honnef with a
power output of 20 MW (Hau, 2006)).
In 1939, the World War II began. From the end of World War I (1918) to 1939, the interest in wind
power in Denmark decreased (Vestergaard, Brandstrup et al., 2004). The price of coal and oil had
become sufficiently low to decelerate the interest in wind energy. However, with the beginning of the
World War II, the increase in the price of fossil fuels led, once again, to the rise of wind energy interest.
Some old commercial LaCour machines were reactivated and some new wind turbines were built.
Also in Denmark, a new company (F.L. Smidth Company) developed a new model during this period,
relying in a small, conservative design. In addition, new concepts of modern aerodynamics (modern
airfoils) were applied to the blades (Ackermann and Söder, 2002). Under this philosophy, reliable
models with power outputs ranging from 50 kW to 70 kW were developed, with rotor diameters of
17.5 m and 24 m, respectively. It is interesting to note that the model with the higher power output was
designed with a concrete tower, not a very common solution at the time.
The models developed by the Smidth Company are an example of a development trend, choosing
simple designs over heavy and complicated models, in opposition to the Smith-Putnam turbine design.
In fact, the Smidth wind turbines are considered the forerunners of modern wind turbine generators
by some authors and one of the pioneers of the Danish wind energy market (Shepherd, 1990;
Ackermann and Söder, 2002). By the first half of the 20th century, the wind energy market was already
a reality due to the number of small wind turbines installed, being the primary energy source in some
areas (Hau, 2006).
31
W Turbines
a
a) b)
Figure 2.8 – Smith wind turbines (Hau, 2006):

2 a) 50 kW
W model with steel lattice towe
er; b) 70 kW moodel with concrete
tower
During the 1930s and 1940s, much theoretical aand design acctivity in the wind energyy field was held h in
Germany. T This developm ment was un ndertaken byy the German n Reich with the aim of a self-sufficien ncy in
fuel power (Hau, 2006). For this purpose,
p thee “Reichsarbeitsgemeinscchaft Windkkraft” (RAW W) was
h renowned scientists,
created with s tecchnicians an nd industrialss. Among thee projects dev
eveloped und der the
RAW progrram, the pro oject led by the engineerr Franz Kleiinhenz, in 19 937, stands oout, reflectin
ng the
ambition off building large machinees to reduce the cost of energye produuction. This wind turbin ne had
remarkable characteristtics like a 130 m rotor ddiameter with h a three or four-bladedd downwind rotor.
This rotor w
would have a power outp put of 10 MW W at a tip speed of 5 and the
t generatorr would havee been
from the diirect drive tyype. The designed towerr was a guyed d tubular steeel tower, wiith the uppeer part
yawing withh the rotor. The
T concept was improvved from 19338 to 1942 an nd was readyy for constru uction.
However, thhe war preveented its consstruction (H au, 2006). It is interesting to note thaat, even nowadays,
some of theese characteriistics are stilll not widely iimplemented d in commerrcial wind turrbines
Apart fromm Denmark and a USA, oth her regions aalso showed some intereest in the expploitation off wind
energy, alth
hough in a sm mall scale. In
n the Unitedd Kingdom, there
t was a wind
w energy program ru unning
from 1948 to the earlyy 1960s (Speera, 1994). A Among some different designs
d deveeloped durin ng the
program, th he Enfield-A
Andreau wind d turbine prooject stands out from th
he others duee to its innoovative
design, althoough associaated with low
w efficiency (H
Hau, 2006))..
nce, there waas some interest in the ddevelopment of experimental wind tuurbines durin
Also in Fran ng this
time. The 30.1 m rotor diameter, 8000 kW Best-R Romani turb
bine and the 35 m rotor ddiameter, 1.0
0 MW
Vadot turbiine are examples of projects from thiss time (Hau, 2006).
During the post-war perriod in Germmany, Ulrich Hutter was the main name in the fieeld of wind power.
p
Hutter, whoo had alreadyy been distin
nguished on previous papers on the theory
t of win
ind turbines (Hau,
2006), cond
ducted some experimentaal tests durinng the 1950s. His studies culminated in the 34 m rotor
diameter W
W-34 wind turrbine with an n AC synchrronous generrator (Gasch and Twele, 22012) with a rated
power output of 100 kWW at a wind speed of 8 m//s (Figure 2.99) (Johnson, 2001). This tturbine, ereccted in
32
Chapter 2
1957,, had a two o-bladed, dow wnwind rotoor with a siimple hollow w pipe toweer with 22 m of heightt
suppoorted by gu uy wires (Jo ohnson, 20001). One off Hutter’s main m prioritiies was the lightweightt
consttruction. Forr this reason
n, the bladess were madee of an advaanced glass-ffibre compossite materiall
(Hauu, 2006) and they were aerodynamic
a cally refined, being connected to a teeetering hub b (see Figuree
3.6). T
The solution
n found for th his connectioon was anothher innovatioon: instead oof a more com
mplex design n
like tthe one usedd in the Smmith-Putnam wind turbin ne, the teeteering movem ments of thee rotor weree
aeroddynamically damped by mechanically
m y coupling th
he teetering angle to the blade-pitch angle (Hau,,
2006)). The pitch mechanism
m of
o the bladess changed theeir angles at higher
h wind speed to keeep a constantt
speedd of rotation of the rotorr (Vargo, 19774). This win
nd turbine is considered as an examp ple of a third
d
desiggn philosophhy, relying on n the reducttion of the structure
s weeight and onn the optimization of itss
aeroddynamics.
Figure 2.9 – W-34 wind turbine (Hau, 20006) Figure 2.1

10 – Gedser winnd turbine (Hau
u, 2006)
One oof most inno ovative and reliable

r windd turbines buuilt during th
his time was tthe Gedser wind
w turbinee
(Figuure 2.10), dessigned by Johhannes Jull aand located in
i Denmark.. This wind tturbine was in operation n
betweeen 1958 an nd 1967 (Lun ndsager, Fraandsen et al., 1980), con nnected to thhe Danish public
p powerr
systemm (Johnson,, 2001). The Gedser win nd turbine haad a 24 m diameter, threee-bladed, upwind
u rotorr
with an induction n generator (rather
( than the more co onventional, at
a that time, synchronouus generator))
(Man nwell, McGo owan et al., 2010)
2 achievving a powerr output of 200 kW at a wind speeed of 15 m/ss
(Lund dsager, Franndsen et al., 1980). Thee tower had a height off 24 m and was made of concrete..
Althoough these ch haracteristics were not soo impressivee, this wind turbine had ssome major innovations,,
such as the poweer control th hrough aeroddynamic staall: its bladess had an aerrodynamic design
d whichh
causeed flow separration underr strong windd conditions,, acting as a completely ppassive poweer limitation,,
avoidding the neeed for pitch control, likke the Smidtth wind turb bine. For aeerodynamic braking,
b thee
turbin ne was equiipped with turnable bladde tips which were activ vated by cenntrifugal forcces at a grid
d
failurre (Gasch and Twele, 201 12). Comparred to the Sm mith-Putnam wind turbinne, the Gedseer was much h
less ccomplex, froom a mechaanical point of view (V Vestergaard, Brandstrup et al., 2004 4). All thesee
charaacteristics an
nd innovation ns led to a peeriod of operation withoout major maaintenance (V Vestergaard,,
Brand dstrup et al.., 2004). The Gedser wiind turbine was decomm missioned inn 1967 and it was laterr
33
3
From Windmills to the Modern Wind Turbines
refurbished in the mid-1970s at the request of NASA for measurements, in the framework of an
American program (Hau, 2006).
Although Hutter and Juuls’ projects were considered a success, the wind turbine projects in general
were not yet reliable, with numerous problems and faults to be dealt with (Hau, 2006). Besides these
reasons, the extremely low prices of primary fuels after the World War II, along with the increased
investments in nuclear energy, demonstrated that the wind generated electricity was too expensive to
compete with other sources (Gasch and Twele, 2012). Due to this, the wind energy industry suffered
another breakdown during the 1960s.
2.1.1.3 The Modern Wind Turbine Concept

In the late 1960s - beginning of 1970s, the awareness of dependency on the oil exporting countries
started to take place. The oil crisis in 1973 and 1979 confirmed this misgiving, having led to supply
problems and an increase in the oil prices in the western economies. With the aim of reducing this
situation, national programs for the development of renewable energy sources were created.
In the USA, the Federal Wind Energy Program began in 1972 (Ramler and Donovan, 1979). The
National Aeronautics and Space Administrator (NASA) was selected to manage the technology
development and initial deployment of large wind turbines in the Lewis Research Center. This project
represents one of the most extensive development program in the wind energy field, in which several
variations of structural configurations and mechanisms were tested.
The first wind turbine developed and built by this program was the NASA/DOE Mod-0 in 1975. This
wind turbine was a two-bladed, 38 m rotor diameter with a power output of 100 kW at a wind speed of
8 m/s (Ramler and Donovan, 1979). This turbine had a 30 m high tower. The Mod-0 wind turbine was
extensively tested with a large number of modifications along its operation life: with a teeter and rigid
hub, different blade materials, with pitch and stall control, with stiff and more flexible supporting
towers, different kinds of generators, among others (Spera, 1994) (Figure 2.11). The turbine operated,
in most of the time, as a downwind rotor, although the upwind solution had also been tested, as well as
a one-bladed rotor solution (Corrigan and Ernsworth, 1986). One of the most noticeable characteristic
of this wind turbine was the influence of the tower in the blade vibration when the rotor was
downwind, one important dynamic characteristic of this type of turbines. The Mod-0 was dismantled
in 1987 (Spera, 1994).
34
Chapter 2
Figure 2.111 – MOD-0 wind

d turbine variattions
The ssecond devellopment of the

t NASA/D DOE program m was the Mo od-0A wind turbine. Thee purpose off
this m
model was to t study, ideentify and soolve technicaal and operaational utilityy interconneection issuess
(Sperra, 1994). Th
his model waas basically a version of the Mod-0 with w a largerr gearbox an nd generatorr
(200 kW) (Ramller and Don novan, 19799). Four pro ototypes were built for four different locationss
betweeen 1977 and d 1980. Each
h location reepresented a different sceenario both for grid inteegration and d
wind conditions. From this prroject, it was proven that, while not yet cost-effecttive, wind tu
urbines couldd
be su
uccessfully in
ntegrated intto the normaal operations of a utility y, producingg high-qualityy AC powerr
(Sperra, 1994). The MOD-0A project
p endedd in 1982 (Sppera, 1994).
The M MOD-1 project entered in service in 11979. The wiind turbine had h a 61 m di diameter dow wnwind rotorr
(the llargest ever built
b (Spera, 1994)) with two steel blades (Ramler and Donovvan, 1979) (F Figure 2.12)..
This m model had a rated powerr of 2.0 MW at a wind speed of 14.6 m/s m (Johnsonn, 2001). Thee tower had a
steel truss design n and was 42.74 m highh. This wind d turbine was
w considere red the first operationall
megaawatt sized wind
w turbine in the worldd (Collins, Sh haltens et al., 1982). How
wever, the MO OD-1 modell
faced
d some probllems. It was too t heavy an nd, consequeently, too exp pensive, mainnly due to itts stiff tower,,
to itss heavy, commplex and laarge full spaan pitch con ntrolled blades and to itts bedplate (Ramler
( and
d
Donoovan, 1979). The effect of the towerr on the blaades created high impulssive loads to o the rigidlyy
moun nted blades, becoming cllear that thiss design philosophy would not be suiitable for a 20 2 - 30 yearss
periood of life (Sppera, 1994). A new moddel, MOD-1A A, was desig gned with soome solution ns aiming too
avoidd these problems but, duee to funding issues, it was not built (C Collins, Shalttens et al., 19
982). Besidess
technnical and ecoonomic prob blems, there were also so ome complaints from loccal residents about noisee
and ttelevision intterference (CCollins, Shalttens et al., 19982). The MO OD-1 projectt ended in 1988 with thee
disasssemble of thee wind turbin ne (Johnson,, 2001).
35
5
W Turbines
Figure 2.12
2 – MOD-1 w
wind turbine (W
Wikipedia, 2009
9)
The next iteeration of thee NASA proggram started in 1977 with h the MOD-2 2 project (Goordon, Andrews et
al., 1983) (F
Figure 2.13). This project was design ned as a sequuence of the findings andd developmeents of
the previous NASA/DO OE projects (M MOD-0, MO OD-0A and MOD-1)
M and
d, for that reaason, it is referred
to as a secoond generatiion machinee (Ramler an nd Donovan, 1979). Thee project of the MOD-2 wind
turbine wass a design evvolution of the t proposedd design forr the MOD-1A model. T The MOD-2 wind
turbine wass a two-bladeed, upwind rotor
r with a ddiameter of 91.4
9 m. Thiss model had a power output of
2.5 MW at a wind speed Gordon, Anddrews et al., 1983).
d of 12.3 m/s and used a ssynchronouss generator (G
Three of thee major mod difications fro
om the MOD D-1 design were
w (Ramler and Donovaan, 1979):
o Thee power conntrol. Instead
d of the heavvy and comp
plex full-span pitch conttrol of the blades,
b
pow o the wind turbine was handled thrrough the pitch of just thhe outer 30 % the
wer control of
blad
de span;
o Thee tower desiggn philosoph hy. The desiggned tower was w considerrably more fflexible than those
used d in previous projects. It consisted inn a 61 m high h steel tower with a cylinddrical shell design.
d
In tthis design, the natural frrequency of tthe tower is lower
l than th
he frequencyy of rotation of the
bladdes. For this reason this design
d is calleed “soft-soft”” (see 3.3.4);
o Thee load reducttion. The rottor was desiggned to allow p to 5o in annd out of the plane
w teeter of up
of th
he rotor.
This design
n approach permitted
p to
o save weighht and, conseequently, red
duce costs. C
Comparing to
t the
MOD-1 dessign (which was smallerr), the MOD D-2 was arou und 10 % liighter (Ramller and Don
novan,
1979).
Three MOD D-2 machinees were built and tested in n a cluster in
n 1980 in Goodnoe Hills,, Washington n. The
disposition of the thrree wind tu urbines was defined in n order to study possiible aerodyn namic
interferencees between laarge turbines in downwiind (wake efffects) (Speraa, 1994). Tw wo other macchines
were also bbuilt for utiliity companiees. The MO D-2 machin nes suffered some techniical problem ms and
economic sttudies showeed that the costs were stilill not compeetitive. Howeever, this prooject proved that a
group of larrge wind turbbines could operate
o in a ttotally autom
matic and unaattended moode (Spera, 19 994).
36
Chapter 2
The tthird generaation of horizzontal-axis w wind turbinees from the NASA/DOE E program inncluded fourr
projeects: MOD-33, MOD-4, MOD-5A M annd MOD-5B B (although the MOD- 3 and MOD D-4 projectss
stoppped in an inittial phase) (SSpera, 1994).. The main goal
g of the MOD-5 prograam, started in i 1980, was,,
once again, to devvelop cost coompetitive, laarge wind turrbines (Baldw
win and Kennnard, 1985).
The MMOD-5A design, develop ped by Geneeral Electric Company
C forr NASA/DO OE, had a 122
2 m diameterr
rotorr with two bllades and thee hub was pllaced 76 m above
a the ground (Baldw win and Kennnard, 1985)..
The p power outpuut of this maachine was 7..2 MW with a rotor rotaation of 13 oor 17 rpm, deepending on n
wind conditions (Johnson, 20 001). Howevver, this macchine was no ot built due to low persp
pectives of a
comm mercial succeess from General Electriic which withdrew from the project.. Neverthelesss, this wind d
turbinne would rep
present, for nowadays
n staate-of-art, on
ne of the mosst powerful pprojects.
The M MOD-5B, developed
d byy Boeing forr NASA/DO OE, was builtt in 1987 (SSpera, 1995). This wind d
turbin
ne followed thet design co oncept of thee MOD-2 maachine: a two o-bladed, upw wind, teetereed rotor withh
a parttial span pitcch control off the blades (SSpera, 1995). The turbinee was supporrted by a cyliindrical steell
shell and had a rated
r power of 3.2 MW W and a roto or diameter of 97.5 m (FFigure 2.14). The majorr
innovvation of thee model was the t ability off its turbine to
t operate su
uccessfully att variable speeed (from 133
to 17.3 rpm).
Figure 2.13 – MOD-2

M wind tu
urbines (Wikipeedia, 2009) gure 2.14 – MO
Fig OD-5B (Wikipedia, 2009)
Alonggside with thhe NASA/DO OE, the Depaartment of th he Interior of

o the USA (DDOI) was alsso interested
d
in wind energy to o operate witth its hydroeelectric turbin
nes (Spera, 1995). For thiis reason, the DOI asked
d
NASA A for projecct management and techhnical suppo ort and a jo oint-venture of two com mpanies (onee
Amerrican and otther Swedish h) was selectted to design n, fabricate and
a install twwo megawattt scale wind d
turbinnes (designaated as “systeem verificatioons units” - SVUs). In 1982, the WT TS-4 wind tu urbine began
n
to op
perate. This machine
m wass a horizontaal axis wind turbine
t with a 79.2 m diaameter, dow wnwind rotorr
(Figu
ure 2.15). It had
h two fiberrglass blades mounted on n a teetered hub
h with full--blade pitch control. Thee
steel shell tower followed
f a “ssoft-soft” dessign with 80 m of height. This wind tturbine was, at that time,,
the m
most powerfu ul ever madee, with a pow wer output of
o 4.0 MW. The second SVU installeed was a 2.55
MW MOD-2.
37
7
W Turbines
F
Figure 2.15 – WTS-4
W wind turb
bine (Wikipediaa, 2009; Schwerrin, 2010)
During the 1970s, theree was also so ome relevantt research in wind powerr outside thee USA. In Caanada,
the research h efforts weere towards the verticall axis wind turbines (V VAWT), speecifically witth the
Darrieus deesign. Amongg others (sm mall) turbiness, the Nation
nal Research Council (NR RC) developeed the
Éole VAWT T. The installlation of thiss turbine wass complete in uebec, Canadda (Spera, 1994). It
n 1987 in Qu
was the firstt megawatt-cclass Darrieuus turbine, wi
with a 64 m diiameter and with an origginal power output
o
of 4.0 MW, operating att variable speeed (Figure 22.16). Howevver, to ensuree the longeviity of the machine,
the rated poower was lowwered to 2.5 MW.M The exxperience gained and the results achieeved were no ot very
encouragingg and the Caanadian proggram was teerminated (H Hau, 2006). Some
S researcch in the Daarrieus
concept field was also peerformed in the USA in tthe Sandia National
N Labo
oratories in A
Albuquerque.
a) b)
Figure 2.16
6 – Éole VAWT: a) general view
w (Wikipedia, 2005);
2 b) illustra
ation (Spera, 19994)
38
Chapter 2
In Euurope, some Governmen nts from Norrdic and Cen ntral Europeaan countries also develop ped research
h
progrrams during this period. In Denmarkk, an ambitio ous programm started in 11974. The main goal wass
“thatt it should bee possible to generate
g 10%
% of the Danish power requirement frrom wind energy withoutt
creatiing particulaar problems in the publiic power grid” (Hau, 20 006). For thiis purpose, the
t 200 kW W
Gedser wind turb bine was refufurbished forr a joint stud
dy with NAS SA. Furthermmore, two 63 30 kW wind d
turbinnes with 40 m diameter were builtt in Nibe, which w had th
heir first rotaation in 197
79 and 19800
(Interrnational En nergy Agencyy (IEA), 19884). These windw turbines had a threee-bladed, upwind rotorr
with steel blade spars
s and bo oth had a cooncrete toweer (Spera, 1994) (Figure 2.17). Each of the wind d
turbinnes had a diffferent poweer control meechanism: on ne had a tip--controlled rrotor while th
he other had
d
a full-span pitch control.
c Thesse wind turbiines operatedd for 15 yearrs.
Previiously in 1975, a 2.0 MW wind turbin
ne was built in
n Tvind, Den
nmark (Hau,, 2006) (Figu
ure 2.18).
Figure 2.17
2 – Nibe A wind
w turbine Fig
gure 2.18 – Tvinnd wind turbine
e
Anotther Nordic country

c with
h important developmen nts was Swed den. The Nattional Swedissh Board forr
Energgy Source Deevelopment wasw foundedd in 1975 and d two large wind
w turbiness were built (Hau, 2006)..
The 22.0 MW, 75 m diameter WTS-75 Naasudden wind d turbine witth two bladees on a rigid hub upwind d
of the cylindricall concrete toower had itss first rotation in 1983 (Figure 2.199) (International Energyy
Agen ncy (IEA), 19984). This tu
urbine had aan unusual feature:
fe a carrriage assembbly mounted d on verticall
rails oon the side of
o the towerr, for raising or loweringg major comp ponents. Thiis feature eliiminated thee
need for a large crane
c (Spera, 1994). The ssecond largee wind turbin ne erected inn Sweden wass the WTS-33
Maglarp wind turrbine in 1982 (Internatioonal Energy Agency (IEA A), 1984). Thhis machine was built byy
the saame consortiium that buillt the WTS-44 for the NASA/DOI program. The W WTS-3 had a rated powerr
of 3.00 MW with a 78 m diam meter, two-blaaded, downw wind rotor (F Figure 2.20).. It also had a “soft-soft””
towerr with a height of 80 m (Internationaal Energy Agency (IEA), 1984).
39
9
W Turbines
Fig
gure 2.19 – Nassudden wind tu
urbine Figure 2.20 – Maglarp windd turbine
The Germaan state su ubsidized program on the develop pment of wind w energy started in 1974
(Internationnal Energy Agency
A (IEAA), 1984). TThe most em mblematic prroject was thhe Growian wind
turbine, whhich was asseembled in 19 982 (Thiele, 1984). This wind
w turbinee was, at thaat time, the largest
l
Figure 2.21). It had a 100 m diameter , two-bladed
ever built (F d, downwind rotor whichh was supportted by
a 100 m talll steel tower stabilized
s byy several guy cables (Spera, 1994). Thee carbon-filaament blades had a
full-span pittch control. The
T Growian n machine wwas also the fiirst wind turbine to attemmpt variable--speed
operation aalthough it was
w not well succeed. Addditionally, the Growian n wind turbiine suffered other
problems, iincluding faatigue in major compo nents which h conducted d to its disaassemble in 1987
(Internationnal Energy Agency
A (IEA), 1989). De spite of the low success of the Grow wian project, some
smaller macchines operaated satisfacttorily (such as the 370 kW k Monopteros with a one-bladed, 48 m
diameter rootor (Figure 2.22)
2 (Spera, 1994).
Fiigure 2.21 – Growian wind turbine Figure 2.22 – Monopteros

M w
wind turbine
40
Chapter 2
The large wind turbines developed during the 1970s and 1980s failed, in general, one of their main
purposes: to produce energy at a competitive price. They were not as reliable as expected, with short
life periods due to several failures. In (Johnson, 2001), it is referred the fact that smaller turbines (in
the 100 kW range) could be built at lower costs and with better performance than the larges turbines as
one of the reasons to the end of the American program. In fact, the seek for larger and powerful wind
turbines underestimated the difficulty of building reliable machines of this size. The Growian wind
turbine is an example in which the size of the turbine is too large for the time in which it was built.
On the other hand, for the first time in the history of the wind energy, there was a consistent
experimental and theoretical knowledge of wind turbines. This knowledge was fundamental for the
beginning of the commercial market of wind turbines as it is known today.
In hindsight, it is possible to evidence three distinct groups of approaches on the structural and control
level over the evolutionary history of wind turbines (Thresher and Dodge, 1998):
o Turbines designed to withstand high wind loads. These models were considerably robust, with
low rated power and were optimized for reliability. An example of this philosophy was the
Gedser wind turbine;
o Turbines designed to be compliant and shed loads. These models were optimized for
performance. Their design had an important concern with the reduction of weight and with
the optimization of the aerodynamic profile of the blades. An example of this design concept
was the W-34 wind turbine;
o Turbines designed to manage loads mechanically and/ or electrically. These turbines were
optimized for control, with several mechanical and electrical innovations (such as hinged
blade connections and variable speed generators). An example of this concept was the Smith-
Putnam turbine.
The first approach was indubitably the most durable and reliable. However, the increase of knowledge
in the industry has led to less conservative designs, with the aim of weight reduction (to reduce cost)
and control (to optimize the production). Thus, a joint of the second and third approaches is currently
the main driver of the design concepts.
An illustration of the principal projects held during this period of development is shown in Figure
2.23. The evolution of the dimension of the wind turbines is notorious.
41
Figure 2.23 – Main projects developed during the 1970s and 1980s (adapted from (Spera, 1995; Gasch and Twele, 2012))
Height [m]
150
(Wind Direction)
125
100
100
79.2
75
75 78
97.5
54 91.4
50
61
38
25
Name MOD-0 Tvind MOD-1 MOD-2 WTS-4 WTS-3 Growian WTS-75 MOD-5B
Year 1975 1975 1979 1980 1982 1982 1982 1983 1987
Rated Power
0.20 2.00 2.00 2.50 4.00 3.00 3.00 2.00 3.20
[MW]
42
Chapter 2
2.1.1.4 Current situation and future trends

Nowadays, wind turbines present a mature state of technology. In fact, wind is considered the
renewable source of energy with the widest and most successful deployment over the last two decades
(Arántegui, Corsatea et al., 2013). The current state-of-art of onshore wind turbine is composed of:
o Horizontal axis turbines;
o Steel, concrete or hybrid shell towers;
o Upwind rotor, with three blades;
o Rotor with variable speed (this feature enhances the efficiency of the system);
o Yaw system for rotor orientation according to wind direction;
o Active pitch control of blades (for high wind speed regulation);
o Synchronous or induction generator;
o Utilization of a gearbox (although some models with direct-driven generators are available).
In order to better understand the state of technical evolution of the wind energy industry, a survey
about the currently available wind turbine models was conducted. In this research, the onshore models
presented in the portfolio of the most relevant manufacturers were considered1.
Figure 2.24 presents the main results obtained, namely the relationship between rotor diameter and
rated power for available wind turbines. As can be seen, the majority of the models available have a
rated power below 4.0 MW, although there are already some models with higher power outputs
(achieving a maximum value of 7.58 MW). One interesting aspect is the apparent stagnation of the
rotor size of the largest wind turbines. However, due to the limited number of models with power
outputs higher than 4.0 MW currently available, this is a trend that will need further investigation.
Another relevant characteristic is the large variation of rotors size for the same rated power. This is
partly explained by the development of models designed for sites characterized by low-wind speed
conditions. These models usually present high towers (around 120 m to 146 m of height) and larger
rotor diameters.
The data collected is also used to illustrate the “position” of the two onshore models studied in this
work, the 2.0 MW Senvion MM82 and the Izar Bonus 1.3MW/62 (indicated with a red circle in Figure
2.24). It is visible that, while the latter is located in the low range of power capacity, the Senvion model
is more in line with the common available models between 2.0 MW and 3.0 MW.
1
In this study, the manufacturers considered were: Alstom, AMSC, Darwin, Enercon, Gamesa, GE Energy, GoldWind, MY
Wind, Nordex, Samsung, Senvion, Siemens, Sinovel, Suzlon and Vestas.
43
Onshore Wind Turbine Models

140
120
Rotor Diameter [m] 100
80
Senvion MM82
60
Izar Bonus 1.3MW/62

40
0 1 2 3 4 5 6 7 8
Rated power [MW]
Figure 2.24 – Rated power and rotor diameter of onshore wind turbines
2.1.2 THE OFFSHORE DEVELOPMENT

The transfer of wind energy technology from land to the sea was a natural development. The
implementation of wind turbines at sea presents some important advantages (Henderson, Morgan et
al., 2003):
o Wide continuous areas available for large wind farm projects;
o Higher wind speed, which generally increases with the distance from the shore;
o Less turbulence, which allows the turbines to harvest the available energy more effectively,
with reduced fatigue loads on the turbine;
o Lower wind shear (i.e. the boundary layer of slower moving wind close to the surface is
thinner), thus allowing the use of shorter towers;
o Possibility of faster rotational speeds than on land, due to lower noise restrictions.
On the other hand, some disadvantages are naturally present in offshore turbines:
o Being a more recent technology, the design of wind turbines is still in an initial phase.
Consequently, the costs are still quite high when compared to onshore technology;
o The structures are subjected to a higher variety of dynamic loads;
o The maintenance is more difficult and costly.
A description of the offshore evolution is briefly presented in the next sections.
2.1.2.1 The first offshore projects

The first turbines were installed at sea in the early 1990s. At that time the developed machines already
presented some reliability, with power outputs in the 500 kW range.
The countries bordering the Baltic and North Sea were the first to install offshore wind turbines. Both
the Baltic and the North Sea present very good conditions for the implementation of offshore
technology: they both have shallow waters and good wind speeds.
44
Chapter 2
The ffirst offshoree turbine was a single dem

monstration erected
e in No ogersund, Swweden, in 199 91. Howeverr
it wass abandoned d due to a fire in 1998 (H
Henderson, Morgan
M et al., 2003). The first offshorre wind farm
m
was tthe Vindeby wind farm, commissione
c ed in 1991 at
a the Baltic Sea
S (Denmarrk) (Hendersson, Morgan n
et al.,, 2003). This wind farm consisted
c in 111 turbines with
w a rated-p power of 4500 kW each, im mplementedd
in verry low water depth.
Durinng the follow
wing years, 4 more wind farms were built in Sweeden, the Nettherlands an
nd Denmark,,
with w
wind turbinees with indivvidual power output of arround 500 kW
W.
In 20000, a new dimension started
s for ooffshore win nd energy with
w the impl
plementation of the firstt
megaawatt class wind
w turbiness. During thhis year, threee new offshore wind farrms were co
ommissioned d
(inclu
uding the firrst in the UKK) (Hendersson, Morgan et al., 2003)). Among thhese, the Mid ddelgrundenn
wind farm, in Deenmark, deseerves greaterr prominence. It was com mposed of 200 wind turbines with an
n
indivvidual power output of 2 MWM locatedd, once again,, in very shallow waters.
Durinng the next years, the co onstruction oof new multti-megawatt sized wind ffarms contin nued. Speciall
relevaance must beb given to thhe Horns Reev (Figure 2..25) and Nyssted wind faarms (both in n Denmark))
comm missioned in 2002 and 20 003, respectivvely. These wind
w farms represented
r a big step in the offshoree
indusstry, with ratted powers off around 1600 MW.
Figure 2.255 – Horns Rev offshore

o wind farm
f
2.1.2..2 Currentt situation an

nd future tren
nds
Nowaadays, the offfshore wind d turbine secttor is alreadyy an implem
mented indusstry. The devvelopment off
offshoore wind turrbines is, in itts essence, an
n adaptation
n and evolutio
on of the lannd-based mod
dels. Amongg
otherrs, some tech
hnical modifications conccerning offsh hore-oriented
d design incluude:
o Strengtheening of the tower
t to han
ndle the addittional loads (Musial
( and Ram, 2010);
o Higher po
ower outputss and tips speeeds;
o Higher co
omponent reedundancy (H
Henderson, Morgan
M et all., 2003);
o Improved
d protection against harshh marine envvironment (iincluding corrrosion);
45
5
o Personnel access platforms;

o Special design of the foundation and the transition piece (component between the foundation
and the tower – see Chapter 3).
Some other technical features are often also included (Musial and Ram, 2010):
o Monitoring systems for better diagnostic information of the structure and turbine;
o Onboard service cranes;
o Bearing-lubrication and oil-temperature regulation systems;
o Sometimes a heliport is included at the top the structure.
The current state-of-art is similar to the onshore model in terms of mechanical solutions, although the
dimensions of rotor diameter and the rated power tend to be higher. This is justified by the absence of
some limitations present on the transportation on land (such as bridges height, road bends and short
transition curves in vertical alignments). Beside these limitations, the installation of wind turbines with
higher rated power allows to reduce the cost per MW, which is an important aspect to counterbalance
the high costs associated with the installation of offshore turbines. Another feature of offshore models
is the lower hub height due to the lower wind shear. In most of the times, tower height is defined for
the specific site of installation. However, the smaller experience in offshore installations still leads to a
lower standardization of the models compared with the onshore variant. For this reason, there is still a
wide variety of offshore wind turbine foundations (see Chapter 3).
The same market study presented in section 2.1.1.4 for onshore wind turbines was also developed for
the offshore models. The offshore models were characterized once again by their rotor diameter and
rated power. The main results are presented in Figure 2.26 (including the Vestas offshore model
studied in this work).
As can be observed, the current offshore models present larger rotor diameters and higher power
output than the onshore models (Figure 2.24), with several models in the 4.0 MW – 6.5 MW range.
Nevertheless, there are already some models presenting capacities higher than 7.0 MW.
Offshore Wind Turbine Models
200
175
Rotor Diameter [m]
150
125
100
75
Vestas V90-3.0MW
50
2 3 4 5 6 7 8 9 10 11
Rated power [MW]
Figure 2.26 – Rated power and rotor diameter of offshore wind turbines
Another important aspect in the offshore development is related to the distance to the shore and water
depth at which the wind farms are installed. Figure 2.27 presents the average distance to shore and
water depth of European wind farms, alongside with their dimension (defined by the size of the
46
Chapter 2
circles). Over the years, the distance to shore has increased, as well as the water depth of the sites of the
projects. The average distance to shore was 32.9 km and the average water depth of wind farms was
22.4 m, at the end of 2014 (European Wind Energy Association (EWEA), 2015a). Looking at Figure
2.27, it is possible to assume that the trend of installation of wind farms away from the shore will
continue.
120
Online
100
Consented
80
Distance to shore [km]
Under
construction
60
40
20
0 10 20 30 40 50 60
Water depth [m]
Figure 2.27 – Average water depth and distance to shore of online, under construction and consented European wind
farms, at the end of 2014 (adapted from (European Wind Energy Association (EWEA), 2015a))
The main design drivers for the future of offshore wind turbines will be the cost reduction and the
implementation of turbines at higher water depths. Furthermore, the increase of the turbine power
output and of their reliability will continue, alongside with a better understanding of the influence of
the environment on the structure of the wind turbine.
Currently, the levelized cost of offshore energy is slightly higher than twice the onshore cost (World
Energy Council (WEC), 2013; Bloomberg New Energy Finance, 2015; U.S Energy Information
Administration, 2015). In order to become viable, the capital and operational expenditure of offshore
investments need to be reduced. A part of the reduction of the initial investment is achieved with the
decrease of the structure weight (Butterfield, Musial et al., 2007). With light-weight structures, the
loads tend to decrease through all the structure, including in the foundation. Some new and more
expensive materials might be used for this purpose, such as, in special cases, carbon in the blades
(Butterfield, Musial et al., 2007). However, less material (and then less stiffness and mass) leads to a
more flexible structure which, considering the dynamic loads, requires a thorough understanding of
the dynamic behaviour of the structural system.
Another possibility for weight reduction is the adoption of two-bladed rotors (Butterfield, Musial et al.,
2005). This solution was almost abandoned for the onshore turbines due to lower yield and for
aesthetic reasons, as well as some technical problems that were not solved in the past - like the one
described for the Jacobs wind turbine (Butterfield, Musial et al., 2007). However, the reduction of one
blade (with the consequent weight decrease) could be balanced by the increase in blade tip speed of the
offshore turbines. In addition, the higher rotor speeds lead to a lower input torque and lower gear
ratios, which permit to reduce some mechanical components located in the nacelle. The increase in
rotor tip speed could also lead to the utilization of direct-drive generators (drivetrains without
gearboxes) (Musial and Ram, 2010).
47
In the future, another major change may take place for the offshore turbines. The continuous
increasing of rotor diameters may lead to a definition of the blade design by a deflection limit in
upwind rotors, instead of a load limit criterion (Butterfield, Musial et al., 2007). This is due to the
increased levels of strength and flexibility achieved in blades. For this reason, downwind rotors may be
an option for future offshore wind turbines because, contrary to the upwind solution, the blades
deflect away from the tower. The downwind solution was abandoned in the past due to vibrations and
noise problems (Butterfield, Musial et al., 2007) with the blades passing through the tower shadow.
Nevertheless, for large distances from the shore, noise disturbances are not a problem.
Some other perspectives for future developments of wind turbines are presented in (European
Commission, 2011). In the report of the UpWind program, some considerations are made for the
design of a giant 20 MW wind turbine. Although the program was not exclusively for the offshore
market, the dimensions of the turbines under analysis make them suitable for the implementation at
offshore locations. Among other considerations, some important aspects are referred for the feasibility
of turbines with this dimension:
o Reduction of the magnitude of fatigue loads. For this, advanced rotor control strategies are
required (the concept of “smart turbine” is introduced);
o Implementation of control systems for detection and evaluation of upcoming wind loads
(gusts or vortex) with the aim of load alleviation (such as LIDAR systems mounted at the
nacelle);
o Individual pitch control of each blade;
o Application of materials with lower mass to strength ratio;
o Introduction of monitoring technologies to assess the structural integrity of the system;
o Due to transportation issues, the blades might have to be splitted into two pieces since the
rotor of the studied 20 MW wind turbine has a diameter of 252 m. For this reason, this joint
might be an important focus of attention;
o The need for optimization of the offshore substructure (foundation and transition piece).
The offshore substructure is another important topic for the future of offshore wind turbines.
Nowadays there is a large variety of commercially implemented, tested and concept designs (see
Chapter 3). The future trend, as stated earlier, is to increase the distance from the shore. In this
situation, due to a larger water depth, floating structures seem to be the most suitable option. Until
today, a few number of experimental projects have been deployed. For the next years, many floating
wind turbine projects are schedule to start in several countries (Main(e) International Consulting LLC,
2012).
48
Chapter 2
2.2 BRIEF MARKET ANALYSIS

2.2.1 THE START OF THE DANISH AND AMERICAN MARKET
The wind energy market gained relevance in the second half of the 20th century (Hau, 2006). In
Denmark in the early 1980s, some small manufacturers of agricultural machines entered in the market
with simple wind turbines with power outputs around 30 - 75 kW (Hau, 2006; Gasch and Twele,
2012). These machines followed the Juul’s concept with a three-bladed, stall-regulated, upwind rotor
working at a fixed speed with an induction generator (Burton, Sharpe et al., 2001). This market was
initially sustained by an appropriate feed-in tariff set by the government (Gasch and Twele, 2012). A
consequence of the development of the Danish market was the creation of a certificate by the Wind
Turbine Test Station in Risø to attest the maturity and safety of the wind turbines.
On the other hand, the American market was created mainly due to politic measures. Consecutive
measures took by the Senate that culminated in the Public Utilities Regulatory Policy Act (PURPA) in
1978 (Hau, 2006), conducted to a favourable situation of subsidies for wind energy investments. The
state which provided the best incentives, and which also had good wind conditions, was California
(Manwell, McGowan et al., 2010). Thus, the first wind farms were built there between 1979 and 1980.
Due to economic viability, these wind farms were constituted by 100 or more turbines (Johnson,
2001). These small wind turbines, with power outputs up to 100 kW, were initially manufactured by
American companies. However, the initial operation of these wind turbines was characterized by
several difficulties due to the technological immaturity of the models (Manwell, McGowan et al.,
2010). Sometime later, Danish wind turbines entered in the American market and became more
successfully than native companies, mainly due to their larger experience.
After the initial problems, the number of wind farms increased very quickly around 1981, with an
installed capacity over 1500 MW (Johnson, 2001). However, in the years between 1986 and 1987, the
economic situation changed in California (Hau, 2006). The conjugation of the expiration of the tax
credits for the investors, alongside with the lower supply tariffs offered by the utilities, led to a
slowdown (and, sometime later, to the stagnation) of the growth of wind farms (Hau, 2006).
In the rest of Europe, the wind energy market started in the 1990s with a support scheme mainly based
in feed-in tariffs for wind power generation (Ackermann and Söder, 2002).
2.2.2 ONSHORE WIND TURBINE MARKET

The analysis of the evolution of the wind energy market and its current situation is a clear way to attest
the success of this industry in the last 15 years. In this section, a brief market analysis is made for the
entire market (onshore together with offshore), since the latter still has a minor importance in terms of
total installation numbers (around 2.4 % of the total installed capacity).
Analysing the cumulative installed capacity in the world (Figure 2.28), it is possible to observe the
consistent growth of the wind energy market, with a total installed capacity already exceeding the 350
GW. Qualitatively, it is also possible to observe that Europe and Asia are the leaders in the installed
capacity. In fact, Asia became the region in the world with the largest installed wind park during 2014.
The main driver of this situation is China, which has become the world leader in installed capacity
when it surpassed the USA during 2010 (Global Wind Energy Council (GWEC), 2011). The top 10
countries with the largest cumulative installed capacity are presented in Figure 2.29.
49
Cumulative Installed Capacity

400000
Pacific Region [1.2 %]
350000
Cummulative Installed Power [MW]
300000 North America [21.1 %]
250000
Latin America & Caribbean
[2.3 %]
200000
Europe [36.3 %]
150000
100000 Asia [38.4 %]
50000
Africa & Middle East [0.7 %]
0
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Figure 2.28 – Cumulative installed capacity (the share at the end of 2014 results is presented in brackets) (EWEA 2007,
2008, 2009, 2010, 2011b, 2012b, 2013c, 2014b, 2015b) (GWEC 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015)
Top 10 Cumulative Capacity (2014)
Brazil
Rest of the
France Italy 2% world China
2% 2% 16% 31%
Canada
3%
UK
3%
India USA
6% Germany 18%
Spain 11%
6%
Figure 2.29 – Top 10 Countries in cumulative installed capacity in the end of 2014 (European Wind Energy Association
(EWEA), 2015b; Global Wind Energy Council (GWEC), 2015)
On the other hand, European investments stagnated during the last years which can be justified by the
financial crisis installed in this continent (Figure 2.30). In the opposite side, China was responsible,
during 2013 and 2014, by more than 45 % of the new installations (Figure 2.31). These recent
investments led China to a position of market domination.
50
Chapter 2
Annual Installed Capacity

60000
Pacific Region [1.1 %]
NewInstalled Power [MW]
50000
North America [14.3 %]
40000
Latin America & Caribbean
[7.3 %]
30000
Europe [25.0 %]
20000
Asia [50.5 %]
10000
Africa & Middle East [1.8 %]
0
2006 2007 2008 2009 2010 2011 2012 2013 2014
Figure 2.30 – Annual installed capacity (the share at the end of 2014 results is presented in brackets) (EWEA 2007, 2008,
2009, 2010, 2011b, 2012b, 2013c, 2014b, 2015b) (GWEC 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015)
Top 10 New Installed Capacity (2014)
France
2% Turkey
2% Rest of the
Sweden world
2% 13%
UK China
Canada
3% 45%
4%
India
5%
Brazil
USA
5%
9% Germany
10%
Figure 2.31 – Top 10 Countries in new installed capacity in the 2014 (European Wind Energy Association (EWEA), 2015b;
Global Wind Energy Council (GWEC), 2015)
In Portugal, large investments were done in wind energy since the first wind farm was commissioned
at Madeira in 1986. Portugal is the 12th country in the world with largest installed power and the 8th in
Europe with a cumulative installed capacity of 4 914 MW (European Wind Energy Association
(EWEA), 2015b; Global Wind Energy Council (GWEC), 2015) (Figure 2.32).
However, the investments have been declining in recent years. This situation may be explained by the
financial crisis, by the minor incentives given by the Government and by some limits imposed for
penetration in the national electrical system.
51
Cumulative and Installed Capacity in Portugal

6.000
5.000
Cummulative
installed Capacity
Power output [MW]
4.000
3.000
Annual installed
capacity
2.000
1.000
?
0
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Figure 2.32 – Annual cumulative and installed capacity in Portugal (since 2005) (EWEA 2007, 2008, 2009, 2010, 2011b,
2012b, 2013c, 2014b, 2015b)
Figure 2.33 shows the geographical distribution of wind farms in the continental territory of Portugal.
As can be observed, the major part of wind farms is located in the north and centre of the country,
mainly in interior highlands.
Wind Farms
Total Power Output [MW]
[
[0.5 – 1.9]
[2.0 – 9.9]
[10.0 – 24.9]
[25.0 – 49.9]
>50
Figure 2.33 – Installed wind farms in Portugal at the end of 2014 (Institute of Mechanical Engineering and Industrial
Management (INEGI) and Portuguese Renewable Energy Association (APREN), 2015)
52
Chapter 2
2.2.3 OFFSHORE WIND TURBINE MARKET

Compared to the onshore market, offshore wind power is still in the beginning, with slightly less than
9 GW installed, corresponding to 2.4 % of the total installed capacity worldwide (Figure 2.34). At the
end of 2014, there were only 17 countries in the world with offshore wind turbines installed. From
these, the major part is from Europe. The implementation in this continent represents more than 90 %
of the world installed capacity.
Annual Cumulative Power US [0.00 %]
10.000 Korea [0.06 %]
Portugal [0.06 %]
9.000
Annual Cumulative Power [MW]
Norway [0.02 %]
8.000 Spain [0.06 %]
7.000 Portugal [0.06 %]
Norway [0.02 %]
6.000 Ireland [0.29 %]
5.000 Japan [0.57 %]
Finland [0.57 %]
4.000
Sweden [2.42 %]
3.000 Belgium [8.13 %]
2.000 Germany [11.98 %]
China [7.51 %]
1.000
Netherlands [2.82 %]
0 Denmark [14.51 %]
2009 2010 2011 2012 2013 2014 UK [51.32 %]
Figure 2.34 – Annual cumulative capacity of offshore wind turbines .The share values in brackets are referred to the end
of 2014 (EWEA 2011a, 2012a, 2013b, 2014a, 2015a) (GWEC 2012, 2013, 2014, 2015)
In today’s market, the UK is clearly dominant with more than 50 % of the entire global capacity. The
UK has been doing a large investment in the offshore market in the last years, as can be seen in Figure
2.35. Portugal has a small portion of the entire market with just one 2.0 MW experimental floating
offshore wind turbine.
Anual Installed Power US [-]
2000 Korea [-]
Portugal [-]
1750 Norway [-]
Annual Installed Power [MW]
Spain [-]
1500 Portugal [-]
Norway [-]
1250
Ireland [-]
1000 Japan [-]
Finland [0.01 %]
750 Sweden [-]
Belgium [8.76 %]
500 Germany [32.84 %]
China [7.89 %]
250
Netherlands [-]
0 Denmark [-]
2010 2011 2012 2013 2014 UK [50.51 %]
Figure 2.35 – Annual installed capacity of offshore wind turbines (data from 2013 is only referred to European countries).
The share values in brackets are referred to the end of 2012 (EWEA 2011a, 2012a, 2013b, 2014a, 2015a) (GWEC 2012,
2013, 2014, 2015)
The dominance of the offshore market by the Northern countries of Europe is a clear evidence. For
this fact, the water depth conditions present in these countries should be considered as an important
driver. Analysing Table 2.1, it is evident the large portion of shallow waters present in the north of
Europe, in opposition to the situation at the south regions, where the water depth increases rapidly
53
with the distance from the coast. In addition, it is also possible to note the apparent good conditions
for the USA to install offshore wind farms. However, in the last years, investments in this area were not
considered a priority by this country.
Table 2.1 – Percentage of water depths in different regions up to 100 km offshore (Henderson, 2003)
Water Depth
Region
< 25 [m] 25 – 50 [m] 50 – 100 [m] 100 – 300 [m]
North Europe 21 26 32 20
South Europe 16 11 23 49
Japan 22 9 18 51
USA 50 26 13 11
54
Chapter 2
2.3 FINAL CONSIDERATIONS

This chapter described the evolution of wind turbines, since the first attempts to modern utility-scale
models. In the same manner, the emergence and development of offshore turbines, from land-based
models to especially designed models to be installed offshore, was covered. Apart from the efficiency in
the exploitation of energy from the wind, two important aspects defined the success of some models
over others: reliability and cost effectiveness.
The second part of the chapter briefly introduced the wind energy market. It was attested the
importance of this sector, with a consistent growth sustained by both consolidated and emergent
markets. In addition, the potential growth of the offshore market, still in an early phase of deployment,
was referred.
This chapter is especially important to frame the scope of the present work. A monitoring system, able
to detect structural damages at an early stage and perform an estimation of the fatigue condition of the
support structure, represents an important tool to remotely control the condition of the wind turbine
and, thus, its reliability. At the same time, this continuous assessing provides helpful information to
timely prepare maintenance actions, as well as possible retrofitting needs, allowing to optimize the
reduction of OPEX (cost over the period of operation – see Chapter 3).
55
56
Chapter 3
3
BASICS OF WIND TURBINES
3.1 GENERAL LAYOUT

3.1.1 WIND TURBINE TYPOLOGY
Historically, wind turbines can be defined according to several design configurations. Concerning
conceptual design aspects, a major distinction is made according to the position of the axis of rotation
of the wind rotor: the axis can be positioned vertically or horizontally.
Vertical-axis wind turbines, such as the Savonius and Darrieus rotor (Figure 2.4 and 2.5), are rarely
used in modern large wind turbines. Although this kind of turbines presents some advantages, some
negative aspects such as the low tip-speed ratio, the inability to self-start and the difficulty to control
the rotor speed (Hau, 2006) led to low interest in their development during the 1970s and 1980s.
The horizontal-axis wind turbine is the most common solution. This type of wind turbines can be
subdivided into two main groups considering the configuration of the rotor with respect to the wind
flow direction: upwind and downwind rotors (being the latter almost not used nowadays).
In this work, the main focus is oriented towards the horizontal-axis upwind turbines, the typical
nowadays configuration. Notwithstanding, some considerations are made about other solutions. The
layout of a nowadays common onshore and offshore wind turbine is shown in Figure 3.1. The
represented offshore example refers to a monopile foundation solution (see section 3.1.1.5). In Figure
3.2 the terminology for the translational and rotational degrees of freedom of a wind turbine is
introduced.
A brief description of the main components of nowadays onshore and offshore wind turbines is
presented in the following sections.
57
Basics of Wind Turbines
Blade Blade
Wind Direction
Hub
Nacelle Nacelle
Rotor
Tower
Tower
Boat
Landing
Transition Support
(Water Level)
Piece Structure
J-tube
Scour Foundation
Protection Foundation
a) b)
Figure 3.1 – Main wind turbine components of horizontal-axis upwind wind turbines: a) offshore; b) onshore
Flapwise Edgewise
Side-Side Fore-Aft
Yaw Roll Twist

(torsion)
Tilt
Pitch
Figure 3.2 – Terminology of the principal axes of translation and rotation of a wind turbine structure
58
Chapter 3
3.1.1.1 Rotor
The rotor is the key component of a wind turbine, converting the kinetic energy of the wind into
mechanical energy. It is composed by the blades which are connected to the hub.
The design of a blade is one of the most complex tasks in the development of a wind turbine. Blades
have to be designed in order to optimize the energy extraction from the wind flow. However, at the
same time, the progressive increase in the power output of the turbines demands increasingly larger
blades (with models of 83.5 m already being produced (Quilter, 2013)).
For these reasons, blades present complex cross sections (named airfoils) which vary along their length
(Figure 3.3). In addition, blades have to be sufficiently light, stiff and strong, at the same time, to not
contribute with enormous gravitational loads (for the blade itself and for the tower), to not deflect too
much (which would drive to loss of efficiency and, even more important, could lead to a collision with
the tower) and to resist to a very high number of cycles during their lifetime (of around 107 to 108
(Hau, 2006)). In order to accomplish all these prerequisites, composite materials are used (usually
epoxy resins with glass fibres, although carbon fibres may be also used in the blade composition).
In Figure 3.4 two common solutions for the structure of a blade are presented.
Figure 3.3 – Variation of the airfoil along blade length
Considering the rotor position, wind turbines may have a wide variety of configurations. As previously
referred, the rotor can be placed in front (upwind) or behind (downwind) the tower. This latter
solution is rarely used today because of noise, vibrations and efficiency problems. Due to the presence
of the tower upstream of the rotor, the flow is disturbed before it reaches the blades and, every time a
blades pass behind the tower, the momentary absence of aerodynamic force leads to additional
vibrations (shadow effects). In addition, large downwind rotors also need a yaw system (which is not
necessary in small downwind turbines). Several examples of downwind wind turbines were presented
in Chapter 2.
The number of blades is another aspect that contributes for the variety of rotor configurations.
Historically, wind turbines have one, two or three blades. The difference between two or three-bladed
rotors implies, besides differences in tip-speed and efficiency, different vibration problems. Figure 3.5
presents an example of the yaw moment created by an asymmetrical wind flow on wind turbines with
different rotor solutions (Hau, 2006). As can be seen, one and two-bladed rotor wind turbines suffer
from a noteworthy time-varying yaw moment while this excitation is almost undetected with rotors
with three or more blades. This effect is due to the variation of the rotor moment of inertia with the
blades revolution. Whereas in one or two-bladed rotors, this asymmetry in the moment of inertia is
59
relevant and leads to a pulsating load profile during a revolution (Hau, 2006), three or more blades
confer an almost symmetrical moment of inertia to the rotor.
Shear web
Trailing Edge
Leading Edge
a)
Box girder
Trailing Edge
Leading Edge
b)
Figure 3.4 – Two common design concepts for blade cross sections with different load-carrying solutions (adapted from
(Tong, 2010)): a) shear web; b) box girder
3000
2000
1 blade
1000
Yaw Moment [kNm]
2 blade
0 3 blade
4 blade
-1000
-2000
-3000
0 90 270 180 360
Angle of Rotation [º]
Figure 3.5 – Yaw moment caused by asymmetrical wind flow on wind turbines with different number of blades (adapted
from (Hau, 2006))
With the aim of alleviating these loads created by two-bladed rotors, some solutions were developed.
These solutions focus on the flexibility of the connections between blades and the hub (Figure 3.6).
The simplest solution is the rigid hub connection. This solution connects the blades to the hub at fixed
positions and, consequently, the loads are directly transmitted to the drive train by the hub. This is the
common solution for three-bladed rotors and for some rotors with two blades (such as the 1983 WTS-
75 Nasudden, see Figure 2.19)).
A more flexible solution is the teetering hub. This hub connection allows the rotor to teeter about the
rotor shaft due to a hinge located at the hub. This solution permits to balance out some asymmetrical
loads. This solution is intended for two-bladed rotors (and was used in the 1982 Growian wind
turbine, see Figure 2.21).
60
Chapter 3
Another flexible solution is the flapping hinge rotor. In this solution, a hinge is installed in each blade
allowing the individual rotation of the blades (the 1941 Smith-Putnam wind turbine used this solution,
see Figure 2.7)).
Fixed Hub Teetering Hub Flapping Hinge Rotor
Figure 3.6 – Hub solutions
3.1.1.2 Nacelle
The nacelle is the component of a wind turbine which encloses the mechanical and some electrical
elements (Figure 3.7).
In the nacelle, the rotor shaft (or main shaft) receives the torque (mechanical energy) from the hub. In
nowadays most common turbines, this shaft is connected to a gear box. The gear box increases the
rotational speed from the rotor shaft to a more convenient speed to the generator. The connection
between the gearbox and the generator is made by the generator drive shaft. The generator produces
electric energy from the mechanical energy. There is also a mechanical brake with the aim of stopping
the turbine when necessary and a yaw system that mechanically orients the nacelle to the wind
direction. The components of the nacelle are supported by the bedplate.
Although the previous description refers to the most common nacelle layout, there are also some
different concepts, such as the absence of a gear box (direct-driven generators).
Bearings
Rotor Shaft Gearbox Suspension

Drive Train
Rotor Brake
Rotor Hub Gearbox
Generator
Blade Pitch Electrical Switch Boxes

Mechanism and control systems
Bedplate
Yaw System Rotor Hub
Generator Drive Shaft
a) b)
Figure 3.7 – a) Main components inside the nacelle. b) Detail of the drive train (adapted from (Hau, 2006))
61
3.1.1.3 Tower
Wind turbine tower elevates the rotor to the desire height. Since the wind flow presents higher speed
and lower turbulence at higher heights, the towers tend to be very tall. In addition, wind turbine
towers are also very slender which, alongside with the rotor motion, leads to complex dynamic
problems.
The more common towers are of four types: steel tubular, steel lattice, concrete tubular and hybrid
solutions. There are also some special designs, such as guyed supported towers. However, since these
are almost not used nowadays, they are not covered in this work.
Prestressing
tendons
Steel
Concrete
a) b) c) d)
Figure 3.8 – Various wind turbine tower types: a) steel tubular; b) lattice; c) concrete tubular; d) hybrid solution (adapted
from (Tricklebank, Halberstadt et al., 2007))
Steel Tubular Tower

The most currently used solution is the steel tubular tower. It is composed by steel sheets which are
rolled into a circular shape and then welded together to form a circular section with a length of around
20 m – 30 m. These sections are assembled on-site by bolted connections. Thus, no on-site welding
processes are needed.
The main advantages of this solution are its short on-site assembly and erection time. From a design
point of view, steel tubular towers also present some benefits, such as an uniform behaviour for all yaw
directions (due to the circular cross section) and a good control over the dynamic characteristics, since
it is a solution with a large accumulated experience.
On the other hand, steel tubular tower have shown some difficulties in following the growth in
dimensions for onshore wind turbines. With higher tower height, also a larger tower base diameter is
usually required. Since a diameter of around 4.0 m – 4.5 m is the common limit for road
transportation, alternative solutions need to be considered. Besides this, the increase in height also
62
Chapter 3
introduces some difficulties concerning the stiffness of the tower. For that reason, this solution is not
used in some models due to the required structural stiffness that a steel tubular tower cannot deliver.
Steel Lattice Tower

Steel lattice towers were the preferred solution of the first experimental turbines and of the first smaller
commercial turbines (see for example Figure 2.6 and 2.7) (Hau, 2006). This solution is constituted by
legs and bracings connected by bolted or welded joints.
Lattice towers can be a solution for high towers, since they can be transported disassembled. They
present a satisfactory stiffness and strength with regard to the material consumed. They are also lighter
than the tubular solutions and usually present a higher structural damping due to the numerous joints
of the structure.
In contrast, the assembling of the structure is very time-consuming and complex. Also the expenditure
with maintenance is considerably higher than for tubular sections due to the high number of joints. In
the past, this solution was considered as “ugly”, not being endorsed by populations. This solution is
not very common in nowadays installations.
Concrete Tubular Tower

Concrete solutions for wind turbine towers have a long tradition in some countries, such as Denmark
(Hau, 2006). It is generally used with precast concrete, alongside with pre or post-stressed tendons,
although concrete mixed on-site is also used.
Concrete towers present the main advantage of higher stiffness and damping than steel tubular. This is
the reason why some very large wind turbines use this solution. Also, if the concrete is mixed on-site,
the problem of transportation of large components does not arise.
Some disadvantages are also inherent to this solution. If the concrete is mixed on-site, the construction
process is time consuming and might be dependent on the weather conditions. Also, the post and pre-
stressing operations are complex and require special precautions. From a design perspective, the
fatigue analysis is difficult to handle, leading to conservative designs in many situations.
Hybrid Tower
A compromise solution is obtained with tubular hybrid towers. This solution is characterized by a
concrete tower for the lower part of the tower (usually mixed on-site), while the upper portion is made
of steel. Thus, it is possible to increase the stiffness of the structural system with a short period of
concrete curing and without transportation of large steel components. This is a solution mainly for
high wind turbine towers (100 m or higher).
63
3.1.1.4 Transition Piece

The transition piece only exists in offshore wind turbines. It connects the foundation to the tower. This
element is considered one of the most critical points of the structural integrity of offshore wind
turbines.
This element is grouted on the top of the foundation and, in this operation, any lack of verticality is
corrected (Figure 3.9). The transition piece usually attaches a structure for boat landing, platforms,
ladders and the j-tube (a metallic tube that guides the cables to the seabed).
Although the transition piece is utilized in several types of foundations for offshore wind turbines,
other offshore solutions may not need this element.
tower base flange
transition piece
(includ. access platform,
cable ducts, ladder,
grounted joint add-ons)
monopile
submarine cable
scour protection
Figure 3.9 – Grouted joint for the connection between the foundation (monopile) and the transition piece. In this
situation, the verticality is corrected by the connection (Gasch and Twele, 2012)
3.1.1.5 Foundation
The foundation is the component that best distinguishes onshore and offshore wind turbine
structures. While onshore turbine foundations follow traditional in-land civil engineering solutions,
offshore models have been progressively adopting solutions from oil and gas offshore industry.
However, both onshore and offshore foundations have the same function: to ensure stability and drive
the loads from the structure to the ground, which means, to prevent the structure from sinking due to
gravitational loads, from sliding due to horizontal loads and from fall over due to base overturning
moments. Besides this, it is important to bear in mind that the definition and implementation of the
designed foundation will influence the dynamic behaviour of the entire structural.
Onshore Solutions
The foundations for onshore wind turbines employ well-known and controlled solutions. There are
mainly two possibilities: slab foundations or pile foundations (Figure 3.10).
64
Chapter 3
a) b)
Figure 3.10 – Examples of foundations for onshore wind turbines: a) slab foundation; b) pile foundation
Slab foundations are used when the ground near the top is of good quality. They consist on reinforced
concrete slabs which usually have a polygonal geometry (such as an octagonal shape) with a tapered or
constant thickness.
This kind of foundation resists to the overturning moment with the eccentricity of the reaction. The
vertical and horizontal loads are resisted by the dispersion of the load through the foundation area and
by the friction between the ground and the foundation, respectively.
Pile foundations are used for weaker soils. They are constituted by a pile cap from which a certain
number of piles are connected and extended until a good soil layer is reached. In this situation, the
loads are absorbed by the soil through axial and lateral pile resistance against the soil. Once again,
reinforced concrete is used in this solution.
Offshore Solutions
In contrast to onshore turbines, there is the need for diversified offshore solutions, depending on the
water depth. In addition, these solutions are considered far more complex than for onshore.
Offshore solutions for wind turbines can be divided in three categories: for shallow waters (up to 25 –
30 m), for transitional waters (from around 30 to around 60 m) and for deep waters (more than 60 m).
The most appropriate solutions for each water depth are presented in Table 3.1.
65
Basics of Wind TTurbines
T
Table 3.1 – Offsshore foundatio
on solutions co
oncerning wate
er depth
Shallow waterrs Transsitional waters Deep w

waters
Solutio
on
(0 – 30 m) (3
30 – 60 m) (+ 600 m)
Gravity-b
based ✓
Monop
pile ✓
Tripod ✓
Jacke
et ✓
Tripile ✓
ng
Floatin ✓
Gravity-bassed foundatio on is the second most ussed solution for current offshore
o winnd turbine prrojects
(Figure 3.111). It consistss on a caisson
n (usually maade of concrete) which iss filled with m
material to acchieve
the requiredd weight. Th herefore, a large footpriint is used to
o absorb mo oments due to environm mental
loads. The ccaisson is faabricated on land and thhen transportted to the sitte, where it is submergeed and
filled with ballast. For the operatio on on the siite, heavy veessels are required to seet the found dation.
Besides the need for heaavy machinerry, another ddisadvantage is the necesssary preparattion of the seeabed.
This operatiion ensures compaction
c and homogeenization of the
t soil to prevent unevenn settling.
This solutioon is considered, with reggard to the viibrational ch
haracteristics, as “stiff”. FFor this reaso
on, the
aerodynamiic damping of o the rotor do not contrribute in largge measure to t alleviate thhe response of the
structure duue to the dyn
namic loads (see
( section 44.5) (Hau, 20006).
Gravity fouundations aree indicated fo
or very shalllow waters (aaround up too 10 m – e.gg. the Nysted
d wind
farm in Den nmark). How wever, there are examplees of implemmentation at deeper
d waterrs (around 27 m –
e.g. the Thoornton Bank wind farm, ini Belgium (T Thomsen, Fo orsberg et al., 2007)).
Tower
(Se
ea Level)
(SSeabed)
Figure 3.11 – Gravity-based foundation ( photo (Scots Renewables,

R 2011) and illustraation)
66
Chapter 3
Monoopile foundaation is anotther alternattive solution for shallow w waters andd is, nowadayys, the mostt
used solution in offshore
o wind turbines (FFigure 3.12).. It consists on
o a free-stannding steel pipe
p which iss
driven or drilled (depending
( on
o soil qualitty) into the seabed
s by meeans of speciial and heavyy equipment..
In sittuations in which the monopile
m is implementeed by driving, hydraulicc hammers are a used forr
penettration. Sincce the pressu ure introducced by the hammers
h is very high, appurtenancces (such ass
outsidde ladders an nd the j-tubee) cannot bee installed directly on thee monopile. FFor this reasson, they aree
installled on the trransition piece. The prevvious preparaation of the seabed is not necessary foor monopiless
although significaant boulders should be avvoided on the foundation n area.
Contrary to the gravity-based
g d solution, m
monopile fou
undations arre consideredd as “soft”, in
i respect off
their dynamic beh
haviour (Hauu, 2006).
Monoopile foundaations are useed in deeper waters than gravity-baseed solutions. However, a water depth
h
of aroound 25 m to
t 30 m is a consensual llimit for the implementaation of this ffoundation. This type off
founddation was used
u in the Horns Rev win nd farm (in Denmark).
D
Tower
Transition Piiece
(Sea Levvel)
(Seabeed)
Figure 3.12
2 – Monopile fo
oundation (pho
oto (LORC, 2012
2b) and illustraation)
A diffferent appro oach under development

d for the foun
ndation of tu
urbine towerss is the suctiion bucket, a
technnology alread dy used in th he oil and gaas industry (Figure
( 3.13)) (Byrne, Hooulsby et al.,, 2002). Thiss
technnology consissts of skirted d shallow fouundations, wh hich inside water
w is pummped out, resulting in thee
sinkinng of the bu ucket into the soil due too the created d pressure diifferential. Suuction buckeets present a
savin
ng in materiaal (compared d to common n monopiles), and a fasteer and simpller installatioon, in whichh
hamm mering is avvoided (Byrne, Houlsby eet al., 2002).. However, there t is somee concern reegarding thee
lateraal stability an
nd the verticaality of the toower (Tong, 2010).
67
7
This type of foundatiion is still in an expeerimental ph hase. There are alreadyy some exaamples
implementeed (in Fredeerikshavn, Denmark) (Ibbsen, Liingaaard et al., 20
005) and moore recently at the
Dogger Bannk (UK) (C Carbon Trust, 2013)). H
However a project
p was cancelled ddue to installlation
problems in
n 2005 (Enerccon GmbH, 2005).
Tower
Transition Piece
(Sea Level)
(Seabed)
Figure 3.13 – Suction

S buckett foundation (p
photo and illusttration)
Water deptths of around 25 m – 30 0 m are the limit of currrent commo on commerccial offshore wind
turbines. Beeyond this liimit, the app
plicability off the presentted solutionss becomes veery expensivve and
demanding on material weight and on o machinerry for implem mentation. Fo or this reasonn, new found
dation
technologiees are under development
d t nowadays ((for transitional and deep p waters).
Tripod founndation is on
ne of the alteernatives for transitional waters. It is an expansioon of the mon
nopile
concept witth a larger footprint
f (Figure 3.14). T Tripods connsist on a strructure withh three legs which
w
diverge from
m a single no
ode. These thhree points arre anchored in the seabed d.
The tripod ggeometry alllows this solu
ution to be liight-weight and
a “stiff”. Since the loadds are mainlyy axial
in the legs, scour protection is geneerally not neeeded (Tongg, 2010). In addition,
a onlly reduced seabed
s
prearrangem ment is requ
uired. On thee other handd, these stru uctures are complex to bbuild (leading to a
high producction expendditure) and difficult
d to trransport (Haau, 2006). Also, the mainn joint of thee three
legs is comp
plex and a po
otential focuss of fatigue pproblems.
Tripod foun ndations are suitable for water depthhs up to arou
und 50 m – 60 m. This kiind of foundations
has been alrready implem
mented on th he Alpha Ven ntus wind farrm, in Germaany (Bartschh, 2012).
68
Chapter 3
Tower
Transition Piece
(SSea Level)
(Seabed)
Figure 3.14
4 – Tripod foun
ndation (photo (Wikipedia, 2012) and illustraation)
Anotther alternatiive for transitional water depths is thee jacket foun

ndation (Figuure 3.15). This solution iss
pplication off a common design empployed by thee oil and gass industry foor offshore structures.
an ap s Itt
consiists on a truss tower with three or fouur legs connected to each other by braacings.
Jackeet foundationns are light-w
weight and effficient strucctures, with a very good rresistance to overturningg
mom ments. Althou ugh they rep present a nottable reduction of material (steel), thhey are expensive due to o
the n
necessary welldings. Its insstallation proocess is simiilar to the tripod. The jaccket is transp
ported to thee
desiree site where the structuree is implemeented. The an nchoring of thet structuree is also realiized through
h
piles or with suctiion buckets (LORC,
( 20122a).
This type of strucctures is now
wadays ideallized for watter depths upp to around 50 m – 60 m. m However,,
this ttechnology iss employed for
f deeper wa waters on the oil and gas industry,
i so iit can be also
o potentiallyy
used for implemeentation of offshore
o turb ines in deeper waters. Th
he Beatrice w
wind farm (iin Germany))
is an example of implementat
i ion of wind tturbines on jacket
j foundations.
69
9
Tower
Transition Pie
ece
(Sea Level)
L
(Se
eabed)
Fig
gure 3.15 – Jacket foundation
n (photo (LORC
C, 2012a) and illustration)
The tripile iis other foun

ndation techn
nology installled nowadayys for medium
m water deptth (Figure 3.16). It
consists on a natural evvolution fromm the monoppile: instead of just one steel
s pile, thrree monopilles are
installed and connected at the top byy a transition
n piece.
This design n has the addvantage of increasing tthe stiffness of the mon nopile solutioon. Howeveer, the
installation of the three piles is of major
m import ance and needs a rigorou
us approach.. In addition
n, each
pile needs too be hammerred which is extremely n oisy (LORC,, 2012c).
The tripile solution is indicated fo
or water deppths up to 50 m. This technology has alreadyy been
implementeed in the BAARD Offshoree 1 wind farmm and in thee Hooksiel wind
w farm (B
BARD Engineeering
GmbH).
70
Chapter 3
Tower
(
(Sea Level)
(Seabed)
F
Figure 3.16 – Trripile foundatio
on (photo (BARD Engineering GmbH) and illuustration)
As reeferred in Ch hapter 2, onee of the todaay’s main fiellds of develo opment in thhe offshore wind
w turbinee
indusstry is the floating strructure for deep waterr depths. Nowadays, N thhere are already somee
experrimental pro ojects under developmen nt, however there
t is still a long way of developmment for thee
feasibbility of commmercial offsh
hore wind tuurbines basedd on floating foundationss.
Besid
des the possib
bility of intro
oduction of wwind turbinees at deep water depths, floating con ncepts enablee
a maass-productioon of structtures since they can achieve a high independdence from the seabed d
condiitions. Anoth
her advantagge of floatingg platforms iss the possibility of being ttowed to their location.
The design of floating
f stru
uctures folloows mainly three diffeerent conceppts to achieeve stabilityy
(Buttterfield, Musiial et al., 2005):
o Ballast – This kind ofo solution uuses the weiight of ballasst below a ceentral buoyaancy tank to o
create a righting
r momment and higgh inertial reesistance to pitch
p and rolll. The elongaated shape off
the struccture helps to
o minimize tthe heave mo otion. This so olution has bbeen used in the offshoree
oil indusstry for manyy years (Musiial, Butterfield et al., 2004
4);
o Mooringg Lines - Plattforms usingg this technollogy achieve stability throough the usee of mooringg
lines und
der tension;
o Buoyanccy – This lastt concept rellies on a disttributed buoy
yancy, achievved through the use of a
weighted
d plane area for
f righting m moment (Bu utterfield, Mu
usial et al., 20005).
Althoough each deescribed conncept presentts a differentt physical principle to acchieve stabiliity, in realityy
floatiing platforms usually preesent a hybriid system. Commonly, so olutions are designed based on all off
71
the three concepts, although generally relying on one primary source for stability (Butterfield, Musial
et al., 2005).
A representative design of each of the listed concepts is presented in Figure 3.17.
Tower
(Sea Level) (Sea Level)
(Sea Level)
Tower Tower
Mooring
Mooring Lines
Mooring
Lines Lines
(Seabed) (Seabed) (Seabed)
a) b) c)
Figure 3.17 – Floating solutions for offshore wind turbines(adapted from (Butterfield, Musial et al., 2005)): a) ballast; b)
mooring lines; c) buoyancy
3.1.1.6 Monitoring and Control System

Modern wind turbines are usually equipped with a monitoring and control system named SCADA
(Supervisory Control and Data Acquisition system). This system consists of a microprocessor, together
with several sensors, to control the operation and monitor the performance of the turbine.
Under this system, several data is registered through sensors that record, among others:
o Wind speed and direction;
o Rotor speed;
o Nacelle orientation;
o Generator speed;
o Pitch angle.
With this data, the control system is responsible for operations like the start and stop of rotor rotation,
emergency shut-down, adjustment of nacelle orientation and pitch angle. The system is also used to
monitor the power production from a remote location, since it is usually connected through any
means of communication (such as internet network). The information provided by the SCADA system
can be also used to monitor the condition of several components (see Chapter 5). This data represents
an important part in the architecture of the developed monitoring system (see Chapter 7).
3.1.2 COSTS BREAKTHROUGH

The investment costs of wind farms are of paramount importance for the feasibility of wind energy.
These costs are divided into capital costs (CAPEX - including all the costs for the implementation of
the farm) and variable costs (OPEX - including all the costs over the period of its operation – e.g.
operation and maintenance, land rental, insurance, among others).
72
Chapter 3
The NREL study present in (Tegen, Hand et al., 2012) describes some costs related to onshore models
of U.S. wind projects during 2010. Figure 3.18 presents the principal results of this study, namely the
capital cost share of the principal components of a wind turbine located in wind farm. These costs are
grouped in larger groups, namely:
o Turbine – comprises the wind turbine itself, which includes the tower, nacelle and rotor
components (Blanco, 2009);
o Balance of plant – including all the infrastructure except the turbine (foundations, buildings,
roads, among others) (Department of Energy & Climate Change, 2010);
o Installation and commissioning – including the installation and commissioning of the turbine
and balance of plant (Department of Energy & Climate Change, 2010);
o Development and consent – includes the multifaceted process of taking a wind farm from
inception through to the point of financial close or commitment to build, depending on the
contracting model, including several studies and contracts (Department of Energy & Climate
Change, 2010).
73
Onshore Wind Turbine Balance of Plant

Installed Capital Costs Installed Capital Costs
Engineering
Other Costs Foundations
& permits
9% 14%
6%
Turbine
transportation
9%
Balance of Electrical
Plant interface and
23% connections
37%
Turbine
Roads & civil
68%
work
20%
Turbine assembly &

installation 14 %
Turbine
Installed Capital Costs
Blades
14%
Tower
22%
Hub
Control, safety 4% Pitch mechanism &
system, and condition bearings
monitoring 4%
2%
Low-speed shaft
Nacelle cover 3%
1% Bearings
Electrical 2%
Hydraulic, cooling
connections
system
6% Gearbox
1%
11%
Main frame
10%
Variable-
speed Generator
electronics 7% Mechanical brake,
9% High sped shaft
Yaw drive & bearing coupling
2% 0%
Figure 3.18 – Installed capital costs for onshore wind turbine farms (Tegen, Hand et al., 2012)
The main installed capital costs of offshore wind turbines are shown in Figure 3.19 (Department of
Energy & Climate Change, 2010). These values are adjusted in order to give representative results for a
500 MW+ wind farm of approximately 5 MW turbines on jacket foundations in typical UK east coast
conditions around 80 km from shore in water depth of 30 m and considering market dynamics
prevailing at the start of 2010.
74
Chapter 3
Offshore Wind Turbine

Installed Capital Costs
Balance of
Plant
31% Turbine
39%
Development
Installation and
and Consent
Commissioning
4%
26%
Turbine Balance of Plant

Installed Capital Costs Installed Capital Costs
Tower Blades
15% 18% Offshore
Other Costs Hub Substations
12% Assembly 23%
9%
Foundations
Cables
Electrical 52%
16%
System Gearbox
20% 26%
Onsohre
Electrical
9%
Figure 3.19 – Installed capital costs for offshore wind turbine farms (Department of Energy & Climate Change, 2010)
Since offshore models are largely dependent on the water depth of implementation, a qualitative
representation of costs for different substructures is presented in Figure 3.20 (Musial and Ram, 2010).
Tripods,
Jackets,
Trusses Floating
Structures
Substructure Cost
Monopiles,
Gravity Foundations
Shallow-Water Transitional Deep-Water
Technology Technology Technology
0
0 20 40 60 80 100 120 140 160
Water Depth (meters)
Figure 3.20 – Cost of offshore wind turbine substructures with water depth (adapted from (Musial and Ram, 2010))
75
As can be seen from the previous figures, the main costs come from the turbine group. However, for
offshore turbines, the costs are more evenly distributed mainly due to the increased cost of
foundations and installation of the turbines (Department of Energy & Climate Change, 2010).
Another interesting aspect is the high relative cost importance of the tower, blades and gearbox. This is
highlighted in Figure 3.21 (capital costs of a 5.0 MW turbine from Senvion).
Turbine Components
Brake System
1%
Cables
1% Screws
1% Other Costs
Nacelle Housing 12% Tower
1% 26%
Transformer
4%
Power Converter
5%
Pitch System
3% Rotor Rotor Hub
Bearings 1%
Yaw System 1%
1% Generator
3% Main Frame Main Shaft
3% 2%
Gearbox
13%
Rotor Blades
22%
Figure 3.21 – Share of capital costs of the Senvion 5M wind turbine (Wind Directions, January/ February 2007)
3.1.3 WIND TURBINE COMPONENTS RELIABILITY

Wind turbines are located, to a large extent, in remote sites. Considering the case of Portugal as an
example (see Figure 2.33), most onshore wind farms are located on mountainous locations, with
difficult access. This fact makes the access of people and necessary machinery/ equipment for
maintenance and/ or reparation much more difficult and costly. If the offshore situation is considered,
in addition to the higher costs compared to the onshore solutions, the access to the structure is also
dependent on the weather conditions (Nielsen and Sørensen, 2011).
Considering all these constraints, a rigorous reliability analysis is necessary to optimize power
production (and profitability). Several strategies for maintenance exist today (Faulstich, 2010), from
simple derivation from past experience to more complex models with the help of installed sensors.
However, a simple separation of strategies can be used (Sørensen and Sørensen, 2011):
o Corrective maintenance – this model follows a reactive approach, which means that
maintenance is performed if a component has failed;
o Proactive maintenance – this strategy aims to anticipate and avoid the failure of components.
This kind of maintenance can be subdivided into two categories: preventive and predictive.
Preventive maintenance (also known as time-based) is referred to a strategy based on
76
Chapter 3
scheduled activities defined according to statistical models based on the historic of failures. On
the other hand, predictive maintenance is a strategy based on the assessment of the
components health through sensors installed in the components/ structure. The monitoring
system developed and described in this work falls in this category. This is the strategy expected
for future large onshore and offshore wind turbines.
Modern wind turbines achieve a high availability1, of around 97 %, when the maturity and teething
problems of the turbines are solved (Harman, Walker et al., 2008). However, several failures still occur,
implying unscheduled down times (Faulstich, 2010).
The UpWind project (Faulstich, 2010) analysed wind turbine failures in the German market between
1989 and 2006. It presents two major groups of failure causes:
o External causes (such as lightning, grid outage, icing, storm). These failures mostly affect
electric subassemblies and occur more frequently. However, the downtime associated with
each occurrence is usually short. For these cases, design optimization is suggested as solution
for prevention;
o Operation causes (such as wear out and relaxation). This type of failure mainly affects
mechanical components and, although seldom, usually implies large downtimes. An
assessment of the condition of the components through the use of sensors is recommended for
prevention of these failures.
Among different components, electrical systems represent the most common failures for onshore wind
turbines (Figure 3.22). However, rotor blades (including pitch system), the yaw system, the gearbox
and also electrical systems are the ones which imply longer downtime, according to Figure 3.23
(Faulstich, 2010; Wilkinson and Hendriks, 2011).
40 %
Power Module
Contribution to overall failure rate (failures/ turbine/ year) [%]
35 %
30 %
Rotor Module
25%
20 % Control &
Comms
15 %
Nacelle
10 %
Drive Train Auxiliary System
5% Structure
0%
HYDRAULIC SYSTEM
TOP
UNKNOWN
ANCILLARY EQUIPMENT
HIGH SPEED SHAFT TRANSMISSION
UNKNOWN
GENERATOR SILENT BLOCKS
eLECTRICAL PROTECTION & SAFETY
SERVICE CRANE
GROUNDING
FIREFIGHTING SYSTEM
ELECTRICAL AUXILIARY CABLING
FOUNDATIONS
WIND FARM SYSTEM
SENSORS
COMMUNICATION SYSTEM
SAFETY CHAIN
CONTROLLER H/W
CONTROLLER S/W
HUMAN & OPERATIONAL SAFETY ...
UNKNOWN
GEARBOX ASSEMBLY
MECHANICAL BRAKE
MAIN SHAFT
COOLING SYSTEM
WTG METEOROLOGICAL STATION
LIGHTNING AND POWER POINTS
GROUND
LIFT
ELECTRICAL CABINETS
UPS CABINET
BEACON
UNKNOWN
HUB CABINET
TOWER
COMMON FACILITIES
UNKNOWN
DATA LOGGER
CONDITION SENSORS & CABLES
PROTOCOL ADAPTER CARD FOR...
FREQUENCY CONVERTER
GENERATOR ASSEMBLY
LV SWITCH GEAR
MV SWITCH GEAR
TRANSFORMER
POWERFEEDER CABLES
UNKNOWN
POWER CABINET
PORTECTION CABINET
PITCH SYSTEM
BLADES
HUB
SLIPRINGS
HUB COVER
UNKNOWN
BLADE BEARINGS
SENSORS
YAW SYSTEM
NCAELLE COVER
NACELLE BEDPLATE
LIGHTNING PROTECTION SYSTEM

NACELLE SENSORS
Figure 3.22 – Normalized failure rate of sub-systems and assemblies for turbines of multiple manufacturers (for onshore
wind turbines) (adapted from (Wilkinson and Hendriks, 2011))
1
Availability time is defined as the ratio between the time during which the wind turbine is available to produce (with
none component failure) and the entire time under analysis.
77
40 %
Power Module
35 %
Contribution to overall downtime (hours lost) [%]
Rotor Module
30 %
25%
20 %
15 %
Control &
Comms
10 %
Nacelle Drive Train
5% Auxiliary System
Structure
0%
HYDRAULIC SYSTEM
TOP
UNKNOWN
ANCILLARY EQUIPMENT
HIGH SPEED SHAFT TRANSMISSION
UNKNOWN
GENERATOR SILENT BLOCKS
eLECTRICAL PROTECTION & SAFETY
SERVICE CRANE
GROUNDING
FIREFIGHTING SYSTEM
ELECTRICAL AUXILIARY CABLING
FOUNDATIONS
WIND FARM SYSTEM
SENSORS
COMMUNICATION SYSTEM
SAFETY CHAIN
CONTROLLER H/W
CONTROLLER S/W
HUMAN & OPERATIONAL SAFETY ...
UNKNOWN
GEARBOX ASSEMBLY
MECHANICAL BRAKE
MAIN SHAFT
COOLING SYSTEM
WTG METEOROLOGICAL STATION
LIGHTNING AND POWER POINTS
GROUND
LIFT
ELECTRICAL CABINETS
UPS CABINET
BEACON
UNKNOWN
HUB CABINET
TOWER
COMMON FACILITIES
UNKNOWN
DATA LOGGER
CONDITION SENSORS & CABLES
PROTOCOL ADAPTER CARD FOR...
FREQUENCY CONVERTER
GENERATOR ASSEMBLY
LV SWITCH GEAR
MV SWITCH GEAR
TRANSFORMER
POWERFEEDER CABLES
UNKNOWN
POWER CABINET
PORTECTION CABINET
PITCH SYSTEM
BLADES
HUB
SLIPRINGS
HUB COVER
UNKNOWN
BLADE BEARINGS
SENSORS
YAW SYSTEM
NCAELLE COVER
NACELLE BEDPLATE
LIGHTNING PROTECTION SYSTEM

NACELLE SENSORS
Figure 3.23 – Normalized hours lost per turbine per year due to faults in sub-systems and assemblies for turbines of
multiple manufacturers (for onshore wind turbines) (adapted from (Wilkinson and Hendriks, 2011))
Although the tower is not considered as a major source of problems, some aspects must be analyzed,
since the support structure of a wind turbine is fundamental for the integrity of the whole system. The
main problems assigned to the tower refer to corrosion and cracking (Liu, Tang et al., 2010), alongside
with more routine maintenance tasks, such as bolt tightening.
Some dramatic failures are registered in literature. In Devon, England, a small 50 kW wind turbine
collapsed in 2013. The cause was said to be the lack of resistance of the tower to the registered high
wind speeds (Gray, 2013). In Taiwan, a large wind turbine collapsed in 2008 due to insufficient
strength of bolts (Chou and Tu, 2011). Other cases of tower buckling due to a typhoon are reported in
(Li, Chen et al., 2013). A list of tower collapses is presented in (Chou and Tu, 2011), however, in some
cases, the causes are not available. In all these cases, the wind turbine was completely destroyed. The
share of wind turbine accident causes up to 2009 is presented in Figure 3.24. As can be seen, structural
failure represents a significant portion.
78
Chapter 3
Other
13%
Environmental
damage Blade failure
9% 23%
Human injury
5% Structural failure
Fatal 12%
accidents
9%
Fire
Ice rainfall Transport
19%
4% 6%
Figure 3.24 – Failure types of wind turbine accidents (Chou and Tu, 2011)
Figure 3.25 presents the estimated maintenance, repair and replacement costs of wind turbines (Gasch
and Twele, 2012). It classifies the support components (tower and foundations) as the elements with
the higher number of years between significant investments. This is expectable, since tower and
foundations are designed to maintain their structural integrity during the entire life period of the wind
turbine. However, this figure also demonstrates the high investment expected for the tower before the
expected period of 20 - 25 years, which means that, in average terms, this expenditure would have to be
considered.
180
Rotor blade
160
Maintenance and/ or replacement costs in €/ kW
140
Critical components
120
Uncritical components
100
Generator Tower
Pitch system
80
Gearbox
Control and
60 electronics Transformer Foundation
Yawing system station
40
Brakes
Shafts
Main bearing
20
Hydraulic system
Today’s period of experience Source: DEWI 2002
0
0 2 4 6 8 10 12 14 16 18 20
Number of years between significant investments in maintenance or replacement
Figure 3.25 – Estimated maintenance, repair and replacement costs of wind turbines (adapted from (Gasch and Twele,
2012))
79
3.2 WIND TURBINE PERFORMANCE

3.2.1 ROTOR POWER CHARACTERISTICS (PHYSICS)
The understating of the operation of a wind turbine has been continuously improved. The axial
momentum theory, initially developed by Rankine in 1865 (Rankine, 1865), was the first explanation of
power extraction at a rotor from a free-stream. In 1878, Froude originally developed the blade element
theory, where he studied the effects of the flow on blades. Based on the axial momentum theory, Betz
in 1920 (Betz, 1920) (and also, independently, Lanchester in 1915 and Joukowsky in 1920 (van Kuik,
2007)) calculated the maximum efficiency of an ideal wind turbine rotor, known as the Betz limit.
Besides these initial works, other authors such as Glauert and Prandtl also contributed with subsequent
developments (Burton, Sharpe et al., 2001). In this section, the combination of the axial momentum
theory with the blade element theory is presented, which constitute the base of the nowadays procedure
to analyze the wind turbine system aerodynamic performance.
3.2.2 AXIAL MOMENTUM THEORY (THE ACTUATOR DISC)

The axial momentum theory (also known as the actuator disc theory) assumes that a stream tube passes
through a disc, where the kinetic energy of the flow is extracted. In this theory, it is assumed that just
the mass of air that passes through the disc is affected (Figure 3.26). The mass flow rate remains
constant during the whole process. So, with the approach to the disc, the air stream slows down and,
consequently, expands its cross section area (due to the principle of linear momentum).
Flow Direction
Figure 3.26 – The energy extracting stream tube of a wind turbine (adapted from (Burton, Sharpe et al., 2001))
This theory considers the following assumptions (Manwell, McGowan et al., 2010):
o Homogenous, incompressible, steady state fluid flow;
o No frictional drag;
o An in-plane continuous disc (similar to an infinite number of blades);
o Uniform thrust over the disc;
o A non-rotating wake;
o The static pressure far upstream and far downstream of the rotor is equal to the undisturbed
ambient static pressure.
A schematic illustration of the flow mechanism according to this theory is presented in Figure 3.27.
80
Chapter 3
Far Ahead Actuator Disc Far Wake
Stream Tube
U0
Actuator Disk U0
UW
FT UD
U0
Axis
Section 0 Section D,0 Section D,W Section W

U0 UD UD UW
p0 pD,0 pD,W pW
A0 AD AD Aw
Stream Tube
Actuator Disc
Velocity
Velocity
Pressure
Axis
Pressure
Figure 3.27 – Actuator disc with stream tube. Velocity and pressure development along the stream tube (adapted from
(Tempel, 2006))
Considering the variation of linear momentum of the volume of air in the stream tube and the mass
conservation principle, it is possible to calculate the axial thrust force of the rotor ( ), considering the
variation of the cross section between 0 and ), represented in Figure 3.27:
. . . . (3.1)
with:
Axial thrust force of the rotor
Mass density of air
∗ wind velocity (in section *)
∗ Area of cross section (in section *)
∗ Section far ahead of the rotor
∗ Section far wake of the rotor
81
Since the flow is considered as in a steady state, it is possible to consider:
. . . . (3.2)
with:
Mass flow rate
The thrust force can then be defined as:
(3.3)
Applying the Bernoulli equation between the sections far ahead and in the vicinity ahead of the rotor
(∗ , ), and between the vicinity of the rotor wake (∗ , ) and the far wake of the rotor, it becomes:
1 1
. . , . . ,
2 2
(3.4)
1 1
, . . , . .
2 2
with:
∗ Pressure (in section *)
As exposed in Figure 3.27, the flow velocity is the same immediately before (section , 0) and after
(section , ) the actuator disc and the pressure is identical far away of the rotor disc (ahead and
after).
So, defining the thrust force according to the variation of pressure on each vicinity of the disc:
, , (3.5)
and introducing (3.4) in equation (3.5), one obtains:
1
. . (3.6)
2
82
Chapter 3
From equation (3.6) and considering . . in equation (3.3), it is possible to conclude that
the velocity in the disc section is the average value of the upstream and downstream flow speed:
(3.7)
2
Introducing the axial induction factor ( ) as:
(3.8)
it is possible to rewrite equation (3.7) as:
1
(3.9)
1 2.
As shown in the previous equations, defines the degree of slowdown of the wind at the actuator disc.
From equation (3.9), it is concluded that, in the extreme case, the induction factor can be = 1/2.
However, in this situation, the flow would cease and, thus, the flow mechanism would not be possible.
Since the power output ( ) is obtained as the rate of work done by the force at the actuator disc, it can
be written as:
1
. . . . . 4. 1 (3.10)
2
where the term defined in the first square brackets represents the power that is present in the flux,
while the second square brackets represents the efficiency of the power capturing.
Considering equations (3.6) and (3.9), the thrust force is defined as:
1
. . . . 4. 1 (3.11)
2
83
Usually, both and parameters are characterized by non-dimensional coefficients: the power
coefficient ( ) and the thrust coefficient ( ), respectively:
1
. . . . 4. 1
2 4. 1
1 (3.12)
. . .
2
1
. . . . 4. 1
2 4. 1
1 (3.13)
. . . .
2
In Figure 3.28 both coefficients are represented. As expected, the maximum values of the coefficients
are attained for different values of (as already referred, values of greater than 1/2 are unrealistic).
CT
(Dimensionless Amplitudes)
0.8
C P, C T
0.6 CP
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
a (Induction factor)
Figure 3.28 – Variation of and with the axial induction factor
The calculation of the induction factor for the maximum values of the coefficients is obtained with:
d 16 1
0→ , ≅ 0.593, (3.14)
d 27 3
d 1
0→ , 1, (3.15)
d 2
While the results obtained for the thrust coefficient are not of too much importance (due to the fact
that the maximum value is obtained for an unrealistic situation), the result of equation (3.14)
represents an important achievement. The value of 16/27, named as Betz Limit, is considered the
theoretical maximum limit for power extraction from the flow. This means that, no matter the
efficiency of the machine, no wind turbine can extract more than 59.3 % of the kinetic energy of the
wind.
Obviously, this theory does not accurately characterize the real physical situation: the rotor is not
composed by a uniform disc with only axial thrust acting on it and the wind does not continue as an
undisturbed axial flow after passing through the rotor, among other things. To overcome these
84
Chapter 3
simplifications, the variation of angular momentum in the flow is considered in the theory presented
in the next section.
3.2.3 ANGULAR MOMENTUM THEORY

The angular momentum theory also studies the extraction of power from the wind flow. However, this
theory, unlike the axial momentum theory, assumes the rotation of the rotor and, consequently, the
creation of a rotational wake flow. This rotational wake flow has an opposite direction from the rotor
rotation in reaction to the torque exerted by the flow on the rotor.
An important difference from the axial momentum theory is that the energy extracted from the flow is
lower. This is due to the energy employed in the rotor and wake rotation. It is assumed that the
angular velocity of the rotor (Ω) is high when compared to the angular velocity of the wake flow ( ).
Since the tangential speed of the rotor is not the same for all radial positions, annular rings of radius
and thickness d are considered, defining areas of 2. . . d (Figure 3.29).
Ω
Axis
R
dr U0 (1-a)
r
ω.r
Figure 3.29 – Annular ring in the stream tube and decomposition of wake wind speed (adapted from (Manwell, McGowan
et al., 2010))
As shown in Figure 3.29, the wake flow has an axial speed of 1 (like in the axial momentum
theory) and a rotational speed of .
Defining an angular induction factor as:
′ (3.16)
2. Ω
85
the angular velocity of the wake flow can be written as:
2. Ω. ′ (3.17)
The torque of the annular stream is defined as the rate of change of angular momentum (due to the
rotational speed):
d . 2. . . d . 1 . 2. Ω. .
(3.18)
4. . 1 . . Ω. . .d
with:
Torque on the actuator disc
With the defined rotor torque, it is possible to calculate the power output due to it:
d T. Ω (3.19)
Defining the relationship between the tangential speed of the rotor at a radial distance and the wind
speed as the local speed ratio (or tip speed ratio, for the case of ):
Ω.
(3.20)
Ω.
The power output (3.19) for each ring of radius and thickness d can be written as:
1
d . . . 2. . . d . 4. ′ 1 . (3.21)
2
Once again, the term defined in the first square brackets represents the power of the flux, while in the
second square brackets, the efficiency of the capturing element is quantified (also known as blade
element efficiency (Burton, Sharpe et al., 2001)).
86
Chapter 3
As shown in (3.21), the power output extracted from the wind when the rotational wake is considered,
depends not only on and , but also on the tip speed ratio. The power coefficient is defined by:
d 8
d ⇒ ′ 1 d
1 (3.22)
. . .
2
Although the calculation of the integral in the equation (3.22) is not presented in this work (it can be
consulted in (Manwell, McGowan et al., 2010)), it is important to note two things:
o The power coefficient ( ) depends on the tip speed ratio (λ);
o The maximum power that can be extracted from the wind tends to the Betz Limit (16/27).
Figure 3.30 illustrates the previous statements.
Betz limit
0.6
0.5
0.4
CP
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18 20
λ (Tip speed ratio)
Figure 3.30 – Evolution of with λ according to angular momentum theory
Considering the previous statements, it is possible to conclude that the evolution from rotors with
fixed-speed to rotors with variable-speed was a major evolution step for the efficiency in energy
harvesting. Whereas in a fixed-speed, the optimum tip-speed ratio is only achieved for a specific wind
speed, with a variable-speed turbine, it is possible to have different optimum points of operation for
each wind speed (Figure 3.31).
87
1200
variable speed
1000
800
fixed
speed 13 m/s
P [kW]
600
400 11 m/s
200 9 m/s
7 m/s
0
0 10 20 30 40 50 60 70 80 90 100
Rotational speed [RPM]
Figure 3.31 – Power output as function of rotational speed for a fixed- and variable-speed wind turbine for different wind
speed classes (adapted from (Tempel, 2006))
Although the previous momentum theories already explain the power extraction from a wind flow, the
problem is still not solvable yet due to unknown axial and angular induction factors ( and ). For
this reason, an additional approach to this problem is needed: the blade element theory.
3.2.4 BLADE ELEMENT THEORY

In this theory, the forces acting on the blades are expressed as a function of their shape and
orientation. The following assumptions are adopted:
o Each blade is divided into elements along its longitudinal axis;
o There is no radial flow (the aerodynamic interaction between elements is ignored);
o The forces acting on the blades are determined only according their airfoil characteristics: lift
( ) and drag ( ) coefficients;
o Three-dimensional effects are neglected.
Figure 3.32 presents a different perspective of the rotor annular ring of the Figure 3.29. It also presents
a detailed illustration of a blade, with an arbitrary airfoil and the wind and rotor velocities associated
with one element (grey area).
88
Chapter 3
Ω.r.a’
δr
r
U0(1-a)
r Ω.r
Figure 3.32 – Rotor annular ring and detail of a blade element (adapted from (Burton, Sharpe et al., 2001))
As previously introduced, a blade element with a radius and a radial width d is subjected to an axial
wind speed of 1 and a rotational wind speed2 of Ω. created by a rotor angular speed of Ω.
Since tangential speeds are obtained from rotational speed multiplied by , the resultant relative
velocity between the blade element and the wind is:
1 Ω. 1 ′ (3.23)
An illustration of the relative velocity, its components and the associated lift ( ) and drag ( ) forces
is presented in Figure 3.33 (where the chord of the airfoil is represented by , represents the pitch
angle and represents the angle of inflow).
Ω.r(1+a’)
θ
α Ф
URel
FL
Wind Direction
U0(1-a)
c FD
Figure 3.33 – Relative wind velocity applied to the airfoil and the subsequent lift and drag forces
2
The rotational speed of the blade element is obtained assuming a linear increase from 0 (at the far ahead location) to
2. Ω. ′ (in the far wake location).
89
Having in mind the relative velocity defined in equation (3.23), it is possible to define the drag and lift
forces:
1
d . . . . .d (3.24)
2
1
d . . . . .d (3.25)
2
with:
Drag force
Lift force
Angle of attack
3.2.5 BLADE ELEMENT MOMENTUM

The Blade Element Momentum (BEM) is defined as a joining of the momentum and blade element
theories, in order to solve the blade load problem.
Summarizing the results obtained with the previous deductions, one gets from the axial momentum
theory:
d 4. 1 . . . .d ,
(3.26)
Considering: d 2. . . d
and from the angular momentum theory:
d 4. . 1 . . Ω. . .d (3.27)
Considering now the results presented for the blade element theory (equations (3.24) and (3.25)) and
calculating their effects on the rotor, the following expressions for d and are obtained:
1
d . . .d (3.28)
2
1
d . . . .d (3.29)
2
with:
Number of blades
90
Chapter 3
Taking into account the following relationships:
.
(3.30)
2. .
1
(3.31)
with:
′ Local solidity
it is possible to obtain expressions for the axial and angular induction factors, equating (3.11) with
(3.28) and (3.18) with (3.29), respectively:
′ . .
(3.32)
1 4.
′ ′ . .
(3.33)
1 4. .
(According to some authors (Wilson and Lissaman, 1974), it is acceptable to set as zero)
However, it is important to consider that this theory refers to an independence of each annular section
of the blade to the adjacent ones. Although this does not correspond to reality, for normal operations
tip speed ratios, the error is considered as acceptable. Beside this aspect, some corrections should be
introduced to the BEM theory:
o According to equation (3.9), the axial induction factor has to be lower than 0.5, otherwise a
reversal of the wake flow would be induced. Since this cannot occur, a different explanation is
needed for this situation. In reality, for high values of (which lead to high values of ), the
permeability of rotor is almost comparable to a disc. Some authors, such as Glauert (Glauert,
1926), proposed a correction for the axial induction factor curve (Figure 3.34);
2. 0
Glauert empirical relation
1. 5
CT
1. 0 CT = 4a(1-a)
0. 5
Turbulent state wake
0. 0
0.0 0. 2 0.4 0. 6 0. 8 1.0

a (Induction factor)
Figure 3.34 – Trend of thrust coefficients for high values of axial induction factor (adapted from (Manwell, McGowan et
al., 2010))
91
o Another correction that should be taken into account is the loss of power production near the
tip of the blades - called tip-loss factor. This is due to the discrete number and finite length of
real blades, which is not accounted in the classical approach of the BEM theory. The most
common approach to this problem is through the Prandtl’s approximation.
1.5
Tip/root-loss factor
0.5
0
0 0.2 0.4 0.6 0.8 1
r/ R
Figure 3.35 – Span-wise variation of combined tip/root-loss factor for a three-blade turbine optimized for a tip speed ratio
of 6 and with a blade root at 20% span (Burton, Sharpe et al., 2001)
3.2.6 CONTROL STRATEGIES OF WIND TURBINES

At this point, the most important aspects of the theories explaining the performance of a wind turbine
were introduced. Having those in mind, it is now possible to introduce some aspects related to control
in the operation of the turbine. These control operations have an important influence in the dynamic
behaviour of the overall structural system.
The performance of wind turbines requires some kind of control (passive or active) in order to
adequately operate as expected. This is a crucial aspect mainly due to two reasons: to maximize the
power production and to avoid excessive loads that could jeopardize the integrity of the structure.
In this subsection, three types of control will be briefly described: blades aerodynamic control (stall
and pitch control) and rotor speed. The purpose of the description is to present the basic aspects
related to control operations behind the different operating regimes of the wind turbines that will be
later described in this work.
3.2.6.1 Stall Control

Stall control is a solution to control the aerodynamic loads acting on the rotor. This solution is based
on the separation of the flow at the surface of the blades with increasing the wind speed for a constant
tangential velocity of the rotor (without changing the blades pitch angle). This flow separation leads to
a decrease of the lift force of the blade, limiting the aerodynamic power input.
Historically, stall control is a passive solution. As can be seen in Figure 3.36, with the increase of the
wind speed, the angle of attack increases for a fixed blade pitch angle. The graph in the lower left
corner of Figure 3.36 shows the evolution of the lift coefficient with the angle of attack for a common
airfoil. With this graph, it is visible that the increase of the angle of attack beyond a certain limit leads
to the creation of instability, dropping the lift force.
92
Chapter 3
Ω.r(1+a’) Ω.r(1+a’)
θi = θf θi = θf
URel,i αi URel,f α f > αi
Wind Direction
U0,i(1-a) U0,f(1-a) > U0,i(1-a)

CL = f(α)
2.0
1.0
CL
0.0
-10 0 10 20 30
α[º]
Figure 3.36 – Aerodynamic stall effect on a blade profile
Wind turbines with this design present a lower aerodynamic efficiency below rated regime than with a
solution based on pitch control, since the rotor must operate closer to stall.
Although stall control seems to be a simple solution, it has some drawbacks, such as a less precise
control of the power production and the inability to drop the thrust force of the rotor at high wind
speeds (when compared to the pitch control solution) (Hau, 2006). For this reasons, the stall control is
usually only used with turbines with a power output up to 1.5 MW.
With large wind turbines, the attempts to introduce this control technology have been made with a
more complex construction, with the use of an active blade pitch angle adjustment. That is the case of
the Izar Bonus 1.3MW/62 studied in Chapter 5.
3.2.6.2 Pitch Control

Pitch control of the blades is the nowadays standard solution in the wind industry. With this design,
the blades are able to turn about their longitudinal axis with the help of actuators, allowing adjusting
the angle of attack of the blade in order to regulate the rotor speed while keeping the load torque
constant. Rotor blades are able to pitch until reach a feathered position (around 90º). This mechanism
is illustrated in Figure 3.37.
Ω.r(1+a’) Ω.r(1+a’)
θ αf θ
URel αi URel
Wind Direction
U0(1-a) U0(1-a)
Pitch rotation
Figure 3.37 – Pitch control mechanism
93
Pitch control of the blades allows a better control of the rotor speed, resulting in a more constant
power output. In Figure 3.38, the power curves of two wind turbines with stall and pitch control are
shown. As can be seen, the last part of the curve (after the rated speed), is considerably more flat with
pitch control. This is the solution implemented in both the Vestas V90-3.0 MW and Senvion MM82
studied in Chapters 5 and 7.
600
Pitch regulated
500 turbines
400 Stall regulated

turbines
Power [kW]
300
200
100
0 Rotor diameter = 40 m
0 5 10 15 20 25 30
Wind Speed [m/s]
Figure 3.38 – Power curve of a 500 kW wind turbine with rotor regulation by active pitch and passive stall (adapted from
(Burton, Sharpe et al., 2001))
3.2.6.3 Rotor Speed Control

In the case of variable-speed generators, it is possible to control, under a certain range, the speed of the
rotor (or generator). As previously referred, this is an important advantage since the optimum
production efficiency is achieved at the optimum tip speed ratio. Thus, a (theoretically) continuous
adjustment of the rotor speed would be required to keep up with the variation of the wind speed
(according to equation (3.20)). According to equations (3.10) and (3.20), the power output can be
defined by:
1
. . π. . .Ω (3.34)
2
The maximum aerodynamic power is defined by introducing the optimum tip speed ratio (λ ) and
the maximum power coefficient ( ) in equation (3.34), while the rotor speed is optimum
(Ω ). Equation (3.34) can then be rewritten as:
.Ω (3.35)
where
1
. . π. . (3.36)
2
94
Chapter 3
Equation (3.35) shows that the maximum aerodynamic efficiency is proportional to the cubic of the
rotor speed. In the same sense, the torque is defined as proportional to the square of the rotor speed:
.Ω (3.37)
Considering these equations, it is concluded that, under steady state conditions (and assuming that the
variables from equation (3.36) are at their optimum values), the maximum efficiency can be achieved
by varying the rotor torque according to the squared value of the rotor speed. This is the control
strategy for maximum efficiency. However, some restraints prevent the operation of the turbines
following this strategy over the whole operating regimes.
Figure 3.39 illustrates the relationship between the torque and speed of the generator from a variable-
speed wind turbine with pitch control. For this turbine, the referred strategy is followed between the
operating points B and C. In this operating regime, the generator torque is proportional to the square
of the generator speed, in order to be in line with the optimum power coefficient (red line in Figure
3.39). This means that, (theoretically) optimal power extraction occurs within this operating zone.
Due to aerodynamic noise, loads and other design constraints, it is not possible to follow the optimum
curve for all wind speed regimes, being the maximum allowable rotor speed achieved at a relatively
low wind speed (Burton, Sharpe et al., 2001). For that reason, when reaching the operating point C, a
different control strategy is followed. At this point, the increase of power production (with the increase
of wind speed) is obtained by increasing the torque, while the rotor speed is kept at its rated value (line
CE). When the rated generator rotor torque value is achieved, the power production is maintained at
the same level by adjusting the pitch angle of the blades, which allows controlling the input torque
generated by higher wind speeds (point E).
It should be noted that the described chain of operating regimes are referred to a situation in which no
limitations are present for control. In reality, some constraints are present in the definition of the
control algorithms, leading to the need of transitional regimes. Thus, the dashed line A’B’C’D’E’
defines a chain of operating conditions more in line with the operation of wind turbines in real
conditions.
180
Rated speed
18 m/s
160
16 m/s
Generator torque
140
14 m/s
Rated torque
120 12 m/s
D’ E
100
10 m/s
80
8 m/s
C
60
6 m/s
C’
40
4 m/s
20
B
Optimum Cp
0
0. 00 200.00 40 00 600. 800.00 1000.00 1200.00 1400.00 1600.00 2000.00
A’ A Generator speed
Figure 3.39 – Illustration of the torque-speed curve for a variable-speed wind turbine with pitch control (adapted from
(Burton, Sharpe et al., 2001))
95

This chapter starts with the description of the main structural elements of wind turbines. A special
focus is given to the support structure (tower and foundation – in onshore models; tower, transition
piece and foundation – in offshore models). This element conditions, in a large extent, the modal
properties and, consequently, the dynamic behaviour of wind turbines.
The chapter continues with a depiction of the main costs related to a wind farm project for onshore
and jacket-based offshore models. A high relative cost is found for the support structure and blades.
Specially in the case of offshore installations, the foundation represents an important share of the
overall cost.
Also the reliability of the various elements of wind turbines is analysed. It is seen that accumulated
experience demonstrated that the support structure in onshore wind turbines is not a commonly
problematic element. However, the increase of the wind turbine dimensions, together with the
installation of turbines in offshore harsh environments, is expected to introduce new problems. The
evaluation of the reliability of the wind turbine components is a key aspect on the definition of the
maintenance procedures, including new strategies based on data collected by installed monitoring
systems.
Lastly, a brief description of the wind turbine performance based on the axial momentum and blade
element theories is introduced. This theoretical background represents a helpful insight for the
comprehension and definition of the various operational conditions of a wind turbine. The variation
of the conditions requires the definition of control strategies for the proper operation of the turbine. In
that sense, simplified aerodynamic and rotor torque control strategies are included in this chapter. As
will be seen in the following chapters, the variations imposed by the various regimes of operation of a
wind turbine influence its modal properties.
96
Chapter 4
4
STRUCTURAL BEHAVIOUR OF WIND
TURBINES
4.1 STRUCTURAL DYNAMICS

Wind turbine structures are highly dynamic systems. As illustrated in Figure 4.1, the normal operation
of a wind turbine is a process in which different types of dynamic excitation take place. However, two
main sources of excitation are evident: environmental conditions (wind loading and offshore
conditions, if applicable) and the rotating machinery during operation, mainly the rotor.
Wind
Wake effects
Waves
Tides and
Currents
Figure 4.1 – Sources of dynamic excitations of an offshore wind turbine (for onshore turbines, tides, currents and waves
are, obviously, not applicable)
97
Structural Behaviour of Wind Turbines
Thus, an accurate description of the dynamic behaviour is extremely important in this context,
otherwise resonance phenomena are very likely to occur due to several sources.
A complete wind turbine structure can be idealized as an assembly of coupled multi degree-of-freedom
mass-spring-damper systems. With this in mind, the structure is defined according to its mass,
damping and stiffness properties (Clough and Penzien, 1995):
. . . (4.1)
with:
Mass matrix
Damping matrix
Stiffness matrix
Displacement vector
Excitation force
Although the complete characterization of a wind turbine might be complex, simplifications for some
components can be made for initial considerations. Within the scope of this thesis, the most important
aspects related with the dynamic characterization of the tower, blades and the tower-blades coupling
are presented.
4.1.1 DYNAMIC CHARACTERIZATION OF THE TOWER

The tower structure of a wind turbine is a component for which a rigorous dynamic characterization is
essential. However, simple geometric sections are usually used (namely, tubular sections) and a
sufficiently accurate characterization of the material is also often achievable with simple methods
(mainly for steel structures). Furthermore, the rotation of the rotor introduces small influence on the
first natural frequency of the tower, which enables its calculation with the rotor at stand-still position
with sufficient rigorous results. Taking this into account, a simple cantilevered column with
distributed mass and a concentrated mass at the top (Figure 4.2) can be used for initial assessment of
the natural frequency of the first tower bending mode.
Figure 4.2 – Simplified structure for calculation of the natural frequency of the tower bending mode
98
Chapter 4
With this configuration, approximate values of the tower first natural frequency can be obtained
through:
, .
2. (4.2)
. , . .
with:
Tower first natural frequency
, Constant value: 1.732 (Young and Buynas, 2002) or 1.7436 (Vugts, 2000), depending
on literature sources
, Constant value: 0.236 (Young and Buynas, 2002) or 0.227 (Vugts, 2000), depending
on literature sources
Tower elastic modulus
Tower section inertia
Tower top mass (including nacelle, all the equipment in it and the rotor)
Tower mass per unit length
Tower height
Figure 4.3 presents an illustration with the mode shape configuration of the first vibration mode for
the fore-aft (1 FA) and side-side (1 SS) directions.
1 FA 1 SS
Figure 4.3 – First tower vibration mode for fore-aft and side-side directions: undeformed (grey) and deformed (black)
structure (adapted from (Skjoldan, 2011))
4.1.2 DYNAMIC CHARACTERIZATION OF THE ROTOR

The definition of the dynamic properties of the rotor blades presents a more complex nature than for
the tower structure. Besides the material composition and geometry of the blades, it is important to
realize that the dynamic behaviour of the rotor acts as a whole, which means that the stiffness of the
shaft supporting bearings has a direct influence on the vibration modes. It is also important to
understand that, when the wind turbine is operating (and the rotor is spinning), two particular effects
take place which affect the dynamic behaviour of the structural system: gyroscopic and centrifugal
stiffening effects.
99
The rotor modes of wind turbines can be divided into two groups: flapwise and edgewise (see Figure
3.2). For each order of these modes, three “sub-modes” need to be considered, namely, one symmetric
and two asymmetric modes in each direction. These modes are presented in Figure 4.4. The symmetric
mode is characterized by the deflection of the three blades in the same direction:
o Flapwise modes: the blades deflect in the rotor out-of-plane direction (1 FS);
o Edgewise modes: the blades deflect in the rotor plane direction (1 ES).
On the other hand, the two asymmetric modes are defined accordingly to the direction of the motion
of the blades:
o Flapwise modes: tilt (1 FT) and yaw mode (1 FY);
o Edgewise modes: horizontal (1 EH) and vertical mode (1 EV).
1FS 1FT 1FY
1 ES 1EH 1EV
Figure 4.4 – First order rotor modes (for low values of pitch angle): undeformed (grey) and deformed (black) structure
(adapted from (Skjoldan, 2011))
These asymmetric modes are associated with translation and rotation of the rotor. This is the reason
why the stiffness of the supporting elements of the rotor has an important role in these modes.
When the wind turbine is under operation and the rotor is spinning, two particular effects need to be
accounted for: gyroscope effects and centrifugal stiffening. These effects should be considered in the
equation of motion (equation (4.1)) of the rotor as:
. (4.3)
with:
Gyroscopic matrix
Geometric stiffness matrix
100
Chapter 4
Gyroscope Effect
The rotation of the rotor introduces a well-known characteristic in the context of rotor dynamics: the
gyroscopic effect. This effect is due to the flexibility of supporting bearings of the rotor which leads to
deviation of the rotating axis from the bearing centre line (Yoon, Lin et al., 2013). Essentially, when the
rotor is rotating at the shaft axis, the motion in the two orthogonal axes is coupled. This is evident by
the skew-symmetric nature of the gyroscopic matrix (Ewins, 2000; Yoon, Lin et al., 2013). Thus, under
rotation, the rotor suffers two rotational movements: the rotation of the shaft itself; and a whirl
rotation due to gyroscopic effects. There are two types of whirl rotation, namely, forward (when the
rotation is in the same direction of the shaft) and backward (when the rotation is in the opposite
direction). Figure 4.5 illustrates these two phenomena for two different situations: one where the
rotor/ stator system is symmetric and another where the system is non-symmetric.
Ω Ω Ω Ω
a.1) a.2) b.1) b.2)

Figure 4.5 – Motion of the rotor system (for a fixed reference) including the gyroscopic effect (the smaller trajectory
represents the rotation of the shaft around itself, while the larger path indicates the whirl motion): a) represents a
symmetric rotor system; b) represents a non-symmetric system; .1 represents the backward whirl mode; .2 represents the
forward whirl movement (adapted from (Ewins, 2000))
The dynamic behaviour of wind turbine rotors is similar to what was described above. With the
rotation of the rotor, the two asymmetric modes couple (the tilt with the yaw flapwise mode; and the
horizontal with the vertical edgewise mode), resulting in the development of the whirl modes. The
frequency of these whirl modes (backward – BW - and forward - FW) drift, from a fixed-frame
reference, due to the rotor rotation. This situation is illustrated in Figure 4.6. As exposed in Figure 4.5,
the forward and backward definition of the whirl modes refers to the rotation of the rotor centre of
mass in the same or against the rotor direction, respectively (Skjoldan, 2011).
101
4 2. flapwise forward whirling
2 FT
2 FY 1. edgewise forward whirling
2. flapwise backward whirling
1 EH -Ω
3
1 EV +Ω
Natural frequency [Hz] 1. edgewise backward whirling
2 1. flapwise forward whirling

1. symmetric flapwise
1 FS -Ω
1 FT
1 FY +Ω
1. flapwise backward whirling
1
1. longitudinal tower bending
1. lateral tower bending
1. fixed-free drivetrain torsion
0
0 5 10 15 20 25
Rotor speed [rpm]
Figure 4.6 – Campbell diagram of fixed-frame natural frequencies (O) for the first 10 structural modes of a 600 kW three-
bladed wind turbine. Lines denote the centre frequencies of the rotor whirling modes given Ω added to the fixed-frame
natural frequencies (adapted from (Hansen, 2007))
Figure 4.6 illustrates some important phenomena that should be considered when studying the
dynamic behaviour of wind turbine rotors.
When the rotor is stopped, modes 4 and 5 (first flapwise yaw and flapwise tilt modes) are very close.
The difference between them is due to different elastic supports in the corresponding directions.
Usually, the yaw mode is lower than the tilt mode because towers are stiffer in tilt than in yaw
direction (Hansen, 2003). The 6th mode, the flapwise symmetric mode, is slightly higher than the
asymmetric ones. This is justified by the fact that this mode, in which all the blades vibrate
simultaneous in-phase, beholds also some fore-aft motion of the tower in out-of-phase (Hansen,
2003). When the rotor starts to operate, a change in the frequency value of the modes is evident. As
highlighted in Figure 4.6, the pair of asymmetric modes starts to couple and became pairs of whirling
modes. Thus, FW and BW modes start to drift, increasing and decreasing their frequency values,
respectively, in an amount equal to the rotor speed (±Ω). This splitting phenomenon, identified
considering a fixed reference of observation, is due to the gyroscopic effect.
Modes 7 and 8 have a similar behaviour as the described for the 4th and 5th mode. These frequencies are
also slightly different when the rotor is stopped due to different flexibilities of the rotor supports.
When the rotor starts to spin, a similar splitting effect is noticeable.
In the last two modes (2 flapwise FW and BW modes), the difference at parked situation of these two
frequencies indicates an even greater difference in the requested flexibility of the support in both
directions. Although the behaviour of these two modes is very similar to modes 4 and 5, it is important
to refer that it is less pronounced due to the lower contribution of the gyroscopic reaction forces
(Hansen, 2007).
102
Chapter 4
The evolution of the frequencies with the rotor speed usually leads to situations where two different
modes stay very close to each other. In the presented situation, the 1st flapwise FW with the 1st
symmetric mode and the 1st edgewise FW with the 2nd flapwise BW feature very close frequencies. In
these situations, the vibrations modes interact with each other, presenting a mode shape with a mixed
configuration.
One last aspect should also be mentioned. If Figure 4.6 is observed carefully, it is possible to verify that
the frequency of the 1 FS mode, which does not contain any gyroscopic effect, also increases with the
rotor speed. This situation is also noticeable for the centre lines of the whirl modes (solid line) which
imply that another effect is present that increases the value of the frequencies too. This phenomenon is
called centrifugal stiffening.
Centrifugal Stiffening Effect

As introduced in the previous section, the centrifugal stiffening effect is responsible for the increase of
the rotor modes frequencies with the rotor speed. As referred, this effect is very pronounced in the
flapwise modes, being the responsible by the curved path of these frequencies in the Campbell diagram
(Figure 4.6).
This effect is justified by an increase in stiffness due to centrifugal forces acting on the rotating blades.
This centrifugal stiffening is then defined by the additional tensile forces installed on the blades for a
position along the blade (Naguleswaran, 1994):
. .Ω d (4.4)
with:
Centrifugal force
Density of the blade
Cross section area of the blade
Ω Rotational frequency of the rotor
Blade length
Rotor hub radius
Position of the blade under consideration
Alongside with the tensile force presented in the blade due to the centrifugal forces, the influence of
the gravity (self-weight) can also be introduced. This effect, which is reflected by an axial force acting
on the blade, depends on the position of the blade (Murtagh, Basu et al., 2005). Thus, the effect of the
self-weight of the blade should be added or subtracted to its stiffness, either it is a tensile or a
compressive force, respectively. However, it is referenced that this gravity effect is negligible when
compared to the effect of the centrifugal stiffening (Arrigan, Pakrashi et al., 2011).
103
Several authors, among which (Naguleswaran, 1994) and (Yoo and Shin, 1998), studied the problem of
determination of natural frequencies of simple beams rotating on a hub taking into account the
centrifugal stiffening. However these studies focused on simple beams, with constant parameters along
the length of the beams, which diverge substantially from a common wind turbine blade. More
complex studies were developed for the specific case of the blades, e.g (Hansen, 2003), however they
are outside the scope of this work. Nevertheless, a rough estimation of the increase of the natural
frequency of a blade due to centrifugal stiffening is presented (Putter and Manor, 1978):
, .Ω (4.5)
with:
, Angular frequency of a rotating blade
Angular frequency of a blade at stand-still position
Southwell coefficient: 1.73 (Madsen, Frandsen et al., 1984)
Although this effect is important and should be considered when studying the dynamic behaviour of
the rotor vibration modes, it is not very noticeable in edgewise modes. This is justified by the softening
effect that is also introduced by the centrifugal forces on the linear stiffness of deflection in the rotor
plane that almost cancels the stiffening effect (Hansen, 2007).
104
Chapter 4
4.2 FOUNDATION SOIL STIFFNESS

The consideration of the soil stiffness is an important aspect when trying to accurately simulate the
behaviour of tall, slender structures. Thus, wind turbines are no exception. As an example, the
guidelines of the Danish standard (Det Norske Veritas (DNV), 2002) refer a reduction between 0 %
and 5 % up to a maximum of 20 %, when comparing a model with fixed boundary conditions and
another with the consideration of the soil stiffness. A larger variation is presented by Hau (Hau, 2006)
for an onshore wind turbine with a slab foundation (Figure 4.7).
0.5
1st Bending eigenfrequency [Hz]
0.4
0.3
m = 250 t
0.2
ground shear module 106 N/m2 t = 18 mm
H = 70 m
D = 5.0 m
- sand, loose, round 20 - 50
- sand, loose, sharp 40 - 80
0.1
- sand, med. dense, round 50 - 100
- sand, med. dense, sharp 80 - 150
- pebbles, without sand 100 - 200
12 m
- brash, sharp 150 - 300
0
0 100 200 300
Ground shear module ES 106 N/m2
Figure 4.7 – Influence of the soil stiffness on the first natural bending frequency of a wind turbine tower (Hau, 2006)
For the modelling of the soil stiffness, the recommendations defined in the DNV guidelines (Det
Norske Veritas (DNV), 2002) are then presented. In order to introduce the effect of the soil, four
foundation springs must be included:
o Vertical spring stiffness - ;
o Horizontal spring stiffness - ;
o Rotational (rocking) spring stiffness - ;
o Torsional spring stiffness - .
It is important to note that soil behaves in a non-linear manner. But instead of using non-linear
springs, it is common to model the springs according to the strain level of the soil and then using
springs with linear behaviour. DNV (Det Norske Veritas (DNV), 2002) suggests three different levels
of dynamic loading (and corresponding shear strains ):
o From earthquakes: large shear strains up to 10-2 to 10-1;
o From wind and ocean waves: moderate shear strains up to 10-2, typically 10-3;
o From rotating machines: small shear strains, usually less than 10-5.
105
After deciding which load case is of interest, it is necessary to compute the shear modulus of the soil.
This is usually accomplished through the use of the plot presented in Figure 4.8 together with the
initial shear modulus . The value of can be obtained according to the empirical expression:
3
′ (4.6)
1
with:
Initial shear modulus (in kPa)
Coefficient depending on material (size, angularity of grains, …) - 3000 1000
Void ratio
′ Effective stress (in kPa)
Over consolidated ratio for clay soil (equal to 1.0 for sand)
Coefficient depending on the plasticity index (according to Table 4.1)
Once the shear modulus of the soil is obtained, the Young modulus of soil is calculated according to:
2. 1 (4.7)
with:
Young modulus
Poisson’s ration (according to Table 4.3)
Table 4.1 – Coefficient k Table 4.2 – Coefficient K Table 4.3 – Poisson ratio
Plasticity Index
k Type of soil K Type of soil ν
(I )
0 0 Loose sand 8 Dense sands 0.25 – 0.30
20 0.18 Dense sand 12 Loose sands, stiff

0.35 – 0.45
40 0.30 Very dense sand 16 clays
60 0.41 Very dense sand and Saturated clays ≃ 0.50

30 - 40
80 0.48 gravel
> 100 0.50
106
Chapter 4
Two simpler alternative expressions for the determination of can be used for sand:
1000. . ′ (4.8)
with:
Coefficient depending in the type of soil (according to Table 4.2)
and for clay:
2600. (4.9)
with:
Undrained shear strength
After computing the value of the initial shear modulus, the value of shear modulus ( ) and damping
( ) of the soil can be obtained with the help of Figure 4.8. Once these values are calculated, the stiffness
of the foundation springs for circular slab footing are given by the equations presented in Table 4.4.
1.00
0.80 0.20
Typical
Range
G( γ)/G 0
0.60 0.15
ζ(γ)
0.40 0.10
0.20 0.05
0 0
10-5 10-4 10-3 10-2 10-1 10-5 10-4 10-3 10-2 10-1
γ γ
Figure 4.8 – Shear modulus and damping ratio according to strain level (adapted from (Det Norske Veritas (DNV), 2002))
It is important to refer that the values obtained from these tables are related to static stiffness (for
excitation frequencies approaching zero). This assumption is correct for majority of the dynamic
loading acting on wind turbines, although may introduce some errors for high frequency excitations.
After the foundation stiffness is calculated, springs should be introduced in the wind turbine model
according to Figure 4.9 in order to more accurately reproduce the boundary conditions between
structure and soil.
107
KR
KT
KH
KV
Figure 4.9 – Illustration of the foundation springs
108
Chapter 4
Table 4.4 – Stiffness of foundation springs for circular footing on different substratum conditions (adapted from (Det Norske Veritas (DNV), 2002))
R R R
D
G
Foundation
H ν
G G1
H H ν1
spring ν
Bedrock
G2
Bedrock ν2
D/R < 2
0 ≤ G1/G2 ≤ 1 D/H < 1/2
4. . 4. . 1 1.28. 4. . ⁄
Vertical 1 1.28 . ;1 ⁄ 5 1 1.28. 1 1 0.85 0.28.
1 1 1 1.28. . 1 2. 1 ⁄
4. . 8. . 1 8. . 2. 5.
. 2.
Horizontal 1 1.28 1 ;1 ⁄ 4 1 1 1
1 1 . 1 2. 3. 4.
2.
8. . 8. . 1 8. .
. 6.
Rotation 1 3 1 ;0.75 ⁄ 2 1 1 2. 1 0.7.
3 1 6. 1 . 3 1 6.
6.
16. . 16. . 8.
Torsion - 1
3 3 3.
109
4.3 TOWER-ROTOR DYNAMIC COUPLING

The previous characterization of the tower and rotor components (see sections 4.1.1 and 4.1.2) permits
to analyse possible situations of dynamic coupling between these components. This analysis is of high
importance since towers are becoming increasingly flexible.
Under operation, blades suffer with the action of the wind, besides cyclic gravitational forces. The
resultant forces of the blades are then transmitted to the tower. However, when the forces “reach” the
tower, they were already “filtered”. In other words, considering only the wind loading, the spectral
content of the wind changes when passing through the rotor. It is changed, in a larger extent, by the
rotation of the blades, and, in a smaller fraction, by the dynamic properties of the blades.
In a situation of a rotor diameter sufficiently larger than the gusts size, the distribution of the wind
energy spectrum when experienced by a rotating blade can differ significantly from the spectrum of
the same wind observed by a fixed object (Connel, 1980). This alteration is reflected on a shifting of
energy from lower frequencies to higher frequencies of the spectrum, with peaks located at integer
multiples of the rotational frequency of the rotor (Figure 4.10) (Murtagh, Basu et al., 2005). This
situation is explained by the fact that when the blades are spinning, they pass through a gust once per
rotor revolution (if the gust is large enough) and, consequently, suffer a load peak with a frequency
value of Ω. For the support structure (like the tower), the excitation will thus be . Ω (3Ω for the
common three-bladed wind turbine) since, within the period of 1/Ω, the same gust will be transversed
times.
This “changed” wind turbulence spectrum will act on the rotor blades and, subsequently, will be
detected on the response of the tower (see section 4.6).
Ampltiude
1Ω 2Ω 3Ω 4Ω
Log Frequency
Figure 4.10 – Illustration of the spectral energy distribution for rotationally sampled spectrum. The peaks refer to integers
of the rotor frequency of rotation (adapted from (Murtagh, Basu et al., 2005))
110
Chapter 4
4.4 DEFINITION OF TOWER STIFFNESS

In Chapter 2, during the introduction about the historic developments of the wind turbine technology,
reference was made about the different design concepts of the tower. In fact, the tower stiffness is a
major consideration in the overall design of a wind turbine. In a wind turbine system, it is imperative
to avoid a match of the natural frequency of the tower with the frequency of rotor rotation in order to
prevent resonant problems. Thus, it would be desirable to have very stiff towers. In this situation,
resonant situations would be avoided, since the frequency of rotation of the rotor (and its harmonics)
would be considerable lower than the tower natural frequency. This was the strategy followed in the
MOD-0 and WTS-75 models in the past (Hau, 2006). However, the increase of the size of the rotors
and the need to reduce costs led to the need of reduction of the tower weight. Thus, a progressive
increase in the tower flexibility led to the usually defined “soft-soft” towers. These soft-soft towers are
characterized by a natural frequency lower than the rotor rotational frequency. This was the design
strategy followed for the MOD-2 and WTS wind turbines. There is also an intermediate solution
(named as “soft-stiff”), where the tower natural frequency is tuned in order to be placed between the
rotational frequency (1Ω) and its first integer multiple (2Ω or 3Ω, for, respectively, two- or three-
bladed wind turbines). In Figure 4.11 a simple scheme is presented with these three configurations.
Soft-Soft Soft-Stiff Stiff-Stiff
1Ω 3Ω
Rotor Rotational Frequency
Figure 4.11 – Frequency intervals for tower stiffness for a fixed rotational speed wind turbine
Figure 4.11 refers to a situation of a fixed-speed wind turbine. For these cases, since there is only one
operational rotational speed, the range of possible values for the tower natural frequency is wider.
However, in the case of variable-speed wind turbines, further considerations had to be taken into
account. In these situations, the range for the tower natural frequency becomes much narrower. With
the purpose of a clearer explanation, a Campbell diagram of a 1.5 MW three-bladed variable-speed
wind turbine is presented in Figure 4.12. In it, the natural frequencies of several components of the
wind turbine are plotted against the rotational speed. The range of operation of the rotor is also shown.
As can be seen, the first natural frequency of the tower remains above the rotational speed (1Ω) and
below the integer multiple 3Ω. Thus, this is considered a soft-stiff tower.
111
Natural and rotational

frequenc y in Hz
Operating range
4 1 st Tower torsion
9Ω
2 nd Flapw
Flapwise bending 8Ω
3 7Ω
2 nd To wer bending 6Ω
2 1 st Edgewise bending 5Ω
4Ω
3Ω
1 1 st Flapw.
2Ω
bending
1st Tower Longitudinal 1Ω
bending Lateral
Rotational speed
in rpm
0 5 10 15 20 25
0 0.1 0.2 0.3 0. 4 in Hz
Figure 4.12 – Campbell diagram of a 1.5 MW three-bladed rotor, variable-speed wind turbine (adapted from (Gasch and
Twele, 2012))
112
Chapter 4
4.5 DAMPING
In the dynamic characterization of the wind turbine system, other important parameters that need to
be determined, besides the value of the natural frequencies, are the damping coefficients. With the aim
of illustrate the importance of these parameters on the supporting structure, a simple single degree of
freedom (SDOF) structure, defined by its mass , stiffness and damping , is presented in Figure
4.13 a). Its dynamic amplification factor (DAF – which represents the magnification the structure
response suffers with the application of a dynamic load relative to the response to a static force) is
defined by (Clough and Penzien, 1995):
(4.10)
1 2. .
with:
Applied load frequency
Natural frequency of the structure
Damping ratio of the structure
In Figure 4.13 b), the DAF for a simple SDOF structure was calculated for two different values of
damping ratios. As can be observed, the DAF value is highly dependent on the damping value. The
value of the structural response in resonance is considerably higher for a structure with a damping
ratio of 0.2 % (a common value for a wind turbine in an idling or parked situation) than for a structure
with = 4 % (a conservative value for a wind turbine under production). For this reason, it is easy to
understand that, for structures subjected to dynamic excitations (such as wind turbines), their
damping value is a key-factor to handle with the problem of fatigue.
The damping of a wind turbine can be considered as linear sum of different damping sources:
structural, aerodynamic, from devices, soil and hydrodynamic. From their nature, the last two are only
referred to offshore wind turbines.
3
10
2
10 ξ = 0.2 %
1
DAF
10
ξ = 4.0 %
0
10
1
-1
10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
f /f
load 1
a) b)
Figure 4.13 – a) SDOF structure. b) DAF values for different damping ratios
113
4.5.1 STRUCTURAL DAMPING

Structural damping is referred to the portion of damping that is intrinsic to the structure. It is due to
the absorption of vibrations by internal friction and conversion into heat.
Structural damping depends on the type of material (e.g. concrete or steel) and on the type of
contacting surfaces (e.g. types of connections in the structure). Also, offshore wind turbines usually
present a higher damping value than onshore models, mainly due to the grouted connection (Tarp-
Johansen, Andersen et al., 2009) or higher complexity of the foundation. Thus, different tower
concepts and connections lead to different damping values.
Typical values of structural damping ratios are in the order of 0.2 % to 0.8 % (Det Norske Veritas
(DNV), 2002; Tarp-Johansen, Andersen et al., 2009).
4.5.2 AERODYNAMIC DAMPING

One of the most important components of damping is the aerodynamic contribution. Aerodynamic
damping has its origin in the wind load acting on the rotor or, more accurately, in the interaction
between the wind flow and the motion of the structure. This phenomenon, which occurs for flexible
structures, can be a source of instability under certain situations. However, for the along-wind
direction, it is favourable and should be accounted for during the design stage and the evaluation of
structure integrity along its lifetime. This interaction between wind flow and wind turbines can be
explained through the simple description presented in (Kuhn, 2001):
“A motion of the tower top in the wind direction (…) results in a smaller angle of attack at the rotor
blades [see Figure 3.33] because the apparent out-of-plane velocity component [U 1 a ] is reduced by
the tower top velocity. A smaller angle of attack corresponds, for attached flow conditions, to lower
aerodynamic lift and drag forces [equations (3.24) and (3.25)] and to a reduction of the thrust force (…).
Likewise a movement of the nacelle against the wind direction increases both the angle of attack and the
thrust force. In both situations the alternation of the thrust is oriented opposite to the disturbing tower
top motion and is experienced as aerodynamic damping. (…)
For higher angles of attack, stall occurs and the dynamic damping is lower or may even become negative
because the slope of the lift curve is reduced. Under such conditions aero-elastic instability can occur if
insufficient damping exists.”
For the purpose of illustration, Figure 4.14 presents the experimental results for lift and drag
coefficients for the NACA 63-415 airfoil obtained in a wind tunnel by Bak et al (Bak, Fuglsang et al.,
2000). As can be seen, the lift coefficient has an almost linear increase for the initial range of angle of
attack. Then, the slope of the curve starts to decrease until a peak is reached. After this point, the
variation of the lift coefficient with the angle of attack becomes negative. The values of drag
coefficients for angle of attack of around 15º are considerable low.
114
Chapter 4
2.0 0.5
1.5 0.4
1.0 0.3
CD
CL
0.5 0.2
0.0 0.1
-0.5 0.0
-10 0 10 20 30 -10 0 10 20 30
α[º] α[º]
Figure 4.14 – Variation of lift ( ) and drag ( ) coefficients with the angle of attack (adapted from (Bak, Fuglsang et al.,
2000))
Garrad (Kuhn, 2001) presented a simple equation (4.11) for the calculation of the aerodynamic
damping for a wind turbine support structure. It assumes that the wind turbine is operating under
stationary rotor dynamics at a high tip speed ratio, where the inflow angle ( ) is small and the drag
forces are negligible compared to lift forces (Freris, 1990; Salzman and Tempel, 2005).
. .Ω d
. . d (4.11)
8. . . d
with:
Frequency of the first fore-aft vibration mode
Modal mass of the first fore-aft mode
Length of the blade
Airfoil chord
From the previous equation, some important aspects should be highlighted. The aerodynamic
damping is inversely proportional to the natural frequency (and modal mass of the structure). This
mean that structures with different philosophies of tower stiffness design (as presented in section 4.4)
will lead to very different values of aerodynamic damping. On the other hand, the increase of the
rotational speed of rotor increases the value of this additional portion of damping.
Some other important aspects should be also mentioned:
o The aerodynamic damping is specially noted for the first bending mode of the tower in the
rotor out-of-plane direction (the first fore-aft vibration mode). This is expected since the
aerodynamic forces act mainly in this direction;
o In the rotor plane direction (the side-side direction) the effect of the aerodynamic damping is
expected to be negligible since there is a very low level of aerodynamic forces in this direction;
o In idling or parked situations, the effect of aerodynamic damping is expected to vanish for
both orthogonal directions. This situation is due to the fact that, in these situations, the blades
are usually in a feathered position (with high angles of attack). Nevertheless, some effect could
115
be noted for the side-side direction due to the opposition of the blade to the wind flow in this
direction;
o Variable and fix rotor speed turbines present different values for aerodynamic damping
(Salzman and Tempel, 2005). Fix speed turbines operate at high tip speed ratios in low wind
speeds and lower tip speed ratios at high wind speeds (due to the constant rotor speed). For
these reasons, in low wind speed the angle of attack is small and, consequently, the flow is
attached to the blade. When the wind speed is higher, the angle of attack increases which lead
to higher angles of attack and some separation of the flow at the blade. In the case of variable
speed rotors, since they operate with their blades close to stall under rated wind speed, there is
some flow separation which leads to an aerodynamic damping below the optimal value. For
the higher wind speed, the flow stays attached to the blade and the damping is higher.
4.5.3 TOWER DAMPERS

One artificial method for additional introduction of damping in the support structure can be achieved
with the installation of special devices, typically in the form of a mass pendulum. This technique is a
widely implemented solution in tall structures. Ideally, it consists in the attachment of an additional
mass, linked to the structure through a spring and a damper (Figure 4.15). The damper system is tuned
according to the vibration mode which response is desired to be attenuated. Commonly, solutions
consisting in a pendulum immersed in high viscous oil (Damgaard, Ibsen et al., 2013a), a tuned mass
damper (Shirzadeh, Devriendt et al., 2013) or a liquid damper (Tarp-Johansen, Andersen et al., 2009)
are used in wind turbines, usually installed offshore.
The damping value introduced by these devices depends on the implemented damper system and on
the tuning accuracy. In (Colwell and Basu, 2009), a numerical study about the efficiency of a tuned
liquid column damper in an offshore monopile wind turbine is presented.
k1 k2
m1 m2
c1 c2
Figure 4.15 – Idealized structural system with the damper system (index 2) attached to the main system (index 1)
4.5.4 SOIL DAMPING

The operation of a wind turbine introduces vibrations at the ground level, causing cyclic motion of the
surrounding foundation soil. In offshore wind turbine, the effect of the soil in the introduction of
damping should be analysed. It is considered one of the most complex components of damping to be
determined (Germanischer Lloyd (GL), 2005). Soil damping is formed by a combination of two parts:
geometrical dissipation and material damping. Geometrical dissipation is due to wave propagation
into the soil. However, it is referred that this effect is insignificant at frequencies below 1 Hz
(Andersen, 2010). On the other hand, material damping is due to the internal friction of soil grains
and should be accounted for.
116
Chapter 4
The damping introduced by the soil is dependent on the type of soil and also, to some extent, on the
type of foundation. Along with the inherent complexity of this phenomenon, it is comprehensible that
there is a quite wide range of values between published works. Damping ratio values between 0.23 %
and 1.00 % for monopile foundations are referred in (Tarp-Johansen, Andersen et al., 2009;
Versteijlen, Metrikine et al., 2011; Damgaard, Ibsen et al., 2013a). Damgaard et al (Damgaard, Ibsen et
al., 2013b) also studied the influence of soil damping in an offshore prototype wind turbine with a
bucket foundation and found a value of 0.16 % for the damping ratio during power production.
4.5.5 HYDRODYNAMIC DAMPING

In offshore wind turbines, there is also another component of damping, the hydrodynamic damping.
This component is essentially constituted by two terms: one due to wave radiation and another due to
hydrodynamic drag. In order to assess the relative importance of each term, the value of the Keulegan-
Carpenter number is usually analysed:
.
(4.12)
with:
Flow velocity amplitude
Wave period
Characteristic length (diameter of cylinder section for a pile)
For small values of (below 2 (Naess and Moan, 2012)), the radiation damping is dominant; while
for high value of , the drag component prevails. Thus, the contribution of each damping term
depends on the type and size of the foundation solution (and also on the sea conditions).
Some authors refer the larger contribution of the radiation damping with respect to the viscous drag
component for monopile wind turbine foundations (Tarp-Johansen, Andersen et al., 2009; Shirzadeh,
Devriendt et al., 2013). Some typical values for the wave radiation component of the hydrodynamic
damping are present in published works. Leblanc and Tarp Johansen (LeBlanc and Tarp-Johansen,
2011) present a value of 0.12 % for the damping ratio of an offshore wind turbine with a pile diameter
of 4.7 m, with a water depth of 20 m and a natural frequency value of 0.3 Hz. Also in (Germanischer
Lloyd (GL), 2005), the value of 0.11 % for damping ratio of a monopole wind turbine is presented.
Likewise, Tarp-Johansen et al (Tarp-Johansen, Andersen et al., 2009) present a value of 0.22 %,
following a numerical procedure.
Besides the hydrodynamic effect of the pile, the presence of mooring lines in floating wind turbines
also represents an important contribution to the damping of the system (Hall, Buckham et al., 2013).
117
4.6 CYCLIC VIBRATION LOADS

As mentioned earlier, wind turbines are subjected to a large variety of dynamic loads (Manwell,
McGowan et al., 2010). In this section, a special reference is made to cyclic loads. These loads hold a
large importance when studying the response of the structure. With a dynamic monitoring system, it is
expected to capture the resonance peaks of the structure in order to identify its natural frequencies.
However, as cyclic loads are acting on the structure, some peaks will be present in the records that do
not correspond to the dynamic properties of the structure. For this reason, there is the need for
previous identification of these peaks. There are mainly three sources of this kind of loads:
aerodynamic forces, mass unbalance or wave forces.
In Figure 4.16, the frequency domain spectrum response (acceleration) at the top of a three-bladed,
fixed-speed wind turbine (obtained in a numerical analysis) is presented. This turbine was subjected to
a three-dimensional turbulent wind field. As can be seen, several peaks spaced by 3Ω (highlighted by
the red dashed line) are easily detected.
0
10
Tower 1st natural frequency
-2
10
Amplitude
-4
10
-6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frquency [HZ]
Figure 4.16 – Frequency domain spectrum of the acceleration of the top of the tower of a wind turbine due to turbulent
wind inflow
In the situation of the Figure 4.16, the cyclic loads are only due to aerodynamic forces of three types:
o Rotational sampled turbulence spectrum: As already explained in section 4.3, for each
revolution, each blade passes through a gust once. This will lead to an excitation of 1Ω for the
blades and of .Ω for the structures that support the rotor (3Ω in the case of Figure 4.16);
o Tower shadow: For an up-wind rotor, the presence of the tower disturbs the wind flow,
retarding the wind speed. The decrease in the wind speed leads to a decrease of the force on
the blade every time it passes in front of the tower, causing a cyclic excitation on the blade
(and, consequently, on the tower). For down-wind rotors, the effect of the tower is more
severe. The retardation of the flow is felt at larger distances and lead to strong excitations;
o Vertical wind shear: The mean wind velocity increases with height (Figure 4.17). With today’s
large rotor diameters, blades are long enough to experience a reasonable variation of wind
speed from top to bottom position. Thus, blades will complete a cycle from the maximum to
the minimum mean wind load in each rotor revolution.
118
Chapter 4
Figure 4.17 – Vertical wind shear
Unbalanced rotors represent another source of cyclic loads. It leads to a rotating force that can excite
lateral vibrations of the nacelle and tower (Gasch and Twele, 2012). As examples of causes of
unbalanced rotors, manufacturing defects and accumulation of ice can be pointed out. In Figure 4.18, a
comparison of the response between an unbalanced mass rotor and a balanced rotor is shown. The
time history was obtained through a numerical analysis, where the unbalanced rotor was defined by an
abnormal mass at the tip of one of the blades with the purpose of evidence the described phenomenon.
As can be observed, the unbalance presented in the rotor introduced an excitation with a frequency of
1Ω in the support structure.
2
10
Balanced rotor
Unbalanced rotor
0
10
Amplitude
-2
10
-4
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 4.18 – Time history of the tower top accelerations for a balance and unbalance mass rotor
For offshore wind turbines, another source of excitation forces needs to be accounted for. Wave
excitation acts on support structures in a wide range of frequencies, usually bellow the tower natural
frequency (Tempel, 2006). Thus, time histories of the motion of an offshore tower will likely record
these excitations, which need to be analysed and discharged when a dynamic continuous monitoring
system is installed in a wind turbine. The usual wave excitation range of frequency is illustrated in
Figure 4.19. As can be seen, the frequencies of excitation are not very well defined as are for
aerodynamic forces.
119
100
f1
80
Ocurrence [%] 1Ω 3Ω
60
40
20
0
0 0.2 0.4 0.6 0.8 1.0 1.2
Frequency [HZ]
Figure 4.19 – Range of frequencies of wave excitation on an offshore wind turbine (Tempel, 2006)
120
Chapter 4
4.7 NUMERICAL MODEL OF A WIND TURBINE

A numerical model of a wind turbine was developed with the HAWC2 software (Larsen and Hansen,
2007) in order to illustrate some of the methods presented throughout this thesis. HAWC2 is an
aeroelastic code developed at Risø DTU and intended to calculate wind turbine response in the time
domain. It uses a multibody formulation to model the wind turbine structure through beam elements,
allowing to model each component of the wind turbine as an independent body. Each body is
implemented in the model according to the finite-element theory.
HAWC2 code allows simulating the structural response of wind turbines at onshore and offshore
environments, subjected to hydrodynamic and aerodynamic loading. Since the main case study of this
work is an onshore wind turbine, only aerodynamic loads were considered. The aerodynamic model
implemented in the HAWC2 code is based on the blade element theory (briefly introduced in Chapter
3). The implemented model is extended to account for the dynamic and skew inflow, wind shear effect,
the effect due to large deflections of the blades, tip loss effect and dynamic stall. In addition, the tower
shadow effect, referred in section 4.6, is also included.
The developed numerical model of the onshore wind turbine is based on the NREL 5MW reference
wind turbine (Jonkman, Butterfield et al., 2009), which is available on (HAWC2 web site, 2013). This
model was used as a basis for the numerical model described in the case study of the section 7.3.
The NREL 5MW reference wind turbine is a conceptual model of a representative utility-scale wind
turbine. It is a three-bladed, upwind rotor with a variable-speed turbine and with pitch control
mechanism. The power curve and the relationship between the wind speed and the rotor speed of the
NREL 5MW wind turbine is presented in Figure 4.20.
The main dimensions of the model are presented in Figure 4.21, alongside with the position of the
sensors considered to collect the accelerations of the structure.
5000 12
126 m
+ 87.600
4000 S3
9
Power Output [kW]
Rotor speed [rpm]
3000 S2
+ 60.000
6
2000 90 m
S1
3 + 30.000
1000
0 0
0 5 10 15 20 25 + 0.000
Wind speed [m/s]
Figure 4.20 – Power curve (in blue) and relationship Figure 4.21 – Illustration of the numerical model of the
between rotor speed and wind speed (in red) of the NREL NREL 5MW and position of the accelerometers
5MW
The aeroelastic analysis was performed for a time length of 700 s, with a time step of 0.02 s. The first
100 s were disregarded to avoid transient responses from the start of operation of the turbine.
121
During the analysis, the wind turbine was subjected to artificially generated wind speed time series
with a mean wind speed of 15 m/s (Figure 4.22). The turbine was operating at the rated power output
conditions. At this regime, the pitch angle is used to control the rotor speed and, for that reason, it
varies along the analysed 600 s. The rotor mean speed is 12.1 rpm. The evolution of the rotor speed
and of the blades pitch angle throughout the analysis is shown in Figures 4.23 and 4.24.
Wind speed at hub height 6 Wind speed at hub height
17 10
16 4
10
Wind spedd [m/s]
Amplitude
15
2
10
14
0
10
13
-2
12 10
100 200 300 400 500 600 700 0 1 2 3 4 5
Time [s] Frequency [Hz]
Figure 4.22 – Generated wind speed at the hub height (time and frequency domain)
Rotor speed Blades pitch angle

12.3 12
12.2 11
Rotor speed [rpm]
12.1 10
Pitch angle [º]
12 9
11.9 8
11.8 7
11.7 6
100 200 300 400 500 600 700 100 200 300 400 500 600 700
Time [s] Time [s]
Figure 4.23 – Rotor speed Figure 4.24 – Blade pitch angle
The acceleration time series obtained from the analysis of the considered sensors are presented in
Figure 4.25, in both time and frequency domain. In this figure, only the considered 600 s are shown. It
is interesting to note that the vibration levels recorded by sensor S2 (height + 60.000) are the highest
from the three sensors.
In the frequency domain plots, the frequencies of the 3Ω, 6Ω, …, 24Ω harmonics are indicated with
vertical dashed lines. The influence of the first 5 harmonics in the structural response of the wind
turbine is clearly visible.
122
Chapter 4
Sensor S1 3 Sensor S1
10
0.2
2
10
Acceleration [m/s2]
0.1
1
Amplitude
10
0
0
10
-0.1
-1
10
-0.2
-2
10
100 200 300 400 500 600 700 0 1 2 3 4 5
0.4 10
0.3
2
10
0.2
Acceleration [m/s2]
0.1 1
Amplitude
10
0
0
-0.1 10
-0.2 -1
10
-0.3
-2
-0.4 10
100 200 300 400 500 600 700 0 1 2 3 4 5
10
0.2
2
10
Acceleration [m/s2]
0.1
1
Amplitude
10
0
0
10
-0.1
-1
10
-0.2
-2
10
100 200 300 400 500 600 700 0 1 2 3 4 5
Figure 4.25 – Wind turbine acceleration captured by the sensors (time and frequency domain)
In order to ease the comprehension of some procedures presented throughout this thesis, only the two
first tower vibration modes were considered. The dynamic properties of the wind turbine were
obtained through an eigensolver implemented in the HAWC2 code. The natural frequency and the
modal damping ratio of these two modes are presented in Table 4.5. It is important to note that the
damping values presented are only referred to the structural component of damping and, for that
reason, do not necessary reflect the effective damping of the modes during the analysis since the
aerodynamic damping is not considered in the eigensolver and, consequently, in the values of shown
in Table 4.5.
123
Table 4.5 – Modal properties of the HAWC2 model
Natural frequency Damping ratio

Mode
[Hz] [%]
1st tower bending mode

0.333 0.24
(1 FA)
2nd tower bending mode

2.686 1.91
(2 FA)
The configuration of the considered vibration modes is illustrated in Figure 4.26. The 1st tower
bending mode is clearly a pure tower bending mode, with a first order configuration of the tower
motion (Figure 4.26 a)). On the other hand, the 2nd tower bending mode beholds, besides the tower, an
important contribution of the rotor to the global motion of the mode (Figure 4.26 b)).
a) d)
Figure 4.26 – Illustration of the considered vibration modes: a) 1 FA; b) 1 FS and c) 2 FA
124
Chapter 4

In this chapter, a dynamic characterization of the wind turbine structural system is developed. It starts
with a description of the tower and rotor blades dynamic behaviour, including the commonly
identified vibration modes. In addition, the importance of the soil condition to the dynamic properties
of the structure is also highlighted. These properties of the support structure are usually used to define
the type of wind turbine based on its stiffness. It can be defined as “soft-soft”, “soft-stiff” and “stiff-
stiff”, depending on the relationship between the frequency of the first natural frequency of the tower
and the rotor speed. This analysis is fundamental to avoid resonance phenomena in the wind turbine
structural systems.
Afterwards, a characterization of the damping in wind turbines is performed. This is a property of
utmost importance in the dynamic analysis of these structures since it is directly involved in the wear
caused by fatigue loading. The several components of the damping that might be present in a wind
turbine are referred, namely structural, aerodynamic, from external devices, soil and hydrodynamic.
The influence of the rotor blades rotation and rotor unbalances in the load spectrum of the support
structure is also described. These two phenomena modify the wind loading spectrum, imposing well-
defined excitations around the rotation harmonic frequencies of the rotor. The influence of these
imposed loading on the identification of the modal properties of wind turbines is addressed in the next
chapter.
Finally, a numerical model of the NREL 5MW used in the context of this thesis is described. A
numerical simulation subjecting the model to an artificially generated wind speed time series is
presented. The response obtained with the analysis will be used, in the following chapters, to illustrate
some concepts introduced in this thesis.
125
126
Chapter 5
5
OPERATIONAL MODAL ANALYSIS OF
WIND TURBINES
5.1 INTRODUCTION TO OPERATIONAL MODAL ANALYSIS

The experimental identification of modal properties of structures is a subject of study with a long
tradition. It was initially applied to small structures of mechanical engineering (most of the times in
laboratory environment), in order to estimate modal frequencies, damping ratios and mode shapes. In
these tests, an input force was applied to the structure that reacts with a particular response (output).
In this type of traditional Experimental Modal Analysis (EMA), both the input and the output were
measured in order to identify the modal properties of the structure. For this reason, this type of modal
identification procedure is usually named as input-output method.
Later, EMA techniques were applied to larger civil engineering structures, like bridges, dams and
buildings (Magalhães and Cunha, 2011). However, the transposition from small to large structures
introduced critical difficulties regarding the excitation of the structure. Heavy and expensive means are
often necessary to perform shaker, impact or pullback tests used for this kind of procedure (Felber,
1993). Some examples of application of EMA on large structures are presented in (Cunha and Caetano,
2006).
In the early 1990’s, a remarkable evolution in the field of experimental identification gained great
attention (Zhang, Brincker et al., 2005). With the technological developments of sensors and A/D
converters, it became possible to measure lower levels of vibration alongside with lower levels of noise.
Thus, sources of operational (like traffic) and ambient (like wind) excitation became sufficient to
analyse the dynamic response of the structure. For this reason, new identification methods requiring
only the measurement of the structure response were developed. These were named output-only
methods. Compared to EMA, this new philosophy of experimental identification, usually also called
Operational Modal Analysis (OMA) or Ambient Vibration Test, presents several advantages (Zhang,
Brincker et al., 2005):
o OMA is cheaper and faster to conduct, avoiding the necessity of large masses or heavy
excitation devices;
o Dynamic characteristics of the complete system under real loading conditions can be obtained;
o OMA can be used not only for dynamic design and structural control but also for vibration-
based health monitoring and damage detection.
127
Operational Modal Analysis of Wind Turbines
Nevertheless, OMA presents some disadvantages inherent to its concept (Magalhães and Cunha,
2011):
o Although the excitation is assumed to be of broad band nature, the frequency content may not
cover the whole spectrum of interest, mainly in the case of structures with high natural
frequencies. Also, if the excitation does not correspond to a flat distribution of energy along
the frequency spectrum, some misidentification might occur;
o The modal mass is not estimated.
OMA has been widely used in the last years in several engineering domains. In civil engineering, this
technique has been applied in diverse structures. Bridges (Cunha, Caetano et al., 2001), (Caetano,
Magalhães et al., 2007), (Brownjohn, Magalhães et al., 2010), dams (Rodrigues, 2004)) and buildings
((Ventura, Finn et al., 2003), (Shi, Shan et al., 2012), (Gentile and Saisi, 2007) are among the most
commons examples of structures tested with OMA methodology. In mechanical engineering, OMA is
also applied in different structures, such as helicopters (Peeters, Cornelius et al., 2007), (Ameri,
Grappasonni et al., 2013), aircrafts (Mevel, Benveniste et al., 2006), (Hermans and Auweraer, 1999) or
vehicles (Goursat, Döhler et al., 2010), (Hermans and Auweraer, 1999).
128
Chapter 5
5.2 APPLICATION OF OMA TO WIND TURBINES

The dynamic behaviour of wind turbines is a major concern, as exposed in the previous chapters.
Thus, an experimental dynamic characterization of the system became a required tool in order to
prevent possible resonance effects.
Wind turbine presents three important particularities that highlight the importance of application of
OMA in these structures:
o Considering the typology and size of wind turbines, it is easy to understand the difficulty in
providing means to adequately excite these structures;
o Wind turbines are located in windy locations. Thus, wind can be used as an ambient source of
excitation (with a broadband spectrum);
o It is important to adequately identify the modal characteristics of the structural system, both in
parked and in real operation conditions, since rotor operation can introduce important
variations in these properties.
Therefore, it is not surprising that modal identification of wind turbines played an important role in
the development of OMA.
The first formalized modal testing of blades and turbines began in the late 1970s (Carne and James Iii,
2010). In 1982, (Carne, Lobitz et al., 1982) developed a finite element procedure to be applied to
vertical-axis wind turbines. In order to validate the procedure, an input-output test was performed in a
2 m height, research vertical axis wind turbine at parked condition and at different operating speeds.
In the parked test, the excitation was achieved through a traditional impact hammer and
accelerometers were used as sensors. For the tests realized under operation, it is referred that there
were some difficulties in exciting the structure (Carne and James Iii, 2010). The solution adopted
consisted on a pretensioned cable attached between the tower and one blade. When the cable was
suddenly released, it created a broad-brand step-relaxation excitation (Carne and Nord, 1983). In this
test, accelerometers were also used, alongside with strain gauges. The identified modal parameters
consisted in frequency values, damping ratios and mode shapes, and the correlation obtained between
numerical and experimental results was very good. However, the procedure for the step-relaxation
technique was very time-consuming because, every time a new test was performed, the turbine had to
be brought to parked condition.
Later in 1986, Carne et al performed a modal test of the 110 m height Éole VAWT in parked
conditions (machine introduced in Chapter 2) (Carne, Lauffer et al., 1988). Since it was extraordinary
higher than the 2 m height VAWT previously tested, a different approach was followed in this test: a
traditional input-output test, where the excitation consisted in two forces of 45 kN and 135 kN applied
on one blade and on the tower, respectively; and a second test in which the wind turbine was only
subjected to wind loading and only the structural response was recorded. The later consisted in one of
the first ambient vibration tests in which a full modal analysis was performed, including mode shapes,
frequencies, and damping values (Carne and James Iii, 2010). In this test, four separate setups of
response measurements were recorded, with four reference channels repeated in every setup. Then,
auto- and cross-spectral densities were computed in order to extract the modal properties of the Éole
turbine. The obtained results showed a very good correlation between the two experimental tests
129
(Carne, Lauffer et al., 1988), evidencing the advantages of ambient vibration tests (considerably faster
and more practical) with regard to forced vibration tests.
The success achieved with the previous application led to the development of a new identification
method named Natural Excitation Technique (NExT) (James, Carne et al., 1993). This technique was
further validated and used in numerical (FloWind Corporation 19 m VAWT) and experimental tests,
both in parked - FloWind Corporation 19 m VAWT - and in operation – DOE/ Sandia 34 m VAWT -
situations (James, Carne et al., 1993). Reference should also be made to the application of NExT to a
rotating 100 kW horizontal axis wind turbine (James, 1994).
After these initial tests, other authors have also performed some modal identification analysis on wind
turbine structures. Molenaar (Molenaar, 2003) performed an experimental modal test on a 750 kW
horizontal axis wind turbine, under parked conditions, in order to compare the results with a
numerical model. For the test, the wind turbine was excited through the step-relaxation technique by
applying a static load on the tower top and the structural response was measured with 19
accelerometers mounted both on tower and blades.
In (Osgood, Bir et al., 2010), continuing earlier works (Osgood, 2001), a parked 600 kW three-bladed
horizontal axis wind turbine was experimentally tested following both forced and ambient vibration
test procedures. For the forced vibration test, two independent hydraulic actuators were used, which
excited the structure at the top in both orthogonal directions. The turbine was heavily instrumented
with accelerometers, totalling 75 channels distributed in the tower, rotor and drive train. A very good
correlation was obtained with the two tests (although it was possible to detect additional modes in the
forced vibration test).
(Hansen, Thomsen et al., 2006) explored different excitation techniques for wind turbine testing.
Experimental results obtained with two different types of harmonic excitation were confronted with
the results from an ambient vibration test. The harmonic excitation was induced by blade pitch angle
variation and by electrical torque variation. However, it was concluded that the excitation of wind
turbines with these techniques presents some problems, namely:
o The excitation of high-frequency and highly damped modes is not technically possible due to
limitation of the blade pitch actuators;
o The excited structure vibrations are not pure modal vibrations and the estimated damping is
therefore not the actual modal damping. Since it is not possible to isolate the desired modes,
the estimation of damping for modes with close frequencies will result in considerable errors.
In the opposite way, the authors (Hansen, Thomsen et al., 2006) concluded that the results obtained
with the adopted output-only tool confirmed its suitability for identifying closely-spaced modes.
Griffith et al (Griffith, Mayes et al., 2010) conducted a modal identification test on a small 60 kW
VAWT under parked conditions. As in the above references, an ambient vibration and forced
vibration (excitation from an impact force, from step-relaxation and from human random excitation)
were performed and both accelerometers and strain gauges were used. The results revealed good
agreement between all the tests.
Chauhan et al (Chauhan, Tcherniak et al., 2011) and Carcangiu et al (Carcangiu, Tcherniak et al.,
2012) experimentally studied a 3 MW, 100 m diameter wind turbine. In this work, OMA techniques
were applied in order to extract the main vibration modes of the tower and rotor. For that purpose,
130
Chapter 5
several accelerometers were placed on the tower, nacelle, gearbox and generator. Several vibration
modes were identified and presented for different rotor speed regimes. A numerical aeroelastic model
was also developed and the results obtained were correlated with the experimental ones.
Ozbek et al (Ozbek, Rixen et al., 2010; Ozbek and Rixen, 2012) monitored the dynamic response of a
2.5 MW HAWT using conventional instrumentation (strain gauges) and optical measurement systems
(laser interferometry and photogrammetry). The wind turbine was tested under parked (both with
strain gauges and laser interferometry) and rotating (both with strain gauges and photogrammetry)
conditions. Although good results were achieved, two important drawbacks associated with the use of
optical measurement systems are referred: laser interferometry is not suitable for rotating conditions
testing; and the available camera systems used for the photogrammetry testing had a reduced memory,
allowing to record only small measurement segments.
Also Marinone et al (Marinone, Cloutier et al., 2014) performed both experimental and operational
modal analysis on two 225 kW wind turbines, in parked condition. A very complete instrumentation
was used, with 80 accelerometers installed on the blades, hub, nacelle and tower and 6 seismic
accelerometers installed at the foundation. The authors of the study were able to identify 18 vibration
modes in the 0 – 15 Hz frequency range and the correlation between the results from the EMA and
OMA was very good.
Apart from the entire wind turbine structure testing, blades properties are important features that
need to be identified (Yang and Sun, 2013). (Thomsen, Petersen et al., 2000) tried to identify damping
of edgewise blade modes of a 600 kW wind turbine under operation. For that purpose, the authors
installed eccentric rotating masses in the nacelle to excite the desired modes. Modal identification tests
are also often conducted under laboratory conditions (Griffith, Smith et al., 2006; Griffith and Carne,
2010; White, Adams et al., 2010).
Yang and Allen (Yang and Allen, 2012) also performed a modal analysis on a small wind turbine blade
based on a laser scanning technique. The test was performed under parked conditions and the authors
accomplished to identify some vibration modes (frequencies and mode shapes).
Another important field of application of OMA techniques on wind turbines is in offshore situations,
since different foundation designs lead to different structural behaviours (section 3.1.1.5). Moreover,
the contribution of the different types of damping is also difficult to predict (see section 4.5). Thus, the
extraction of the modal parameters under operating conditions reveals as an essential tool for the
correct comprehension of the dynamic behaviour of these structures.
Some work in this area has already been published. In (Devriendt, Magalhães et al., 2014), the modal
identification of a 3.0 MW offshore wind turbine with monopile foundation is introduced. This study
analyses the acceleration data recorded during 2 weeks (under non-operating conditions). The modal
results obtained were then confronted with collected environmental data, such as tidal level and wind
speed. This wind turbine is also study in section 5.9 of the present work.
In (Shirzadeh, Devriendt et al., 2013), the same wind turbine is experimentally tested. The main goal of
the test was to calculate the damping value of the fundamental fore-aft (FA) vibration mode. For this
purpose, both ambient and rotor-stop tests were performed. Rotor-stop tests (sometimes named
“overspeed stop tests”) are a type of forced vibration test, in which the turbine is operating at a normal
situation until a sudden pitch-out of the blades (Figure 5.1). In this procedure the wind turbine is
highly excited by the sudden absence of aerodynamic loads on the blades. Furthermore, the
131
aerodynamic damping contribution from the rotor almost does not interfere in the decay response of
the structure, since the blades are pitched-out. The authors were able to identify the first pair of
vibration modes (in FA and SS directions) and the obtained value of the damping in the FA direction
was similar in both ambient and forced vibration tests. An aeroelastic model was also developed, in
which the two tests were simulated with good results.
0.8 90
Fore-Aft Acceleration [m/s2]
0.6 80
0.4 70
Blades Pitch [°]

60
0.2
50
0.0
40
-0.2
30
-0.4 20
-0.6 10
-0.8 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [s] Time [s]
a) b)
Figure 5.1 – Example of fore-aft acceleration a) and blade pitch variation b) on a rotor-stop test (adapted from
(Damgaard, Ibsen et al., 2013))
In (Damgaard, Ibsen et al., 2013) the dynamic characteristics of offshore wind turbines with monopile
foundations are also studied. In this study, the value of damping of the fundamental mode for the side-
side direction was investigated. Once again, ambient vibration and rotor stop tests were performed
with the aim of extracting frequency and damping values for this vibration mode. In the analysis of the
results, some considerations are also made concerning the damping introduced by the soil.
Like the previous authors, (Versteijlen, Metrikine et al., 2011) studied the damping effect present in an
offshore wind turbine located in the Burbo Banks wind farm. 12 rotor stop tests were realized in order
to assess the influence of the soil.
In (Ibsen and Liingaard, 2006), a different type of offshore wind turbine is studied. A 3 MW wind
turbine with a prototype suction caisson foundation is tested in three different conditions: idling;
without blades; and without blades and nacelle. For the test, 15 accelerometers were installed along the
tower, and the structure was subjected to ambient excitation. The experimental results were compared
with a numerical model of the structure.
132
Chapter 5
5.3 PARTICULARITIES OF OMA APPLICATION TO WIND TURBINES

In the previous section, the main modal identification tests performed on wind turbines were
presented. Although some refer to wind turbines under operation, there is an important part that only
discusses the modal properties of the structure under parked or idling condition. This might seems a
contradiction to one of the main advantages of OMA techniques – the ability to extract the modal
characteristics under real operating conditions - but there is one main reason for that: the application
of OMA to wind turbines under operation violates some of its basic assumptions (Tcherniak, Chauhan
et al., 2010a; Ozbek, Meng et al., 2013).
There are mainly four assumptions for the applicability of OMA (Tcherniak, Chauhan et al., 2010a):
o The structure does not vary in time;
o The forces should have broadband frequency spectra;
o The forces have to be uncorrelated;
o The forces have to be distributed over the entire structure.
When in operation, wind turbines can be idealized as a set of connected substructures with rotations
with respect to each other: the tower is fixed, the nacelle rotates on the top of the tower and the rotor
spins relative to the nacelle. This is an obvious case of violation of the time invariance assumption of
the structure.
Addressing the previous assumptions related to the input force, it is correct to assume that the wind, as
exciting force, would validate all the conditions listed. This is the reason why OMA works in a
straightforward way for wind turbines under parked situations: the structure is fixed (there is no
relative movement between substructures) and the acting wind is the only excitation force. However,
when the rotor is operating, the spectrum of the forces acting on the rotor changes from a flat
spectrum to one with several peaks (as explained in Chapter 4).
Tcherniak et al (Tcherniak, Chauhan et al., 2010a) presents an analytical study in which evidences the
violation of the temporal and spatial randomly distributed nature of the excitation forces, due to the
effect of the operating rotor. The authors studied the temporal auto-correlation at different blade
radius. The main results are presented in Figure 5.2.
2
10
Power density spectra
1
10
0
10
r=0
r = 26.42m
-1
10 r = 54.40m
-2
10
0.001 0.01 0.1 1p 2p 3p 4p 1
Frequency [Hz]
Figure 5.2 – Normalized power spectral density of the temporal auto-correlation functions computed for points on the
same blade (for different radii cases) (adapted from (Tcherniak, Chauhan et al., 2010a))
133
As can be seen, the power spectral density functions (for radii higher than 0) are not flat as in a
common wind spectrum (Kármán, 1948). Moving from the centre of the rotor to the blade tip, peaks
with a frequency equal to the frequency of rotation of the rotor (and its harmonics) become more
pronounced. This is motivated by the same phenomena exposed in Chapter 4:
o Every time a blade passes through a region with a determined wind speed (gust), the action on
the blade is somehow “similar” to the prior passage (considering the gust is long enough) –
this fact justifies the existence of the peaks;
o With the increase of radius of the point on the blade, the ratio between tangential speed of the
point and the mean wind speed increases – this fact justifies the increase of the peaks with the
increase of the radius of the point.
This is the reason why the second assumption presented earlier is violated. On the other hand, the
results obtained for the point with = 0, since it is stationary, do not present any peak. This is the
reason why this assumption is not violated under parked conditions.
The third assumption of OMA (spatial non-correlation of the input excitation) was also investigated in
(Tcherniak, Chauhan et al., 2010a). For this study, the authors computed the coherence ( )
between the wind speed fluctuations at different points ( and ) on the same and different blades:
| |
(5.1)
.
with:
Coherence
Auto- or cross-spectrum of wind turbulence at points i and j (respectively, if i j and i j)
The results presented in (Tcherniak, Chauhan et al., 2010a) for points with different radii are shown in
Figure 5.3. It is once again evident that several peaks are presented at rotor frequency (and its
harmonics), leading to the violation of the third OMA assumption.
1
2
γAB (rA = 26. 42m , rB = 54. 40m , same blade)
2
0.8 γAB (rA = 26. 42m , rB = 54. 40m , diff. blades )
2
γAB (rA = 26. 42m , rB = 26. 42m , diff. blades )
AB
2
Coherence γ2
0.6 γAB (rA = 54. 40m , rB = 54. 40m , diff. blades )
0.4
0.2
0
0 1p 2p 3p 4p 1 6p 7p 8p 9p 2 11p 2.5
Frequency [Hz]
Figure 5.3 – Coherence functions for 2 points under different study scenarios (adapted from (Tcherniak, Chauhan et al.,
2010a))
134
Chapter 5
One important observation that should be noted is the wide shape of the peaks in Figure 5.3 (and also
noted in Figure 5.2). This situation, in which the peaks have “thick tails”, leads to the spread of the
assumption violation for a quite wide frequency band around the harmonics. Contrary to a situation
with narrow peaks, in which the signal is only “polluted” in a very narrow band of frequency around
the harmonic, Figure 5.3 highlights the fact that the band between harmonics in which the coherence
is approximately 0 is very narrow. This fact may increase the difficulty in applying OMA to wind
turbines.
The last assumption, in which is referred that the forces have to be distributed over the entire
structure, is almost fully fulfilled. Although the top of the supporting structure is subjected to a
considerable higher loading when in operation (due to the rotor), the remaining part is also excited by
the wind. Hence, this assumption does not introduce any additional concern in the modal analysis of
wind turbines.
In spite of these difficulties, it is still possible to apply OMA to operating wind turbines. Section 5.4
describes the modal identification algorithms used in this work. Some possible solutions to overcome
the problem of the assumptions violation are presented in section 5.6.
135
5.4 OUTPUT-ONLY STOCHASTIC IDENTIFICATION METHODS

In this section, an overview of the output-only algorithms implemented in Matlab (MathWorks, 2012)
and used in this work is presented.
5.4.1 ALGORITHMS BASED ON IDENTIFICATION OF STATE-SPACE MODELS

This section introduces two different time-domain, parametric modal identification algorithms: the
SSI-COV and the SSI-DATA. They both aim to identify a state-space model of the structural system
from the measured signals (in this case, accelerations). For a better understanding of the methods,
some considerations about state-space models and Kalman filters are also presented in this section.
5.4.1.1 State-Space model

A stochastic state-space model is an alternative time domain formulation of a dynamic system. In its
discrete version, a stochastic state-space model is characterized by (Juang, 1994):
. .
(5.2)
. .
with:
State vector (which contains the displacement and velocity vectors of the system)
Output vector (which contains the measured output values)
Input vector
State matrix
Input matrix
Output matrix
Direct transmission matrix
Vector containing noise due to disturbances and modelling inaccuracies
Vectors containing noise due to sensor inaccuracy
Among all the elements previously presented, a special reference should be made to the matrix since
it characterizes the structural system (it is possible to extract the modal properties of the system from
its eigenvalues and eigenvectors). This is the reason why the main goal of these algorithms is to
estimate this matrix.
136
Chapter 5
In situations in which the inputs are unknown, as is the case of OMA, the terms related to can be
grouped together with the noise terms. In this situation, equation (5.2) can be rewritten as:
.
(5.3)
.
The state-space model from equation (5.2) can also be defined in a modal basis. Considering a special
case of similarity transformation:
Ψ. , (5.4)
with:
Ψ Matrix with eigenvectors of in each column
, Modal state vector at instant
it is possible to define the modal state-space model:
, Λ. , Ψ .
(5.5)
. ,
with:
Λ Modal state matrix (Λ Ψ . . Ψ)
Modal output matrix ( . Ψ)
One important property of the modal state-space model is that, due to the diagonal structure of the Λ
matrix, it is possible to decouple the contribution of each vibration mode to the response of the
structure. In the same way, the matrix contains the observable components of the mode shapes in
each column, in accordance with the eigenvalues from the Λ matrix.
Due to these characteristics, it is possible to reduce the order of the state-space model in order to
consider only the desired modes. This property will be applied and discussed in section 5.5.2.3.
5.4.1.2 SSI-COV algorithm

The COVariance driven Stochastic Subspace Identification method is a time-domain, parametric
method that identifies a stochastic state-space model from the output covariance matrix (or
correlation, since the mean value of the signals is assumed to be zero). The algorithm presented in this
work was developed and described in (Peeters, 2000).
137
The algorithm is based on the principal assumption that the correlation functions of the response of a
structure excited by white noise are defined by a sum of sinusoids with exponential decrement. Since
these sinusoids are related with the impulsive structural response, the dynamic characteristics of the
structure can be then extracted from these functions.
The method starts with the computation of estimates of the output correlation matrix for 2. -1 lags,
being an input of the method to be tuned by the analyst.
These correlation matrices are organized in a Toeplitz matrix:
…
…
| (5.6)
… … … …
. . …
with:
Refers to the reference sensors. Reference sensors are a subset of sensors that should present relevant
∗
modal ordinates of the modes to be identified. All sensors might be assumed as reference sensors.
Correlation matrix evaluated at time lag , between all sensors and reference sensors
Considering that the inputs are idealized as realizations of white noise processes and that the state
vector can be characterized by a stationary stochastic process with zero mean, each correlation matrix
can be factorized according to (Overschee and Moor, 1996):
. . (5.7)
with:
Next-state output correlation matrix
Considering the application of the property presented in equation (5.7) to the Toeplitz matrix, one
obtains:
.
| . . … . .Γ (5.8)
…
.
with:
Extended observability matrix
Γ Reversed extended stochastic controllability matrix
138
Chapter 5
Computing the singular value decomposition of the same Toeplitz matrix, the following factorization
is obtained:
0
| . . . . . . (5.9)
0 0
Comparing now equations (5.8) and (5.9), it is possible to define O and Γ matrices in a different way:
. (5.10)
Γ . (5.11)
Considering that matrix A characterizes the dynamic behaviour of the structural system, and observing
equation (5.8), it is possible to state that, with the equalities from equations (5.10) and (5.11), the
identification problem is solved. Since it is known that the matrix is defined by the first rows of the
matrix ( represents the number of measured outputs), the matrix can be obtained through several
ways. In this work, the following formulation was implemented:
. .
. . . .
. ⟺ . (5.12)
… … … …
. . . .
with:
∗ Moore-Penrose pseudo-inverse
Computing the eigenvalues ( ) and eigenvectors decomposition of matrix , the results in discrete
time are obtained:
A Ψ. Λ. Ψ (5.13)
with:
Ψ Matrix with eigenvectors in each column
Λ Matrix with eigenvalues ( ) in the diagonal
139
The natural frequency values and modal damping ratios for each vibration mode are then obtained
taking into account the conversion of the eigenvalues from discrete time to continuous time model:
ln
∆ (5.14)
2.
ln
∆
ln (5.15)
∆
with:
Frequency value of the nth vibration mode
ξ Damping ratio of the nth vibration mode
Sampling interval of the outputs
∗ Absolute value of *
∗ Real part of *
With the values of the frequency and damping identified, the only missing result is the mode shape of
each vibration mode. The observable components of each mode shape are thus obtained with:
. (5.16)
with:
Matrix with the mode shape configurations at the measured degrees of freedom
Some further considerations should be mentioned concerning the application of the SSI-COV
algorithm to real structures. Considering the system identification of theoretical structures, the
application of the Singular Value Decomposition to the Toeplitz matrix, leads to singular values equal
(or very close) to zero after a certain order (see equation (5.9)). In other words, there should be a clear
gap between singular values that defines the order of the model. However, in real structures, due to the
noise contained in the signals and due to the fact that the used correlation functions are estimates,
there is not a clear gap between singular values. Thus, it is common to calculate the results for different
orders of the matrices and . With this method, a group of modal results is obtained for each order.
Then, these results are compared with the results from the previous model order by analysing the
variation between the estimated modal parameters. If the variation is within a previously defined limit,
the result (or pole) is considered stable. The results obtained for different orders are usually presented
in a stabilization diagrams as illustrated in Figure 5.5.
Example
The SSI-COV algorithm was applied to the acceleration time series obtained with the numerical model of
the NREL 5MW wind turbine introduced in section 4.7. For this analysis, the three sensors were defined
140
Chapter 5
as reference and 257 points were considered for calculation of the correlation functions (2. -1 = 257). The
obtained correlation functions are presented in Figure 5.4.
-3 R11 -3 R12 -4 R13
x 10 x 10 x 10
3 5
2
5
Amplitude
Amplitude
Amplitude
1
0 0
0
-1
-2
-5
-3 -5
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Time [s] Time [s] Time [s]
-3 R21 -3 R22 -3 R23

x 10 x 10 x 10
5 10
2
5
Amplitude
Amplitude
Amplitude
1
0
0
0
-5 -1
-5
0 1 2 3 4 5 0 1 2 3 4 5 0 2 4
-4 R31 -3 R 32 -3 R33
x 10 x 10 x 10
2 4
5
1 2
Amplitude
Amplitude
Amplitude
0 0 0
-1 -2
-5
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Figure 5.4 – Correlation functions
Considering the following stabilization criteria:

o Maximum variation of natural frequency (Δ ): ≤ 1 %;
o Maximum variation of damping ratio (Δ ): ≤ 5 %;
o Minimum MAC value ( ): 0.99;
o Maximum damping ration value ( ): 10 %.
a stabilization diagram considering results from models with orders between 1 and 50 was obtained
(Figure 5.5). As can be seen, there are several clear alignments of stable poles. The alignments referred to
the two first tower bending modes are well identified and are highlighted by red boxes.
In addition, other stable alignments are also noticed. Some of them coincide with frequency values
corresponding to rotor rotation harmonics (3Ω to 12Ω). These alignments are identified at the top of the
plot. As shown in Figure 5.5, these harmonics are misidentified by the algorithm as regular vibration
modes. The remaining stable alignments are related to vibration modes from the rotor, which are not
analysed in this example.
141
The modal damping of the stable poles is shown in Figure 5.6. In this figure, the poles related with the two
considered vibration modes are also highlighted by red boxes.
Stabilization diagram All poles Stable Freq. Stable Damp. MAC

3Ω 6Ω 9Ω 12Ω
50
40
Model Order
30
20
10
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.5 – Stabilization diagram obtained with the SSI-COV method (average spectra at the background)
Stabilization Frequency - Damping

10
8
Damping ratio [%]
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.6 – Modal damping ratio estimates for the stable poles
The identified modal properties from the two vibration modes are shown in Table 5.1.
Table 5.1 – Modal properties: SSI-COV results vs reference results
SSI-COV results Reference results
Natural frequency Damping ratio Natural frequency Damping ratio

Mode
[Hz] [%] [Hz] [%]
1 0.332 7.15 0.333 0.24
2 2.639 2.08 2.686 1.91
142
Chapter 5
The results obtained for the natural frequency are in a good agreement with the reference values. On the
other hand, the identified damping ratio for the 1st vibration mode is considerably higher than the value
obtained with the HAWC2 eigensolver. This discrepancy is due to the fact that the identified damping is
related to the effective damping of the mode during the 600 s. Thus, besides the structural damping, the
identified value also includes the aerodynamic component of the damping. The results obtained with the
SSI-COV algorithm clearly show the influence of the aerodynamic damping on the 1st vibration mode of
the tower. This effect is not relevant for the 2nd vibration mode of the tower.
5.4.1.3 Kalman filter

The Kalman filter is an important aspect of the SSI-DATA algorithm. In that sense, a brief
introduction to the non-steady-state Kalman filter is presented. A more detailed description can be
found in (Juang, 1994).
The main goal of the Kalman filter is to compute an optimal estimation of the state vector at the
instant ( ) based on the observations up to 1 and on the system matrices. Considering an
initial state estimate and the initial covariance of the state estimate . , the non-steady-
state Kalman filter estimates are obtained with the recursive utilization of the equations:
. . (5.17)
. . . . (5.18)
. . . . . . . . (5.19)
with:
Kalman filter gain at instant
Kalman error covariance matrix
Correlation matrix of the outputs (with zero time lag)
The Kalman filter state sequence organizes the Kalman filter state estimates obtained with equations
(5.17) to (5.19):
… (5.20)
5.4.1.4 SSI-DATA algorithm

The DATA-driven Stochastic Subspace Identification algorithm is another time-domain, parametric
algorithm. Just like the SSI-COV algorithm, it aims to identify a state-space model in order to compute
the modal properties of the system. However, instead of using the covariance between signals, SSI-
DATA uses the projection of row space of future output into the row space of the past reference
sensors. The concept behind this operation is the retention of all past information that is important to
predict the future.
143
The version of SSI-DATA algorithm implemented in this work is presented in (Peeters, 2000), which
in turn is based on (Overschee and Moor, 1996).
The SSI-DATA algorithm starts with the organization of the data in an Hankel matrix:
…
…
… … … …
1 ↕ . " "
… |
(5.21)
√ … | .
↕ . " "
…
… … … …
. . … .
with:
2. 1 Total number of data points (per channel)
Number of outputs sensors
Parameters of the method to be defined by the analyst
It is common to divide the Hankel matrix into a past reference and a future part, the upper and lower
part of the matrix, respectively. Considering this division, the projection operation can then be defined
as:
/ (5.22)
The definition of the is a very important aspect of the algorithm. In fact, the SSI-DATA algorithm
relies on fact that the can be factorized as product of the extended observability matrix and the
Kalman filter state sequence :
.
. . … (5.23)
…
.
The most computational efficiency way to calculate is through the application of the 1
factorization to the Hankel matrix:
. (5.24)
1
In a factorization, the is an orthonormal matrix ( . . ) and is a lower triangular matrix.
144
Chapter 5
Omitting the columns filled only with zeros in the matrix, the and matrices can be organized as
follow:
. 1
↔ ↔ ↔ ↔ ↔
. ↕ 0 0 0 ↕ .
↕ 0 0 ↕
. (5.25)
↕ 0 ↕
1 ↕ ↕ 1
Computing the singular value decomposition of , the following factorization is obtained:
0
. . . . . . (5.26)
0 0
From the factorization process, and can be calculated according to:
/
. (5.27)
. (5.28)
On the other hand, the projection of future row spaces into past row spaces can be expressed with the
help of the and submatrices:
. (5.29)
After the estimation of , it is necessary to compute the state sequence :
. (5.30)
The extended observability matrix is obtained after deleting the last rows of , while is
given by:
. (5.31)
145
At this point, the system matrices can be estimated through the application of the system of equations
from the state-space model (see equation (5.3)):
. (5.32)
|
where the | matrix can be defined resorting, once again, to the and submatrices:
0
| . (5.33)
Once all matrices all calculated, equation (5.32) can be used to estimate matrices and . Since it is an
overdetermined problem, it can be solved through:
. (5.34)
|
Once state-space matrices and are obtained, the modal properties of the structural system can be
calculated with the already presented equations (5.13) to (5.16).
An important property of the SSI-DATA algorithm is that it allows to recover the noise covariance
matrices , and from the state-space model equations:
. (5.35)
Like in the case of the SSI-COV, it is common to calculate the state-state matrices ( and ) for
different orders. Then a similar strategy is followed, where the results from the different orders are
compared in order to identify the stable poles.
Example
The simulated acceleration time series from the NREL 5MW wind turbine were used to illustrate the SSI-
DATA algorithm. For this analysis, three sensors were considered as reference and a value of = 80 was
used.
The stabilization criteria defined in the SSI-COV example were also used in this example. The obtained
stabilization diagram is presented in Figure 5.7. The damping values of the stable poles are shown in
Figure 5.8.
146
Chapter 5

3Ω 6Ω 9Ω 12Ω
50
40
Model Order
30
20
10
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.7 – Stabilization diagram obtained with the SSI-DATA method (average spectra at the background)

10
8
Damping ratio [%]
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.8 – Modal damping ratio estimates for the stable poles
The values of natural frequencies and damping ratios of the two founded vibration modes are shown in
Table 5.2. The values are very similar to those previously obtained with the SSI-COV algorithm. Once
again, the contribution of the aerodynamic damping is noticeable for the 1st vibration mode.
Table 5.2 – Modal properties: SSI-DATA results vs reference results
SSI-DATA results Reference results

Mode
[Hz] [%] [Hz] [%]
1 0.332 6.85 0.333 0.24
2 2.625 1.98 2.686 1.91
147
5.4.2 P-LSCF
The third and last identification method covered in this work is the poly-reference Least Squares
Complex Frequency Domain method (p-LSCF), also known as Polymax. This is a parametric,
frequency-domain method which was initially developed in order to identify the modal characteristics
of a system from its frequency response functions (Guillaume, Verboven et al., 2003; Peeters,
Auweraera et al., 2004). Later, the previously developed methodology was adapted by (Peeters and
Auweraer, 2005) to work based on the output half-spectrum functions, an output-only version of this
method. In this work, only this last version is presented.
5.4.2.1 Half-Spectrum
Prior to the application of the p-LSCF routine, the half spectrum matrix should be estimated.
Considering the input excitation as white noise, the input spectrum ( ) is constant and not
dependent on frequency, and the output spectrum is obtained by:
. . (5.36)
Considering the modal decomposition of the frequency response function (Heylen, Lammens et al.,
2007),
∗
. .
. ∗ (5.37)
.
with:
Number of vibration modes in the frequency range under analysis
Observable components of mode shape
Modal participation factor of mode
∗
System poles ( , . . 1 . )
it is possible to define the output spectrum as a summation of the contribution of structural vibration
modes (Peeters, 2000):
∗ ∗
. . . .
. ∗
. ∗ (5.38)
. .
with:
Operational reference vector of mode
148
Chapter 5
As can be observed, the previous equation has four poles for each mode which impose the use of
models with twice the order of the system defined by equation (5.37). For this reason, the positive (or
half) spectrum is usually used:
∗
. .
. ∗ (5.39)
.
Since the frequency response function (5.37) and the half spectrum (5.39) present similar modal
decomposition, they both can be parameterized in exactly the same way (Peeters and Auweraer, 2005).
In this work, the Correlogram approach, restricted to the positive time lags of the correlation function,
is used for the estimation of the half spectrum:
0 . . .
.Δ (5.40)
2
with:
Number of time lags to analyse
With the Correlogram approach, an exponential window should be applied to the correlations before
the use of equation (5.40). This window is important mainly for two reasons: to reduce the effect of
leakage and to reduce the influence of the higher time lags, which have a larger variance (Peeters and
Auweraer, 2005). The used exponential window follows the equation:
. .
, for 1 (5.41)
with:
Decay rate of the window
The application of the exponential window leads to biased values of damping for the computed poles.
For this reason, the damping values should be corrected according to (Cauberghe, 2004):
(5.42)
5.4.2.2 Right Matrix-Fraction Description

The p-LSCF method uses a mathematical representation of the dynamic system called right matrix-
fraction description (RMFD). The RMFD is used to model linear time-invariant systems,
149
parameterizing the transfer function as a right division of two polynomial matrices and
(Reynders, 2009):
(5.43)
. . .
with:
, Polynomial matrices
, Model order of the and matrices, respectively
Complex variable
For the implemented version of the p-LSCF method, it is of great importance to refer the conversion
from a RMRD model to a state-space model (Reynders, 2009). Assuming that matrices and
have the same order ( ) and that the degree of the polynomial with the determinant of s is equal
to . (where refers to the number of structure inputs), the matrices , , and from the state-
space model are obtained through (the subscript ∗ is dropped for simplification):
. . … . .
0 … 0 0
⋮ ⋱ … ⋮ ⋮
0 0 … 0
(5.44)
0
. . … . .
with:
State matrix (of order )
Input matrix (of order )
Identity matrix
150
Chapter 5
5.4.2.3 p-LSCF algorithm

The p-LSCF method models the half spectrum matrix using a right matrix-fraction description
(RMFD) in the discrete-time frequency domain, the z-domain (Magalhães, 2010). This is achieved
through the use of matrices and . Considering both matrices as polynomials of the same order ,
they are defined as:
. . . . . .
. . . . (5.45)
Bearing in mind that these matrices contain the modal parameters of the dynamic system, the main
goal of this method involves the calculation of and that reduce the approximation error
associated to equation (5.45). The nonlinear least squares problem can be simplified to a linear
problem according to (Guillaume, Verboven et al., 2003):
. . . . . .
. . . (5.46)
Considering the polynomial basis functions evaluated at each frequency (ω ) organized in one row
with ( 1) components:
Ω Ω Ω … Ω . . . . . .
… . . . (5.47)
equation (5.46) can be re-written as:
Ω . Ω . Ω . … Ω . . (5.48)
⋮ ⋮
with:
Line of the , or matrix (which varies from 1 to the number of measured outputs - )
151
Introducing the following definitions:
⋮ (5.49)
, with 1, 2, … , (5.50)
⋮
equation (5.48) can be written in a more compact form:
, . (5.51)
where:
Ω
⋮ (5.52)
Ω
Ω … Ω ⊗
… , with 1, 2, … , (5.53)
Ω … Ω ⊗
with:
⊗ Kronecker product operator
, Limits of the frequency range under analysis
Expanding equation (5.51) with all the parameters from matrices α and β, one obtains:
0 … 0
0 … 0
⋮ ⋮ ⋱ ⋮ ⋮ . ⋮ . (5.54)
0 0 …
152
Chapter 5
The solution for the identification problem can be solved through . 0. However, this is very
computationally demanding. For this reason, a different approach is usually followed. It consists on
determining the model parameters through a linear least squares cost function obtained by adding all
the squared elements of the matrix evaluated at each discretized frequency:
∗
, . , (5.55)
with:
From equations (5.52) and (5.53), (5.55) can be written as:
, , . , . .
(5.56)
. ; . ; .
with:
∗ Trace of the matrix *
Thus, the minimum of the cost function is obtained with:
,
2 . . 0 (5.57)
,
2 . . 0 (5.58)
Eliminating the unknowns , the size of the system of equations can be reduced:
2 . . 0⇔ . . (5.59)
Then, the α matrix is obtained through:
2 . . 0⇔ . 0 (5.60)
153
Operational Mod
dal Analysis of Wind
W Turbines
In order too avoid the trivial

t on α 0, a constraint should
solutio s be immposed on th the parameteers. In
(Cauberghee, 2004), it is suggested to
o impose (from the α matrix) as an identity m
matrix. Bearin
ng this
in mind, α iis obtained with:
w
. 0 ⇔ . 0 ⇔ . ⇔
. ⋮ ⋮
0 (5.61)
. ⇒
.
Once α is d
defined, it iss possible to compute thhe matrixx using equattion (5.59), w
which defines the
RMFD mod del.
The next steep in the algo
orithm is thee definition oof the companion matrix:
. . … . .
0 … 0 0
(5.62)
⋮ ⋱ … ⋮ ⋮
0 0 … 0
If matrix from equatiion (5.44) is compared too matrix from equatiion (5.62) it is easily obsserved
their similarrity. Thus, representss the matrix from the state-space
s model.
m The reemaining maatrices
from the moodel can be computed
c through equatiion (5.44).
With the sttate-space model
m defined
d, the modall parameters of the dynaamic system can be calcu
ulated
using, once again, the eq
quations (5.1
13) to (5.16).
As referred in the case of
o the SSI-CO OV algorithm
m, there is no wledge about what model order
o prior know
( ) will prooduce the beest results. For
F this reasson, several orders are computed
c annd the resullts are
usually pressented in a sttabilization diagram.
d
Exam
mple
The aapplication of the p-LSCF algorithm
a to tthe simulated
d acceleration signals obtainned for 3 degrrees of
freedoom of the NR REL 5MW wiind turbine m model was peerformed from m correlation functions with 256
pointss. Then, the half-spectrum
h functions werre obtained foor subsequentt estimation oof matrices anda .
Thesee functions aree shown in Figgure 5.9.
154
Chapter 5
S+11 S+12 S+13

-3 -3 -4
10 10 10
-4 -4 -5
Amplitude
Amplitude
Amplitude
10 10 10
-5 -5 -6
10 10 10
-6 -6 -7
10 10 10
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
S+21 S+22 S+23

-3 -3 -4
10 10 10
-4 -4 -5
Amplitude
Amplitude
Amplitude
10 10 10
-5 -5 -6
10 10 10
-6 -6 -7
10 10 10
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
S+31 S+32 S+33

-4 -4 -3
10 10 10
-5 -5 -4
Amplitude
Amplitude
Amplitude
10 10 10
-6 -6 -5
10 10 10
-7 -7 -6
10 10 10
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Figure 5.9 – Half-spectrum functions
The stabilization diagram for models with order between 2 and 35 was created considering the same
stabilization criteria defined for the SSI-COV and SSI-DATA algorithms (Figure 5.10). The modal
damping ratio results associated with the stable poles are shown in Figure 5.11.

3Ω 6Ω 9Ω 12Ω
35
30
25
Model Order
20
15
10
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.10 – Stabilization diagram obtained with the p-LSCF method (average spectra at the background)
155

10
8
Damping ratio [%]
0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 5.11 –Modal damping ratio estimates for the stable poles
The final results obtained with p-LSCF algorithm are presented in Table 5.3. The results are similar to
those previously obtained with the other two identification algorithms.
Table 5.3 – Modal properties: p-LSCF results vs benchmark results
p-LSCF results Benchmark results

Mode
[Hz] [%] [Hz] [%]
1 0.339 6.791 0.333 0.24
2 2.641 1.99 2.686 1.91
156
Chapter 5
5.5 MODAL RESPONSE ESTIMATION

One interesting post-processing methodology of the presented stochastic identification methods
allows using the computed state-space matrices to reconstruct the measured acceleration signals as a
sum of modal contributions. With this tool, the measured acceleration signals are splited into several
modal responses. A modal response is defined as the acceleration response of a SDOF with the same
modal properties of the considered mode.
This result is obtained with the definition of a forward innovation model from the state-space model
matrices. This methodology is presented for the three parametric stochastic identification models: SSI-
DATA, SSI-COV and p-LSCF.
5.5.1 FORWARD INNOVATION MODEL

The state-space model, presented in 5.4.1.1, can also be represented under different forms. One of
these representations is the forward innovation model. This representation is obtained with the
application of the Kalman filter to the stochastic state-space model (Overschee and Moor, 1996):
. .
(5.63)
.
with:
State vector
Innovation sequence
The innovation sequence is a white noise vector, with the covariance matrix defined as:
. . (5.64)
The forward innovation model is thus defined with the matrices and , in addition to the state-
space matrices A and C already defined. These matrices can be obtained through the solution of the
discrete Riccati equation [(Peeters, 2000)]:
. . . . . . . . (5.65)
with:
Forward state covariance matrix ( . )
Next-state output correlation matrix
Correlation matrix of the outputs (with zero time lag)
157
After calculation of the forward state covariance matrix, the covariance matrix of the innovation
sequence is obtained as:
. . (5.66)
and, finally, the Kalman gain is computed according to:
. . (5.67)
5.5.2 APPLICATION TO THE STOCHASTIC IDENTIFICATION ALGORITHMS

The methodology herein presented for modal response estimation is applied to the three parametric
identification algorithms.
Initially, the methodology for the SSI-DATA algorithm is introduced. It is based on the technique
presented in (Peeters, 2000). Then, an adaptation of this methodology for the SSI-COV and p-LSCF is
introduced. This constitutes one of the new developments included in the present work, with practical
relevance in the context of the continuous dynamic monitoring (see Chapter 7).
5.5.2.1 Application to the SSI-DATA algorithm

The definition of the forward innovation model after the SSI-DATA algorithm (introduced in section
5.4.1.4) requires the computation of the next-state output correlation matrix and the correlation
matrix . These matrices can be directly calculated once the noise covariance matrices , and
from the state-space model are recovered (equation (5.35)).
Then, the Lyapunov equation is used to obtain Σ:
Σ . Σ. (5.68)
Finally, the and matrices can be computed as:
. Σ. (5.69)
. Σ. (5.70)
Once these matrices are obtained, it is possible to define a forward innovation model (equation (5.63))
through the application of equation (5.65) to (5.67).
158
Chapter 5
The forward innovation model from equation (5.63) can then be defined in the modal basis:
, Λ . , .
(5.71)
. ,
with:
, Modal state vector ( , Ψ . )
Modal Kalman gain ( Ψ . )
The advantage of using the model in a modal basis is that, since Λ is a diagonal matrix, ,
represents the separate contribution of each vibration mode to the total response signal. Rearranging
equation (5.71), it can be written as:
, Λ . , .
(5.72)
. ,
At this point, it is possible to use the system of equations (5.72) as a state-space model in order to
estimate the modal state sequence , and the innovation sequence from the measured
acceleration . The measured acceleration can then be reconstructed as the sum of an estimated
response part ( ) with an estimation error part ( ):
. , (5.73)
In the previous equation, represents the estimated contribution to the response from the vibration
modes, while represents the estimation error, i.e., the part of the response that cannot be explained
by the stochastic state-space model.
In addition, the acceleration signal estimation can be also thought as a sum of contributions from
single vibration modes. The estimated modal response can be split in single modal responses as:
. , (5.74)
with:
Estimated modal response of the -th vibration mode
Column vector with the observable components of the mode shape of the -th vibration mode
, Array vector from the modal state vector corresponding to the -th vibration mode
159
Naturally, is a complex modal response. However, if the complex modal response of the pair of
complex-conjugated poles from the same vibration mode is combined, the real modal response is
obtained.
5.5.2.2 Application to the SSI-COV algorithm

The methodology presented in 5.5.2.1 uses specific results obtained with the SSI-DATA algorithm to
estimate the modal response. The adaptation of this methodology to the SSI-COV cannot be done
directly since the noise covariance matrices ( , and ) from the state-space model are not obtained
with this method.
Thus, a procedure to estimate the modal response from the identified state matrix and output matrix
is introduced in this section. In its development, some steps from the SSI-DATA algorithm are used
in order to estimate the noise covariance matrices from the state-space model.
The procedure starts with the computation of the Hankel matrix with the projection of row space of
future output into the row space of the past reference sensors (equation (5.21). Then, the
factorization of the Hankel matrix is performed (equation (5.24) in order to compute the matrices ,
and | , respectively equations (5.28), (5.30) and (5.33).
Then, the residuals from the state-space model can be computed with the and matrices obtained
with the SSI-COV algorithm and with the calculated matrices:
. (5.75)
|
Finally, the noise covariance matrices are obtained with:
. (5.35)
At this point, it is now possible to use the methodology introduced in 5.5.2.1.

The validation of this procedure is illustrated through a simple example in the Appendix A.
5.5.2.3 Application to the p-LSCF algorithm

The procedure introduced in section 5.5.2.1 can also be used with the results from the p-LSCF, since it
is possible to define a state-space model from a RMFD model (section 5.4.2.2). However, some
additional challenges are present in its application.
The main restriction to the application of this procedure to the state-space matrices obtained with the
p-LSCF algorithm is that the resolution of the Riccatti equation requires that all poles (see equation
(5.37)) from the state matrix are stable (Overschee and Moor, 1996).
160
Chapter 5
In order to evaluate the stability of the poles and exclude the unstable poles (i.e. associated with
negative modal damping ratios), the unit circle can be used. The unit circle limits the stable region of
the poles in the -domain. The radius of pole in the unit circle is given by:
. . (5.76)
with:
Natural frequency (of the pole )
Δ Time sampling interval
The unit circle is illustrated in Figure 5.12.
(imaginar y axis)
j i stable pole
t
unstable pole e j j
e i i t
t (real axis)
t
e i i
j e j j
unstable pole stable pole

i
Figure 5.12 – Identification of a stable and unstable pole in the unit circle
Since it is desired to only account for the stable poles (and the respective vibration modes), it is
necessary to eliminate the unstable poles from the and state-space model matrices.
In this work, a truncation based methodology of the modal state-space model is used. The identified
and matrices are initially transformed into the modal basis:
Λ Ψ . .Ψ (5.77)
.Ψ (5.78)
As referred in section 5.4.1.1, the Λ is a diagonal matrix with the eigenvalues of the system and is a
matrix with the observable components of the vibration modes in the columns. After the
transformation from equations (5.77) and (5.78), the eigenvalues corresponding to the unstable poles
can now be removed, deleting the corresponding lines and columns from matrix Λ. Similarly, the
columns from the matrix can now be removed in accordance to the deleted eigenvalues. This
truncation operation is illustrated in Figure 5.13.
161
Operational Mod
W Turbines
…
Λ V
poles in the maatrix Λ and

Figure 5.13 – Elimination off the unstable p (trruncation operration)
uncation is complete,
After the tru c thee state-space model can be
b recovered from the moodal basis through
the inversioon of the equ
uations (5.77
7) and (5.78)). After this step,
s it is possible to folllow the proccedure
presented foor the SSI-CO
OV algorithmm in section 5.5.2.2.
The validatiion of this prrocedure is illlustrated thrrough a simp
ple example in
i the Appenndix A.
Exam
mple
With the aim of illlustrate the methodology
m presented to
o decompose the acceleratiion time seriees into
modaal responses, thhe procedure was applied too the accelerattion time seriees numericallyy generated with
w the
NREL L 5MW wind turbine modeel. For that puurpose, the reesults obtained d with the appplication of th
he SSI-
DATA A algorithm (presented in n the examplle of the secttion 5.4.1.4) were used aas reference for f the
validaation of the po
ost-processingg tool applied w
with the otherr identification
n algorithms.
A statte-space modeel of order 50,, obtained withh the SSI-DATA, was used to define the forward inno ovation
modeel. Since the reesonance response of the w wind turbine att the harmoniics frequenciees was identifieed as a
regulaar vibration mode
m (Figure 5.7), it is alsoo possible to estimate
e the response
r of thhe structure at these
frequeencies.
Figuree 5.14 presentts the modal responses
r of toop sensor S3 for
f the two traacked vibratioon modes (1 FAF and
2 FA) and the two first
f harmoniccs (3Ω and 6Ω ). The responnses obtained illustrate
i importance off the 1st
the im
F and the 3Ω harmonic eexcitation to th
vibrattion mode (1 FA) he motion at the
t top of the w wind turbine tower.
162
Chapter 5
1 FA 1 FA
0.2 0.05
SSI-DATA SSI-DATA
0.15 SSI-COV
p-LSCF
0.1 0.025
0.05
Acc. [m/s 2 ]
Acc. [m/s 2 ]
0 0
-0.05
-0.1 -0.025
-0.15
-0.2 -0.05
0 100 200 300 400 500 600 375 380 385 390 395 400
Time [s] Time [s]
3Ω 3Ω
0.2 0.15
0.15
0.1
0.1
0.05
0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05
-0.05
-0.1
-0.1
-0.15
-0.2 -0.15
0 100 200 300 400 500 600 375 380 385 390 395 400
Time [s] Time [s]
6Ω 6Ω
0.2 0.05
0.15
0.1 0.025
0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05
-0.1 -0.025
-0.15
-0.2 -0.05
0 100 200 300 400 500 600 375 380 385 390 395 400
Time [s] Time [s]
2 FA 2 FA
0.2 0.01
0.15
0.1 0.005
0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05
-0.1 -0.005
-0.15
-0.2 -0.01
0 100 200 300 400 500 600 375 380 385 390 395 400
Time [s] Time [s]
Figure 5.14 – Modal responses of the top sensor S3 (only the 1 FA and 2 FA modes and the 3Ω and 6Ω harmonics were
considered)
163
5.5.3 QUANTIFICATION OF THE MODAL CONTRIBUTIONS

An interesting outcome from the decomposition of the original signal into several modal acceleration
responses is that the relative contribution from each of these modes can be quantified (Cara, Juan et
al., 2013).
The procedure starts after the estimation of the modal acceleration response from each vibration mode
and the associated error:
(5.73)
with:
Time series of the original signal
y Time series of the estimated modal response
Time series of the innovations
Equation (5.73) can be also written as a sum of the estimated vibration mode acceleration:
(5.73)
with:
Time series of the estimated modal response of the -th vibration mode
Multiplying each element from the previous equation by the transpose of and retaining only the
diagonal elements of the matrices, the following is obtained:
. . . ⇒
(5.79)
. . . ⋯ . .
with:
∗ Diagonal elements of matrix *
164
Chapter 5
In the previous equation, the diagonal elements of . and . represent, respectively, the
covariance between the estimated modal response from mode and the original signal and the
covariance between the estimation error and the original signal. Normalizing equation (5.79), one
obtains:
. . . . . . . . . ⇒
(5.80)
1 Δ Δ ⇒ 1 Δ Δ ⋯ Δ Δ
with:
1 Vector filled with elements equal to one
Δ Vector with the contribution of the estimated modal response of the -th vibration mode
Δ Vector with the contribution of the estimation error
The procedure presented so far is referred to an acceleration signal from a single sensor. Thus, the
contribution of each vibration mode to the original signal is only referred to that single sensor.
However, it is common to have more than one sensor installed on the structure. In that case, a global
assessment of the modal contribution can be obtained with the mean value of the Δ and Δ vectors
obtained from the l installed sensors:
1
Δ (5.81)
1
Δ (5.82)
with:
Global contribution of the -th vibration mode
Global contribution of the estimation error
Example
The acceleration signals obtained with the NREL5 MW wind turbine model were once again used to
illustrate the procedure presented in this section.
In that sense, the modal responses obtained with the forward innovation model defined with the state-
space model of order 50 from the SSI-DATA algorithm was used to estimate the modal contribution to
the recorded acceleration from the 3 sensors. Once again, only the 1 FA and 2 FA modes and the two first
rotor harmonics were considered.
Figure 5.15 presents the contribution of each considered modes/ harmonic excitations for the three
sensors, together with the estimated error (Δ ). It should be noted that the sum of the presented
165
contributions (including error) is not equal to 1, since some residual contributions from the other modes/
harmonic excitations are not presented in the figure.
The acceleration time series recorded by the sensors S1 and S2 are clearly dominated by the 2 FA. It is an
expected result because the mode shape of this mode presents higher modal amplitudes at the positions of
these sensors than the 1 FA mode. On the other hand, the acceleration signal captured by the S3 sensor is
dominated by the 3Ω harmonic excitation and, in a smaller scale, by the 1 FA mode. This is justified by
the fact that motion configuration of this mode/ harmonic presents its higher modal amplitude at the top
of the tower.
0.5
0.45 S1
S2
0.4 S3
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1 FA 3Ω 6Ω 2 FA ∆e
Figure 5.15 – Modal contributions from the 1 FA and 2 FA modes, 3Ω and 6Ω harmonics and error to the acceleration
signals obtained for sensors S1, S2 and S3 with the NREL 5MW wind turbine model
166
Chapter 5
5.6 AUXILIARY TOOLS TO OVERCOME OMA VIOLATIONS

In section 5.3, some violations of the OMA assumptions were presented and discussed. It was justified
why, under operating conditions, OMA identification algorithms (such as the ones presented in
section 5.4) cannot be directly applied to wind turbine signals in the same way as when under parked
conditions. Thus, alongside with the application of the identification algorithms, some auxiliary tools
may need to be used. Reference should be made that for some strategies presented below, some input
information, usually obtained with the data collected by the SCADA system, is required.
5.6.1 COORDINATE TRANSFORMATION - NACELLE ROTATION

There are some alternative procedures to handle the problem of structure time-variance due to nacelle
rotation. The choice between then should be made according to the quality of the data available from
the SCADA system.
Considering sensors installed in the tower, framed into a global stationary reference, it is necessary to
apply a coordinate transformation to the collected signals so that these will follow the nacelle
orientation, defined by the fore-aft (FA) and side-side (SS) directions (moving reference). On the other
hand, if sensors are installed in the nacelle, no transformation is required since they are framed into
the moving reference.
Yaw angle data acquisition rate is high enough

When yaw angle data is available with an acquisition rate sufficiently high (equal to the acquisition rate
of the sensors or high enough to track the nacelle rotation with a good precision), the raw data
obtained from the sensors should suffer a coordinate transformation according to:
′ ,
.
′ ,
(5.83)
with:
′, ′ Nacelle (moving) reference coordinates
, Tower (stationary) reference coordinates
Rotation matrix
Angle between the tower sensors referential and nacelle orientation (see Figure 5.16)
This transformation leads to the projection of the collected acceleration time series in the FA and SS
directions (Figure 5.16).
167
Y
Y’ ≡ SS X’ ≡ FA
Figure 5.16 – Illustration of the angle between the sensors referential and nacelle orientation (θ)
Yaw angle data consists of time-averaged values

When the data from yaw angles is only available in the form of time-averaged values (e.g. 10 minutes
mean values), there are two possible strategies to proceed with the coordinate transformation:
o The same rotation matrix (equation (5.83)) should be used for the entire raw data from that
period;
o As an alternative, OMA algorithms can be applied directly to the raw data without any pre-
processing. Then, the configuration of the mode shapes should be multiplied by the same
rotation matrix of equation (5.83).
This approach only provides adequate results when the nacelle does not suffer important rotations
during the adopted time interval.
5.6.2 COORDINATE TRANSFORMATION - ROTOR ROTATION

The spinning rotor, as referred earlier, is another source of violation of the structure time-invariance
requirement. The rotation of the rotor with respect to the nacelle introduces time-dependence in the
mass, stiffness and gyroscopic matrices (Ewins, 2000). In experimental testing, a coordinate
transformation is required in situations where sensors are located in the rotating components of the
rotor (such as blades), under operating conditions.
168
Chapter 5
The referred violation can be faced through Multi-Blade Coordinate (MBC) transformation (also
named as Coleman transformation) (Hansen, 2007; Bir, 2008). The MBC transformation helps
integrate the dynamics of individual blades into the whole structural system, converting the motion of
each blade in the rotating blade frame into the ground-fixed frame (the same of the tower coordinate
system). With this transformation, the periodic terms in the equations of motion are eliminated
(Hansen, 2007). The MBC transformation is defined, for the most common case (3-bladed rotor wind
turbines), as (Bir, 2008):
1
3
2
. cos
3 (5.84)
2
. sin
3
with:
, , Multi-blade coordinates (ground-fixed frame)
Deflection of the -th blade (rotating frame)
Azimuth location of the -th blade: Ω. 1 . (it is assumed that for 0, the blade is
vertically up)
The new degrees of freedom in the ground-fixed frame ( , and ) represent now a cumulative
behaviour of all rotor blades, instead of each individual blade ( ). Two examples of the physical
meaning of the new coordinates is presented in (Hansen, 2007), for operating conditions with a low
value of blades pitch angle:
If → flapwise blade deflection shape (Figure 5.17) If → edgewise blade deflection shape
→ simultaneous flapwise motion of all blades → simultaneous edgewise motion of all blades
→ tilt motions of the rotor → horizontal motions of the rotor
→ yaw motions of the rotor → vertical motions of the rotor
b1
a0 a1
Figure 5.17 – Multi-blade coordinate when qi represents flapwise motion (Skjoldan, 2011)
169
The inverse transformation (from ground-fixed coordinates to the rotating frame) is given by:
. cos . sin (5.85)
In (Tcherniak, Chauhan et al., 2010b), a routine for the application of MBC transformation to records
from sensors located on the rotor is proposed:
1. The signals of points located on the blades and tower (and also other locations) and the
azimuth angle are acquired as time histories;
2. The signals data from the rotating parts are subjected to MBC transformation using the
azimuth data. The signals data from the non-rotating components (tower and nacelle) are not
transformed;
3. The new signals obtained with MBC transformation and the signals from non-rotating
components are the inputs to OMA identification algorithms;
4. The frequencies and damping values are determined;
5. The resulting mode shapes are subjected to inverse MBC transformation (5.85); the results can
be directly animated overlaid by the rotor rotation.
Although MBC is an useful tool to eliminate the periodic effect of the rotor, there are some practical
aspects that may complicate its application (Ozbek, Meng et al., 2013):
o If the measurements are only taken from one blade, the transformation cannot be applied;
o The measurements must be taken at equal radial locations for all the blades;
o The application of the MBC transformation requires the knowledge of the instantaneous
position of the blade (azimuth angle).
Two last comments should be made about the use of MBC transformation to wind turbines. There is
one fundamental assumption of this method (verified for almost all wind turbines): the rotor has to be
isotropic, which means, the all the blades are identical, identically pitched and symmetrical mounted
on the hub (Hansen, 2003). Another important aspect about the effects of utilization of this technique
is that, although it removes the periodicity from the matrices, it does not eliminate the periodicity
character of the excitation forces.
5.6.3 PERIODIC NATURE OF THE EXCITATION FORCES

The periodic nature of the excitation is probably the most difficulty OMA assumption violation to
overcome. Since the structural response contains harmonic components due to rotor rotation, the
modal identification process can be considerable hindered (or even become impossible). Thus,
auxiliary processing tools may be important to remove the pollution in the signal due to the presence
of harmonics. In that sense, three methodologies are introduced.
170
Chapter 5
5.6.3.1 Kurtosis-based Methods

One alternative method associated to the detection of harmonics is based on the properties of kurtosis.
Kurtosis is a statistical measure that can be used to estimate how peaked or flat a distribution is when
compared to the normal distribution. It is defined as:
̅
| ,
(5.86)
with:
Expectation value
̅ Mean value of the distribution
Standard deviation of the distribution
where > 3, = 3 and < 3 define distributions as leptokurtic (distribution more peaked than
normal distributions), mesokurtic (normal distribution) and platykurtic (more flat than normal
distribution), respectively. It is common to express this measure in the form of excess kurtosis:
∗
̅
3 (5.87)
| ,
∗
With this designation, leptokurtic, mesokurtic and platykurtic distributions will be defined as > 0,
∗
= 0 and ∗ < 0.
This kurtosis can also be used as an indicator of the presence of harmonic components (Brincker,
Andersen et al., 2000):
o It is assumed from the central limit theorem that any linear combination of a large number of
random variable tends to a normal distribution independently of the statistical properties of
the variables. Thus, the stochastic structural response, due to loading from a sufficient large
number of independent sources, will be approximately Gaussian (see Figure 5.18 for the
Gaussian probability function). For this reason, the value of excess kurtosis will be ∗ = 0
(mesokurtic distribution);
o In the case of a pure harmonic, the probability function presents a characteristic shape with
two symmetric peaks approaching infinity (Figure 5.18). As result, it will present a value of
excess kurtosis of ∗ = - 1.5.
171
PDF
1
Normal
Harmonic
0.8
0.6
0.4
0.2
0
-3 -2 -1 0 1 2 3
Figure 5.18 – Normalized probability density function of a pure structural mode (blue line) and pure harmonic
component (red dashed line)
Based on these properties, some methodologies were proposed (Jacobsen, Andersen et al., 2006;
Andersen, Brincker et al., 2007; Jacobsen, Andersen et al., 2008). However, some difficulties are
expected regarding the application of these methods to the modal identification of wind turbines. In
fact, it is referred that when the harmonic is located closed to the natural frequency of a structural
mode, the results may deteriorate (Jacobsen, Andersen et al., 2008).
5.6.3.2 Cepstral Method

The Cepstral method is another methodology aiming to eliminate the effect of harmonics in a signal.
The concept of cepstrum was initially introduced with the aim of detecting echoes in seismic analysis
(Bogert, Healy et al., 1963). This method consists in editing the vibration signals in the cepstrum
domain. The Cepstral method proposed in (Randall and Sawalhi, 2011) consists in editing the real
cepstrum while keeping the original phase. The cepstrum can be obtained as:
log (5.88)
where:
| |. .
(5.89)
with:
| | Magnitude of the Fourier transformation of
Phase angle of the Fourier transformation of
172
Chapter 5
and, consequently:
log ln | | . (5.90)
Equation (5.88) will result in the “complex cepstrum”. On the other hand, if the phase of is set to
zero, “real cepstrum” is obtained.
The main steps of this methodology are presented in Figure 5.19. It consists on the combination of the
edited “real cepstrum” (where influence of the harmonics is removed) with the unmodified phase
information. The complex spectrum is then retrieved by exponentiation.
Input FFT + Edited log Exp. Comple x
Phase
signal spectrum spectrum
+ IFFT
Log Tim e
amplitude signal
IFFT
Real
cepstrum
Edit
Edited
cepstrum
FFT
Edited log
amplitude
cepstrum
Figure 5.19 – Algorithm of the Cepstral method used to eliminate the effect of harmonics in signals (adapted from
(Randall and Sawalhi, 2011))
The application of this methodology to vibration signals from wind turbine is described in the
literature. The results indicated that the use of the cepstral method is still not a definite solution to
solve this problem. In some cases (Manzato, White et al., 2013), the application of the cepstral method
led to an improve of the results, while in other situations (Manzato, Devriendt et al., 2014), its
application spoiled the modal identification.
5.6.3.3 Time Synchronous Averaging

The Time Synchronous Averaging (TSA) method aims to transform a signal containing time-varying
frequency components (as varying rotor speed) into a constant frequency component. The main
feature of this method is the digitally resample of the original signal (acquired with a constant time
interval – time domain) to a constant angular spacing – angular domain. This resampling process is
illustrated with the help of Figure 5.20. After this step, the “varying” harmonic, transformed into a
“constant” frequency harmonic, should be eliminated with a filter.
173
Frequency
Frequency
Frequency
Time Time Time
Sample Time Axis Sample Time Axis Sample Time Axis
Sample Angle Axis Sample Angle Axis Sample Angle Axis
Sampled Waveform Sampled Waveform Sampled Waveform
a) b) c)
Figure 5.20 – Graphical representation: a) sine with constant frequency; b) sine with increasing frequency with constant
Δ ; c) sine with increasing frequency and constant Δ (adapted from (Blough, 2006))
Time synchronous averaging represents an useful tool to identify a varying frequency of excitation
along time. As example, Peeters et al (Peeters, Cornelius et al., 2007) refers a successfully case study in
which this technique was applied to an helicopter.
However, the application of this method for the analysis of vibration signals from wind turbines
presents some drawbacks, including the need to know the position or frequency/ radial speed of the
rotor with very good precision, which implies the installation of additional sensors at the turbine.
Other important difficulties are also referred in literature regarding the application of TSA to wind
turbines. Tcherniak et al (Tcherniak, Chauhan et al., 2010a) refers that it should not work properly
since this method suits better in situations where the harmonics are identified as well-defined sharp
peaks. As shown in section 5.3, the peaks obtained from the harmonic excitation present the
previously introduced “thick tail” characteristic. This difficulty was experimentally confirmed in
(Manzato, White et al., 2013).
Although different methodologies to identify and remove the effect of harmonics in vibration signals
have been developed, their application to wind turbine signals have not revealed as a definite solution
for this problem. In that sense, other types of strategies are usually employed, as shown in Chapter 7.
174
Chapter 5
5.7 AUTOMATED OPERATIONAL MODAL ANALYSIS

In the implementation of a dynamic monitoring system, it is desirable to have the least possible human
intervention. In that sense, an additional procedure has to be added to the output-only identification
algorithms presented in section 5.4 to avoid the intervention of an experienced operator in the analysis
of stabilization diagram. For this reason, an algorithm for automated extraction of the desired modal
parameters from the stable poles on a stabilization diagram is necessary.
According to (Reynders, Houbrechts et al., 2012), three major types of strategies have been proposed
to automate the identification process: histogram analysis, non-hierarchical and hierarchical
clustering.
In histogram analysis, with the aim of identify vertical alignments of stable poles, the frequency axis is
divided into a certain number of bins and the poles contained in each bin are counted. Scionti et al
(Scionti, Lanslots et al., 2003) presented an algorithm based on this procedure.
Another approach to perform an automatic interpretation of stabilization diagrams is cluster analysis.
Cluster analysis is a multivariate statistical analysis technique of classification, in which the objects
(poles) are partitioned into a certain number of classes (Gordon, 1999). This partition of data into a set
of disjoint classes (or clusters) leads to the integration (of similar objects) and separation (of unrelated
objects) of the data. Concerning the actual problem presented in this work, clustering analysis is used
as a tool to group physical poles that refer to the same vibration mode. This can be accomplished by
considering poles as objects with quantitative parameters – the modal parameters. Clustering analysis
can be divided in mainly two categories: hierarchical and non-hierarchical algorithms.
Non-hierarchical clustering procedures allocate data poles into a pre-specified number of clusters.
Initial seeds are defined by the user among which the clusters will be created (Reynders, Houbrechts et
al., 2012). The poles will be then assigned to the cluster in order to optimize the criterion of high
internal homogeneity (within clusters) and high external heterogeneity (between clusters). A famous
mathematical algorithm is the k-means method (Gordon, 1999). The major drawbacks of non-
hierarchical algorithms are the need of a priory definition of the number of clusters (vibration modes)
and the selection of the seeds (although the latter may be overcome by a random selection of the seeds
(Magalhães, 2010)). An illustration of this algorithm is presented in Figure 5.21). Some examples of
application of this algorithm for automated classification of poles on system identification are
presented in (Verboven, Parloo et al., 2002; Scionti and Lanslots, 2005).
- Initial seed
E
- Poles
B
- Cluster
A
A C
H D
G
F
Figure 5.21 – Non-hierarchical clustering
175
The last described strategy is the hierarchical clustering analysis. With this strategy, each initial pole is
considered a cluster. Then, and in each step, the closest poles are aggregated into the same cluster until
all poles are combined in one large cluster (Figure 5.22). The algorithm of a hierarchical clustering
methodology is as follows:
1. Definition of distance between poles. This property will measure the closeness between poles;
2. Calculation of similarity between every clusters. The criteria for similarity calculation has to be
chosen (the more common are single link, complete link, group average link, Ward’s method,
centroid method and median (Gordon, 1999));
3. Linking of the clusters in an hierarchical tree;
4. Definition of a rule to cut the hierarchical tree. This rule will define the final number of
clusters.
It should also be referred that, contrary to what can happens in non-hierarchical methods, a pair of
poles grouped together at a certain step cannot be separated in subsequent steps. Pappa et al (Pappa,
III et al., 1997) is said to be the first author to report the utilization of this kind of strategy for
automatic system identification (Reynders, Houbrechts et al., 2012). Other descriptions are presented
in (Verboven, Guillaume et al., 2003) and (Magalhães, Cunha et al., 2009) with examples of application
with the LSCF and SSI-COV method, respectively.
A,B,C,D,E,F,G,H
cut level - Cluster
- Final Cluster
A,B,C,D E,F,G,H
A,B F,G,H
F,G
C,D
A B C D E F G H
Figure 5.22 – Hierarchical clustering
5.7.1 IMPLEMENTED HIERARCHICAL CLUSTER ANALYSIS

The strategy implemented in the monitoring system described in this work is based on a hierarchical
clustering algorithm. The aim of application of this algorithm is to automatically separate spurious
poles from the physical ones and to aggregate the latter in clusters according to the different vibration
modes.
Taking into account the algorithm presented earlier, the initial step of the analysis comprises the
computation of the distance between all pairs of poles. For this calculation, the modal properties of
each pole should be used. However, the utilization of the modal damping usually presents two
important drawbacks: different modes may present the same value of damping and the extracted
values of damping ratios present high variance. For these reasons, the use of the modal damping as a
176
Chapter 5
pole property for calculation of similarity between poles was dropped. Thus, the frequency value and
the mode shape (more specifically, the MAC value) were the properties used for distance calculation
between poles (Magalhães, Cunha et al., 2009):
1 , (5.91)
with:
, Frequency value of pole or
, Modal assurance criterion between mode shapes of pole and
The different parameters of equation (5.91) can be computed with weighting factors, which allows
assigning different importance to each property. However, this possibility was not used in the
monitoring system. As can be concluded from the observation of equation (5.91), the shorter the
distance is, the greater the similarity between poles and the higher the probability that both are
referred to the same vibration mode.
After the calculation of the distance between poles, it is necessary to decide a criterion to assess the
similarity between clusters. In the implemented strategy, the single link was used. This criterion
defines the distance between two clusters as the smallest distance between poles contained in different
clusters. This distance is the same as defined in equation (5.91).
Once the hierarchical tree is constructed, it is necessary to define another criterion for the cut level.
This decision cannot be based on a hard criterion, like a final number of clusters: it is not possible to
know how many clusters will be formed containing spurious poles; besides, there are structural modes
which will not be detected in every setup. These aspects introduce an uncertainty about the final
number of physical and spurious clusters. Thus, the chosen criterion imposes a degree of homogeneity
inside the cluster: a maximum distance between any point and its closest point in the same cluster.
However, this maximum distance has to be carefully evaluated, since it can lead to the split of similar
poles into different clusters (if the distance is too tight) and the aggregation of different poles into the
same cluster (if the distance is too large).
After the application of the previous criterion, it is necessary to pick the final clusters containing
physical poles. Considering that poles referring to physical modes are consistent along the different
orders of the system identification algorithms, an initial cut-off criterion should be implemented to
eliminate clusters with a reduced number of poles. Then, the modal properties of the remaining cluster
can be compared with the reference properties of the vibration modes intended to be tracked. These
reference properties can be defined during an initial training period.
Example
The results obtained with the application of the SSI-COV algorithm to the acceleration time series from
the NREL 5MW wind turbine model were used to illustrate the implemented hierarchical cluster
methodology. In this example, the maximum distance between poles was set to 0.02 and a minimum
number of poles in a cluster in order to be considered was defined as 5. The results obtained with the
177
cluster analysis are presented in Figure 5.23, where the considered cut level of 5 poles is represented by the
dashed red line).
The natural frequency and damping ratio of the poles found after the consideration of the cut level are
shown in Figure 5.24. In this figure, the small circles represent the properties of the stable poles, while the
large circles indicate the mean values of the poles inside each cluster.
25 9
2 FA 8
20 7
Damping ratio [%]

6
9Ω
15
Nº of poles
3Ω 5
1 FA 4
6Ω
10
3
2
5
1
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Frequency [Hz] Frequency [Hz]
Figure 5.23 – Number of poles obtained for each cluster Figure 5.24 – Natural frequency and damping ratio of the
defined according to the implemented hierarchical cluster selected clusters (the small circles represent the properties
analysis (the red dashed line indicates the cut level) of the pole, while the large circles indicate the mean
values of the poles inside each clusters
178
Chapter 5
5.8 CASE STUDY – IZAR BONUS 1.3MW/62
5.8.1 INTRODUCTION
This section presents the most relevant results from the experimental identification of the dynamic
properties of a Izar Bonus 1.3MW/62 wind turbine. This analysis is integrated in a project to upgrade
the wind turbine with an extension of the blades, hereafter designated as Rotor Blade Extension (RBE).
This project was led by Energiekontor Windpower Improvement GmbH and was developed in
collaboration with the Institute of Science and Innovation in Mechanical and Industrial Engineering
(INEGI).
In the scope of the RBE project, several tasks were developed in order to assess the feasibility of the
rotor upgrade. Among them, the bending stress of the blades, the evaluation of debris in the gear box
oil and the definition of the wind turbine power curve before and after the rotor extension were
analysed. The Laboratory of Vibrations and Monitoring (ViBest) of FEUP was invited to perform
dynamic tests of the wind turbine blades (Oliveira, Magalhães et al., 2014).
In the context of the project, the assessment of the blades dynamic properties before and after the
extension allows to analyse three important aspects:
o Assessment of the natural frequency of the rotor blades modes – it is of the utmost importance
to guarantee that the frequency values of the extended blades are not close the tower modes
and to the frequency of rotation of the rotor;
o Assessment of the damping ratios from the rotor blades modes – it is important to assess if
there is significant change in the structural damping of the blades after the extension;
o Independent assessment of the natural frequency of each blade – this analysis is intended to
check if the RBE does not change the structural similarity between the blades.
Within the RBE project, data collected by two different measurement systems were used: one system
based on fibre Bragg grating sensors and another based on accelerometers.
The results presented in this section are mainly focused on the assessment of the dynamic properties of
the wind turbine blades before and after the rotor extension. However, it was given the opportunity to
performed an ambient vibration test (before the implementation of the RBE), including simultaneous
measurements at the tower and blades. In this context, a rotor-stop test was also performed in order to
assess the damping ratio of the first and second tower bending modes in the FA direction.
5.8.2 DESCRIPTION OF THE WIND TURBINE

The Izar Bonus 1.3MW/62 wind turbine presents a traditional configuration, with a three-bladed
upwind rotor with a diameter of 62 m (Figure 5.25). The tower is in steel and the hub is 68 m height.
The wind turbine is part of the Penedo Ruivo wind farm, located in the northern part of Portugal, in a
highly complex terrain. The turbine started operating in 2005.
179
Operational Mod
W Turbines
Figure
e 5.25 – Photo of the Izar Bon
nus 1.3MW/62
The generattor is of the fixed-speed

f type
t and preesents two different rotatiional operatiing speeds: 13
1 and
20 rpm. It presents a rated-output
r power of 1..3 MW. Thee rotor bladees have an aactive-stall co ontrol
system.
The analyseed data was collected
c durring periods oof time when
n the rotor was
w parked annd the bladess were
pitched by an angle off about 97º. Under this situation, th he blade motion in the fflapwise direection
occurs in thhe rotor planne while the motion
m in thhe edgewise direction
d is oriented
o to oout of the plaane of
the rotor. TThis configurration is opp
posite to the one under operation.
o Fig
gure 5.26 prresents two photos
p
illustrating bboth configu
urations.
a) b)
Figu
ure 5.26 – Blade
e orientation: aa) operating situ
uation; b) parke
ed situation
180
Chapter 5
5.8.3 MEASUREMENT SYSTEM

As previously referred, two distinct measurement systems were used in this analysis: one based on
fibre Bragg grating sensors and another based on accelerometers.
5.8.3.1 Measurement System Based on Fibre Bragg Grating Sensors

The measurement system based on fibre Bragg grating sensors is a permanent system installed in the
wind turbine. It is composed by strain and temperature sensors installed at each blade; all sensors are
connected in the hub to an interrogation unit which processes the data and delivers it in a local
industrial PC which stores the information.
The system is intended to measure strains at the blades root due to bending stresses. For each of the
three blades, four sensors, referred here as A, C, E and K, were installed at their roots, measuring the
strains at these positions. Sensors A and E are referred to the motion in the flapwise direction (bending
around the weaker axis), while sensors C and K are related to the edgewise direction (bending around
the stronger axis). Each pair of sensors measures the strains at opposite positions of the same direction.
Figure 5.27 illustrates the position of the sensors on the blades.
The data was acquired with a sampling frequency of 100 Hz and then decimated to focus the analysis
in the frequency range 0 – 5 Hz.
Sensor A
Sensor C
Sensor K
Sensor E
Figure 5.27 – Position of the sensors of the measurement system based on fibre Bragg grating sensors
5.8.3.2 Measurement System Based on Accelerometers

In the context of the RBE project, it was also possible to perform an ambient vibration test (including
measurements at the tower and blades) and a rotor-stop test of the wind turbine (with measurements
at the tower). This test took place only before the extension of the blades.
During the ambient vibration test, six tri-axial 24-bit strong motion recorders GeoSIG GSR-24 (with
internal force balance accelerometers), synchronized by GPS, were used to collect acceleration time
series at different position of the wind turbine. Three data sets of 16 minutes were recorded at a
sampling frequency of 100 Hz and then decimated in order to focus the analysis in the frequency range
of 0 – 5 Hz.
181
In each setup, a different wind turbine blade was monitored, alongside with the instrumentation
installed at the tower. The test was conducted under low wind speed conditions and the nacelle
presented a misalignment of around 110º to the wind direction. The elements monitored in each setup
are identified in Table 5.4.
Table 5.4 – Measurement setups and corresponding instrumented structural elements during the ambient vibration test
Data set Time length [min] Sensors Instrumented elements
S1 to S4 Tower
1 16
S5 and S6 Blade A
S1 to S4 Tower
2 16
S5 and S6 Blade B
S1 to S4 Tower
3 16
S5 and S6 Blade C
Throughout the test, four accelerometers (S1 to S4), were kept fixed at tower positions at the platform
levels 6, 8 and 10, as schematized in Figure 5.28. Both sensors S3 and S4 were placed at the top
platform in order to identify possible torsion components of the vibration modes.
Lateral view Top view
Level 6 Level 8
62 m
y S1 y S2
S3/S4 Level 10 x z x z
S2 Level 8
68 m Level 10
S1 Level 6
S4 S3
y y
x z x z
+ 0.000
Figure 5.28 – Accelerometers position kept fixed at the tower during the three data sets
In each set, a different wind turbine blade was monitored with two accelerometers (S5 and S6). The
blade to be instrumented was oriented in the horizontal position and the two accelerometers were
placed inside the parked blade in the positions characterized in Figure 5.29 (the mechanical brake was
activated and the blades were pitched with an angle of 97º). The position of the sensors at the blade for
the three setups is presented in Figure 5.30.
182
Chapter 5
a) b))
Figuree 5.29 – Position of the instrum
mented blade d
during the reco
orded setups: a)
a position for tthe blade A and
d C (data sets 1
and 3); b) p
position of the blade B (data set
s 2)
Top view
4.00 m 6.30 m
x S6 x S5
z z
Blade A
y y
5..30 m 4.70
0m
S5 S6
y
x
y
z
Blade B
x z
3.30 m 4.60 m
x S6 x S5
y z y z Blade C
Figure
e 5.30 – Positio
on of the sensors inside the different blades
Figurre 5.31 preseents an illusstration of thhe monitoreed elements of the windd turbine. The blue linee
repreesents the in
nstrumented tower, whille the red lin ne denotes thet portion of the bladee which wass
moniitored. The crosses
c stand
d for the possitions of thee installed seensors. On thhe other han
nd, the blackk
dasheed lines repreesent the parrt of the windd turbine thaat was not insstrumented.
183
3
a) b)
Figure 5.31 – Instrumented elements of the wind turbine: a) isometric view; b) front view
The performed rotor-stop test was conducted at the end of the day, when the wind speed was around
4.5 m/s. The test consisted in the sudden stop of the rotor, when the wind turbine was operating at
normal conditions, with a rotor speed of 13 rpm. The stop was accomplished through the activation of
the mechanical brake and the change of the blades pitch angle. With this operation, it was possible to
create an impulse at the top of the wind turbine in the FA direction.
In the scope of the rotor-stop test, only the acceleration at the tower was recorded. For that reason,
only the sensors S1 to S4 from Figure 5.28 were considered.
5.8.4 ANALYSIS BEFORE THE INSTALLATION OF THE RBE

The identification of the dynamic properties of the wind turbine blades, prior to the installation of the
RBE, was performed with data from the installed fibre Bragg grating sensors and data recorded by the
accelerometers.
The signals collected with the fibre Bragg grating system were used as primary source of data, since the
response of the three blades is recorded simultaneously. The test with accelerometers placed both at
the tower and blades allowed to better understand the configuration of the vibration modes.
5.8.4.1 Analysis of fibre Bragg grating sensors measurements

The data collected by the measurement system based on fibre Bragg grating sensors corresponds to a
setup of 18 min. It is referred to a period when the wind turbine was parked, with the mechanical
brake activated and the blades pitched with an angle of 97º. The mean wind speed, during the 18
minutes was, roughly, 10 m/s.
In this analysis, the signal from the sensor E installed in blade C was not recorded and, for that reason,
only 11 channels were considered. Figure 5.32 shows the recorded signals from the three blades during
the considered period. As can be seen, for the considered period, the vibration amplitude in the
flapwise direction is higher than in the edgewise direction.
184
Chapter 5
Blade A Blade A
Flapwise Edgewise Flapwise Edgewise Flapwise Edgewise Flapwise Edgewise
25 20
20 15
15
10
10
5 5
με
με
0 0
-5 -5
-10
-10
-15
-20 -15
-25 -20
0 200 400 600 800 1000 988 990 992 994 996 998 1000 1002
Time [s] Time [s]
Blade B Blade B
Flapwise Edgewise Flapwise Edgewise Flapwise Edgewise Flapwise Edgewise
25 15
20
15 10
10
5
5
με
με
0 0
-5
-10 -5
-15
-10
-20
-25 -15
0 200 400 600 800 1000 988 990 992 994 996 998 1000 1002
Time [s] Time [s]
Blade C Blade C
Edgewise Flapwise Edgewise Edgewise Flapwise Edgewise
15
25
20
10
15
10 5
5
με
0
με
0
-5
-5
-10
-15
-10
-20
-25 -15
0 200 400 600 800 1000 988 990 992 994 996 998 1000 1002
Time [s] Time [s]
Figure 5.32 – Strain measured by the fibre Bragg grating sensors: a) complete time series; b) zoom of a period of 14
seconds
As preliminary analysis, the averaged normalized power spectrum density of the time series collected
by each sensor of the three blades was computed. It is presented in Figure 5.33. It is visible how several
resonance peaks can be identified in this figure.
185
0 Averaged Normalized Power Spectral Density

10
Sensor E
Sensor C
Sensor A
-2 Sensor K
10
Amplitude
-4
10
-6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 5.33 – Averaged normalized power spectrum density from each sensor (the dashed lines indicate the main peaks
of the spectra)
The identification of the modal parameters was based on the application of two different parametric
identification algorithms: the SSI-COV (see section 5.4.1.2) and p-LSCF (see section 5.4.2). In this
analysis, the 11 collected signals were processed all together.
The results obtained with the two algorithms were very similar and in accordance with the spectra
from Figure 5.33. The stabilization diagram obtained with the p-LSCF algorithm is presented in Figure
5.34.
Stabilization diagram (p-LSCF)
40
35
30
Model order
25
20
15
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 5.34 – Stabilization diagram obtained with the p-LSCF algorithm for the data obtained with the measurement
system based on fibre Bragg grating sensors
From the analysis of the stabilization diagrams obtained with both methods, it was possible to estimate
the modal parameters of the 13 vibration modes presented in Table 5.5, including three pairs of closely
spaced modes (around 0.40 Hz, 1.14 Hz and 3.76 Hz). The results obtained with the two algorithms
were very similar. Notwithstanding, the use of the p-LSCF algorithm allowed the identification of three
additional vibration modes.
186
Chapter 5
Table 5.5 – Comparison of results obtained with the SSI-COV and p-LSCF methods
SSI-COV p-LSCF

Mode
[Hz] [%] [Hz] [%]
1 - - 0.441 0.38
2 - - 0.456 0.17
3 0.610 7.22 0.610 8.51
4 1.136 2.44 1.129 1.93
5 1.146 3.05 1.146 2.51
6 1.328 0.76 1.329 0.59
7 1.447 0.60 1.449 0.47
8 2.035 0.71 2.039 0.54
9 - - 2.108 1.00
10 3.404 1.61 3.399 1.14
11 3.512 0.55 3.512 0.55
12 3.745 1.44 3.745 1.31
13 3.776 0.47 3.776 0.42
Since the available layout of sensors only assessed the modal curvature at the blades root, a complete
description of the mode shapes is not possible. Nevertheless, considering that the shape of the first
modes of wind turbine rotors usually follows a typical configuration, an estimation of the first modes is
presented. The modal curvatures at the blades root from each identified vibration mode, alongside
with an illustration of the theoretical shapes of the first 8 modes, are presented in Figure 5.35. Since the
data from channel E from blade C was not available, the symmetric value obtained for the sensor A of
the same blade was used for purposes of illustration (identified with the white bar).
The mode shapes from the first two vibration modes do not resemble any typical rotor mode shape
configuration. These two modes present natural frequency values in accordance with typical first tower
bending modes in the FA and SS directions. Nevertheless, this fact was only confirmed with the
analysis of the results obtained with the ambient vibration test (section 5.8.4.2).
187
= 0,44 Hz = 0,46 Hz
Mode 1 Mode 2
Blade A Blade B Blade C 1 SS Blade A Blade B Blade C 1 FA
1 Tower Bending 1
Tower Bending
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
E A C K E A C K E A C K E A C K E A C K E A C K
Flapwise Flapwise Flapwise Flapwise Flapwise Flapwise
Edgewise Edgewise Edgewise Edgewise Edgewise Edgewise
= 0,61 Hz = 1,13 Hz
Mode 3 Mode 4
Blade A Blade B Blade C In plane Blade A Blade B Blade C
1 1 In plane
(symmetric)
0.8 0.8 (vertical)
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
= 1,15 Hz = 1,33 Hz
Mode 5 Mode 6
Blade A Blade B Blade C In plane Blade A Blade B Blade C Out of plane
1 (horizontal) 1
(yaw)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
= 1,45 Hz = 2,04 Hz
Mode 7 Mode 8
Blade A Blade B Blade C Out of plane Blade A Blade B Blade C Out of plane
1 1
(tilt) (symmetric)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
Figure 5.35 –Amplitude of the modal curvature obtained for the identified vibration modes and respective illustration
(the rotor configuration during the measurement period may not exactly match the illustration)
188
Chapter 5
It is noticed that the illustrations presented for the rotor vibration modes are referred to the wind
turbine under parked condition. Under this condition, the flapwise motion occurs in the rotor plane,
while the edgewise motion is related to motion out of plane of the rotor. This configuration is opposite
to the one when the wind turbine is operating and, for that reason, these illustrations should not be
extrapolated for that condition.
Furthermore, the third identified vibration mode ( = 0.61 Hz) is probably dominated by the drive
train inertia and by the support condition of the rotor (the mechanical brake was activated).
5.8.4.2 Analysis of accelerometers measurements

In order to clarify the source of the resonance peaks around 0.4 Hz and 4.0 Hz, and to validate the
previous results obtained with the fibre Bragg grating sensors , an ambient vibration test and a rotor-
stop test were performed with a set of tri-axial accelerometers.
Ambient Vibration Test

Firstly, an independent analysis was performed for the data collected from the blades and from the
tower.
The three data sets recorded during the ambient vibration test by the accelerometers located at the
blades are shown in Figure 5.36, together with the normalized power spectra density. The analysis
permitted to conclude that the spectra obtained for each blade are very similar. The shape of the
spectrum obtained for blade A shows a larger difference for the range 2 – 4 Hz when compared with
the other two data sets, though. This fact may be explained by the impulsive excitation on the blade A
at around 400 s, which may have excited, in a larger extent, the modes in this frequency range. This
impulsive excitation was due to a strong gust that occurred during the measurement period.
189
Data set 1 (blade A) Data set 1 (blade A)

S5 - Edgewise S5 - Flapwise S5 - Edgewise S5 - Flapwise
S6 - Egdewise S6 - Flapwise S6 - Egdewise S6 - Flapwise
0
0.6 10
0.4
-2
10
Acceleration [m/s 2]
0.2
Amplitude
-4
0 10
-0.2
-6
10
-0.4
-8
-0.6 10
0 200 400 600 800 1000 0 1 2 3 4 5
Data set 2 (blade B) Data set 2 (blade B)

0
0.15 10
0.1
-2
10
0.05
Amplitude
-4
0 10
-0.05
-6
10
-0.1
-8
-0.15 10
0 200 400 600 800 1000 0 1 2 3 4 5
Data set 3 (blade C) Data set 3 (blade C)


0
0.15 10
0.1 -2
10
0.05
Amplitude
-4
0 10
-0.05
-6
10
-0.1
-8
-0.15 10
0 200 400 600 800 1000 0 1 2 3 4 5
a) b)
Figure 5.36 – Recorded data sets from the accelerometers placed at the blades, during the ambient vibration test: a) time
domain; b) normalized power spectral density
The preliminary analysis of the data sets recorded with the sensors located at the tower was performed
considering two different signals combination for the tower top sensors S3 and S4: half-sum and half-
190
Chapter 5
difference of the horizontal accelerations. With this strategy, it is possible to identify modes with
relevant bending (with the half-sum) and torsion (with the half-difference) components of the tower.
The averaged normalized power spectral density of the acceleration time series in the FA direction
recorded by the sensors placed at the tower top is presented in Figure 5.37 a). On the other hand, the
averaged normalized power spectral density of the accelerations in the SS direction is presented in
Figure 5.37 b). Both spectra clearly evidence one important resonance peak around 0.4 Hz. In addition,
the spectra referred to the SS direction evidence a second important peak around 4.0 Hz. It is also
visible, for the half-difference signal combination in the SS direction, a resonance peak at 3.3 Hz that is
not seen for the half-sum combination. Thus, this peak is probably due to a vibration mode with an
important torsion component.
FA SS
Plat. 6 Plat.10 (Half-sum 6 Plat.10 (Half-sum
0 Plat. 8 Plat.10 (Half-difference) 0 Plat. 8 Plat.10 (Half-difference)
10 10
-2 -2
10 10
Amplitude
-4 Amplitude -4
10 10
-6 -6
10 10
-8 -8
10 10
0 1 2 3 4 5 0 1 2 3 4 5
a) b)
Figure 5.37 – Averaged normalized power spectral density of the of the acceleration time series recorded by the sensors
placed at the tower in the a) FA and b) SS direction
The identification of the modal parameters was also based on the application of the SSI-COV and p-
LSCF algorithms. In this analysis, each data set was analysed separately and all signals from tower and
blades were considered together. With the purpose of illustrating the obtained results, Figure 5.38
shows the stabilization diagram associated with the acceleration time series collected during setup 2.
30
25
20
Model order
15
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 5.38 – Stabilization diagram obtained with the p-LSCF algorithm for the data collected with the measurement
system based on accelerometers during setup 2
191
The analysis of the three data sets led to the identification of 16 vibration modes of the wind turbine.
Table 5.6 summarizes the modal properties from these vibration modes identified with the p-LSCF
algorithm. The obtained results present a good agreement for the different analysed data sets. It is
observed that, apart from some modes, the majority of them simultaneously evidence motion of the
rotor blades and tower. From the identified modes, only the 1st, 2nd, 15th and 16th mode presented a
dominant tower motion (without important motion of the blades). For that reason, it was concluded
that they represent the 1st and 2nd pair of tower bending modes.
The obtained results are in good agreement with the ones estimated with the measurement system
based on fibre Bragg grating sensors. However, the modal properties of 3rd vibration mode present an
important deviation between the two measurement systems. This may be due to a worst identification
of this mode with the accelerometers since the resonance peak of this mode is almost not noticeable in
Figure 5.36.
192
Chapter 5
Table 5.6 – Natural frequencies ( ) and modal damping ratios ( ) of the identified vibration modes with the data recorded
by the accelerometers and comparison with the results obtained with the fibre Bragg grating measurement system
(F.B.G.) using the p-LSCF algorithm
Data set 1 Data set 2 Data set 3 F.B.G.

Mode Description
[Hz] [%] [Hz] [%] [Hz] [%] [Hz] [%]
1 0.452 2.51 0.445 1.93 0.455 1.89 0.441 0.38 Tower SS
2 0.456 0.13 0.459 0.22 0.456 0.17 0.456 0.17 Tower FA
3 0.520 10.95 0.530 10.30 0.542 9.41 0.610 8.51 Blade flapwise
Blade flapwise
4 1.128 0.28 1.134 0.54 1.120 0.47 1.13 1.93 Tower SS
Blade flapwise
5 1.140 0.52 1.140 1.18 1.156 1.18 1.146 2.51
Tower SS
6 1.326 0.76 1.324 0.89 1.318 0.64 1.329 0.59 Blade edgewise
Blade flap/edgewise
7 - - 1.474 0.41 1.492 0.61 1.447 0.47 Tower FA
8 1.994 1.56 2.016 0.79 2.020 0.91 2.029 0.54 Blade flap/edgewise
9 2.078 0.81 2.062 1.02 2.062 0.86 2.108 1.00 Blade flapwise
Blade edgewise
10 3.260 0.90 3.330 1.02 3.310 0.86 3.399 1.14
Tower torsion
Blade flapwise
11 3.485 0.33 3.498 0.35 3.490 0.42 3.512 0.55
Tower FA
Blade flap/edgewise
12 3.735 0.29 3.742 0.36 3.738 0.38 3.776 0.42
Tower FA
13 4.038 0.35 4.034 0.33 4.034 0.36 - - Tower SS
14 4.874 0.32 4.862 0.37 4.8868 0.40 - - Tower FA
The mode shapes of some of the identified vibration modes are shown in Figure 5.39, following the
illustration introduced in Figure 5.31 (only the instrumented elements are plotted). Although the
measurement system based on accelerometers allows identifying two additional vibration modes
(when compared to the fibre Bragg grating system), the interpretation of the mode shapes with
important rotor motion presents greater difficulties.
193
= 0.45 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
FA direction SS direction FA direction
= 0.46 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
= 0.53 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
Figure 5.39 – Mode shapes of the first three vibration modes identified with the measurement system based on
accelerometer
194
Chapter 5
Rotor-stop Test
In addition to the ambient vibration test, a rotor-stop test was also performed with the aim of
identifying the damping ratio of the first tower bending mode in the FA direction. The acceleration
time history recorded by the sensors oriented in the FA direction is shown in Figure 5.40. A maximum
value around 0.6 m/s2 was achieved. The motion of the wind turbine after the sudden stop is mainly in
the FA direction; however movement in the orthogonal direction is also noticed (Figure 5.40).
Rotor-stop test Sensor S4
0.8 0.2
S1-FA
0.6
S2-FA
0.4 S3-FA 0.1
SS acceleration [m/s 2]
Acceleration [m/s2]
S4-FA
0.2
0 0
-0.2
-0.4 -0.1
-0.6
-0.8 -0.2
0 50 100 150 200 250 300 350 -0.2 -0.1 0 0.1 0.2
Time [s] FA acceleration [m/s 2]
a) b)
Figure 5.40 – a) Time history from the recorded acceleration by the sensors oriented to the FA direction. b) Movement
seen from above from the accelerations at S4 sensor (top sensor)
The main modes excited with the rotor-stop test are the 2nd vibration mode (1st tower FA mode) and
the 8th vibration mode (the identified rotor symmetric out of plane mode) in Table 5.6. The spectra
obtained with the recorded acceleration time series are presented in Figure 5.41, where the resonance
peaks of the referred modes are evidenced.
-2
10
S1-FA
S2-FA
-4
10 S3-FA
S4-FA
Amplitude
-6
10
-8
10
-10
10
0 1 2 3 4 5
Frequency [Hz]
Figure 5.41 – Power spectral density from the acceleration time series recorded during the rotor-stop test
195
In order to determine the damping level of the 1st FA tower bending mode, the signals recorded by
sensors were filtered and an exponential function was adjusted to the envelope of the free decay signal.
The used exponential function is defined according to:
. . . .
. (5.92)
For this analysis, a low-pass band filter with a cut of frequency of 1 Hz was used. Considering that the
filtered response does not present a perfect theoretical free decay response, three different amplitude
ranges were considered to adjust an exponential window: the initial part, the last part and the whole
decay. Figure 5.42 a) to c) presents the fittings performed to the acceleration time series recorded by
sensor S4, considering the three signal ranges.
The fitting obtained with the exponential function considering the first part of the signal is clearly
misfit. It is visible from Figure 5.42 a) that the function does not follow the progress of the signal. On
the other hand, the other two fittings seems to properly envelope the free decay signal.
Mode 2 Mode 2
0.3 0.3
S4-FA S4-FA
0.2 0.2
0.260*e-0.0077*2*π*0.453*t 0.201*e -0.0032*2*π*0.453*t
0.1 0.1
0 0
-0.1 -0.1
-0.2 -0.2
-0.3 -0.3
0 100 200 300 400 0 100 200 300 400
Time [s] Time [s]
a) b)
Mode 2 Mode 8
0.3 0.15
S4-FA S4-FA
0.2 0.1
0.238*e -0.0038*2*π*0.453*t
0.079*e-0.0066*2*π*2.044*t
0.1 0.05
0 0
-0.1 -0.05
-0.2 -0.1
-0.3 -0.15
0 100 200 300 400 0 50 100 150 200
Time [s] Time [s]
c) d)
nd th
Figure 5.42 – Identification of the modal damping ratio of the 2 and 8 vibration mode through filtering and fitting of
an exponential function to the envelope of the free decay response
196
Chapter 5
The same procedure was also applied to the 8th vibration mode. This mode is characterized by a
symmetric in-phase bending of the three blades in the FA direction and, for that reason, was clearly
excited during the test. For this mode, a band-pass filter for the frequency range 1.5 – 2.5 Hz was
considered since the influence of other modes in this range seems negligible. The filtered response
obtained for this mode presents an almost perfect free decay shape and, for that reason, only one
adjustment was performed. Figure 5.42 shows the fitting operation performed for the 8th vibration
mode.
In addition to the referred analysis, a second procedure was also used to estimate the damping value of
the 2nd and 8th modes. This procedure is based on the application of the SSI-COV algorithm
(introduced in section 5.4.1.2). This algorithm is adapted to consider the free decay signals as input of
the method, instead of the correlation functions calculated from the ambient responses (Magalhães,
Cunha et al., 2010). This procedure was applied to the same range of points previously presented for
the two modes.
The results obtained with both procedures are shown in Table 5.7. For the 2nd vibration mode (1st FA
mode), the fitting obtained with the initial part of the decay, associated with higher vibration levels,
provided an higher value, which can be an indication that the damping may increase with the
increasing of the vibration amplitude. The other two fittings performed for this mode represent a good
estimation of the damping value for lower vibration amplitudes with a good agreement between the
two used procedures. This value is considerable higher than the value obtained with the ambient
vibration test, which is consistent with hypothesis of the damping dependency with the vibration
amplitude
The damping value obtained for the 8th mode is coherent with the values obtained with the ambient
vibration test.
It should be noted that damping values associated with operating condition can be considerable
different, and generally higher, than these ones identified with parked conditions.
Table 5.7 – Natural frequencies (݂) and modal damping ratios (ߦ) of the 2nd and 8th vibration modes identified with the
rotor-stop test with the filtering and fitting procedure and with the SSI-COV method
Filtering and fitting SSI-COV
Considered Natural frequency Damping ratio Natural frequency Damping ratio

Mode
range [Hz] [%] [Hz] [%]
2 Initial 0.453 0.77 0.452 0.68
2 Last 0.453 0.32 0.453 0.29
2 All 0.453 0.38 0.453 0.32
8 - 2.044 0.66 2.043 0.81
197
Operational Mod
W Turbines
5.8.5 ANA
ALYSIS AFTER
R THE INSTAL THE RBE
LLATION OF T
The RBE coonsists on a blade tip thhat is attacheed to the oriiginal bladess in order too extend the rotor
blades lenggth. With thee increase of
o the swept rotor area, this is a sollution to im mprove the energy
e
production on wind farmms. An illusttration of thee RBE is pressented in Figu
ure 5.43.
The RBE in nstallation coonsists on thhe overlappiing of the original rotorr blade tip bby two half shells.
These two shells are thhen fixed witth adhesive resin. With this solution n, the lengthh of each blade is
extended wiith 1.5 m len
ngth tips, inccreasing the sswept area off the rotor in
n 8.6 %. Eachh blade tip weights
w
45 kg, whicch representss an increase of 1 % of the blade mass.
m The maain propertiees of the RB BE are
summarized d in Table 5.88.
For the anaalysis of the dynamic
d chaaracteristics oof the wind turbine
t bladees after the R
RBE, only the data
collected byy the measureement system
m based on fi fibre Bragg grrating system
m was considdered.
Table 5.8 – Main characte

eristics of the RBE
R
applied
d to the Izar Bon
nus 1.3MW/62
Leength of RBE 3.0 m
Extenssion of the blad

de 1.5 m
Upd
dated diameterr 65 m
Increasee swept rotor area

a 8.6 %
Weight of RBE (per blaade) 45 kg Figure 5.43

3 – Illustration of
o the RBE (Salaavessa, 2014)
Variation of blade maass +1%
5.8.5.1 Analysis of thee fibre Braggg grating senssors measureements

The data coonsidered fo
or the analyssis corresponnds to a setu
up of 18 minutes in whhich the roto or was
parked. Thee mean windd speed obserrved during tthe setup wass 7.5 m/s. In this data sett, all the 12 seensors
were availabble.
The averageed normalizeed power speectrum densiity of each seensor is preseented in Figuure 5.44. As can
c be
seen, the sppectra presennt a shape thhat is similaar with the one
o shown in n Figure 5.333 (related to
o data
collected prrior to the in
nstallation off the RBE). HHowever, thhere is a decrrease in the natural freq
quency
values of thee vibration modes.
m
198
Chapter 5
0 Averaged Normalized Power Spectral Density

10
Sensor E
Sensor C
Sensor A
-2 Sensor K
10
Amplitude -4
10
-6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 5.44 – Averaged normalized power spectrum density from each sensor (the dashed lines indicate the main peaks
of the spectra)
The decrease in the natural frequency values is more evident when the signals collected before and
after the RBE are confronted. Figure 5.45 shows the averaged normalized power spectrum density
functions of the signals collected before and after the extension of the blades. Although, the global
shape of the spectra is kept, there is an evident decrease of the frequency values of the resonance peaks.
0 Flapwise -1 Edgewise
10 10
Before Before
After After
-2
10
-2
10
Amplitude
Amplitude
-3
10
-4
10
-4
10
-6 -5
10 10
0 1 2 3 4 5 0 1 2 3 4 5
Figure 5.45 – Averaged normalized power spectrum density of the three blades before and after the RBE
The identification of the modal properties of the wind turbine after the RBE was performed with the
same output-only algorithms. Once again, the results obtained with both algorithms were similar. In
order to illustrate these results, the stabilization diagram obtained with the p-LSCF algorithm is
presented in Figure 5.46.
199

40
35
30
25
Model Order
20
15
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 5.46 – Stabilization diagram obtained with the p-LSCF algorithm for the data obtained after the RBE
Several vibrations modes were again identified. However, due to the difficulty in the interpretation of
their mode shapes, only the first 8 modes were considered. The results are presented in Table 5.9. The
correspondence between the modes identified before and after the RBE was done considering the type
of mode (in plane or out of plane) and the proximity between natural frequencies
From the results, it can be seen that the first two modes (tower bending modes) are only slightly
affected by the RBE. However, the accuracy on the identification of these modes is lower than for the
others since the main element of motion (the tower) is not instrumented.
The remaining modes, mainly related with motion of the rotor, present a consistent decrease on the
value of the natural frequencies. It was noted that for modes 6 and 7 there is a change on the mode
configuration (from yaw to tilt and vice versa). This might be motivated by a different position of the
blades in the rotor plane during the two testing periods.
200
Chapter 5
Table 5.9 – Natural frequencies ( ) and modal damping ratios ( ) from the first 8 identified vibration modes before and
after the RBE (p-LSCF algorithm)
Before RBE After RBE

Mode Δ [%]
[Hz] [%] Description [Hz] [%] Description
1 0.441 0.38 1st Tower SS 0.432 2.69 1st Tower SS -2.0
2 0.456 0.17 1st Tower FA 0.456 0.10 1st Tower FA 0.0
In plane (symmetric) In plane (symmetric)

3 0.610 8.51 + 0.589 6.39 + -3.4
drive train drive train
4 1.129 1.93 In plane (vertical) 1.068 2.96 In plane (vertical) -5.4
5 1.146 2.51 In plane (horizontal) 1.095 2.98 In plane (horizontal) -4.5
6 1.329 0.59 Out of plane (yaw) 1.286 0.75 Out of plane (tilt) -3.2
7 1.449 0.47 Out of plane (tilt) 1.425 0.76 Out of plane (yaw) -1.7
Out of plane Out of plane

8 2.039 0.54 1.949 0.70 -4.4
(symmetric) (symmetric)
5.8.6 CONCLUSIONS
This section presented the work carried out in the experimental identification of the dynamic
properties of an Izar Bonus 1.3MW/62 wind turbine. In this study, different measurement systems
were used to collect vibration signals from the tower (accelerometers) and blades (fibre Bragg grating
sensors and accelerometers).
The analysis of the data allowed to identify 14 vibration modes, including the first two pairs of tower
bending modes and the 6 first order rotor blades modes. The results obtained with both measurement
systems revealed to be consistent.
After the installation of the Rotor Blade Extension, and the consequent increase of blades mass in 1 %,
the decrease of the rotor modes natural frequencies was clearly identified, showing the ability of the
used tools to identify small changes of the structural elements of the wind turbine.
A rotor-stop test was also performed, enabling the analysis of the modal damping ratio of the 1st tower
bending mode in the FA direction and the 1st symmetric flapwise mode with greater precision. In that
sense, two alternative procedures were used. A good coherence of results was obtained with both
methods.
201
Operational Mod
W Turbines
5.9 CASEE STUDY – VESTAS V90-3.0MW
5.9.1 INTTRODUCTION
N
This section
n presents th
he results obtained with the analysis of 80 acceleeration data ssets collected
d by a
dynamic monitoring syystem installeed at a Vestaas V90-3.0MMW offshore wind turbinne. This stud dy was
developed wwith the aim
m of verifying the suitabbility of the developed
d toools when appplied to offfshore
wind turbinnes.
The data ussed in this analysis
a was kindly proviided by the Offshore Wiind Infrastruucture Appliccation
Lab (OWI-LLab - www.o owi-lab.be/).
5.9.2 WIN
ND TURBINE DESCRIPTION
The Vestas V90-3.0MWW wind turbinne is part of the Belwind d wind farm, 46 km off thhe Belgian co
ost, on
the Bligh B
Bank, North Sea (Figure 5.47). This offshore wiind farm is composed bby 55 Vestas V90-
3.0MW turbbines foundeed on monop
piles, leadingg to a total po
ower output capacity
c of 1 65 MW.
• Belwind
B
BLIG H
•B
ZA
ANDBANK
46 km
• Ze
eebrugge
• Brugge
• Oost en
nde
Figure 5.47
5 – Location of the Belwind
d wind farm (Be
elwind Offshore Energy, 20099)
The studied
d Vestas wind d turbine is characterized
c d by a 90 m diameter, up
pwind rotor, with a rated
d rotor
speed of 166.1 rpm. Thee pitch anglee of the threee rotor blad
des is contro
olled by a piitch actuatorr. The
power curvee of the turbiine is introdu
uced in Figurre 5.48.
202
Chapter 5
3000
2600
2200
Power [kW]
1800
1400
1000
600
200
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Wind sp
peed [m/s]
Figure 5.488 – Vestas V90--3MW power cu

urve
The ssupport struccture of the wind

w turbinee is a monop pile with a diameter of 5 m. Its penetration depth h
is 20..6 m. The trransition piece is 25 m hhigh and thee interface beetween the ttower and th he transition
n
piecee is 17 m abo ove sea level. The water ddepth at the turbine location is 22.9 m and the hub height is,,
on avverage, 72 m above sea leevel. Figure 55.49 illustratees the main dimensions
d oof the wind turbine. Thee
soil aat the location
n is considerred stiff, mosstly consisting of sand (Devriendt, Weeijtjens et al., 2014).
72 m
17 m
LAT 25 m
22.9 m
20.6 m
Fiigure 5.49 – Vestas V90-3.0MW

W wind turbinee installed at th
he Belwind wind
d farm (Belwindd Offshore Ene
ergy, 2009)
203
3
5.9.3 MONITORING SYSTEM DESCRIPTION

The Vestas wind turbine is instrumented with several sensors, among which accelerometers, strain
gauges, load cells and displacement sensors (Weijtjens, 2014). Alongside, a subset of the SCADA data
was made available by the wind farm owner. In the scope of this study, only the accelerometer data,
together with the respective SCADA data, was considered.
The accelerometers are installed at the tower and transition piece, using 10 sensors distributed along 4
levels of measurements (19 m, 27 m, 41 m and 69 m above sea level). At the 4 levels of measurement, 2
sensors record the horizontal motion in both orthogonal directions, while 2 additional accelerometers
are installed at the top level to capture the torsional motion of the support structure (Devriendt,
Magalhães et al., 2014). The measurement locations are illustrated in Figure 5.50.
The data acquisition system is mounted at the transition piece. The accelerometer data is acquired with
a sampling rate of 5 kHz and then re-sampled with a sampling frequency of 12.5 Hz. Lastly, the
coordinate transformation introduced in section 5.6.1 is performed to the recorded signals to orientate
them according to the FA and SS directions.
Figure 5.50 – Position of the accelerometers at the different levels of the wind turbine (Devriendt, Magalhães et al., 2014)
5.9.4 ACCELERATION DATA SETS

The data analysed in this section is referred to 80 setups of 10 minutes, grouped into 4 different
operation conditions.
Case 1 is referred to a situation in which the turbine is not operating. Figure 5.51 presents an example
of a data set from this case. As can be seen in the SCADA data plotted in Figure 5.52, the wind turbine
is idling with large values of blades pitch angle. Under these conditions, the wind turbine support
structure mainly vibrates in the SS direction probably due to the larger opposition of the blades to the
air flow in this direction (due to large pitch angle values) and to some possible misalignment of the
rotor to the wind (due to the fact that the turbine is not operating). This is visible in the bottom plot of
Figure 5.51, where the RMS values of the acceleration from each sensor are shown (sensors 3 and 4
were not considered in the plot).
204
Chapter 5
09/05/2012 03:57
Rotor speed
0.01 0.4
[rpm]
Acc. [g]
0.2
0 0
0 5 10 15 20
-0.01 Index
+69 FA 88.6
Pitch [º]
+69 SS 88.4
0 100 200 300 400 500 600
+41 FA 88.2
Time [s] 0 5 10 15 20
+41 SS Index
-3 +27 FA 168
Yaw [º]
x 10 RMS +27 SS 166
5
+19 FA 164
4 0 5 10 15 20
+19 SS Index
Acc. [g]
Wind speed
3 8
[m/s]
2 6
1 4
0 5 10 15 20
Index
0 5 10 15 20
Data set
Figure 5.51 – Example of measured acceleration time series (top Figure 5.52 – SCADA data of the 20 data sets from
plot) and RMS values of the 20 data sets from Case 1 (bottom plot) Case 1
The remaining 3 cases are referred to production periods under different conditions. The data sets
from Case 2 consist of acceleration time series during periods when the turbine is operating close to
the rated wind speed, with the highest wind speed values from the 4 cases (Figure 5.54). The
acceleration levels from the 20 data sets of this case present also the largest values (Figure 5.53). Under
these conditions, with the rotor orientated to the main wind direction and the blades with low pitch
angles, the turbine support structure vibrates mainly in the FA direction. It is also interesting to note
that the measurement level with the highest vibration level is not the top level (+ 69 m) like in the data
from Case 1, but the + 41 m level. In fact, it is observed that the top level present similar vibration
levels to the +27 m level. This phenomenon is analysed in Chapter 7 in another case study.
01/05/2012 02:16
Rotor speed
0.05
16.04
[rpm]
Acc. [g]
16.02
0
16
0 5 10 15 20
-0.05 Index
+69 FA 20
Pitch [º]
+69 SS 0
0 100 200 300 400 500 600
+41 FA 20
Time [s] 0 5 10 15 20
+41 SS Index
RMS +27 FA 80
Yaw [º]
0.02 +27 SS 60
+19 FA 40
0.015 0 5 10 15 20
+19 SS Index
Acc. [g]
Wind speed
20
0.01
[m/s]
15
0.005 10
0 5 10 15 20
Index
0 5 10 15 20
Data set
plot) and RMS values of the 20 data sets from case 2 (bottom plot) case 2
Case 3 consists of data sets with varying rotor speeds (Figure 5.56). Similarly to case 2, the turbine also
presents the highest acceleration levels of the support structure at +41 m level (Figure 5.55).
205
04/05/2012 01:06
0.04
Rotor speed
0.02 16
[rpm]
Acc. [g]
14
0 12
0 5 10 15 20
-0.02 Index
+69 FA -2
Pitch [º]
-0.04
0 100 200 300 400 500 600 +69 SS -2.5
Time [s] +41 FA -3
0 5 10 15 20
+41 SS Index
-3 +27 FA 300
Yaw [º]
x 10 RMS
+27 SS 280
8 +19 FA 260
0 5 10 15 20
+19 SS
Acc. [g]
6 Index
Wind speed
10
[m/s]
4
8
2 6
0 5 10 15 20
0 5 10 15 20 Index
Data set
Lastly, Case 4 comprises the time series under operating conditions with the lowest acceleration levels
(Figure 5.57). This case is referred to data sets in which the turbine is operating with a nearly constant
rotor speed of 10 rpm (Figure 5.58). The only exception is the 6th data set, where a stoppage of the
operation seems to occur.
04/05/2012 09:16
0.02
Rotor speed
15
[rpm]
Acc. [g]
10
0 5
0 5 10 15 20
Index
+69 FA 10
Pitch [º]
-0.02 +69 SS 0
0 100 200 300 400 500 600
+41 FA -10
Time [s] 0 5 10 15 20
+41 SS Index
-3 +27 FA 350
Yaw [º]
x 10 RMS +27 SS 300

5
+19 FA 250
4 0 5 10 15 20
+19 SS Index
Acc. [g]
Wind speed
3 6
[m/s]
2 4
1 2
0 5 10 15 20
Index
0 5 10 15 20
Data set
The colour map shown in Figure 5.59 illustrates the distribution of energy in the analysed frequency
range (0 – 4.5 Hz) using the first singular value spectra of the 80 acceleration data sets within the
frequency range 0 - 4.5 Hz. The frequency values corresponding to the harmonics 3Ω to 24Ω are also
represented by dashed lines. From an initial analysis of the figure, a vertical alignment with high
energy around 0.36 Hz is immediately noticed throughout the 4 studied cases. This alignment is
probably due to the 1st pair of support structure bending modes, since its value is within the common
values of these modes. It is also interesting to observe the large scattering of energy around the
frequency values below this alignment. This situation is probably due to the wave excitation. As
206
Chapter 5
introduced in Chapter 4, the range of frequencies of wave excitation is usually spread along a relative
wide range of frequencies below the 1st pair of bending modes.
The presented colour map also highlights an important aspect in the analysis of the data from this
offshore wind turbine. The energy scattering of the 4 cases is very different from each other. The first
20 data sets (referred to Case 1), present a clean spread of energy, with three well-defined alignments
around 1.20, 1.44 and 1.50 Hz and three additional not so well-defined alignments around 3.24, 3.60
and 3.90 Hz, besides the aforementioned alignment around 0.36 Hz. However, for the remaining 3
cases (corresponding to operating conditions), only the alignment around 0.36 Hz is clearly identified,
due to the important influence of the rotor harmonics in the spectra. Notwithstanding, a frequency
range of high energy is still seen around 1.50 Hz, although with a considerably higher dispersion than
in Case 1. Alignments of high energy corresponding to the possible modes around 3.24, 3.60 and 3.90
Hz are also hard to identify in Figure 5.59 in these 3 operating cases.
3Ω 6Ω 9Ω 12Ω 15Ω 18Ω 21Ω 24Ω
80
70
60
50
Data set
40
30
20
10
0.5 1 1.5 2 2.5 3 3.5 4

Frequency [Hz]
Figure 5.59 – Colour map with the 1st singular value spectra of the 80 data sets (the regions with the hotter colours
represent the highest energy)
Figure 5.60 illustrates the difference in the spectra from operating and non-operating conditions. As
expected, the presence of harmonics is clearly evident. The 3 alignments around 1.20, 1.44 and 1.50 Hz
visible in Figure 5.59 are also easily identified in Figure 5.60 for non-operating conditions. Under
operating conditions, the obtained spectrum is apparently changed in this frequency range. The peak
corresponding to the vibration mode around 1.20 Hz is not present in this frequency. At the same
time, a new (but smaller) peak with a slightly lower frequency value (around 1.17 Hz) is found for the
turbine under operating conditions. This may be a sign that this mode corresponds to a rotor vibration
mode, changing its frequency value according to the pitch angle of the blades. Two very close peaks
around 1.42 Hz and 1.46 Hz are found in the spectra corresponding to operating conditions, while
only the latter is identified in the spectra from non-operating conditions. In addition, the resonance
peak around 1.50 Hz found under non-operating conditions is apparently not identified under
operating conditions due to the interference of the 6Ω harmonic in this frequency range.
The three modes around 3.24, 3.60 and 3.90 Hz, identified in the colour map referred to the Case 1 are
also identified in Figure 5.60 for both situations. However, it is expected that some harmonics may
affect the quality of the identification process of these modes. It is interesting to note that these 3
modes do not present a well identified resonance peak as the other modes do. This is the reason why
the energy alignments of these modes in Figure 5.59 are not as clear as the others.
207
Lastly, the spectra in Figure 5.60 also allow assessing the influence of the wave excitation in the
dynamic response of the wind turbine support structure. This excitation is spread in small peaks with
frequency values below the first resonance peak (around 0.36 Hz).
0 3Ω 6Ω 9Ω 12Ω 15Ω
10
Non-operating
Operating
-2
10
Amplitude
-4
10
-6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Frequency [Hz]
Figure 5.60 – ANPSD from two different setups: under non-operating conditions (rotor speed = 0 rpm; wind speed = 6
m/s) and operating conditions (rotor speed = 16.0 rpm; wind speed = 17 m/s). The vertical dashed lines indicate the
frequency values of the rotor speed 3Ω and its harmonics
5.9.5 MODAL RESULTS

In the context of the analysis of the modal properties of the Vestas V90 offshore wind turbine, the SSI-
COV and the p-LSCF modal identification algorithms were used. Figure 5.61 presents the stabilization
diagrams obtained with the p-LSCF algorithms to a data set corresponding to non-operating
conditions. From the figure, it is possible to clearly identify 8 alignments of stable poles. These
alignments are in agreement with the previous considerations made for non-operating conditions. The
right side of Figure 5.61 shows a zoom of the same stabilization diagram between 0.30 and 0.40 Hz,
attesting the presence of two closely spaced alignments (corresponding to the first pair of support
structure bending modes). Furthermore, it is also possible to observe the presence of some additional
poles mainly due to the wave excitation.
Poles due to
Non-operating conditions Poles Stab. freq Stab. damp Stab. MAC ANPSD wave excitation 1 FA 1 SS
50
50
40 40
Model Order
30 30
20 20
10 10
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0.3 0.35 0.4
Figure 5.61 – Stabilization diagram obtained with the p-LSCF algorithm in a data set corresponding to non-operating
conditions
208
Chapter 5
Figure 5.62 introduces the stabilization diagram obtained for a data set recorded during operating
conditions with the p-LSCF algorithm. From this figure, it is visible that the rotor harmonics are
identified as alignments of stable poles, just like real vibration modes. Notwithstanding, both the first
pair of bending modes and the modes around 3.24, 3.60 and 3.90 Hz have clear alignments of stable
poles.
The right plot of Figure 5.62 shows a zoom of the stabilization diagram between 1.1 and 2.2 Hz. From
this figure, it is possible to see the existence of stable alignments around 1.17, 1.75 and 2.10 Hz. While
the former is close to a resonance peak identified during non-operating conditions (around 1.20 Hz),
the last two are only identified when the turbine is operating. In addition, two close alignments of
poles are also visible around 1.41 and 1.44 Hz. This situation is different from the one obtained with
the turbine under parked/ idling conditions, where only one vibration mode was identified in this
frequency range. As will be seen ahead, these two modes present a similar mode shape.
Operating conditions Poles Stab. freq Stab. damp Stab. MAC ANPSD
6Ω
3Ω 6Ω 9Ω 12Ω 15Ω
50 50
40 40
Model Order
30 30
20 20
10 10
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 1.2 1.4 1.6 1.8 2 2.2
Figure 5.62 – Stabilization diagram obtained with the p-LSCF algorithm in a data set corresponding to operating
conditions
Since only a limited number of data was analysed, it is not possible to clearly attest the nature of some
vibration modes, as well as the correspondence of some modes between parked/ idling and operating
conditions. Thus, it is not possible to elaborate a clarifying Campbell diagram, which could help to
better interpret the identified vibration modes. If a proper Campbell diagram could be obtained, it
would be possible to check which modes are related to the blades rotor motion as these modes are
highly dependent on the rotor speed when fixed sensors are used, as explained in Chapter 4.
Notwithstanding, some modes are clearly identified and their nature is indubitably understood. This is
the case of the first pair of support structure bending modes, which natural frequencies are kept fairly
constant during non-operating and operating conditions.
The mode shapes of the 8 vibration modes identified under non-operating conditions are presented in
Figure 5.63. These modes are identified by the index “non” (from non-operating conditions). The two
first modes are indubitably the first pair of bending modes (in FA and SS direction). The 3rd, 4th and 5th
identified modes present a second-order configuration for their mode shapes in the FA, SS and FA
direction, respectively. However, it is noticed that 3rd mode presents practically null displacement at
the top sensor, while the other two show some motion at this level. The last three identified modes also
present a similar configuration, potentially a 3rd order shape configuration. However, since the first
209
level of measurement is around + 19 m, it is not possible to accurately estimate the configuration of

this mode in the lower part of the support structure.
, = 0.37 Hz (FA) , = 0.37 Hz (SS) , = 1.21 Hz (FA)

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-20 -20 -20

-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
, = 1.44 Hz (SS) , = 1.57 Hz (FA) , = 3.24 Hz (FA)

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-20 -20 -20

-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
, = 3.60 Hz (SS) , = 3.91 Hz (FA)

80 80
60 60
40 40
20 20
0 0
-20 -20
-1.2 0 1.2 -1.2 0 1.2
Figure 5.63 – Tower mode shapes of the identified vibration modes under non-operating conditions
210
Chapter 5
, . = 0.37 Hz (FA) , . = 0.37 Hz (SS) , . = 1.17 Hz (SS)

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-20 -20 -20

-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
, . = 1.39 Hz (FA) , . = 1.45 Hz (FA) , . = 1.75 Hz (SS)

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-20 -20 -20

-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
, . = 2.09 Hz (SS) , . = 3.27 Hz (SS) , . = 3.43 Hz (FA)

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-20 -20 -20

-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
, . = 3.92 Hz (SS)
80
60
40
20
-20
-1.2 0 1.2
Figure 5.64 – Tower mode shapes of the identified vibration modes under operating conditions
211
Figure 5.64 introduces the mode shapes of the 10 identified vibration modes during operating
conditions, identified with the index “op.”. The two first modes are the 1st pair of support structure
bending modes, also identified during non-operating conditions. The 3rd, 6th and 7th are modes
vibrating in the SS direction that were not identified with the data corresponding to Case 1. Thus,
these are modes potentially linked to asymmetric edgewise blades rotor motion, since this type of
modes creates a reaction at the top of the tower in the SS direction, due to the configuration of the
blades with low pitch angles (contrary to the configuration in standstill conditions, where the pitch
angle is usually around 80º - 90º). On the other hand, the configurations of 4th and 5th modes (both in
FA direction) resemble the ones from the 3rd and 5th modes under parked/ idling conditions. However,
a mode with a configuration similar to the 4th mode (in the SS direction) was not identified in the
analysed data sets, turning it difficult to assess the nature of these modes. Lastly, the 8th, 9th and 10th
modes show a similar configuration to the 6th, 7th and 8th modes identified under non-operating
conditions. However, it is interesting to note that all of them changed their direction of vibration. The
reasons behind this phenomenon are probably related to the interference of the blades pitch angle in
the motion of these modes. A more profound comprehension of these modes would require the
analysis of a more extended period of data.
After the main vibration modes were preliminary analysed, all the data was automated processed in
order to obtain the modal properties of the introduced modes. After the application of the output-only
identification algorithms to the recorded acceleration time series, the cluster algorithm described in
section 5.7.1 was applied to the obtained stable poles. A maximum distance of 0.02 was defined for the
single linkage criterion. Figure 5.65 shows the computed clusters from the stabilization diagrams
represented in Figures 5.61 and 5.62, quantifying the average frequency value of the poles included in
each cluster and the number of poles of each cluster. In this analysis, it was decided to exclude the
clusters with a number of poles lower than 6, as identified by the red dashed line.
Non-operating conditions Operating conditions
50 50
40 40
30 30
Nº of poles
Nº of poles
20 20
10 10
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 5.65 – Characterization of the clusters: average frequency and number of poles included in each cluster
A representative value of the modal properties from each cluster (frequency value, damping ratio and
mode shape) was kept. Figure 5.66 shows the representative frequency value of the considered clusters
for each data set. In this figure, each selected results was defined as FA or SS, according to the main
direction of vibration of its mode shape. This figure is illustrative of the increased difficulty in the
identification of the vibration modes under operating conditions. While in Case 1 the considered poles
present a very stable and clear evolution of their frequency values, a part from some disturbance
212
Chapter 5
around the 1st pair of bending modes due to the wave excitation, the same does not occur for the
remaining cases. Under operating conditions, the appearance of relevant clusters due to the presence
of the harmonics is evident. Besides the harmonics, this figure allows to confirm some previous
referred considerations. For example, it is evident that the last 3 modes detected for both under
parked/ idling and operating conditions inverted theirs main direction of vibration. Furthermore, both
the 6, . and 7, . modes are clearly distinguished once the turbine is operating (although not in
every case). Lastly, the two closely spaced modes 4, . and 5, . are also identifiable only in cases 2, 3
and 4.
FA
SS
4.5
4 24Ω
3.5 21Ω
3 18Ω
Frequency [Hz]
2.5 15Ω
2 12Ω
1.5 9Ω
1 6Ω
0.5 3Ω
0
0 10 20 30 40 50 60 70 80
Data set
Figure 5.66 – Considered poles from each data set according to the main direction of vibration of the mode shape (the
dashed lines refer to the frequency value of the harmonics of each data set)
In order to identify the selected reference vibration modes from the poles obtained with the
identification algorithms, the modal properties presented in Figures 7.63 and 7.64 were used as
reference properties of the modes. Thus, for each data set, the clusters with a frequency value close to a
reference natural frequency value were selected. The frequency range adopted to consider a pole varied
between 10 % (for the 1st pair of bending modes) and 20 % (for the others). Within the considered
poles, the one with highest correlation with the reference mode shape (i.e. the highest value of MAC
coefficient) is selected. Only modes with MAC values higher than 0.80 were considered.
This procedure was applied together with the p-LSCF and SSI-COV modal identification algorithms,
as already stated. For the p-LSCF method, positive time lags of the correlation with 1024 points were
used to calculate the spectra, together with an exponential window with a factor of 0.1. A maximum
model order of 50 was considered. In order to assess the stability of a pole, the criteria exhibited in
Table 5.10 were followed. These settings were already used for the results previously presented in this
section.
213
Table 5.10 – Stable pole criteria for models of consecutive orders
Modal parameter Maximum allowed variation
Frequency Δf ≤ 1 %
Damping Δξ ≤ 5 %
Mode Shape MAC ≥ 0.97
Figure 5.67 presents the results obtained after the comparison of the selected cluster properties with
the properties of the reference modes. From this figure, it can be seen that the modes were successfully
identified throughout all the cases. Only the identification of the 3, . mode shows a poor success
rate, being only identified in case 2. In this figure, the colours selected for the tracked modes from Case
1 purposely does not match with the colours from the other Cases since it was not intended to link the
identified vibration modes from non-operating conditions with the ones from operating conditions.
Case 1 Case 2 Case 3 Case 4 Case 1 Case 2, 3, 4
4.5
1,non (FA) 1,op. (FA)
4 2,non (SS) 2,op. (SS)
3,non (FA) 3,op. (SS)
3.5
4,non (SS) 4,op. (FA)
3 5,non (FA) 5,op. (FA)
Frequency [Hz]
6,non (FA) 6,op. (SS)

2.5 7,non (SS) 7,op. (SS)
2 8,non (FA) 8,op. (SS)
9,op. (FA)
1.5 10,op. (SS)
1
0.5
0
0 10 20 30 40 50 60 70 80
Data set
Figure 5.67 – Tracked vibration modes
The identified modal properties of the vibration modes with the p-LSCF algorithm throughout the 4
cases are summarized in Table 5.11. From these results, it is interesting to note that, under non-
operating conditions, the 2nd mode (the 1st bending mode in the SS direction) presents higher damping
values than its pair in the FA direction. However, under operating conditions, this situation is
inverted, being the mode in the FA direction the one with the highest damping values. This situation is
caused by a possible deviation of the rotor from the wind direction and to the high pitch value of the
blades when the turbine is parked/ idling (Case 1). Under this configuration, the blades present a
larger opposition to the air in the SS direction, leading to the appearance of some contribution of
aerodynamic damping in this direction. When the turbine is operating (Case 2, 3 and 4), the rotor is
oriented to the main wind direction and the blades present a low pitch angle. Under these
circumstances, the main opposition to the air by the rotor is presented in the FA direction, vanishing
the aerodynamic contribution in the SS direction. Since the aerodynamic damping component
increases with the increase of the wind speed (assuming similar blades pitch angle), the damping
values of the mode in the FA direction shows its highest values in Case 2, corresponding to the data
sets with the highest wind speed.
214
Chapter 5
Another interesting aspect of the results obtained is related with the damping values of the last 3 mode
shapes of both parked/ idling and operating conditions. It is noted that for non-operating conditions,
the SS vibration mode (7th mode) presents the highest damping values, while the other 2 modes
vibrating in the FA direction present similar damping values. When the turbine is operating, the two
SS modes present similar damping values (8th and 10th modes), while the 9th mode (in the FA direction)
shows the highest damping ratio. This correspondence reinforces the idea of these modes found in
Case 2, 3 and 4 may be linked to the last three modes found in 1 case.
Table 5.11 – Mean values of natural frequency and damping ratio obtained with the p-LSCF algorithm
Case 1 Case 2 Case 3 Case 4
Mode ݂ [Hz] ߦ [%] ݂ [Hz] ߦ [%] ݂ [Hz] ߦ [%] ݂ [Hz] ߦ [%]
1 0.374 1.77 0.364 4.68 0.375 2.16 0.373 2.87

2 0.366 2.23 0.368 1.66 0.361 1.56 0.365 1.40
3 1.207 0.65 1.174 0.77 - - - -
4 1.444 1.20 1.393 1.39 1.390 1.52 1.428 0.88
5 1.571 1.01 1.453 1.00 1.457 1.30 1.487 0.89
6 3.239 0.56 1.746 1.10 - - - -
7 3.600 1.70 2.085 1.09 2.126 0.80 2.054 1.16
8 3.907 0.89 3.269 0.67 3.278 0.93 3.290 0.79
9 - - 3.423 1.19 3.484 1.49 3.503 1.65
10 - - 3.910 0.49 3.958 1.07 3.960 0.52
The same strategy was also followed with the SSI-COV algorithm. For this method, 512 points of the
correlation functions were used. The state-space models were defined with a maximum order of 100.
Table 5.12 sums up the main results obtained with this algorithm. Although similar, the results
attained with this algorithm were not as good as the ones obtained with the p-LSCF, for the analysed
data. This is particularly relevant due to non-identification of 3rd and 7th vibration mode during
operating conditions.
215
Table 5.12 – Mean values of natural frequency and damping ratio obtained with the SSI-COV algorithm
Case 1 Case 2 Case 3 Case 4
Mode [Hz] [%] [Hz] [%] [Hz] [%] [Hz] [%]
1 0.366 1.66 0.365 6.78 0.370 5.19 0.366 4.23

2 0.366 2.01 0.369 2.07 0.365 2.40 0.364 1.91
3 1.207 0.57 - - - - - -
4 1.444 1.29 1.397 1.23 1.399 1.24 1.413 0.90
5 1.573 1.12 1.444 1.60 1.457 0.79 1.473 1.19
6 3.236 0.72 1.741 0.95 - - - -
7 3.591 1.51 - - - - - -
8 3.902 1.02 3.273 0.57 3.281 0.61 3.288 0.61
9 - - 3.458 1.93 3.494 2.01 3.483 1.13
10 - - 3.923 0.56 3.935 0.72 3.978 0.21
5.9.6 CONCLUSIONS
This section presented the results obtained with the analysis of vibration data collected during a short
period of time from a Vestas V90-3.0MW offshore wind turbine. In this analysis, four different
scenarios of operation were studied, including idling and power production conditions.
The analysis undertaken permitted to identify 8 vibration modes during non-operating conditions and
10 modes during power production conditions. It was noticed that the dynamic properties of the wind
turbine are considerably different depending if the system is operating or not.
An automated procedure was successfully applied to the data to automatically track the modal
properties of the identified vibration modes throughout the different cases. The results obtained gave a
good indication about the possibility of development of a generic dynamic monitoring system to be
installed in both onshore and offshore, as exposed in (Oliveira, Weijtjens et al., 2014).
Due to the restricted period of analysis, it was not possible to attest the accuracy of the monitoring
system to detect small damages on the support structure of the wind turbine. Notwithstanding, work
developed on the same wind turbine using a long period of data revealed a good accuracy in the
detection of structural changes in the foundation (Weijtjens, Verbelen et al., 2015).
216
Chapter 6
6
FATIGUE ASSESSMENT OF WIND
TURBINES
6.1 INTRODUCTION TO FATIGUE PROBLEMS IN WIND TURBINES

Wind turbines are structures predominantly subjected to dynamic loads. During a life period of 20
years, long and flexible blades from a large rotating component (the rotor) suffer from both the
turbulent wind and the gradient wind shear. This rotating equipment is placed at the top of a slender
structure (prone to resonance problems), suffering from a cyclic wind flow disturbance (the tower
shadow effect). This scenario illustrates the harsh environment that the wind turbine structure has to
withstand.
From a fatigue point of view, the design of a wind turbine is a very complex task. Wind turbines are
composed by large structural components designed with several materials, leading to different
properties and assessment procedures (Sutherland, 2000). Also, the definition of the loading scenarios
is a difficult and extensive task (especially for offshore turbines). Apart from quasi-static loads (such as
sea current), environmental loads acting on the wind turbines show a dynamic, non-deterministic
nature, leading to a considerable uncertainty in their quantification and frequency of occurrence.
Some published studies refer the complexity and uncertainty associated with the fatigue assessment of
wind turbines support structures. In (Tempel, 2006), the fatigue assessment of the monopile support
structure was performed using a numerical frequency-domain approach. In this work, a large number
of environmental conditions were analysed, aiming to reproduce the wind and sea states.
The study present in (Dong, Moan et al., 2012) shows a different approach for fatigue analysis of
welded connections on offshore jacked support structures for wind turbines. In this work, the
uncertainties associated with several design aspects, such as loading scenarios, dynamic behaviour of
the structure and corrosion, are taken into account to perform a fatigue reliability analysis.
The numerical assessment of the fatigue damage in a floating wind turbine is described in (Kvittem
and Moan, 2015). In this work, the need for a correct tuning of the dynamic properties is said to be
very important to avoid a rapid increase of fatigue damage at the support structure.
The work described in (Weijtjens, Iliopoulos et al., 2015) presents the initial results obtained with the
installation of fibre Bragg gratings strain sensors at the tower and transition piece of the Vestas V90-
3.0 MW introduced earlier in Chapter 5. In the present research, a clear relationship between fatigue
damage rate and wind turbulence intensity is evidenced.
In this work, the fatigue assessment is focused on the tower of the wind turbine.
217
Fatigue Assessment of Wind Turbines
6.2 FATIGUE DESIGN LOADS

Wind turbines are subjected to a large variety of dynamic loads contributing to the fatigue wear of the
structure. There are mainly three sources of fatigue loads for wind turbines:
o Wind;
o Rotor operation (and all the consequences of its rotation, such as the tower shadow effect);
o Waves.
Wind loading is the primary external source of excitation of wind turbines. In Chapter 3, a brief
theoretical explanation about the rotor thrust force introduced by the wind was already developed. In
the context of fatigue analysis, it is also of interest to understand how it dynamically affects the wind
turbine structure.
Time varying wind speed time series are usually assumed as the sum of two components: a mean wind
speed and a turbulent component . Considering the HAWC2 example introduced in Chapter 4, the
two aforementioned components of the wind speed at the hub height are illustrated in Figure 6.1.
19 4
Turbulent component [m/s]

17 2
Mean wind speed [m/s]
15 0
17 -2
11 -4
100 200 300 400 500 600 700
Time [s]
Figure 6.1 – Mean wind speed and turbulent component of the wind speed at the hub height (simulated in the context of
the NREL 5MW wind turbine example)
The turbulent component of the wind can thus be seen as the component responsible for the dynamic
excitation of the wind turbine. It is considered a non-deterministic variable and, for that reason, it is
usually defined by statistical means. A common way to define the turbulence intensity of the wind
in the flow direction is through the ratio between the standard-deviation of the wind speed and its
mean value:
(6.1)
In addition to the turbulence in the mean wind direction, also the lateral and vertical turbulence are
defined similarly.
The turbulence is usually represented in the spectral form. This representation allows understanding
the frequency regions with higher excitation energy. The most commonly used turbulence spectra
218
Chapter 6
models are the von Kármán and the Kaimal spectra (Burton, Sharpe et al., 2001). Figure 6.2 presents
the von Kármán turbulence autospectral density function for several mean wind speeds. As can be
seen, the energy is essentially concentrated at the very low frequency values.
For common wind turbine structures, whose lowest vibration mode is usually around 0.30 – 0.40 Hz,
the energy from the lowest part of the spectrum is mainly responsible for the quasi-static response of
the structure, i. e., with high amplitude displacement cycles with large periods of occurrence. On the
other hand, the remaining frequency range of the spectrum is responsible for the resonance driven
motion of the wind turbine. This means that, in this frequency range, the motion of the wind turbine is
commanded by resonance of the main support structure vibration modes.
The mean component of the wind may be also important for fatigue analysis of some structural details.
For example, the assessment of fatigue damage of some materials (such as composite or concrete) or
elements (such as prestressed bolted connections) is dependent on the mean value of the stress cycles.
Notwithstanding, other elements (such as steel towers) do not require any consideration about the
mean stress level for fatigue analysis according to the main standards/ guidelines (section 6.3.2).
2 Turbulence autospectral density function (longitudinal)
10
0
10
Amplitude
28 m/s
24 m/s
20 m/s
-2 16 m/s
10
12 m/s
8 m/s
-4 4 m/s
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 6.2 – Von Kármán longitudinal turbulent spectra for various mean wind speeds (I = 10 %)
The rotor operation is another source of excitation of wind turbines. As stated in section 4.6, the
rotation of the rotor introduces cyclic loads at integer multiples of the rotor speed. For the wind
turbine support structure, the energy of this excitation will be mainly located at the frequency of 3Ω
and its integer multiples. For a common variable-speed wind turbine (with a rotor operating regime
between 8 and 20 rpm), the 3Ω harmonic excitation presents a frequency between 0.4 Hz and 1.0 Hz.
Figure 6.3 illustrates the change introduced in the aerodynamic rotor thrust by the rotor rotation using
data from the NREL 5MW wind turbine example. In this figure, the (theoretical) aerodynamic force
that would be obtained if the harmonic effect is vanished (“aerodynamic force”) is compared with the
actual aerodynamic rotor thrust calculated by the HAWC2 code (“aerodynamic rotor thrust”). These
forces represent the load transmitted by the rotor blades to the support structure of the wind turbine.
The effect introduced by the rotor rotation is clearly visible, with well-defined peaks at the harmonic
frequencies. It is thus concluded that, during operating periods, the wind load really applied on the
support structure does not follow a turbulence spectrum like the one from Figure 6.2 but one similar to
the presented in Figure 6.3.
219
0
10
Aero. force
-2 Aero. rotor thrust
10
Normalized amplitude
-4
10
-6
10
-8
10
-10
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 6.3 – Averaged normalized spectrum of the aerodynamic force and aerodynamic rotor thrust
At offshore conditions, wave excitation is another important source of dynamic loading of the wind
turbine support structure. Waves are mainly originated as result of the wind action along the surface of
the sea. Thus, just like the wind, waves also present an apparent random behaviour (wave loading is
considered a non-deterministic process too).
The formation of waves can be seen as a combination of several regular waves, defining the sea surface.
It is thus helpful to define the energy distribution in the frequency domain. In that sense, two
important statistical properties should be referred: the significant wave height ( ) and the peak
period ( ). The significant wave height is related to the magnitude of the waves. It is approximately
four times the standard deviation of the sea surface height (Manwell, McGowan et al., 2010). The peak
period is the wave period associated with the most energetic waves (which can be also defined in terms
of frequency - ).
The wave spectrum is an essential tool to assess the loading environment to which the wind turbine
will be subjected. The two most commonly used wave spectra are the Pierson-Moskowitz and the
JONSWAP spectra.
The shape of the Pierson-Moskowitz wave spectrum was fitted to measurements from the Atlantic
Ocean during long periods of constant environment conditions (Pierson and Moskowitz, 1964). This
spectrum only considers the wind speed as input. Later, this spectrum was updated to consider and
as input.
The JONSWAP spectrum corresponds to a modified version of the Pierson-Moskowitz spectrum
(Hasselmann, Barnett et al., 1973). It is based on measurements performed in the North Sea. The
JONSWAP spectrum is characterized by a higher and narrower peak (when compared to the Pierson-
Moskowitz spectrum). For that reason, the JONSWAP spectrum is often used for extreme event
analysis (Manwell, McGowan et al., 2010).
Figure 6.4 presents the 22 representative sea state conditions for the 20 m water depth Offshore Wind
farm Egmond aan Zee (OWEZ), located in the Dutch North Sea (Tempel, 2006). For this figure, the
Pierson-Moskowitz wave spectrum was used. It is seen that the highest energy is related to low
frequency values, below the common values of natural frequency of the 1st support structure bending
mode. Notwithstanding, it is also visible that, although the peak may present a wide spread of energy
220
Chapter 6
for low values of wave height, the energy is almost zero for very low frequencies. This is an indication
that the wave loading may not induce a quasi-static response of the wind turbine.
14
12 P=0.39 %
Power Spectral Density [m2 .s]

10
8
P=0.15 %
6 P=2.34 %
4 P=5.28 %
P=1.79 %
2 P=3.26 %
P=10.48 %
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Frequency [Hz]
Figure 6.4 – Wave spectra from the OWEZ site (the probability of occurrence of the first 7 states with higher wave height
is shown)
221
6.3 APPLICATION TO WIND TURBINE SUPPORT STRUCTURES

6.3.1 MAIN STRUCTURAL DETAILS
Wind turbine structures, due to their geometry and assembly process, have several details that need to
be verified against fatigue. Considering only the tower structure, there are mainly three spots of
interest: the flanged connection between tower segments (including bolts), the welded connection
between the cylindrical sheet and the flange; and the stress concentration due to the door opening at
the bottom of the tower.
a) b) c)
Figure 6.5 – a) Bolted connection between flanges. b) Welded connection between cylindrical sheet and the flange. c)
Door opening at the bottom of the tower
The connection between the tower and the foundation is also another sensitive location for fatigue
(Currie, Saafi et al., 2015). Considering onshore wind turbines, the constant vibration of the structural
system can lead to cracking of the concrete around the ring imbibed in the foundation (see Figure 6.6).
The problem of fatigue is even more severe in offshore wind turbines. The grouted connection of the
transition piece is a very delicate location for fatigue and is subjected to very high stresses (see Figure
3.9). In addition, also some offshore foundation models (such as the tripod and jacket foundations)
need an accurate analysis of fatigue efforts, due to their complicate geometry and high number of
joints.
Figure 6.6 – Detail of the connection between the insert ring and the slab foundation
222
Chapter 6
6.3.2 FATIGUE ANALYSIS ACCORDING TO STANDARDS/ GUIDELINES

The main standards/ guidelines of wind turbines, including (International Electrotechnical
Commission (IEC), 2009; International Electrotechnical Commission (IEC), 2014), (Det Norske
Veritas (DNV), 2011) and (Germanischer Lloyd (GL), 2010; Germanischer Lloyd (GL), 2012), suggest
the application of the linear accumulation of damage for fatigue assessment of elements of the support
structure. These standards refer the use of a S-N curve according to the detail under analysis, together
with the Rainflow Counting method for determination of the histogram with the stress cycles.
For the design of steel support structures, all the standards/ guidelines define similar S-N curve
equations for fatigue analysis. As example, (Germanischer Lloyd (GL), 2010) defines the same S-N
curve as the EC 3 but disregards the threshold value of the fatigue strength. The S-N curve defined in
EC 3 and GL for a particular detail is presented in Figure 6.7.
3
10
Stress range ( Δσ R )
2
10
Threshold defined in
EC3
1
10
4 5 6 7 8 9
10 10 10 10 10 10
Number of cycles (N)
Figure 6.7 – S-N curve defined in GL and EC 3 (detail 71)
In addition, corrections to the S-N curves are proposed in some standards/ guidelines, considering
thickness, corrosion and material effects, among others.
The consideration of the mean stress in fatigue analysis is also referred in the standards/ guidelines,
since different structural details required different approaches:
o The fatigue analysis of concrete elements, such foundations and grouted joints, requires the
knowledge of the mean stress;
o The study of fatigue state of the pre-loaded bolted joints also requires the consideration of the
mean stress level. This is due to the non-linear relation between the stress state of the tower
and the forces in the bolts (Sørensen and Sørensen, 2011);
o For the analysis of steel elements from the support structure, the mean value of the stress
cycles is usually disregarded. Some standards/ guidelines present a correction coefficient to
consider the mean stress effect. However, this consideration only introduces a positive effect
when a portion (or the whole range) of the stress range is in the compression side. Thus, in
this context, the disregarding of this coefficient (and consequently, of the mean stress effect) is
conservative.
These documents also define partial safety factors to account for the uncertainties about material,
loads and consequences of failure (although some documents do not define each value separately).
223
Both IEC, GL and EC 3 consider the safety factors that increase the loads. On the other hand, DNV
defines a safety factor (named Design Fatigue Factor - DFF) which is considered as a magnification
factor of the accumulated damage . As example, the safety factors defined in the referred documents
for the steel elements of an onshore tower are summarized in Table 6.1.
Table 6.1 – Partial safety factors defined in standards/ guidelines
Standard/ Consequences of
Load Material Total
Guideline failure
GL 1.0 - 1.15 1.15
EC 3 1.0 1.15 - 1.15
IEC* 1.0 1.1 1.15 1.265
DNV* - - - 2.00
* These documents are referred to offshore wind turbines
224
Chapter 6
6.4 FATIGUE ASSESSMENT USING ACCELEROMETERS

The structural response due to fatigue loading is usually monitored using strain gauges. The use of
strain gauges requires the installation of sensors in each section under study. This usually requires
scrapping the point of the structure, which is not well regarded by the wind farm owners. Other
important drawback about using strain gauges is their sensitivity to harsh environment. Thus, it is
difficult and expensive to estimate the stress condition of structural details of offshore wind turbine
under the water level with these sensors. This represents an important issue, since the inspection of
these elements is only possible using underwater equipment.
In that sense, a method enabling the analysis of the fatigue condition along the entire wind turbine
support structure would represent a useful tool for the management of a wind farm. This would be
especially important for offshore turbines, since it would allow estimating the condition of underwater
structural details.
One important aspect of the wind turbines structures is that they are highly dynamic systems,
subjected to important dynamic loading excitations, as referred in section 6.2. Consequently, the use of
a sensor capable of recording this dynamic motion would likely be an important source of information
for fatigue analysis. Thus, it seems legitimate to use accelerometers to evaluate the dynamic response of
wind turbine support structures and then perform the fatigue analysis with this input.
Displacements (and, consequently, stresses) can theoretically be computed from acceleration data.
Considering the acceleration function given by , the velocity and displacement functions are given, in
the time domain, by the single and double integration of , respectively:
(6.2)
with:
, , Displacement, velocity and acceleration functions (in time domain)
Initial velocity
Initial displacement
Nonetheless, the computation of displacements from accelerations cannot be performed directly, as it

may lead to considerable errors. Observing the previous equation, it is immediately noted that the two
initial conditions ( and ) are not known and, if they are not null, they will introduce errors in the
double integration process. These errors are illustrated in Figure 6.8. Both illustrated displacement
functions departed from the same acceleration function but with different initial conditions. As can be
seen, there is an initial offset (due to ) and a drift along the function (due to ) that leads to a
considerable lag between the two functions.
225
1
y0 = 0.2
.
y0 = 0.06
0.5
Displacement
0
y0 = 0
.
y0 = 0
-0.5
-1
0 2 4 6 8 10 12 14
Time
Figure 6.8 – Results from double integration of acceleration functions with different initial conditions
Notwithstanding, there are some methodologies published regarding the estimation of the stress time
history from accelerometers. However, they are focused on the dynamic component of the
displacements, i. e., the quasi-static displacements, which depend on the initial conditions and on the
static loads, are disregarded.
In (Hjelm, Brincker et al., 2005) and (Aenlle, Skafte et al., 2013), an interesting methodology is
presented. This procedure uses the measured accelerations to estimate the dynamic properties of the
structure. Alongside, the measured accelerations are integrated twice to obtain an estimation of the
displacements. Using the estimated displacements and the experimentally identified vibration modes,
the modal coordinates can be estimated according to:
, . , (6.3)
with:
Estimated displacements
, Mode shapes experimentally obtained
, Modal coordinates
Number of considered vibration modes
Equation (6.3) can be solved directly if the number of mode shapes equals the measured degrees of
freedom. If the measured degrees of freedom exceed the number of vibration modes considered, the
equation can be solved with the least square method.
226
Chapter 6
Once the dynamic characteristics are known, a finite-element model of the structure is developed,
which is calibrated and updated based on the experimental results. With the numerically obtained
mode shapes, it is possible to estimate the displacements at any point of the structure through:
, . , (6.4)
with:
, Mode shapes obtained from the finite-element model
, Modal coordinates obtained from equation (6.3)
Considering that the relationship between bending moment and curvature at a given section of the
structure is defined by:
. . (6.5)
with:
Young modulus of the material
Second moment of inertia of the cross section
and that the bending stress is given by:
(6.6)
with:
Bending stress
Bending moment
Distance from neutral axis to the point of interest of the section
227
the bending stress at any point along the height of the structure ( axis) is then calculated according to
the combination of equations (6.5) and (6.6):
,
, . (6.7)
with:
Mode shape from -th vibration mode
Modal coordinate from -th vibration mode
This formulation implies that the function describing each vibration mode has to be derived twice (in
order to the height) to obtain the stress at any point of the structure. An example of application of the
methodology to a reduced two story building model is presented in (Aenlle, Hermanns et al., 2013).
The described procedure is developed with the support of a finite-element model. However, its
application to wind turbines presents some important drawbacks. Firstly, the need of a properly
calibrated finite-element model of a wind turbine can be an obstacle, since geometrical and material
characteristics of the nacelle and rotor are usually not provided by the manufacturer. In addition, the
direct use of the experimentally obtained mode shapes in equation (6.7), avoiding the need of a finite-
element model, may lead to important errors in the double derivative operation. Lastly, the quasi-static
component of the displacements, mostly due to wind action, is not considered.
In order to overcome these obstacles, a procedure to assess the stress condition at any location of a
wind turbine support structure is presented in next section.
6.4.1 PROPOSED PROCEDURE FOR FATIGUE ASSESSMENT

The procedure for fatigue assessment at any location of the wind turbine support structure is proposed
in this section. The procedure is inspired on the one presented in the previous section, albeit with
some modifications.
The procedure is based on the estimation of the acceleration modal responses at any point of the wind
turbine support structure. These (modal) acceleration time series are later integrated into
displacements, which are used to estimate the stress condition of the support structure.
The procedure starts after the estimation of the modal properties of the wind turbine (with application
of the algorithms presented in section 5.4). The estimated state-space matrices and are then used
to define a forward innovation model to estimate the modal acceleration responses at the sensors
position, as explained in section 5.5.
228
Chapter 6
Once the modal acceleration responses are computed, the modal displacements can be estimated
through a double integration process. In this work, the acceleration time series are transformed into
the frequency domain and thereafter double integrated. The modal displacement of the structure
motivated by the -th vibration mode at one of the sensors position is thus defined as:
(6.8)
with:
Displacement (in the frequency domain)
Acceleration (in the frequency domain)
The obtained displacements are then transformed back to the time domain.
After the displacement field for each vibration mode is achieved, it is possible to compute equivalent
forces imposing the same (modal) deformation. The equivalent forces acting along the height of the
support structure at the time step from the -th vibration mode are obtained according to:
. (6.9)
with:
Equivalent forces from the -th vibration mode
Stiffness matrix of the support structure of the wind turbine
Vector with displacements regarding the -th vibration mode
Time step
Equation (6.9) requires the definition of the stiffness matrix of the wind turbine support structure.
Contrary to the nacelle and rotor, the support structure can be defined through very simple models.
The tower structure can be assumed as a simple cantilevered beam, once the material and structural
characteristics (variation of the bending stiffness along the height) are known.
The total equivalent forces acting along the height of the support structure at the time step are
obtained as a sum of the contribution from the identified vibration modes:
(6.10)
Equation (6.9) defines equivalent loads at the measurements points (where the modal amplitudes of
the mode shapes are known) imposing the same modal deformation of mode . Thus, if the mode
shape is defined by a limited number of points (as is usual the case, since the number of accelerometers
placed along the structure is reduced), equation (6.9) will conduct to large concentrated loads at few
229
points of the structure. In this case, the stress condition obtained for the support structure will not
correspond to reality.
In order to overcome this problem, an interpolation of the experimentally found mode shapes is used
(e.g. spline interpolation) to obtain a more detailed estimation of the modal amplitudes of the mode
along the height of the support structure. As referred in section 5.5, the information about the mode
shapes is contained in the matrix from the forward innovation model (see equation (5.71)). When
the mode shapes are interpolated, a new matrix is defined, named , which can be used to
estimate the accelerations at the interpolation points:
. , (6.11)
Once the acceleration time series at the interpolation points are obtained, the presented procedure can
be used to estimate equivalent loads at these points. Considering that a fine resolution is used in the
interpolating operation, the loads will resemble a distributed load, leading to a stress field in
accordance with the reality.
The interpolating operation has the advantage of avoiding the need of a well calibrated finite-element
model. However, the suitability of the interpolation must be carefully analysed. In situations where the
number of sensors is limited and vibration modes with high order modal configurations are important
for the displacement response, this operation may introduce important errors. In the case of support
structures of wind turbines, the displacement response is mainly conditioned by the first bending
mode and low order harmonics. These mode shapes (and operating deflection shapes) present smooth
configurations. Thus, a large number of sensors is not required for their correct definition.
Figure 6.9 illustrates the application of the presented methodology to a wind turbine support structure,
using three levels of measurement. In this figure, the red circles represent the identified modal
amplitudes from the considered mode shapes, which are used to obtain the interpolation presented by
the thick line. The stress at any point of the structure is then obtained as a sum of the stress
contribution from each vibration mode. However, it is important to refer that the equivalent forces
computed with equation (6.9) should not be confused as real loads acting on the structure, since they
are just a gimmick used to recreate the deformed structure.
230
Chapter 6
= + +
Figure 6.9 – Illustration of the proposed methodology to estimate the equivalent forces along the tower height
One critical step when using accelerometers for fatigue assessment of structures is the double
integration operation. From equation (6.8), it is noticed that the result is prone to errors for frequency
values close to zero. Figure 6.10 illustrates a typical result obtained with the double integration of
acceleration time series from wind turbines support structures. It presents the estimation of the
displacement at the top position (computed according to equation (6.8)) from a simulated acceleration
(sensor S3) of the HAWC2 example. The displacement directly obtained with the numerical analysis,
from which the mean value was subtracted, is also presented. The comparison between the two signals
evidences a clear distortion of the estimated signal in the time domain. The estimated displacement is
governed by an erroneous large period cycle, masking the real displacement time series. This illustrates
the difficulty in estimating displacements from accelerations.
Sensor S2 - Displacement 5 Sensor S2 - Displacement
2 10
Original
1.5
Estimated
0
1 10
Displacement [m]
0.5
Amplitude
-5
0 10 0 0.15 0.3
-0.5
-10
-1 10
Original
-1.5
Estimated
-15
-2 10
100 200 300 400 500 600 700 0 1 2 3 4 5
Figure 6.10 – Time and frequency representation of the displacement (from which the mean value was subtracted)
obtained with the HAWC2 model (original) and estimated displacement computed through the double integration
process without any filtering window
231
It is thus comprehensible that this procedure is not capable of estimating the stress associated with the
very low frequency range of the spectrum. This part of the response spectrum, usually related to the
quasi-static component of the wind loading, will be referred in this work as “quasi-static” component
of the stress.
On the other hand, the described procedure is adequate to estimate the stress condition associated
with the remaining part of the spectrum, which includes the contribution of the vibration modes and
sinusoidal excitations (such as waves and rotor rotation driven harmonics). This component will be
referred as the “dynamic” component of the stress.
A complementary approach is thus required to estimate the quasi-static component of the stress. This
approach takes advantage of the fact that most of the wind action is applied on the support structure as
a thrust force at the hub level. Thus, it can be assumed that the quasi-static deformation of the
structure is similar to a cantilevered beam with a concentrated load at the top (Figure 6.11). The
stiffness matrix defined for the estimation of the dynamic component of the stress can then be used to
estimate this deformation.
Once this deformation shape is known, the knowledge of the quasi-static condition of a reference
point along the support structure is enough to estimate the stress condition along the entire structure.
This reference quasi-static condition can be the stress (measured with strain gauges), the displacement
(measured with GPS antenna) or the curvature (measured with inclinometers) evolution along time at
an adequate section of the support structure.
The use of a single measurement point to estimate the condition of the whole support structure is
illustrated in Figure 6.11. Assuming the deformed shape of the structure and a known reference
condition, the bending moment from the quasi-static component (and, consequently, the bending
stress) can be extrapolated for the rest of the structure.
Deformed shape Bending moment

Thrust force
Figure 6.11 – Thrust force applied to the wind turbine and illustration of the equivalent deformation and bending
moment of the support structure due to the quasi-static component of the loading
232
Chapter 6
6.4.1.1 Estimation of the Quasi-Static and Dynamic Stresses Contributions to Fatigue

During the proposed procedure for fatigue assessment of wind turbine support structures, the
contribution for fatigue damage of the quasi-static and dynamic components of stress are splitted.
Each component is analysed by different methodologies in order to estimate the bending stresses along
the structure. It is thus necessary to define which parts of the response spectrum correspond to the
quasi-static and dynamic components and how the response should be separated.
As previously referred, the main problem associated with the integration of accelerations is related to
the lowest range of frequencies of the spectrum. When performing the double integration operation
with equation (6.8), the estimation of the displacement corresponding to this range loses accuracy.
In that sense, the separation of the structural response in the two components should eliminate this
problem. The dynamic component should then be defined only for frequency values in which these
errors are not present. This can be achieved by applying a high-pass filter to the response acceleration
signal. At the same time, this frequency range should include all the contributions from vibration
modes and external dynamic loading excitations. On the other hand, a low-pass filter should be
applied to the measured reference quasi-static signals in order to contain only the frequency content
associated to the quasi-static component of the wind loading.
Based on these requirements, two window filters were designed. The high-pass window was design to
despise the frequency content below the frequency cut . From this point, a cosine taper window is
defined between and . From , the signal is kept unaffected. The filter is intended to be used
with the acceleration signal, in the frequency domain, before the double integration process
introduced with equation (6.8). On the other hand, the low-pass filter is designed as the
complementary of the high-pass, in such a way that no energy is lost during the process. Both filter
windows are shown in Figure 6.12.
Filter windows
1
Low-pass
0.8 High-pass
Anplitude
0.6
0.4
0.2
0
0 fc1 fc2
Frequency
Figure 6.12 – Low-pass and high-pass filter window
The filter parameters ( and ) should be adjusted according to each situation. The main
constraints to define the parameters are the lowest natural frequency of the wind turbine, the first
detected harmonic excitation, the wave loading spectrum (if the turbine is located offshore) and the
eventual frequency limits for the linear behaviour of the accelerometers. The filter should then be
defined in order to consider these effects in the dynamic component of the displacements.
233
The main steps of the proposed procedure for fatigue assessment of the wind turbine support structure
are summarized in Figure 6.13. As previously described, the procedure starts with the computation of
the low-pass filtered quasi-static signal from the quasi-static input data (stress, displacement or
curvature). Similarly, the dynamic acceleration data is obtained with the application of the referred
high-pass filter to the collected acceleration signals. From this point, two independent procedures are
applied in order to compute the two contributions to fatigue.
The quasi-static contribution to fatigue is obtained assuming the applied forces acting solely on the
rotor blades (and then transmitted to the tower through the hub). Since the deformed shape of a
cantilever beam with a concentrated force at the top is known, the stress condition along the support
structure can be estimated based on the quasi-static condition of a reference point (installed sensor).
The procedure to estimate the dynamic contribution to fatigue requires a more complex methodology.
It starts with the decomposition of the “dynamic” acceleration data into modal (and harmonic, if the
turbine is operating) responses. These acceleration time series are integrated to obtain modal (or
harmonic) displacement responses. Then, the modal (or harmonic) displacement responses are
estimated at non measured locations through an interpolation of the experimentally found mode
shapes. Lastly, equivalent forces are obtain, for each vibration mode (or harmonic), imposing the same
(modal) deformation.
Once the quasi-static and dynamic contributions to stress are estimated, the fatigue condition at an
arbitrary position of the wind turbine support structure can be obtained by summing these two
components.
234
Chapter 6
Input Data
Quasi-static Dynamic
contribution to fatigue contribution to fatigue
“Quasi-static” filtered signal “Dynamic” Acceleration Data
Known deformed
Acceleration
shape Time
Acceleration
modal response
Acceleration
Time
Esmated deformed Vibraon

Harmonics
shape Modes
Acceleration
Time
Double
Displacement
integration
Time
Displacement modal response

Displacementt
Time
Vibraon
Mode 1
Mode 1 Harmonics
Modes
Displacement
Time
Interpolation of the
displacements along
Displacement
p
Time
tower height
Displacement
p
Time
Displacement
p
Time
Displacement
Time
Mode 1
Vibraon
p
Time
Mode 1 Harmonics
Displacement
Modes
p
Time
Displacement
Computation of
equivalent forces
Equivalent forces
Mode 1
Vibraon
Mode 1 Harmonics
Modes
Esmated stress Esmated stress

condion condion
Estimated stress condition

(arbitrary position)
Figure 6.13 – Chain of events included in the proposed procedure for fatigue assessment at any position of the wind
turbine support structure
235
Example
In order to illustrate the proposed methodology, an additional sensor was introduced in the equipment
layout shown in Figure 4.21. This sensor is required to obtain the reference quasi-static condition. It is
assumed that the displacement at the position of sensor S1 is known.
For the analysis, the filter parameters ݂௖ଵ and ݂௖ଶ were defined as 0.10 Hz and 0.15 Hz, respectively. The
measured displacement and the signal obtained after the application of the low-pass filter are presented in
Figure 6.14.
0.04
Measured disp.
Filtered disp.
0.035
Displacement [m]
0.03
0.025
0 100 200 300 400 500 600
Time [s]
Figure 6.14 – Measured displacement and low-pass filtered displacement time series at S1 sensor position
Once the displacement at the S1 position is known, it is possible to estimate the quasi-static condition at
any position along the support structure. Figure 6.15 shows the comparison between the real (“original” -
which was filtered with the same low-pass window), and estimated quasi-static displacements at the S2
and S3 sensors position. As expected, the agreement between the signals is very good.
Sensor S2 - Low-pass filtered displacements Sensor S3 - Low-pass filtered displacements
0.16 0.32
Original Original
0.15 Estimated Estimated
0.3
Displacement [m]
Displacement [m]
0.14
0.28
0.13
0.26
0.12
0.24
0.11
0.1 0.22
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time [s] Time [s]
Figure 6.15 – Original and estimated low-pass filtered displacements at S2 and S3 sensors position
Together with the stiffness matrix, the low-pass filtered signal of the measured displacements at S1
position is used to estimate quasi-static component of the bending moment at the foundation level. The
obtained results are shown in Figure 6.16. An almost perfect correlation is observed between the
estimated and the original quasi-static base bending moment.
236
Chapter 6
4 4
x 10 Base bending moment x 10 Low-pass filtered base bending moment
5 5
Original
4.8 Original
Estimated (filt). 4.8
Estimated
4.6 4.6
Bending moment [kN.m]

4.4 4.4
4.2 4.2
4 4
3.8 3.8
3.6 3.6
3.4 3.4
3.2 3.2
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time [s] Time [s]
a) b)
Figure 6.16 – Comparison of the estimated quasi-static component of the base bending moment with: a) original
bending moment; b) low-pass filtered bending moment
After the estimation of the quasi-static component, it is necessary to compute the dynamic component of
the bending moment. In this example, the SSI-COV modal identification algorithm was used. Using the
modal acceleration responses estimated with the algorithm presented in section 5.5, the dynamic
component of the displacements at the sensors position was computed. The results were then compared
to the displacements obtained with the HAWC2 simulation after application of the high-pass filter. Figure
6.17 shows the very good agreement between both results.
237
-3 -3
x 10 S1 - High-pass filtered displacements x 10 S1 - High-pass filtered displacements
4 4
Original Original
3 3
Estimated Estimated
2 2
Displacements [m]
Displacements [m]
1 1
0 0
-1 -1
-2 -2
-3 -3
-4 -4
0 100 200 300 400 500 600 100 110 120 130 140 150
Time [s] Time [s]
S2 - High-pass filtered displacements S2 - High-pass filtered displacements

0.015 0.015
Original Original
0.01 Estimated 0.01 Estimated
Displacements [m]
Displacements [m]
0.005 0.005
0 0
-0.005 -0.005
-0.01 -0.01
-0.015 -0.015
0 100 200 300 400 500 600 100 110 120 130 140 150
Time [s] Time [s]
S3 - High-pass filtered displacements S3 - High-pass filtered displacements

0.03 0.03
Original Original
Displacements [m]
Displacements [m]
0.01 0.01
0 0
-0.01 -0.01
-0.02 -0.02
-0.03 -0.03
0 100 200 300 400 500 600 100 110 120 130 140 150
Time [s] Time [s]
Figure 6.17 – Original and estimated high-pass filtered displacements at the sensors position (the right hand plots are a
zoom of the left hand figures)
During the integration process in the frequency domain, numerical errors are introduced in the
displacement signal due to the application of the filter window over the acceleration time series. In order
to minimize these errors, a window is applied to the acceleration signal (in time domain) to attenuate the
amplitude of the signal in the beginning and end, prior to the integration. Figure 6.18 compares the
dynamic displacement obtained without and with the application of the attenuate window. It is visible
that, if the window is not used (Figure 6.18 a)), an error is present in the signal. Although the use of the
window may distort even further the initial and last part of the signal, it allows to confidently control the
length of these erroneous segments. With this strategy, it is thus possible to disregard these two time
segments when computing the bending stress for fatigue analysis purposes.
238
Chapter 6
-3 -3
x 10 S1 - High-pass filtered displacements x 10 S1 - High-pass filtered displacements
4 4
Original Original
3 3
Estimated Estimated
2 2
Displacements [m]
Displacements [m]
1 1
0 0
-1 -1
-2 -2
-3 -3
-4 -4
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time [s] Time [s]
a) b)
Figure 6.18 – Estimated dynamic displacements at sensor S1: a) without using the attenuate window; b) using the
attenuate window
The estimated dynamic displacements are then used to compute the dynamic component of the bending
moment at any position of the support structure. In this example, the foundation level was chosen.
Through the procedure introduced in section 6.4.1, the dynamic component of the bending moment at
the foundation level was estimated. The result is presented in Figure 6.19 alongside with the high-pass
filtered bending moment obtained as results of the numerical analysis. In this figure, the effect of the
applied attenuating window is seen in the beginning and end of the estimated time series. These two
segments should then be disregarded.
High-pass filtered base bending moment
6000
Original
4000 Estimated
2000
-2000
-4000
-6000
0 100 200 300 400 500 600
Time [s]
Figure 6.19 – Comparison of the estimated dynamic component of the base bending moment with the original high-pass
filtered bending moment
With both quasi-static and dynamic components estimated, the total bending moment can be obtained
with a sum between the two components. The total bending stress at the foundation level is shown in
Figure 6.20, where the initial and last transient segments were already eliminated.
239
Base bending moment 900

70 Original
800
Original Estimated
65 Estimated 700
Bending stress [MPa]
600
Nº occurrence
60
500
55
400
50 300
200
45
100
40 0
100 200 300 400 500 0 5 10 15 20
Base bending moment [kN.m] Bending stress [MPa]
Figure 6.20 – Comparison between the original and Figure 6.21 –Histogram with the number of cycles from
estimated base bending stress the original and estimated bending stress at the
foundation level
At this point, the histogram of the loading cycles can be computed. This histogram is compared with the
one corresponding to the original bending moment in Figure 6.21 (from which the same initial and last
parts were also not considered).
It would be interesting to characterize the relative contribution of each component (quasi-static and
dynamic) to the total fatigue condition of a wind turbine. In that sense, numerical analyses were
performed to assess each contribution for different wind conditions. However, it was noticed that the
contributions are dependent on the control system (mainly, on the pitch control) and, for that reason,
they are not generalizable to any wind turbine.
6.4.2 VIRTUAL SENSORS

The methodology presented in the previous section requires the installation of a certain number of
accelerometers along the wind turbine support structure. Nevertheless, nowadays wind turbines are
usually equipped with a bi-directional accelerometer at the top as part of their SCADA system. Thus,
the use of this single sensor for dynamic monitoring of the wind turbines would avoid the need of
installing additional sensors (and, consequently, reduce the cost of the monitoring system).
The use of a top sensor as the only measurement sensor of the monitoring system introduces an
important limitation, concerning the fatigue assessment of the wind turbine support structures. Since
it is only possible to obtain a single modal amplitude per direction, it is not possible to accurately
estimate (and, later, interpolate) the configuration of vibration modes along the height of the structure.
In order to overcome this problem, two solutions may be regarded: the use of a calibrated finite-
element model; or the characterization of the mode shapes during a training period.
The use of a calibrated finite-element model can be used to obtain the configuration of the vibration
modes. The contribution of each mode to the response can then be scaled according to the modal
amplitude obtained with the single accelerometer. Then, the procedure presented in section 6.4.1 can
be used. Notwithstanding, the development of a finite-element model of a wind turbine presents
important challenges as previously referred.
240
Chapter 6
The characterization of the mode shapes during a training period was used in this work. With this
approach, a suitable number of temporary sensors is installed at the wind turbine support structure
during a short period of time (e.g. between two scheduled maintenance operations). Throughout this
period, the modal tracking of the wind turbine modes is accomplished with all the sensors (permanent
plus temporary), allowing to characterize the modal amplitudes at the measurement positions, under
different operating conditions. In addition, the characterization of the operational deflection shapes of
the wind turbine vibrating at harmonics frequencies along the operating regime of the turbine is also
necessary. The number of installed sensors during this period has to be sufficient to obtain a
satisfactory characterization of these vibration configurations.
After the training period, the mode shapes (and operational deflection shapes) are well characterized at
the positions of the permanent and temporary sensors. At this point, the continuous monitoring of the
wind turbine is performed with the single sensor. As result, the identified modes shapes are
characterized by a single modal amplitude per direction (at the permanent sensor position). These
modal amplitudes are used to scale the contribution of mode shapes (and operational deflection
shapes) characterized during the training period. With this step, a new matrix is defined and the
estimated acceleration time series at the unmeasured locations are obtained with:
. , (6.12)
Example
The methodology based on virtual sensors was applied to the numerical NREL 5MW wind turbine
example. In this example, the recorded acceleration time series from one sensor was used to predict the
acceleration at the location of the other two sensors. The mode shapes obtained with the analysis in which
all sensors were considered were adopted to define the matrix, simulating the results from a training
period.
Three different analyses were then performed, considering in each one a different single sensor. Initially,
the single acceleration time series was used as input for the output-only modal identification algorithm.
For this example, the SSI-COV algorithm was selected. From the identification results, a suitable model
order was considered to define the state-space model.
Considering the selected state-space model, the forward innovation model was obtained as introduced in
section 5.5.2. Then, the acceleration time series at the location of the non-considered sensors were
estimated through equation (6.12).
Figure 6.22 illustrates the results obtained with this analysis. In this figure, each column is referred to an
analysis where a single sensor was used to estimate the acceleration from the other two. For example, the
first column is referred to the analysis where the S1 sensor was used as single sensor. The predicted
acceleration to the considered single sensor is also presented in each column to show the error associated
with application of the model (see equation (5.73)).
The results presented in this figure show a good agreement between the predicted and original time series
at the unmeasured places. However, it is noticed that the consideration of the top sensor S3 as single
sensor produces the worst results. This situation is due to the fact that the acceleration response at the S1
and S2 location is mainly dominated by vibration modes with reduced modal amplitude at the tower top.
241
Consequently, the identification of these modes will be worse when a sensor at the tower top is considered
as single sensor, leading to a poorer agreement between the predicted and the real acceleration signal.
Virtual S3 Virtual S3 S3
Virtual S2 S2 Virtual S2
S1 Virtual S1 Virtual S1
Original
Prediction
Sensor S1 Virtual Sensor S1 Virtual Sensor S1
0.4 0.4 0.4
Acceleration [m/s2]
0.2 0.2 0.2
0 0 0
-0.2 -0.2 -0.2
-0.4 -0.4 -0.4

160 165 170 175 180 230 235 240 245 315 320 325 330
Virtual Sensor S2 Sensor S2 Virtual Sensor S2
0.4 0.4 0.4
Acceleration [m/s2]
0.2 0.2 0.2
0 0 0
-0.2 -0.2 -0.2
-0.4 -0.4 -0.4

160 165 170 175 180 230 235 240 245 315 320 325 330
Virtual Sensor S3 Virtual Sensor S3 Sensor S3
0.4 0.4 0.4
Acceleration [m/s2]
0.2 0.2 0.2
0 0 0
-0.2 -0.2 -0.2
-0.4 -0.4 -0.4

160 165 170 175 180 230 235 240 245 315 320 325 330
Figure 6.22 – Comparison between the recorded and estimated acceleration at the sensors position when a single sensor
is used to estimate the time series of the other two sensors. Each column is referred to the results obtained with the
consideration of a different single sensor
242
Chapter 6

The fatigue assessment of wind turbine structures is an important indicator for the evaluation of their
structural integrity. Due to the highly dynamic nature of the loads, the structure is almost continuously
subjected to cyclic excitations, requiring a special attention to guarantee a proper operation
throughout its life.
In that sense, this chapter introduced the main aspects related to the fatigue assessment of the wind
turbine support structure. The main loads acting on the support structure were described for both
onshore and offshore wind turbines. Also, a survey covering the main standards and guidelines about
the design method for fatigue assessment of the support structure was developed.
A methodology for estimation of the fatigue state at any point of the wind turbine support structure
was proposed in this chapter. The method is based on the independent estimation of the quasi-static
(using sensors measuring strains, displacements or curvatures) and dynamic (using accelerometers)
contributions to fatigue. This procedure was illustrated using the results obtained with the NREL
5MW numerical example.
Lastly, the possibility of using a single level of measurement of acceleration to estimate the dynamic
contribution to fatigue was addressed. It was noticed that the sensor usually installed in the nacelle is
not enough to accurately estimate the acceleration at other positions of the support structure. Two
alternative positions for installation of a single level of measurement were tested, which provided
better results.
The proposed methodology to estimate the fatigue condition at any point of the wind turbine support
structure is further exploited in the next chapter, where experimental data is used to test the accuracy
of the method. In the same way, the ability to reduce the number of sensors and, at the same time,
maintain the level of accuracy is also addressed in the next chapter.
243
244
Chapter 7
7
STRUCTURAL MONITORING OF
WIND TURBINES
7.1 INTRODUCTION TO STRUCTURAL MONITORING OF WIND TURBINES

The assessment of the integrity of wind turbines and their components is a field of study with a high
level of interest. As exposed in section 3.1.3, some mechanical and electronic components present
relative high rate of failures and consequent downtime. This non-production time (together with the
cost of repair/ replacement of the components) considerable increases the revenue losses, both for
onshore and offshore locations. This is even more critical for high power rated turbines. It should also
be highlighted that the specificity of offshore environment introduces some new variables in the cost
function:
o Offshore turbines are expected to become considerably larger than onshore ones (in
dimension and rated power). This situation introduces two important variables: downtime
corresponds to a greater loss of income; and the components to be replaced are heavier and
larger, which makes their transportation and installation more difficult;
o The need of adequate weather conditions to reach the structure at offshore locations, leads to
longer downtimes.
Considering the reasons presented above, the continuous monitoring of the condition of the support
structure (foundation, tower and, in some situations, transition piece) can introduce some important
advantages. Firstly, it should be noticed that if the support structure fails and collapse, the entire
investment is lost. Thus, a system able to anticipate such dramatic damage at the tower is very
important since it allows earlier reparation, preventing the failure of the entire system. Besides this,
three important aspects should also be considered for the installation of a monitoring system at the
support structure:
o The qualitative and quantitative assessment of vibration levels might lead to the identification
of high-vibrational events and/ or periods of time corresponding to reduced turbine
performance;
o If the natural frequency of the tower is close to the rotational frequency of the rotor (or its
multiples), Germanischer Lloyd Guidelines (Germanischer Lloyd (GL), 2010; Germanischer
Lloyd (GL), 2012) refer the need of a continuous monitoring vibration system. With this
system, it is possible to track the value of the natural frequency along the life of the wind
turbine and evaluate possible resonance problems;
245
Structural Monitoring of Wind Turbines
o The continuous assessment of the structural condition and operation of the support structure
might give a good support to decide about the extension of operation after the initial
predefined period of life or, in the other hand, if the structure should be dismantled (or
reinforced) before the expected period of life due to unforeseen loads and/ or degradation.
Considering specifically offshore wind turbines, the installation of a monitoring system at the support
structure is even more advantageous. Being this type of structures fairly recent, the degree of
knowledge is not as mature as for onshore systems. Mainly for this reason, some additional
motivations should be considered for the installation of a monitoring system:
o The effects of the sea acting at the structure (mainly waves) are still not mastered. For this
reason, there is some uncertainty when dealing with the fatigue life of these structures.
Furthermore, some structural aspects, like foundations and the grouted joint in the transition
piece (Rücker, 2007), are prone to rapid wear, which may require specific structural health
assessment procedures;
o One specific problem of offshore wind turbines is scour around the foundation piles. This
problem increases the flexibility of the entire structure and leads to reduction of the support
structure natural frequency (Zaaijer, Tempel et al., 2002; Prendergast, Gavin et al., 2015). This
situation may lead to resonance problems if the natural frequencies of important support
structure vibration modes are reduced to values close to the harmonics of the rotor speed.
Figure 7.1 shows the evolution of the 1st and 2nd natural frequency values of a monopile
support structure. It is seen that these modes (especially the 2nd mode) are very sensitive to this
problem;
o Conversely, the increasing of the foundation stiffness might also represent a problem when the
correspondent increase of the frequency values brings them close to harmonics. This is the
case of soft-stiff solutions. A real case study in which a progressive increase of the modes
frequency values was noted is presented in (Weijtjens, Verbelen et al., 2015). It is referred that
this phenomenon may be related to soil stiffening due to cyclic loading.
1
S - Scour depth
0.95 D - Monopile diameter
Relative natural frequency
0.90
0.85
0.80
0.75 f1 (Loose sand) [Prendergast, et al.]

0.70 f1 (Med. dense sand) [Prendergast, et al.]
f1 (Very dense sand) [Prendergast, et al.]
0.65 f1 (general scour) [Zaaijer, et al.]
f2 (general scour) [Zaaijer, et al.]
0.60
0 0.5 1 1.5 2 2.5
Relative scour depth [S/D]
Figure 7.1 – Variation of natural frequencies of offshore wind turbines with scour (adapted from (Zaaijer, Tempel et al.,
2002) and (Prendergast, Gavin et al., 2015))
Thus, the implementation of a monitoring system in wind turbines allowing the assessment of the
actual condition of the components and support structure would introduce considerable
246
Chapter 7
improvements in the definition of an Operation and Maintenance (O&M) strategy, following a

predictive maintenance (see section 3.1.3). With this kind of maintenance, a continuous assessment of
the condition of the components is available, which permits to detect damages at an early stage. At this
stage, different strategies can be considered, such as derating of the turbine (Griffith, Yoder et al.,
2013) with the goal of operating with minimal increase of the damage and of preventing the damage of
one component from spreading to others (Hyers, McGowan et al., 2006) (e.g. if the pitch of one blade
is not working properly, the aerodynamic asymmetry created will introduce additional loads on the
drive train which may damage this component). Besides this, with the early damage detection, it is
possible to order replacement components and schedule maintenance in a timely manner with
minimum downtime (Adams, White et al., 2011) and avoid the abrupt shutdown of the plant.
Considering the potential benefits of the introduction of monitoring systems in wind turbines, it is
comprehensible the interest of all industry agents involved, including insurers. Insurers play an
important role in the wind energy business, since a significant portion of O&M are due to insurance
costs (Morthorst and Jacobsen, 2004). Thus, the acceptance by insurer institutions that monitoring
systems have an important role in maintenance management could contribute for better conditions for
wind turbine owners. Receptivity to this change by insurers started around 2002, when some German
insurers introduced a review clause in which it was required that all roller bearings in a drive train
have to be replaced after either 40 000 operating hours or five years (whichever is earlier) unless an
appropriate Condition Monitoring System had been installed (Becker and Poste, 2006). Gothaer
(Wind Power Monthly, 2003), GCube (Appleyard, 2013) and Allianz (Gellermann, 2012) also refer
monitoring system as a need for proactive maintenance measures. The latter also highlights the
importance of monitoring of other components, such as blades, tower and foundation (Gellermann,
2012).
Alongside with the insurances perspectives, another key aspect is the economic benefit of the
installation of a monitoring system. McMillan and Ault (McMillan and Ault, 2007) studied this subject
for a predictive maintenance strategy (with the introduction of a condition monitoring system for the
generator, gearbox, blades and power electronics system) against the more common periodic
maintenance for a 5 MW offshore wind turbine. They concluded that the predictive strategy
introduces economic benefits, if the monitoring system provides accurate diagnosis, of around 60 % to
80 %.
The CONMOW project also studied the costs of preventive and corrective maintenance for offshore
wind turbines (Wiggelinkhuizen, Watson et al., 2007). It was concluded that the costs of corrective
maintenance are a factor of 2 higher than for preventive maintenance.
The monitoring of the condition of wind turbine components can be performed by various techniques,
in particular Condition Monitoring (CM) and Structural Health Monitoring (SHM). Conceptually,
these two techniques refer to different fields of operation: CM is oriented to the machinery, mainly
rotating components (such as drive trains), whilst SHM deals with major structural components (such
as foundations, tower and blades). This is the guidance that will be used throughout this work.
Nonetheless, the scope of application of the two concepts is not universal (Carden and Fanning, 2009)
and there seems to be a unification of the monitoring systems around the name Condition Monitoring
(e.g. GL guidelines (Germanischer Lloyd (GL), 2003) encompasses the monitoring system for every
components of a wind turbine, such as gear boxes and towers, under the CM definition).
247
7.1.1 CONDITION MONITORING OF WIND TURBINES

Condition monitoring is referred as “the process of monitoring a parameter of condition in machinery,
such that a significant change is indicative of development failure” (Wiggelinkhuizen, Watson et al.,
2007). It is a technique used to assess the condition and detect early damages and/ or malfunctions in
components machinery. Condition monitoring is widely used in other fields of industry (Şeker, Ayaz
et al., 2003; Fast and Palmé, 2010).
As referred previously, Condition Monitoring has been steadily gaining some support from the
industry, especially due to the considerably high rates of failure from some components. A sign of
confidence in the application of monitoring systems in wind turbines is the requirement, by some
guidelines/ standards, of their installation in offshore turbines. For example, GL guidelines
(Germanischer Lloyd (GL), 2003) define as mandatory, as minimum components to be monitored,
main bearings, gearbox and generator.
Condition Monitoring can be applied to a large variety of components. However, the components of
major concern are usually rotors, pitch mechanisms, drive trains (including gearboxes and bearings)
and generators (Hyers, McGowan et al., 2006). For this, various techniques have been developed to
detect problems related to these components, which stand, among others:
o Vibration-based analysis;
o Acoustic emission;
o Ultrasonic testing techniques;
o Oil analysis;
o Strain measurements;
o Electrical effects;
o Thermography;
o Performance monitoring.
A more complete list of techniques and processing tools can be accessed in (Hameed, Hong et al.,
2009; Liu, Tang et al., 2010; García Márquez, Tobias et al., 2012; Yang, Tavner et al., 2014; Liu, Tang et
al., 2015). Although currently there is a long list of CM techniques in the context of wind turbines,
Germanischer Lloyd Guidelines (Germanischer Lloyd (GL), 2003) states that, to be within its
certification standards, the monitoring system should be defined by a vibration-based analysis. This
requirement is justified by the fact that this technique allows an earlier detection (Figure 7.2).
no visible visible severe

damage damage damage
Condition monitoring (vibration)

Damage
Particle count
SCADA
(temperature)
Life time
Figure 7.2 – Detection of damage by different systems defined by GL Guidelines (Germanischer Lloyd (GL), 2003)
248
Chapter 7
7.1.2 SHM OF WIND TURBINES

A possible definition of the Structural Health Monitoring concept is presented in (Boller, Chang et al.,
2009): ”SHM is the integration of sensing and possibly also actuation devices to allow the loading and
damaging conditions of a structure to be recorded, analysed, localized, and predicted in a way that
nondestructive testing becomes an integral part of the structure.”. Thus SHM methods deal with the
problem of assessing the structure integrity through the installation of sensors with the aim of
detecting early damage. This kind of monitoring concept dates back from earlier the 1980s (Boller,
Chang et al., 2009). Over the time, the knowledge in this field increased at a very high rate, mainly due
to the easier and cheaper access to various sensors. Some theoretical definitions were developed (see,
for example, the fundamental principles of SHM presented in (Worden, Farrar et al., 2007)). One
important idea presented by Rytter (Rytter, 1993) is referred to the different levels in damage detection
(Figure 7.3).
LEVEL 1 – DETECTION (qualitative assessment that damage is present (or not) in the structure)
LEVEL 2 – LOCALIZATION (probable location of the damage)
LEVEL 3 – ASSESSMENT (the extent of damage is evaluated)
LEVEL 4 – CONSEQUENCE (information about the actual safety of the structure, prognosis of the
remaining lifetime)
Figure 7.3 – Levels in damage assessment
SHM systems have been applied for years in different types of civil engineering structures. Magalhães
et al (Magalhães, Cunha et al., 2009) implemented a monitoring system in concrete arch bridge. Hu et
al (Hu, Moutinho et al., 2012) implemented a similar SHM system in a footbridge. Brownjohn
(Brownjohn, 2007) presents several examples of SHM applications in a variety of structures, among
which dams, offshore installations, nuclear installations and tunnels. Some examples of very complete
SHM systems installed in bridges are presented in (Boller, Chang et al., 2009).
Lately, the application of SHM to wind turbines has been one of the main topics of study. Blades have
been the preferred substructure of study due to their high propensity for failures. However, recently
also the support structure (tower and foundation) have focused some attention concerning safety
issues. This is mainly due to the increasingly higher dimensions of the tower, alongside with larger
loads and masses at the top, due to larger rotor diameters, and due to the lack of knowledge of the
environmental situations at sea and greater wear of structures in offshore installations.
Several methods of SHM for application in wind turbines have already been developed. An exhaustive
list of methods (referred to blades) is introduced in (Boller, Chang et al., 2009) and (Ciang, Lee et al.,
2008). Also in (Wymore, Van Dam et al., 2015), a general overview of the state of development of
health monitoring systems is presented. When considering SHM for support structures, essentially two
249
methods are used: vibration-based and strain measurement. The present work deals with the
application of SHM to the support structure of wind turbines through the application of vibration-
based methods.
Some numerical studies about the feasibility of using the frequency values of the vibration modes to
identify structural damage in wind turbines are present in the literature. Adams et al (Adams, White et
al., 2011) simulated three different types of damage in different locations in a numerical model: blade
root section, low-speed shaft and yaw joint. Through a sensitivity analysis, the variations of the first 14
natural frequencies were calculated for each damage scenario. Also, some developments have been
made in order to assess the evolution of the frequency values of the support structure with scour
(Zaaijer, Tempel et al., 2002; Sørensen and Ibsen, 2013; Prendergast, Gavin et al., 2015).
A good example of the use of vibration-based methodologies for damage detection on the support
structure is given in (Pelayo, López-Aenlle et al., 2011). In this work, high levels of vibration, together
with cracks at the foundation, were detected in two wind turbines. For diagnosing the cause of this
anomaly, operational modal tests were performed in these two turbines, alongside with a similar but
healthy wind turbine. For these tests, accelerometers were placed on the foundation and on the lower
part of the tower. The results showed that both damaged wind turbines presented lower values of
natural frequencies than the healthy turbine (around 7 % to 3 %). After this assessment, the
foundations of the two wind turbines were reinforced. Then, the tests were once again performed and
the results indicated an increase in the natural frequencies (in line with the healthy wind turbine).
Also, in (Currie, Saafi et al., 2015), problems related to high levels of vibration at the foundation level
of onshore wind turbines, due to damage around the bottom flange of the embedded ring, were
described.
In the experimental field, some cases of implementation of vibration-based monitoring systems are
described in literature. During the HISTWIN project (Veljkovic, Heistermann et al., 2012), a
monitoring system was implemented on a wind turbine for almost one year. For the instrumentation,
accelerometers, strain measurements, temperature sensors and inclinometers were used, together with
some SCADA data. During this period, the modal identification of the tower was performed, as well as
the fatigue assessment of the tower and bolts from the flanged connections between segments.
A different approach for a monitoring system is described by Braam et al (Braam, Obdam et al., 2012).
In this work, a strategy of monitoring an entire wind farm with the installation of additional sensors in
just a few reference turbines is presented. On the reference turbines, fibre optic sensors were installed
on the blades with the aim of measure strains (load indicators) and evaluate the fatigue state of these
turbines. The relations between these load indicators and the parameters measured by the SCADA
system (present in every turbine of the wind farm) were then established, in order to extrapolate the
load accumulation of the non-reference turbines. This methodology allows defining a maintenance
strategy considering the most mechanically loaded wind turbines. In (Guo and Infield, 2012), also a
more economical SHM system is presented, since it only uses built-in SCADA parameters. It consists
in the processing of data from the SCADA system, mainly 10 min. averaged values of the tower top
and drive train accelerometers. This data is then correlated with other SCADA data and analysed in
the context of the different operational regimes of the rotor.
Also in (Hu, Thöns et al., 2015b; Hu, Thöns et al., 2015a), a study about the implementation of a
vibration-based monitoring system at a prototype 5 MW wind turbine with tripod foundation installed
onshore is described in two parts. For this system, 8 accelerometers distributed along the support
250
Chapter 7
structure were used. In the first part, the main results concerning the dynamic characterization of the
wind turbine are presented. Then, in the second part, the modal results obtained during a period of 2
years were used to assess the detection of damage at the tower and foundation.
The offshore Vestas V90-3.0 MW, introduced in section 5.9, is also used in the context of a project
aiming to develop monitoring systems based on the dynamic properties of wind turbines. The results
obtained up to date are described in (Weijtjens, Verbelen et al., 2015).
The assessment of the variation of natural frequencies in order to detect blade damage is also a field of
current study. The majority of published works, both in the experimental (Wang, Liang et al., 2014;
Ulriksen, Tcherniak et al., 2015) and numerical domain (Hoell and Omenzetter, 2014), refer the small
sensitivity of the frequency values of the first order blades vibration modes to common small damages.
Notwithstanding, in (Marulo, Petrone et al., 2015), an experimental study of the sensitivity of the
dynamic properties of a small-scale blade to a very severe damage is described. The blade was tested to
buckling and its modal properties were analysed before and after the buckling. In this case, the
variation was notorious, exposing that the application of this kind of monitoring technique may be
only effective for detection of heavy damages.
251
7.2 VIBRATION-BASED SHM SYSTEM

The structural health monitoring system presented in this work is developed based on a vibration-
based methodology. The system is composed by two modules, aiming to present a solution for two
important problems within the scope of structural condition assessment of the support structure of
wind turbines: detection of damage at an early stage and fatigue life estimation.
One of its components (damage detection), follows a modal approach (Fritzen, 2005). In this system,
the modal parameters of the wind turbine structure (natural frequencies, damping ratios and mode
shapes) are continuously extracted and tracked over time. Thus, anomalous variations of modal
parameters, which may be caused by the occurrence of damage, are detected. As damages are normally
associated with stiffness reduction, and assuming a negligible variation of mass (which is perfectly
admissible on wind turbines), abnormal reductions of the natural frequencies are assumed as presence
of damage. Thus, natural frequencies are considered a prime indicator for damage detection in wind
turbines, as already exposed in previous sections. Some examples of similar monitoring system
implemented on civil engineering structures are presented in (Peeters, 2000; Magalhães, 2010; Hu,
2011).
The second module of the monitoring system is referred to the estimation of the fatigue life condition
of the support structures of wind turbines. This component is based on the procedure introduced in
Chapter 6. Once a continuous estimation of the loading cycles of the structure is obtained, an on-line
evaluation of the increment of damage in virtually every structural detail can be achieved. Then, it is
possible to perform some estimation regarding the real fatigue life of the support structure.
7.2.1 DAMAGE DETECTION

The damage detection module of the monitoring system aims to detect abnormal structural changes at
an early stage. In that sense, the monitored frequency values of the vibration modes are used as
damage indicators. In this section, the procedures implemented to reduce the variability of the tracked
frequency values and to identify possible situations of damage are introduced.
7.2.1.1 Removing of Environmental and Operational Effects on Natural Frequencies

Wind turbines are subjected to a large variability of environmental effects, such as ambient
temperature, wind, boundary conditions, among others (Sohn, 2007). Alongside, the operation of
wind turbines also introduce important modifications on their structural condition, such as variation
of pitch and yaw angle, varying rotor speed, among others (see section 3.2.6). These effects are an
important source of variability of the modal properties of the support structures, often introducing
variations with greater amplitude than the damage itself. Consequently, if it is desirable to detect small
scale damage at an early stage, the influence of external factors has to be “cleaned” from the modal
properties. Usually, only the natural frequencies are used.
Some practical cases studied are also presented in the literature. Magalhães et al (Magalhães, Cunha et
al., 2012) studied the variation of natural frequencies due to temperature in a concrete arch bridge,
while Hu et al (Hu, Caetano et al.) and Moser and Moaveni (Moser and Moaveni, 2011) performed a
similar work in footbridges. For its turn, interesting correlations between the modal properties of the
offshore Vestas V90-3.0 MW and the tidal level are presented in (Devriendt, Magalhães et al., 2014).
252
Chapter 7
In order to understand and remove the influence of environmental and operational effects on the
modal properties of the structure, three different types of approaches can be followed: physical models,
models whose environmental effects are used as input and models which do not consider these inputs.
The last two models were tested in the context of the dynamic monitoring system described in this
work. These methods are considerable easier and quicker to implement and operate, however they
required a period of training in which a representative variation of the environmental and operational
effects should be included.
The methods considering the measurement of the environmental and operational effects in the
conception of the model are defined as input-output methods. The application of these methods for
removal of the external influences in the context of wind turbines seems, from a conceptual point of
view, very suitable. Since the SCADA system collects a variety of operational and environmental data
related to the wind turbine in which it is installed, the access to these records is easier and do not
require the installation of any additional sensor. In this approach, two methodologies are described
based on Multivariate Linear Regression Models and Dynamic Regression Models.
On the other hand, the approach not requiring the measurement of the operational/ environmental
effects is referred as output-only methods. In the context of wind turbines, output-only methods can
be also used in association with the previously presented input-output methodologies with the aim of
achieving a “cleaner” time evolution of the natural frequencies. This can be obtained since there may
be external factors not recorded by the SCADA system that also influence the modal properties.
Within this scope, the approach based on the Principal Component Analysis was tested (Johnson and
Wichern, 2002). However, it was observed that the application of this method did not improve the
results obtained with the linear regression models. For that reason, this method was not considered in
the context of the monitoring system described in this work.
Multivariate Linear Regression Models

Multivariate Linear Regression Models deal with the problem of predicting values of response
(dependent) variables from a collection of predictor (independent) variable values (Johnson and
Wichern, 2002). These models can also be used to assess the effects of the predictor variables on the
responses. In order to expose a clearer explanation, these models will initially be address to the
situation of just one dependent variable and, at the end, the generalization for multiple response
variables is introduced.
For the present case of study, the natural frequency values of one vibration mode is considered the
response variable, while the predictor variables are represented by the operational/ environmental
effects. Initially, the degree of influence in the variation of the response is assessed and quantified for
each environmental effect (predictors). Then, the defined model is ready to predict response values
from a particular moment according to known predictors of the same moment. Finally, real and
predicted response values can be compared and the magnitude of the deviation evaluated.
253
The mathematical definition of a Multivariate Linear Regression Model is provided by (Johnson and
Wichern, 2002):
. (7.1)
with:
Vector with observations of the dependent variable
Matrix with the values of the predictors variables (for operational/ environmental effects)
Vector with weight parameters of the predictor variables
Vector with the values of the random error
The vector , which represents the random error due to measurement issues and due to the existence
of variables not explicitly considered in model, presents the following properties (Johnson and
Wichern, 2002):
0
(7.2)
. .
with:
∗ Expected value operator
∗ Transpose operator
∗ Standard deviation
Identity matrix
When defining the matrices from equation (7.1), some considerations are employed. The first column
of is filled by ones, so this matrix contains 1 columns. Also, the values of are usually
normalized, being each element subtracted from the mean values and then divided by the standard
deviation of the variable:
(7.3)
with:
Normalized value of
With this configuration, the first element of the vector contains the mean value of the observed
variable.
254
Chapter 7
Before the definition of the model, it is needed to assess which predictors influence the response
variable. For this step, the predictors with high correlations with the response variable and low
correlations with the other predictors should be selected (Magalhães, 2010). A suitable tool for the
evaluation of the correlation between two variables and is the Pearson’s product-moment
correlation coefficient (Johnson and Wichern, 2002):
,
. (7.4)
where values close to zero represent a low linear correlation, while values close to one (in absolute
value) might indicate a good correlation. It should be noted that this tool only considers linear
correlations. Thus, if a zero value is obtained, it does not imply that a non-linear relation is also not
present.
Once the choice of the relevant predictors is complete, the vector β should be estimated. This vector
(named as ) is obtained through application of the least squares method and defined in (Johnson and
Wichern, 2002):
. . . (7.5)
Since the weight factors of the predictor variables are already defined, the vector should then be
evaluated. In that sense, an estimation of the residual random errors is given by the difference between
the real and estimated response variable (Johnson and Wichern, 2002):
̂ . . . . (7.6)
Two ways for assessing the quality of the model adjustment can be used. The first one is through the
computation of the estimation of variance of ̂:
1
. (7.7)
However, this indicator has the disadvantage of being dependent of the absolute values of y variable.
To solve this problem, a coefficient is presented in (Johnson and Wichern, 2002) that avoids this
inconvenience:
∑ ∑
1 (7.8)
∑ ∑
255
The coefficient is a ratio of variances that aims to quantify the proportion of variance of
“explained” by the selected predictor variables (it varies from 1 to 0, where 1 holds to situations where
the adjust passes through all data points – a “perfect adjust” -, while 0 corresponds to a total
independence between response and predictor variables).
After the assessment of the residual errors is complete and the results are accepted, the model is ready
to predict future values of the response variable ( ) based only on the predictors not considered for its
definition ( ):
. (7.9)
At this point, the model is capable of predicting natural frequencies (the response variable), based on
the environmental effects (the predictor variables) from the same period. Thus, observed and
forecasted values can be compared. Since errors ( ) from this difference are inevitable, the definition of
a range in which the errors are acceptable is required. In that sense, if the errors are considered as
normally distributed, a confidence interval of 100 1 α % is given by (Johnson and Wichern, 2002):
. 1 . . (7.10)
2
with:
2 The upper 100 1 th

percentile of a t-distribution with n-p-1 degrees
The previous equation set a range ; with a defined confidence level in which
the new values of response variable should lay in. If the previously referred considerations about the
construction of the matrix are met (first column of is filled with ones and the elements of the other
columns are normalized), the confidence interval can be written as (Magalhães, 2010):
. (7.11)
with:
New response variable value (which was not considered for the definition of )
The previous equation can be interpreted in the context of the monitoring situation under study. Once
the training period for the definition of matrix is complete, predictions of the frequency values ( )
from the subsequent periods can be computed. These predictions will be formulated considering the
entire set of environmental effects covered during the training period. For this reason, predicted
frequencies will present variations that are interpreted as normal, i.e., this variations do not refer to
any kind of damage since the training period is assumed to have occurred during a healthy stage of the
structure. When comparing the predicted and real values of the natural frequencies from the
256
Chapter 7
subsequent periods, their difference should be evaluated. If it is abnormally high (i.e., the value is
outside the confidence interval defined in equation (7.11)), the variation cannot be explained by the
environmental effects considered during the training period, and an abnormal condition affecting the
structure, such as damage, may be present.
The generalization of the presented method is easily performed for several dependent variables
(frequency values of several vibration modes). This can be achieved by formulating the previous
methodology as many times as response variables or through the matrix equation:
. (7.12)
with:
Matrix with n observations of the m dependent variables
Matrix with the n values of the p predictors variables (p environmental effects)
Matrix with weight parameters of the p predictor variables
Matrix with the values of the random error
One last property from these models should also be referred: Multivariate Linear Regression Models
are not valid for extrapolations. For that reason, a period of training with a wide range of values from
the environmental effects should be used.
Dynamic Regression Models

An extension of the previous approach is given by the Dynamic Regression Models. In the previous
section, it was shown that a Multivariate Regression Model, in order to predict a response value at a
time instant , uses the observations of the predictor variables at the same time instant – it is then
considered a static model. On the other hand, a Dynamic Regression Model considers that
observations of predictor variables at previous time instants also influence the foreseen of the
dependent variables. This consideration is very useful when dealing with response variables with a
large inertia to variations of predictors.
The implementation of Dynamic Regression Models follows the same procedure as the Multivariate
Models. The only change in the methodology is the introduction of additional predictors consisting of
“past” observations of the independent variables.
7.2.1.2 Damage detection

The described statistical techniques allow estimating a variable (the residual error) which can be used
as an indicator of abnormal conditions (usually damage) of a monitored wind turbine structure.
However, since it is desired that the monitoring system operates in an autonomous way, an automatic
procedure to identify situations of damage is required.
A general procedure to control the residual error from the frequency values of several vibration modes
is obtained with the application of control charts. Control charts, one of the simplest techniques of
257
statistical process control, are used to evaluate if one or more variables are kept within predefined
limits, which indicates that the process is occurring without abnormalities. For this reason, control
charts are a very suitable technique to monitor possible frequency deviations due to damage.
A typical control chart is presented in Figure 7.4. It consists of a graphical display, with horizontal
development (usually associated with time), in which observations are plotted. There are also some
control parameters: a centre line, which represents the average value of the quality characteristic under
study; an upper (UCL) and lower control line (LCL) defining an interval in which the observations are
considered “in-control”. If an observation has a value considered as abnormal, it falls outside this
interval and is considered as “out-of-control”.
Out-of-control
Upper control limit
Sample quality characteristic
Center line
Lower control limit

Out-of-control
Sample number or time
Figure 7.4 – Control chart
A common control chart to assess the evolution of a single variable is the Shewhart -chart. In the
process control, the points (in the plot) under control represent a single observation or the mean of a
group of observation (subsample). For a -chart, the definition of the upper and lower control lines is
given by (Montgomery, 2009):
̅ .
(7.13)
̅ .
with:
̅ Mean value from all observations
Standard deviation. When each point represent one single observation, is the standard deviation
from all observations; if each point represents the mean of a subsample with elements, is given by
the quotient between the standard deviation from all observations and √
“Distance” between the control lines and the centre line. If 3 is defined in a distribution
considered as normal, the control limits correspond to a confidence level of 99.7 %
When dealing with a situation with more than one variable (like vibration modes), it is required a
slightly evolution from a univariate technique (such as the -chart) to a multivariate technique. A
similar technique of the Shewhart -chart for monitoring more than one parameter is the control
258
Chapter 7
chart. The control chart is a simple tool to detect deviations in the characteristics under evaluation.
Although very similar to -charts, control charts present two distinct characteristics: the vertical
axis does not represent the quality characteristic under evaluation, but a statistic test named ; and
the LCL is always zero.
This multivariate statistical parameter control can be also applied to individual observations or
subgrouped data (subsamples). When every observation is checked (i.e. each point of the plot
represents one observation), the - statistic and the UCL are computed according to (Montgomery,
2009):
. ̅ . . ̅
1
(7.14)
1 .
. ,
3
with:
Number of observations collected during the reference period
Individual observation (vector with components)
Covariance matrix
, percentage point of the distribution with and degrees of freedom
If, instead of checking all the observations, subsamples with observations are verified, the -
statistic and the UCL are computed according to (Montgomery, 2009):
. ̅ ̅ . . ̅ ̅
. 1 . 1 (7.15)
. , .
. 1
with:
̅ Subgroup average
̿ Process average
Number of groups collected during the reference period
7.2.2 FATIGUE MONITORING

Wind turbines are structures predominantly subjected to dynamic loads throughout their period of
life. In that sense, the continuous assessment of the fatigue condition of structural components can
anticipate a fatal damage. Although failures from fatigue damage are also not expected to occur (if the
structural design is correct), there is a large uncertainty about the real loading condition that the wind
turbines will face. Thereby, the conditions considered during the design stage might not correctly
recreate the actual conditions, leading to an over or underestimation of the components life. For these
reasons, the implementation of a fatigue monitoring system can be an important advantage for the
management of wind farms.
259
Support structures, as seen in Chapter 3, are a fundamental component of wind turbines that cannot
fail at any circumstance, otherwise the whole system is jeopardized. Furthermore, their design is
usually very conservative with respect to fatigue, which may lead to real fatigue life considerably higher
than 20 years (the usually considered expected fatigue lifetime). Thus, the implementation of a fatigue
monitoring system for assessment of the real solicitation of the support structure can introduce several
advantages. If the monitoring system indicates that damage is not present in the structure, the fatigue
monitoring system can be extremely useful in the following aspects:
o Estimation of the evolution of real fatigue condition of the support structure and its suitability
with the expected lifetime (usually 20 years);
o Since an estimation of the real condition of fatigue damage is known, the results of the fatigue
monitoring system could be an essential element for a decision about extending the lifespan of
the structure, leading to a higher profitability of the investment;
o Possibility of repowering or overpowering of the system, while keeping the support structure.
If the fatigue monitoring system indicates a considerable margin of safety at the end of the
period of life initially stipulated, together with a situation of non-damage detection by the
dynamic monitoring system, one of these two options could be considered. For that purpose, a
careful inspection of the structure would have to be carried out on-site. However, the results
obtained with both components of the monitoring system, could provide a factual evidence of
the overall good condition and low level of structural loading during the first period of life.
The decision about repowering and overpowering could greatly reduce the cost of tower +
foundation portion, which contributes with 18 % and 23 % of the capital cost for onshore and
offshore installations, respectively (see section 3.1.2).
The description of fatigue monitoring systems and publication of results from long periods of data
acquisition is not very common in literature. In the context of the HISTWIN project (Veljkovic,
Heistermann et al., 2012), a 2.1 MW wind turbine tower was monitored with strain rosettes, among
other sensors. Results from two periods of 139 and 159 days for the most unfavourable structural detail
lead to the conclusion of a considerable overdesign of the fatigue strength.
Another work considering the problem of fatigue in the support structure of a wind turbine is reported
in (Pollino and Huckelbridge, 2012). A 100 kW wind turbine was monitored with strain gauges during
approximately one year. The authors studied the principal structural details of the tower and also
concluded that the fatigue life of these elements were significantly higher than 20 years.
An interesting study is described by Thies et al (Thies, Johanning et al., 2014) about the fatigue damage
assessment of the mooring lines of an offshore floating marine energy converter. This equipment, that
also comprises a 1.2 m diameter wind turbine, was monitored with load cells to measure the tension
forces in the mooring system.
Outside the scope of wind turbine support structures, there is a greater range of studies about the
assessment of fatigue and prediction of the remaining period of life on steel structures. As an example,
Ye et al (Ye, Ni et al., 2012) presented a study about a long-term monitoring data of dynamic strain on
a steel bridge (the Tsing Ma Bridge).
260
Chapter 7
7.2.2.1 Estimation of the Remaining Lifetime

The results of fatigue damage of the support structure obtained during a representative time period
may be used to estimate the remaining fatigue life of the structure. A direct way of doing this is
considering the monitored period as representative of the whole period of life. In that case, the
estimated accumulated damage at the end of the design fatigue life of the structural detail is obtained
as:
. (7.16)
with:
T Design fatigue life (usually 20 years)
Period of time which was considered for fatigue monitoring
Accumulated damage throughout the monitored period
However, this extrapolation assumes that the monitored period is representative of the loading
scenarios from the whole period of operation of the wind turbine. Considering that the monitored
period may not be representative of the environmental loading along the estimated period of life, a
more complex approach should be followed.
An alternative approach can be obtained if the information about the environmental data is recorded.
In that case, loading scenarios should be defined according to the simultaneous occurrence of the
various environmental parameters. Important parameters such as mean wind speed, turbulence
intensity and wind direction should be considered. Also, the significant wave height and peak period
should be included for offshore installations. The variability of the parameters is usually reduced into
bins with a representative magnitude in order to ease the computation of probability. Figure 7.5 shows
part of the scatter diagram from the OWEZ site, where the probability of occurrence of loading
scenarios with a mean wind speed of 10 m/s combined with various wave conditions is shown. In this
figure, the wave loading condition is defined by the significant wave height (introduced in section
6.2) and the mean zero-crossing period (instead of the peak period ). This period is referred to
the mean value of the zero up-crossings of a point at the sea surface (Tempel, 2006).
261
6
5.5
5
4.5
4
3.5
Hs 3
2.5 0.000038 0.000038
2 0.000342 0.000342
1.5 0.017727 0.001674 0.019401
1 0.003918 0.089550 0.093468
0.5 0.048009 0.004831 0.052840
0 0.000266 0.000038 0.000304
0 1 2 3 4 5 6 7 0.166394
0.166394
Vw = 10m/s Tz
Figure 7.5 – Part of scatter diagram corresponding to the mean wind speed of 10 m/s (the values represent the
probability of occurrence of the loading scenarios) (Tempel, 2006)
Once a representative probability of occurrence of the environmental loading scenarios is obtained,

the accumulated damage at the end of the design fatigue life of the structural detail can be estimated
according to:
. , ,… . (7.17)
with:
T Design fatigue life (usually 20 years)
Period of time which was considered for fatigue monitoring
Probability of occurrence of the load scenario
Accumulated damage of load scenario during the monitoring period ( )
The monitoring period of the structure (which might be much shorter than the monitoring period of
the loading conditions) should include all the important load conditions for the quantification of D .
7.2.3 OVERVIEW OF THE DYNAMIC MONITORING SYSTEM MAIN STEPS

At this point, all the important aspects of the dynamic monitoring system were already introduced.
This section briefly summarizes the chain of processes throughout the developed system.
The chart with the chain of procedures is presented in Figure 7.6. The processing starts with the
acquisition of the 10 min. acceleration time series and data from environmental and operational
conditions. Initially, the acceleration signals are decimated and a coordinate transformation is applied
to these signals according to the yaw angle recorded by the SCADA system, in order to orientate the
signals according to the FA and SS direction.
After this preprocessing, the module related to structural damage detection starts with the automated
modal analysis. The first step is referred to the application of the output-only identification algorithms
262
Chapter 7
(SSI-COV, SSI-DATA and p-LSCF) to the acceleration signals. These algorithms were introduced in
section 5.4. The results obtained are then analysed through a cluster analysis (section 5.7.1), in which
the poles with similar modal properties are gathered into the same clusters. Lastly, these clusters are
compared with reference modal properties (frequency values and mode shapes) of the vibration modes
intended to be tracked in order to separate the clusters referred to these modes from the other ones.
These reference properties are defined during a training period prior to the automated processing.
Once they are defined, these properties are saved in a database, being used in every processing cycle.
From the moment the modal tracking is complete, it is necessary to remove the influence of the
operational and environmental effects on the frequency values of the identified modes. In that sense,
multivariate linear regression models have to be defined during a representative period (again, prior to
the automated process). Once these models are set, they are applied to the identified frequency values
in order to reduce their variability. This process was presented in section 7.2.1.1.
The last step of the module is referred to the assessment of important deviations of the frequency
values (damage detection). The implemented procedure, based on control charts, was introduced in
section 7.2.1.2. A caveat should be made at this level that, if subsamples with more than one
observation are used, the damage detection step is only processed after the required number of
observations is achieved.
The module related to the fatigue assessment of the wind turbine support structure starts after the
application of the output-only identification algorithms to the acceleration signals. At this point, it is
necessary to select an order of the state-space model. This selection should take into consideration the
order with the highest number of identified modes with important contribution to the support
structure motion. For that reason, the results from the modal tracking represent an important
information. Furthermore, the selected order should contain poles related to the low order harmonics
(when the turbine is operating). Thus, the data from the SCADA system is also important at this point.
Once the state-space model order is selected, it is possible to define the forward-innovation model.
With the transformation of the model into the modal basis, it is thus possible to obtain the modal state
vector, essential information to estimate the acceleration at unmeasured locations (this procedure was
presented in section 5.5). At this point, a detailed estimation of the mode shapes along the height of the
support structure is required to estimate the acceleration time series at every location. Two options are
available:
o If the number of installed sensors at the structure is enough to estimate the mode shapes of the
most important modes along the height of the support structure, an interpolation of the
experimentally found mode shapes can be used. This methodology was referred in section
6.4.1;
o On the other hand, if a reduced number of sensors is used, reference support structure mode
shapes of the important modes should be used. In addition, operational deflection shapes of
the harmonics should also be used for situations under operation. Both mode shapes and
operational deflection shapes must be defined after a period of training and should consider
the different operating conditions of the wind turbine. This procedure was described in section
6.4.2.
263
With the possibility of estimate the acceleration time series at every position of the wind turbine
support structure, the fatigue damage of the structure is assessed with the procedure proposed in
section 6.4.1. During this process, the stiffness matrix of the support structure is required and should
be previously calculated according to the mechanical properties of the support structure.
Lastly, the fatigue damage estimations from the 10 min. time series should be continuously summed in
order to obtain a cumulative damage of the support structure. Once representative damage
accumulation events from the various environmental loading scenarios are recorded, it is possible to
perform an estimation of the damage condition at the end of the designed fatigue life and, thus, predict
the real lifetime of the structure. This procedure was introduced in section 7.2.2.1.
Acceleration Data SCADA Data
Signal Decimation
Coordinate Transformation
Automated Modal Analysis
Fatigue Assessment
Definition of the
Modal Identication
State-Space Model
Definition of the
Cluster Analysis
Forward-Innovation Model
Database Database
Mode Shape/ Operational Reference Mode Shapes/

Reference Modal Properties Modal Tracking
Deflection Shape Estimation Operational Deflection Shapes
Removal Operational/
Multivariate Regression Model Fatigue Damage Stiffness Matrix
Environmental Conditions
Accumulated Damage/
Damage Detection
Lifetime Prediction
Figure 7.6 – Chain of events included in the dynamic monitoring system (the dashed box is only referred to situations in
which a reduced number of sensors is used)
264
Chapter 7
7.3 CASE STUD

DY – SENVIO
ON MM82
7.3.1 INTRODU
UCTION
This section intro

oduces the main
m results oobtained with
h the applicattion of the viibration-baseed structurall
h monitoring methodolo
health ogy to a 2.0 M
MW Senvion
n MM82 win nd turbine. Inn that sense, a long-term m
studyy was perform
med for this wind
w turbinee.
The ggoal of this sttudy was to demonstrate
d e the suitabiliity of dynamic monitorinng systems to
o be installed
d
at win
nd turbines. The work co ontemplated several stepss:
o Installatio
on of the mo
onitoring equuipment;
o Implemen
ntation of th
he structural hhealth monittoring tools in
i an automaated way;
o Full comp
prehension of
o the dynam
mic behaviou
ur of the win
nd turbine inn different op
perating and
d
environm
mental condittions;
o Identificaation of the main
m vibratioon modes;
o Assessmeent of the capability of thhe implemen
nted dynamiic monitorinng system to detect smalll
damages;
o Fatigue asssessment off the wind tuurbine supporrt structure;
o Study ab
bout the feasibility of thhe implementation of monitoring
m systems witth a smallerr
number of
o sensors.
7.3.2 WIND TURBINE

U DESCRIPTION AND
D NUMERICA
AL MODELS
The SSenvion MM
M82 is a 2.0 MW
M onshore wind turbin ne. This wind
d turbine starrted operatin
ng in 2007 att
the T
Torrão wind farm,
f in the north
n of Porttugal (Figuree 7.7).
Figure 7.7 – Location of th

he Torrão wind farm
265
5
Structural Monittoring of Wind Tu
urbines
This wind tturbine is ch haracterized by 82 m diaameter, up-w wind rotor with

w variablee speed operration,
with three ppitch controolled blades. The generattor of this model
m is of th
he asynchronnous type an
nd the
hub is locateed at 80 m high.
Figu
ure 7.8 – Senvio
on MM82 wind turbine installe
ed at the Torrão wind farm
The supporrt structure comprises

c a reinforced
r cooncrete slab foundation and a steel tutubular towerr. The
tower is composed by three segmeents, totalizinng 76 m in height. Thesse segments are connectted by
bolted flangged connections. The maain technical characteristics of the wind turbine aare summarizzed in
Table 7.1.
266
Chapter 7
Table 7.1 – Main technical characteristics of the Senvion MM82
Model Senvion MM 82
General
Rated-power [MW] 2.0

Cut-in wind speed [m/s] 3.0
Rated Wind speed [m/s] 13.0
Cut-out wind speed [m/s] 25.0
IEC class IIa
Rotor
Nº blades 3
Horizontal axis
Rotor orientation
Up-wind rotor
Rotor speed range [rpm] 8.5 – 17.1 (±16 %)
82.0
Rotor diameter [m]
(Blade length: 40 m)
Control systems Pitch and rotor speed control
Machinery
Variable-speed
Generator type
asynchronous generator
3 stage Planetary/ helical gear
Gearbox
(ratio: 105.4)
Yaw system Active system with disc brake
Tower
Type Steel tubular

Hub height [m] 80
Base diameter [m] 4.300
Top diameter [m] 2.955
Aiming to obtain an initial approach of the modal properties of the Senvion MM82 wind turbine, two
numerical models were developed. Initially, the structural behaviour of the tower was modelled with
the ANSYS software (Ansys Inc., 2011), according to technical drawings made available by the
manufacturer. Four-node shell elements with six degrees of freedom at each node were used to model
the shell components of the tower, while the flanged bolt connections were simulated with beam
elements. For the nacelle and rotor blades components, a point element with a representative
concentrated mass was employed on their centre of gravity. The connection between tower and
foundation was defined as fixed. Figure 7.9 presents a global view and a detail from the door opening
of the finite element model.
267
1 1
ELEMENTS ELEMENTS
NOV 21 2012 NOV 21 2012
MAT NUM 15:10:03 MAT NUM 15:15:14
PLOT NO. 1 PLOT NO. 1
Y
X
Z
Z X
Figure 7.9 – Global view and detail of the door opening of the Senvion MM82 finite element model
The second numerical model of the Senvion MM82 wind turbine was developed using the HAWC2
aeroelastic code (Larsen and Hansen, 2007). For this model, an improvement on the previous model
developed using ANSYS was introduced. In addition to modelling the tower structure according to
technical drawings, the nacelle and rotor blades were also modelled. Since the characteristics of these
elements were not provided, the characteristics from the NREL 5MW reference wind turbine
(Jonkman, Butterfield et al., 2009) were scaled down to meet the characteristics of the Senvion MM82
wind turbine. The connection between tower and foundation was also defined as fixed.
The results obtained after the modal analysis of both models for the frequency values of the first 2 pairs
of tower bending modes are shown in Table 7.2. It is observed that the results achieved with both
models are coherent, giving a good indication about the expected frequency values of these modes.
Table 7.2 – Natural frequency values of the first 2 pairs of tower bending modes obtained with the numerical models
developed using ANSYS and HAWC2
Natural frequency [Hz]
Modes ANSYS HAWC2
1 SS 0.378 0.366
1 FA 0.378 0.361
2 FA 2.963 2.771
2 SS 2.967 2.830
The mode shapes of the referred modes are illustrated in Figure 7.10. It is seen that second pair of
bending modes includes an important participation of rotor blades motion, while this contribution is
imperceptible in first pair.
268
Chapter 7
1 SS
1
NODAL SOLUTION
MX NOV 21 2012
STEP=1 14:32:38
SUB =2 PLOT NO. 1
FREQ=.378684
USUM (AVG)
RSYS=0
DMX =.002921
SMX =.002921
Y
Z X
MN
0 .649E-0
3 .001298 .00194
7 .002597
.325E-03 .974E-03 .001623 .002272 .002921
Senvion MM 82
1 FA
1
NODAL SOLUTION
NOV 21 2012
STEP=1 14:30:44
SUB =1 PLOT NO. 1
FREQ=.378154
USUM (AVG)
RSYS=0
DMX =.002921
SMX =.002921
Y
Z MN X
0 .649E-0
3 .001298 .00194
7 .002597
.325E-03 .974E-03 .001623 .002272 .002921
Senvion MM 82
2 FA
1
NODAL SOLUTION
NOV 21 2012
STEP=1 14:32:53
SUB =3 PLOT NO. 1
FREQ=2.96267
USUM (AVG)
RSYS=0
DMX =.004349
SMX =.004349
MX
Y
Z MN X
0 .966E-0
3 .001933 .00289
9 .003865
.483E-03 .00145 .002416 .003382 .004349
Senvion MM 82
2 SS
1
NODAL SOLUTION
NOV 21 2012
STEP=1 14:32:14
SUB =4 PLOT NO. 1
FREQ=2.96982
USUM (AVG)
RSYS=0
DMX =.004357
SMX =.004357
MX
Y
Z X
MN
0 .968E-0
3 .001936 .00290
5 .003873
.484E-03 .001452 .00242 .003389 .004357
Senvion MM 82
Figure 7.10 – Mode shapes of the first 2 pairs of tower bending modes obtained with the numerical models (left column:
ANSYS; right column: HAWC2)
269
7.3.3 OPERATING CONDITIONS

The Senvion MM82 operates within a wind speed regime between 3 and 25 m/s, reaching its rated
power at a wind speed of 13 m/s. This power production condition is expressed in the power curve of
the turbine (Figure 7.11).
Region Region Region Region Region Region
1 2 3 4 5 6
2000
1500
Power [kW]
1000
500
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
Wind speed [m/s]
Figure 7.11 – Senvion MM82 power curve
Figure 7.11 also indicates the various operating conditions of the wind turbine. The first region is
upper-limited by the cut-in speed of the turbine and is characterized by very low wind speeds. In this
region, the wind speed is not enough to create a consistent rotor torque and, thus, the turbine does not
operate and the rotor is usually parked or idling. While in this region, the control strategy of the
turbine involves the analysis of the wind condition in order to evaluate a feasible start of the turbine
operation.
When the wind speed is higher than the cut-in value, the turbine enters in Region 2. In this region, the
wind turbine starts the power production regime in which the rotor operates close to an optimal tip-
speed ratio. This regions corresponds to line BC’ in Figure 3.39 and characterized by an important
variation of rotor speed.
Once the Region 3 is achieved, the wind turbine enters in a regime defined by line C’D’ in Figure 3.39.
This is a transition condition in which the rotor torque achieves its rated value while the rotor speed
slightly increases. During both Regimes 2 and 3, the pitch angle of the blades is kept constant on its
minimum value and the system acts on the yaw actuator to keep the rotor orientated to the main
direction of the wind.
Region 4 is defined by the start of operation of the blades pitch actuator. In this region, the rotor
reaches its rated speed and the blade pitch angle is adjusted to keep the rotor speed under defined
values.
Once the turbine is operating at its rated power, the Region 5 is reached. In this region, the main goal
of the control system is to keep constant this production level. For that goal, the blade pitch angle is
controlled over this region, increasing its value with the increase of the wind speed. The actuation of
the control system over the blade pitch permits to control the rotor speed, keeping the production
constant, as well as to avoid exceeding electrical and mechanical load limits.
270
Chapter 7
The last region of Figure 7.11 is referred for wind speeds higher than the cut-out value. At this point,
the wind load acting on the system is excessive and the production is stopped. In order to avoid
extreme loads, the turbine is shutdown, the blades are pitched out of the wind.
These six regions define, in a simplistic way, the control system of a wind turbine for different wind
speed conditions. Thus, if it is desired to track the modal properties of the wind turbine over its period
of life, the variations occurring within each region need to be considered. For that reason, the data
recorded by the SCADA system has an important role in the monitoring system.
The different operating regions are also evident in the evolution of the rotor speed with the wind speed
(Figure 7.12). In this figure, the rotor speed range from the variable-speed generator system is evident.
During the period in analysis, Region 6 was not detected.
Region Region Region Region Region
1 2 3 4 5
18
16
14
Rotor speed [rpm]
12
10
0
0 5 10 15 20 25
Wind speed [m/s]
Figure 7.12 – Rotor speed vs wind speed
The behaviour of the pitch angle with the rotor speed is depicted in Figure 7.13. In this figure, the
different operating conditions of the wind turbine system are also visible. Up to the wind cut-in speed,
the pitch angle is kept with high values (around 70º to 90º). Under this condition, the blades are only
able to slowly rotate, which is desire since the wind speed is too low for production. From the moment
the wind speed is higher than the cut-in speed, the pitch decreases and the generator starts operating
(from around 8.8 rpm). The pitch angle is kept low until the Region 4 is reached. Since the pitch
actuator acts practically at the rated rotor speed, the increase of the pitch angle in Region 4 and 5 is not
completely perceptible in the plots of Figure 7.13.
271
100 25
80 20
15
Pitch angle [º]
Pitch angle [º]

60
40 10
20 5
0 0
0 5 10 15 8 10 12 14 16
Rotor speed [rpm] Rotor speed [rpm]
a) b)
Figure 7.13 – Pitch angle vs rotor speed: a) all regimes; b) only operating regimes
The illustration of evolution of the blades pitch angle actuator with the wind speed (Figure 7.14) is
helpful to understand the pitch mechanism in Region 4 and 5. In this figure, the start of the actuator
around 11 m/s is clearly visible. From the observation of Figure 7.14 b), it is also possible to attest that
the increase rate of the pitch angle with the wind speed is lower in Region 5 than in Region 4.
Region Reg. Region Reg. Region
1 2 3 4 5
100 25
80 20
15
Pitch angle [º]
Pitch angle [º]
60
40 10
20 5
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Wind speed [m/s] Wind speed [m/s]
a) b)
Figure 7.14 – Pitch angle vs wind speed: a) all regimes; b) only operating regimes
The analysis of the recorded SCADA data also permitted to conclude that the main wind incidence
direction is around 110º. It is concluded, from the wind rose of Figure 7.15, that the higher wind speed
regimes are also obtained for this direction.
272
Chapter 7
0º
NORTH
15%
10%
N
5% W E
Wind speed
110º
270º 90º [m/s]:
WEST EAST S
22 - 24
20 - 22
18 - 20
16 - 18
14 - 16
12 - 14
10 - 12
8 - 10
6-8
4-6
SOUTH 2-4
0-2
180º
Figure 7.15 – Wind rose histogram of mean wind speed Figure 7.16 – Local orography on the wind turbine site
The local orography at the wind turbine site is represented in simplified form in Figure 7.16. In this
figure, the origin of the referential represents the wind turbine location. It is interesting to note that at
the direction of 110º, the terrain presents an apparent depression.
7.3.4 MONITORING SYSTEM DESCRIPTION

The implemented monitoring system is based on a central acquisition system to which all sensors are
connected. It is composed by 9 uni-axial accelerometers distributed along the tower height and at the
foundation level, according to the scheme presented in Figure 7.17. All the accelerometers were fixed
to the tower through fixing clamps on the tower flanges. The central acquisition system is located at
the first platform, around +20.000 m. All the equipment is placed inside the wind turbine.
With the considered instrumentation layout, the acceleration on both horizontal directions at three
different tower sections is measured. In addition, the pair of sensors S5 and S7 enables the
characterization of motion associated with torsion modes.
+ 0.000 + 21.772
S1
S3
S9 S4
+ 74.988
S8 S2
S6
+ 48.392
S5 S7 + 48.392 + 74.988
S4 S6
S3 + 21.772 S7
S9
S5 S8
S1 S2
+ 0.000
a) b)
Figure 7.17 – Position of the accelerometers at the different levels of the wind turbine: a) front view; b) top view of the
instrumented sections
273
Structural Monittoring of Wind Tu
urbines
The installeed sensors are

a force balaance accelerrometers, mo odel CMG-5 5S from Gurralp (Figure 7.18).
This accelerrometer pressents a dynaamic range oof 140 dB, wiith a frequen
ncy band froom DC to 1000 Hz.
The measurring range vaaries from േ0.1g
0 to 4.0g.
The centrall acquisition system is co
omposed by a CR-5P unit from GeoS SIG (Figure 7.18). This unit
u is
composed bby a digitizerr (CR-5PAD24) and an iinternal com mputer (CR-55PRHDx). Thhe digitizer allows
a
the connecttion of up to 9 uni-axiaal accelerom meters and is equipped with
w a 24-bitt analog-to-d digital
converter. T
The internal computer co ontrols the reecording opeerations.
The CR-5P unit was configured to record 10 m min. accelerattion time series with a fre
requency sam
mpling
of 50 Hz. T
The length off the time serries was deciided in order to match thhe 10 min. pperiod used by
b the
SCADA sysstem to comp pute the meaan values of thhe monitored parameters.
a) b)
c) d)
wer flange; and c) foundation flange. d) Centtral
Figure 7.18 – Accelerometers installed at: a) tower fla nge; b) top tow
acquisition
a systtem with the CR
R-5P unit
274
Chapter 7
The dynamic monitoring system is complemented by the SCADA system of the wind turbine. This
system records the mean, maximum and minimum value from 10 min. period of several operational
and environmental parameters. Among them, some were important for the context of this structural
health monitoring project:
o Wind speed and direction;
o Rotor speed;
o Yaw angle;
o Blades pitch angle;
o Outdoor temperature.
Since it was not possible to connect the central acquisition system to an external GPS antenna, the data
recorded by the CR-5P was manually synchronized to the SCADA data. This operation was tuned
using the date information of startup and shut-down events of the turbine recorded by the SCADA
data, which were correlated to events of sudden increase or decrease of the tower vibration levels.
The dynamic monitoring system was installed at the wind turbine on the 22th of July of 2013.
Unfortunately, problems related to electrical power supply of the central acquisition system prevented
a continuous operation. Moreover, it was only possible to have access to the equipment during
maintenance periods, which annulled the possibility to make rapid interventions once the problems
were noticed. For that reason, it was not possible to avoid some periods with a low rate of recorded
setups (or even inactivity of the system). A total number of 37 685 data sets were collected during this
period. The percentage of recorded setups for each month between 22/07/2013 and 21/07/2014 is
evidenced in Table 7.3.
Table 7.3 – Percentage of recorded setups between 22/07/2013 and 21/07/2014
Year 2013 2014
Month 7 8 9 10 11 12 1 2 3 4 5 6 7
Recorded
100 100 63 0 23 89 89 100 100 94 67 59 53
Setups [%]
7.3.5 PRELIMINARY RESULTS

The processing of the continuously collected data comprehends an initial task of filtering and
resampling to reduce the sampling frequency to 25 Hz. After this step, a coordinate transformation is
applied to the acceleration time series measured at the tower in order to obtain signals that are always
aligned with the FA and SS directions.
Examples of the different excitation conditions that the wind turbine is subjected to are illustrated in
Figures 7.19 to 7.21. Figure 7.19 shows the acceleration time series of the recorded setup with the
highest observed mean wind speed (23.3 m/s). On the other hand, Figures 7.20 and 7.21 present two
situations of sudden increase and decrease of the vibration level related to a situation of startup and
shutdown of the rotor, respectively.
275
10/02/2014 00:50 31/07/2013 12:40 24/07/2013 04:40

2 0.8 0.4
1 0.4 0.2
Acc. [m/s2]
Acc. [m/s2]
Acc. [m/s 2]
0 0 0
-1 -0.4 -0.2
-2 -0.8 -0.4
0 200 400 600 0 200 400 600 0 200 400 600
time [s] time [s] time [s]
Figure 7.19 – Acceleration time series Figure 7.20 – Acceleration time series Figure 7.21 – Acceleration time series
from a setup with the highest during a startup event during a shutdown event
observed mean wind speed (23.3 m/s)
The variation of the vibration levels during the monitored period is shown in Figures 7.22 and 7.23 for,
respectively, the RMS and maxima values from each recorded 10 min. time series. The maximum
acceleration recorded in FA direction was 3.73 m/s2 while in the SS direction was 3.10 m/s2. As
expected, the vibration levels at the foundation are considerably lower than in the tower.
RMS
0.6
+74.988 FA
Acc. [m/s 2]
0.4 +74.988 SS
+48.392 FA
0.2 +48.392 FA
+48.392 SS
+21.772 FA
0 +21.772 SS
0 0.5 1 1.5 2
Index 4
x 10
-3
x 10 RMS
5
Acc. [m/s 2]
2.5 Found. 6
Found. 7
0
0 0.5 1 1.5 2
Index 4
x 10
Figure 7.22 – Time evolution of the RMS values of the acceleration time series collected during the monitored period
276
Chapter 7
Maxima
4
3 +74.988 FA
Acc. [m/s 2] +74.988 SS
2 +48.392 FA
+48.392 FA
1 +48.392 SS
+21.772 FA
0 +21.772 SS
0 0.5 1 1.5 2
Index 4
x 10
Maxima
0.03
Acc. [m/s 2]
0.02
Found. 6
0.01 Found. 7
0
0.5 1 1.5 2
Index 4
x 10
Figure 7.23 – Time evolution of the maxima values of the acceleration time series collected during the monitored period
It is interesting to observe that the higher values of acceleration do not occur at the tower top but at
+48.392 m level. This is a common situation when the turbine is operating (the majority of the time).
However, when the turbine is parked or idling, the highest vibration levels were recorded by the top
sensors (+ 74.988 m). This behaviour was also noticed in the analysis of the offshore Vestas V90 wind
turbine (section 5.9).
Figure 7.24 shows the RMS values of acceleration in the FA direction for operating and non-operating
situations. It can be seen that, when the turbine is operating, the acceleration level at the +48.392 m
height is clearly higher than at the other two levels. Indeed, the vibration levels at the top (+ 74.988 m)
and +21.772 m are similar. On the other hand, when the turbine is not operating, the level of vibration
usually increases from the lowest part to the top of the tower structure. Nevertheless, the vibration at
the +48.392 m sometimes exceeds the value from the top sensor. The causes for this behaviour will be
further analysed in section 7.3.6.4.
Operation Parked/ Idling
0.5 0.07
+74.988 FA +74.988 FA
+48.392 FA 0.06 +48.392 FA
0.4 +21.772 FA +21.772 FA
0.05
Acc. [m/s 2 ]
Acc. [m/s 2 ]
0.3 0.04
0.2 0.03
0.02
0.1
0.01
0 0
4000 4100 4200 4300 4400 4500 1250 1300 1350 1400 1450 1500 1550 1600
Index Index
a) b)
Figure 7.24 – Time evolution of the RMS values of the acceleration time series during: a) operating conditions and; b)
parked or idling conditions
277
The wind excitation is, naturally, the main driver of wind turbine vibration. Figure 7.25 presents the
time evolution of the RMS values at the +74.988 m level in the FA direction and of the wind speed. As
expected, there is a clear coherence between acceleration level and wind speed.
The relationship between the RMS values of acceleration and the wind speed is also illustrated in
Figure 7.26. From the figure, it seems that the acceleration continuous increases with the increase of
the wind speed.
0.25 25 0.6
+74.988 FA
+48.392 FA
0.5
0.2 20 +21.772 FA
0.4
Wind speed [m/s]
Acc. [m/s 2]
Acc. [m/s 2]
0.15 15
0.3
0.1 10
0.2
0.05 5
0.1
0 0 0
8600 8800 9000 9200 9400 9600 0 5 10 15 20 25
Index Wind speed [m/s]
Figure 7.25 – RMS values of acceleration of the +74.988 m Figure 7.26 – Correlation of the RMS values of the sensors
sensor in the FA direction vs the wind speed in the FA direction with the wind speed
The variation of the vibration amplitude with the yaw angle is an important analysis to check the
directions with greater wear of the support structure. As expected, the vibration level is highly
dependent on the rotor orientation (which is conditioned by the wind direction). Figure 7.27 presents
the RMS values of the acceleration in FA direction according to the yaw angle. As can be seen, this
figure is coherent with the main wind directions illustrated in the wind rose of Figure 7.15.
0.6
+74.988 FA
+48.392 FA
0.5
+21.772 FA
0.4
Acc RMS [m/s 2]
0.3
0.2
0.1
0
0 40 80 120 160 200 240 280 320 360
Yaw angle [º]
Figure 7.27 – RMS values of the acceleration in the FA direction according to the yaw angle
Figure 7.27 clearly indicates the directions according to which the structure vibrates with the highest
levels. However, as the wind speed is not evenly distributed across all the directions (as shown in
Figure 7.15), it is not possible to make any consideration about the heterogeneity of the support
structure just with this observation.
278
Chapter 7
In order to compare the vibration amplitude for different angles of nacelle orientation, only the
recorded setups from a narrow range of wind speeds, under operating conditions, were selected. With
this consideration, it can be assumed that the wind excitation is roughly similar across all directions.
Still, differences can exist in terms of wind turbulence.
Figure 7.28 presents the box plots obtained for the RMS values of acceleration in the FA direction for
the top sensor considering only setups when the turbine was operating and the wind speed was
between 6 and 7 m/s. For the analysis, the yaw angles were grouped in 18 ranges of 20º, from 0º to
360º. In this figure, the line connects the median value of each group, while the edges of the box
represent the 25th and 75th percentiles. The whiskers were extended to the most extreme points not
considered as outliers. In the background, the data considered for the analysis is plotted in light grey.
This figure clearly indicates that, for roughly similar wind conditions, the structure presents higher
vibration levels for the ranges 100º – 140º and 180º – 240º than for remain directions. The first range
coincides with the main direction of the wind.
Wind speed: 6 - 7 m/s 0.12
0.2 Wind speed [m/s:]
3-4
0.1 4-5
5-6
0.15 6-7
0.08
7-8
Acc. [m/s 2]
Acc. [m/s 2]
0.1 0.06
0.04
0.05
0.02
0 0
0 40 80 120 160 200 240 280 320 360 0 40 80 120 160 200 240 280 320 360
Yaw angle [º] Yaw angle [º]
Figure 7.28 – Box plots of the RMS values of acceleration of Figure 7.29 – Median values of the RMS values of
the top sensor in the FA direction according to the yaw acceleration of the top sensor in the FA direction
angle, considering setups when the turbine was operating according to the yaw angle, considering setups when the
and the wind speed was in the 6 – 7 m/s range turbine was operating and several wind speed ranges
This analysis is further extended to ranges of wind speed whose number of recorded setups is
representative of all yaw sectors. For that reason, ranges with the largest wind speeds were not
considered since they mainly occur in a few yaw sectors. Figure 7.29 presents the results obtained with
this analysis, considering wind speed ranges between 3 and 8 m/s. In this figure, only the median
values are shown. The results obtained are in line with the conclusion already introduced for Figure
7.28. It is interesting to note that, for example, the highest values obtained with the wind speed range 4
– 5 m/s are similar to the values obtained for the 7 - 8 m/s in the 300º - 360º region. From this analysis,
it seems that the foundation is less stiff along the directions that are more excited by the wind.
However, it should be noticed that this analysis should have taken into account the turbulence of the
wind flow, which also influences the dynamic excitation of the support structure. Since this
information is not provided by the SCADA system, it is not possible to confirm the presented
considerations.
It is also interesting to assess the evolution of RMS values of acceleration with the rotor speed (Figure
7.30). As expected, the vibration levels tend to increase with the increase of the rotor speed.
279
0.6
+74.988 FA
+48.392 FA
0.5
+21.772 FA
0.4
Acc. [m/s 2]
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16
Rotor speed [rpm]
Figure 7.30 – RMS values of acceleration vs the rotor speed
The colour map presented in Figure 7.31 illustrates the frequency content of the acceleration signals
during a period of one month. This figure is a top view of the first singular value spectra of the
spectrum matrices obtained from each recorded setup. The regions with the hotter colours represent
the highest energy. Two vertical alignments with high energy are clearly visible around 0.35 Hz and
2.80 Hz (indicated by arrows). These frequency values are coherent with the results obtained for the
first and second pairs of tower bending modes of the numerical models described in section 7.3.2.
Albeit with less energy, two additional alignments are also visible around 1.30 Hz and 1.80 Hz (also
indicated by arrows).
In the same figure, the rotor speed frequencies from the 1Ω, 3Ω and 6Ω harmonics are represented by
dashed lines. As expected, the excitation introduced by the rotor rotation is visible, mainly for the 3Ω
harmonic. In a small scale, the 6Ω harmonic is also detected and seems to cross the 1.30 Hz alignment
several times. The 1Ω harmonic alignment is very tenuous and is only visible in a few setups.
1Ω 3Ω 6Ω
Time
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Frequency [Hz]
Figure 7.31 – Colour map with the variation of the signals frequency content during the 15/02/2014 and 15/03/2014
280
Chapter 7
7.3.6 CONTINUOUS CHARACTERIZATION OF THE DYNAMIC PROPERTIES

The design of a dynamic monitoring system requires the previous knowledge of the wind turbine
modal properties. In that sense, the results obtained with an ambient vibration test (Oliveira,
Magalhães et al., 2013) and with the analysis of initial recorded data were used to characterize the
dynamic properties of the wind turbine. This initial characterization was used to tune the dynamic
monitoring system to detect possible damage situations.
7.3.6.1 Dynamic Characterization of the Wind turbine

As presented in section 7.3.3, wind turbines are time-varying structures subjected to different
operating conditions. As example, Figure 7.32 presents the averaged normalized power spectrum
density of two different setups, one referring to a non-operating situation and another referred to an
operating condition. The shape of the spectra are considerable distinct. The most distinguished feature
is the appearance of peaks at the harmonics frequencies in the operating setup. These peaks are due to
the external excitation caused by the tower shadow effect. From the figure, it is also visible a small peak
around 1Ω. As expected, this peak is considerable smaller than the ones associated with other
harmonics.
Besides the harmonic peaks, other interesting characteristics are also evident. The two peaks around
1.10 Hz and the peak at 2.30 Hz, clearly identified under non-operating conditions, are practically not
visible in the operating setup. Furthermore, the peaks from the non-operating conditions spectrum at
1.30 and 1.80 Hz seem to move to slightly higher frequency values.
0 1Ω 3Ω 6Ω 9Ω 12Ω 15Ω 18Ω
10
Non-operating
Operating
-2
10
Amplitude
-4
10
-6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency [Hz]
Figure 7.32 – ANPSD of two different setups: under non-operating conditions (rotor speed = 0 rpm; wind speed = 1.9 m/s)
and operating conditions (rotor speed = 16.4 rpm; wind speed = 10.7 m/s). The vertical dashed lines indicate the
frequency values of the rotor speed and its harmonics
It is thus important to perform a preliminary analysis of the properties of the most relevant vibration
modes and their variation throughout the various operating conditions. In that sense, the recorded
setups were initially processed with the modal identification algorithms introduced in section 5.4.
Then, an automatic interpretation of the produced stabilization diagrams was performed by the
algorithm based on the hierarchical clustering presented in section 5.7.1. The same datasets used to
plot the ANPSD in Figure 7.32 are used to illustrate this preliminary analysis. The methodology is
similar to the one described in section 5.9 for the case study related to the Vestas V90.
281
Figure 7.33 shows the stabilization diagrams associated with the selected datasets provided by the SSI -
COV algorithm, including an averaged spectrum at the background. As expected, the spectrum and
the stabilization diagram associated with the parked condition are clearer. Still, in the other
stabilization diagram several vertical alignments of stable poles can be easily identified.
Poles Stab. freq Stab. damp Stab. MAC ANPSD

70 70
60 60
.
50 50
Model Order
Model Order
40 40
30 30
20 20
10 10
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 7.33 – Stabilization diagrams obtained with the SSI-COV algorithm for operating and non-operating conditions
The automatic processing of the stabilization diagrams generated by the identification algorithms (the
three tested methods produce similar stabilization diagrams) is then obtained with the cluster
algorithm. Figure 7.34 characterizes the obtained clusters by the average frequency of the poles
included in each cluster and by the number of poles. At this preliminary stage, it was decided to
exclude just the clusters with a very low number of poles (less than 6, as marked by the red dashed
line).
30 25
25
20
20
15
Nº of poles
Nº of poles
15
10
10
5
5
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 7.34 – Characterization of the clusters: average frequency and number of poles included in each cluster
For each considered cluster, a representative value of frequency, damping and mode shape was
retained. Each cluster was then classified as FA or SS, according to the direction of its representative
mode shape. After this preliminary analysis, it is possible to plot all the considered clusters in a
Campbell diagram, considering the main direction of vibration (Figure 7.35). In this figure, several
282
Chapter 7
alignments of clusters are clearly identified, corresponding to vibration modes. However, it also shows
some clusters located around the diagonal dashed lines (corresponding to the harmonics associated
with the rotor rotation). These clusters, are only present when the turbine is in operation (rotor speed
higher than 8.7 rpm), corresponding to poles motivated by the harmonic excitation.
FA
SS
4.5
4 FA
4 3 SS*
3.5 3 FA/ 3 SS
3 2 SS
Frequency [Hz]
2 FA
2.5
2 2 SS*
1.5
1 1 SS*
0.5 1 SS
1 FA
0
0 2 4 6 8 10 12 14 16 18
Rotor speed [rpm]
Figure 7.35 – Campbell diagram of considered clusters of stable poles. The vertical dashed lines separate the non-
production situation, a transition state and the operating regimes of the wind turbine. The diagonal dashed lines
represent the harmonic frequencies associated with the rotor rotation (multiples of 3Ω)
From this figure, some initial considerations should be highlighted. The two well defined vertical
alignments around 0.35 Hz and 2.80 Hz in Figure 7.31 are clearly visible in the Campbell diagram
along the whole rotor speed regimes. In fact, they both correspond to two pairs of closely spaced
vibration modes in the FA and SS directions, as initially indicated by the numerical models. For the
pair of cluster around 2.80 Hz it is interesting to note that for high values of rotor speed (higher than
15 rpm), the number of SS clusters of this pair tend to decrease. On the other hand, FA poles appear in
the same alignment of the SS poles. This phenomenon is clearly visible in Figure 7.36, where the
clusters are compared with the wind speed. Apparently, after a certain point of operation (Figure 7.37),
the mode vibrating in the SS direction tends to change its main direction of vibration to FA (although
not perfectly aligned to the FA direction as its FA clusters pair). Looking at Figures 7.13 and 7.14, it
can be attested that this point of operation coincides with the start of the blades pitch actuation.
283
2.95 2.95
FA FA
SS SS
2.9 2.9
Frequency [Hz]
Frequency [Hz]
2.85 2.85
2.8 2.8
2.75 2.75
2.7 2.7
3 5 7 9 11 13 15 17 19 21 23 25 9 10 11 12 13 14 15 16 17
Wind speed [m/s] Rotor speed [rpm]
Figure 7.36 – Zoom of the clusters around 2.70 Hz and 2.95 Figure 7.37 – Zoom of Campbell diagram around 2.70 Hz
Hz vs. wind speed and 2.95 Hz
In addition, the other two referred alignments (1.30 Hz and 1.80 Hz) are also distinguished as SS
clusters in all operating conditions. It is interesting to note that the frequency value of these two modes
seems to decrease and increase, respectively, with the increase of the rotor speed (Figure 7.38). As
referred in section 4.1.2, this is an expected behaviour of rotor whirling modes when considering a
fixed reference (as is the case of the implemented monitoring system with sensors at the tower).
Nonetheless, this figure also evidences that, with the start of the blades pitch actuation, both modes
tend to decrease their natural frequency, changing their behaviour. It is also interesting to note that
these two modes do not seem to be affected by the rotor harmonics, even when their frequency value is
close (or coincident) with the one from the harmonic.
1.9 1.1
FA FA
1.8 SS 1.05 SS
1
1.7
Frequency [Hz]
Frequency [Hz]
0.95
1.6
0.9
1.5
0.85
1.4
0.8
1.3 0.75
1.2 0.7
9 10 11 12 13 14 15 16 17 0 2 4 6 8 10 12 14 16 17
Rotor speed [rpm] RPM
Figure 7.38 – Zoom of Campbell diagram around 1.20 Hz Figure 7.39 – Zoom of Campbell diagram around 0.70 Hz
and 1.90 Hz and 1.10 Hz
The alignment of clusters around 3.70 Hz is composed by FA and SS clusters (although it is not clearly
visible in Figure 7.35). A detailed analysis of this alignment shows that it is referred to only one mode,
since the FA and SS cluster do not appear at the same time. This phenomenon is analysed in more
detail in the next section.
The analysis of Figure 7.35 is also important to assess the influence of the rotor excitation on the
detection of the vibration modes throughout the operating regimes. It is observed that some stable
284
Chapter 7
horizontal alignments of cluster are only visible until the start of the turbine operation. The alignment
of FA clusters around 0.80 Hz is an example of this phenomenon (Figure 7.39). This alignment may be
related to vibration modes whose influence on the tower motion is too low to be noticed when the
rotor starts spinning. Another example of the harmonic effect is illustrated with the horizontal
alignment of FA clusters around 2.30 Hz. From the rotor speed of 8.7 rpm, it is noticed that this
alignment tend to fluctuate according to the harmonics that cross it.
Considering this initial analysis, 9 modes were decided to be followed in the scope of the dynamic
monitoring system. This decision was made based on the importance of the mode to the wind turbine
motion and on influence of the harmonics over the mode.
A nomenclature is also given for the monitored modes. It is referred to the number of the order of the
mode and to the main direction of vibration. In addition, when the evolution of the identified
frequency value of a mode is clearly dependent on the rotor speed, an “*” is added to the name since it
may be referred to a rotor mode. The vibration mode with frequency value around 3.70 Hz is labelled
as 3 FA/ 3 SS, since, as referred, it seems to vibrate either way.
Considering the results obtained with this preliminary analysis, 6 operating regimes were adopted to
define the reference properties of the monitored vibration modes. For each regime, reference values of
frequency value and mode shape of the 9 considered modes were kept. These regimes are characterized
in Table 7.4.
Table 7.4 – Regimes considered for reference modal properties of the vibration modes
Operating regime Wind turbine condition
1 Parked or idling (with high pitch angle values higher than 72º)
2 Parked or idling (with pitch angle value lower than 72º)

Transition regime from non-operation to operation (rotor speed
3
between 0 and 8.7 rpm)
Operating regime, defined by the lowest operating rotor speed and the
4 point where the pitch angle starts to increase to avoid excessive rotor
torque values (around mean wind speed of 10 m/s)
Operating regime, between regime 4 and the highest operating rotor
5
speed
6 Wind speed higher than cut-out speed
Regime 1 is referred to setups when the pitch angle of the blades is at its maximum level. As
introduced in Figure 7.14, under certain non-operating conditions, the pitch angle of the blades is set
to values around 90º. Regime 2 is also referred to non-operating situations, but the pitch angle is
defined to values of around 70º.
The region defined as a transition state in Figure 7.35, related to setups where the turbine starts
operating, is named as Regime 3. Since the SCADA system records 10 min. mean values, the rotor
speed values of this regime are between 0 and the lowest operating rotor speed (8.7 rpm).
Regime 4 is regarded to setups in which the rotor speed is higher than 8.7 rpm and the pitch angle is
kept at its lower level, i.e., before the pitch actuator is actively varying the angle of attack of the blades.
285
Once the pitch angle starts increasing in order to avoid excessive rotor speed values, it is considered as
Regime 5. This Regime extends to the limit operating conditions of the turbine.
Lastly, Regime 6 is referred to situations when the wind speed is higher than the turbine cut-out wind
speed. However, this regime was not identified during the monitoring period.
With the aim of illustrating the 9 considered vibration modes, the reference mode shapes from Regime
4 are presented Figure 7.40. At the bottom of each mode shape, a polar plot shows the magnitude and
phase angle of the modal amplitude of the sensors orientated to the main direction of vibration. It can
be seen that some are not perfect real modes, since some amplitudes present phase angles slightly
different from 0º or 180º.
From the illustrations of Figure 7.40, it is possible to conclude that the two pairs of closely-spaced
modes around 0.35 Hz (modes 1 SS and 1 FA) and 2.80 Hz (2 FA and 2 SS) are, respectively, the first
and second pairs of tower bending modes, which is in accordance with the results of the numerical
models described in section 7.3.2 (Oliveira, Magalhães et al., 2013).
The remaining modes are certainly related to motion from other structural elements besides the tower,
such as the rotor blades. This fact is in line with the behaviour observed in the Campbell diagram for
the modes 1 SS*, 2 SS* and 3 SS*.
Mode 1 SS Mode 1 FA Mode 1 SS*

80 80 80
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
90
90 120 90
120 60 1 60 120
1 60
150 30
150 30 150 30
180 90 180 90 180 90
330 210 330

210 210 330
240 300 240 300 240 300

270 270 270
286
Chapter 7
Mode 2 SS* Mode 2 FA Mode 2 SS

80 80 80
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
90 90 90
120 60 120 120
1 1 60 1 60
150 30 150 30 150 30
180 90 180 180 90
210 330 210 330 210 330
240 300 240 300 240 300

270 270 270
Mode 3 FA/ 3 SS Mode 3 SS* Mode 4 FA

80 80 80
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
-1.2 0 1.2 -1.2 0 1.2 -1.2 0 1.2
90 90 90
120 120 60
1 60 1 120 1 60
150 30 150 30 150 30
180 90 180 90 180 90
210 330 210 330 210 330
240 300 240 300 240 300

270 270 270
Figure 7.40 – Tower mode shapes of the tracked vibration modes
287
7.3.6.2 Automated Identification of the Modal Parameters

Three output-only modal identification algorithms were used to perform an automated analysis of the
modal parameters of the wind turbine: SSI-COV, SSI-DATA and p-LSCF. Once the reference modal
properties of the vibration modes intended to be tracked were defined, it was possible to perform the
identification in a completely automated way.
The strategy introduced in previous sections was followed. For each identification algorithm, a pole is
classified as stable in a stabilization diagram if it respects the limits presented in Table 7.5 for
variations between models of consecutive orders. Once the stable poles are identified, the hierarchical
cluster algorithm is applied. For this process, a maximum distance of 0.02 was defined for a single
linkage criterion.
Table 7.5 – Stable pole criteria for models of consecutive orders
Modal parameter Maximum allowed variation
Frequency Δ ≤1%
Damping Δ ≤5%
Mode Shape MAC ≥ 0.97
From the obtained clusters, only the ones with a number of poles higher than 6 were considered. Then,
for each reference mode, the clusters presenting an average natural frequency that did not differ more
than a predefined percentage value (10 % to 20 % depending on the mode type) from the reference
natural frequency value were selected. From those, it is selected the one that presents the average mode
shape with the highest correlation with the reference mode shape (evaluated with the MAC
coefficient). Modes with MACs lower than 0.80 are not considered.
The stated parameters were tried on several initial datasets and proved to deliver good and coherent
results. Figure 7.41 shows the Campbell diagram after the tracking (comparison of the cluster
properties with the reference properties of the vibration modes intended to be monitored). It can be
seen that, in this particular application, this simple tracking procedure was adequate to eliminate the
influence of the harmonics in the modes under analysis. For that reason, it was not necessary to use the
methods presented in section 5.6.3.
288
Chapter 7
Campbel Diagram
4.5
4 FA
4 3 SS*
3.5 3 FA/SS
3 2 SS
Frequency [Hz] 2.5 2 FA
2 2 SS*
1.5
1 1 SS*
0.5 1 SS
1 FA
0
0 2 4 6 8 10 12 14 16 18
RPM
Figure 7.41 – Campbell diagram with the tracked vibration modes (p-LSCF algorithm)
Naturally, it was also necessary to define the input parameters for each modal identification algorithm.
Initial datasets were used to tune these parameters.
The SSI-COV algorithm requires two input parameters: the number of points of the correlation
function and the maximum order of the model. It was decided to adopt correlation functions of 128
points and a maximum order of 70.
Table 7.6 presents the main statistics related to the performance of this algorithm in the automated
identification of the 9 considered vibration modes. A success rate above 60 % is consistently obtained,
except for the 3 FA/3 SS, 3 SS* and 4 FA modes. Also, the 1 FA mode shows a considerably lower
success rate than its pair. This fact is probably due to the high values of damping of this mode which
hinders the identification of this mode by the algorithms. For the 3 FA/3 SS mode, the low success rate
is explained by the fact that this mode presents an unusual behaviour, which tends to change its main
direction of vibration according to the nacelle orientation. Thus, since a minimum MAC value for the
modal tracking was defined, only a small portion of the clusters identified in the frequency range of
this mode were considered. On the other hand, the low value obtained for the identification of the 3
SS* and 4 FA modes are mainly related to non-operating conditions (Regimes 1 and 2) and to the
Regime 5. In these three regimes, the success rate was considerably low.
From the analysis of Table 7.6, it is interesting to check that within the first pair of tower bending
modes, the 1 SS mode presents a much smaller dispersion of its natural frequency than the 1 FA mode.
Again, the high values of damping of the 1 FA mode may have influenced the quality of the detection
of this mode.
289
Table 7.6 – Statistics related to the results obtained with the SSI-COV algorithm
Success Rate
Mode [Hz] [Hz] . .
[%]
1 SS 81.2 0.354 0.002 0.005
1 FA 67.4 0.355 0.007 0.020
1 SS* 63.4 1.325 0.051 0.039
2 SS* 78.3 1.791 0.060 0.034
2 FA 75.8 2.792 0.024 0.009
2 SS 83.7 2.828 0.030 0.011
3 FA/ 3SS 9.1 3.673 0.042 0.011
3 SS* 29.5 3.782 0.043 0.011
4 FA 38.3 4.236 0.072 0.017
The second time domain identification algorithm used was the SSI-DATA. For this algorithm, two
parameters were defined: the number of blocks of the Hankel matrix and the maximum order of the
model. After initial tuning, it was concluded that = 70 blocks, together with a maximum model order
of 70 led to the best results.
The main statistics about the automated modal identification with the SSI-DATA are introduced in
Table 7.7. The results are in line with the ones obtained with the SSI-COV. On the identification of the
4 FA mode, a considerably increase was obtained.
Table 7.7 – Statistics related to the results obtained with the SSI-Data algorithm
Success Rate
Mode [Hz] [Hz] . .
[%]
1 SS 84.7 0.355 0.002 0.006
1 FA 67.4 0.358 0.008 0.022
1 SS* 84.9 1.330 0.050 0.038
2 SS* 89.8 1.793 0.057 0.032
2 FA 74.0 2.789 0.024 0.009
2 SS 60.2 2.829 0.030 0.011
3 FA/ 3SS 14.9 3.690 0.039 0.011
3 SS* 44.9 3.789 0.047 0.012
4 FA 82.3 4.266 0.079 0.019
The last identification algorithm used in the dynamic monitoring system was the p-LSCF. Positive
time lags of the correlation with 1024 points were used to calculate the spectra. An exponential
window with a factor of 0.1 was also used. A maximum model order of 40 was considered. The
obtained results show that this algorithm provided much higher success rates for the modes not so
successfully identified with the other two algorithms.
290
Chapter 7
Table 7.8 – Statistics related to the results obtained with the p-LSCF algorithm
Success Rate
Mode [Hz] [Hz] . .
[%]
1 SS 80.8 0.354 0.002 0.006
1 FA 64.7 0.354 0.007 0.019
1 SS* 87.4 1.330 0.049 0.037
2 SS* 85.5 1.793 0.058 0.032
2 FA 80.1 2.796 0.017 0.006
2 SS 85.5 2.831 0.026 0.009
3 FA/ 3SS 43.2 3.700 0.030 0.008
3 SS* 67.8 3.809 0.058 0.015
4 FA 95.1 4.276 0.081 0.019
Once the modal parameters are identified, it is possible to analyse the variation of the parameters with
the operational and environmental factors. Figure 7.42 characterizes the evolution of the natural
frequencies of the 9 monitored vibration modes along the monitoring period. The variability of the
frequency values is evident.
4.5
3.5
3
Frequency [Hz]
2.5
1.5
0.5
Time
Figure 7.42 – Variation of the of the monitored natural frequencies
Figure 7.43 a) shows a zoom of the frequency variation of 1 FA and 1 SS vibration modes along the
recorded setups. It is visible that the variability of the 1 FA mode is much higher than of the 1 SS,
which is in agreement with the results presented in Tables 7.6 to 7.8. Figure 7.43 b) evidences the
variation of the natural frequencies of the 1 SS* and 2 SS* modes, mainly due to their dependence on
the rotor speed.
291
0.4
1.8
0.38
1.7
Frequency [Hz]
Frequency [Hz]
1.6
0.36
1.5
0.34
1.4
1.3
0.32
1.2
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Index 4 Index 4
x 10 x 10
a) b)
Figure 7.43 – Zoom of the natural frequencies along the recorded setups: a) 1 FA and 1 SS; b) 1 SS* and 2 SS*
In order to better understand the behaviour of the modal parameters of the 9 tracked modes, Figure
7.44 shows the variation of the natural frequencies for each operating regime. This graphic evidences
the suitability of defining different Regimes for modal tracking. For example, the variation of the
frequency values of the modes 1 SS* and 2 SS* is notorious. Apart from these modes, also the 2nd pair
of tower bending modes shows a different behaviour when the turbine starts operating, with a better
separation of the two modes.
4.5
3.5
3
Frequency [Hz]
2.5
1.5
0.5
0
Regime Regime Regime Regime Regime
1 2 3 4 5
Figure 7.44 – Variation of the natural frequency, for each operating regime, of the monitored vibration modes
Figure 7.45 a) presents a closer look of the first pair of tower bending modes. It is interesting to note
that the variability of the 1 FA mode is smaller than the 1 SS mode for non-operating conditions. Once
the turbine starts operating, the dispersion of the frequency values from this mode greatly increases.
On the other hand, the variability of the 1 SS mode is kept almost constant throughout the different
operating regimes. These facts reinforce the idea that the high values of damping of the 1 FA mode,
during operating conditions, may have influenced the accuracy of the detection of this mode. Figure
7.45 b) evidences the better separation of the 2nd pair of tower bending modes under operating
conditions.
292
Chapter 7
0.4 2.95
2.9
0.38
2.85
Frequency [Hz]
Frequency [Hz]
0.36
2.8
0.34 2.75
2.7
0.32
2.65
Reg. Reg. Reg. Reg. Reg. Reg. Reg. Reg. Reg. Reg.
1 2 3 4 5 1 2 3 4 5
a) b)
Figure 7.45 – Variation of the natural frequency, for each operating regime, of: a) 1 FA and 1 SS; 2 FA and 2 SS
The results obtained with the identification of the natural frequencies and damping ratios of 9
monitored vibration modes with the p-LSCF algorithm are shown in Tables 7.9 and 7.10 for each
regime.
Table 7.9 – Results obtained for each operating regime: natural frequencies
1 2 3 4 5
Mode
[Hz] [Hz] [Hz] [Hz] [Hz] [Hz] [Hz] [Hz] [Hz] [Hz]
1 SS 0.353 0.003 0.353 0.003 0.354 0.002 0.355 0.001 0.355 0.002
1 FA 0.352 0.001 0.352 0.001 0.355 0.004 0.357 0.006 0.349 0.006
1 SS* 1.258 0.011 1.277 0.011 1.400 0.020 1.362 0.033 1.294 0.018
2 SS* 1.625 0.011 1.812 0.017 1.729 0.025 1.793 0.045 1.844 0.013
2 FA 2.767 0.012 2.784 0.013 2.781 0.022 2.799 0.015 2.796 0.014
2 SS 2.792 0.020 2.790 0.021 2.819 0.022 2.842 0.016 2.841 0.012
3 FA/
- - 3.659 0.042 3.664 0.038 3.702 0.028 3.702 0.026
3SS
3 SS* 3.668 0.050 3.683 0.055 3.748 0.053 3.812 0.036 3.870 0.018
4 FA 4.107 0.033 4.167 0.046 4.253 0.068 4.301 0.040 4.349 0.048
293
Table 7.10 – Results obtained for each operating regime: damping ratios
1 2 3 4 5
ߦ௠௘௔௡ ߦ௦௧ௗ ߦ௠௘௔௡ ߦ௦௧ௗ ߦ௠௘௔௡ ߦ௦௧ௗ ߦ௠௘௔௡ ߦ௦௧ௗ ߦ௠௘௔௡ ߦ௦௧ௗ
Mode
[%] [Hz] [%] [Hz] [%] [Hz] [%] [Hz] [%] [Hz]
1 SS 0.974 0.631 0.733 0.438 0.707 0.590 0.630 0.394 1.096 0.548
1 FA 0.351 0.482 0.213 0.356 1.490 1.128 3.659 1.390 7.868 2.169
1 SS* 0.502 0.207 0.646 0.316 0.845 0.232 0.932 0.236 0.860 0.152
2 SS* 0.543 0.261 0.746 0.297 0.746 0.271 0.916 0.247 1.113 0.211
2 FA 0.175 0.100 0.195 0.120 0.624 0.231 0.884 0.320 1.291 0.284
2 SS 0.301 0.127 0.254 0.167 0.344 0.162 0.451 0.225 0.662 0.181
3 FA/
- - 0.778 0.275 1.013 0.350 1.395 0.427 1.820 0.407
3SS
3 SS* 0.572 0.230 0.551 0.228 0.660 0.224 0.719 0.149 0.888 0.232
4 FA 0.476 0.170 0.457 0.178 1.079 0.433 1.598 0.374 1.566 0.467
The evolution of the 1 SS* and 2 SS* modes with the rotor speed is illustrated in Figure 7.46. This
figure resembles the typical behaviour the of rotor whirling modes. It is visible that, when the turbine
is consistently rotating (for rotor speed higher than 8.7 rpm), the 1 SS* and 2 SS* modes have a similar
behaviour with, respectively, a backward and a forward whirling mode. Considering that, when in
operation, these modes are detected by tower motion in the SS direction, i. e., in the direction of the
rotor plane, these two modes are most likely related to edgewise vibration modes. From the three
configurations of first order edgewise modes shown in Figure 4.4, only asymmetrical modes create a
reaction force at the rotor and, thus, are capable of inducing a detectable tower motion. For these
reasons, it expected that 1 SS* and 2 SS* modes are in fact the first backward and forward edgewise
modes, respectively.
A regression model for each vibration mode is also presented in Figure 7.46 in dashed lines. These
regression models were adjusted to the frequency values obtained for the two modes during events
regarding Regime 4. Regime 5 was not considered for the definition of the models because this regime
implies the variation of the blades pitch angle which introduces changes in the behaviour. The
intercept of the regression models with the origin (rpm=0) gives an approximation of the frequency
values of these two modes under parked conditions and with a pitch angle around 0º. Values of 1.526
Hz and 1.558 Hz were obtained for, respectively, the backward and forward mode. However, this
configuration does not occur in the monitored wind turbine. When under parked conditions, the
blades are usually oriented with a pitch angle between 72º and 90º, changing the direction of vibration
of the edgewise modes from in-plane rotor motion (when in operation) to out-of-plane (when in non-
operation conditions). Consequently, the support conditions of the blades are also changed with this
transition. Due to these reasons, modes vibrating at the referred frequencies were not detected. It was
then investigated if, during events from Regime 1 and 2, there were modes with natural frequencies
next to these frequencies and with a tower mode shape similar to the ones from the 1 SS* and 2 SS*
modes but vibrating in the FA direction (as referred, during non-operating conditions, the edgewise
modes vibrate out of the plane rotor). The cluster referred to the found vibration modes during
Regime 1 and 2 are shown in Figure 7.46 in lighter colours. It is seen that, while for the 1 SS* mode, the
identified modes have natural frequencies close to each other during Regime 1 and 2 (݂ ≈ 1.26 Hz), the
same does not happen for the 2 SS* (݂ = 1.63 Hz and ݂ = 1.81 Hz for, respectively, Regime 1 and 2). In
294
Chapter 7
fact, the identified clusters for the 2 SS* mode present a considerably change in the natural frequency
between Regime 1 and 2. Unfortunately, this situation could only be deeper investigated if
instrumentation at the blades would be available. Even though, the referred clusters for Regime 1 and 2
were considered for monitoring purposes.
2
2 SS*
Frequency [Hz] 1.8
1.6
1.4
1.2 1 SS*
1
0 2 4 6 8 10 12 14 16
Rotor speed [rpm]
Figure 7.46 – Evolution of the frequency values of the 1 SS* and 2 SS* with the rotor speed
Likewise, although on a smaller scale, the natural frequencies of the second pair of tower bending
modes (2 FA and 2 SS) also shows an increasing trend with the operating regimes. On the other hand,
the 1 FA and 2 SS modes do not have a very distinct trend over the regimes, as shown in Figure 7.47
0.39 0.37
0.38
0.365
0.37
0.36
Frequency [Hz]
Frequency [Hz]
0.36
0.35 0.355
0.34
0.35
0.33
0.345
0.32
0.31 0.34
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Rotor speed [RPM] Rotor speed [rpm]
0.39 0.37
0.38
0.365
0.37
0.36
Frequency [Hz]
Frequency [Hz]
0.36
0.35 0.355
0.34
0.35
0.33
0.345
0.32
0.31 0.34
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 4 5 6 7 8 9 11 13 15 17 19 21
a) b)
Figure 7.47 – Median (and box plots) of the frequency values: a) 1 FA mode; b) 1 SS mode
295
Globally, the damping ratio of the FA modes increases with the increase of the wind speed (which
usually implies a higher rotor speed). This is mostly due to the contribution of the aerodynamic
component of damping. As expected, this increase of the damping values is not notorious in the SS
modes due to the lower opposition to the wind flow in this direction. It is however noticed that the
damping of SS modes increases in Regime 5. This is probably due to the aerodynamic change
introduced by the increase of the blades pitch angle.
The analysis of the evolution of the damping of the two pairs of bending modes illustrates the complex
dynamics of a wind turbine. For non-operating conditions (Regimes 1 and 2), the damping values of
the SS modes are higher than for the FA modes. This situation is due to the deviation of the rotor from
the main wind direction and to the high pitch value of the blades. Under these conditions, the blades
present a larger opposition to the wind flow in the SS direction, leading to the appearance of
aerodynamic damping in this direction.
Once the turbine starts operating, the rotor is orientated to the main wind direction and the blades
pitch angle is set to its minimum value. These conditions, together with the higher wind speed, lead to
the increase of the aerodynamic damping in the FA modes and the almost vanishing of this
contribution in the SS modes. Since the aerodynamic damping component increases with the increase
of the wind speed, the damping values of the 1 FA and 2 FA modes are consistently growing over the
operating regimes. This effect is more noticeable for modes with large modal amplitude at the tower
top, where the wind force is higher. For this reason, the 1 FA mode presents higher values than the 2
FA (whose modal amplitude at the top is very small).
The top plots of Figure 7.48 illustrate the evolution of the damping ratio of the 1 FA and 1 SS vibration
modes with the rotor speed. The increase of the damping values of the 1 FA mode when the turbine
starts operating is evident. It is also noted that the damping increases when reaching the final part of
the rotor speed range. Notwithstanding, these figures only shows the evolution of damping until the
wind rated speed is achieved (around 13 m/s). For that reason, the lower graphic of Figure 7.48
presents the evolution of the damping of the 1 FA mode with the wind speed. The shape of the
adjusted line clearly shows a drop in the damping value around 15 – 17 m/s. This situation is
consequence of the increase of the pitch angle in Regime 5 which reduces the thrust force at the tower
top and, consequently, reduces the rotor blades opposition to the wind flow in the FA direction.
On the other hand, the adjusted line for the damping values of the 1 SS mode only shows small
variations over the different operating conditions, with a reduction when the turbine starts operating
and an increasing when reaching its rated speed.
296
Chapter 7
15 7
12.5 6
5
10
Damping [%]
Damping [%]
4
7.5
3
5
2
2.5 1
0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Rotor speed [m/s] Rotor speed [RPM]
15 7
12.5 6
5
10
Damping [%]
Damping [%]
4
7.5
3
5
2
2.5 1
0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
a) b)
Figure 7.48 – Median (and box plots) of the damping ratios: a) 1 FA mode; b) 1 SS mode
The situation illustrated in Figures 7.36 and 7.37, about the possible change of mode shape orientation
of the 2 SS mode for high wind speed conditions, was further investigated during the modal tracking.
In that sense, an additional reference “mode” named 2 SS(/FA), based on the 2 SS mode but with an
orientation between the FA and the SS direction, was considered. As results, it was possible to confirm
that this “mode” was identified on setups in which the 2 SS was missing. Considering the number of
setups in which the 2 SS(/FA) “mode” was tracked, the success rate of the 2 SS increases from 67.1 % to
85.5 % (with the p-LSCF algorithm).
2.9 2.9
2 FA 2 FA
2 SS 2 SS
2 SS(/FA)
2.85 2.85
Frequency [Hz]
Frequency [Hz]
2.8 2.8
2.75 2.75
5 10 15 20 25 5 10 15 20 25
a) b)
Figure 7.49 – a) Evolution of the 2 FA and 2 SS modes with the wind speed. b) Evolution of the 2 FA, 2 SS and the 2 SS/(FA)
297
7.3.6.3 Damping Estimation from Free Decay Responses

The previous section showed the evolution and characterization of the modal properties of the wind
turbine main vibration modes under different operating conditions. One of the most important
features obtained with this analysis is the characterization of the modal damping, mainly the damping
of the first tower bending (1 FA) mode. This is a crucial property with strong influence on the fatigue
lifetime of the support structure of a wind turbine.
During the monitored period, several recorded events related to shutdown situations were detected.
These are events in which the wind turbine is initially operating and, at a certain point, the wind
turbine cease power production, stopping the rotor. The reasons behind these events are usually
associated to situations when there is a wind speed reduction to values below the cut-in or to very high
wind speed (above the cut-out speed) and it is accomplished through the increase of the blades pitch
angle to values close to 90º.
Under these circumstances, when the blades pitch angle is changed, an instantaneous impulse acts at
the top of the tower, imposing a free decay response of the structure. This decay can then be used to
accurately estimate the damping of the vibration modes. This methodology usually provides more
accurate estimates of the damping than with the ambient vibration response [(Magalhães, Cunha et al.,
2010)].
Taking advantage of the information provided by the SCADA system, the acceleration events with a
maximum and minimum rotor speed values, respectively, above and below the minimum rotor speed
were initially selected. From these, the ones regarded to start up events were dismissed. The remaining
ones were then visually inspected to assess the quality of the free decay response. Only the acceleration
time series containing a clean exponential decay shape of the acceleration response were kept. Figure
7.50 a) illustrates an example of a considered free decay response, where the referred exponential decay
shape can be observed. On the other hand, Figure 7.50 b) shows an example of a disregarded shutdown
event. As can be seen, the decay response does not resemble the referred decay shape. This fact may be
due to the lack of a sudden absence of the thrust force at the top of the tower.
0.2 0.3
+74.988 FA +74.988 FA
0.15
+48.392 FA 0.2 +48.392 FA
0.1 +21.772 FA +21.772 FA
0.1
0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05
-0.1
-0.1
-0.2
-0.15
-0.2 -0.3
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time [s] Time [s]
a) b)
Figure 7.50 – Events in which a shutdown was detected: a) structural response following an exponential decay; b)
example of structural response not following the common exponential decay
298
Chapter 7
From the initially pre-selected events, 51 were chosen to be analysed. The majority of the events are
referred to situations in which the wind speed dropped to values below the cut-in. In addition, 7 events
were detected whose measured maximum wind speed was above the rated speed and the minimum
wind speed as also above the cut-in value. However, in none of these events, the maximum wind speed
is above the cut-out value. Figure 7.51 illustrates the minimum and maximum wind speed measured
by the SCADA system for the selected events.
25 Cut-out
Minimum
Maximum
20
Wind speed [m/s]
15
Rated
10
5
Cut-in
0
0 5 10 15 20 25 30 35 40 45 50
Index
Figure 7.51 – Measured maximum and minimum wind speed occurred in the selected 10 min. events containing a free
decay response
Figure 7.52 illustrates the diversity of the acceleration levels observed within the selected free decay
responses. Figure 7.52 a) shows the decay response from an event whose maximum and minimum
wind speed was, respectively, 19.4 m/s and 10.1 m/s. On the other hand, Figure 7.52 b) presents the
acceleration time series of a decay response from a period in which the wind speed varied between 4.2
m/s and 1.0 m/s, which means that the turbine stopped because the wind speed was too low for power
production. Comparing both figures, the difference in the level of acceleration is notorious.
1.5 0.08
+74.988 FA +74.988 FA
+48.392 FA
0.06
1 +48.392 FA
+21.772 FA +21.772 FA
0.04
0.5
0.02
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.02
-0.5
-0.04
-1
-0.06
-1.5 -0.08
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time [s] Time [s]
a) b)
Figure 7.52 – Free decay response from: a) a high wind speed event; b) a low wind speed event
During the free decay events, the motion of the tower is mainly in the FA direction. Figure 7.53 shows
the evolution of the FA and SS acceleration at the tower top of the decay responses represented in
Figure 7.52 for the time segments enclosed by the vertical dashed lines. In this figure, corresponding to
299
a top view of the tower motion, the hotter colours correspond to the beginning of the time segment
while the cold colours correspond to the final instants of the analysed period.
+74.988 m +74.988 m
1.5 0.05
1
SS acceleration [m/s2]
SS acceleration [m/s2]
0.025
0.5
0 0
-0.5
-0.025
-1
-1.5 -0.05
-1.5 -1 -0.5 0 0.5 1 1.5 -0.05 -0.025 0 0.025 0.05
FA acceleration [m/s2] FA acceleration [m/s2]
a) b)
Figure 7.53 – Evolution of the FA and SS acceleration (top view) from: a) a high wind speed event; b) a low wind speed
event (the hotter colors correspond to the beginning of the time segment while the cold colors are referred to the final
instants)
Initially, the two events introduced in Figure 7.52 were analysed by fitting an exponential function to
the free decay envelope, in the same manner as in section 5.8.4.2 for the rotor-stop tests performed for
the Izar Bonus 1.3MW/62 wind turbine. A low-pass filter with a frequency cut of 0.5 Hz was applied to
both signals. The exponential function was then fitted to the local maxima points of the filtered signals,
as illustrated in Figure 7.54. The results obtained for the frequency value and damping ratio of the 1
FA mode are presented in Table 7.11.
+74.988 m +74.988 m
1.5 0.08
1.203*e-0.0058*2*π*0.349*t 0.047*e-0.0060*2*π*0.352*t
0.06
1
0.04
0.5
0.02
Acc. [m/s 2 ]
Acc. [m/s 2 ]
0 0
-0.02
-0.5
-0.04
-1
-0.06
-1.5 -0.08
150 200 250 300 350 400 200 250 300 350 400
Time [s] Time [s]
a) b)
Figure 7.54 – Identification of the modal damping ratio of the 1 FA vibration mode through filtering and fitting of an
exponential function to the envelope of the free decay response from: a) a high wind speed event; b) a low wind speed
event
The two free decay responses were also analysed through the application of the SSI-COV algorithm
adapted to consider the measured free decays as input. This method was also used in section 5.8.4.2 to
study the structural response of the Izar Bonus 1.3MW/62 wind turbine to the rotor-stop test. The
300
Chapter 7
time segments defined to fit the exponential function in Figure 7.54 were used as inputs of the
algorithm. The obtained stabilization diagrams are shown in Figure 7.55. Besides the one referred to
the 1 FA mode, other vertical stable alignments are also evident. The results obtained for the 1 FA
mode are summarized in Table 7.11.
Stabilization diagram Stabilization diagram
20 20
15 15
Model Order
Model Order
10 10
5 5
0 0
0 1 2 3 4 5 0 1 2 3 4 5
a) b)
Figure 7.55 – Stabilization diagram obtained with the application of the SSI-COV algorithm
The results obtained with the two procedures presented very similar values for frequency and damping
ratio. It is also visible that there is no apparent influence of the wind speed on the damping values of
the mode. It should be noted that in both free decays, due to change of orientation of the blades (from
low angles of pitch to values to close to 90º), the contribution of the aerodynamic damping for the FA
mode is not relevant.
Table 7.11 – Natural frequency ( ) and modal damping ratio ( ) of the 1 FA vibration mode identified with the filtering
and fitting procedure and the SSI-COV method
Exponential window SSI-COV
[Hz] [%] [Hz] [%]
high wind speed event 0.349 0.58 0.349 0.59

low wind speed event 0.351 0.60 0.352 0.62
The described analysis was then extended to all the selected free decay events. For the procedure based
on filtering and fitting, a routine was developed to find the best exponential function to fit the decay
response of the wind turbine. The results obtained were then visually analysed in order to validate the
results. The obtained results are presented in Figure 7.56.
The segment of the acceleration time series used for the filtering and fitting procedure was also
considered as input for the SSI-COV algorithm. In order to automate the procedure based on this
method, a cluster analysis was performed after each run of the algorithm. With this processing
solution, the procedure becomes similar to the one described for the continuous modal identification
in section 7.3.6.1. The results obtained are also shown in Figure 7.56.
301
1st FA mode
0.356
filt./ fitting
0.355
SSI-COV
0.354
Frequency [Hz] 0.353
0.352
0.351
0.35
0.349
0.348
0 5 10 15 20 25 30 35 40 45 50
Index
1st FA mode
1
filt./ fitting
0.9 SSI-COV
0.8
Damping ratio [%]
0.7
0.6
0.5
0.4
0.3
0 5 10 15 20 25 30 35 40 45 50
Index
Figure 7.56 – Results obtained for the 1st FA mode with the filtering and fitting procedure and SSI-COV algorithm for
frequency value (top plot) and damping ratio (bottom plot)
The results obtained show a good agreement between the two methods. Some variability in the results
is noticed though. It was thus decided to assess if the operational and environmental parameters could
explain this variance. Figure 7.57 illustrates the evolution of the modal properties identified with the
SSI-COV methodology with maximum absolute values (MAX) of the selected segments of the
acceleration decay responses. From these figures, it is visible that the majority of the decays is related
to events with values smaller than 0.4 m/s2 which difficulties the identification of a clear trend.
Notwithstanding, there is an apparent decrease of the frequency values of the 1 FA mode with the
acceleration. However, the number of events with MAX values higher than 0.4 is too low to confirm
this trend.
On the other hand, the identified values of modal damping do not present a distinct dependency on
the acceleration level. This conclusion is contrary to what usual occurs in common civil structure,
where the damping tends to increase with the acceleration level.
Theoretically, the study of the free decay events could allow the study of the structural damping of the
wind turbine. However, comparing the damping values identified with the free decay events with the
results from the automated modal tracking, it is visible that they are considerably higher than the ones
from the non-operating Regime 1 (in which the blades orientation is equal to the observed in the free
decay events after the impulse) and lower than the ones observed in Regime 3 (in which the wind
conditions are similar to the free decay events).
302
Chapter 7
1st FA mode 1st FA mode

0.356 1
0.355
0.9
0.354
Frequency [Hz]
0.8
Damping [%]
0.353
0.352 0.7
0.351
0.6
0.35
0.5
0.349
0.348 0.4
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Acc. MAX [m/s ] 2
Acc. MAX [m/s2]
a) b)
Figure 7.57 – Evolution of the frequency value (a)) and damping ratio (b)) of the 1st FA mode with the maximum absolute
values of acceleration
Although the study of the frequency values indicates some possible relations with the vibration level, it
is not sufficient to explain the variation detected along the 51 decay events. This variation is mostly
related to the temperature dependency of this modal property. This relation is shown in Figure 7.59,
where a clear trend is visible. In this figure, some points (enclosed in a red circle) are visible with
frequency values lower than what the trend suggests. These points are referred to events with high
acceleration amplitudes, suggesting that, in fact, the vibration level may interfere in the frequency
values of this mode.
With regard to the damping, it cannot be identified the source of the observed scatter.
1st FA mode
0.356
0.355
0.354
Frequency [Hz]
0.353
0.352
0.351
0.35
Acc. RMS > 0.20
0.349
0.348
5 10 15 20 25 30
Temperature [ºC]
Figure 7.58 – Evolution of the frequency value of the 1st FA mode with temperature
7.3.6.4 Participation of the Vibration Modes/ Harmonics to the Measured Dynamic Response
The methodology presented in section 5.5 to decompose the measured acceleration time series into
several modal and stationary responses (due to harmonics) was implemented in the developed
dynamic monitoring system.
In the application of the modal response estimation, it is required the selection of a representative
model order. The state-space matrices computed for that model order will then be used to define the
303
forward innovation model. While in the example of section 5.5, a manual selection was performed
based on the observation of the stabilization diagram, an automated strategy for choosing the best
model order is required as part of a monitoring system. The implemented strategy is thus based on the
selection of the order with the highest number of stable poles whose clusters were chosen during the
tracking procedure. When the turbine is operating (Regimes 4 and 5), cluster referred to the
harmonics were also considered in the selection of the model order. For a cluster to be considered as
representative of the stationary response due the harmonic excitation, its frequency value has to be
within a range of 10 % of the multiples of the mean rotor speed from the setup under analysis. In
cases in which there is more than one cluster meeting this condition, the cluster containing the higher
number of poles is chosen. If this cluster has more poles than a pre-defined minimum number, it is
defined as representative of the wind turbine response due to the harmonic excitation. In this
application, a minimum number of 6 poles was defined. The 1Ω, 3Ω and 6Ω harmonics were
considered in this analysis, although the first was only identified in 5.6 % of the operating setups.
Figure 7.59 characterizes the RMS values of the modal contributions associated with the modes
vibrating in the FA direction, alongside with the harmonics, at the +74.988 m level. It is visible that the
dynamic motion is dominated mainly by the 3Ω harmonic and the 1 FA mode. As expected, the
acceleration response in 1Ω harmonic frequency represents a very low level. It is also interesting to
note that the 1 FA mode (and also the 6Ω harmonic) present a different behaviour from the other
responses, do not continuously increasing with the increase of the rotor speed (or wind speed). It is
observable that, when the turbine starts operating, there is a steep increase on the values from the 1 FA
mode, which tend to disappear when the turbine reaches the 10 rpm. This increase in the modal
acceleration response of the 1 FA is probably due to the excitation of this mode by the 3Ω harmonic.
Considering Figure 7.35, it is possible to attest that the 3Ω harmonic is close to the 1 FA mode at rotor
speeds between 8.7 and 10 rpm which can lead to the excitation of this mode and, consequently, to the
increase of the vibration amplitudes. A similar situation seems to occur with 6Ω harmonic response.
For wind speed around 3 m/s (corresponding to the start of the turbine), an increase of the
acceleration levels is also visible in Figure 7.59 b). This situation is probably justified by the interaction
between this harmonic and the non-tracked FA mode with a frequency around 0.80 Hz (Figure 7.35).
Sensor: 1 (+74.988 m) +74.988 m
0.3 0.3
1Ω
0.25 3Ω
0.25
6Ω
0.2 0.2 1 FA
Acc. [m/s2]
Acc.[m/s 2]
2 FA
0.15 0.15 4 FA
0.1 0.1
0.05 0.05
0 0
0 2 4 6 8 10 12 14 16 17 0 5 10 15 20 25
Rotor speed [rpm] Wind speed [m/s]
a) b)
Figure 7.59 – RMS values from the modal and stationary responses in the FA direction at level +74.988 m with: a) rotor
speed; b) wind speed
304
Chapter 7
Once the recorded acceleration time series are decomposed into modal (and stationary) responses, it is
possible to use the procedure introduced in section 5.5.3 to quantify the relative contribution of each
responses to the measured signal. Likewise in the example from section 5.5.3, the acceleration
responses due to the harmonic excitation were also considered in this analysis. This analysis is helpful
to better understand the changes in the dynamic behaviour of the wind turbine structure over the
different conditions.
Figure 7.60 shows the evolution of the relative modal contribution of the FA modes and 1Ω, 3Ω and 6Ω
harmonics to the measured acceleration at the tower top sensor with the wind speed. The median
values (from each range of 0.5 m/s of wind speed) from the biggest contributors responses (1 FA
mode, 3Ω and 6Ω harmonics) are represented by the lines in light colour. For low wind speed
conditions (non-operating scenarios), the dynamic behaviour at the tower top is clearly dominated by
the 1 FA mode, with participations of almost 100 % in certain setups. Once the turbine starts
operating, the response is partially dominated by the 3Ω harmonic and by the 1 FA mode, whose
participation is still representative mainly due to the excitation of this mode by the 3Ω harmonic, as
previously explained. From the moment this effect vanishes, it is observed that the acceleration is
practically dominated by the 3Ω harmonic. From this point, the participation of this harmonic and the
1 FA mode tend to converge to a participation of 30 % each with the increase of the wind speed. The
participation of the 6Ω harmonic is only relevant when the turbine starts its operation, decreasing
from this point to participation levels below 5 % with the increase of the wind speed.
+74.988 m
1
1Ω
3Ω
0.75 6Ω
1 FA
Participation
2 FA
0.5 4 FA
0.25
0
0 5 10 15 20 25
Wind speed [m/s]
Figure 7.60 – Variation of the participation of the modal and stationary responses in the measured acceleration in the FA
direction at level +74.988 m with wind speed
The analysis presented to the sensor at the tower top was also performed at the +48.392 m level in the
FA direction. The results are presented in Figure 7.61. The acceleration, in the FA direction, is
dominated by the 2 FA mode with RMS values clearly higher than the ones presented in Figure 7.59
for the 1 FA mode. This result is probably justified by the excitation introduced by the blade rotation
(due to the tower shadow effect) at the level +48.392 m, corresponding to the maximum modal
amplitude of the 2 FA mode. Anyway, the dynamic displacements would be expected to be higher at
the tower top than at +48.392 m due to the lowest natural frequency of the 1 FA when compared to the
2 FA mode. This is confirmed later in section 7.3.7.
305
+48.392 m +48.392 m
1 1
1Ω 1Ω
3Ω 3Ω
0.8 0.8 6Ω
6Ω
1 FA 1 FA
Acc.[m/s 2]
Acc.[m/s 2]
0.6 2 FA 0.6 2 FA
4 FA 4 FA
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 10 12 14 16 0 5 10 15 20 25
Rotor speed [rpm] Wind speed [m/s]
a) b)
Figure 7.61 – RMS values from the modal and stationary responses in the FA direction at level +48.392 m with: a) rotor
speed; b) wind speed
The quantification of the contribution of the modal and stationary responses to the measured
acceleration time series was also assessed. The results are shown in Figure 7.62. In this figure, besides
the modes considered in Figure 7.60, the stationary response due to the 9Ω harmonic was also
included. However, it is important to note that around a rotor speed of 14 rpm, this harmonic crosses
the potential untracked FA vibration mode, identified with a frequency of around 2.30 Hz (Figure
7.35). Since this potential mode was not considered during the tracking procedure, the automated
algorithm developed to identify clusters related to the harmonics cannot distinguish, for this region of
the Campbell diagram, a cluster representative of the 9Ω harmonic from one representative of this
mode. Thus, it is possible that clusters referred in Figure 7.62 as due to the 9Ω harmonic are actually
representative of this potential mode.
The results presented in Figure 7.62 clearly show that, under low wind speed conditions (non-
operating regimes), the acceleration is dominated by the 1 FA mode. Some sparse points from the 1 FA
with high modal contribution are also visible for high wind speeds but these are referred to setups with
no power production. When the rotor starts operating, the contribution of the 1 FA drops drastically,
keeping low for every wind speed condition. As expected, the acceleration is dominated by the 2 FA
mode while the remaining the modes and harmonics are kept low during all operating conditions.
Notwithstanding, it is observed that, for wind speeds around 7 and 13 m/s, the relative contribution of
the 2 FA decreases slightly. This region coincides with the increase of the contribution of the 9Ω
harmonic. This increase is due to the cross of this harmonic with the referred potential vibration mode
with frequency of 2.30 Hz, leading to a possible interaction or resonance of this mode. For wind speeds
higher than 13 m/s, the relative contribution of the 2 FA increases to levels similar to the ones obtained
to winds speed below 7 m/s and the contribution of the 9Ω harmonic drops, meaning that this effect
was reduced.
306
Chapter 7
+48.392 m
1
1Ω
3Ω
0.75 6Ω
9Ω
Participation
1 FA
0.5 2 FA
4 FA
0.25
0
0 5 10 15 20 25
Wind speed [m/s]
Figure 7.62 – Variation of the participation of the modal and stationary responses in the measured acceleration in the FA
direction at level +48.392 m with wind speed
7.3.6.5 Removal of Operational and Environmental Effects

The characterization of the influence of the operational and environmental factors on the modal
properties of the tracked vibration modes is an important step to remove a significant part of their
variability. As these effects may introduce variations on the modal parameters with higher amplitudes
than the damage itself, the removal of the influence of these effects is crucial to detect small damages at
an early stage.
Wind turbines are equipped with sensors measuring several operational and environmental
conditions. This data is usually accessed through the SCADA system, as is the case of the Senvion
MM82 wind turbine. Some of the results obtained with this data were used to illustrate the operating
conditions of the turbine in section 7.3.3. For that reason, the application of methods to remove the
influence of the effects that take into account measurements of the predictors (input-output methods)
represents a suitable option to be applied to wind turbines.
The definition of the regression models is usually performed considering a period of data large enough
to include all the important variations of the predictors. In that sense, a period of one year is usually
used to define the regression model (Magalhães, Cunha et al., 2012). This model is then used to
evaluate the possible occurrence of damage in the following years. Since it was only possible to acquire
acceleration data corresponding to a period of about one year, a slightly different strategy was followed
in this work. Thus, the whole data was split into two different periods, consisting each one in data
from intercalary days. With this strategy, both periods contain results within a period of one year, but
from different days. Period 1 was then used to define a suitable multivariate regression model, while
Period 2 was used to assess the existence of damage.
The influence of some operational factors on the modal properties was already presented throughout
section 7.3.6.2. As example, the effect of the rotor speed on the frequency values of the modes, namely
for the 1 SS* and 2 SS*, was shown. The changes on the behaviour of the natural frequencies with the
different operating regimes were also presented in Figure 7.44. The variability shown between regimes
evidences the effect that different factors have on the wind turbine. For that reason, a different
multivariate linear regression model was established for each operating regime, reflecting the different
307
conditions. The operational and environmental factors recorded by the SCADA system and
considered in this analysis were:
o Wind speed (w. s.);
o Outdoor temperature (temp.);
o Rotor speed (r. s.);
o Blade pitch angle (pitch).
The operating regime 1 is characterized by values of blades pitch angle (around 90º) with small
variations, during non-production setups. Thus, the considered predictors influencing the most the
variability of the frequency values are the temperature and, in a small scale, the wind speed. These
expected results are confirmed by the obtained correlation coefficients (computed according to
equation (7.4)) and presented in Table 7.12. This coefficient permits to highlight the modes whose
natural frequencies are more (linearly) influenced by the factors under study. Figures 7.65 and 7.66
show two examples of the correlation between natural frequency and the wind speed and temperature
for two different modes, respectively.
Table 7.12 – Correlation coefficients between natural frequencies and operational/ environmental factors (Regime 1)
w. s. temp. r. s. pitch
1 SS 0.12 -0.23 0.08 0.08

1 FA 0.38 -0.28 0.34 -0.03
1 SS* -0.49 -0.36 -0.44 0.09
2 SS* 0.19 -0.57 0.01 0.46
2 FA 0.25 -0.64 0.15 -0.02
2 SS -0.13 -0.43 -0.13 -0.08
3 SS* 0.29 -0.70 0.21 -0.11
4 FA -0.36 -0.49 -0.35 -0.01
1.3 2.85
1 SS* 2 FA
Linear fit 2.825 Linear fit
1.28
2.8
Frequency [Hz]
Frequency [Hz]
1.26
2.775
1.24
2.75
1.22
2.725
1.2 2.7
0 5 10 15 20 25 0 5 10 15 20 25 30 35
Wind speed [m/s] Temperature [ºC]
Figure 7.63 – Natural frequency of 1 SS* mode (Regime 1) Figure 7.64 – Natural frequency of 2 FA mode (Regime 1)
vs wind speed vs temperature
308
Chapter 7
Regime 2 is similar to Regime 1 to the extent that they are both referred to non-production setups, in
which the blades pitch angles is kept almost constant (although at lower values in Regime 2). Thus, it is
comprehensible that the predictors with the highest correlation with the natural frequencies are again
the wind speed and the temperature, as seen in the results of Table 7.13.
1 SS 0.27 -0.28 0.33 -0.06

1 FA -0.11 -0.64 0.01 0.03
1 SS* -0.45 -0.36 -0.39 0.06
2 SS* -0.19 -0.39 -0.13 -0.03
2 FA -0.06 -0.57 0.03 0.03
2 SS -0.18 -0.36 -0.09 -0.05
3 SS* -0.12 -0.35 0.04 -0.12
4 FA -0.07 -0.31 -0.01 -0.15
The correlation coefficients obtained between natural frequencies from Regime 3 and operational/
environmental factors are shown in Table 7.14. Operating regime 3 corresponds to transition situation
in which the turbine changes from parked/ idling configuration to operating conditions, or vice-versa.
For that reason, both the rotor speed and pitch angle show important changes during these setups.
However, apart from some modes (such as 1 SS*, 2 SS* and 4 FA modes), these changes do not seem to
have an important influence on the frequency values.
1 SS -0.09 -0.33 -0.08 -0.01

1 FA 0.03 -0.19 0.22 -0.23
1 SS* -0.70 -0.37 0.09 -0.23
2 SS* 0.56 -0.26 -0.01 0.19
2 FA 0.02 -0.38 -0.02 0.00
2 SS -0.03 -0.31 0.25 -0.22
3 SS* 0.25 -0.12 0.05 0.02
4 FA 0.17 -0.20 0.32 -0.23
The number of identified setups regarded to Regime 4 is considerably higher than for the first 3
regimes. This fact permitted to obtain a better characterization of the variations of the natural
frequencies for this regime. The results obtain for the correlation analysis are presented in Table 7.15.
309
1 SS 0.22 -0.50 0.27 0.08

1 FA -0.40 0.02 -0.39 -0.14
1 SS* -0.93 0.15 -0.95 -0.26
2 SS* 0.91 -0.42 0.97 0.15
2 FA 0.44 -0.51 0.51 0.05
2 SS 0.23 -0.54 0.34 -0.05
3 SS* 0.84 -0.53 0.88 0.15
4 FA 0.41 -0.55 0.48 0.10
Both the rotor speed and wind speed (which are themselves correlated) present interesting high values
of correlation with the frequencies. As expected, the pitch angle does not represent an interesting
source for explanation of the frequencies variability, since its value is practically constant during this
Regime. On the other hand, the temperature presents important linear relationship with the
frequencies, which are confirmed by visual observation of the results. The evolution of the frequency
values of the 2 FA mode with the temperature is illustrated in Figure 7.65.
The highest correlation values are obtained with the rotor speed. Due to the nature of the whirl modes,
this fact is especially important for the 1 SS* and 2 SS* modes. This phenomenon is illustrated for the 2
SS* mode in Figure 7.66.
Although slightly lower than with the rotor speed, the correlation coefficients obtained with the wind
speed are also relevant. However, this fact is mainly related to the relation between the wind speed and
the rotor speed than to physical relations between the wind and the natural frequencies.
3.95
2 FA 1.95
2 SS*
3.9 Linear fit
1.9 Linear fit
3.85 1.85
Frequency [Hz]
Frequency [Hz]
1.8
3.8
1.75
3.75
1.7
3.7
1.65
3.65 1.6
0 5 10 15 20 25 30 35 9 10 11 12 13 14 15
Temperature [”C] Rotor speed [m/s]
Figure 7.65 – Natural frequency of 3 SS* mode (Regime 4) Figure 7.66 – Natural frequency of 2 SS* mode (Regime 4)
vs temperature vs rotor speed
Finally, Regime 5 is related to production setups with the strongest wind speed conditions. It is
characterized by an increase of the blades pitch angle with the wind speed in order to control the rotor
speed. This Regime is then commanded by the pitch angle, which in turn is defined according to the
wind speed. The relation between the pitch angle and the wind speed was already introduced in Figure
7.14. Again, the 1 SS* and 2 SS* present a large variability of their frequency values due to operational
310
Chapter 7
factors. The variation introduced by the pitch angle in the frequency values of the 1 SS* mode is shown
in Figure 7.67. Also the temperature seems to be an important predictor of the natural frequency of the
1 SS mode. This relationship is illustrated in Figure 7.68.
1 SS -0.02 -0.47 0.00 -0.01

1 FA -0.31 -0.27 -0.34 -0.34
1 SS* -0.91 -0.19 -0.68 -0.92
2 SS* -0.68 -0.32 -0.42 -0.67
2 FA -0.22 -0.56 -0.26 -0.22
2 SS -0.06 -0.57 -0.05 -0.04
3 SS* -0.22 -0.47 -0.01 -0.19
4 FA 0.42 -0.11 0.33 0.43
1.34 0.37
1 SS* 1 SS
1.32 Linear fit 0.365 Linear fit
1.3 0.36
Frequency [Hz]
Frequency [Hz]
1.28 0.355
1.26 0.35
1.24 0.345
1.22 0.34
-5 0 5 10 15 20 25 0 5 10 15 20 25 30
Pitch angle [º] Temperature [ºC]
Figure 7.67 – Natural frequency of 1 SS* mode (Regime 5) Figure 7.68 – Natural frequency of 1 SS* mode (Regime 5)
vs blades pitch angle vs temperature
Once the influence of the predictors was analysed for each regime, two alternative regression models
were considered: SM1 and SM2.
SM1 only considers the predictors with important physical relations with the natural frequencies. For
that reason, even predictors with high correlation coefficients (such as wind speed in Regime 4 and 5)
were not considered, since the coefficient value is not related to a physical relation with this predictor
but to a correlation between other predictors, as previously referred. The predictors considered for this
model are presented in Table 7.17. On the other hand, the SM2 model considers all predictors, aiming
to define a general regression model with potential application without any previous analysis (Table
7.17).
311
Table 7.17 – Considered predictors for regression models SM1 and SM2
SM 1 Regime SM 2 Regime
Predictors 1 2 3 4 5 Predictors 1 2 3 4 5
w. s. x x x w. s. x x x x x
temp. x x x x x temp. x x x x x
r. s. x r. s. x x x x x
pitch x pitch x x x x x
The suitability of both models to produce forecasts was quantified through the computation of the
coefficient of determination , presented in equation (7.8). For its calculation, the Period 2 of data
was used to assess the quality of the forecasts (as previously referred, this period of data was not
considered in the construction of the regression models). The results obtained for the two models are
very similar (Table 7.18). In that sense, the SM 2 was chosen since it considers all predictors, making
its use more general to be implemented in any wind turbine in any regime.
It is observed from Table 7.18 that the modes related to rotor motion (1 SS*, 2 SS* and 3 SS*) present
the highest values of . The good quality of the predictions of these modes is justified by the fact
that the variability of these modes is mostly driven by the rotor speed, which is a known predictor. On
the other hand, the 1st tower bending modes (1 SS and 1 FA) presents the lower coefficients of
determination of all modes, showing that their variability is less dependent on the considered
operational/ environmental effects. One possible cause to the low coefficients obtained for some
modes might be related to the potential influence of asymmetric soil conditions on the natural
frequency values, which could be considered by using the yaw angle as predictor in the regression
model. However, it was decided not to use this predictor in order to assess the accuracy of the
monitoring system to detect foundation asymmetries (see section 7.3.6.6). Notwithstanding, in
situations in which the installation of the monitoring system is performed at an early stage of the
turbine operation, the consideration of the yaw angle in the model would increase the accuracy of the
system to detect abnormal structural changes.
Table 7.18 – Coefficients of determination obtained with the regression models
Modes SM 1 SM 2
1 SS 0.295 0.297
1 FA 0.377 0.391
1 SS* 0.965 0.967
2 SS* 0.950 0.952
2 FA 0.458 0.458
2 SS 0.706 0.719
3 SS* 0.860 0.864
4 FA 0.777 0.780
312
Chapter 7
Although steel structures do not usually present a significant thermal inertia, concrete structures do.
Considering the importance of the support conditions (i. e. concrete foundation) in the definition of
the dynamic properties of wind turbines, alternative dynamic models were also analysed in order to
assess the influence of temperature records from previous time instants on the structure (see section
7.2.1.1). In that sense, 5 dynamic models were considered. These models consider all the predictors
from the SM 2 model (including the temperature at the instant under consideration), along with the
temperature records referred to different time delays of 3, 6, 12, 18 and 24 hours. For each dynamic
model referred to a temperature measurement from a previous time instant, the temperatures from the
more recent time instants were also considered. This means that, for a dynamic model considering
temperature record with a delay of 12 hours, the temperature measurements from the current time
instant and from time delays of 3, 6 and 12 hours were considered. The results obtained are shown in
Table 7.19. In this table, the dynamic models are identified by the largest time delay considered. It can
be seen that, although some improvements are noticed for the 1 FA and 3 SS* modes, smaller values
were obtained for the 2nd pair of tower bending modes. Since the 2 FA and 2 SS modes are very
sensitive to structural changes in the foundation condition and, consequently, represent important
features to assess the existence of damage, the static model SM 2 was considered to be adequate to
eliminate the environmental and operational effects on the frequency values.
Table 7.19 – Coefficients of determination obtained with the dynamic models (the largest time delay considered for each
model is indicated within parenthesis)
DM 1 DM 2 DM 3 DM 4 DM 5
Modes
(3 hours) (6 hours) (12 hours) (18 hours) (24 hours)
1 SS 0.296 0.293 0.292 0.284 0.299
1 FA 0.396 0.399 0.392 0.388 0.391
1 SS* 0.966 0.967 0.966 0.964 0.965
2 SS* 0.949 0.949 0.951 0.947 0.947
2 FA 0.450 0.445 0.436 0.440 0.428
2 SS 0.716 0.713 0.721 0.707 0.710
3 SS* 0.866 0.866 0.864 0.871 0.875
4 FA 0.775 0.774 0.771 0.767 0.770
In order to illustrate the quality of the SM 2 model, Figure 7.69 presents the predicted values of natural
frequency of the 1 SS* and 4 FA modes for some data sets from Period 2, alongside with the identified
values of these modes with the monitoring system. The accuracy of the prediction obtained with this
model is clearly visible in this figure.
313
1 SS* 4 FA
Identified 4.6 Identified

1.4 Prediction 4.5 Prediction
4.4
Frequency [Hz]
Frequency [Hz]
1.35 4.3
4.2
1.3
4.1
4
1.25
3.9
1.2 3.8
1.265 1.27 1.275 1.28 1.285 1.29 1.44 1.445 1.45 1.455 1.46 1.465 1.47 1.475
Index 4 Index 4
x 10 x 10
Figure 7.69 – Identified and predicted frequency value of the 1 SS* and 4 FA using the SM 2 mode during (data
corresponding to Period 2)
Using the regression model SM2, the effect of operational and environmental effects on the natural
frequencies of the modes can be then minimized. Figure 7.70 shows the time evolution of the
frequency values of the 1 SS* mode before and after the removal of these effects during period 2. These
effects (mainly the rotor speed) have a large influence on the variation of the frequency values of this
mode. Notwithstanding, the SM2 model is capable of mitigating the influence of these effects, as seen
in the bottom figure.
Before
1.5
1.45
1.4
Frequency [Hz]
1.35
1.3
1.25
1.2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Index
After
1.5
1.45
1.4
Frequency [Hz]
1.35
1.3
1.25
1.2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Index
Figure 7.70 – Time evolution of the frequency values of the 1 SS* mode before (top) and after (bottom) the removal of the
operational and environmental effects during period 2
314
Chapter 7
Figure 7.71 presents the frequency values of the 2 FA vibration mode along the time. Although this
mode does not present a variability as high as the 1 SS* mode, it is also possible to attest its reduction
after the correction of the frequency values with the regression model SM 2.
Before
2.95
2.9
Frequency [Hz] 2.85
2.8
2.75
2.7
2.65
0 2000 4000 6000 8000 10000
Index
After
2.95
2.9
Frequency [Hz]
2.85
2.8
2.75
2.7
2.65
0 2000 4000 6000 8000 10000
Index
Figure 7.71 – Time evolution of the frequency values of the 2 FA mode before (top) and after (bottom) the removal of the
operational and environmental effects during period 2
The reduction of variability of the frequency values of the vibration modes is also evident in the
histograms of Figure 7.72. As can be seen, the 1st tower bending mode in the SS direction presents the
lowest reduction in its variability from the three considered modes. This result is in accordance with
the results from Table 7.18. The reduction obtained with the correction of the frequency values of the 2
FA mode is notorious in the histogram, where the corrected frequencies are concentrated in a
narrower range. Lastly, the histogram of the 1 SS* mode illustrates the large reduction of variability
obtained with the regression model. It is also interesting to note the considerable different shape of the
histogram of this mode before the correction, illustrating the different nature of these rotor modes in
comparison with tower modes.
315
After After After

1 SS Before 1 SS* Before 2 FA Before
2500
1500
2000 1500
1000 1500
1000
1000
500 500
500
0 0 0
0.34 0.35 0.36 0.37 1.25 1.3 1.35 1.4 1.45 2.7 2.75 2.8 2.85 2.9
Frequency [Hz] Frequency [Hz] Frequency [Hz]
Figure 7.72 – Histograms of the natural frequencies of the 1 SS, 1 SS* and 2 FA modes before and after the removal of
operational and environmental effects
7.3.6.6 Damage Detection

The definition of the static regression model in the previous section and the subsequent removal of the
influence of the operational and environmental effects allowed reducing the variability of the
frequency values of the tracked vibration modes. Thus, it is now possible to assess the variations of
these values in order to detect the occurrence of small structural changes.
One important issue related to structural changes in wind turbines is the possible reduction of the
foundation stiffness along time. Although this problem is more related to offshore turbines, due to
scour, it is also relevant in onshore since problems in the connection of the tower with the foundation
have already been observed in several towers. Therefore, it is interesting to assess if a variation of
frequency is noticed in the monitored Senvion wind turbine. As referred in section 7.1, the 2nd pair of
tower bending mode is one of the most sensitive modes to changes in the stiffness of the foundation. In
that sense, the frequency values of the 2 FA and 2 SS modes from periods 1 and 2 were corrected
according to the model SM 2 and plotted against the yaw angle. The results obtained are presented in
Figure 7.73. In these figures, also the median values and the box plots corresponding to the 25th and
75th percentile are shown. It is seen that, although small, both plots present coherent fluctuations. The
mode in the FA direction presents the highest values of frequency at 150º and at 330º (= 150º + 180º).
On the other hand, the smallest values are found for yaw angles of 70º and 240º (≈ 70º + 180º). It is
interesting to note that the highest values of frequency of the 2 SS mode are obtained for the directions
of the smallest values of 2 FA mode and vice-versa. This means that it is very likely that the foundation
or the connection of the tower to the foundation is not structurally homogeneous, with the stiffest
direction in the 150º - 330º direction and a (practically) orthogonal softest direction in the 70º-240º
direction.
316
Chapter 7
2 FA
2.86
2.84
2.82
Frequency [Hz]
2.8
2.78
2.76
2.74
0 40 80 120 160 200 240 280 320 360
Yaw angle [º]
2 SS
2.9
2.85
Frequency [Hz]
2.8
2.75
0 40 80 120 160 200 240 280 320 360
Yaw angle [º]
Figure 7.73 – Corrected frequency values of the 2 FA (top) and 2 SS (bottom) from Period 1 and 2 vs yaw angle
Although in a smaller scale, it is also possible to identify the same variation of the frequency value of
the 1 SS vibration mode with the yaw angle (Figure 7.74). This variation is similar to the one presented
by the 2 SS, which reinforces the idea of a possible asymmetry of the foundation stiffness. The 1 FA
mode do not evidences a noticeable variation, probably due to its larger variability which probably
hinders possible fluctuations due to asymmetric foundation conditions.
Nevertheless, it is not possible to say, at this moment, that these conclusions attest the existence of
damage in the foundation or, more likely, in the tower-foundation connection. It is possible that this
non-homogeneity of the foundation is already present since the beginning of operation of the turbine
and that is related to slightly different soil conditions under the large foundation of the support
structure. In that sense, the installation of a continuous dynamic monitoring at the initial stage of the
turbine operation would lead to conclusive results, permitting to identify if this non-homogeneity is a
consequence of structural changes occurred throughout its period of life or if it just represents the
stiffness of the foundation conditions.
317
1 SS 0.12
Wind speed [m/s:]
0.36
3-4
0.1 4-5
0.358 5-6
6-7
0.08
7-8
Acc. [m/s 2]
Frequency [Hz]
0.356
0.06
0.354
0.04
0.352 0.02
0.35 0
0 40 80 120 160 200 240 280 320 360 0 40 80 120 160 200 240 280 320 360
Yaw angle [º] Yaw angle [º]
Figure 7.74 – Corrected frequency values of the 1 SS from Figure 7.75 – Median values of the RMS values of
Period 1 and 2 vs yaw angle acceleration of the top sensor in the FA direction
according to the yaw angle, considering setups when the
turbine was operating and several wind speed ranges
In the previously presented Figure 7.29, the variation of RMS values of acceleration from narrow
ranges of wind speeds with the yaw angle was analysed. This plot is again presented in Figure 7.75. It is
interesting to see that the direction with the highest acceleration RMS values (around 180º - 240º)
coincides with the softest direction. Still, the acceleration levels around 70º are very low, which may
seems a contradiction. However, the number of recorded operating events with the nacelle oriented to
the yaw sector between 0º and 70º is smaller than for the other sectors, which may not be
representative and, thus, hinder the effect of the softest foundation when the turbine is oriented to this
sector.
The yaw sector between 110º and 140º is also a sector with high acceleration RMS values. This is the
direction of the main wind incidence, as shown in Figure 7.15. This sector is located in a transition
location between the stiffest and softest directions of the foundation. For that reason, the high
acceleration levels detected might be also related to the wind action itself (higher levels of turbulence)
than just motivated by a possible loss of stiffness of the foundation.
As previously referred, damage in wind turbine support structures is usually related to foundation
problems. For offshore turbines, scour problems are likely to occur, decreasing the strength of the
foundation and potentially decreasing the expected fatigue life of the structure (Sørensen and Ibsen,
2013). Although in a small scale, some problems of excessive levels of vibrations were also detected due
to damage in the interface connection between tower and foundation (Pelayo, López-Aenlle et al.,
2011; Currie, Saafi et al., 2015).
Apart from foundation problems, rotor blades are also elements prone to damage and with a costly
replacement, as shown in section 3.1.3. In that sense, it would be interesting to assess the sensitivity of
the identified rotor vibration modes to small structural damages of the blades.
In order to evaluate the ability of the dynamic monitoring system to detect the presence of small
damage in the previously referred elements of a wind turbine, a strategy based on the introduction of
artificial damage was followed. For the assessment of damage at the foundation, two scenarios were
studied. The damage scenario D1 is related to scour problems in offshore wind turbines. In that sense,
the evolution of the first natural frequency of the support structure with the scour depth presented in
318
Chapter 7
(Prendergast, Gavin et al., 2015) was considered (see Figure 7.1). A small scour depth of around 0.075
times the base diameter of the monopile is considered. For the case of the Senvion MM82, with a tower
base diameter of 4.300 m, this damage corresponds to a scour depth of around 0.32 m, while for the
offshore Vestas V90 wind turbine represents a depth of 0.38 m. This damage is said to correspond to a
decrease of the first tower bending mode of around 0.2 %. However this study only considers the
evolution of the first natural frequency, while nothing is said about the second mode of the support
structure. To overcome this situation, results of the studies introduced in (Sørensen and Ibsen, 2013)
and (Zaaijer, Tempel et al., 2002) were considered. These studies present a ratio between the decrease
of the second tower bending mode relative to the first tower bending mode of 2.1 times to 2.6 times,
respectively, for monopile foundations. Following these results, a conservative value of 2.0 was used in
this damage scenario. Since it is not expected that rotor blades modes are influenced by this damage,
their frequency values is not changed.
The second case of foundation damage (scenario D2) reproduces the frequency variation detected in
(Pelayo, López-Aenlle et al., 2011). This study describes the dynamic tests performed in three wind
turbines, from which two of them presented anomalous vibration levels. It was also found that these
turbines presented cracks at the foundation. The test indicated that the frequency values of the
identified vibration modes from these two turbines were consistently lower than the ones from the
third healthy turbine. Initially, the frequency deviations of the first and second tower bending modes
presented by the turbine with the smallest amplitude of variations were considered in this scenario.
However, since this damage scenario was clearly identified, a smaller variation was studied. Thus, a
variation 5 times smaller than the variation reported in the study was used for this damage scenario.
Due to the difficulty in the comparison between the 4 FA mode and the vibration modes presented in
this study, this mode was also not considered. Again, the frequency value of the rotor blades modes
was not changed for this scenario.
The last damage scenario (D3) is related to blade damage. As referred in section 7.1.2, the majority of
published studies refer the small sensitivity of the natural frequencies of the blade modes to small,
common damages. In that sense, the model of the Senvion MM82 developed in the HAWC2 code
(introduced in section 7.3.2) was used to assess the sensitivity of the 1 SS* and 2 SS* modes to
structural damage at the blades. Considering that the eigenvalue analysis performed by the HAWC2
code is related to the turbine with the blades according to an operating configuration but under parked
conditions, the expected frequency value of these two modes, introduced in section 7.3.6.2, was used to
tune the stiffness of the blades. Due to the high uncertainties in modelling the structural elements
located inside the nacelle, as well as the rotor blades, the estimation of the frequency value of the 3 SS*
mode was not possible. The comparison between the frequency values obtained with the numerical
model and with the monitoring system is introduced in Table 7.20.
Table 7.20 – Comparison between the expected frequency values of the 1 SS* and 2 SS* modes during the monitored
period (non-operating conditions) with the results obtained with the HAWC2 model
Natural frequency [Hz]
Modes Monitoring HAWC2

1 SS* 1.526 1.530
2 SS* 1.558 1.556
319
The presence of damage at just one (or two) blades leads to an anisotropic rotor. In that sense, the
identification of such damage would be more easily detected through the identification of the 1Ω
harmonic (as illustrated in Figure 4.18) or through the identification of phenomena such as the ones
described in (Ramírez, Tcherniak et al., 2015), where a clear distinction in the singular values spectrum
of the structure is identified between an isotropic and an anisotropic rotor.
Therefore, the damage scenario D3 is referred to a similar deterioration of three blades, which may be
caused by continuous wear of the blades due to operation. The stiffness of the three blades was
decreased in 15 % over a length of 2 m, corresponding to 5 % of the total length of the blade. The
damage was located at around 33 % in chord length from the blade root, which is said to be the
location more prone to damage (Ciang, Lee et al., 2008). The imposed variation of stiffness led to a
decrease of -0.65 % of the 1 SS* and 2 SS* rotor vibration modes. It was also noticed that the imposed
stiffness variation was not significantly reflected in the other modes. For that reason, only the rotor
modes 1 SS* and 2 SS* were used to detect this damage scenario.
The variation of the natural frequencies of the vibration modes with the considered damage scenarios
is resumed in Table 7.21. It is considered that the damage from each scenario occurred at the middle of
Period 2. The artificial damage is thus introduced in the detected frequencies from the second part of
Period 2 by reducing their values according to the values presented in Table 7.21.
Table 7.21 – Variation of the natural frequencies of the modes associated with the damage scenarios
Δ [%]
Modes D1 D2 D3
1 SS -0.20 -0.50 0.00

1 FA -0.20 -0.50 0.00
1 SS* 0.00 0.00 -0.65
2 SS* 0.00 0.00 -0.65
2 FA -0.40 -0.24 0.00
2 SS -0.40 -0.24 0.00
3 SS* 0.00 0.00 -
4 FA - - -
The methodology followed to assess the presence of damage in the wind turbine is based on the use of
control charts, already introduced in section 7.2.2. This multivariate technique allows assessing the
evolution of several vibration modes in the same chart. For the construction of these control charts,
the residual error was used. As referred, the residual error is the difference between the identified
frequency value of a vibration mode and its predicted value according to the defined SM 2 regression
model, as illustrated in Figure 7.69.
The control charts obtained for an undamaged situation and the three damage scenarios are presented
in Figure 7.76. In their construction, groups with 36 observations were considered (corresponding to a
quarter of a day). It is seen that the damages are clearly detected in all situations. However, a different
upper limit control of the chart from the one defined in equation (7.15) was used due to the large
number of false alarms obtained with this limit. This situation is most likely due to the fact that
320
Chapter 7
features considered do not respect all the assumptions adopted for the definition of the limits
(Magalhães, Cunha et al., 2012). To overcome this limitation, a different limit was imposed. This limit
is defined by considering that 95 % of the values from the training period are considered within the
safety region.
Undamaged scenario D1
300 300
250 250
200 200
T2
150
T2
150
100
100
50
50
0
0
Time
Time
D2 D3
300 300
250 250
200 200
T2
150
T2
150
100
100
50
50
0
0
Time
Time
Figure 7.76 – control charts associated with a scenario where no damage is present and with the three defined
damage scenarios
This is a very relevant result, since it demonstrates that quite small damages in the foundation (and,
although with less accuracy, in the blades) can be detected with the developed monitoring
methodology. In the case of scenario D1, it is relevant to note that the identified damage
(0.075 ) is considerably smaller than the recommended design value of 1.30 (Det
Norske Veritas (DNV), 2011).
It should be further referred that a longer monitoring period would even permit the detection of
smaller frequency variations due to a better and more complete regression model and due to the
possibility of using more observations to compute the control chart, leading to more robust results.
321
7.3.7 FATIGUE ASSESSMENT RESULTS

The characterization of the fatigue condition of the wind turbine support structure was performed
according to the procedure introduced in section 6.4.1. Unfortunately, it was not possible to install,
during the period of this thesis, additional sensors to quantify the quasi-static motion of the tower.
Consequently, this work is only focused on the estimation of the dynamic component of fatigue
damage.
The procedure is illustrated with 10 min. acceleration time series recorded during a period of strong
wind conditions, with a mean wind speed of 20.9 m/s and maximum (instantaneous) speed of 35.9
m/s. During this period, the turbine operated at its rated rotor speed (around 16.5 rpm) and the blades
pitch angle varied between 11.1º and 27.9º, in order to control the rotor torque. The acceleration time
series collected by the dynamic monitoring system in the FA direction are presented in Figure 7.77.
2.5
+74.988
2
+48.392
1.5
+21.772
Acceleration [m/s2]
1
0.5
0
-0.5
-1
-1.5
-2
-2.5
0 100 200 300 400 500 600
Time [s]
Figure 7.77 – Acceleration time series in the FA direction
Prior to the integration process, the acceleration time series were multiplied by a time window with the
aim of softening the start and end of the time series as discussed in section 6.4.1.1. These initial and
last parts are then disregarded in the integrated displacement time series. After an initial analysis of
several setups, it was considered that the disregarded of the initial and last 30 seconds would guarantee
a safety margin to avoid transient errors.
For the double integration process, the filter introduced in section 6.4.1.1 (Figure 6.12) was used. The
constants of the filter ( and ) were set to only consider the dynamic component of the
displacements.
A good solution to tune these two parameters would be possible with the temporary use of
instrumentation to measure the stress condition of the tower at an arbitrary condition. Since it is
possible to estimate the stress condition at any position of the support structure with the presented
methodology, the results obtained with the temporary instrumentation would be compared with the
ones estimated with the accelerometers in order to tune the parameters. Since it was not possible to
install this instrumentation, a compromise between the amplification of possible low frequency errors
and the loss of relevant dynamic component information was studied with the help of numerical tests
performed with the HAWC2 code (Larsen and Hansen, 2007). As result, the constants f and f were
defined as, respectively, 0.10 Hz and 0.15 Hz. In that sense, the analysis described in this section
regarding the fatigue assessment of the Senvion MM82 wind turbine is focused on the validation of the
322
Chapter 7
developed procedure considering only the dynamic component of the fatigue, and not on the
estimation of the real fatigue condition of the structure.
Figure 7.78 shows the obtained displacements at the measurement levels in the FA direction. As
expected, the displacements are higher at the top, in opposition to what was obtained with the
acceleration time series (Figure 7.77). This result was expected since the displacements are mainly
driven by the 1st pair of tower bending modes. In this figure, the transient initial and last parts are
indicated by the vertical dashed lines and correspond to the referred 30 seconds in each part.
0.15
+74.988
+48.392
0.1 +21.772
Displacements [m]
0.05
-0.05
-0.1
0 100 200 300 400 500 600

Time [s]
Figure 7.78 – Dynamic displacement time series in FA direction (the vertical dashed lines indicate the disregarded initial
and last part)
Once the dynamic displacements along the turbine support structure are obtained, it is possible to
estimate the dynamic component of the stress condition at any level of the tower. As an example, the
bending stress at one point of the tower section at the foundation level was computed using the
procedure based on the SSI-DATA identification algorithm described in section 6.4.1.1. The result is
presented in Figure 7.79.
40
30
20
10
-10
-20
-30
-40
0 100 200 300 400 500 600
Time [s]
Figure 7.79 – Estimated dynamic bending stress at the foundation level due to motion in the FA direction (the vertical
dashed lines indicate the disregarded initial and last part)
323
Lastly, the load cycles are counted with the Rainflow algorithm. The histogram with the number of
bending stress cycles associated with different stress amplitudes is shown in Figure 7.80. The damage
due to fatigue can then be assessed with the S-N curve defined for the detail under analysis.
350
300
250
200
Nº cycles
150
100
50
0
0 10 20 30 40 50 60 70 80
Figure 7.80 – Histogram with the number of bending stress cycles
Considering that sensors installed at the tower structure are capable of detecting the motion in both
orthogonal directions, the stress due to bending in the SS direction can also be computed with this
methodology. With the stress condition computed in both FA and SS directions, it is thus possible to
estimate the bending stress condition at any point of the tower cross section. Initially, the bending
moment vectors in the FA and SS directions are reoriented according to the original SCADA
referential (N-S, E-W). This coordinate rotation, illustrated in Figure 7.81, depends on the yaw angle.
σP, E-W
N N
FA
MN-S MN-S
direction SS
direction
MSS P σP,N-S
R.cos(β) β
W E W E
ME-W R.sin(β) ME-W
MFA
S S
Figure 7.81 – Coordinate rotation from FA-SS referential to Figure 7.82 – Calculation of the bending stress at a generic
NS-EW referential point P
324
Chapter 7
After defining the moment vectors in the original SCADA referential, it is then possible to estimate the
stress condition at any point of the support structure cross section. This operation is illustrated in
Figure 7.82 for a generic point . The stress condition at the point is then defined according to:
. sin . cos
(7.18)
with:
Area moment of inertia of the cross section
Radius of the cross section
A discretization of the cross section in 18 sectors was considered in order to assess the fatigue
condition at the foundation level. The detail 71 from Eurocode 3 (European Comitee for
Standardization (CEN), 2005) was used to study the welding-on flange connection between the tower
and the foundation (Sørensen and Sørensen, 2011). The S-N curve corresponding to this detail was
already introduced in Figure 6.7.
Alongside with the bending stress estimation, it is also of utmost importance to have confidence on the
obtained results. Since the procedure estimates the acceleration time series as a sum of vibration
modes/ harmonics, it is necessary to evaluate if the estimations do not underestimate the motion of the
structure (due to a badly defined state-space model) or, on the other hand, overestimate the
importance of some frequency content (due to numerical residues). In that sense, the residual error ∆
from the forward innovation model (equation (5.80)) can be used to assess the quality of the estimated
acceleration time series. However, in the context of stress estimation, in which a double integration is
performed to estimate the displacement time series, the use of residual error as the only quality control
parameter is not sufficient.
Figure 7.83 shows an example of an estimated acceleration time series obtained from the forward
innovation model, using all the 7 installed sensors. The residual error ∆ obtained is low (2 %),
indicating that the estimation is good. The good agreement between the two signals is visually verified.
325
0.03 0.03
Measured
0.02 Estimated 0.02
0.01 0.01
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.01 -0.01
-0.02 -0.02
-0.03 -0.03
0 100 200 300 400 500 600 325 330 335 340 345 350
Time [s] Time [s]
Figure 7.83 – Measured and estimated acceleration time series at the level +21.772 in the SS direction, considering all
sensors as input for the forward innovation model (the right hand side figure shows a zoom of the time segment
identified in the left hand side figure by the dashed lines)
However, looking at these results in the frequency domain, important errors in the lowest part of the
frequency range are observed. Figure 7.84 shows the power spectral density of the measured and
estimated acceleration time series illustrated in Figure 7.83 (the spectrum is averaged with segments of
1024 points in order to facilitate the explaination). From this figure, it is possible to attest that,
although a good estimation is obtained around important resonance peaks (namely the 1st and 2nd
tower bending modes, as well as the 3Ω and 6Ω harmonics), there is an erroneous increase of the
amplitude of several peaks in the low part of the spectrum. Since the response is clearly dominated by
the 2nd bending mode, these errors are almost imperceptible in time domain, as evidenced in Figure
7.83. This is the reason why the error ∆ is low.
-4 Measured
10
Estimated
-6
10
Amplitude
-8
10
-10
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Frequency [Hz]
Figure 7.84 – Measured and estimated acceleration time series (in frequency domain) at the level +21.772 in the SS
direction, considering all sensors as input for the forward innovation model
The main problem related to the bad estimation of the lower part of the spectrum occurs during the
integration process to estimate the displacement time series. As referred in section 6.4.1.1, during this
operation, the amplitude of vibrations related to the low frequencies is increased. Consequently, the
displacement time series obtained from the measured and estimated acceleration will lead to
considerably different results. Figure 7.85 illustrates the displacement time series obtained from the
326
Chapter 7
measured and estimated acceleration signals represented in Figure 7.83. The erroneous majoration of
the displacements obtained from the estimated signals is evident. Under these circunstancies, the
estimation of the stress condition would naturally lead to erroneous results.
-3
x 10
1.5
Measured
1 Estimated
0.5
Disp. [m]
-0.5
-1
-1.5
0 100 200 300 400 500 600
Time [s]
Figure 7.85 – Displacement time series obtained with the double integration process based on the measured and
estimated acceleration signals
Due to possible occurrence of situations such as the one described, an additional quality index of the
estimation is required to assess the adjustment in the lowest part of the spectrum. An index ∆
based on the relative difference of the energy countained in the lowest part of the frequency range of
the estimated and measured acceleration signals was chosen. This index is defined as the ratio between
the area defined by the difference of the spectra of the estimated and measured acceleration and the
area defined by the spectrum of the measured acceleration signal (dark grey area in Figure 7.86). Only
the frequency range between and is considered to compute this index. In the present work, these
values were defined as 0.10 Hz (equal to ) and 0.35 Hz, in order to evaluate the adjustment of the
signal for the frequency range below the 1st tower bending mode.
f1 f2
-4 Measured
10
Estimated
-6
10
Amplitude
-8
10
-10
10
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2

Frequency [Hz]
Figure 7.86 – Definition of the area between the spectra of the estimated and measured acceleration signals and the area
from the measured signal
The results obtained for the quality indices using data sets collected during one year of monitoring
were initially assessed. It was observed that the quality index ∆ is not very restrictive. Figure 7.87 a)
327
shows the evolution of percentage of selected 10 min. time series events with the value adopted for ∆
considering the results obtained with the three tested modal identification algorithms. It is seen that
for the method based on the SSI-DATA algorithm almost all events are considered for an admissible
level of 0.15. For the other algorithms, it is visible that the estimations are apparently not so good.
100 100
Percentage of considered events [%]

80 80
60 60
40 40
SSI-DATA SSI-DATA
20 20 SSI-COV
SSI-COV
p-LSCF p-LSCF
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.5 1 1.5 2 2.5 3
Quality index Δe Quality index Δfreq
a) b)
Figure 7.87 – Percentage of considered events with the evolution of: a) ∆ index; b) ∆ index
The values computed for the second quality index ∆ , referred to the quality of the adjust on the
lowest frequency range of the spectrum, were also initially assessed. The percentage of events
considered with the evolution of the ∆ index is presented in Figure 7.87 b). It was concluded that
most results contain an increase of the energy in the frequency range considered for this index (0.1 –
0.35 Hz). It was thus noticed that the biggest problem associated with the bad estimation of this part of
the spectrum is related to the unrealistic increase of the computed stress levels of the structure. It is
interesting to note that the quality of the results provided by the alternative identification methods is
inverse to the quantified by the index ∆ . Nevertheless the difference is not very large.
Based on this preliminary analysis of the quality indices, a strategy was defined to accurately disregard
bad estimations of the fatigue damage. In that sense, it was decided to use the ∆ index to identify and
dismiss events in which the defined forward innovation model was clearly not capable to estimate the
acceleration time series as a sum of contributions of vibration modes/ harmonics. A value of ∆ = 0.20
was thus chosen, since it considers almost all events and, at the same time, imposes a limit that
guarantees the quality of the estimates. The maximum admissible value of the ∆ was defined to
admit a maximum error ∆ = 0.90. With this value, a rate of considered events of more than 90 %
was achieved when the combination of the best estimations obtained from the three algorithms is
computed.
At this point, the fatigue damage estimation from the recorded acceleration time series can be
computed. Figure 7.88 shows the evolution of accumulated damage calculated with the SSI-DATA
algorithm at the welding-on flanged connection between the tower and the foundation using the S-N
curve from the GL standard. From this figure, two periods of time with a rapid increase of damage are
noticed. The first period corresponds to events occurred during November 2013, while the second
occurs in February 2014. It is interesting to note that these two periods correspond to two long periods
of consecutive data sets with high mean wind speeds.
328
Chapter 7
It is also interesting to confirm that the most damaged sectors are the 100º-120º (and its symmetric
280º-300º) which is in accordance with the main wind direction of the wind (around 110º, see Figure
7.15).
-4
x 10
8
[0º-20º]/[180º-200º]
7 [20º-40º]/[200º-220º]
[40º-60º]/[220º-240º]
6
Accumulated Damage
[60º-80º]/[240º-260º]
5 [80º-100º]/[260º-280º]
[100º-120º]/[280º-300º]
4 [120º-140º]/[300º-320º]
[140º-160º]/[320º-340º]
3
[160º-180º]/[340º-360]º
2
0
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul
2013 2013 2013 2013 2013 2013 2014 2014 2014 2014 2014 2014 2014
Time
Figure 7.88 – Time history of the accumulated damage at the foundation level (with S/N curve from GL standard) using
the methodology based on the SSI-DATA algorithm
A direct extrapolation of the accumulated damage to the expected life of the wind turbine (20 years)
can be performed according to equation (7.16). The results obtained with the three methods are
presented in Table 7.22. As expected, the values obtained with the S-N curve from the EC3 code are
slightly lower due to the consideration of the threshold referred in section 6.3.2 (Figure 6.7).
Table 7.22 – Direct extrapolated accumulated damage at the end of the design life
SSI-DATA SSI-COV p-LSCF
GL 0.0420 0.0422 0.0213

EC3 0.0319 0.0339 0.0195
Success rate [%] 76.8 64.0 46.7
The results obtained present very low values of accumulated damage. Nevertheless, some important
aspects should be highlighted. Firstly, it should be noted that the estimated damage is only referred to
the dynamic component of the bending stress. Large loading cycles from the quasi-static frequency
regime, which might have important contributions to the fatigue damage of the structure are not
considered in these results. Two additional causes for this low accumulated damage should be also
referred. The location of the Senvion MM82 wind turbine is considered a low-turbulence site, which
means that high amplitude stress variations are not expected to occur frequently. This aspect has a
direct impact on the estimation of the fatigue life of the analysed structural detail, since it only depends
on the measured amplitude of the stress cycles and not on the mean stress level. In addition, technical
problems occurred at the beginning of October, preventing the operation of the dynamic monitoring
system during this month and almost the entire month of November. Thus, considering the detected
high rate of damage during the end of November, it is very likely the occurrence of important events
329
during this period that were not measured, which may have masked the real value of accumulated
damage at the end of the monitoring period.
Considering the confidentiality usually associated with projects related to the measurement of real
fatigue damage values, it is not easy to frame the obtained results. Nevertheless, the work presented in
(Loraux and Brühwiler, 2015), related to the fatigue analysis of a wind turbine tower using strain
gauges, highlights the conservatism of the current codes, based on a preliminary assessment of
experimentally obtained accumulated fatigue damage. Although the results are not comparable, it is
interesting to note that, even when the static, quasi-static and dynamic responses are taken into
account, a considerably higher fatigue lifetime than the design value of 20 years is expected.
The direct extrapolated estimation of total accumulated damage due to the dynamic component of the
bending stress shows a considerably variation, with values between 0.0420 and 0.0213 (using the S-N
curve from the GL standard) and 0.0319 and 0.0195 (using the S-N curve from the EC3 standard). This
discrepancy is mainly related to the different acceleration time series considered in each algorithm for
fatigue estimation (due to the quality indices restrictions), leading to different results at the end. This
situation is partially shown by the different success rate presented at the last line of the table. The
methodology using the SSI-DATA presents a higher rate. These values are comprehensible since the
original methodology is based on the SSI-DATA, while the other two are adaptations of the method.
The methodology based on the p-LSCF presents the worst success rate probably because it requires the
manipulation of the state-space matrices to remove the unstable poles.
One consequence of using different acceleration events to extrapolate the accumulated damage at the
end of the expected fatigue life of the structure is that different wind loading conditions are
considered. Figure 7.89 a) shows the number of events for each mean wind speed in the sector 100º -
120º (corresponding to the main wind incidence direction) during one year using the data recorded by
the SCADA system. Alongside, Figure 7.89 b) presents the probability density function (PDF) of the
same data. In the same figures, the distribution of the events considered in the fatigue estimation with
each algorithm is also shown. It is seen that, while the events considered for the SSI-DATA algorithm
seems to follow the wind speed distribution, the other two methods diverge considerably. The SSI-
COV shows an increase in the occurrence of events in the low (lower than 7 m/s) and high wind
speeds (higher than 15 m/s), while there is a notorious lack of events in the range 7 - 15 m/s. On the
other hand, the events considered with the p-LSCF algorithm focus on the low wind speed region.
These facts contribute to the discrepancy in the results presented in Table 7.22.
330
Chapter 7
500 25 %
1 year 1 year
SSI-DATA SSI-DATA
Percentage of occurrence [%]

400 SSI-COV 20 % SSI-COV
p-LSCF p-LSCF
Nº of events
300 15 %
200 10 %
100 5%
0 0
0 5 10 15 20 25 0 5 10 15 20 25
a) b)
Figure 7.89 – Histogram a) and probability density function b) of mean wind speed (in the sector 100º-120º) during one
year according to the data recorded by the SCADA system and to the data considered for fatigue estimation with the
three algorithms
A simple way to attest the quality of the results obtained with the three algorithms is the computation
and comparison of damage estimated from the same acceleration time series events. In that sense, the
results obtained with the SSI-DATA were considered as reference, since they consider more data sets,
and only the events that were both considered for the other algorithms and the SSI-DATA were used
to compute the estimated damage. The accumulated damage obtained with SSI-COV and p-LSCF
algorithm is directly compared to the one computed with the SSI-DATA since they are related to the
same acceleration events. Table 7.23 shows the results obtained with this study, presenting the
deviation obtained relative to the reference method (SSI-DATA algorithm). As can be seen, the
agreement is now much better, especially for the estimation based on the SSI-COV where an error
lower than 8 % was achieved.
Table 7.23 – Deviation of accumulated damage of coincident events relative to the methodology based on SSI-DATA
algorithm
GL - + 7.5 % - 19.8 %
EC3 - + 7.8 % - 20.6 %
The advantage of using different algorithms to estimate the fatigue damage for the same period of time
is the ability to choose the best estimations obtained with each method for each acceleration time
series event according to the quality indices. The results obtained for the direct extrapolated damage
(according to the S-N curve from the GL and EC3 standards) using the best estimations from the three
algorithms are shown in Table 7.24. It is visible the increase of the success rate to 91.0 %.
331
Table 7.24 – Estimation of accumulated damage at the end of design life considering a direct extrapolation and an
adjusted extrapolation according to the PDF of the wind speed using the best estimations from the three algorithms
Direct extrapolation Adjusted to wind speed PDF
GL 0.0371 0.0357
EC3 0.0283 0.0271
Success rate [%] 91.0
Another advantage of using different algorithms to estimate the fatigue damage for the same period of
time is the possibility of consider more loading scenarios and, consequently, approximate the
considered wind conditions to the actual distribution of wind conditions. This advantage is illustrated
in Figure 7.90, where the PDF of the mean wind speed in sector 100º - 120º obtained with the data
considered when the best estimations from the three algorithms were used is compared with the PDF
defined with the data recorded by the SCADA system during one year. Both PDFs seem to be similar
enough to allow an adjustment of the results obtained with the best estimations to follow the “real”
distribution of wind speed (as introduced in section 7.2.2.1). In this case, it was only possible to
quantify the damage accumulation associated with the mean wind speed of 10 min. period. The results
of accumulated damage at the end of the design life after adjusting the loading scenarios according to
the PDF of the mean wind speed are shown in Table 7.24.
12
1 year
10 Best Estimation
0
0 5 10 15 20 25
Wind speed [m/s]
Figure 7.90 – Probability density function of mean wind speed (in the sector 100º-120º) during one year of data recorded
by the SCADA system (1 year) and the events included in the best estimation obtained with the three algorithms
Naturally, the estimation of damage due to the dynamic component of fatigue would improve if more
data from the monitoring system was available or, in case it is not possible, if a characterization of the
wind conditions over a longer period was known.
7.3.8 OPTIMIZATION OF THE MONITORING SYSTEM

The installation of the dynamic monitoring system is desirable if it adds value to the investment. As
referred in section 6.4.2, nowadays wind turbines are already equipped with a bi-directional
accelerometer at the nacelle level as part of its own control system. Thus, it would be interesting to
assess the level of accuracy that can be achieved with a dynamic monitoring system with a reduced
332
Chapter 7
number of sensors, both for modal tracking and fatigue assessment. At the most extreme solution, it
would be interesting to only use the top sensor as part of the monitoring system.
This section introduces the results obtained with the two modules of the monitoring system using
three different solutions of sensors layout. The first solution uses only one biaxial sensor at the top
level (+74.988 m) to perform both modal tracking (for damage detection) and fatigue analysis. This
represents the aforementioned most interesting solution from the economic point of view. The second
proposed solution uses only one biaxial sensor at the +48.392 m level of measurement (around 2/3 of
the tower height). Lastly, the third solution is based on the use of two biaxial sensor placed at +74.988
m and at +48.392 m levels.
7.3.8.1 Layout 1 (+74.988 m level)

The option based on the use of only the top level of measurement (sensors S8 and S9) is the most
interesting solution. Since wind turbines are nowadays equipped with a bi-directional accelerometer
sensor at this position, the use of this sensor for damage detection and fatigue assessment of the
support structure would represent an important benefit without any relevant investment.
Although interesting from an optimization point of view, this solution presents an important
drawback: due to the very low modal amplitude of the 2nd pair of support structure bending modes at
this level (2 FA and 2 SS modes from Figure 7.40), it is not possible to accurately identify and track
these two modes. This handicap has consequences for both modules of the dynamic monitoring
system. In the context of damage detection, this optimization solution prevents the tracking of the
most interesting indicators of foundation damage. This situation will necessarily influence the degree
of precision in the detection of small damages at the foundation level. Moreover, it is not possible to
consider these two modes for fatigue assessment, since it is very difficult (or even impossible) to use a
very small modal amplitude as a reference value to scale a mode shape.
Modal Tracking and Damage Detection

The acceleration data recorded by the sensors at the top level was used to perform the tracking of the
vibration modes. The strategy presented in section 7.3.6.2 was followed to process to the recorded data
during the monitored period. The same input parameters for the three modal identification algorithms
were used and the same criteria to assess the stability of the poles (Table 7.5) were followed. However,
instead of a minimum MAC value of 0.80 in order to consider a cluster of poles to the tracking
procedure, a minimum value of 0.50 was defined. This value was chosen due to the poor definition
that is achieved with only one bi-directional sensor (representing only one modal amplitude per
direction) and corresponds to a differentiation between modes in the FA and SS direction.
Figure 7.91 illustrates the Campbell diagram obtained after the tracking procedure. From this figure, it
is visible that, as expected, the 2 FA and 2 SS modes are badly estimated and, for that reason, should be
disregarded. On the other hand, it is interesting to note that the influence of the harmonics is still
disregarded after the tracking procedure. This fact is due to the complex nature of the operational
deflection shape exhibited by poles referred to the harmonics, in contrast to the more real nature of the
vibration modes.
333
Campbel Diagram
4.5
4 FA
4 3 SS*
3.5 3 FA/SS
3 2 SS
Frequency [Hz]
2.5 2 FA
2 2 SS*
1.5
1 1 SS*
0.5 1 SS
1 FA
0
0 2 4 6 8 10 12 14 16 18
Rotor Speed [RPM]
Figure 7.91 – Campbell diagram with the tracked vibration modes with Layout 1 (SSI-COV algorithm)
The statistics related to the success in the identification of the 9 vibration modes are introduced in
Table 7.25. Comparing these results with the ones obtained when all sensors are used (Tables 7.6 to
7.8), it is concluded that the success rate of the first 4 modes have increased when only the top sensors
are considered. Although this might seem a contradiction, it is explained by the lower MAC condition
imposed to consider a cluster of poles. With this MAC condition, it is only possible to distinguish
between FA and SS modes which are, in fact, the only information that can be extracted from the mode
shapes with this sensors layout. The coefficient of variation of the frequency values of these modes
does not present an important increase relative to the solution using all sensors, which is a good
indicator for their use in damage assessment.
As regards the last 3 modes, the success in their identification was rather low and, for that reason, these
modes were not considered for damage detection.
Table 7.25 – Success rate and coefficient of variation obtained with the identification algorithms (with Layout 1)
SSI-COV SSI-DATA p-LSCF

Mode Success Rate Success Rate Success Rate
. . . . . .
[%] [%] [%]
1 SS 93.8 0.005 95.5 0.005 86.3 0.005
1 FA 86.6 0.022 87.3 0.019 43.3 0.018
1 SS* 90.5 0.037 92.0 0.037 56.5 0.035
2 SS* 96.9 0.031 96.9 0.031 93.6 0.029
2 FA - - - - - -
2 SS - - - - - -
3 FA/ 3SS 31.7 0.012 39.7 0.018 42.8 0.012
3 SS* 19.3 0.022 29.6 0.019 30.6 0.019
4 FA 27.5 0.017 56.7 0.019 84.8 0.019
The ability of the dynamic monitoring system to detect damage was then assessed using the first pair
of bending modes (1 FA and 1 SS) and the rotor blades modes 1 SS* and 2 SS*. Initially, the same
334
Chapter 7
damage scenarios shown in Table 7.21 (for a sensor layout with all sensors) were tested. It was verified
that the damage scenario D1, related to the scour problems in an offshore wind turbine, was not
properly detected. This scenario is especially relevant to the second pair of bending modes which are
not detected with this solution. For that reason, a new scenario D1* related to scour damage was
defined, reflecting the minimum detectable frequency variation of the first pair of modes. It was
concluded that, in order to be detectable, the scenario D1* would have to imply a frequency variation
of -0.50 % of the first pair of support structure bending modes (an increase of 2.5 times when
compared to the original D1 scenario). In terms of scour damage, the new damage scenario D1*
represents a depth of 0.162 times the base diameter of the monopile, i. e., a depth of 0.70 m and 0.82 m
if the Senvion MM82 tower base and the Vestas V90 monopile are used as reference, respectively.
Although this value is more than twice the value achieved when all sensors are used, it still represents a
very early state of damage, being considerably lower than the recommended design value of 1.30.
(Det Norske Veritas (DNV), 2011).
The second damage scenario D2, reproducing a frequency variation 5 times smaller than the one
described in a test of damaged onshore wind turbines (Pelayo, López-Aenlle et al., 2011), was also
analysed and it was concluded that, even with this layout solution, it would be detected. In fact, the
variation imposed by this damage scenario for the first pair of tower bending modes is similar to the
one observed in the scenario D1*.
The last damage scenario (D3) is related to blade damage. It was concluded that, for the same imposed
stiffness variation described in section 7.3.6.6, the damage is still detected with this layout solution.
Table 7.26 – Variation of the natural frequencies of the tracked modes associated with the damage scenarios
Δ [%]
Modes D1* D2 D3
1 SS -0.50 -0.50 0.00

1 FA -0.50 -0.50 0.00
1 SS* 0.00 0.00 -0.65
2 SS* 0.00 0.00 -0.65
2 FA - - -
2 SS - - -
3 SS* - - -
4 FA - - -
Figure 7.92 introduces the control charts obtained for the referred damage scenarios with Layout 1. As
can be seen, the damage is easily detected which attest the feasibility of this solution. Although some
accuracy was lost in the damage scenario D1, it is seen that it provides a good solution for detection of
damage at an early stage.
335
Undamaged scenario D1* / D2

300 250
250
200
200
150
T2
T2
150
100
100
50
50
0 0
Time Time
D3
250
200
150
T2
100
50
Time
damage scenarios (Layout 1)
Fatigue Results
The estimation of fatigue damage using a single level of measurement is an interesting application for
the methodology introduced with the virtual sensors in section 6.4.2. With this methodology, it is
possible to estimate the response of the wind turbine support structure at any point of the structure,
using only a (“well positioned”) single reference measurement point. In the case of the Layout 1, this
reference point is the top measurement level. However, as was previously referred, this measurement
position cannot be considered as “well positioned” since it coincides with a level of very low modal
amplitude of an important vibration mode. Thus, it is not possible to accurately estimate the mode
shape of this mode. In that sense, it was decided to despise all vibration modes and harmonics with a
frequency value higher than 2 Hz. With this imposition, not only the second pair of bending modes is
disregarded, but also some integer multiples of the harmonics are avoided since their operational
deflection shapes also present a very low modal amplitude at the top of the tower.
The importance of the second pair of tower bending modes in the context of the dynamic behaviour of
the wind turbine structure was already described in section 7.3.6.4. It is thus expected that, using this
sensor layout, the estimation of the acceleration at the middle and lower part of the tower will be
underestimated. Figure 7.93 illustrates this problem, using the acceleration time series collected by the
S8 and S9 sensors during a production period (rotor speed = 12.4 rpm). The acceleration signals were
then processed by the methodology based on the p-LSCF algorithm. The top figures compare the
336
Chapter 7
original signal recorded by the sensor in the FA direction with the estimated signal as a sum of modes/
harmonics at the measured level in time and frequency domain. On the other hand, the plots at the
middle and bottom of Figure 7.93 are referred to the estimations of the acceleration at unmeasured
levels. Taking advantage of the non-used sensors (at levels +21.722 m and +48.392 m), the estimated
acceleration time series at unmeasured locations are compared with the real measured time series.
+74.988 (FA) - measured 2 +74.988 (FA) - measured
0.2 10
Original Original
0.15 1
Estimated 10 Estimated
0.1
0.05 0
Acc. [m/s2]
Amplitude
10
0
-1
-0.05 10
-0.1 -2
10
-0.15
-3
-0.2 10
0 100 200 300 400 500 600 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
+48.392 (FA) - unmeasured 2 +48.392 (FA) - unmeasured

0.2 10
Original
0.15 1
Estimated 10
0.1
0.05 0
Acc. [m/s2]
Amplitude
10
0
-1
-0.05 10
-0.1 -2
10
-0.15 Original
-3 Estimated
-0.2 10
0 100 200 300 400 500 600 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
+21.772 (FA) - unmeasured 2 +21.772 (FA) - unmeasured

0.1 10
Original
Estimated
0.05
0
10
Acc. [m/s2]
Amplitude
-2
10
-0.05
Original
-4 Estimated
-0.1 10
0 100 200 300 400 500 600 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 7.93 – Original and estimated acceleration time series at the measured level (+74.988 m) and unmeasured levels
(+48.392 m and +21.772 m) in the FA direction
The observation of the plots in Figure 7.93 clearly shows that, as expected, the disregard of the energy
contribution from the spectrum for frequency values higher than 2 Hz leads to an important
underestimation of the acceleration at the lowest 2/3 of the tower. Nevertheless, it is seen that, for the
337
considered frequency range, the agreement between the estimated and the real acceleration signal is
quite good. Even the contribution of the 3Ω harmonic (represented by the large peak around 0.62 Hz),
which requires the estimation of its operational deflection shape, shows a good agreement.
The processing for fatigue damage estimation was then applied to the estimated acceleration signals. In
order to illustrate the quality of the results obtained with this layout configuration, the dynamic
component of the bending stress from a 10 min. period acceleration event at the foundation level was
computed. The results were compared with the stress estimation obtained with the methodology based
on the SSI-DATA considering the three levels of measurement, since it is considered the most accurate
of the implemented procedures. The bending stress computed with the two procedures is shown in
Figure 7.94. It is visible that the bending stress is underestimated mostly due to the non-consideration
of the energy of the spectrum for frequency values higher than 2 Hz. Computing the damage
associated with the two estimations (using the S-N curve from the GL standard), it is concluded that
the value obtained with the procedure based on the p-LSCF with Layout 1 is 3.2 times lower than the
value achieved with the SSI-DATA methodology using all sensors.
2 9
10
SSI-DATA (all sensors
1.5 SSI-DATA (all sensors
p-LSCF (Layout 1) 8
10 p-LSCF (Layout 1)
1
7
0.5 10
Amplitude
6
0 10
-0.5 5
10
-1
4
10
-1.5
3
-2 10
0 100 200 300 400 500 600 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 7.94 – Estimated dynamic bending stress at the foundation level due to motion in the FA direction (the vertical
dashed lines indicate the disregarded initial and last part)
The same quality indices introduced in section 7.3.7 (∆ and ∆ ) were used to despise erroneous
estimations of fatigue damage. Naturally, these indices are only referred to the results associated with
the considered S8 and S9 sensors. Figure 7.95 illustrates the evolution of percentage of acceleration
events with the increase of the indices. Comparing these results with the ones presented in Figure 7.87,
it is visible with the index ∆ that the errors associated with the quality of estimation of the
acceleration time series are higher. Qualitatively, it is seen that the procedure based on the SSI-DATA
algorithm shows the best results, followed by the SSI-COV and, lastly, the p-LSCF, just like when all
sensors were used. In relation to the index ∆ , the p-LSCF algorithm shows, again, the best results.
However, with this sensor layout solution, the difference to the other methods is higher than when all
sensors are used.
338
Chapter 7
100 100

80 80
60 60
40 40
SSI-DATA SSI-DATA
20 SSI-COV 20 SSI-COV
p-LSCF p-LSCF
0 0
0 0.1 0.2 0.3 0.4 0.5 0 1 2 3 4
a) b)
Figure 7.95 – Percentage of considered events with Layout 1 with the evolution of: a) ∆ index; b) ∆ index
As referred in section 7.3.7, the comparison of the direct extrapolated damage obtained with each
methodology does not lead to a correct estimation of the quality of the results since it compares
estimations of damage from different events. It is visible in the probability density function of Figure
7.96 a) that the considered events for each methodology differ considerably from each other.
20 12
1 year 1 year
SSI-DATA 10 Layout 1
SSI-COV
15
p-LSCF
8
10 6
4
5
2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
a) b)
Figure 7.96 – Probability density function of wind speed (in the sector 100º-120º) recorded by the SCADA system and
from the data considered for fatigue estimation with Layout 1 with: a) the thee algorithms; b) best estimations from the
thee algorithms
Thus, the estimated damage obtained with each methodology was compared with the damage
estimated with the SSI-DATA using all sensors (considered the reference methodology) for the same
events. The deviations obtained with the three methodologies are presented in Table 7.27.
339
Table 7.27 – Deviation of accumulated damage of coincident events between results with Layout 1 and the methodology
based on SSI-DATA algorithm using all sensors
SSI-DATA (All sensors) SSI-DATA SSI-COV p-LSCF
GL - -67.4 % -69.3 -41.2 %

EC3 - -68.9 % -71.1 -42.0 %
Success rate [%] 76.8 60.7 37.4 26.7
From the results, it is evident that the damage estimations with Layout 1 were not very good. The
damage obtained with the three methodologies is clearly underestimated. Also the success rate
achieved is low, with only one methodology (based on the SSI-DATA) showing a rate higher than 50
%.
Lastly, the fatigue damage using the best estimations from the three algorithms was computed. With
this estimation, the distribution of considered wind speed events is close to the distribution obtained
with the SCADA data (Figure 7.96 b)). The results are introduced in Table 7.28, where they are
compared to the ones obtained with the best estimation using all sensors. Again, the values are
considerably underestimated.
Table 7.28 – Estimation of accumulated damage at the end of design life considering an adjusted extrapolation according
to the PDF of the wind speed using the best estimations from the three algorithms
All sensors Layout 1 Error
GL 0.0371 0.015 -57.8 %

EC3 0.0283 0.013 -52.4 %
Success rate [%] 91.0 79.4 -
7.3.8.2 Layout 2 (+48.392 m level)

Layout 2 represents an optimization of Layout 1. While the use of the top level as single level of
measurement showed that it is not possible to consider the 2nd pair of bending modes, Layout 2
optimizes the selection of the single level of measurement. It was decided to use the measurement level
at around 2/3 of the tower height (+48.392 m), since it (theoretically) enables the identification of all
the important vibration modes. Looking to Figure 7.40, it is visible that none mode shape presents a
reduced modal amplitude at this level.

The same procedure described in section 7.3.6.2 was followed using the acceleration data collected by
sensors S6 and S7 (at level +48.392 m). Just like in the case of Layout 1, a minimum MAC value of 0.50
was imposed during the modal tracking in order to distinguish between FA and SS modes. The
Campbell diagram defined by the results obtained with the modal tracking procedure based on the p-
LSCF algorithm is presented in Figure 7.97. From this figure, it is visible the good quality of the
identification of the second pair of tower bending modes, confirming the suitability of the
measurement location. On the other hand, it is visible that the 15Ω harmonic hinders the identification
340
Chapter 7
of the 3 SS* mode. In fact, some clusters referred to this harmonic are misidentified as being from this
mode. As for the other vibration modes, no further erroneous modal identifications were detected.
Campbel Diagram
4.5
4 FA
4 3 SS*
3.5 3 FA/SS
3 2 SS
Frequency [Hz]
2.5 2 FA
2 2 SS*
1.5
1 1 SS*
0.5 1 SS
1 FA
0
0 2 4 6 8 10 12 14 16 18
Rotor Speed [RPM]
Figure 7.97 – Campbell diagram with the tracked vibration modes with Layout 2 (p-LSCF algorithm)
The success rate in the identification of the 9 vibration modes is summarized in Table 7.29. It is
concluded that the results are good for the 1 SS, 1 SS*, 2 SS*, 2 FA and 2 SS modes, with high rates of
success in the tracking process. On the other hand, the identification of the 1 FA presents a very low
success rate, especially for the p-LSCF.

. . . . . .
[%] [%] [%]
1 SS 90.0 0.005 94.8 0.005 64.6 0.005
1 FA 46.0 0.018 61.0 0.019 18.6 0.016
1 SS* 48.7 0.040 84.7 0.037 29.6 0.034
2 SS* 84.6 0.031 93.4 0.031 72.5 0.029
2 FA 90.6 0.009 92.1 0.009 90.3 0.006
2 SS 73.5 0.012 83.3 0.011 72.8 0.010
3 FA/ 3SS 7.2 0.017 12.2 0.016 18.4 0.016
3 SS* 55.2 0.019 61.6 0.021 63.9 0.018
4 FA 30.6 0.018 76.2 0.018 96.2 0.018
After the modal tracking, the environmental and operational effects were removed from the frequency
values, using a multivariate analysis (similar to the procedure described in section 7.3.6.5). Thus, using
the frequency values as damage indicators, the three damage scenarios presented in section 7.3.6.6
(considering the dynamic monitoring system with three levels of measurement) were tested (see Table
7.21).
341
The control charts obtained are shown in Figure 7.98. It is evident that the three damage scenarios are
well identified. Notwithstanding, the detection of damage referred to scenario D1 is not as clear as
when all sensors are used.
300 300
250 250
200 200
T2
T2
150 150
100 100
50 50
0 0
Time Time
D2 D3
300 250
250
200
200
150
T2
T2
150
100
100
50
50
0 0
Time Time
Figure 7.98 – ܶ ଶ control charts associated with a scenario where no damage is present and with the three defined
Fatigue Results
Considering the position of the selected measurement level, it is expected that the problems noticed
during the estimation of the acceleration time series with Layout 1 would disappear with this solution.
In that sense, the frequency cut of 2 Hz defined with the Layout 1 was not imposed.
An example of the estimation of the acceleration time series obtained during production regimes (with
a mean rotor speed of 11.7 RPM) in introduced in Figure 7.99. In this figure, the estimation of the
acceleration at the measured level (+48.392 m) is presented alongside with the estimations at two
unmeasured levels. Again, the +74.988 m and +21.772 m levels were selected in order to be possible to
compare the estimations with the real acceleration time series. In this example, the methodology based
on the SSI-COV algorithm was used. The very good agreement of the estimations at the unmeasured
spots is notorious.
342
Chapter 7
+74.988 (FA) - unmeasured +74.988 (FA) - unmeasured

0.15 0.15
Original Original
0.05 0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05 -0.05
-0.1 -0.1
-0.15 -0.15
0 100 200 300 400 500 600 105 110 115 120 125 130
Time [m] Time [m]
+48.392 (FA) - measured +48.392 (FA) - measured

0.2 0.2
Original Original
0.15 0.15
Estimated Estimated
0.1 0.1
0.05 0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05 -0.05
-0.1 -0.1
-0.15 -0.15
-0.2 -0.2
0 100 200 300 400 500 600 105 110 115 120 125 130
Time [m] Time [m]

0.08 0.08
Original Original
0.06 0.06
Estimated Estimated
0.04 0.04
0.02 0.02
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-0.08 -0.08
0 100 200 300 400 500 600 105 110 115 120 125 130
Time [m] Time [m]
Figure 7.99 – Original and estimated acceleration time series at the measured level (+48.392 m) and unmeasured levels
(+74.988 m and +21.772 m) in the FA direction
Naturally, the same quality indices were used to discharge erroneous estimations. It is noted that,
qualitatively, the results are similar to the ones obtained with Layout 1. Nevertheless, there is an
important improvement in the index ∆ for the methodology based on the SSI-COV algorithm.
343
100 100

80 80
60 60
40 40
SSI-DATA SSI-DATA
20 SSI-COV 20 SSI-COV
p-LSCF p-LSCF
0 0
0 0.1 0.2 0.3 0.4 0.5 0 1 2 3 4
a) b)
The considered events for each methodology are again not enough to independently reproduce the
density function of the wind speed, as seen in Figure 7.101 a). Thus, the same strategy previously
followed to compare only the coincident events between the reference results (methodology based on
SSI-DATA, using all sensors) and the various methods using Layout 2 was followed. The results are
summarized in Table 7.30.
20 12
1 year 1 year
SSI-DATA 10 Layout 2
Percentage of Occurrence [%]
SSI-COV
15
p-LSCF
8
10 6
4
5
2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
a) b)
thee algorithms
The results obtained confirm the good premise indicated by the quality of the estimations. Both the
methodologies based on the SSI-DATA and SSI-COV show damage estimations very close to the ones
obtained with the reference results. The success rate achieved with the SSI-COV algorithms should
also be highlighted, which is considerably higher than the with SSI-DATA. This is an indication of the
quality of the adaptation of the post-processing tools presented in section 5.5 (originally developed to
the SSI-DATA) to the SSI-COV algorithm. Again, the p-LSCF presented the worst results, both for the
success rate and damage estimation.
344
Chapter 7
Table 7.30 – Deviation of accumulated damage of coincident events between results from Layout 2 the methodology
GL - -8.1 % -1.8 % -47.8 %

EC3 - -7.7 % +2.3 % -49.4 %
Success rate [%] 76.8 46.5 69.4 44.7
Lastly, the best estimations from the three algorithms were selected to estimate damage. With this
consideration, the success rate increased to values similar to the ones obtained when the three levels of
measurement are considered. At the same time, the distribution of the wind speed becomes similar to
the one recorded by the SCADA system during one year (Figure 7.101 b)). Under these circumstances,
the estimated damage is again close to the estimation using all sensors, with an error lower than 15 %.
Considering the sensitivity of the S-N curve to small variations of the load cycles and the reduced
number of recording channels used with this solution, this result has to be considered a very good
approximation.
GL 0.0371 0.0305 -14.5 %

EC3 0.0283 0.0233 -14.1 %
Success rate [%] 91.0 91.0 -
7.3.8.3 Layout 3 (+74.988 m and +48.392 m levels)

Lastly, Layout 3 is a natural consequence of the two previous solutions. Layout 1 showed to be
impractical for fatigue damage estimation, while Layout 2 demonstrated that it is a good choice for a
single measurement level. However, opting for Layout 2 requires the installation of an additional level
of measurement, while the top sensors are already installed. This means that the additional cost of
using the Layout 2 or a solution based on the use of both levels of measurement is the same. For that
reason, Layout 3 explores this possibility. It consists on the use of the two levels of measurement: at the
top (+74.988 m) and around 2/3 of the tower height (+48.392 m).

Again, the procedure previously described for the other solutions was applied to this case. The
acceleration data recorded by sensors S6, S7 S8 and S9 was used as input for the monitoring system.
The same input parameters described in section 7.3.6.2 were implemented. Also, since two levels of
measurement enable a good level of detail of the mode shapes, a minimum MAC value of 0.80 was
imposed to consider a cluster for the tracking procedure (the same value used when all sensors were
considered).
345
The Campbell diagram obtained for the analysed period of data with this layout solution is shown in
Figure 7.102. The results are in good agreement with the ones obtained when all sensors were used.
However, it is noted a small influence of the 15Ω harmonic on the identification of the 3 SS* vibration
mode, although in a smaller extent than in Layout 2.
Campbel Diagram
4.5
4 FA
4 3 SS*
3.5 3 FA/SS
3 2 SS
Frequency [Hz]
2.5 2 FA
2 2 SS*
1.5
1 1 SS*
0.5 1 SS
1 FA
0
0 2 4 6 8 10 12 14 16 18
Rotor speed [RPM]
Figure 7.102 – Campbell diagram with the tracked vibration modes with Layout 3 (p-LSCF algorithm)
The statistics about the identification of the 9 vibration modes are summarized in Table 7.32. A
significant improvement is clearly noticed in the success rate of the identification of almost all
vibration modes. In fact, for some modes, the results are better than the ones obtained when all sensors
are used. Nevertheless, the identification of the 1 FA mode by the p-LSCF algorithm is slightly lower
than when all sensors are considered.

. . . . . .
[%] [%] [%]
1 SS 94.2 0.005 95.9 0.005 90.6 0.005
1 FA 82.7 0.021 80.2 0.019 59.7 0.020
1 SS* 80.7 0.038 89.3 0.037 85.9 0.037
2 SS* 93.9 0.031 95.4 0.030 93.3 0.030
2 FA 91.1 0.009 92.2 0.008 90.5 0.006
2 SS 74.1 0.012 82.3 0.011 69.4 0.010
3 FA/ 3SS 12.4 0.013 17.6 0.012 48.3 0.010
3 SS* 44.1 0.018 56.6 0.018 81.8 0.018
4 FA 19.3 0.019 56.8 0.020 97.0 0.019
Considering the quality of the results achieved, the same damage scenarios defined in section 7.3.6.6
were also tested for this layout solution (see Table 7.21).
346
Chapter 7
The control charts obtained for the three damage scenarios are exposed in Figure 7.103. It is visible
that the damage events are clearly identified, with results in line with the ones presented for a layout
solution based on all sensors.
300 300
250 250
200 200
T2
T2
150 150
100 100
50 50
0 0
0 0 50 100 150
Time Time
D2 D3
250 200
200
150
150
T2
T2
100
100
50
50
0 0
0
Time Time
Fatigue Results
Given the distribution of sensors in Layout 3 and the good results achieved with the modal tracking, it
is expected that the estimation of the acceleration time series at unmeasured locations would be also
good. Just like in the Layout 2, the frequency cut imposed for Layout 1 was not implemented in this
solution.
The good estimation of the acceleration time series at unmeasured locations can be attested with
Figure 7.104, referred to an event during a production period (with a mean rotor speed of 14.1 RPM).
In this figure, the estimated accelerations at measured (+74.988 m and +48.392 m) and unmeasured
(+21.772) positions in the FA direction are compared with the recorded time series by the sensors at
the same positions. For this example, the methodology based on the SSI-DATA algorithm was used. It
is visible that the estimated acceleration time series at the unmeasured position is very good.
347

0.15 0.15
Original Original
0.05 0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05 -0.05
-0.1 -0.1
0 100 200 300 400 500 600 100 105 110 115 120 125
Time [s] Time [s]

0.3 0.3
Original Original
0.1 0.1
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.1 -0.1
-0.2 -0.2
0 100 200 300 400 500 600 100 105 110 115 120 125
Time [s] Time [s]

0.15 0.15
Original Original
0.05 0.05
Acc. [m/s2]
Acc. [m/s2]
0 0
-0.05 -0.05
-0.1 -0.1
0 100 200 300 400 500 600 100 105 110 115 120 125
Time [s] Time [s]
Figure 7.104 – Original and estimated acceleration time series at the measured levels (+74.988 m and +48.392 m) and
unmeasured levels (+21.772 m)
Figure 7.105 evidences the evolution of the number of events with the increase of the two considered
quality indices ( and ) referred to the 4 used sensors. The results obtained with this layout
solution are similar to the ones achieved with all sensors.
348
Chapter 7
100 100

80 80
60 60
40 40
SSI-DATA SSI-DATA
20 20
SSI-COV SSI-COV
p-LSCF p-LSCF
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.5 1 1.5 2 2.5 3
a) b)
The deviation associated with the direct comparison of the damage estimated with each methodology
(using Layout 3) and the damage estimated with the reference methodology (SSI-DATA algorithm,
using all sensors) for coincident events was again used to assess the quality of the results. The
deviations are presented in Table 7.33. It is seen that results obtained with the SSI-DATA algorithm
represent a very good agreement with the reference results, also showing a good success rate. Both the
other two algorithms show lower success rates. Notwithstanding, they both show a good agreement of
the damage estimation with the reference results, especially the methodology based on p-LSCF
algorithm.
30 0.12
1 year 1 year
SSI-DATA Layout 3
25 0.1
SSI-COV
p-LSCF
20 0.08
15 0.06
10 0.04
5 0.02
0 0
0 5 10 15 20 25 0 5 10 15 20 25
a) b)
thee algorithms
349
Table 7.33 – Deviation of accumulated damage of coincident events between results from Layout 3 the methodology
GL - +7.9 % -20.7 -13.3 %

EC3 - +9.8 % -20.7 -13.1 %
Success rate [%] 76.8 70.8 39.4 30.5
Using the best estimations provided by the three algorithms to increase the range of damage scenarios
included in the analysis, it is possible to extrapolate the damage expected at the end of the design
fatigue life according to the measured probability density function (PDF) of the mean wind speed.
Figure 7.106 compares the PDF of the mean wind speed measured for the sector 100º-120º for the
acceleration events considered when the best estimations from the three algorithms are used with the
PDF obtained with the data recorded by the SCADA system during one year. The results obtained with
the best estimation were then adjusted to follow the PDF of the mean wind speed obtained from the
SCADA system. The results obtained are summarized in Table 7.34. It is visible that the estimation
obtained with this layout solution is very similar to the estimated using all sensors (deviations smaller
than 7 %), while a good success rate is also achieved.
GL 0.0371 0.0374 +4.93 %

EC3 0.0283 0.0289 +6.87 %
Success rate [%] 91.0 86.8 -
7.3.9 CONCLUSIONS
This section presented a complete analysis of the installation and implementation of a vibration-based
structural health monitoring system in a 2.0 MW wind turbine. It starts with the presentation of the
main characteristics of the wind turbine, together with a description of its operating conditions.
Alongside, the modal results obtained with two developed finite element models are presented, serving
for an initial assessment of the wind turbine vibration modes.
Then, the dynamic behaviour of the wind turbine structure under operational and parked conditions
was analysed. The application of the implemented algorithms (SSI-COV, SSI-DATA and p-LSCF) for
modal identification was successfully employed, allowing to identify 9 vibration modes, among tower
and rotor blades modes. The continuous assessment of the wind turbine dynamic properties permitted
to completely characterize its modal parameters throughout the different operating regimes. The
developed methodology allowed to remove the influence of the harmonics in the identification of the
most important vibration modes.
The analysis of the collected data during more than one year permitted to identify 51 data sets
corresponding to free decay events. These acceleration events were analysed using two different
350
Chapter 7
methodologies: one based on the adjustment of an exponential window and another based on the SSI-
COV algorithm. The results obtained with both methods presented very coherent outcomes.
Nevertheless, the values obtained for the damping of the 1 FA mode showed some variability which
could not be fully explained.
An analysis about the relative contribution of the vibration modes and rotor harmonics to the
measured acceleration was also performed following the procedure described in section 5.5. It was
noticed that the participation of the modes/ harmonics on the measured signal is clearly dependent on
the operating conditions and on the tower section under analysis.
After this, the identified natural frequency values of the tracked vibration modes were used to assess
the influence of operational and environmental effects. In that sense, static and dynamic models were
tested. Among these, the static model SM2 provided the best results, demonstrating the ability of the
implemented methodology to minimize the influence of these external effects on the measured natural
frequencies.
Using the outcomes obtained with the application of the regression model to the identified natural
frequencies, the ability of the developed methodology to detect abnormal structural changes was
tested. It was demonstrated the accuracy of the system to identify small variations of the tower bending
modes along the yaw angle. This represents an important achievement since it allows to identify
possible asymmetric conditions of the foundation stiffness. In addition, the methodology also proved
being able to detect frequency variations associated with small damages on the foundation of onshore
and offshore wind turbines and, although with a smaller precision, on the blades.
The fatigue condition of the wind turbine tower was also assessed following the procedure introduced
in section 6.4.1.1. Due to the impossibility to install sensors to quantify the quasi-static motion of the
tower, it was only possible to estimate the condition of the structure due to the dynamic component of
fatigue. In that sense, the considerations made in this context only intended to validate the proposed
methodology and did not aim to analyse the real fatigue condition of the structure. It was
demonstrated that, for this scenario of study, the methodology based on the SSI-DATA and SSI-COV
algorithms delivered the best results. A strategy based on the use of the best estimation from the three
algorithms was also tested with good results.
Finally, an optimization of the dynamic monitoring system considering different sensors layout was
proposed and tested. The first solution, based on a biaxial sensor at the tower top, revealed some
deficiencies on the identification of some modes. Even though, it has proved successful in the
identification of small damages (although without the precision of the solution based on all sensors).
On the other hand, this layout performed very poorly for the fatigue assessment of the tower. Both
layout solutions 2 and 3, based on one and two biaxial sensors, respectively, evidenced the good ability
for detection of damage at an early stage and for fatigue assessment of wind turbine towers.
In short, this section proved the suitability of the developed vibration-based monitoring system to be
implemented in wind turbines. The usefulness of the damage detection module was clearly
demonstrated, while the component regarding the fatigue assessment of the tower is still at a
development stage, requiring additional testing. Nevertheless, the results already obtained permit to
anticipate the potential of this solution.
351
352
Chapter 8
8
CONCLUSIONS AND FUTURE
RESEARCH
8.1 CONCLUSIONS
This work explored the potential of using operational modal analysis in the monitoring of wind
turbines. In this scope, several processing routines were developed and implemented to extract the
maximum information that a monitoring system based on the dynamic response of wind turbines can
deliver. The developed methodology was implemented into a vibration-based monitoring system,
capable of operating continuously and in an automated way, that transforms the collected data into
useful information for the wind farm owners.
The thesis starts with a global perspective of the wind energy industry. An overview of the main
technical evolutions of wind turbines was presented, focusing on two crucial aspects that defined the
success of some models: reliability and cost effectiveness. Then, a description of the main structural
elements of wind turbines is made, with a special emphasis on their reliability and cost. A brief
description of the main theories describing the performance of wind turbines is also introduced,
aiming to facilitate the comprehension of the response of the turbine under different operating
conditions.
Then, the focus of the thesis is orientated to the analysis and characterization of the dynamic
behaviour of wind turbines. This characterization was helpful to study the applicability of the OMA
techniques, used in the context of the work (SSI-DATA, SSI-COV and p-LSCF), to wind turbines.
Several methodologies developed and implemented in the context of the dynamic monitoring system
are also described:
o Automated procedure for operational modal analysis;
o Computation of the relative contribution of the vibration modes and harmonics to the
measured acceleration response;
o Statistical tools for detection of damage;
o Fatigue assessment of the support structure using a reduced number of sensors.
353
Conclusions and Future Research
The processing tools developed in this thesis were validated using three case studies. The first case
study comprises the analysis of vibration data collected during two distinct periods on an Izar Bonus
1.3MW/62 wind turbine. With this study, some important achievements should be referred:
o The ability of the SSI-COV and p-LSCF algorithms to accurately identify the main tower and
rotor blades vibration modes of the Izar Bonus 1.3MW/62 wind turbine using both
accelerometers at the tower and fibre Bragg grating sensors at the blades was demonstrated;
o Analysis of the data collected before and after the Rotor Blade Extension was used to
demonstrate the precision of the implemented modal identification algorithms to detect small
variations of frequency, being able to detect the frequency shift associated with an increase of 1
% of the blades mass.
The second case study is an offshore Vestas V90-3.0MW wind turbine. For this study, a short period of
data collected with a dynamic monitoring system was considered. The main conclusions regarding this
analysis were:
o The potential of the SSI-COV and p-LSCF algorithms to identify the most important vibration
modes of an offshore wind turbine was demonstrated for operating and non-operating
conditions;
o The developed methodology for automated identification of the modal properties of the wind
turbine was successfully employed, giving a good indication about the possibility of its
application to generic onshore and offshore wind turbines.
Lastly, the third case is the most complete study of this thesis. It comprises the analysis of the
acceleration data collected during a period of one year at a Senvion MM82 wind turbine. The major
achievements obtained with this analysis are:
o The methodology developed for continuous assessment of the modal properties was
successfully tested in the Senvion MM82 wind turbine. With this analysis, the suitability of this
methodology for identification of the properties of the main vibration modes under several
operating conditions was demonstrated;
o The analysis of the relative contribution of the vibration modes and harmonics for the
acceleration response of the wind turbine showed the influence of the varying operating
conditions on the dynamic behaviour of the turbine. Depending on the tower section under
analysis and on the conditions of operation, the acceleration response may be dominated by
the structural response due to the harmonic excitation 3Ω or by the 1 FA or 2 FA vibration
modes;
o The use of a regression model demonstrated to be a suitable solution to minimize the
operational and environmental effects. A model composed by individual static regression
models for each operating regime proved to deliver the best results;
o The developed monitoring methodology showed its ability to detect small variations of the
natural frequencies, associated with damages at a very early stage on foundation of onshore
and offshore turbines. A lower precision was achieved in the detection of blades damage;
354
Chapter 8
o The proposed procedure for assessment of the wind turbine tower due to the dynamic
component of fatigue was validated. Additional testing is required to extract meaningful
results from this methodology;
o An optimization of the dynamic monitoring system considering a reduction of the number of
sensors was tested. It was concluded that significant results can be obtained for detection of
damage using only one biaxial sensor at the tower top. However, in order to achieve results for
both damage detection and fatigue assessment with a precision similar to the ones obtained
when all sensors are considered, a solution based on one biaxial sensor installed on a
convenient position or a solution based on two biaxial sensors are required. This last solution
may, in fact, only require the installation of one additional sensor if the sensor incorporated in
turbine control system can be used.
355
Conclusions and Future Research
8.2 FUTURE RESEARCH

The work presented in this thesis contains useful contributions for the development of a vibration-
based monitoring system to be implemented in wind turbines. Notwithstanding, some future research
is still needed to benefit from all the potential that a dynamic monitoring system can provide. Some
interesting topics are presented in the following paragraphs:
o By continuing the collection of data of the Senvion MM82 wind turbine, it would be possible
to assess the increase in precision in the detection of damage using a more complete regression
model (improving the mitigation of the external effects over the natural frequencies) and using
more observations to compute the control chart (which would lead to more robust results);
o The installation of sensors in the rotor blades would permit to assess the increase in precision
in the detection of damage in these structural elements, with regard to the accuracy obtained
with the sensors installed at the tower. However, it is expected an additional difficulty in the
analysis of the blades vibration data due to the influence of the rotor rotation. In that sense,
effective tools to remove this effect need to be developed;
o The fatigue assessment of the wind turbine support structure requires further developments.
In that sense, the installation of sensors for measuring the quasi-static motion of this structure
is needed. Using this data, it would be possible to integrally apply the proposed procedure for
fatigue assessment described in section 6.4.1;
o The accuracy of the estimated displacements obtained by integration of the acceleration time
series requires an experimental validation. Thus, the installation of strain gauges in the wind
turbine support structure would permit to study the precision obtained with this process
(mainly for the lower frequency range of the spectrum);
o Alternative sensors to measure the dynamic response of wind turbines, such as MEMS sensors,
should be also tested, in order to lower the installation costs. The precision obtained with this
low-cost sensors should be then compared with the one obtained with more expensive sensors,
such as the ones used in this work. With this information, a cost-benefit analysis would give an
important insight about the possibility of making the system more accessible, allowing its
installation on a larger number of turbines.
356
Appendix A
APPENDIX A
MODEL RESPONSE ESTIMATION -
VALIDATION
A.1. INTRODUCTION
In order to validate the concepts introduced in section 5.5, related to the decomposition of the
recorded acceleration signal into several modal (and harmonic) responses, a simple example is
presented in this section. In that sense, a simple small-scale model of a two-floor frame structure is
used. The analysis is conducted using only one direction, as shown in Figure A.1 a). The stiffness ,
mass and damping matrices of the structural system are introduced in Figure A.1 b).
f2(t) y2(t)
14.655 7.328
/
7.328 7.328
f1(t) y1(t) 0.0018 0

0 0.0018
0.0046 0.0023
. /
0.0023 0.0023
a) b)
Figure A.1 – a) Illustration of the two-floor frame structure; b) Structural matrices of the model
With the previously introduced matrices, two vibration modes with well separated natural frequency
values are obtained. The natural frequency and modal damping ratio values of the system as presented
in Figure A.2, as well as illustrations of the mode shapes.
357
Model Response Estimation - Validation
6.336 16.587
0.618 % 1.618 %
Figure A.2 – Modal characteristics of the two-floor frame structure
A.2. NUMERICAL SIMULATION

Using the introduced structural system, a numerical simulation was performed using the methodology
presented in (Magalhães, 2010). Initially, a broadband excitation was artificially generated with
normally distributed random numbers. This excitation was applied at the floor levels ( and ).
The response of the structural system was obtained based on the discrete-time state-space model from
the structural system, using a frequency sampling of 50 Hz. The considered time length of the
simulation was 600 s. The simulated acceleration time series at the floor levels (y t and y t ) were
computed as outputs.
Once the acceleration signals from the two floors were obtained, the methodology described in section
5.5 was applied using the three different modal identification algorithms. In that sense, it was required
to tune the input parameters from each method. After an initial period of training, the parameters
shown in Table A.1 were selected. The first line of the table (“Modal Identification”) refers to the
different input parameters required for each algorithm, while the second line (“Modal Response”)
denotes the values of the parameter for the post-processing tool, according to section 5.5. Since the
SSI-DATA algorithm uses the parameter in the modal identification process, it is not necessary to
define a new value for the post-processing step.
Table A.1 – Input parameters of the different algorithms used for modal identification and modal decomposition of
the signal
Modal Nº of points of
100 2. 1 257 512
Identification the Half-spectra
Modal Response - - 90 100
358
Appendix A
The results obtained with the application of the modal response post-processing tool to the output
acceleration time series from the first floor are presented in Figure A.3. In this figure, the simulated
acceleration is shown at the top, while the modal responses corresponding to the 1st and 2nd mode are
illustrated in the middle and bottom plots. It is possible to attest that the estimated modal responses
obtained with the different algorithms are almost coincident.
Level 1
0.3
0.2
0.1
Acc. [m/s2]
-0.1
-0.2
-0.3
0 100 200 300 400 500 600
Time [s]
Level 1 - Mode 1
0.15
SSI-DATA
0.1 SSI-COV
p-LSCF
0.05
Acc. [m/s2]
-0.05
-0.1
-0.15
0 100 200 300 400 500 600
Time [s]
Level 1 - Mode 2
0.2 SSI-DATA
SSI-COV
p-LSCF
0.1
Acc. [m/s2]
-0.1
-0.2
0 100 200 300 400 500 600

Time [s]
Figure A.3 – Output acceleration time series at level 1 (top) and estimated responses for the 1st and 2nd vibration modes
obtained with the different algorithms during the whole period of simulation
359
Aiming to highlight the good initial impression given by Figure A.3, Figure A.4 presents the same
results from the previous figure for a shorter time length (10 seconds). Again, the top plot refers to the
numerically obtained acceleration, while the other two shows the modal responses from to the 1st and
2nd modes obtained with the different algorithms. In this figure, the very good agreement between the
modal responses is evident.
Level 1
0.3
0.2
0.1
Acc. [m/s2]
-0.1
-0.2
-0.3
380 381 382 383 384 385 386 387 388 389 390
Time [s]
Level 1 - Mode 1
0.15
SSI-DATA
0.1 SSI-COV
p-LSCF
0.05
Acc. [m/s2]
-0.05
-0.1
-0.15
380 381 382 383 384 385 386 387 388 389 390
Time [s]
Level 1 - Mode 2
0.15
SSI-DATA
0.1 SSI-COV
p-LSCF
0.05
Acc. [m/s2]
-0.05
-0.1
-0.15
380 381 382 383 384 385 386 387 388 389 390
Time [s]
obtained with the different algorithms during a period of 10 seconds
360
Appendix A
The analysis previously presented for the results achieved for the first floor of the structural system was
also performed with the acceleration output of the second floor.
Level 2
0.3
0.2
0.1
Acc. [m/s2]
-0.1
-0.2
-0.3
0 100 200 300 400 500 600
Time [s]
Level 2 - Mode 1
0.2 SSI-DATA
SSI-COV
p-LSCF
0.1
Acc. [m/s2]
-0.1
-0.2
0 100 200 300 400 500 600

Time [s]
Level 2 - Mode 2
0.15
SSI-DATA
0.1 SSI-COV
p-LSCF
0.05
Acc. [m/s2]
-0.05
-0.1
-0.15
0 100 200 300 400 500 600
Time [s]
obtained with the different algorithms during the whole period of simulation
361
A detailed representation of the results is introduced in Figure A.6. In this figure, the good agreement
obtained for the modal responses is clearly noticed.
Level 2
0.3
0.2
0.1
Acc. [m/s2]
-0.1
-0.2
-0.3
380 381 382 383 384 385 386 387 388 389 390
Time [s]
Level 2 - Mode 1
0.2 SSI-DATA
SSI-COV
p-LSCF
0.1
Acc. [m/s2]
-0.1
-0.2
380 381 382 383 384 385 386 387 388 389 390
Time [s]
Level 2 - Mode 2
0.15
SSI-DATA
0.1 SSI-COV
p-LSCF
0.05
Acc. [m/s2]
-0.05
-0.1
-0.15
380 381 382 383 384 385 386 387 388 389 390
Time [s]
obtained with the different algorithms during a period of 10 seconds
362
Appendix B
APPENDIX B
IZAR BONUS 1.3MW/62 WIND
TURBINE – MODE SHAPES
B.1. INTRODUCTION
In this annex the mode shapes of the identified vibration modes of the Izar Bonus 1.3MW/62 wind
turbine are presented.
B.2. ANALYSIS OF ACCELEROMETERS MEASUREMENTS

The mode shapes estimated from the acceleration measurements included ordinates at the
instrumented blade and at the instrumented sections of the tower. For each mode, two lateral views
(along the FA and SS direction) and one top view are presented in Figure B.1.
݂ଵ = 0.45 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
363
Izar Bonus 1.3MW/62 Wind Turbine – Mode Shapes
݂ଶ = 0.46 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ଷ = 0.53 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ସ = 1.13 Hz
70 70
30
60 60
20
50 50
10
40
SS direction
40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
364
Appendix B
݂ହ = 1.14 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂଺ = 1.32 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂଻ = 1.47 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
365
଼݂ = 2.02 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ଽ = 2.06 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ଵ଴ = 3.33 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
366
Appendix B
݂ଵଵ = 3.45 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ଵଶ = 3.74 Hz
70 70
30
60 60
20
50 50
10
40
SS direction
40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
݂ଵଷ = 4.03 Hz
70 70
30
60 60
20
50 50
10
SS direction
40 40
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
367
݂ଵସ = 4.86 Hz
70 70
30
60 60
20
50 50
10
40 40
SS direction
0
30 30
-10
20 20
-20
10 10
-30
0 0
-20 0 20 -20 0 20 -30 -20 -10 0 10 20 30
Figure B.1 – Mode shapes of the vibration modes identified with the measurement system based on accelerometers
B.3. ANALYSIS OF FIBRE BRAGG GRATING SENSORS MEASUREMENTS (AFTER RBE)

The amplitude of the modal curvatures from the first 6 rotor related vibration modes identified after
the RBE are compared with correspondent modes identified before the RBE in Figure B.2. The modes
related with tower motion are not presented since the influence of the installation of the RBE is more
relevant in the rotor modes.
Before installation of RBE After installation of RBE
݂ଷ = 0.61 Hz ݂ଷ = 0.59 Hz
Mode 3 Mode 3
Blade A Blade B Blade C Blade A Blade B Blade C
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
݂ସ = 1.13 Hz ݂ସ = 1.07 Hz
368
Appendix B
Mode 4 Mode 4
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
݂ହ = 1.15 Hz ݂ହ = 1.10 Hz
Mode 5 Mode 5
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
݂଺ = 1.33 Hz ݂଺ = 1.29 Hz
Mode 6 Mode 6
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
݂଻ = 1.45 Hz ݂଻ = 1.43 Hz
369
Mode 7 Mode 7
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
଼݂ = 2.04 Hz ଼݂ = 1.95 Hz
Mode 8 Mode 8
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
Figure B.2 – Comparison of the amplitude of the modal curvature of the first 6 identified rotor related modes before and after
the RBE
370
References
REFERENCES
Ackermann, T. and L. Söder (2002). An overview of wind energy-status 2002. Renewable and
Sustainable Energy Reviews 6(1–2): 67-127.
Adams, D., J. White, M. Rumsey and C. Farrar (2011). Structural health monitoring of wind turbines:
method and application to a HAWT. Wind Energy 14.
Aenlle, M. L., L. Hermanns, P. Fernández and A. Fraile (2013). Stress Estimation in a Scale Model of a
Symmetric Two Story Building. 5th International Operational Modal Analysis Conference (IOMAC).
Guimarães, Portugal.
Aenlle, M. L., A. Skafte, P. Fernández and R. Brincker (2013). Strain Estimation in a Glass Beam using
Operational Modal Analysis. XXXI International Modal Analysis Conference (IMAC). Florida, USA.
Ameri, N., C. Grappasonni, G. Coppotelli and D. J. Ewins (2013). Ground vibration tests of a helicopter
structure using OMA techniques. Mechanical Systems and Signal Processing(35): 35-51.
Andersen, L. (2010). Assessment of lumped-parameter models for rigid footings. Computers &
Structures 88(23–24): 1333-1347.
Andersen, P., R. Brincker, C. Ventura and R. Cantieni (2007). Estimating Modal Parameters of Civil
Engineering Structures subject to Ambient and Harmonic Excitation. Experimental Vibration Analysis
for Civil Engineering Structures (EVACES). Porto, Portugal.
Anon (1890). Mr. Brush's Windmill Dynamo. Scientific American Vol. LXIII(No. 25): cover and p.
389.
Ansys Inc. (2011). ANSYS Academic Research. Release 14.0.
Appleyard, D. (2013). Wind Insurance Claim Analysis Reveals Need for Monitoring. 2013, from
http://www.renewableenergyworld.com/rea/news/article/2013/08/wind-insurance-claim-analysis-
reveals-monitoring-need.
371
References
Arántegui, R. L., T. Corsatea and K. Suomalainen (2013). 2012 JRC Wind Status Report, European
Commission. Joint Research Institute (JRC). Institute for Energy and Transport.
Arrigan, J., V. Pakrashi, B. Basu and S. Nagarajaiah (2011). Control of flapwise vibrations in wind
turbine blades using semi-active tuned mass dampers. Structural Control and Health Monitoring 18:
840-851.
Bak, C., P. Fuglsang, J. Johansen and I. Antoniou (2000). Wind tunnel tests of the NACA 63-415 and a
modified NACA 63-415 airfoil. Technical Report Risø–R–1193. Roskilde, Denmark, Risø National
Laboratory.
Baldwin, D. H. and J. Kennard (1985). Development of Large, Horizontal-Axis Wind Turbines.

DOE/NASA/20320-62 NASA TM-86950
BARD Engineering GmbH. BARD - Wind Offshore Projects. 2013, from http://www.bard-
offshore.de/en/projects.html.
Bartsch, C. (2012). Fact-Sheet alpha ventus. EWE, E-ON and Vattenfall.
Becker, E. and P. Poste (2006). Keeping the blades turning: Condition monitoring of wind turbine gears.
Refocus 7(2): 26-32.
Belwind Offshore Energy (2009). Belwind - Wind farm in the North Sea (Press information).
Betz, A. (1920). Das Maximum der theoretisch möglichen Ausnutzung des Windes durch Windmotoren
(The maximum of the theoretically possible exploitation of the wind by wind motors). Zeitschrift f. d.
gesamte Turbinenwesen 17.
Betz, A. (1926). Windenergie und Ihre Ausnutzung durch Windmühlen,. Göttingen, Germany,
Vandenhoeck.
Bir, G. (2008). Multiblade Coordinate Transformation and Its Application to Wind Turbine Analysis.
ASME Wind Energy Symposium. Nevada, USA.
Blanco, M. I. (2009). The economics of wind energy. Renewable and Sustainable Energy Reviews 13(6–
7): 1372-1382.
Bloomberg New Energy Finance (2015). Sustainable Energy in America: 2015 Factbook. London, UK.
Blough, J. R. (2006). Adaptive Resampling - Transforming From the Time to the Angle Domain. XXIV
International Modal Analysis Conference (IMAC). Missouri, USA.
372
References
Bogert, B. P., M. J. R. Healy and J. W. Tukey (1963). The Quefrency Alanysis of Time Series for Echoes:
Cepstrum, Pseudo-Autocovariance, Cross-Cepstrum, and Saphe Cracking. Time Series Analysis(15):
209-243.
Boller, C., F.-K. Chang and Y. Fujino (2009). Encyclopedia of Structural Health Monitoring, John Wiley
& Sons.
Braam, H., T. S. Obdam and T. W. Verbruggen (2012). Low Cost Load Monitoring for Offshore Wind
Farms. International Conference on Noise and Vibration Engineering (ISMA). Leuven, Belgium.
Brincker, R., P. Andersen and N. Møller (2000). Indicator for Separation of Structural and Harmonic
Modes in Output-only Modal Testing. XVIII International Modal Analysis Conference (IMAC). San
Antonio, USA.
Brownjohn, J. M. W. (2007). Structural health monitoring of civil infrastructure. Philosophical

Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 365(1851): 589-
622.
Brownjohn, J. M. W., F. Magalhães, E. Caetano and Á. Cunha (2010). Ambient vibration re-testing and
operational modal analysis of the Humber Bridge. Engineering Structures(32): 2003-2018.
Burton, T., D. Sharpe, N. Jenkins and E. Bossanyi (2001). Wind Energy Handbook, John Wiley & Sons.
Butterfield, C. P., W. Musial and J. Jonkman (2007). Overview of Offshore Wind Technology, National
Renewable Energy Laboratory (NREL).
Butterfield, S., W. Musial and J. Jonkman (2005). Engineering Challenges for Floating Offshore Wind
Turbines, National Renewable Energy Laboratory (NREL).
Byrne, B., G. Houlsby, C. Martin and P. Fish (2002). Suction Caisson Foundations for Offshore Wind
Turbines. Wiind Engineering 26(3): 145-155.
Caetano, E., F. Magalhães, Á. Cunha, O. Flamand and G. Grillaud (2007). Comparison of Sochastic
Identification Methods Applied to the Natural Response of Millau Viaduct. Experimental Vibration
Analysis for Civil Engineering Structures (EVACES). Porto, Portugal.
Cara, F. J., J. Juan, E. Alarcón, E. Reynders and G. De Roeck (2013). Modal contribution and state space
order selection in operational modal analysis. Mechanical Systems and Signal Processing 38(2): 276-
298.
373
References
Carbon Trust. (2013). Successful installation of first suction bucket foundation at Dogger Bank. 2013,
from http://www.carbontrust.com/news/2013/02/successful-installation-of-first-suction-bucket-
foundation-at-dogger-bank.
Carcangiu, C. E., D. Tcherniak, S. Chauhan, J. Basurko and M. Rossetti (2012). Numerical and
Experimental Modal Characterization of a 3MW Wind Turbine. EWEA 2012 Annual Event.
Copenhagen, Denmark.
Carden, E. P. and P. Fanning (2009). Vibration Based Condition Monitoring: A Review. Structural
Health Monitoring(4): 355-377.
Carne, T. G. and G. H. James Iii (2010). The inception of OMA in the development of modal testing
technology for wind turbines. Mechanical Systems and Signal Processing 24(5): 1213-1226.
Carne, T. G., J. P. Lauffer and A. J. Gomez (1988). Modal Testing the EOLE. New Mexico, USA, Sandia
National Laboratories.
Carne, T. G., D. W. Lobitz, A. R. Nord and R. A. Watson (1982). Finite Element Analysis and Modal
Testing of a Rotating Wind Turbine. New Mexico, USA, Sandia National Laboratories.
Carne, T. G. and A. R. Nord (1983). Modal Testing of a Rotating Wind Turbine. New Mexico, USA,
Sandia National Laboratories.
Cauberghe, B. (2004). Applied frequency-domain system identification in the field of experimental and
operational modal analysis. PhD thesis, V.U.Brussel.
Chauhan, S., D. Tcherniak, J. Basurko, O. Salgado, I. Urresti, C. E. Carcangiu and M. Rossetti (2011).
Operational Modal Analysis of Operating Wind Turbines: Application to Measured Data. IXXX
International Modal Analysis Conference (IMAC). Florida, USA.
Chou, J.-S. and W.-T. Tu (2011). Failure analysis and risk management of a collapsed large wind
turbine tower. Engineering Failure Analysis 18(1): 295-313.
Ciang, C. C., J.-R. Lee and H.-J. Bang (2008). Structural health monitoring for a wind turbine system: a
review of damage detection methods. Measurement Science and Technology 19: 1-20.
Clough, R. W. and J. Penzien (1995). Dynamics of Structures. Berkeley, USA, Computers & Structures,
Inc.
Collins, J. L., R. K. Shaltens, R. H. Poor and R. S. Barton (1982). Experience and Assessment of the
DOE/NASA MOD-1 2000 kW Wind Turbine Generator at Boorne, North Carolina.
374
References
Colwell, S. and B. Basu (2009). Tuned liquid column dampers in offshore wind turbines for structural
control. Engineering Structures 31(2): 358-368.
Commission of the European Communities (2008). 20 20 by 2020 - Europe's climate change

opportunity. t. C. Communication from the Commision to the European Parliament, the European
Economic and Social Committee and the Committee of the Regions. Brussels, Switzerland.
Connel, J. R. (1980). Turbulence spectrum observed by a fast-rotating wind turbine blade, Battelle
Pacific Northwest Laboratory.
Corrigan, R. D. and C. B. F. Ernsworth (1986). Design and Initial Testing of a One Bladed 30-Meter
Diameter Rotor on the NASA-DOE Mod-0 Wind Turbine. DOE/NASA/20320-70 NASA TM-88810.
Cunha, Á. and E. Caetano (2006). Experimental Modal Analysis of Civil Engineering Structures. Sound
and Vibration(40): 12-20.
Cunha, Á., E. Caetano and R. Delgado (2001). Dynamic Tests on a Large Cable-Stayed Bridge. An
Efficient Approach. Journal of Bridge Engineering 6(1): 54-62.
Currie, M., M. Saafi, C. Tachtatzis and F. Quail (2015). Structural integrity monitoring of onshore wind
turbine concrete foundations. Renewable Energy 83: 1131-1138.
Damgaard, M., L. B. Ibsen, L. V. Andersen and J. K. F. Andersen (2013a). Cross-wind modal properties
of offshore wind turbines identified by full scale testing. Journal of Wind Engineering and Industrial
Aerodynamics 116: 94-108.
Damgaard, M., L. B. Ibsen, L. V. Andersen, P. Andersen and J. K. F. Andersen (2013b). Damping

Estimation of a Prototype Bucket Foundation for Offshore Wind Turbines Identified by Full Scale
Testing. 5th International Operational Modal Analysis Conference (IOMAC). Guimarães, Portugal.
Darrieus, G. J. (1931). Turbine Having Its Rotating Shaft Transverse to the Flow of The Current. U. S. P.
Offcie.
Department of Energy & Climate Change (2010). Value breakdown for the offshore wind sector.
Det Norske Veritas (DNV) (2002). Guidelines for Design of Wind Turbines.
Det Norske Veritas (DNV) (2011). DNV-OS-J101 - Design of Offshore Wind Turbine Structures.
Devriendt, C., F. Magalhães, W. Weijtjens, G. De Sitter, Á. Cunha and P. Guillaume (2014). Structural
health monitoring of offshore wind turbines using automated operational modal analysis. Structural
Health Monitoring 13(6): 644-659.
375
References
Devriendt, C., W. Weijtjens, M. El-Kafafy and G. D. Sitter (2014) Monitoring resonant frequencies and
damping values of an offshore wind turbine in parked conditions. IET Renewable Power Generation 8,
433-441.
Dong, W., T. Moan and Z. Gao (2012). Fatigue reliability analysis of the jacket support structure for
offshore wind turbine considering the effect of corrosion and inspection. Reliability Engineering &
System Safety 106: 11-27.
Enercon GmbH (2005). Installation of E-112 nearshore in Wilhelmshaven stopped.
European Comitee for Standardization (CEN) (2005). EN 1993-1-9 Eurocode 3: Design of steel
structures - Part 1-9: Fatigue.
European Commission (2011). UpWind - Design limits and solutions for very large wind turbines, The
Sixth Framework Programme for Research and Development of the European Commission (FP6).
European Commission (2013). EU energy in figures - Statistical Pocketbook 2013. Luxembourg.
European Commission. (2015a). 2030 climate & energy framework. from

http://ec.europa.eu/clima/policies/strategies/2030/index_en.htm.
European Commission. (2015b). 2050 low-carbon economy. from

http://ec.europa.eu/clima/policies/strategies/2050/index_en.htm.
European Commission (2015c). The European Union leading in renewables. Brussels, Belgium.
European Wind Energy Association (EWEA) (2007). Powering Change - EWEA 2006 Annual Report.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2008). Delivering Eenery and Climate Solutions - EWEA
2007 Annual Report. Brussels, Belgium.
European Wind Energy Association (EWEA) (2009). Winning with European Wind - EWEA 2008
annual report. Brussels, Belgium.
European Wind Energy Association (EWEA) (2010). Wind in Power - 2009 European statistics.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2011a). The European offshore wind industry key trends
and statistics 2010. Brussels, Belgium.
376
References
European Wind Energy Association (EWEA) (2011b). Wind in Power - 2010 European statistics.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2012a). The European offshore wind industry key 2011
trends and statistics. Brussels, Belgium.
European Wind Energy Association (EWEA) (2012b). Wind in Power - 2011 European Statistics.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2013a). The European offshore wind industry - key
trends and statistics 2012. Brussels, Belgium.
European Wind Energy Association (EWEA) (2013b). Wind in Power - 2012 European Statistics.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2014a). The European offshore wind industry - key
trends and statistics 2013. Brussels, Belgium.
European Wind Energy Association (EWEA) (2014b). Wind in power - 2013 European statistics.
Brussels, Belgium.
European Wind Energy Association (EWEA) (2015a). The European offshore wind industry - key trend
and statistics 2014.
European Wind Energy Association (EWEA) (2015b). Wind in power - 2014 European statistics.
Brussels, Belgium.
Ewins, D. J. (2000). Modal Testing: Theory, Practice and Application. Baldock, England, Research
Studies Press Ltd.
Fast, M. and T. Palmé (2010). Application of artificial neural networks to the condition monitoring and
diagnosis of a combined heat and power plant. Energy 35(2): 1114-1120.
Faulstich, S. (2010). Component Reliabilty Ranking with Respect to WT Concept and External
Environmental Conditions Project UpWind.
Felber, A. J. (1993). Development of a Hybrid Bridge Evaluation system. PhD thesis, The University of
British Columbia.
Freris, L. L. (1990). Wind Energy Conversion System. New York, USA, Prentice-Hall.
377
References
Fritzen, C.-P. (2005). Vibration-Based Structural Health Monitoring – Concepts and Applications. Key
Engineering Materials 293 - 294(3): 3 - 20.
García Márquez, F. P., A. M. Tobias, J. M. Pinar Pérez and M. Papaelias (2012). Condition monitoring
of wind turbines: Techniques and methods. Renewable Energy 46(0): 169-178.
Gasch, R. and J. Twele (2012). Wind Power Plants - Fundamentals, Design, Construction and
Operation, Springer.
Gellermann, T. (2012). Extension of the scope of Condition Monitoring Systems for multi-MW and
offshore wind turbines1. VDI-Conference Schwingungen von Windenergieanlagen 2012. Bremen,
Germany.
Gentile, C. and A. Saisi (2007). Ambient vibration testing of historic masonry towers for structural
identification and damage assessment. Construction and Building Materials(21): 1311-1321.
Germanischer Lloyd (GL) (2003). Guideline for the Certification of Condition Monitoring Systems for
Wind Turbines. Rules and Guidelines - IV Industrial Services. Hamburg, Germany.
Germanischer Lloyd (GL) (2005). Overall Damping for Piled Offshore Support Structures. Note on
Engineering Details.
Germanischer Lloyd (GL) (2010). Guideline for the Certification of Wind Turbines. Hamburg,
Germany.
Germanischer Lloyd (GL) (2012). Guideline for the Certification of Offshore Wind Turbines. Hamburg,
Germany.
Glauert, H. (1926). The Analysis of Experimental Results in the Windmill Brake and Vortex Ring States
of an Airscrew. Aeronautical Research Committee Reports and Memoranda.
Global Wind Energy Council (GWEC) (2007). Global Wind 2006 Report. Brussels, Belgium.
Global Wind Energy Council (GWEC) (2011). Global Wind Report - Annual Market Update 2010.
Brussel, Belgium.
378
References
Global Wind Energy Council (GWEC) (2012). Global Wind Report - Annual market update 2011.
Brussel, Belgium.
Global Wind Energy Council (GWEC) (2013). Global Wind Report - Annual market update 2012.
Brussel, Belgium.
Brussels, Belgium.
Brussels, Belgium.
Gordon, A. D. (1999). Classification, Chapman and Hall/CRC.
Gordon, L. H., J. S. Andrews and D. K. Zimmerman (1983). Mod-2 Wind Turbine Development.
DOE/NASN20305-9 NASATM-83460.
Goursat, M., M. Döhler, L. Mevel and P. Andersen (2010). Crystal Clear SSI for Operational Modal
Analysis of Aerospace Vehicles. XXVIII International Modal Analysis Conference (IMAC). Florida,
USA.
Government of Portugal (2015). Green Growth Commitment. Lisbon, Portugal, Ministry of

Environment Spatial Planning and Energy,.
Gray, L. (2013). Wind turbine collapses in high wind. 2013, from

http://www.telegraph.co.uk/earth/energy/windpower/9837026/Wind-turbine-collapses-in-high-
wind.html.
Griffith, D. T. and T. G. Carne (2010). Experimental Modal Analysis of 9-meter Research-sized Wind
Turbine Blades. XXVIII International Modal Analysis Conference (IMAC). Florida, USA.
Griffith, D. T., R. L. Mayes and P. S. Hunter (2010). Excitation Methods for a 60 kW Vertical Axis Wind
Turbine. XXVIII International Modal Analysis Conference (IMAC), Florida, USA.
Griffith, D. T., G. Smith, M. Casias, S. Reese and T. W. Simmermacher (2006). Modal Testing of the
TX-100 Wind Turbine Blade. New Mexico, USA, Sandia National Laboratories.
Griffith, D. T., N. C. Yoder, B. Resor, J. White and J. Paquette1 (2013). Structural health and
prognostics management for the enhancement of offshore wind turbine operations and maintenance
strategies. Wind Energy.
379
References
Guillaume, P., P. Verboven, S. Vanlanduit, H. V. d. Auweraer and B. Peeters (2003). A Poly-Reference

Implementation of the Least-Squares Complex Frequency-Domain Estimator. XXII International Modal
Analysis Conference (IMAC). Florida, USA.
Guo, P. and D. Infield (2012). Wind Turbine Tower Vibration Modeling and Monitoring by the
Nonlinear State Estimation Technique (NSET). Energies 5: 5279-5293.
Hall, M., B. Buckham and C. Crawford (2013). Evaluating the importance of mooring line model fidelity
in floating offshore wind turbine simulations. Wind Energy: n/a-n/a.
Hameed, Z., Y. S. Hong, Y. M. Cho, S. H. Ahn and C. K. Song (2009). Condition monitoring and fault
detection of wind turbines and related algorithms: A review. Renewable and Sustainable Energy Reviews
13(1): 1-39.
Hansen, M. H. (2003). Improved Modal Dynamics of Wind Turbines to Avoid Stall-induced Vibrations.
Wind Energy 6: 179-195.
Hansen, M. H. (2007). Aeroelastic Instability Problems for Wind Turbines. Wind Energy 10(6): 551-
577.
Hansen, M. H., K. Thomsen and P. Fuglsang (2006). Two Methods for Estimating Aeroelastic Damping
of Operational Wind Turbine Modes from Experiments. Wind Energy(9): 179-191.
Harman, K., R. Walker and M. Wilkinson (2008). Availability trends observed at operational wind
farms. European Wind Energy Conference (EWEC). Brussels, Belgium.
Hasselmann, K., T. P. Barnett, E. Bouws, H. Carlson, D. E. Cartwright, K. Enke, J. A. Ewing, H.

Gienapp, D. E. Hasselmann, P. Kruseman, A. Meerburg, P. Müller, D. J. Olbers, K. Richter, W. Sell and
H. Walden (1973). Measurements of wind wave growth and swell decay during the Joint North Sea
Wave Project (JONSWAP). Hamburg, Germany, Deutsches Hydrographisches Institut.
Hau, E. (2006). Wind Turbines - Fundamentals, Technologies, Application, Economics, Springer.
HAWC2 web site. (2013). HAWC2 model - NREL 5-MW Reference Wind Turbine. from
http://www.hawc2.dk.
Henderson, A. R. (2003). Support Structures for Floating Offshore. Workshop on Deep Water Offshore
Wind Energy Systems.
Henderson, A. R., C. Morgan, B. Smith, H. C. Sorensen, R. J. Barthelmie and B. Boesmans (2003).

Offshore Wind Energy in Europe - A Review of the State-of-the-Art. Wind Energy 6(doi: 10.1002/we.82):
35-52.
380
References
Hermans, L. and H. V. d. Auweraer (1999). Modal Testing and Analysis of Structures Under
Operational Conditions: Industrial Applications. Mechanical Systems and Signal Processing 13(2): 193-
216.
Heylen, W., S. Lammens and P. Sas (2007). Modal Analysis Theory and Testing. Belgium, Katholieke
Universiteit Leuven.
Hjelm, H. P., R. Brincker, J. Graugaard-Jensen and K. Munch (2005). Determination of Stress Histories
in Structures by Natural Input Modal Analysis. XXIII International Modal Analysis Conference
(IMAC). Florida, USA.
Hoell, S. and P. Omenzetter (2014). Detection and localization of trailing edge disbond in a large wind
turbine blade. International Conference on Offshore Renewable Energy. Glasgow, UK.
Hu, W.-H. (2011). Operational Modal Analysis and Continuous Dynamic Monitoring of Footbridges.
PhD thesis, Faculdade de Engenharia da Universidade do Porto.
Hu, W.-H., E. Caetano and Á. Cunha Structural health monitoring of a stress-ribbon footbridge.
Engineering Structures 57.
Hu, W.-H., C. Moutinho, E. Caetano, F. Magalhães and Á. Cunha (2012). Continuous dynamic
monitoring of a lively footbridge for serviceability assessment and damage detection. Mechanical Systems
and Signal Processing 33: 38-55.
Hu, W.-H., S. Thöns, R. G. Rohrmann, S. Said and W. Rücker (2015a). Vibration-based structural
health monitoring of a wind turbine system Part II: Environmental/operational effects on dynamic
properties. Engineering Structures 89: 273-290.
Hu, W.-H., S. Thöns, R. G. Rohrmann, S. Said and W. Rücker (2015b). Vibration-based structural
health monitoring of a wind turbine system. Part I: Resonance phenomenon. Engineering Structures 89:
260-272.
Hyers, R. W., J. G. McGowan, K. L. Sullivan, J. F. Manwell and B. C. Syrett (2006). Condition

monitoring and prognosis of utility scale wind turbines. Energy Materials: Materials Science and
Engineering for Energy Systems 1(3): 187-203.
Ibsen, L. B. and M. Liingaard (2006). Prototype bucket foundation for wind turbines - natural frequency
estimation. Aalborg, Denmark, Aalborg University - Department of Civil Engineering.
Ibsen, L. B., M. Liingaard and S. A. Nielsen (2005). Bucket Foundation, a status. Copenhagen Offshore
Wind, Copenhagen, Denmark.
381
References
Institute of Mechanical Engineering and Industrial Management (INEGI) and Portuguese Renewable
Energy Association (APREN) (2015). Wind Farms in Portugal - December 2014.
International Electrotechnical Commission (IEC) (2009). IEC 61400-3 Wind turbines – Part 3: Design
requirements for offshore wind turbines.
International Electrotechnical Commission (IEC) (2014). IEC 61400-1 Wind turbines –Part 1: Design
requirements.
International Energy Agency (IEA) (1984). International Energy Agency - Co-operation in the
Development of Large-Scale Wind Energy Systems (Annual Report of 1983), IEA LS WECS Executive
Commitee.
International Energy Agency (IEA) (1989). IEA Large-Scale Wind Energy - Annual Report 1988, IEA
LS WECS Execuiive Committee.
Jacobsen, N.-J., P. Andersen and R. Brincker (2006). Using Enhanced Frequency Domain
Decomposition as a Robust Technique to Harmonic Excitation in Operational Modal Analysis.
International Conference on Noise & Vibration Engineering 2006 (ISMA). Lueven, Belgium.
Jacobsen, N.-J., P. Andersen and R. Brincker (2008). Applications of Frequency Domain Curve-fitting in
the EFDD Technique. XXVI International Modal Analysis Conference (IMAC). Florida, USA.
James, G. H. (1994). Extraction of modal parameters from an operating HAWT using the Natural
Excitation Technique (NExT). 13th Energy-Sources Technology Conference and Exhibition (ETCE) on
Wind Energy. Los Angeles, USA.
James, G. H., T. G. Carne and J. P. Lauffer (1993). The Natural Excitation Technique (NExT) for Modal
Parameter Extraction From Operating Wind Turbines. New Mexico, USA, Sandia National
Laboratories.
Johnson, G. L. (2001). Wind Energy Systems, Prentice Hall.
Johnson, R. A. and D. W. Wichern (2002). Applied multivariate statistical analysis. New Jersey, USA,
Prentice Hall.
Jonkman, J., S. Butterfield, W. Musial and G. Scott (2009). Definition of a 5-MW Reference Wind
Turbine for Offshore System Development National Renewable Energy Laboratory.
Juang, J.-N. (1994). Applied System Identification. New Jersey, USA, Prentice Hall Englewood Cliffs.
382
References
Kármán, T. v. (1948). Progress in the statistical theory of turbulence. Proceedings of the National
Academy of Science of the United States of America 34: 530-539.
Kuhn, M. J. (2001). Dynamics and Design Optimisation of Offshore Wind Energy Conversion Systems.
PhD thesis, Delft University of Technology.
Kvittem, M. I. and T. Moan (2015). Time domain analysis procedures for fatigue assessment of a semi-
submersible wind turbine. Marine Structures 40: 38-59.
Larsen, T. J. and A. M. Hansen (2007). How 2 HAWC2, the user’s manual.
LeBlanc, C. and N. J. Tarp-Johansen (2011). Monopile in sand, stiffness and damping. Oral
Presentation, EWEA Conference. Brussels, Belgium.
Li, Z.-q., S.-j. Chen, H. Ma and T. Feng (2013). Design defect of wind turbine operating in typhoon
activity zone. Engineering Failure Analysis(27): 165-172.
Liu, W., B. Tang and Y. Jiang (2010a). Status and problems of wind turbine structural health monitoring
techniques in China. Renewable Energy 35(7): 1414-1418.
Liu, W., B. Tang and Y. Jiang (2010b). Status and problems of wind turbine structural health
monitoring techniques in China. Renewable Energy(35): 1414-1418.
Liu, W. Y., B. P. Tang, J. G. Han, X. N. Lu, N. N. Hu and Z. Z. He (2015). The structure healthy
condition monitoring and fault diagnosis methods in wind turbines: A review. Renewable and
Sustainable Energy Reviews 44: 466-472.
Loraux, C. and E. Brühwiler (2015). Realistic examination of the fatigue life of a wind turbine tower
using data from long term monitoring. IABSE Conference – Structural Engineering: Providing
Solutions to Global Challenges. Geneva, Switzerland.
LORC. (2012a). The Jacket - A Path to Deeper Waters. 2013, from http://www.lorc.dk/offshore-
wind/foundations/jackets#post-piling.
LORC. (2012b). The Monopile - Close to a Monopole. 2013, from http://www.lorc.dk/offshore-

wind/foundations/monopiles.
LORC. (2012c). Tripile - Three Monopiles in One. 2013, from http://www.lorc.dk/offshore-

wind/foundations/tripiles.
Lundsager, P., S. Frandsen and C. J. Christensen (1980). Analysis of Data from the Gedser Wind turbine
1977-1979. Riso-M-2242. Roskilde, Denmark, Riso National Laboratory.
383
References
Madsen, P. H., S. Frandsen, W. E. Holley and J. C. Hansen (1984). Dynamics and Fatigue Damage of
Wind Turbine Rotors during Steady Operation. Riso-R-512. Roskilde, Denmark, Riso National
Laboratory.
Magalhães, F. (2010). Operational Modal Analysis for Testing and Monitoring of Bridges and Special
Structures. PhD thesis, Faculdade de Engenharia da Universidade do Porto.
Magalhães, F. and Á. Cunha (2011). Explaining operational modal analysis with data from an arch
bridge. Mechanical Systems and Signal Processing 25: 1431-1450.
Magalhães, F., A. Cunha and E. Caetano (2012). Vibration based structural health monitoring of an
arch bridge: From automated OMA to damage detection. Mechanical Systems and Signal Processing
28(0): 212-228.
Magalhães, F., Á. Cunha and E. Caetano (2009). Online automatic identification of the modal
parameters of a long span arch bridge. Mechanical Systems and Signal Processing 23(2): 316-329.
Magalhães, F., Á. Cunha, E. Caetano and R. Brincker (2010). Damping estimation using free decays and
ambient vibration tests. Mechanical Systems and Signal Processing 24(5): 1274-1290.
Main(e) International Consulting LLC (2012). Floating Offshore Wind Foundations: Industry Consortia
and Projects in the United States, Europe and Japan - An Overview.
Manwell, J. F., J. G. McGowan and A. L. Rogers (2010). Wind Energy Explained - Theory, Design and
Application, John Wiley & Sons.
Manzato, S., C. Devriendt, W. Weijtjens, E. Di Lorenzo, B. Peeters and P. Guillaume (2014). Removing
the Influence of Rotor Harmonics for Improved Monitoring of Offshore Wind Turbines. Dynamics of
Civil Structures, Volume 4. F. N. Catbas, Springer International Publishing: 299-312.
Manzato, S., J. R. White, B. LeBlanc, B. Peeters and K. Janssens (2013). Advanced identification
techniques for operational wind turbine data. XXXI International Modal Analysis Conference (IMAC),
California, USA.
Marinone, T., D. Cloutier, B. LeBlanc, T. Carne and P. Andersen (2014). Artificial and Natural
Excitation Testing of SWiFT Vestas V27 Wind Turbines. XXXII International Modal Analysis
Conference (IMAC). Florida, USA.
Marulo, F., G. Petrone, E. D. Lorenzo and A. Cutolo (2015). Operational modal analysis for SHM of a
wind turbine blade. 7th International Conference on Structural Health Monitoring. Torino, Italy.
MathWorks (2012). Matlab. R2012b.
384
References
McMillan, D. and G. W. Ault (2007). Quantification of Condition Monitoring Benefit for Offshore Wind
Turbines. Wind Engineering 31(4): 267-285.
Mevel, L., A. Benveniste, M. Bassevillea, M. Goursatb, B. Peetersd, H. V. d. Auweraerd and A.

Vecchiod (2006). Input/Output Versus Output-Only Data Processing for Structural Identification—
Application to In-Flight Data Analysis. Journal of Sound and Vibration(295): 531-552.
Molenaar, D. (2003). Experimental Modal Analysis of a 750 kW Wind Turbine for Structural Model
Validation. 41st AIAA Aerospace Sciences Meeting and Exhibit. Nevada, USA.
Montgomery, D. C. (2009). Statistical quality control : a modern introduction. New York, John Wiley &
Sons.
Morthorst, P. E. and H. Jacobsen (2004). Wind Energy - The Facts, European Commission (EC)
European Wind Energy Association (EWEA). Volume 2 - Costs & Prices.
Moser, P. and B. Moaveni (2011). Environmental effects on the identified natural frequencies of the
Dowling Hall Footbridge. Mechanical Systems and Signal Processing 25(7): 2336-2357.
Murtagh, P. J., B. Basu and B. M. Broderick (2005). Along-wind response of a wind turbine tower with
blade coupling subjected to rotationally sampled wind loading. Engineering Structures 27(8): 1209-
1219.
Musial, W., S. Butterfield and A. Boone (2004). Feasibility of Floating Platform Systems for Wind
Turbines. 23rd ASME Wind Energy Symposium. Nevada, USA, Nationa Renewable Energy Laboratory
(NREL).
Musial, W. and B. Ram (2010). Large-Scale Offshore Wind Power in the United States - Assessment of
Opportunities and Barriers, National Renweable Energy Laboratory (NREL).
Naess, A. and T. Moan (2012). Stochastic Dynamics of Marine Structures. USA, Cambridge University
Press.
Naguleswaran, S. (1994). Lateral Vibration of a Centrifugally Tensioned Uniform Euler-Bernoulli Beam.

Journal of Sound and Vibration 176(5): 613-624.
Nielsen, J. J. and J. D. Sørensen (2011). On risk-based operation and maintenance of offshore wind
turbine components. Reliability Engineering and System Safety(96): 218-229.
Oliveira, G., F. Magalhães, E. Caetano and Á. Cunha (2013). Modal Identification and FE Model
Correlation of a Wind Turbine Tower. 5th International Operational Modal Analysis Conference
(IOMAC). Guimarães, Portugal.
385
References
Oliveira, G., F. Magalhães and Á. Cunha (2014). Rotor blade extension project – Determination of the
blades natural frequencies. Porto, Portugal.
Oliveira, G., W. Weijtjens, G. D. Sitter, F. Magalhães, Á. Cunha, E. Caetano and C. Devriendt (2014).
Dynamic Monitoring of wind turbines: case studies on- and off-shore. 1st International Conference on
Renewable Energies Offshore. Lisboa, Portugal.
Osgood, R. (2001). Dynamic Characterization Testing of Wind Turbines. Colorado, USA, National
Renewable Energy Laboratory (NREL).
Osgood, R., G. Bir, H. Mutha, B. Peeters, M. Luczak and G. Sablon (2010). Full-scale modal wind
turbine tests: comparing shaker excitation with wind excitation. XXVIII International Modal Analysis
Conference (IMAC). Florida, USA.
Overschee, P. V. and B. D. Moor (1996). Subspace Identification for Linear Systems - Theory,
Implementation, Applications, Kluwer Academinc Publishers.
Ozbek, M., F. Meng and D. J. Rixen (2013). Challenges in testing and monitoring the in-operation
vibration characteristics of wind turbines. Mechanical Systems and Signal Processing.
Ozbek, M. and D. J. Rixen (2012). Operational modal analysis of a 2.5 MW wind turbine using optical
measurement techniques and strain gauges. Wind Energy(16).
Ozbek, M., D. J. Rixen, O. Erne and G. Sanowb (2010). Feasibility of monitoring large wind turbines
using photogrammetry. Energy 35: 4802-4811.
Pappa, R. S., G. H. J. III and D. C. Zimmerman (1997). Autonomous modal identification of the Space
Shuttle tail rudder. Virginia, USA, National Aeronautics and Space Administration (NASA).
Peeters, B. (2000). System Identification and Damage Detection in Civil Engineering. PhD thesis,
Katholieke Universiteit Leuven.
Peeters, B. and H. V. d. Auweraer (2005). PolyMax: a Revolution in Operational Modal Analysis. 1st
International Operational Modal Analysi Conference (IOMAC). Copenhagen, Denmark.
Peeters, B., R. Cornelius, K. Janssens and H. V. d. Auweraer (2007). Removing Disturbing Harmonics in
Operational Modal Analysis. 2nd International Operational Modal Analysis Conference (IOMAC).
Copenhagen, Denmark.
Peeters, B., H. Van der Auweraer, P. Guillaume and J. Leuridan (2004). The PolyMAX Frequency-
Domain Method: A New Standard for Modal Parameter Estimation? Shock and Vibration 11(3-4): 395-
409.
386
References
Pelayo, F., M. López-Aenlle, A. Fernández-Canteli and R. Cantieni (2011). Operational modal analysis
of two wind turbines with foundation problems. 4th International Operational Modal Analysis
Conference. Istanbul, Turkey.
Pierson, W. and L. Moskowitz (1964). A proposed spectral form of fully developed wind seas based on
the similarity theory of S. A. Kitaigorodskii. Journal of Geophysical Research 69(4): 5181-5190.
Pollino, M. and A. Huckelbridge (2012). In-situ Measurements of Fatigue Demands on a Wind Turbine
Support Structure. IEEE Energy Tech 2012. Ohio, USA.
Portuguese Renewable Energy Association (APREN). (2015). Produção de Electricidade por Fonte em
Portugal em 2015. Retrieved 05/01/2016, from http://www.apren.pt/pt/dados-tecnicos-3/dados-
nacionais-2/producao-2/a-producao-de-electricidade-em-portugal-2/fontes-de-producao-de-
eletricidade/.
Prendergast, L. J., K. Gavin and P. Doherty (2015). An investigation into the effect of scour on the
natural frequency of an offshore wind turbine. Ocean Engineering 101: 1-11.
Putter, S. and H. Manor (1978). Natural Frequencies of Radial Rotating Beams. Journal of Sound and
Vibration 52(2): 175-185.
Quilter, J. (2013). World's largest blade begins journey to Scotland. 2013, from
http://www.windpowermonthly.com/article/1191655/picture-gallery---worlds-largest-blade-begins-
journey-scotland?DCMP=EMC-CONWindpowerWeekly&bulletin=windpower-weekly.
Ramírez, O., D. Tcherniak and G. Larsen (2015). Comparative study of OMA applied to experimental
and simulated data from an operating Vestas V27 wind turbine. 6th International Operational Modal
Analysis Conference (IOMAC). Gijón, Spain.
Ramler, J. R. and R. M. Donovan (1979). Wind turbines for Electric Utilities: Development Status and
Economisc. N. T.-. DOE/NASA/1028-79/23, AIAA-79-0965.
Randall, R. B. and N. Sawalhi (2011). A New Method for Separating Discrete Components from a Signal.
Sound & Vibration 45(5): 6 - 9.
Rankine, W. J. M. (1865). On the Mechanical Principles of the Action of Propellers, Transactions of the
Institution of Naval Architects.
Reynders, E. (2009). System Identification and Modal analysis in Structural Mechanics. PhD Thesis, K.
U. Leuven.
387
References
Reynders, E., J. Houbrechts and G. De Roeck (2012). Fully automated (operational) modal analysis.
Mechanical Systems and Signal Processing 29(0): 228-250.
Rodrigues, J. (2004). Identificação Modal Estocástica. Métodos de análise e aplicações em estruturas de

engenharia civil. PhD thesis, Faculdade de Engenharia da Universidade do Porto (FEUP)
Laboratório Nacional de Engenharia Civil (LNEC).
Rücker, W. (2007). Offshore wind energy plants – Problems and possible solutions. Experimental
Vibration Analysis for Civil Engineering Structures (EVACES). Porto, Portugal.
Rytter, A. (1993). Vibrational Based Inspection of Civil Engineering Structures. PhD thesis, Aalborg
University.
Salavessa, R. (2014). Improving instead of repowering is a technical, ecological and economic solution.
Prototype - Rotor Blade Extension. European Wind Energy Association (EWEA) 2014 Annual Event.
Barcelona, Spain.
Salzman, D. J. C. and J. v. d. Tempel (2005). Aerodynamic Damping in the Design of Support Structures
for Offshore Wind Turbines. Copenhagen Offshore Wind 2005. Copenhagen, Denmark.
Savonius, S. J. (1931). The S-Rotor and Its Applications. Mechanical Engineering 53(5): 333-338.
Schwerin, B. (2010). Featured Invention: NASA Modeling Innovations Advance Wind-Energy Industry.
Ask Magazine: 22 - 24.
Scionti, M., J. Lanslots, I. Goethals, A. Vecchio, H. V. d. Auweraer, B. Peeters and B. D. Moor (2003).
Tools to improve detection of structural changes from in-flight flutter data. 8th International Conference
on Recent Advances in Structural Dynamics (ISVR). Southampton, UK.
Scionti, M. and J. P. Lanslots (2005). Stabilisation diagrams: Pole identification using fuzzy clustering
techniques. Advances in Engineering Software 36(11–12): 768-779.
Scots Renewables. (2011). No-Tiree-Array Group Visit An Offshore Windfarm. from

http://www.scotsrenewables.com/blog/offshorewind/no-tiree-array-group-visit-an-offshore-
windfarm/.
Sectorov, W. R. (1973). The first aerodynamic three-phase electric power plant in Balakawa, National
Aeronautics and Space Administration (NASA).
Şeker, S., E. Ayaz and E. Türkcan (2003). Elman's recurrent neural network applications to condition
monitoring in nuclear power plant and rotating machinery. Engineering Applications of Artificial
Intelligence 16(7–8): 647-656.
388
References
Shepherd, D. (1990). Historical Development of the Windmill. DOE/NASA/5266-1, National

Aeronautics and Space Administration (NASA).
Shi, W., J. Shan and X. Lu (2012). Modal identification of Shanghai World Financial Center both from
free and ambient vibration response. Engineering Structures(36): 14-26.
Shirzadeh, R., C. Devriendt, M. A. Bidakhvidi and P. Guillaume (2013). Experimental and

computational damping estimation of an offshore wind turbine on a monopile foundation. Journal of
Wind Engineering and Industrial Aerodynamics 120: 96-106.
Skjoldan, P. F. (2011). Aeroelastic modal dynamics of wind turbines including anisotropic effects. Risø-
PhD-Report-66. Roskilde, Denmark, Riso - DTU.
Sohn, H. (2007). Effects of environmental and operational variability on structural health monitoring.
Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences
365(1851): 539-560.
Sørensen, J. D. and J. N. Sørensen (2011). Wind Energy Systems: Optimising Design and Construction
for Safe and Reliable Operation. Cambridge, UK
Philadelphia, PA, Woodhead Publishing.
Sørensen, S. P. H. and L. B. Ibsen (2013). Assessment of foundation design for offshore monopiles
unprotected against scour. Ocean Engineering 63: 17-25.
Spera, D. (1994). Wind Turbine Technology - Fundamental Concepts of Wind Turbine Engineering, The
American Society of Mechanical Engineers.
Spera, D. A. (1995). Bibliography of NASA-Related Publications on Wind Turbine Technology 1973-

1995. DOE/NASA/5776-3 NASA CR-195462.
Sutherland, H. J. (2000). A summary of the fatigue properties of wind turbine materials. Wind Energy
3(1): 1-34.
Tarp-Johansen, N. J., L. Andersen, E. D. Christensen, C. Mørch, B. Kallesøe and S. Frandsen (2009).

Comparing Sources of Damping of Cross-Wind Motion. European Offshore Wind Conference &
Exhibition. Stockholm, Sweden.
Tcherniak, D., S. Chauhan and M. H. Hansen (2010a). Applicability Limits of Operational Modal
Analysis to Operational Wind Turbines. XXVIII International Modal Analysis Conference (IMAC).
Florida, USA.
389
References
Tcherniak, D., S. Chauhan, M. Rossetti, I. Font, J. Basurko and O. Salgado (2010b). Output-only Modal
Analysis on Operating Wind Turbines: Application to Simulated Data. European Wind Energy
Conference. Warsaw, Poland.
Tegen, S., M. Hand, B. Maples, E. Lantz, P. Schwabe and A. Smith (2012). 2010 Cost of Wind Energy
Review, National Renweable Energy Laboratory (NREL).
Tempel, J. V. D. (2006). Design of Support Structures for Offshore Wind Turbines. PhD Thesis,
Technische Universiteit Delft.
Thiele, H. M. (1984). Growian Rotor Blades: Production Development, Construction and Test. NASA
TM-77479.
Thies, P. R., L. Johanning, V. Harnois, H. C. M. Smith and D. N. Parish (2014). Mooring line fatigue
damage evaluation for floating marine energy converters: Field measurements and prediction.
Renewable Energy 63(0): 133-144.
Thomsen, J. H., T. Forsberg and R. Bittmer (2007). Offshore Wind Turbine Foundations - The Cowi
Experience. 26th International Conference on Offshore Mechanics and Artic Engineering (OMAE).
California, USA.
Thomsen, K., J. T. Petersen, E. Nim, S. Øye and B. Petersen (2000). A Method for Determination of
Damping for Edgewise Blade Vibrations. Wind Energy(3): 233-246.
Thresher, R. W. and D. M. Dodge (1998). Trends in the evolution of wind turbine generator
configurations and systems. Wind ENERGY 1(S1): 70 - 86.
Tong, W. (2010). Wind Power Generation and Wind Turbine Design. Great Britain, WIT Press.
Tricklebank, A. H., P. H. Halberstadt, B. J. Magee and A. Bromage (2007). Concrete Towers for
Onshore and Offshore Wind Farms.
U.S Energy Information Administration (2015). Levelized Cost and Levelized Avoided Cost of New
Generation Resources in the Annual Energy Outlook 2015. Washington, DC.
Ulriksen, M. D., D. Tcherniak, P. H. Kirkegaard and L. Damkilde (2015). EWSHM 2014: Operational
modal analysis and wavelet transformation for damage identification in wind turbine blades. Structural
Health Monitoring.
van Kuik, G. A. M. (2007). The Lanchester–Betz–Joukowsky limit. Wind Energy 10(3): 289-291.
Vargo, D. J. (1974). Wind Energy Developments on the 20th Centurty. Missouri, USA.
390
References
Veljkovic, M., C. Heistermann, W. Husson, M. Limam, M. Feldmann, J. Naumes, D. Pak, T. Faber, M.

Klose, K.-U. Fruhner, L. Krutschinna, C. Baniotopoulos, I. Lavasas, A. Pontes, E. Ribeiro, M. Hadden,
R. Sousa, L. d. Silva, C. Rebelo, R. Simoes, J. Henriques, R. Matos, J. Nuutinen and H. Kinnunen
(2012). High-strength tower in steel for wind turbines (Histwin). Luxembourg.
Ventura, C. E., W. D. L. Finn, J.-F. Lord and N. Fujita (2003). Dynamic characteristics of a base isolated
building from ambient vibration measurements and low level earthquake shaking. Soil Dynamics and
Earthquake Engineering(23): 313-322.
Verboven, P., P. Guillaume, B. Cauberghe, E. Parloo and S. Vanlanduit (2003). Stabilization Charts
and Uncertainty Bounds for Frequency-Domain Linear Least Squares Estimators. XXI International
Modal Analysis Conference (IMAC). Florida, USA.
Verboven, P., E. Parloo, P. Guillaume and M. Van Overmeire (2002). Autonomous Structural Heath
Monitoring - Part I: Modal Parameter Estimation and Tracking. Mechanical Systems and Signal
Processing 16(4): 637-657.
Versteijlen, W. G., V. Metrikine, J. S. Hoving, E. Smid and W. E. d. Vries (2011). Estimation of the
Vibration Decrement of an Offshore Wind Turbine Support Structure Caused by its Interaction with
Soil. EWEA Offshore Conference. Amsterdam, The Netherlands.
Vestergaard, J., L. Brandstrup and R. D. G. III (2004). A Brief History of the Wind Turbine Industries in
Denmark and the United States. Academy of International Business (Southeast USA Chapter).
Knoxville, USA: 322-327.
Voaden, G. H. (1943). The Smith-Putnam Wind Turbine. Turbine Topics 1(3): 1-6.
Vugts, J. H. (2000). Considerations on the dynamics of support structures for an OW EC. PhD thesis,
Delft University of Technology.
Wang, Y., M. Liang and J. Xiang (2014). Damage detection method for wind turbine blades based on
dynamics analysis and mode shape difference curvature information. Mechanical Systems and Signal
Processing 48(1–2): 351-367.
Weijtjens, W. (2014). Advanced methods for estimating and monitoring the modal parameters of
operational systems subjected to non-white excitation. PhD thesis, Vrije Universiteit Brussel.
Weijtjens, W., A. Iliopoulos, J. Helsen and C. Devriendt (2015). Monitoring the consumed fatigue life of
wind turbines on a monopile foundation. EWEA Offshore 2015. Copenhagen, Denmark.
Weijtjens, W., T. Verbelen, G. De Sitter and C. Devriendt (2015). Foundation structural health
monitoring of an offshore wind turbine-a full-scale case study. Structural Health Monitoring.
391
References
White, J. R., D. E. Adams and M. A. Rumsey (2010). Modal Analysis of CX-100 Rotor Blade and Micon
65/13 Wind Turbine. XXVIII International Modal Analysis Conference (IMAC). Florida, USA.
Wiggelinkhuizen, E. J., S. J. Watson, J. Xiang, G. Giebel, E. J. Norton, M. C. Tipluica, A. J. Christensen

and E. Becker (2007). Conmow Final Report. ECN-E--07-044, Energy Research Centre of the
Netherlands (ECN).
Wikipedia. (2005). Eoliennes Gaspesie. 2013, from

http://en.wikipedia.org/wiki/File:Eoliennes_Gaspesie.jpg.
Wikipedia. (2009). NASA wind turbines. 2013, from

http://en.wikipedia.org/wiki/NASA_wind_turbines.
Wikipedia. (2012). Tripod wind energy Brhv. 2013, from

http://commons.wikimedia.org/wiki/File:Tripod_wind_energy_Brhv.jpg.
Wilkinson, M. and B. Hendriks (2011). Report on Wind Turbine Reliability Profiles, Reliawind.
Wilson, R. E. and P. B. S. Lissaman (1974). Applied Aerodynamics of Wind Power Machine. Oregon
State University.
Wind Directions (January/ February 2007). European Wind Energy Asscociation (EWEA).
Wind Power Monthly. (2003). Wind and insurance industries head for peace treaty. 2013, from
http://www.windpowermonthly.com/article/955632/wind-insurance-industries-head-peace-treaty.
Worden, K., C. R. Farrar, G. Manson and G. Park (2007). The fundamental axioms of structural health
monitoring. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Science
463(2082): 1639-1664.
World Energy Council (WEC) (2013). World Energy Perspective - Cost of Energy Technologies. London,
UK.
Wymore, M. L., J. E. Van Dam, H. Ceylan and D. Qiao (2015). A survey of health monitoring systems
for wind turbines. Renewable and Sustainable Energy Reviews 52: 976-990.
Yang, B. and D. Sun (2013). Testing, inspecting and monitoring technologies for wind turbine blades: A
survey. Renewable and Sustainable Energy Reviews(22): 515-526.
Yang, S. and M. S. Allen (2012). Output-only Modal Analysis using Continuous-Scan Laser Doppler
Vibrometry and applicationto a 20 kW wind turbine. Mechanical Systems and Signal Processing 31:
228-245.
392
References
Yang, W., P. J. Tavner, C. J. Crabtree, Y. Feng and Y. Qiu (2014). Wind turbine condition monitoring:
technical and commercial challenges. Wind Energy(17): 673-693.
Ye, X. W., Y. Q. Ni, K. Y. Wong and J. M. Ko (2012). Statistical analysis of stress spectra for fatigue life
assessment of steel bridges with structural health monitoring data. Engineering Structures 45(0): 166-
176.
Yoo, H. H. and S. H. Shin (1998). Vibration Analysis of Rotating Cantilever Beam. Journal of Sound
and Vibration 212(5): 807-828.
Yoon, S. Y., Z. Lin and P. E. Allaire (2013). Control of Surge in Centrifugal Compressors by Active
Magnetic Bearings. London, England, Springer.
Young, W. C. and R. G. Buynas (2002). Roark's Formulas for Stress and Strain. The United States of
America, McGraw-Hill.
Zaaijer, M., J. v. d. Tempel, P. Fish, T. Topper and A. Vos (2002). Tripod support structure - pre-design
and natural frequency assessment for the 6 MW DOWEC. Delft, the Netherlands, TUDelft.
Zhang, L., R. Brincker and P. Andersen (2005). An Overview of Operational Modal Analysis: Major
Development and Issues. International Operational Modal Analysis. Copenhagen, Denmark.
393

VIBRATION-BASED STRUCTURAL HEALTH MONITORING OF WIND TURBINES Gustavo

Uploaded by

Copyright:

Available Formats

VIBRATION-BASED STRUCTURAL HEALTH MONITORING OF WIND TURBINES Gustavo

Uploaded by

Document Information

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

VIBRATION-BASED STRUCTURAL HEALTH MONITORING OF WIND TURBINES Gustavo

Uploaded by

Copyright:

Available Formats

Gustavo Miguel Cameira da Silva Oliveira

A dissertation presented to the Faculty of Engineering of the University of Porto

Supervisors: Álvaro Cunha (Full professor); Filipe Magalhães (Assistant professor);

‫ܣ‬ Area of cross section

‫ܣ‬, ‫ܤ‬, ‫ܥ‬, ‫ܦ‬ State-space model matrices

‫ܥ‬ଵ Damping matrix

‫ܥ‬஽ Drag coefficient

‫ܥ‬௅ Lift coefficient

‫ܥ‬௉ Power coefficient

‫்ܥ‬ Thrust coefficient

‫ܦ‬ Diameter, Damage

‫ܧ‬ Young modulus

‫ܨ‬ Excitation force

‫ܩ‬ Next-state output correlation matrix, Gyroscopic matrix

‫ܪ‬ Frequency response function

‫ܪ‬௦ Significant wave height

‫ܫ‬ Identity matrix, Second moment of inertia of the cross section

‫ܭ‬ Stiffness matrix, Kalman gain

‫ܭ‬஼ Geometric stiffness matrix

‫ܭ‬௠ Modal Kalman gain

‫ܯ‬ Mass matrix, Bending moment

ܱ Extended observability matrix

ܳ, ܴ, ܵ Noise covariance matrices

ܴ Correlation matrix, Radius, Rotation matrix

ܵ Spectrum matrix, Scour depth

ܶ Torque, Period of time

Mean zero-crossing period

Modal output matrix

Matrix with predictor variables

Axial induction factor

Angular induction factor

Number of outputs sensors

Mass flow rate

Number of reference outputs

Position of the blade under consideration

Vectors containing noise due to sensor inaccuracy

Vector containing noise due to disturbances and modelling inaccuracies

, Modal state vector at instant

̅ Mean value of the distribution

Output vector, displacement (in time domain)

, Modal state vector

Γ Reversed extended stochastic controllability matrix

Λ Modal state matrix

Ψ Matrix with eigenvectors

Ω Rotational frequency of the rotor

Vector with weight parameters of the predictor variables

Tip speed ratio, system poles

Circular frequency, angular velocity of the wake

∗ Diagonal elements of a matrix

⊗ Kronecker product operator

CAPEX Capital Expenditure

FEM Finite Element Model

LCL Lower Control Limit

MAC Modal Assurance Criterion

MBC Multi-Blade Coordinate

OMA Operational Modal Analysis

OPEX Operational Expenditure

PDF Probability Density Function

RBE Rotor Blade Extension

RMS Root Mean Square