Doctoral theses at NTNU, 2020:356

Doctoral thesis
​

John Marius Hegseth

John Marius Hegseth

​

Efficient Modelling and Design

Optimization of Large Floating
Wind Turbines
​
​

ISBN 978-82-326-5058-3 (printed ver.)

ISBN 978-82-326-5059-0 (electronic ver.)
ISSN 1503-8181

NTNU
Norwegian University of
Science and Technology
Thesis for the degree of
Philosophiae Doctor
Faculty of Engineering
Department of Marine Technology
Doctoral theses at NTNU, 2020:356
John Marius Hegseth

Efficient Modelling and Design

Optimization of Large Floating
Wind Turbines
​
​

Thesis for the degree of Philosophiae Doctor

​
Trondheim, November 2020
​
Norwegian University of Science and Technology
Faculty of Engineering
Department of Marine Technology
NTNU
Norwegian University of Science and Technology
​
Thesis for the degree of Philosophiae Doctor
​
Faculty of Engineering
Department of Marine Technology
​
© John Marius Hegseth
​
ISBN 978-82-326-5058-3 (printed ver.)
ISBN 978-82-326-5059-0 (electronic ver.)
ISSN 1503-8181
​
Doctoral theses at NTNU, 2020:356

NO - 1598

Printed by Skipnes Kommunikasjon AS

Abstract

Floating wind turbines (FWTs) are considered a promising solution for wind
energy harvesting in deep water, but are currently too expensive to compete
with other energy sources. Being a relatively new and immature technology
means that there still is a large potential for cost reductions through optim-
ization of the FWT structure. Optimized designs will bring the construction
costs of FWTs down and increase profitability, which is currently the major
challenge for the industry. Optimized designs can also result in increased
reliability, which is an important issue.
A FWT may encounter a large number of different loading situations dur-
ing its lifetime, including operational, parked, fault, start up, and shutdown
events. In each of these situations, various combinations of environmental
loads from wind and waves must be considered. Combined with computa-
tionally expensive response analyses, this results in a comprehensive design
process.
FWTs are also highly multidisciplinary systems, primarily combining the
areas of aerodynamics, hydrodynamics, structural dynamics, and control
theory. The interactions between the disciplines, and between different
components in the system, calls for integrated analysis and design. Due
to the complex dynamics, strong couplings, and potentially large number
of design variables, identifying optimal design solutions become a difficult
task.
The main purpose of this work was to improve the design process for FWTs
and thus contribute to reducing the cost of energy. This is addressed through
two overall research objectives, which consider i) increased computational
efficiency of global design analyses for FWTs, and ii) methods for numerical

iii
iv

design optimization which can help identify cost-effective and reliable design
solutions. The main focus was on the support structure and controller for
10 MW spar-type turbines, considering fatigue and ultimate loads.
A linearized aero-hydro-servo-elastic model was shown to yield good results
for the fatigue loads in the support structure, where agreement within ±
30 % was achieved for the long-term fatigue damage compared to nonlin-
ear time-domain simulations. Acceptable agreement was also observed for
short-term extreme response, especially for the support structure bending
moments, which were quite Gaussian also in harsh environmental condi-
tions. The resonant platform pitch response was overestimated by the linear
model, especially in near-rated conditions.
A gradient-based optimization approach with analytic derivatives was de-
veloped to perform integrated design optimization of the support structure,
blade-pitch controller, and mooring system for an elastic 10 MW spar FWT,
including the scantling design of the hull, where the goal was to minimize
a combination of design costs and rotor speed variation. Different con-
trol strategies were compared through integrated design of the controller
and support structure, which allowed for identification of optimal control
parameters in a lifetime perspective, and fair comparisons between different
strategies.
The impact of environmental modelling on the long-term fatigue reliability,
and associated design costs, of the support structure was assessed through
re-design of the tower and platform. Considering stochastic turbulence in-
tensity, wind-wave misalignment and the wind directional distribution re-
duced the long-term fatigue damage by approximately two-thirds along the
support structure, compared to the base model. Implications of the chosen
fatigue safety factor on the trade-offs between CAPEX and OPEX were also
assessed.
The methodologies for global response analyses and integrated design op-
timization developed in the present work have been shown to be suitable for
preliminary design of spar FWTs, where they can provide a starting design
for later and more detailed design phases. Different modelling and design
aspects for cost-effective and reliable solutions have been identified and as-
sessed. The methodologies can be further extended to account for different
FWT concepts, additional design parameters, and other load cases, and
may help identify novel design solutions.
Acknowledgements

This thesis is submitted in partial fulfilment of the requirements for the

degree of philisophiae doctor (PhD) at the Norwegian University of Science
and Technology (NTNU). The research has been carried out at the Depart-
ment of Marine Technology, in affiliation with the Centre for Autonomous
Marine Operations and Systems (AMOS). Support has been provided by
the Research Council of Norway (NFR) through project number 274827 -
‘Green Energy at Sea’.
Firstly, I would like to express my gratitude to my supervisor, Professor
Erin E. Bachynski. Her commitment and genuine interest in my research
have been invaluable throughout this PhD, and despite a busy schedule, her
door has always been open for discussions or advice. I am hugely grateful
for the patience she has shown with me and my problems along the way,
and for the thorough and prompt feedback she has provided on my work.
Her incredible knowledge and passion for offshore wind energy research have
truly been inspiring.
During my PhD I was invited to spend three months with the MDO Lab
at the University of Michigan. I would like to thank my co-supervisor, Pro-
fessor Joaquim R. R. A. Martins, for welcoming me into his lab, for sharing
some of his vast knowledge about multidisciplinary design optimization, and
for taking me running on the beautiful trails of Pinckney Recreation Area.
The enthusiastic help of Dr. John Jasa was crucial for the work concerning
OpenMDAO. A word of thanks goes also to the rest of the lab for making
my stay so enjoyable.
At NTNU, my co-supervisor Professor Michael Muskulus is acknowledged
for his ideas and feedback. I am also grateful to Professor Bernt J. Leira

v
vi

for discussions and guidance on the work concerning structural reliability.

Furthermore, Bjørn Tore Bach deserves thanks for the hours spent setting up
the computers that was used to perform many of the numerical calculations.
I have enjoyed many interesting conversations with my office mate for three
years, Øyvind Rabliås, spanning everything from potential flow theory to
cross-country skiing. Carlos Eduardo Silva de Souza has been a frequent
discussion partner, especially on the topic of wind turbine control, which
resulted in a co-authored paper (and some sightseeing in Boston). I also
appreciate the friendship and discussions I have shared with my fellow PhD
students and other colleagues at the Department of Marine Technology.
Finally, I would like to thank my family - my parents and my sisters - who
have provided lifelong support and encouragement, and my friends, who
made sure I spent most of my weekends running or skiing rather than at
the office.
Publications

The thesis consists of an introductory part and a collection of papers. The

papers that form the basis of the thesis, as well as additional papers and a
declaration of authorship, is presented in the following.

List of publications
The following five papers are considered part of this thesis:

P1 John Marius Hegseth and Erin E. Bachynski (2019). A semi-analytical

frequency domain model for efficient design evaluation of spar floating
wind turbines. Marine Structures, 64, 186-210.

P2 Carlos Eduardo S. Souza, John Marius Hegseth and Erin E. Bachynski

(2020). Frequency-dependent aerodynamic damping and inertia in lin-
earized dynamic analysis of floating wind turbines. Journal of Physics:
Conference Series, 1452, 012040.

P3 John Marius Hegseth, Erin E. Bachynski and Joaquim R. R. A. Mar-

tins (2020). Integrated design optimization of spar floating wind tur-
bines. Marine Structures, 72, 102771.

P4 John Marius Hegseth, Erin E. Bachynski and Joaquim R. R. A. Mar-

tins (2020). Design optimization of spar floating wind turbines con-
sidering different control strategies. Journal of Physics: Conference
Series. Accepted for publication.

P5 John Marius Hegseth, Erin E. Bachynski and Bernt J. Leira (2020).

Effect of environmental modelling and inspection strategy on the op-

vii
viii

timal design of floating wind turbines. Submitted to Reliability Engin-

eering and System Safety.

The following paper is not regarded as part of the thesis due to scope:

P6 John Marius Hegseth, Erin E. Bachynski and Madjid Karimirad (2018).

Comparison and validation of hydrodynamic load models for a semi-
submersible floating wind turbine. Proceedings of the ASME 2018 37th
International Conference on Ocean, Offshore and Arctic Engineering
(OMAE2018), Madrid, Spain.
 ix

Declaration of authorship
In P1, P3, P4, and P5, J. M. Hegseth had the main ideas for the papers,
developed the numerical models, performed all calculations, and wrote the
manuscripts. Profs. E. E. Bachynski, J. R. R. A. Martins and B. J. Leira
contributed with valuable suggestions and feedback during the processes.
P2 was initiated by C. E. Souza, who also developed the forced oscillation
method and wrote the manuscript. J. M. Hegseth derived the analytic
expressions for the aerodynamic damping and inertia coefficients, developed
the linearized model, and participated in discussions. Prof. E. E. Bachynski
provided guidance and feedback.
x
Contents

Abstract iii

Acknowledgements v

Publications vii

1 Introduction 1
1.1 Motivation and background . . . . . . . . . . . . . . . . . . . 1
1.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Main contributions . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Literature Survey 11
2.1 Global response analyses . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Environmental modelling . . . . . . . . . . . . . . . . 11
2.1.2 State-of-the art simulations . . . . . . . . . . . . . . . 14
2.1.3 Linear models . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.4 Structural reliability . . . . . . . . . . . . . . . . . . . 20
2.2 Design optimization . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Basic optimization theory . . . . . . . . . . . . . . . . 21

xi
xii CONTENTS

2.2.2 Offshore wind substructures . . . . . . . . . . . . . . . 23

2.2.3 Integrated design . . . . . . . . . . . . . . . . . . . . . 25
2.2.4 Formulation of objective and constraints . . . . . . . . 29

3 Numerical Models 35
3.1 Linearized formulation . . . . . . . . . . . . . . . . . . . . . . 35
3.2 System overview . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Structural dynamics . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6 Control system description . . . . . . . . . . . . . . . . . . . . 45
3.7 Response to stochastic input . . . . . . . . . . . . . . . . . . 47
3.7.1 Fatigue damage . . . . . . . . . . . . . . . . . . . . . . 47
3.7.2 Extreme response . . . . . . . . . . . . . . . . . . . . . 49
3.8 Optimization framework . . . . . . . . . . . . . . . . . . . . . 49
3.9 Model development . . . . . . . . . . . . . . . . . . . . . . . . 54

4 Research Findings 55
4.1 Linearized dynamic analyses . . . . . . . . . . . . . . . . . . . 55
4.1.1 Aerodynamic damping and inertia . . . . . . . . . . . 55
4.1.2 Long-term fatigue damage . . . . . . . . . . . . . . . . 59
4.1.3 Short-term extreme response . . . . . . . . . . . . . . 61
4.2 Integrated design optimization . . . . . . . . . . . . . . . . . 64
4.2.1 Multimodality . . . . . . . . . . . . . . . . . . . . . . 64
4.2.2 Cost and power quality . . . . . . . . . . . . . . . . . 65
4.2.3 Support structure design . . . . . . . . . . . . . . . . . 66
4.2.4 Control strategies . . . . . . . . . . . . . . . . . . . . . 72
 CONTENTS xiii

4.3 Environmental modelling . . . . . . . . . . . . . . . . . . . . 75

4.4 Inspection schedule . . . . . . . . . . . . . . . . . . . . . . . . 77

5 Conclusions and Recommendations for Future Work 79

5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Recommendations for future work . . . . . . . . . . . . . . . 81

A Appended Papers 99

B Formulation of Rotor Effective Wind Speed 221

xiv CONTENTS
Chapter 1

Introduction

This chapter gives an overview of the background and motivation for this
work, and states the objectives of the research. A summary of the main
contributions to the scientific community, as well as their relations to the
research objectives and appended papers, are also presented.

1.1 Motivation and background

The high global demand for energy has led to a large consumption of fossil
fuels, with several negative consequences, most notably climate change. The
last decades have seen a large increase in the use of wind energy, which more
recently has moved offshore. Offshore wind is an attractive energy source
due to higher wind speeds with less turbulence, larger areas less susceptible
to conflicts with other interests, and with smaller environmental impact
than its land-based counterpart.
The majority of energy producing offshore wind turbines installed to date
are bottom-fixed, and located in shallow waters. Because suitable sites for
such wind parks are limited, it is desirable to develop offshore wind farms
in deeper waters. In water depths above 50-60 m, bottom-fixed structures
are not economically feasible, and FWTs are expected to be the most cost-
effective option (Musial et al. 2006). As 80 % of the offshore wind resources
in Europe are located at water depths above 60 m (The Carbon Trust 2015),
there is a vast potential market for floating wind power, which could become
a crucial part of the energy mixture of the future.
The technical potential floating wind capacity across Europe, USA, and
Japan is estimated to about 7000 GW, where 4000 GW is located in Europe

1
2 Introduction

Table 1.1: Floating wind farms in Europe with expected commissioning date
within 2022 (WindEurope 2019).

Floater Capacity No. of Water

Country Wind farm
type (MW) turbines depth (m)
Windfloat
Portugal Atlantic Semi-sub 25.0 3 100
Phase 1
France EolMed Barge 24.0 4 55
Provence
France TLP 28.5 3 100
Grand Large
France EFGL Semi-sub 30.0 3 65-80
Eoliennes
France Flottantes TLP 28.5 3 60
de Groix
UK Kincardine Semi-sub 50.0 5 60-80
Hywind
Norway Spar buoy 88.0 11 260-300
Tampen

(The Carbon Trust 2015). However, moving the energy production to deeper
water and larger distances from shore increases the complexity and requires
innovative technology. To date, the world’s only floating wind farm is the
Hywind Scotland pilot park, which consists of five turbines with an installed
capacity of 30 MW. However, more than 30 different FWT concepts have
been proposed (Leimeister et al. 2018), and several floating wind farms are
planned to start operation in the near future. Seven European farms, listed
in Table 1.1, are expected to be commissioned within the next three years,
with a total capacity of 274 MW.
The cost of floating wind is currently higher than for bottom-fixed turbines,
and large reductions are needed to reach a competitive level. Floating wind
turbines have larger construction costs than the bottom-fixed alternative;
however, installation costs are expected to be smaller, because the turbines
can be assembled at port using onshore cranes instead of expensive jack-up
vessels at site (Katsouris and Marina 2016). The current levelized cost of
energy (LCOE) is estimated to e180-200/MWh for pre-commercial projects,
but large reductions are expected as the technology matures and reaches
commercial-scale deployment, and may attain values of e40-60/MWh in
2030 (WindEurope 2018).
Lower LCOE can be achieved in several ways, such as reduction of capital
 1.1. Motivation and background 3

costs (CAPEX) or operation and maintenance costs (OPEX), increase in

the capacity factor, or through lifetime extension. As floating wind is a
relatively new and immature technology, there is still a large potential for
cost reductions through optimization of the FWT structure. A breakdown
of typical CAPEX for floating wind farms is shown in Fig. 1.1, where the
turbine and platform are seen to comprise more than 60 % of the total
costs, suggesting that significant cost reductions in these components may
be necessary to reach a sustainable and competitive LCOE. This is also
supported by feedback from concept designers, who identify platform size
as the most critical technical barrier with respect to cost savings for floating
wind farms (The Carbon Trust 2015).
As the technology matures, turbines grow larger, and during the last decade,
the average capacity of newly installed offshore wind turbines in Europe has
more than doubled, from 3 MW in 2009 to 7.8 MW in 2019 (WindEurope
2019). The turbines are expected to grow further as the technology im-
proves, and could reach 15-20 MW by 2030 (IEA 2019). The increase in
turbine size is expected to be the largest driver of cost reduction, because it
allows the same amount of energy to be captured with fewer machines, and
thus reduces the balance-of-system costs (Smith et al. 2015). In addition, it
results in lower relative costs of the substructure, tower and mooring costs,
compared to a larger number of turbines with smaller rating (Müller et al.
2019). The increase in size also introduces new challenges and opportunities
for design, with innovative turbine and substructure technologies having a
large impact on the achieved cost reductions (Chaviaropoulos et al. 2014).
It may also help new materials, such as concrete, become more attractive
solutions (Müller et al. 2016).
A FWT is subjected to a large number of different loading conditions during
its lifetime, involving a highly stochastic environment with irregular loads
from wind and waves. According to DNV GL (2018a), the structure shall
be designed to sustain all loads during temporary, operating, and damaged
conditions, ensure safe operation of the wind turbine, maintain acceptable
safety for personnel and environment, and have sufficient durability against
deterioration. Because FWT structures are unmanned, the consequences of
failure are normally limited to economic losses. Failures may, however, also
affect the environment or nearby structures, and FWTs should be designed
for an annual target probability of failure of 10−4 (DNV GL 2018a).
Four different limit states are considered in design (DNV GL 2016):
4 Introduction

Figure 1.1: Estimated breakdown of capital costs for commercial scale floating
wind farms (The Carbon Trust 2015).

Ultimate limit state (ULS): Exceedance of maximum load

carrying resistance.
Fatigue limit state (FLS): Failure due to cyclic loading.
Accidental limit state (ALS): Failure due to accidental or
abnormal loads.
Serviceability limit state (SLS): Exceedance of criteria for normal
use or durability.

Offshore wind turbines are highly dynamic systems subjected to static, peri-
odic, stochastic, and transient loads. Combined with compliant structures
and a relatively long design lifetime, this often leads to FLS being design
driving for several components (Vorpahl et al. 2013). Contrary to the design
of conventional offshore structures, which tends to be driven by ultimate
loads, this is also the case for the floating platform (Strach-Sonsalla and
Muskulus 2016). To give a realistic estimate of the fatigue life of the struc-
ture, all relevant environmental conditions throughout the lifetime of the
system must be considered, which leads to a comprehensive set of design
calculations. This is especially the case if directionality of wind and waves
is considered, which is of particular importance for fatigue design (DNV GL
2018a).
In addition to fatigue, the survival of the platform and turbine in extreme
 1.1. Motivation and background 5

conditions must be ensured. Typical ULS criteria consider overturning,

capsizing, yielding, and buckling (DNV GL 2018a). Further constraints may
be applied to prevent water on deck, negative air gap, and sloshing inside
ballast tanks during storm events (Strach-Sonsalla and Muskulus 2016).
Failure caused by accidental loads, or exceedance of ultimate resistance of
damaged structures, is covered by ALS criteria (DNV GL 2018a), and is
not considered in this thesis.
Lastly, SLS deals with criteria to ensure normal operation of the turbine. It
involves limitations on inclinations, accelerations and motions due to sensit-
ive equipment and the power cable (Strach-Sonsalla and Muskulus 2016), in
addition to ensurement of sufficient hydrostatic stability for efficient power
production.
In general, the design criteria can be written

Sd ≤ R d , (1.1)

where Sd is the design load effect, and Rd is the design resistance. These
are usually found using the partial safety factor method, where the different
characteristic load effects and resistances are multiplied by individual load
and material factors to obtain their design values. The safety factors reflect
the uncertainty in the respective design parameters, as well as the required
reliability level.
Probability-based design is an alternative to the partial safety factor method,
where the load and resistance variables are described by their probability
distributions. The design of the system is then performed to meet a safety
level which is expressed by the failure probability:

PF = P [S > R], (1.2)

which should be equal to the annual target failure probability (DNV GL

2018a). In practical applications, the partial safety factor method is by far
the most popular, and probability-based design is mainly used to calibrate
safety factors, or in design of systems where limited experience is available
(DNV GL 2018b).
To evaluate the structure’s margin of safety against the different limit states,
an extensive number of design load cases (DLCs) must be assessed. Relevant
DLCs for FWTs are defined by DNV GL (2016), and consider eight distinct
design situations:

• Power production
6 Introduction

• Power production plus fault

• Start up

• Normal shutdown

• Emergency shutdown

• Parked turbine

• Parked turbine plus fault

• Transport, installation and repair

Within each of these design situations, a subset of load cases are specified.
The subsets consider different wind and wave conditions, fault types, and
limit states, and the total number of load cases in a design process may
therefore add up to several thousands.
The goal of the design optimization process is to find the design that min-
imizes or maximizes a given performance measure, such as costs or power
production, and at the same time meets the appropriate design criteria. In
traditional design optimization, the performance of an initial design is eval-
uated through analyses or experiments. Based on the results, the designer
makes changes to the design, and the process is repeated until a satisfact-
ory design is achieved. This process is typically limited to a few iterations,
where changes to the design to a large extent depend on the experience of
the designer (Muskulus and Schafhirt 2014).
In contrast, numerical optimization techniques can efficiently explore large
parts of the design space to identify improved and possibly non-intuitive
solutions, as illustrated in Fig. 1.2. Such techniques have been widely util-
ized in the automotive and aerospace industries, but have seen limited use
in design of wind turbine structures (Muskulus and Schafhirt 2014). Simil-
arly to a conventional design process, the performance of an initial design is
evaluated based on results from analyses, and iterations are performed until
a satisfactory solution has been achieved. The key differences are that the
design changes are made automatically by an optimization algorithm, and
that the design process does not end until certain optimality criteria are
satisfied. These criteria ensure that the final design not only complies with
the relevant constraints, but that it indeed is the solution that minimizes
or maximizes the chosen performance measure, at least in that particular
neighbourhood of the design space (Martins and Ning 2020).
 1.1. Motivation and background 7

Optimization prob-
lem formulation

Initial design Initial design

Analysis of design Analysis of design

Check performance Check performance

Few iteration steps

Many iteration steps

and constraints and constraints

Yes Yes Satisfaction of

Satisfactory design? Final design
convergence criteria?

No No

Update design Update design using

based on experience optimization algorithm

Figure 1.2: Comparison of main steps in a manual and numerical design optim-
ization process. Adapted from Muskulus and Schafhirt (2014).

Although these techniques enable automation of the design process, they

do not remove the need for field-specific expertise from the designer. The
outcome of a numerical optimization procedure relies heavily on the for-
mulation of the optimization problem. An appropriate objective function
must be chosen, whose value is the only measure of the performance for a
given design. Further, relevant constraints should be identified, and proper
enforcement strategies must be defined. Finally, a parametrization of the
design space is needed, which determines the possibilities for the optimiz-
ation algorithm to modify the design. The successful application of such
methods thus requires profound knowledge about the nature of the system
that is considered, and possibly more so than in a traditional design process
(Martins and Ning 2020).
As a FWT system experiences important interactions between several differ-
ent subsystems and disciplines, integrated design is needed to find optimal
solutions on the overall system level (Muskulus and Schafhirt 2014). In an
8 Introduction

integrated design process, all relevant disciplines are coupled, and the sub-
systems are designed simultaneously. An integrated optimization approach
involves a large number of design variables and load cases, and requires
sophisticated and multidisciplinary numerical tools. Although the overall
system in terms of total cost and energy production is the whole wind farm,
the studies are limited to a single turbine in the present work, and farm-level
effects are not considered.

1.2 Research objectives

The overall aim of the thesis is to improve the efficiency and reliability of
the design process for FWTs, especially catenary-moored spar-type turbines,
and develop methodologies to identify novel and optimized design solutions.
This is addressed by defining two separate research objectives.
During the design process, a numerical model of the FWT system is created
and analysed. The global dynamic response is evaluated using fully coupled
aero-hydro-servo-elastic analyses, considering combined wind-wave loading.
The computational costs associated with such analyses are high, and due to
the variety of design situations and environmental conditions that a FWT
may encounter during its lifetime, a large number of load simulations must
be carried out. There is therefore a need for more efficient analysis tools,
especially during the conceptual and preliminary design phases. While sim-
plified simulation models may not be able to provide accurate response
predictions for all relevant design situations, they should capture important
interactions and response characteristics in the system, and provide consist-
ent design load estimates in comparisons with high-fidelity simulations for
applicable loading conditions.
The first research objective of this thesis is related to reduction of the com-
putational burden associated with global design analyses of FWTs, and is
formulated as:

O1 Suggest and verify methods to increase the computational efficiency

of global design analyses for floating wind turbines.

Due to the complex and multidisciplinary dynamic system, with a large

number of parameters and load cases, the design space cannot be efficiently
explored through traditional design procedures. The important couplings
in the system call for integrated optimization, which further complicates
the design problem. Numerical optimization techniques may help develop
 1.3. Main contributions 9

improved and possibly novel solutions, but have seen limited use for FWT
structures. Consequently, there is currently no software or framework avail-
able for efficient FWT optimization, and there are uncertainties related to
the formulation of the optimization problem. Design methodologies with
improved accuracy and efficiency, which can help identify innovative and
optimized solutions, are expected to result in cost reductions and increased
reliability, and can help floating wind technology become commercially feas-
ible.
The second research objective addresses the potential for design improve-
ments through numerical optimization, and is formulated as:

O2 Develop methodologies for numerical design optimization of floating

wind turbine systems, considering ultimate strength and fatigue, and
identify possible cost-effective and reliable design solutions.

The two research objectives are connected because contributions related

to O1 can be implemented in, and thus improve the feasibility of, models
developed in relation to O2. The objectives are achieved using numerical
simulations, which are performed for a 10 MW wind turbine rotor mounted
on a floating support structure1 . Because a limited amount of available and
relevant measurement data from full- or model-scale FWTs exists, results
from state-of-the-art numerical simulations are used as benchmark for the
models and methods developed over the course of the thesis.

1.3 Main contributions

Based on the research objectives defined in Section 1.2, several contributions
to the scientific community have been made. The main contributions are
further discussed in Chapter 4, and briefly summarized in the following.
The relations between research objectives, papers, and contributions are
illustrated in Fig. 1.3.

C1 A linearized aero-hydro-servo-elastic model for efficient global response

analyses of FWTs, extensively verified against state-of-the-art simula-
tions with regards to both fatigue damage and extreme response.
C2 Derivation of closed-form expressions for frequency-dependent aero-
dynamic damping and inertia, and an assessment of their impact on
frequency-domain calculations.
1
The term support structure corresponds to the combined tower and platform structure.
10 Introduction

C3 A methodology for integrated design optimization of spar FWTs, con-

sidering simultaneous design of platform, tower, controller, and moor-
ing system, using gradient-based optimization with analytic derivat-
ives.

C4 Evaluation and comparison of different control strategies for FWTs,

based on integrated optimization of support structure and blade-pitch
controller with realistic design limits.

C5 Quantification of the impact of environmental modelling on the long-

term fatigue reliability and costs of FWT support structures.

C6 Assessment of trade-off effects between CAPEX and OPEX, and de-

rivation of cost-optimal fatigue safety factors, considering reliability-
based inspection planning.

O1 O2
Increased Numerical
computational design
efficiency optimization

P1 P2 P3 P4 P5

C1 C2 C3 C4 C5 C6
Linearized Damping Integrated Control Modelling of Inspection
model and inertia optimization strategies environment planning

Figure 1.3: Relations between research objectives, papers, and contributions.

Chapter 2

Literature Survey

This chapter summarizes previous work relevant for global response ana-
lyses and numerical design optimization of FWT structures. Methods used
in state-of-the-art and simplified dynamic simulations are described, some
basic optimization theory is given, and specific aspects and challenges re-
lated to design optimization of FWTs are discussed.

2.1 Global response analyses

2.1.1 Environmental modelling
FWTs are subjected to stochastic environmental loading, which require a
proper statistical description for evaluation of lifetime loads in design. Tem-
poral environmental variations occur at many different time scales, which in
engineering applications are separated into short-term and long-term variab-
ility. In short-term conditions, the wind and wave processes are considered
stationary Gaussian stochastic processes, and are generally assumed to be
uncorrelated (DNV GL 2016). Short-term variations in wind speed and
wave elevation are commonly given in terms of their power spectral densit-
ies, which describe the distribution of energy at different frequencies. For
wind, a description of the spatial coherence is also needed. The long-term
statistics consider the variation of the statistical properties describing the
short-term process, and describe the correlation between wind and waves
through a joint probability distribution.
The joint distribution of wind and waves is ideally fitted to measured data
from the actual site over a long period of time, and is used to establish scat-
ter diagrams for fatigue calculations, or identify environmental conditions

11
12 Literature Survey

Figure 2.1: Relative wind-wave direction distributions for two wind speed bins
at a North Sea location 25 km west of Norway.

with a certain return period. As a minimum, the environmental model

should include distributions for the mean wind speed (U ) at a reference
height, the significant wave height (Hs ), and the spectral peak period (Tp ).
A joint distribution based on measurements from the Northern North Sea
was presented by Johannessen et al. (2002), using the marginal distribution
of the 1-h mean wind speed at 10 m (U10 ), and conditional distributions of
Hs and Tp :
fU10 ,Hs ,Tp (u, h, t) = fU10 (u)fHs |U10 (h|u)fTp |U10 ,Hs (t|u, h). (2.1)

It is also of interest to include directional data in the environmental model.

Waves are typically short-crested, which requires a spreading function in ad-
dition to the main propagation direction. Also, the wind and waves are in
general misaligned, and larger misalignment angles typically occur at smal-
ler wind speeds (Fischer et al. 2011). This is illustrated in Fig. 2.1, where
the wind-wave misalignment distribution, based on 60 years of hindcast data
from a location in the North Sea, is shown for two wind speed bins. While
relative wind-wave directions of more than 90◦ are quite rare for U10 > 10
m/s, larger misalignment angles occur more than 25 % of the time for the
lower wind speed bin. The relative wind-wave direction may have a large
effect on the response of the structure, and is of particular importance for
fatigue design (DNV GL 2018a). While misaligned wind and waves tend to
increase the tower fatigue damage for bottom-fixed turbines, the opposite
has been observed for FWTs (Barj et al. 2014, Bachynski et al. 2014).
The wind speed in a given point in space consists of longitudinal, lateral,
 2.1. Global response analyses 13

and vertical components, where the longitudinal component varies around

the mean wind speed, and the other two components typically are assumed
to have zero mean. Short-term variations in the wind speed are known as
turbulence, which are described by so-called turbulence spectra. A measure
of the overall level of turbulence is given by the turbulence intensity, which
is defined as the ratio between the standard deviation of the wind speed and
its mean. In general, the turbulence intensity takes values between 0.1 and
0.4, where higher values occur for lower wind speeds (Manwell et al. 2009).
If site-specific data is unavailable, values for the turbulence intensity can be
taken from design standards. The IEC Class B normal turbulence model
(NTM) (IEC 2005), which is applicable for moderate turbulence levels, is
used in the present work.
In addition to temporal variations, the turbulence also varies spatially, which
is described by the cross-correlations between fluctuations at points separ-
ated laterally and vertically (Burton et al. 2011). The resulting turbulent
wind field is assumed to follow Taylor’s frozen turbulence hypothesis (Taylor
1938), where the turbulence is considered as time-invariant eddies moving
forward with the mean wind speed. To generate the three-dimensional
turbulence box in time domain, several methods are available. A one-
dimensional fast Fourier Transform (FFT) can be applied, based on spectra,
to each of the three turbulence components independently, together with
a coherence function for the spatial cross-correlations. Alternatively, the
Mann model can be applied, where correlated components are generated
using a three-dimensional FFT (Burton et al. 2011). In the present thesis,
the Kaimal spectrum and an exponential coherence function are used to
describe the turbulent wind field (IEC 2005).
The mean wind speed also increases with the height above ground. This
variation is known as shear, and is commonly modelled using either a log-
arithmic or power law. The present work uses a power law with exponent
0.14, which is recommended for offshore locations (IEC 2009).
The physics of ocean waves are highly complex; however, for engineering ap-
plications, several simplifications can be made while still maintaining reas-
onable estimates of the wave dynamics. The waves are typically described
using linear (Airy) wave theory, which assumes potential (irrotational, invis-
cid, and incompressible) flow (Faltinsen 1990). The irregular wave elevation
process can then be generated from linear superposition of a large number
of independent wave components with different frequencies, where the amp-
litude of each component is found from the wave spectrum.
14 Literature Survey

The wave conditions in a sea state can be separated into two categories:
wind seas, which are generated by local wind, and swell, which are waves
that have travelled out of the areas where they were generated (DNV GL
2019a). For wind seas, the Pierson–Moskowitz (PM) and JONSWAP spec-
tra are frequently applied (DNV GL 2019a). The JONSWAP spectrum,
which was developed for North Sea conditions, is used in the present work.
If the sea state is expected to contain swell components in addition to wind-
generated waves, these must also be properly represented (DNV GL 2018a).
This can be done through two-peaked spectra, such as the Torsethaugen
model (Torsethaugen 1996, Torsethaugen and Haver 2004), which take into
account both wind and swell sea. Alternatively, the wind and swell compon-
ents can be separated and described using two JONSWAP spectra, possibly
with different directions, which are superimposed to represent the total wave
elevation process (DNV GL 2018a).
2.1.2 State-of-the art simulations
Due to important interactions between aerodynamics, hydrodynamics, struc-
tural dynamics, and control dynamics, fully coupled aero-hydro-servo-elastic
models are needed to properly simulate the global response of FWTs. Dif-
ferent aspects and interactions considered in such analyses are illustrated
in Fig. 2.2. Over the last decade, several fully coupled simulation tools
for global analysis of FWTs have been developed, often based on existing
codes for either conventional offshore structures or land-based wind tur-
bines. The present thesis uses the SIMA workbench developed by SINTEF
Ocean, which couples the RIFLEX and SIMO programs (SINTEF Ocean
2016a;b). A brief summary of common structural, aerodynamic, hydro-
dynamic, and control capabilities in state-of-the-art analysis tools is given
in the following.
The structural dynamics of FWTs are usually modelled using multibody dy-
namics, modal analysis, finite elements, or a combination of these (Robertson
et al. 2014). While modal analysis is computationally efficient, finite ele-
ments offer the possibility of geometric and material nonlinearities, and
do not require pre-calculation of the system modes. Most finite element
models for FWTs are limited to beam elements, which are applicable for
long, slender structures such as the tower and blades, whereas the large
volume platform typically is considered rigid. The drivetrain is usually
modelled using a multibody dynamics approach, where the shaft flexibility
often is included through a torsional spring-damper to capture the first tor-
sional mode of the drivetrain (Smilden 2019). The mooring lines are either
 2.1. Global response analyses 15

represented using nonlinear force-displacement relationships found from a

quasi-static analysis, or modelled explicitly using finite elements or lumped
masses. While the latter approach is computationally more expensive, it
captures the mooring line dynamics, and the effects of mooring line iner-
tia and damping on the platform response. The importance of mooring
line dynamics is highly dependent on the considered platform concept and
response parameters, but should be included for designs with natural fre-
quencies inside the wave-frequency range and fairleads far away from the
platform’s center of mass (Hall et al. 2014b), or if extreme mooring line
tension is considered (Azcona et al. 2017).
Various methods exists for calculation of aerodynamic loads on wind tur-
bines. For global response analyses of FWTs, the aerodynamic loads on
an operational turbine are commonly computed using blade element mo-
mentum (BEM) theory, due to its computational efficiency. Here, the aero-
dynamic lift and drag forces on the blades are balanced by the change in
axial and angular momentum of the air passing through the swept annulus
(Burton et al. 2011). However, BEM requires several engineering correc-
tions to give satisfactory results, where the most commonly applied are the
Glauert correction for high induction factors, and the Prandtl corrections
for tip and hub loss factors. Further, additional corrections are often im-
posed to account for skewed inflow conditions, dynamic wake, and dynamic
stall effects. The generalized dynamic wake (GDW) method is an altern-
ative to BEM, which is based on the potential flow solution of the Laplace
equation. GDW inherently includes skewed wake, tip losses, and dynamic
wake effects, but is only valid for lightly loaded rotors (high wind speeds),
and does not account for wake rotation (Moriarty and Hansen 2005).
Hydrodynamic loads on large volume floating structures are commonly com-
puted from potential flow theory, where the diffraction and radiation loads
are found numerically using a panel method. For many structures, the
first order solution, which assumes small wave amplitudes and small mo-
tions around the equilibrium position, is sufficient to capture the dominant
wave loads (Faltinsen 1990). The outputs from the hydrodynamic analysis
are frequency-dependent wave excitation, added mass, and wave radiation
damping, which are transformed to the time domain before they are ap-
plied in the simulations. For structures with natural frequencies outside
the wave-frequency range, higher-order wave loads may also be of import-
ance. Second-order loads from potential flow theory, which consist of sum-
and difference-frequency terms, can then also be included through so-called
quadratic transfer functions (QTFs). To capture important viscous effects,
16 Literature Survey

the quadratic drag term in Morison’s equation is often superimposed to the

potential flow solution in the aero-hydro-servo-elastic simulations.
The majority of modern horizontal axis wind turbines are regulated us-
ing variable-speed variable-pitch control. Below the rated wind speed, the
generator torque, and consequently the rotational speed of the turbine, is
varied to maximize the power output. Above the rated wind speed, the
blades are pitched to reduce the structural loads and maintain a constant
rotational speed, while the generator torque is kept constant or varied to
maintain a constant (rated) power output. While the controllers in below-
and above-rated conditions operate independently, they are typically com-
bined in near-rated conditions to limit the tip speed at rated power (Jonk-
man et al. 2009). For wind turbine analysis tools, control system models
should contain, as a minimum, routines for the blade-pitch and generator
torque control, in addition to start-up and shut-down procedures (Smilden
2019).
Despite their extensive use in analysis and design, the accuracy of aero-
hydro-servo-elastic simulation codes is still somewhat uncertain. As iden-
tification of limitations and sources of error in the theoretical models is
crucial to reduce safety factors and achieve safe and efficient design solu-
tions, verification and validation of numerical design tools are important.
Extensive code-to-code verification studies have been conducted through
the OC3, OC4, and OC5 projects (Jonkman and Musial 2010, Popko et al.
2012, Robertson et al. 2014; 2017, Popko et al. 2018; 2019), which have
contributed to an understanding of the applicability of different modelling
approaches, helped resolve errors in the codes, and identified future im-
provement needs (Robertson et al. 2014). The recent OC5 project has also
considered validation against experiments (Robertson et al. 2017) and full-
scale data (Popko et al. 2019).
Limited full-scale measurements from FWTs exist, with very few results
being publicly available, and validation studies must therefore to a large
degree rely on model-scale experiments. Model tests which consider com-
bined wind-wave response are challenging due to scaling issues. Recently,
hybrid testing, where parts of the load picture are calculated numerically
and applied on the experimental model using actuators, has shown prom-
ise. A real-time hybrid testing method for FWTs has been developed by
SINTEF Ocean (Sauder et al. 2016, Bachynski et al. 2016, Berthelsen et al.
2016), where wave basin tests are conducted with aerodynamic loads from
simultaneous numerical simulations. A complimentary approach using wind
tunnel tests with numerically computed hydrodynamic loads has also been
 2.1. Global response analyses 17

Control
system

Drive-
Wind Rotor Grid
train

Nacelle

Rotor-nacelle assembly

Tower

Waves Platform Support structure

Mooring
system

Figure 2.2: Aspects of global aero-hydro-servo-elastic analyses for floating wind

turbines. Based on a figure presented by Vorpahl et al. (2013).

developed at the Politecnico di Milano (Bayati et al. 2017; 2018).

2.1.3 Linear models
Although FWTs are nonlinear systems, linearized formulations are desir-
able for several reasons. In addition to the computational efficiency, meth-
ods based on linear models are extensively used for analysis of eigenvalues,
stability, and control systems, and may provide increased insight in the
properties of the underlying system. Because linearity often is an accur-
ate assumption for small displacements, even for nonlinear structures, such
models are regarded as suitable for fatigue. Assuming harmonic response,
the linear model can be solved very efficiently in the frequency domain.
This is also preferable for gradient-based optimization, since the stochastic
nature of time-domain results, which depend on the random realization of
the environment, complicates the computation of accurate gradients (Zahle
18 Literature Survey

et al. 2015).
Frequency-domain models have a long history for offshore structures, which
is reflected in the early efforts on frequency-domain modelling of FWTs
(Wayman et al. 2006, Tracy 2007, Sclavounos et al. 2008, Philippe et al.
2011), where response amplitude operators (RAOs) were established based
on diffraction and radiation loads from an external potential flow solver. As
hydrodynamic loads from numerical panel codes are computed in the fre-
quency domain, they can be readily applied in such global response models
without modification. A drawback with this approach in a design process
is that it requires an additional hydrodynamic analysis each time changes
are made to the design, which must be run prior to the response analysis,
and prevents quick examination of different design solutions. More compu-
tationally efficient options include Morison’s equation and MacCamy–Fuchs
theory; however, the validity of these methods is limited to simple geomet-
ries.
While early studies on frequency-domain methods for FWTs considered
only rigid-body modes, Bachynski and Moan (2012) compared a frequency-
domain model for a tension-leg platform (TLP) wind turbine with nonlinear
time domain simulations, and found that the linear model was insufficient
for design calculations, partly due to the rigid modelling of the tower. The
effect of tower flexibility on the global response was confirmed by full-scale
measurements from the Hywind Demo FWT, where the tower bending mo-
ment spectra clearly showed peaks around the first bending mode (Skaare
et al. 2015). Kvittem and Moan (2014) used a frequency-domain model,
which included the first tower bending mode, to evaluate the fatigue loads
for a semi-submersible FWT in stochastic wind-wave conditions, where the
shape of the bending mode was found from decay tests in calm air with a
coupled time-domain model. A stochastic linearization procedure, consid-
ering the two first tower modes, was presented by Kluger et al. (2016), who
computed the modes from the eigenvalue problem with an isolated tower.
Although more recent efforts on frequency-domain modelling have included
tower modes, platform flexibility has been little explored. As turbines grow
larger, support structure flexibility is expected to become more important,
and may need to also include the platform to properly capture the correct
global behaviour.
Due to the relatively complex aerodynamic calculations, and because the
design of the rotor usually is separated from the support structure design,
frequency-domain methods have typically focused on decoupling of the rotor
dynamics. Early work on FWTs considered only stochastic response due to
 2.1. Global response analyses 19

waves, while the aerodynamic excitation forces typically were treated as a

constant thrust force on the tower top. However, to properly capture the
combined wind-wave response, turbulent wind excitation loads are needed,
which can be estimated in several ways (Muskulus 2015). A method based
on turbulent wind simulations in the time domain, with a fixed rotor and an
active controller, was presented by Bachynski (2014), who compared results
from linear and nonlinear analyses on a TLP wind turbine. The approach
gave reasonably accurate estimates for the low-frequency wind response in
certain environmental conditions, but the agreement was highly dependent
on the damping in the model.
Calculation of frequency-dependent aerodynamic loads based on time series
has some drawbacks. Firstly, the turbulent wind loads are stochastic, mean-
ing that a large number of time-domain simulations may be needed. Secondly,
the low-frequency response is sensitive to how the thrust force spectrum is
estimated from the time series (Kvittem and Moan 2014). In addition, using
pre-calculated aerodynamic loads means that the controller cannot be mod-
ified without re-running the time domain simulations. Another approach,
which allowed for simultaneous design of the controller, was taken by Sand-
ner et al. (2014), who computed linearized aerodynamic loads on the tower
top as a function of wind speed, rotor speed, and blade pitch angle. The
formulations were based on a Taylor expansion, using pre-computed look-
up tables for the thrust and power coefficients, and a rotor effective wind
speed. More sophisticated linear models, involving blade elasticity, unsteady
aerodynamics, and hundreds of degrees-of-freedom (DOFs), have previously
been developed for wind turbine analysis (Merz 2015, Tibaldi et al. 2016),
and more recently, full-system linearization of FWT models based on the
OpenFAST software has been presented by Jonkman et al. (2018). The
implementation has later been verified against nonlinear simulations from
OpenFAST by Johnson et al. (2019).
For support structure models with decoupled rotor dynamics, the change
in aerodynamic forces due to the motions of the turbine should also be
included. The most important component is the aerodynamic damping,
which typically has been represented using a linear dashpot at the tower
top, and a number of methods has been applied to calculate an appropri-
ate damping coefficient. Early studies on frequency-domain methods for
offshore wind turbines used a value of 4 % of critical based on engineer-
ing estimate (van der Tempel 2006). Bachynski (2014) derived a mean
wind speed-dependent value from the change in thrust for a small change
in wind speed, assuming that the rotor speed and blade pitch angle were
20 Literature Survey

kept unchanged. However, due to interaction with the controller and ro-
tor dynamics, the damping is frequency-dependent, and neglecting these
interaction effects leads to non-conservative damping estimates (Jonkman
2008). As damping is most important for the response around resonance,
individual values for each mode can be found from decay tests with steady
wind and an active controller (Jurado et al. 2018). Model tests have shown
that there also is a frequency-dependent inertia (or stiffness) effect from the
controller, which results in changes in the pitch natural frequency for the
system (Bachynski et al. 2016, Goupee et al. 2017). Souza and Bachynski
(2019) computed frequency-dependent inertia and damping coefficients for
the NREL 5MW turbine numerically, based on forced harmonic oscillations
of the nacelle, and were able to accurately predict pitch decay periods for
three different FWTs, using a simplified 2-DOF model with modified inertia
and damping matrices.
As linearized models assume small displacements, they are seen as unsuit-
able for extreme response. However, results from previous studies have
shown that both the tower base bending moment (Karimirad and Moan
2011) and global motions (Aggarwal et al. 2017) for spar FWTs can be quite
Gaussian in harsh environmental conditions, which suggests that a linear
model also may be used to assess the extreme response in early stages of
design, or reduce the computational burden by quickly identifying critical
load cases. The degree of nonlinearity depends on the considered response
parameter, and also varies with location, floater geometry, environmental
conditions, and design situations. Therefore, the results obtained with a
linear model cannot be generalized to any FWT structure, and such mod-
els must be continuously verified against more sophisticated tools during a
design process.
2.1.4 Structural reliability
In the design of FWTs, stochastic variables such as turbulence intensity,
wind-wave directionality, and fatigue strength are typically replaced by
design values, which are obtained from their characteristic values combined
with a given safety factor. In order to reduce conservatism and thus the
design costs caused by these safety factors, probabilistic analyses can be
performed to assess the safety level of the structure through the probabil-
ity of failure. Such analyses can also be used to evaluate the importance
of different model uncertainties on the resulting structural reliability, and
possibly calibrate appropriate safety factors, which has been the focus of
several studies for bottom-fixed wind turbines. Márquez-Domínguez and
Sørensen (2012) calibrated design fatigue factors (DFFs) for a 2.3 MW off-
 2.2. Design optimization 21

shore wind turbine, and investigated the effect of inspections on the required
DFF values. Horn et al. (2019) quantified the effect of environmental load
uncertainties on the fatigue reliability of a 10 MW monopile, and derived
corresponding differences in DFFs resulting from the increased model fidel-
ity. Velarde et al. (2020) designed a 10 MW monopile with different DFFs
by varying the wall thickness, and used reliability analyses to identify the
necessary safety factor to achieve the target probability of failure during a
service life of 25 years without inspections.
Structural reliability problems consider the probability of failure for the
system, which can be expressed as

PF = P [g(x) ≤ 0] = fX (x) dx, (2.2)
g(x)≤0

where X ∈ Rn is a random vector with joint probability density function

(PDF) fX (x), which represents the uncertainties in the system. g(x) is
commonly known as the limit state function, where g(x) ≤ 0 denotes the
failure domain.
The evaluation of this integral must typically be done numerically, com-
monly through Monte Carlo simulations or the first- or second-order reli-
ability methods (FORM, SORM) (Madsen et al. 1986). In the widely used
and computationally efficient FORM, X is transformed into a vector of in-
dependent standard normal variables (U), and the limit state function is
linearized about the most probable point, which is the point on g(u) = 0
closest to the origin. The reliability index, β, which is the distance from
the origin to the most probable point, can be used to calculate the failure
probability through the following relation:
PF = Φ(−β), (2.3)
where Φ is the standard normal cumulative distribution function (CDF).

2.2 Design optimization

2.2.1 Basic optimization theory
An optimization problem can be formally written as (Nocedal and Wright
2006):
minimize f (x)
with respect to x ∈ Rn
(2.4)
subject to ĉj (x) = 0, j = 1, 2, ..., m̂
ck (x) ≥ 0, k = 1, 2, ..., m
22 Literature Survey

where f is the objective function, x is a vector of design variables, ĉ is a

vector of equality constraints, and c is a vector of inequality constraints.
In general, two approaches can be used to solve the optimization problem,
namely gradient-free and gradient-based methods (Martins and Ning 2020).
Gradient-free methods are relatively easy to combine with existing analysis
tools, and do not require differentiability of the simulation outputs. How-
ever, as illustrated in Fig. 2.3, these methods scale poorly with the number
of design variables, and become infeasible for large-scale optimization prob-
lems. The scaling properties are particularly important when expensive
analyses are needed to evaluate the objective function or constraints.
Gradient-based methods require derivatives to determine the search dir-
ection, which can be acquired in different ways. Numerical gradients are
commonly found using either finite-differences or the complex-step method.
While finite difference formulas are easier to implement and do not require
access to the source code, the complex step approximation is more accurate,
as it is not susceptible to subtractive errors (Martins and Ning 2020). Both
methods, however, scale linearly with the number of design variables, and
thus become prohibitively costly as the problem size increases.
The most preferable way of obtaining gradients is to compute the derivat-
ives analytically. This method requires significant implementation effort, as
modification of the specific components of the source code is needed, but
offers large improvements in efficiency. Analytic gradients can be computed
using either the direct method or the adjoint method, where the preferred
approach depends on the problem. The adjoint method, whose cost is inde-
pendent of the number of design variables, tends to be the most commonly
used approach in practice (Gray et al. 2019). This method is further de-
scribed in Section 3.8.
A function can be either unimodal, meaning that it only has a single min-
imum, or multimodal, which implies that several minima exist. Unless an
optimization problem is known to be convex, it is usually difficult to prove
that a global minimum has been found (Nocedal and Wright 2006). How-
ever, to prove multimodality, it is sufficient to find two separate local min-
ima. Therefore, Martins (2020) argues that ‘an optimum should be assumed
to be the global one until proven otherwise’. This requires that an effort to
search for additional minima has been made by performing the optimization
with a number of different initial designs.
For certain problems, it may be desirable to minimize several performance
 2.2. Design optimization 23

Figure 2.3: Comparison of required function evaluations until convergence to the

optimum using a gradient-free algorithm, finite differences, and analytic derivatives
for a multidimensional generalization of the Rosenbrock function (Ning and Petch
2016).

measures. This is known as a multi-objective problem, and may be useful

to quantify trade-offs between different parameters, or to produce a number
of possible designs when a final decision must be made during a later design
stage (Martins and Ning 2020). For these problems, there will no longer be
a single optimal solution, but an infinite number of solutions which forms a
surface in the objective function space. Multi-objective optimization there-
fore uses the term Pareto optimality, where a design is considered Pareto
optimal if it is impossible to improve the performance in one objective,
without worsening the performance in another (Boyd and Vandenberghe
2004).
2.2.2 Offshore wind substructures
Most of the work on numerical design optimization of offshore wind sub-
structures has considered bottom-fixed turbines. The focus has mainly been
on jacket structures, because of the complex geometry and large number
of potential design variables, which complicates a manual design process,
and increases the potential for cost savings by automation. Hydrodynamic
shape optimization of floating structures has a long history in the offshore
community, using RAOs to identify designs with favourable response to
wave loads. An early study on numerical hydrodynamic shape optimiz-
ation was performed by Clauss and Birk (1996), where various offshore
structures were designed to minimize forces or motions, based on a panel
24 Literature Survey

code and a gradient-free optimizer, and in recent years, similar techniques

have also been applied on FWT structures (Hall et al. 2013, Steinert et al.
2016, Karimi et al. 2017). Most of the work has considered linear frequency-
domain models to limit the computational effort, with an exception of Myhr
and Nygaard (2012), who optimized a floating substructure for two extreme
load cases using nonlinear time-domain simulations. Due to the compu-
tational effort, such optimization studies have rarely included more than
two or three design load cases, and the resulting optimized designs may
therefore be highly dependent on the chosen environmental conditions. In
addition, the applied numerical models may also impact the results. Hall
et al. (2013) performed single- and multi-objective hull shape and moor-
ing line optimization for different platform concepts spanning all stability
classes, and found that the Pareto front was dominated by relatively com-
plex platforms with several legs and taut mooring lines. The work was later
extended by Karimi et al. (2017), using an updated dynamic FWT model
and a new optimization algorithm, which significantly affected the derived
optimal solutions.
One of the challenges related to design optimization of substructures for
floating wind is the parametrization of the geometry. Several platform con-
cepts, with significantly different properties, have been suggested, and there
is currently no consensus on the optimal platform type. An illustration of
three main classes of FWTs is shown in Fig. 2.4. The derivation of a gen-
eral formulation that accurately can describe all platform types, including
different stability classes, various mooring system configurations, and both
single- and multi-hull platforms, possibly connected by braces or pontoons,
is a highly involved and difficult task. Consequently, the main properties of
the system, including the general shape of the platform, are typically chosen
before the optimization problem is formulated, which limits the design space
and reduces the chance of obtaining truly novel design solutions. Hall et al.
(2014a) tried to overcome this issue by introducing a basis function ap-
proach, where the platform design was expressed as a combination of basic
geometries with pre-calculated hydrodynamic coefficients. The design space
had numerous local minima, and the resulting optimized design was thus
highly dependent on the initial starting point. Also, some issues arose when
physical geometries was created based on optimal combinations of basis
designs, as the resulting hydrodynamic behaviour tended to differ from the
superimposed case, due to nonlinearities and interaction effects. While such
generalizations can remove the potential bias towards certain concepts and
allow for a more direct comparison between different solutions, it remains
uncertain whether they are able to fully represent the relevant physics for
 2.2. Design optimization 25

Figure 2.4: Semi-submersible, catenary moored spar, and tension-leg platform

wind turbine concepts (Bachynski 2014).

all platform types, or if more concept-specific formulations are needed.

Several studies have considered shape optimization of FWT platforms, but
little exists in the literature concerning structural optimization. The hydro-
dynamic loads on the hull are typically found from a separate hydrodynamic
analysis and given as resultant point loads, and complicates a structural
optimization procedure based on global response analyses, which requires
distributed hydrodynamic loads combined with a formulation for the scant-
ling design criteria. While comprehensive finite element models may be
necessary for the detailed structural design of the hull, design criteria based
on global forces and moments, such as the formulas provided by DNV GL
(2019b), may be used for structural sizing in early phases of design.
2.2.3 Integrated design
Common to previous studies on FWT design optimization is that they con-
sider the design of the platform, mooring system, or both, whereas the
tower, rotor, and control system are considered fixed during the optim-
ization. A possible reason is that design of wind turbines and offshore
structures, which traditionally have been two completely separated fields, is
performed by designers with specific engineering backgrounds. In addition,
different companies are responsible for the design of different components.
This may lead to partly integrated procedures where loads are exchanged
between the support structure designer and the wind turbine manufacturer
at a pre-defined interface (Vorpahl et al. 2013). A fully integrated design
approach, on the other hand, may be practically impossible in reality, due
26 Literature Survey

to confidentiality issues between the companies. However, because FWTs

are multidisciplinary systems, with important interactions between the dif-
ferent subsystems and disciplines, optimizing components separately will in
general lead to suboptimal, and in many cases infeasible, solutions on the
overall system level (Muskulus and Schafhirt 2014). To find the true op-
timum, an integrated optimization approach should be applied, where all
relevant components are considered simultaneously.
Despite gradient-based optimization being desirable due to its efficiency, a
large majority of the work on FWT design optimization has been performed
using gradient-free methods. Typical analysis tools for FWTs have not been
developed for optimization purposes, which require differentiability, and in
the case of analytic gradients, substantial modifications to the source code.
One exception is Fylling and Berthelsen (2011), who optimized the hull
shape, mooring lines and power cable for a spar FWT using a gradient-
based approach with numerical gradients obtained from finite differences.
While these methods may be feasible for optimization of smaller parts of
the FWT system with a limited amount of design variables and load cases,
integrated design optimization considering lifetime loads would result in a
large number of input variables and expensive function evaluations, which
requires analytic derivatives.
As there are important interactions between the control dynamics and the
support structure response, the control system should preferably be integ-
rated in the design optimization procedure. Numerical optimization of a
land-based wind turbine controller has earlier been performed by Tibaldi
et al. (2014), but has rarely been considered simultaneously to support
structure optimization, which typically decouple the rotor dynamics. Sand-
ner et al. (2014) manually derived optimal blade-pitch controller gains for
three different spar FWT designs, where the goal was to minimize a weighted
combination of rotor speed variation and tower bending. The gains that
minimized the cost function was shown to be highly dependent on the plat-
form design. A more automated approach was presented by Lemmer et al.
(2017), who optimized four parameters for a FWT substructure to min-
imize a combination of material cost and damage-equivalent loads, where
the blade-pitch controller was tuned in each design iteration using a linear
quadratic regulator (LQR). However, the effect of including control system
tuning in the substructure optimization loop was not quantified.
Perhaps the most important interaction between the controller and support
structure response is the potential introduction of negative damping for the
platform motions above rated wind speed (Larsen and Hanson 2007). Sev-
 2.2. Design optimization 27

eral methods have been suggested to resolve the issue, such as detuning the
controller gains (Larsen and Hanson 2007, Jonkman 2008), or to introduce
a feedback term proportional to the pitch velocity (Lackner 2009) or nacelle
velocity (van der Veen et al. 2012, Fleming et al. 2014) to manipulate the
rotor speed reference. For the latter approach, Fleming et al. (2019) sug-
gested that such controllers should not include the wave-frequency range
within their bandwidth, as doing so yielded an increase in the loads around
these frequencies. All methods lead to trade-offs between structural loads,
rotor speed tracking, and blade-pitch actuator use, and an integrated control
and structural design optimization approach is therefore needed to properly
evaluate and compare different solutions.
The shapes and natural frequencies of the tower modes, and thus the tower
response, are dependent on the platform design. In addition, as turbines
grow larger, the flexibility of the platform may become important, which
also can affect the response of the tower. In Fig. 2.5, the first bending mode
for a 10 MW spar is shown for both a rigid and flexible platform. In addition
to the modal shape being clearly affected by the platform flexibility, a 17 %
decrease in the natural frequency is observed. A similar effect was seen for
the OO-Star 10 MW semi-submersible, where the natural frequency of the
first tower mode was reduced by 25 % when the flexibility of the floater
was included (Müller et al. 2018). As the first bending mode of FWT
towers typically is located close to the 3P range, an accurate estimation of
the natural frequency is important to capture the correct modal excitation
forces. Souza and Bachynski (2020) studied the effect of flexible pontoons
for a 5 MW TLP wind turbine, and found that the flexible platform resulted
in a 12.7 % reduction in the weighted 1-h fatigue damage at the tower base,
due to the reduced natural frequency of the tower mode, which resulted in
less 3P excitation.
Commonly, wind turbine towers are designed with linearly tapered diameter
and wall thickness (Jonkman et al. 2009, Bak et al. 2013, Müller et al. 2018),
and are thus fully described by the values at the base and top. However,
the load picture for a FWT tower alters significantly from a land-based
or bottom-fixed offshore tower, with inertial and gravitational loads arising
from large platform motions, as well as possible restrictions at the tower base
due to compliance with the platform. An integrated design process which
considers the actual loads on the FWT system, and consistent reliability
levels, may therefore result in cost reductions.
Some effort has been made to perform integrated design optimization of the
blades and tower for onshore (Ning and Petch 2016) and bottom-fixed off-
28 Literature Survey

Figure 2.5: First support structure bending mode for a 10 MW spar with flexible
and rigid platform, and associated natural periods.
 2.2. Design optimization 29

shore (Ashuri et al. 2014) wind turbines, which experience interaction effects
due to coupled eigenmodes, increased blade loads from tower shadow effects,
and the possibility of tower strike by the blade tips. It is expected that the
inclusion of blade design in an integrated optimization process will affect
the results also for FWTs. However, design optimization of wind turbine
blades is a separate body of literature, which requires high-fidelity aerody-
namic and structural formulations to properly capture the blade response,
and is therefore considered outside the scope of this thesis.
2.2.4 Formulation of objective and constraints
The success of a numerical optimization process depends heavily on the
formulation of objective function and constraints for the design problem.
Because the goodness of a design is measured purely through the value of
a scalar objective function, it is essential that the applied function reflects
a relevant performance measure. The constraints should ideally cover all
important limit states, in addition to any restrictions that may be necessary
to avoid non-physical or impractical solutions. However, the goal of an
optimization process is not necessarily to provide a ready-to-build design,
but may also be used to study the sensitivity of different input parameters,
or identify design trends that are non-intuitive for a human designer.
A common objective function in design optimization of offshore wind sub-
structures is the structural weight (Pasamontes et al. 2014, Chew et al.
2016, Oest et al. 2017), which is assumed to be closely related to the cost
of the design. The weight is a convenient cost parameter in an engineering
optimization problem, because it is easily calculated with small uncertainty.
However, the weight, or associated material costs, may not be a good meas-
ure of the actual cost optimality, and even in the case where only capital
costs are considered, manufacturing costs due to e.g. welding of joints should
be included (Muskulus and Schafhirt 2014). The manufacturing costs are
more uncertain, and are typically expressed as empirically fitted functions
of the structural dimensions. Cost models for manufacturing of general steel
structures are presented e.g. by Farkas and Jármai (2013); however, their
accuracy with respect to FWT structures have not been verified.
Operational costs are rarely considered in design optimization of offshore
wind turbines. For conventional offshore structures, the focus has been on
inspection planning, where a cost optimal schedule for crack inspection and
repair is sought. Madsen et al. (1989) presented an early formulation of an
optimization problem which minimized expected costs related to construc-
tion, inspection, repair, and failure, where a set of structural design para-
30 Literature Survey

meters, as well as inspection times and qualities, were included as design

variables, and a constraint was placed on the reliability index. A small ex-
ample was solved using gradient-based optimization for the outer problem,
combined with FORM for the inner reliability problem. Such approaches
may lead to more cost-effective solutions in a lifetime perspective, but in-
crease the complexity of the design problem considerably. In addition, little
information exists in the literature regarding costs related to inspection and
repair of offshore structural elements, which vary with the inspection tech-
nique, location of potential cracks, and extent of the repairs.
An important difference in design optimization of FWTs, compared to con-
ventional offshore structures, is that it is not the cost of the system which
should be minimized, but rather the cost of energy. The energy production
should therefore be taken into account in the optimization process. Stud-
ies, e.g. by Roddier et al. (2017), have suggested that the movement of the
platform has a small effect on the total energy production, which is mostly
related to the wind farm site. However, the energy quality, which is a meas-
ure of how much the energy output fluctuates, is also a relevant parameter.
High-quality power, which has low variability in the output, is desirable, and
its value is closely related to the rotor speed tracking performance of the
system. The power quality may be regarded as an SLS criterion, and thus
be constrained based on system specifications, or treated as a performance
measure that should be maximized. Because the monetary cost of reduced
power quality is difficult to quantify, multi-objective optimization may be
needed to take this parameter into account. Ideally, an optimization process
should also consider the power quality for the complete wind farm, which
may result in different solutions compared to a single turbine.
A commonly constrained response parameter in FWT design optimization
is the platform tilt angle, which should be limited due to stability require-
ments and SLS criteria for the rotor and nacelle assembly (RNA). Such
criteria are project-specific, and there is currently no consensus on how the
tilt angle should be constrained; both the static and dynamic angle have
been considered in the literature, and different values have been applied.
The dynamic tilt angle constraint has commonly been expressed using the
standard deviation, which is easily obtained from the response spectrum
for a frequency-domain analysis. Typical values for the total angle have
been 10◦ (Hall et al. 2013, Karimi et al. 2017, Leimeister et al. 2020) or
15◦ (Steinert et al. 2016), with possibly different values in operational and
survival conditions (Ramachandran et al. 2017).
The constraint on the maximum static tilt angle is mostly related to the
 2.2. Design optimization 31

Figure 2.6: Side view of the wind speed for a spar FWT without tilt (left) and
with a tilt angle of 10◦ (right). The black dots represent the vertical wake center
position (Wise and Bachynski 2020).

reduction in projected rotor area and consequently the resulting power pro-
duction. The use of a static tilt angle has computational benefits, because
it can be computed without the need of simulations, using the platform
restoring stiffness and the specified rated thrust for the turbine. Common
constraint values in the literature are 5◦ (Matha et al. 2015, Steinert et al.
2016) or 10◦ (Wayman et al. 2006, Tracy 2007, Hall et al. 2013, Karimi et al.
2017), which, given that the produced energy is proportional to the cosine of
the tilt angle to the power three (Matha et al. 2015), result in energy losses
of about 1 % and 4.5 %, respectively. Increased stability is likely to trans-
late into larger design costs, and the static tilt angle varies with the mean
wind speed. Additionally, Wise and Bachynski (2020) showed that larger
tilt reduced the wake loss for a downstream turbine due to vertical wake
deflection, as illustrated in Fig. 2.6. Concept-specific assessments of the
trade-off with energy loss, also considering farm level effects, are therefore
needed to find cost-optimal values for this constraint.
An important consideration in design is to maintain the structural integrity
of the power cable, which is designed for relatively small curvatures. The
maximum offset of the platform is often constrained in lieu of an explicit
dynamic model and design criterion for the power cable. As the relation
between offset and power cable loads is somewhat dependent on the design
of the specific system, no general offset criterion exists in design guidelines,
but limitations are typically expressed as a fraction of the water depth. The
maximum offset is also of interest in relation to the tendon connectors for
TLP wind turbines, where a maximum offset limitation of 10 % of the water
depth has been common industry practice for conventional TLPs (Bachynski
2014).
The offset is to a large degree a function of the mooring system design, which
in addition must possess sufficient breaking strength to withstand the design
32 Literature Survey

tension in the ULS and ALS (DNV GL 2018a). A minimum tension should
also be maintained at all times, as slack line events may result in high snap
loads when the line regains its tension. For drag anchors, which are not
designed to take vertical loads, a constraint may also be needed to prevent
uplift at the anchors for extreme offsets (Brommundt et al. 2012).
Mooring lines should be designed against fatigue failure; however, as the
design life of a mooring line often is shorter than that of the platform and
wind turbine, replacement of mooring lines during the design life of the
platform is an implicit design assumption (DNV GL 2018a). Consequently,
fatigue is not considered a key criterion in mooring system optimization
(Benassai et al. 2014). The effect of including a fatigue life constraint for
the mooring lines was demonstrated by Fylling and Berthelsen (2011), who
found that it resulted in a 20 % more expensive mooring system and about
5 % increase in spar buoy costs for the optimized system.
A challenge related to constraint formulations for FWTs is to derive expres-
sions based on actual design criteria for fatigue and ultimate loads. Often,
constraints are expressed in terms of response parameters which can indic-
ate the level of response, but are only applicable for relative comparison
of solutions. However, to achieve cost-effective structures with a consistent
reliability level, the constraints should reflect the actual limit state criteria.
This requires a large number of simulations, and often, no closed-form solu-
tion is available for the design loads. Steinert et al. (2016) attempted to
properly express ULS constraints, where the 50-year extreme response in
a single parked condition was considered. The constraints were based on
the assumption of Gaussian response, which simplifies the estimation of the
extreme value distribution considerably. Fatigue damage in the tower and
possibly the platform is of concern, but has rarely been assessed directly in
design optimization studies, where proxy variables such as tower top dis-
placement due to bending (Sandner et al. 2014, Lemmer et al. 2017) or
nacelle acceleration (Hall et al. 2013, Karimi et al. 2017) have been used
to assess the support structure loads. Although the large number of re-
quired load cases makes optimization for fatigue challenging, it is typically
the design-driving limit state for large parts of the support structure, and
accurate and consistent fatigue design will thus lead to more effective use
of material.
Another reason for constraining the nacelle acceleration has been the as-
sumption of strong correlation with loads on the drivetrain; however, Nejad
et al. (2019) showed that the accelerations at the tower top were not a good
indicator for either maximum or fatigue loads on the main bearings of the
 2.2. Design optimization 33

gearbox. As the drivetrain is associated with relatively frequent failures

(Nejad et al. 2014), and has been shown to be sensitive to control strategies
(Lee et al. 2020) and platform concept (Nejad et al. 2015), it is desirable to
have a measure of the drivetrain response in the design optimization pro-
cess. Global analyses are not appropriate for detailed drivetrain analysis,
which typically is performed using de-coupled multibody simulations, but
simplified formulations based on global response parameters can be used for
qualitative analysis of the gear, shaft, and pitch actuator response with dif-
ferent control system designs (Smilden et al. 2018). Although these methods
may be inappropriate to assess the actual fatigue damage in the components,
they are applicable for relative comparison between different solutions.
In addition to general requirements, concept-specific constraints may also
be needed. For instance, spar platforms are susceptible to Mathieu instabil-
ity (Haslum and Faltinsen 1999), and TLPs may show ringing responses
(Bachynski and Moan 2014). The critical load cases also depend on the
system components that are considered. While the largest values for the
tower base bending moment, mooring line tension, and global motions of
FWTs may occur in operational conditions, the drivetrain and blades are
more affected by fault conditions (Bachynski et al. 2013). Insight about the
behaviour of the considered system, as well as the specific components that
are to be designed, is therefore needed to ensure that feasible solutions are
achieved.
34 Literature Survey
Chapter 3

Numerical Models

The results presented in Chapter 4 and the appended papers are based on
a linearized numerical model for global response analyses of floating wind
turbines, which was developed over the course of the thesis. The main
features of the model are summarized in the following. A brief description
of the evolution of the model, and the associated publications, is also given.

3.1 Linearized formulation

The coupled dynamics of FWTs are typically analysed using nonlinear time-
domain analyses, which capture the complex interactions between aerody-
namics, hydrodynamics, structural dynamics and control. For a design pro-
cedure which requires several iterations, each involving a large number of
load cases, the process becomes computationally prohibitive, and simplified
models are therefore desirable.
Using Newton’s second law, the nonlinear equations of motion for the dis-
placement vector q is expressed as

Mq̈ = f̃ (q̇, q, u), (3.1)

where M is the mass matrix, and f̃ contains position- and velocity-dependent

terms, in addition to forces due to external inputs (u). By defining a state
 
vector, x = q q̇ , Eq. (3.1) can be written as a set of first order
differential equations (Balchen et al. 2003):

ẋ = f (x, u). (3.2)

Considering small perturbations, Δ, about an operational point defined by

35
36 Numerical Models

(x0 , u0 ), the state and input vectors are written

x = x0 + Δx, u = u0 + Δu, (3.3)

while the system in Eq. (3.2) can be linearized by a Taylor series expansion:
∂f ∂f
ẋ ≈ f (x0 , u0 ) + Δx + Δu. (3.4)
∂x ∂u
If the operational point is chosen such that ẋ0 = f (x0 , u0 ) = 0, the linear-
ized dynamic system for the differential state and input variables can be
written in state-space form as (Chen 2013):

Δẋ = AΔx + EΔu, (3.5)

where A is the state matrix, and E is the input matrix. Given a set of system
outputs, y, which are described by y = g(x, u), a similar linearization can
be applied:
Δy = CΔx + DΔu. (3.6)
Here, C and D are commonly known as the output matrix and feedthrough
matrix, respectively. If the system is time-invariant, the matrices in Eq. (3.5)
and Eq. (3.6) are constant (Balchen et al. 2003).

3.2 System overview

An overview of the considered FWT system is shown in Fig. 3.1. The model
is valid for single-hull spar platforms with circular cross sections, support-
ing an arbitrary three-bladed horizontal-axis wind turbine, where the hull
is partially filled with solid ballast from the bottom to achieve stability.
A catenary mooring system consisting of three lines spread symmetrically
about the vertical axis is used for station-keeping, where the so-called crow-
feet are neglected, and the lines are assumed to have constant cross-sectional
properties and no discrete line components such as buoys or clump weights.
Only 2D response is detailed in the following; however, out-of-plane motions
have also been included in the present work using the same methodology.
The support structure is modelled as a slender flexible beam with varying
cross sectional properties. The model is regarded unsuitable for local blade
response, and rotor design is considered outside the scope of the current
work. The dynamic response of the blades is assumed to have a small ef-
fect on the support structure response, and are therefore considered rigid.
Consequently, the RNA is replaced by a point mass and inertia, with res-
ultant aerodynamic loads acting on the tower top. The considered resultant
 3.2. System overview 37

loads are the thrust force (FT ), the tilting moment (MT ), and the aerody-
namic torque (QA ). In addition, distributed wave excitation forces (dFW )
are applied on the hull.
For each wind-wave condition, the linearization point is found from equilib-
rium when the system is subjected to the mean environmental loads. As the
model neglects current and mean wave loads, the wind forces on the rotor
and tower are the only load components with non-zero mean. The opera-
tional point for a given system design is thus solely a function of the mean
wind speed, which determines the references for the rotor speed, collective
blade pitch angle, and generator torque, as well as the equilibrium position
of the support structure and mooring system.
The model presented here considers three support structure degrees-of-
freedom (DOFs), namely surge, pitch, and the first bending mode. For
deep-draught spar platforms, the first order wave excitation in heave is
small. As a consequence, the heave response is not considered to have a
large effect on the structural loads on the system, and is thus not included in
the model. However, the natural period can be estimated, and subsequently
used in the design process to avoid heave resonance in the wave-frequency
range. Another issue relevant for spar platforms is the phenomenon known
as Mathieu instability, which can occur when the pitch restoring moment
varies harmonically due to large heave motions (Koo et al. 2004). Haslum
and Faltinsen (1999) reported that unstable solutions occur for certain ratios
between the frequency of the heave motion and the pitch natural frequency,
and can thus be prevented assuming that the heave response is dominated
by resonant motions.
In addition to the support structure DOFs described above, a rotor speed
DOF is included. The inputs to the structural system consist of both control
system outputs and disturbances due to wind and wave loads. The control
outputs, i.e. the generator torque and the collective blade pitch angle, are
described by a separate control system model, which can be combined with
the structural model to create a single closed-loop system.
The model considers coupled response of four disciplines: structural dy-
namics, hydrodynamics, aerodynamics, and control dynamics. The applied
formulations for each of the disciplines are detailed in the following subsec-
tions.
38 Numerical Models

MT QA

FT

z y
Line 2
dFW Wind
x Line 1 x

Waves
Line 3

Figure 3.1: Overview of modelled FWT system.

3.3 Structural dynamics

The linear dynamic response of the support structure is found using gener-
alized coordinates. The total displacements are written as a weighted sum
of a set of pre-defined spatial shape functions, ψ, and the accuracy of the
formulation thus depends on how well the actual displacement field is cap-
tured by the shape functions. If the eigenmodes of the system are chosen
as shape functions, the method corresponds to modal superposition. An il-
lustration of the applied shape functions for the present support structure
modes is shown in Fig. 3.2.
The total displacement of the support structure at time t is found by mul-
tiplying the shape functions by time-dependent weight functions, χ, i.e.,
(Naess and Moan 2013)


n
w(z, t) = ψk (z)χk (t) = ψ(z) χ(t). (3.7)
k=0

The weight functions are assumed harmonic, i.e. χ(t) = χ0 eiωt , and are
found by solving the generalized equations of motion. Using the principle
of virtual work, the generalized system matrices can be established as a
combination of continuous and discrete terms. The support structure is
modelled as a slender beam undergoing pure bending deformations, and
 3.3. Structural dynamics 39

z z z

ψ(z) ψ(z) ψ(z)

Surge Pitch 1st bending

Figure 3.2: Generalized DOFs for the support structure response.

is assumed to follow Euler–Bernoulli beam theory. Including only terms

relevant for the considered FWT support structure, the generalized mass,
damping, and stiffness matrices can be expressed as
 l 
Mχ = m(z)ψ(z)ψ  (z)dz + Mp ψ(zp )ψ  (zp )
−d p
 (3.8)
+ Ip ψ ,z (zp )ψ 
,z (zp ),
p

 l 
Bχ = b(z)ψ(z)ψ  (z)dz + Bp ψ(zp )ψ  (zp ), (3.9)
−d p

and
  l
Kχ = Kp ψ(zp )ψ  (zp ) + EI(z)ψ ,zz (z)ψ 
,zz (z)dz
p −d
 l (3.10)
+ N (z)ψ ,z (z)ψ 
,z (z)dz,
−d

respectively, where m and b are distributed masses and dampers, Mp , Ip ,

Bp , and Kp represent discrete masses, inertias, dampers, and springs, EI is
the bending stiffness, N is the axial force, d is the platform draft, and l is
the height of the tower top above the still water line (SWL). Here, the mass
matrix includes both structural and hydrodynamic mass. In addition to
40 Numerical Models

hydro- and aerodynamic damping, the damping matrix contains structural

damping terms, which are described using stiffness-proportional Rayleigh
damping.
The generalized force vector is written:
 l  
F(t) = dF (z, t)ψ(z)dz + Fp (t)ψ(zp ) + Qp (t)ψ ,z (zp ), (3.11)
−d p p

where dF are distributed forces, and Fp and Qp are discrete forces and mo-
ments, respectively. The linear system of equations for the weight functions
then becomes
Mχ χ̈(t) + Bχ χ̇(t) + Kχ χ(t) = F(t). (3.12)

To include the controller and power production in the design process, a

model for the rotor and drivetrain is needed. Here, a rigid drivetrain is
assumed, and a single-DOF model is thus applied for the angular rotation
of the shaft. The differential equation is written:

ID ϕ̈ = QA − Ngear QG , (3.13)

where ID is the drivetrain inertia, Ngear is the gear ratio, QG is the high-
speed shaft (HSS) generator torque, and ϕ̇ is the low-speed shaft (LSS)
rotational speed.
With the structural DOFs described above, the state vector is defined as
 
x = χ χ̇ ϕ̇ . (3.14)

The state matrix in Eq. (3.5) can now be derived from the total mass,
damping, and stiffness matrices for the system (Chen 2013):
 
0 I
A= , (3.15)
−M−1 K −M−1 B

while the input matrix E relates the system inputs to the external forces
acting on the states.
The mean configuration of the mooring system, as well as the linearized
mooring stiffness, for a given mean thrust is found using the elastic catenary
equations. For the dynamic response of the mooring lines, the simplified
dynamic model proposed by Larsen and Sandvik (1990), and later extended
by Lie and Sødahl (1993), is utilized. Here, each line is modelled as a single-
DOF spring/damper system, that consider inertia and drag forces due to
 3.4. Hydrodynamics 41

TD

x
kE
cL
u

kG

Figure 3.3: Simplified dynamic mooring line model as illustrated by Lie and
Sødahl (1993).

the line motion in addition to elastic and geometric stiffness. Given the
displacement of the upper end of the line in the tangential direction, x(t),
the dynamic tension in the line can then be calculated from

TD = kE [x(t) − u(t)] = c∗ |u̇(t)|u̇(t) + kG u(t) + ω 2 m∗ x(t) (3.16)

where u(t) is the generalized displacement of the line, found from a quasi-
static analysis where the upper end is displaced two times the standard
deviation of the platform motion from the equilibrium position (signific-
ant motion). The geometric stiffness, kG , is found as the secant stiffness
when the line is moved from equilibrium to the position of significant mo-
tion. m∗ and c∗ are the generalized inertia and damping coefficients due to
Morison-type hydrodynamic forces, respectively, which are found using the
quasi-static line configuration as the shape function. kE is the elastic line
stiffness. The model is illustrated in Fig. 3.3.

3.4 Hydrodynamics
Assuming a circular cross section for the hull, the first order wave excit-
ation forces can be calculated using the analytic expression developed by
MacCamy and Fuchs (1954). The force per unit length, dFW , for a regular
wave with unit amplitude is then given by

4ρg cosh k(z + h)

dFW (z, ω) = G ei(ωt−α) , (3.17)
k cosh kh
42 Numerical Models

where ρ is the water density, g is the gravitational acceleration, k is the

wave number, h is the water depth, and

1 J1 (ka)
G=  , tan α = . (3.18)
(J1 (ka))2 + (Y1 (ka))2 Y1 (ka)

Here, Jn and Yn are the derivatives of the Bessel function of the first and
second kind, respectively, of order n, and a is the hull radius. The assump-
tion of a circular cross section also allows for simplification of the radiation
forces. The transverse added mass per unit length can be approximated
using the analytical 2D coefficient for a circular cylinder in infinite fluid
(Newman 1977):
a11 (z) = ρπD2 (z)/4, (3.19)
where D is the hull diameter. For a relatively slender structure, radiation
damping will be small and can be neglected, especially at the relevant nat-
ural frequencies.
The calculation of the natural period in heave requires an estimate of the
added mass in the vertical direction. Here, the vertical added mass is ap-
proximated as the value for a 3D circular disc (DNV GL 2017), with the
same diameter as the bottom of the platform (Db ), which was found to
yield good agreement with numerical linear potential flow solutions for sev-
eral considered spar designs:

A33 = ρDb3 /3. (3.20)

The viscous drag forces on the hull are nonlinear, and must therefore be
linearized. While viscous wave excitation is small compared to the wind
excitation forces and thus not included in the model, viscous damping is
important for the low-frequency surge motions. Neglecting the wave particle
velocities, the viscous damping due to the velocity of the structure (ẋ) can
be added based on stochastic linearization of the quadratic drag term in
Morison’s equation (Borgman 1969):

8
Bvisc |ẋ|ẋ ≈ Bvisc σẋ ẋ = Bvisc,lin ẋ, (3.21)
π
where Bvisc is the dimensional quadratic damping coefficient, and Bvisc,lin is
the stochastically linearized coefficient. The standard deviation of the velo-
city, σẋ , depends on the damping forces and vice versa, and must therefore
be found using an iteration scheme.
 3.5. Aerodynamics 43

Second order potential flow forces may be important if eigenfrequencies in

the system are located close to the excitation frequencies of the sum- or
difference-frequency loads. However, as studies have shown that these loads
have limited effect on the response for spar-type FWTs (Roald et al. 2013,
Duarte et al. 2014), they are not considered in the model.

3.5 Aerodynamics
Aerodynamic loads in global response analyses of FWTs are typically com-
puted using BEM theory, where induction factors, and subsequently blade
element forces, are found iteratively (Burton et al. 2011). Assuming that
the rotor design does not change during the course of the design process,
induction factors can be pre-calculated and stored in look-up tables, before
they are used in the response analyses. For each blade element, the induc-
tion factors are functions of the blade pitch angle (θ) and tip-speed ratio
(λ), where the latter is defined as

ϕ̇R
λ= . (3.22)
v

Here, R is the rotor radius, and v is the wind speed. In the present model,
the induction factors are computed from the BEM equations following the
procedure presented by Ning (2013), which includes Prandtl hub and tip
loss factors, and a modified Glauert correction as described by Buhl (2005)
for high induction factors. A quasi-static approach is used, meaning that
dynamic wake and dynamic stall effects are neglected. The calculated in-
duction factors at midspan for the DTU 10 MW blade design are shown in
Fig. 3.4.
Following quasi-steady BEM theory, the aerodynamic forces on a blade ele-
ment are nonlinear functions of (relative) wind speed, rotor speed, and
blade pitch angle, which can be found once the induction factors are known.
Linearizing the equations, the normal and tangential blade element forces
relative to the rotor plane can be expressed as:

∂Fn ∂Fn ∂Fn

Fn (v, ϕ̇, θ) ≈ Fn0 + Δv + Δϕ̇ + Δθ, (3.23a)
∂v ∂ ϕ̇ ∂θ
∂Ft ∂Ft ∂Ft
Ft (v, ϕ̇, θ) ≈ Ft0 + Δv + Δϕ̇ + Δθ. (3.23b)
∂v ∂ ϕ̇ ∂θ

The blade root loads of interest, i.e. the flapwise shear force Fy , the flapwise
bending moment Mz , and the edgewise bending moment My , are found by
44 Numerical Models

(a) Axial induction factor. (b) Tangential induction factor.

Figure 3.4: Calculated induction factors at midspan for the DTU 10 MW blade.

Figure 3.5: Blade coordinate system.

integrating the loads over the length of the blade

 R
Fy = Fn dr, (3.24a)
0
 R
Mz = r Fn dr, (3.24b)
0
 R
My = r Ft dr. (3.24c)
0

Here, the subscripts point to the blade coordinate system, which is defined
as in Fig. 3.5. The resultant rotor loads can then be calculated as:
 3.6. Control system description 45


3
FT = Fy,i , (3.25a)
i=1
3
MT = Mz,i cos ϕi , (3.25b)
i=1
3
QA = My,i . (3.25c)
i=1

As only the resultant loads on the rotor are of interest in the global response
model, it is desirable to simplify the wind field and express it using a single
scalar variable. This is achieved by expressing a rotor effective wind speed
for each of the resultant rotor loads in Eq. (3.25). The rotor effective wind
speed is a spatially constant wind speed which yields identical resultant
loads as the full wind field, and can be established from the incoming wind
spectrum, the spatial coherence function, and the transfer functions between
wind speed and blade element loads from Eq. (3.23). A detailed derivation
of the rotor effective wind speed, based on the thesis by Halfpenny (1998),
is provided in Appendix B.
In addition to the rotor loads, the aerodynamic drag forces on the tower are
taken into account, using a Morison drag formulation. The tower loads are
expressed as a combination of a mean force, which is added directly, and
a frequency-dependent force, which is found using stochastic linearization.
Here, only the excitation forces are considered, meaning that the tower drag
forces arising from the movement of the turbine are neglected.

3.6 Control system description

The baseline control system determines the changes in collective blade pitch
and generator torque based on the rotor speed error, and follows the generic
approach for variable-speed variable-pitch wind turbines described by e.g.
Jonkman et al. (2009). It consists of two separate and independent regions,
namely below and above rated wind speed. Typically, wind turbine control-
lers also include transition strategies to ensure smooth switchovers between
regions, however, for simplicity, this is not included in the model.
In both regions, the rotor speed is low-pass filtered to avoid high-frequency
actuation, using a first-order filter:
ϕ̈ + ωc ϕ̇ = ωc ϕ̇, (3.26)
where ωc is the corner frequency, and ϕ̇ is the filtered rotor speed. Below
rated wind, the blade pitch angle is kept at fine pitch, while the generator
46 Numerical Models

torque is set proportional to the squared rotor speed, to track the optimal
tip-speed ratio and thus optimize the power output:
2
QG = Kg ϕ̇ , (3.27)
where Kg is the generator torque constant. The change in generator torque
for a rotor speed error Δϕ̇ then becomes
∂QG
ΔQG = Δϕ̇ = 2Kg ϕ̇0 Δϕ̇, (3.28)
∂ ϕ̇
where ϕ̇0 is the rotor speed reference. Alternatively, the generator torque
can be regulated using a proportional-integral (PI) controller:
 t
ΔQG = kp,Q Δϕ̇ + ki,Q Δϕ̇ dτ, (3.29)
0
where kp,Q and ki,Q i are the proportional and integral gains for the rotor
speed error feedback, respectively. Above rated wind speed, the blade pitch
angle is found using a gain-scheduled PI controller:
 t
Δθ = ηk kp,θ Δϕ̇ + ηk ki,θ Δϕ̇ dτ, (3.30)
0
where kp,θ and ki,θ are the proportional and integral gains, respectively,
and ηk is the gain-scheduling parameter. For the torque controller, two
common strategies exist for the above-rated regime: (1) constant power or
(2) constant torque. The variation in generator torque then becomes
⎧
⎪
⎨− N Pr ϕ̇2 Δϕ̇, constant power,
gear r
ΔQG = (3.31)
⎪
⎩ 0, constant torque,
where Pr and ϕ̇r are the rated power and rotor speed, respectively.
To avoid the potential problem of negative damping for the platform motions
above rated wind speed, a feedback term proportional to the platform pitch
velocity or nacelle velocity may also be included, to manipulate the rotor
speed reference. Here, a modified rotor speed reference, ϕ̇0 , as defined by
Lackner (2009) is used:
ϕ̇0 = ϕ̇0 (1 + kf ẋf ), (3.32)
where kf is the velocity feedback gain, and ẋf is the nacelle or pitch velocity.
An updated expression for the rotor speed error can then be established as

Δϕ̇ = ϕ̇ − ϕ̇0 = Δϕ̇ − ϕ̇0 kf ẋf . (3.33)
Here, ẋf may also be passed through a low-pass filter before it is fed back
to the blade-pitch controller.
 3.7. Response to stochastic input 47

3.7 Response to stochastic input

After establishing the linearized model, the transfer matrix between system
inputs (u) and outputs (y) is defined using the system matrices in Eq. (3.5)
and Eq. (3.6) (Chen 2013):

H(ω) = C (iωI − A)−1 E + D. (3.34)

The cross spectral density matrix of the response vector y can be calculated
from
Sy (ω) = H(ω)Su (ω)H(ω)H , (3.35)
where (·)H denotes the conjugate transpose (Naess and Moan 2013). The
response spectra of y are then found along the diagonal of Sy (ω).
The input vector is expressed as
 
u = vFT vMT vQA F
W , (3.36)

where vFT , vMT and vQA are rotor effective wind speeds for thrust, tilting
moment and aerodynamic torque, respectively. The wave load vector, FW ,
contains the generalized wave excitation force for each support structure
DOF. Su (ω) is the cross spectral density matrix for the load process, which
has the following structure:
 
S (ω) 0
Su (ω) = wind , (3.37)
0 Swave (ω)

as the wind and wave processes are assumed to be uncorrelated (DNV GL

2016). Both the wind speed and the wave elevation are assumed to be
stationary Gaussian processes, and the response will thus also be a station-
ary Gaussian process. The nth spectral moment of an arbitrary response
parameter y, given the variance spectrum Sy (ω), is defined as
 ∞
mn = ω n Sy (ω)dω. (3.38)
0

The standard deviation of y can then be found from

√
σy = m0 . (3.39)

3.7.1 Fatigue damage

To calculate the fatigue damage, formulations for the fatigue resistance and
fatigue loading are needed. The fatigue resistance is commonly described
48 Numerical Models

by SN curves, where the relation between the stress range (S) and number
of cycles to failure (N ) is given as (DNV GL 2019c):
N = KS −m , (3.40)
where K and m are material parameters dependent on the type of fatigue
detail. Using the Palmgren-Miner hypothesis of linear accumulation of dam-
age, the fatigue damage in time T can be expressed as (Naess and Moan
2013):  ∞
νT m
D= s fS (s)ds. (3.41)
0 K
Here, the fatigue loading is described by the cycle rate, ν, and the PDF of
the stress ranges, fS (s). If the response process is narrow-banded and Gaus-
sian, the stress ranges can be described by a Rayleigh distribution; however,
because the response of FWTs in general is wide-banded, the narrow-band
formulation may result in overly conservative damage estimates. For a gen-
eral wide-banded response, cycle counting methods based on the stress time
series are typically applied, with rainflow counting being considered the
most accurate for estimation of fatigue damage (Naess and Moan 2013).
This method has no closed-form solution in the frequency domain for wide-
banded Gaussian response, and given a stress response spectrum, the stress
range distribution must thus be estimated either by transformation of the
response spectra to the time domain, or directly in the frequency domain us-
ing empirical formulae. For the latter approach, various methods have been
proposed, where the formulae presented by Dirlik (1985) and Benasciutti
and Tovo (2005) have been shown to yield accurate approximations of the
rainflow counting method for a variety of spectral shapes and bandwidths
(Gao and Moan 2008).
The Dirlik method, which is implemented in the current model, uses the
stress response spectrum and empirical factors to fit the PDF of the stress
cycles to a combination of an exponential and two Rayleigh distributions,
and the accuracy of the method is therefore dependent on how well the
rainflow-count can be represented by these distributions. The resulting
stress range distribution is given as:
   2
  2

G1
Q exp − Q
Z
+ G2 Z
R2
exp − 2R
Z
2 + G3 Z exp − Z2
fS (s) = √ , (3.42)
2 m0
where G1 , G2 , G3 , R, and Q are empirical weight factors found from the
spectral moments, and Z is the normalized stress range:
S
Z= √ . (3.43)
2 m0
 3.8. Optimization framework 49

3.7.2 Extreme response

Estimation of extreme values for a stochastic process can be performed using
upcrossing analysis, which is based on the assumption that upcrossings of
high levels are statistically independent (Naess 1984). Consequently, the
number of upcrossings in time T will be Poisson distributed, which, for the
response process y, gives the following CDF for the extreme value Ξ:
 
FΞ (ξ) = exp −νξ+ T , (3.44)

where νξ+ is the mean upcrossing rate of the level ξ. Assuming a stationary
Gaussian process with zero mean, the CDF can be written (Naess and Moan
2013)   
ξ2
FΞ (ξ) = exp −ν0 T exp − 2
+
, (3.45)
2σy

where ν0+ is the mean zero-upcrossing rate, which can be found from the
zeroth and second order spectral moments:
1 m2
ν0+ = . (3.46)
2π m0
The p-fractile of the distribution, i.e. the value which is exceeded with prob-
ability (1 - p) during the time T , becomes (Naess and Moan 2013):
  

 ν0+ T

ξp = σ 2 ln , (3.47)
ln(1/p)

while the most probable maximum value, ξc , also known as the characteristic
largest extreme, can be found approximately from the relation FΞ (ξc ) = e−1 :
 
ξc = σ 2 ln ν0+ T . (3.48)

3.8 Optimization framework

To enable efficient gradient-based optimization, the FWT model is imple-
mented in the OpenMDAO framework (Gray et al. 2019). OpenMDAO is an
open-source framework for multidisciplinary design, analysis and optimiza-
tion, which contains a large number of solvers and other features making it
suitable for large-scale optimization of complex systems. The model is con-
structed in a modular fashion, where the complex multidisciplinary model
50 Numerical Models

Figure 3.6: OpenMDAO problem structure (Gray et al. 2019).

is broken down into smaller units of code (components). The components

are gathered in groups that again can be part of other groups, and thus
forms a hierarchical structure which can be used to organize the model or
define different solver strategies. An overview of the OpenMDAO problem
structure is shown in Fig. 3.6.
The partial derivatives of the component outputs with respect to the com-
ponent inputs can either be calculated numerically using e.g. finite differ-
ences, or be provided as analytic functions by the user. Analytic derivatives
are desirable due to their computational efficiency and accuracy; however,
they require significant implementation effort. In the present work, analytic
derivatives are computed for the complete coupled model, which ensures a
feasible optimization process even as the number of design variables and
load cases become large.
One of the main purposes of the optimization framework is to compute
overall model-level derivatives given the component-level derivatives. For a
model where all relations are given by explicit functions, the total derivatives
can be calculated using the chain rule. However, if implicit relationships
between variables exist, more elaborate methods are needed. Given an
output of interest, f , which is described by the function f = F(x, y(x)),
where x is the vector of input variables, and y is a vector of implicit variables
dependent on x, the total derivative can be found from (Martins and Ning
2020):
df ∂F ∂F dy
= + . (3.49)
dx ∂x ∂y dx
The relationship between y and x is defined by the solution of the residual
equations r = R(x, y(x)) = 0. The total derivative of r with respect to x
can be expressed as:
dr ∂R ∂R dy
= + = 0. (3.50)
dx ∂x ∂y dx
 3.8. Optimization framework 51

It is desirable to calculate df /dx using only partial derivatives, as they do

not require solving the residual equations and thus have a lower computa-
tional cost. The only term in Eq. (3.49) that is not a partial derivative is
dy/dx, which from Eq. (3.50) can be expressed as

∂R dy ∂R
=− . (3.51)
∂y dx ∂x

Given the partial derivatives of R with respect to x and y, dy/dx can then
be found by solving the linear system in Eq. (3.51). This is known as the
direct (forward) method, where the linear system must be solved once for
each model input (Martins and Ning 2020).
Alternatively, dy/dx in Eq. (3.49) can be replaced by the expression in
Eq. (3.51):
 
df ∂F ∂F ∂R −1 ∂R
= − . (3.52)
dx ∂x ∂y ∂y ∂x
  
ψ

Here, ψ is known as the adjoint vector, which can be described through the
following relation (Martins and Ning 2020):
   
∂R ∂F
ψ= . (3.53)
∂y ∂y

Now, ψ can be found by solving the linear system in Eq. (3.53), and used
together with Eq. (3.52) to calculate the total derivative. This is known
as the adjoint (reverse) method. The linear system in Eq. (3.53) must,
contrary to the forward method, be solved once for each model output,
and this method therefore offers increased computational efficiency when
the number of model inputs is larger than the number of model outputs.
OpenMDAO supports both direct and adjoint gradient computations, and
the method is chosen based on the specific problem structure.
To illustrate the extent of the analytic gradient implementation, the required
calculations are shown for a single model component. The component which
calculates the uncoupled heave natural period (Tn3 ), given the structural
mass of the system (M ), the added mass in heave (A33 ), and the heave
restoring stiffness (C33 ), uses the following relation:

M + A33
Tn3 = 2π . (3.54)
C33
52 Numerical Models

The required partial derivatives for the component become:

∂Tn3 π
=  , (3.55a)
∂M C33 MC+A 33
33

∂Tn3 π
=  , (3.55b)
∂A33 C33 MC+A 33
33

M +A33
∂Tn3 π C33
=− . (3.55c)
∂C33 C33
As both M , A33 , and C33 are dependent on the design variables for the
optimization problem, the chain rule in Eq. (3.49) is then needed to compute
the total derivative of Tn3 with respect to an arbitrary design variable p:
dTn3 ∂Tn3 ∂Tn3 dM ∂Tn3 dA33 ∂Tn3 dC33
= + + + . (3.56)
dp ∂p ∂M dp ∂A33 dp ∂C33 dp
If there is an implicit relationship between M , A33 , or C33 , and p, obtaining
the total derivative also requires solving the linear system in Eq. (3.51) or
Eq. (3.53).
The component-level derivatives must be provided for all parts of model, and
may result in rather involved computations, especially for more advanced
components. The complexity and required effort related to the gradient
implementation are thus comparable to the development of the analysis
model itself, and makes up about 70 % of the total code volume in the
present work.
The model layout and data dependencies are illustrated using an extended
design structure matrix (XDSM) (Lambe and Martins 2012) in Fig. 3.7.
Here, the rounded blue box is the optimizer, green boxes are analysis models,
gray parallelograms are data, white parallelograms are fixed input paramet-
ers, and grey lines represent data dependencies. The data moves from top
to bottom and left to right in the upper triangular part, and from bottom
to top and right to left in the lower triangular part.
 Mean wind speed,
Corner freq., SN data, Buckling resistance,
Mean wind speed Water depth Water depth RNA data sig. wave height, Drag coefficient
gain schedule DFFs pitch and offset limits
spectral peak period

Support structure Support structure

Optimizer Mooring variables Platform geometry Controller gains
variables and mooring variables

Linearized Effective wind speed

Aerodynamics Mean loads Mean loads
aerodynamic loads transfer function

Mooring stiffness Breaking strength,

Mooring
and mass anchor uplift offset

Wave excitation
Wave excitation
transfer function

Heave natural period, Mass matrix,

taper angles, Structural added mass matrix,
diameter match, model damping matrix,
Mathieu stability stiffness matrix

Control
Control matrices
model

Environmental
Load matrix
loads

Viscous Linearized viscous

damping damping matrix

Rotor speed std.dev., Dynamic

Structural velocity Stress spectra Response spectra
system stability analysis

Fatigue damage Fatigue

Upcrossing
Extreme response
analysis

Cost
System costs
models

Figure 3.7: XDSM diagram showing the model layout and data dependencies.
3.8. Optimization framework
53
54 Numerical Models

3.9 Model development

The linearized FWT model has been developed over the course of the thesis,
and the versions applied in the different papers have therefore not been the
same. The evolution of the model is briefly summarized in the following.

V1, applied in P1: The structural model for the support structure using
generalized coordinates is established, but assumes a rigid platform, while
the elasticity of the tower is considered. The hydrodynamic formulations
presented in Section 3.4 are implemented, and the quasi-static mooring sys-
tem response is found from the elastic catenary equations. The rotor DOF
is not included, and aerodynamic excitation (thrust and tilting moment) is
computed from time-domain simulations with turbulent wind and an active
controller. The aerodynamic damping is found as the change in thrust due
to a small change in wind speed, assuming no reaction from the rotor or
controller. The Dirlik method for fatigue damage in the support structure
and extreme response based on upcrossing analysis are implemented.

V2, applied in P3: Building off of V1, A single-DOF model for the rotation
of the shaft is implemented, and a flexible platform is included. Linearized
aerodynamic loads are derived as functions of wind speed, rotor speed, and
blade-pitch angle, using quasi-steady BEM theory and a rotor effective wind
speed. Generator-torque and blade-pitch controllers are added to the model.
Dynamic mooring line tension is found based on a single-DOF model which
consider drag and inertia forces due to the motion of the line. Analytic
gradients are calculated for the model, which is implemented in the Open-
MDAO framework.

V3, applied in P4: More advanced control strategies for the blade-pitch
controller are implemented. The new controllers consider velocity feedback
using either the nacelle or the platform pitch velocity, with an optional low-
pass filter on the velocity signal.

V4, applied in P5: The model is extended to account for out-of-plane

response, and thus allow for wind-wave misalignment.
Chapter 4

Research Findings

This chapter presents the main research findings from the papers included
in the thesis, and discusses their relation to the contributions stated in
Section 1.3.

4.1 Linearized dynamic analyses

Findings related to the use of linearized dynamic analyses to predict global
design loads for FWT structures are presented in the following. First,
closed-form expressions for frequency-dependent aerodynamic damping and
inertia are derived. Then, verification of the linear model against nonlin-
ear simulations for fatigue and extreme response is presented, and some
important aspects and limitations to the model are discussed.
4.1.1 Aerodynamic damping and inertia
The motions of the FWT give rise to changes in the aerodynamic forces
on the turbine, which depend on the mean wind speed, motion amplitude,
and oscillation frequency. Commonly, frequency-domain analyses of offshore
wind turbines have considered aerodynamic damping coefficients which are
either constant, or vary only with the mean wind speed. These methods
neglect the impact of the controller and rotor dynamics, and consequently
overestimate the damping at low oscillation frequencies.
In addition, the forces which arise from the platform motions that are out
of phase with the velocity, and thus result in an inertia (or stiffness) effect,
are neglected. For FWTs, which typically experience low-frequency reson-
ant motions, frequency-dependent forces should be considered in response

55
56 Research Findings

calculations. In the present work, closed-form expressions for frequency-

dependent aerodynamic damping and inertia were derived. The key results
are presented in the following, while a detailed derivation can be found in
P2.
Using linearized aerodynamic forces and a 2-DOF wind turbine model, with
the horizontal nacelle displacement (x) and the rotation of the shaft (ϕ),
the following expressions were obtained for the aerodynamic inertia (aaero )
and damping (baero ) coefficients:
 ! "  #
1 ∂FT ∂FT ∂FT ∂FT
aaero (ω) = − 2
Re iω − iω + kP C̃ϕ (ω) ,
+ kI
ω ∂v ∂ ϕ̇ ∂θ ∂θ
(4.1a)
 ! "  #
1 ∂FT ∂FT ∂FT ∂FT
baero (ω) = Im iω − iω + kP + kI C̃ϕ (ω) .
ω ∂v ∂ ϕ̇ ∂θ ∂θ
(4.1b)

Here, FT is the thrust force, v is the wind speed, kP and kI are the propor-
tional and integral gains of the blade-pitch controller, and θ is the blade-
pitch angle. C̃ϕ is the complex transfer function between x and ϕ, i.e.,

ϕ(ω) = C̃ϕ (ω)x(ω). (4.2)

The derived coefficients in Eq. (4.1) were compared to results from nu-
merical simulations using forced oscillations with AeroDyn as described by
Souza and Bachynski (2019), as well as constant damping coefficients, which
neglect the controller and rotor dynamics. In Fig. 4.1, the coefficients are
shown for the low-frequency range, for a mean wind speed of 25 m/s. The
overall agreement between the analytical and numerical solutions was good,
while the constant damping coefficient was found to significantly overes-
timate the aerodynamic damping at low frequencies. The magnitude of the
aerodynamic inertia is low compared to the total mass of typical FWTs, res-
ulting in negligible consequences for the surge response. Due to the height
of the nacelle, however, the contribution to the moment of inertia in pitch
is considerable.
To assess the importance of frequency-dependent coefficients, a frequency-
domain model of the OO-Star 10 MW semi-submersible, which was part of
the the LIFES50+ project (Müller et al. 2018), was created. Results from
the frequency-domain model, using different formulations for the aerody-
namic damping and inertia, are compared to nonlinear time-domain simu-
lations using SIMA. The response spectra for platform pitch and tower base
 4.1. Linearized dynamic analyses 57

(a) Inertia. (b) Damping.

Figure 4.1: Aerodynamic inertia and damping coefficients, 25 m/s mean wind
speed.

Table 4.1: Error in standard deviations compared to time domain simulations.

Response parameter Constant (%) Analytical (%) AeroDyn (%)

Platform pitch -29.6 +55.0 +30.6
Tower base moment -33.0 +19.2 +5.3

bending moment are shown in Fig. 4.2, and relative errors in the standard
deviations are presented in Table 4.1.
While the constant damping coefficient underestimated the responses, the
opposite was observed for the frequency-dependent coefficients. There were
also notable differences between the analytical and AeroDyn models, despite
the good agreement for the damping coefficient, which suggested that the
response was sensitive to resonant motions and consequently the damping
level in the model. Although the actual numbers also depend on the con-
sidered wind-wave conditions, platform concept, and additional damping in
the model, the results demonstrated that the use of constant damping coef-
ficients is non-conservative, and that this can be avoided in a simple manner
through the frequency-dependent coefficients in Eq. (4.1).
The deviations in response between the frequency-domain models in Fig. 4.2
can be attributed to differences in aerodynamic damping coefficients. The
effect of aerodynamic inertia was much less prominent, but can be seen in
the calculated platform pitch natural period. Calculated natural periods in
pitch, with and without aerodynamic inertia, are shown in Fig. 4.3.
58 Research Findings

(a) Platform pitch. (b) Tower base bending moment.

Figure 4.2: Response comparison between SIMA and frequency-domain analyses,

25 m/s mean wind speed.

Figure 4.3: Pitch natural period with and without aerodynamic inertia effect.
 4.1. Linearized dynamic analyses 59

An alternative method of achieving frequency-dependent damping and iner-

tia effects in the support structure analysis is to include a description of the
rotor and controller dynamics in the support structure model, as described
in P3. These effects will then be implicitly included.
4.1.2 Long-term fatigue damage
To assess the applicability of the linearized model in design, the degree of
nonlinearity in the global response of the support structures is evaluated.
Because both the wind speed and the wave elevation in the short term are
assumed stationary Gaussian processes, the linear response of the FWT is
also stationary and Gaussian. One way to evaluate the nonlinearity of the
response is therefore to look at the skewness and (Pearson’s) kurtosis values,
which are 0 and 3 for a Gaussian process, respectively. The skewness and
kurtosis for the tower bending moment, based on nonlinear time-domain
simulations in SIMA, are shown in Fig. 4.4. The results are based on one
wind-wave condition for each mean wind speed from 1 to 30 m/s, using a
bin size of 1 m/s and six random realizations. The most probable values
for Hs and Tp were applied for each mean wind speed, and represent typical
fatigue conditions.
For most wind speeds, the response is quite linear, especially above the
rated wind speed of 11.4 m/s. Here, the wave loads become increasingly
more important relative to the wind loads, which exhibit a larger degree of
nonlinearity. Close to the tower top, the response is dominated by aerody-
namic loads, and overall more non-Gaussian behaviour is observed. In the
rest of the tower, the pattern is more or less identical to the tower base, sug-
gesting that the same load effects are dominating along most of the tower
length.
A comparison of weighted 1-h fatigue damage at the tower base, based on
both linear and nonlinear analyses, is shown as a function of the mean
wind speed in Fig. 4.5, taken from P1. Here, 546 environmental conditions
were considered, which corresponded to all conditions within the operational
range with probability of occurrence greater than 10−4 at the considered site.
The linear model was, for most wind speeds, able to accurately capture the
fatigue response, and the damage estimates based on response spectra using
the Dirlik method were found to agree well with inverse Fourier transform
combined with rainflow counting. Large errors in the damage estimates were
observed at 9 m/s mean wind speed, where the first tower bending mode
coincided with the 3P frequency. In such resonance conditions, nonlinear
effects become important, and the linear model significantly overestimated
60 Research Findings

(a) Skewness. (b) Kurtosis.

Figure 4.4: Skewness and kurtosis values for tower fore-aft bending moment.
 4.1. Linearized dynamic analyses 61

Figure 4.5: 1-h weighted tower base fatigue damage arranged by wind speed.

the bending mode response.

These results suggest that the linearized model should be used with care if
the first bending mode is close to the blade-passing frequency, which is often
the case for FWTs. It also highlights the importance of accurate estimation
of the tower bending mode and associated natural period, as the modal
excitation forces may change rapidly with frequency. This is illustrated for
the tower base bending moment response spectrum in Fig. 4.6, based in the
linearized model, where the natural frequency of the first bending mode is
artificially increased in steps of 0.06 rad/s by modifying the modal stiff-
ness. As the natural frequency approaches the 3P frequency, the resonant
response, and consequently the fatigue damage, is significantly increased.
This effect may also be important in the context of the numerical modelling
of the platform in the global response analyses. For large turbines, the
flexibility of the platform can have a large impact on the tower modes, and
should therefore be included to ensure that the correct dynamic behaviour
is captured.
4.1.3 Short-term extreme response
Similarly to the skewness and kurtosis values for the fatigue conditions,
upcrossing rates were used to assess the linearity of the extreme response
in 50-year conditions. For a Gaussian process, the upcrossing rate of level
y can be calculated from the process’ standard deviation (σ) and mean
62 Research Findings

Figure 4.6: Linearized tower base bending moment response spectrum for U = 15
m/s, Hs = 3 m, and Tp = 12 s, where the natural frequency of the first bending
mode is artificially increased in steps of 0.06 rad/s (≈ 0.01 Hz) by modifying the
modal stiffness. The corresponding increase in calculated 1-h fatigue damage is
shown, together with the 3P frequency at rated speed (black line).

zero-upcrossing rate (ν + (0)):

 
y2
ν (y) = ν (0) exp − 2
+ +
(4.3)
2σ

For the nonlinear time-domain simulations, a function for the tail of the
mean upcrossing rates can be established empirically, as described in P1.
The nonlinear and linear average upcrossing rates (AUR) for surge, pitch,
and tower base bending moment are shown for two 50-year conditions in 4.7.
The plots show the multiplication factor κ, which is related to the extreme
response, y, through the relation

y = μ + κσ, (4.4)

where μ and σ are the mean and standard deviation of the process.
Very good agreement was obtained with an extreme wind speed in the
parked condition, where the response was governed by wave loads. Lar-
ger nonlinearities are observed near the rated wind speed, due to the large
thrust forces in this condition. This was especially the case for the pitch
response. The tower base bending moment, on the other hand, was also
quite Gaussian in the operational condition, due to the large influence of
inertial forces arising from wave-frequency motions.
 4.1. Linearized dynamic analyses 63

(a) Surge.

(b) Pitch.

(c) Tower base bending moment.

Figure 4.7: Upcrossing rates for 11 m/s (left) and 50 m/s (right) mean wind
speed.
64 Research Findings

While these results suggest that a linear model also can be used to evaluate
extreme response in early phases of design, especially for the bending mo-
ments in the support structure, such a conclusion cannot be drawn in the
general case. The accuracy of the linear model depends on the importance
of nonlinearities in the system, which will vary with location, floater geo-
metry, and different operational conditions. The agreement with nonlinear
analyses also depends on how the system is represented in the time-domain
model, and care should be taken to make sure that relevant nonlinearities
are included.

4.2 Integrated design optimization

The following subsections present key results and discussion regarding in-
tegrated numerical design optimization of FWT systems. The findings are
based on the methodology presented in P3, where the tower, spar platform,
blade-pitch controller, and mooring system were optimized simultaneously,
including the scantling design of the hull, using gradient-based optimiza-
tion with analytic gradients. The optimization considered buckling and
long-term fatigue damage along the hull and tower, maximum offset and
pitch angle of the platform during selected extreme conditions, and extreme
mooring line tension and anchor uplift.
4.2.1 Multimodality
During the optimization runs performed in the present work, local minima
were identified for the tower design. The fatigue damage in the support
structure increases significantly if the natural frequency of the first tower
bending mode coincides with the blade passing frequency, and the optimizer
will therefore try to move the tower mode away from the 3P range. The
direction that the mode is moved during the optimization depends on the
initial design, as illustrated in Fig. 4.8. Due to the ‘barrier’ created by the
blade passing frequency, the optimizer will not be able to move over to the
stiff-stiff range if the initial design has a soft-stiff tower, and vice versa.
Early studies with the optimization model suggested that no feasible solu-
tion existed in the soft-stiff range for the applied design problem, and all
subsequent studies were therefore performed with a stiff-stiff initial design.
However, this should not be treated as proof that a soft-stiff solution does
not exist, as the design analyses were based on a simplified model and a
reduced set of environmental conditions. In addition, more advanced con-
trol strategies, such as a speed exclusion zone to prevent tower resonance
in the below-rated regime, could improve the fatigue behaviour. More de-
tailed studies of potential designs are therefore needed to conclude whether
 4.2. Integrated design optimization 65

Barrier

Converges towards Converges towards

3P range
soft-stiff design stiff-stiff design

0.2 0.4 0.6 0.8

Frequency [Hz]

Figure 4.8: Impact of initial tower bending natural frequency on optimization

results.

a feasible soft-stiff tower design exists.

4.2.2 Cost and power quality
The goal of the design optimization was to minimize a combination of system
costs and rotor speed variation, where the latter was considered a measure
of the (inverse) power quality. The monetary cost of reduced power quality
is difficult to quantify, and the terms could therefore not be combined in
a straightforward manner. The optimization was therefore performed with
different weights on the two sub-objectives, to assess the trade-off effects for
the system.
The relative importance of costs and power quality in the objective function
primarily affected the blade pitch control system parameters, as shown in
Fig. 4.9, where the optimized PI gains for different weight factors are plotted
against the resulting normalized rotor speed variation. Increasing the gains
results in a faster controller, leading to lower variation in the rotor speed,
but increased loads in the support structure.
Within the range of the objective function weights considered in P3, the
power quality was found to be much more sensitive to the applied weighting
than the system costs, as shown in Fig. 4.10. While a 15 % decrease in
rotor speed variation was observed from lowest to highest power quality
weighting, the corresponding increase in design costs was only 0.7 %.
66 Research Findings

Figure 4.9: Optimized blade-pitch controller gains plotted against resulting rotor
speed variation.

As the applied design optimization process considered only stationary con-

ditions, relatively slow controllers were obtained. To ensure proper control
action during transient events such as wind gusts, extensions to the presen-
ted methodology are needed, which may result in faster controllers with
larger negative effects on the support structure costs.
4.2.3 Support structure design
The optimized tower and platform design for a selected combination of ob-
jective function weights, based on the integrated approach applied in P3,
is shown in Fig. 4.11, together with fatigue and buckling utilization factors.
The platform took an hour-glass shape below the wave-zone, which increased
the restoring stiffness in pitch and thus improved the low-frequency beha-
viour. The sharp edges and constant taper angles along the platform were
caused by a constraint on the maximum taper angle, which was included
to ensure a platform shape that was captured correctly by the numerical
model. The fact that this constraint was active for most of the platform
length, suggests that reductions in costs could be achieved if the constraint
is relaxed. More comprehensive numerical models, and possibly more in-
formation about the sensitivity of the platform shape on the manufacturing
costs, would be needed in order to relax this constraint.
The tower base diameter was forced to match the top diameter of the plat-
form, which gave rise to conflicting influences. The platform diameter should
be low to limit the wave loads, whereas the tower diameter should be large,
as this is the most cost-effective way to achieve the required fatigue res-
istance. The resulting compromise led to an unconventional tower design,
 4.2. Integrated design optimization 67

Figure 4.10: Trade-offs between rotor speed variation and system costs in the
multiobjective optimization.

where the diameter increased from the base. The importance of integrated
design optimization was demonstrated here, as simultaneous design of the
platform and tower is needed to identify the optimal solution near the in-
terface.
Tower shape
While typical wind turbine towers are linearly tapered, the integrated op-
timization procedure employed in the present work suggested a different and
somewhat unconventional tower design. The reduction in costs achieved by
adopting such a design was quantified by optimizing the tower using three
different strategies. In all strategies, 20 years fatigue life and global buck-
ling in selected 50-year conditions were considered, while a 15◦ pitch angle
constraint was applied on the platform.
In T1, the tower and platform were optimized in an integrated fashion, as
described in previous sections. The same procedure was used in T2, but
here, the tower was constrained to a linearly tapered shape. Finally, T3
considered a procedure where only the tower was optimized with a linearly
tapered shape using a fixed platform design, taken as the optimal design
from T1. The main consequence is then that the tower base diameter is
fixed in the optimization, such that it matches the platform. The resulting
optimized tower designs are shown in Fig. 4.12.
The resulting normalized tower and platform costs are shown in Table 4.2.
The consequence of constraining the tower shape depends on the optimiz-
68 Research Findings

Figure 4.11: Optimized tower and platform design with fatigue utilization (left)
and buckling utilization (right). The wall thickness is scaled by a factor of 40
relative to the diameter for illustration purposes.
 4.2. Integrated design optimization 69

Figure 4.12: Optimized tower designs using three different design strategies,
with fatigue utilization (left) and buckling utilization (right). The wall thickness
is scaled by a factor of 40 relative to the diameter for illustration purposes.
70 Research Findings

Table 4.2: Normalized platform and tower costs using different design strategies.

Design Tower cost Platform cost Total cost

T1 1.00 1.00 1.00
T2 1.00 1.03 1.02
T3 1.29 1.00 1.11

ation strategy. In the event of an integrated optimization (T2), the tower

base diameter is significantly increased to maintain a cost-effective fatigue
design, which actually results in a tower with equal costs as the T1 design.
However, a consequence of the increased base diameter is an increase in
wave loads, which requires a more expensive platform design to satisfy the
15◦ pitch criterion. For the tower optimized with a fixed platform design
(T3), a large wall thickness is needed to satisfy the fatigue criterion at the
base, which results in an inefficient tower design with almost 30 % increase
in costs.
From the results in Figs. 4.11 and 4.12, the presented optimization method-
ology is seen to result in support structure designs which are close to fully
utilized with respect to fatigue over a large part of the length, which min-
imizes the design costs within the applied constraints. However, this affects
the fatigue reliability of the total support structure, as discussed in P5. The
reliability index after 20 years of operation for support structures optimized
with different design fatigue factors (DFFs) is shown in Fig. 4.13(a). Here,
the names of the designs indicate the DFF used for the tower, while values
twice as high were applied for the platform due to more difficult inspection
access. As expected, relatively constant reliability levels are observed along
the tower length.
For structures where a single component (i.e. potential fatigue crack loc-
ation) dominates the failure probability, system effects for the reliability
can be neglected. However, the almost constant fatigue reliability along the
length of the optimized FWT structures means that system reliability must
be considered. The support structure of a spar FWT can be modelled as
a series system, where failure in one of the components results in failure of
the whole structure.
The importance of system effects on the resulting reliability is illustrated in
Fig. 4.13(b), where the system reliability index after 20 years of operation
is shown with different number of components included in the system reli-
ability calculations. For systems with less than 12 components, the most
 4.2. Integrated design optimization 71

(b) System reliability as a function of components

included in the series system.
(a) Reliability as a function of sup-
port structure location.

Figure 4.13: Reliability index after 20 years service life.

critical ones, based on the results in Fig. 4.13(a), were included.

The importance of system effects depended on the correlation between the
components. A common assumption is to use independent fatigue strength
in different components, while the loads are taken as fully correlated. This
assumption, using typical values from the literature, resulted in relatively
low correlation values between the components.
A large reduction in reliability is observed in Fig. 4.13(b) as the number
of components increases, due to the consistent failure probabilities and low
correlation between the components. Based on the results presented here,
the fatigue safety factor needs to be doubled for a system with eight fatigue-
critical components to achieve the same reliability after 20 years as a support
structure with a single critical component.
These effects must be appropriately considered in design, which can be done
in a simple manner by increasing the required DFF along the entire support
structure, and optimizing the structural design with consistent probabilit-
ies of failure in a large number of components. However, the system effects
observed here suggest that more cost-effective designs might be achieved if
less steel-intensive components are designed with longer fatigue life (or lar-
ger DFF), in order to reduce the system effects. To derive truly optimized
fatigue designs for a given safety level, inclusion of system reliability con-
72 Research Findings

siderations within the design optimization problem is required, which has

not been considered in the present work.
4.2.4 Control strategies
Control of FWTs is a complex subject, and several strategies for the blade-
pitch controller have been suggested in the literature. All of these control
strategies result in trade-offs between structural loads, rotor speed variation,
and blade pitch actuator use, which vary with different environmental con-
ditions, and identifying optimal control parameters is a challenging task.
Also, due to the strong interactions, simultaneous design of the control-
ler and support structure should be performed to have a fair comparison
between different strategies. To properly identify and compare optimal solu-
tions, the integrated control and structural designs should be evaluated over
the lifetime of the system, considering actual design limits.
To illustrate the trade-offs between rotor speed variation, blade-pitch ac-
tuator use, and structural loads in the tower, the effect of the blade-pitch
controller PI gains on the linearized system response is shown in Fig. 4.14 for
a given support structure design. To evaluate blade-pitch actuator use, the
actuator duty cycle (ADC), defined as the total number of degrees pitched
divided by the total simulation time, is used. Here, the ADC values are
also normalized by the maximum allowable blade pitch rate for the turbine,
which for the DTU 10 MW is equal to 10 deg/s. The contours are shown for
the area where the closed-loop system is stable (i.e. no negative damping in
any modes). As shown, the optimal solution for an integrated design optim-
ization process depends heavily on the relative importance of the response
parameters.
Integrated optimization of the blade-pitch controller and support structure
design, considering different control strategies, were investigated in P4.
A baseline strategy, using a gain-scheduled PI controller, was considered
in addition to three different velocity feedback controllers, as described in
Table 4.3. In CS4, the nacelle velocity signal was passed through a first or-
der low-pass filter to remove the wave-frequency components before it was
fed back to the blade-pitch controller. On the control side, the considered
design variables were the PI gains (CS1-4), the velocity feedback gain from
Eq. (3.32) (CS2-4), and the corner frequency of the low-pass filter for the
velocity signal (CS4).
In addition to the controller parameters, the design optimization considered
the diameter and wall thickness distribution of the tower, with constraints
 4.2. Integrated design optimization 73

(a) Weighted average rotor speed variation. (b) Long-term tower base fatigue damage.

(c) Weighted average ADC.

Figure 4.14: Response parameters as functions of PI controller gains.

Table 4.3: Considered strategies for blade-pitch controller.

Strategy Description
CS1 PI controller
CS2 PI controller + platform pitch velocity feedback
CS3 PI controller + nacelle velocity feedback
CS4 PI controller + nacelle velocity feedback + low-pass filter
74 Research Findings

on long-term fatigue and global buckling. Because the structural design of

the platform was found to mostly be driven by hydrostatic buckling loads
unaffected by the controller, the study considered only the geometry of
the platform through the draft and diameter distribution, while the wall
thickness was assumed fixed. This procedure thus neglected the potential
cost savings from reduced wall thickness in the uppermost part of the hull,
where fatigue loads were more critical.
The optimization aimed to minimize the costs of the support structure,
while constraints on the rotor speed standard deviation and actuator duty
cycle (ADC) were included to ensure a certain rotor speed tracking and pitch
actuator fatigue performance. As appropriate values for these constraints
are difficult to quantify, values based on simulations with an onshore DTU
10 MW turbine were used.
The velocity feedback controllers increased the aerodynamic damping and
thus reduced the response around the pitch natural frequency. This led to
a decrease in fatigue loads, which were the design-driving constraint for the
tower. For all four design strategies, a 50-year storm condition with parked
turbine was found to be the critical load case for the extreme responses, and
the ULS design was therefore not affected by the controller.
The optimized costs of the tower, platform, and tower plus platform for
CS2-4 are shown in Fig. 4.15(a), compared to the costs for the optimized
CS1. The majority of the cost reductions came from the tower, due to the
improved fatigue behaviour, whereas the platform was less affected due to
the fixed costs related to the 15◦ pitch angle constraint. However, because
a lighter tower resulted in a lower overall center of gravity, the platform
pitch response was somewhat improved, and a small reduction in platform
costs of about 2 % was also achieved. Because the platform accounted for
70-75 % of the total costs for the considered designs, the resulting total cost
reduction was approximately 6 %.
The rotor speed standard deviations and ADCs, normalized by their max-
imum allowable values, are shown in Fig. 4.15(b). For each control strategy,
there exists a limit where no further reduction in cost can be achieved by
increasing the actuator use, as seen from Fig. 4.14. This limit is higher for
the velocity feedback controllers than for a controller using only the rotor
speed error as input, which caused the ADC constraint to be inactive at the
optimum for CS1. For the rotor speed variation, better performance was
achieved with CS4 than with the other control strategies. Since there is a
trade-off between rotor speed variation and structural loads (and thus costs)
 4.3. Environmental modelling 75

(a) Costs relative to CS1. (b) Rotor speed variation and ADC values.

Figure 4.15: Objective and constraint function values for optimized designs.

as previously discussed, it is expected that larger cost reductions could be

achieved with CS4 if the rotor speed constraint is tightened.
Verification against nonlinear simulations in SIMA showed that the tower
base fatigue damage was significantly overestimated for CS1, whereas good
agreement was obtained with the other control strategies. The reason for
the poor agreement was that the resulting aerodynamic damping for the
platform pitch mode was much lower with this control strategy. Con-
sequently, the resonant pitch response became very sensitive to the presence
and amount of additional damping in the system, which was either not con-
sidered or underpredicted in the linear model. The overestimation of fatigue
damage suggests that the cost reductions in Fig. 4.15(a) are highly optim-
istic, and that CS1 may yield a fatigue design similar to that with a nacelle
velocity feedback controller. It also suggests that future optimization stud-
ies using the linearized model from the present work should consider more
advanced control strategies than CS1, to limit the pitch response error.

4.3 Environmental modelling

Due to lack of site-specific data, or to limit the computational effort during
the design process, conservative assumptions about the environmental loads
are often used. The impact of different environmental modelling uncertain-
ties on the fatigue reliability and associated design costs was assessed in P5,
using an environmental model based on hindcast data from a realistic float-
ing wind park site. In addition to the base case, four different environmental
models were considered:

• Stochastic turbulence intensity described by the Weibull distribution

from IEC (2005), rather than the 90 % design value.
• Inclusion of the wind directional distribution.
76 Research Findings

• Inclusion of the relative wind-wave direction distribution.

• A two-peak wave spectrum to account for swell.

The resulting equivalent fatigue factors with the different environmental

models, based on Monte Carlo simulations on the linearized FWT model,
are presented in Fig. 4.16(a). Here, the equivalent fatigue factor is defined as
the maximum expected 20-year fatigue damage around the circumference,
normalized by the base case results. The factors are shown for the fatigue-
critical parts of the support structure, which consisted of the tower and
upper part of the platform. For depths below approximately 10 m, shell
buckling became the design driving constraint, and the differences in fatigue
loads did not affect the design process.
The effect of turbulence and wind-wave misalignment varied significantly
along the length of the structure due to differences in the relative import-
ance of wind and wave response. The tower top response, which was almost
exclusively governed by wind loads, was much more affected by the turbu-
lence modelling than the tower base, and vice versa for the wind-wave mis-
alignment. This also resulted in some location-dependent effects when the
wind directional distribution was included. The inclusion of swell through
a two-peaked wave spectrum had small impact on the fatigue reliability for
the considered support structure.
Applying all the aforementioned environmental model uncertainties resulted
in an almost constant reduction in fatigue damage of about 65 % along the
length of the tower and upper part of the platform. This indicates that an
additional safety factor on fatigue of about three is implicitly applied when
these model uncertainties are neglected in the design process for such FWT
structures.
Reductions in support structure costs with the different environmental mod-
els were quantified through re-design of the support structure, using a nu-
merical optimization scheme as described in the previous sections. To ac-
count for the environmental modelling effects in the design optimization
procedure, the equivalent fatigue factors from Fig. 4.16(a) were included
in the applied fatigue safety factors along the tower and platform. The
resulting cost reductions are shown in Fig. 4.16(b), where total cost reduc-
tions in the order of 5-11 % is observed. These numbers are only indicative,
and depend on the platform concept, considered design constraints, and the
metocean conditions at the actual wind park site.
 4.4. Inspection schedule 77

(b) Comparison of support structure costs for op-

timized designs.

(a) Equivalent fatigue factors.

Figure 4.16: Effect of environmental modelling on long-term fatigue and design

costs.

4.4 Inspection schedule

This section considers derivation of cost-optimal DFFs, including costs re-
lated to construction and inspections during the operational lifetime. The
required fatigue crack inspection interval for a FWT depends on the applied
DFF, which results in trade-offs between CAPEX and OPEX. To minimize
the cost of energy, the DFF resulting in the lowest lifetime costs should be
applied. This issue was investigated in P5, based on probabilistic fracture
mechanics and reliability updating using Bayes theorem. The study was per-
formed with the three different set of DFFs described in Section 4.2.3. To
evaluate lifetime costs of the considered designs, necessary inspection inter-
vals with DFF1 and DFF2 to achieve the same reliability as DFF3 without
inspections at the end of the service life were identified. The inspections
were assumed to occur with fixed intervals for each design.
Figure 4.17(a) shows the accumulated reliability index over the service life of
the system, assuming inspections with no detection of cracks every two and
five years for DFF1 and DFF2, respectively. This corresponds to a total
of nine and three inspections during the lifetime of the structure, which
resulted in reliability indices within 5 % of DFF3 without inspections after
20 years.
Based on the results in Fig. 4.17(a), trade-offs between CAPEX and OPEX
were derived. In Fig. 4.17(b), the lifetime costs are shown for different
78 Research Findings

(a) System reliability index during service (b) Combined design and inspection costs
life, including the effect of inspections. comparison, r=0.05.

Figure 4.17: Reliability and total costs with different DFFs and inspection sched-
ules.

values of the inspection costs, assuming a real interest rate of 5 %. In

this particular case, DFF1 was the most cost-effective solution for costs per
inspection lower than 210 000 e while DFF3 was the cost-optimal design
for costs of more than 650 000 e per inspection. This corresponded to about
2 % and 6 % of the support structure costs with DFF1, respectively.
In the present work, higher DFFs were achieved by increasing the struc-
tural dimensions. The fatigue life could also be extended by reducing the
SCF, which may be a more cost-effective option. However, the cost of im-
proved fabrication tolerances or detail geometry is highly uncertain, and
SCF reduction was therefore not considered.
Chapter 5

Conclusions and
Recommendations for Future
Work

This chapter summarizes the contributions of this thesis to the research

community. Suggestions for future studies are also made, based on limita-
tions identified in the current work.

5.1 Conclusions
This thesis has sought to improve the design process for FWTs through
the development of efficient methods for global dynamic response analyses,
and numerical design optimization techniques. The main focus has been
on the support structure for 10 MW spar-type turbines, considering fatigue
and ultimate loads. Some attention has also been given to the blade-pitch
controller and mooring system, particularly related to coupling effects with
the support structure.
A linearized aero-hydro-servo-elastic model with four structural DOFs was
shown to yield acceptable accuracy for the support structure response with
co-directional wind and waves. Nonlinear effects resulted in poor perform-
ance of the linearized model when the tower bending natural frequency was
close to the 3P range. Some overestimation was also observed for the res-
onant platform pitch response, which caused the model to be conservative
in general compared to nonlinear analyses. Analytical frequency-dependent
aerodynamic damping and inertia coefficients were derived, which removed

79
80 Conclusions and Recommendations for Future Work

the underestimation of the low-frequency response observed with constant

damping coefficients.
A gradient-based optimization approach with analytic derivatives was used
to perform integrated design optimization of the support structure, blade-
pitch controller, and mooring system for a 10 MW spar FWT, including the
scantling design of the hull, where the goal was to minimize the system costs
and rotor speed variation. An hour-glass shape was found to be favourable
for the platform below the wave zone, as it increased the effective rotational
stiffness of the system and improved the low-frequency behaviour in pitch.
The inclusion of structural hull design ensured a realistic distribution of
material and proper penalization of designs with large drafts. Due to com-
pliance with the platform, the tower took an unconventional shape, where
the diameter increased from the base. Integrated optimization is needed to
identify the optimal solution near the interface.
The optimized support structure was fully utilized with respect to fatigue
along most of the tower length, as well as in the uppermost part of the
platform. Below approximately 10 m below the SWL, shell buckling became
design-driving due to the large hydrostatic pressure. The consistently high
fatigue utilization in the upper part of the structure led to important system
effects for the total fatigue reliability. By considering the support structure
as a series system, an additional fatigue safety factor of two was needed
for a system with eight fatigue-critical welds, to achieve the same reliability
after 20 years as a single critical component. System effects may thus be
important for cost-optimal designs with a given target safety level.
Trade-offs between rotor speed tracking, blade-pitch actuator use, and struc-
tural loads make controller design a challenging task, and due to import-
ant interaction effects, simultaneous design of the controller and support
structure should be preferably be performed. The blade-pitch controller
affected the fatigue loads in the tower and upper part of the platform due
to damping of the resonant pitch response; however, because the critical
extreme responses occurred in either below-rated or parked conditions, no
effect was observed on the ULS design. Different control strategies were
compared through integrated design optimization with constraints on rotor
speed variation and blade-pitch actuator use. While appropriate values for
these constraints are difficult to quantify, the methodology allowed for iden-
tification of optimal control parameters in a lifetime perspective, and fair
comparison between different solutions.
Uncertainties in environmental input parameters used in global design ana-
 5.2. Recommendations for future work 81

lyses are often chosen based on conservative assumptions. The impact of

these assumptions on the long-term fatigue reliability and associated costs
of the support structure was assessed using Monte Carlo simulations and
re-design of the tower and platform based on derived fatigue factors. Wind-
wave misalignment, wind directionality, and stochastic turbulence intensity
were all shown to have non-negligible effect on the long-term fatigue dam-
age. The inclusion of all three resulted in an almost constant reduction in
fatigue along the support structure length of about two-thirds, suggesting
that an additional safety factor of three is inherent in the fatigue design
when these effects are neglected.

5.2 Recommendations for future work

Based on the studies conducted in P1-5, the following suggestions are made
for future work:

• Alternative platform concepts:

The developed numerical design optimization methodology is only ap-
plicable for spar platforms, which imposes serious limitations on the
design space for the floater. A more general description of the plat-
form geometry would allow the optimizer to explore other parts of
the design space, and possibly identify non-intuitive and more cost-
effective solutions. This requires a numerical panel method for the
hydrodynamic loads, with the support of gradient calculations. For
structural optimization of a general platform shape, a more sophistic-
ated finite element model, using e.g. shell elements, may be needed.
• Nonlinear and transient events:
The linearized dynamic model developed in the present work cannot
be used to analyse transient or highly nonlinear events. Because such
conditions may be design-driving for certain parts of the system, it is
recommended to focus on computationally efficient methods to include
them in the design optimization process.
• Improved cost models:
The optimal design from a numerical optimization process depends on
the cost models. While material costs can be calculated in a straight-
forward manner, there are large uncertainties related to manufactur-
ing, installation, maintenance, and decommissioning costs. If more
reliable engineering cost models can be established, based on the ac-
tual lifetime costs of the system, this may have a large effect on the
resulting cost-optimal designs.
82 Conclusions and Recommendations for Future Work

• Mooring system and power cable representation:

A more detailed description of the mooring system, including bridle
lines, varying cross sections, and discrete components such as clump
weights and buoys, would enable more complete and realistic optimiz-
ation of the mooring design. Together with wind and wave direction-
ality, such procedures could also consider the layout of the mooring
system. A model describing the power cable dynamics would allow
for simultaneous design optimization of the cable, including explicit
constraints on the curvature.

• Blade and drivetrain response:

A fully integrated FWT design optimization procedure must also con-
sider simultaneous design of the rotor. This requires more detailed
structural and aerodynamic models to accurately capture the blade
and drivetrain response, as well as appropriate cost models and con-
straints for these subsystems.

• Reliability-based design optimization:

FWTs are subjected to a stochastic environment, and uncertainties
in loads and response are typically considered through partial safety
factors. The design conservatism could be reduced through the ap-
plication of probabilistic design, where the structure is optimized to
minimize the cost for a given target probability of failure.

• Farm-level design optimization:

Similarly to an integrated design optimization approach on the turbine
level, design optimization considering farm-level effects could affect
and improve the overall system behaviour. Important considerations
include wake effects, total wind farm costs, system reliability effects,
and possibly shared mooring lines or anchors.
Bibliography

N. Aggarwal, R. Manikandan, and N. Saha. Nonlinear short term extreme

response of spar type floating offshore wind turbines. Ocean Engineering,
130:199–209, 2017. doi: 10.1016/j.oceaneng.2016.11.062.

T. Ashuri, M. B. Zaaijer, J. R. R. A. Martins, G. J. W. van Bussel, and

G. A. M. van Kuik. Multidisciplinary design optimization of offshore
wind turbines for minimum levelized cost of energy. Renewable Energy,
68:893–905, 2014. doi: 10.1016/j.renene.2014.02.045.

J. Azcona, D. Palacio, X. Munduate, L. González, and T. A. Nygaard. Im-

pact of mooring lines dynamics on the fatigue and ultimate loads of three
offshore floating wind turbines computed with IEC 61400-3 guideline.
Wind Energy, 20:797–813, 2017. doi: 10.1002/we.2064.

E. E. Bachynski. Design and dynamic analysis of tension leg platform wind

turbines. PhD thesis, Norwegian University of Science and Technology,
2014.

E. E. Bachynski and T. Moan. Linear and nonlinear analysis of tension

leg platform wind turbines. In Proceedings of the Twenty-second (2012)
International Offshore and Polar Engineering Conference (ISOPE2012),
Rhodes, Greece, 2012.

E. E. Bachynski and T. Moan. Ringing loads on tension leg plat-

form wind turbines. Ocean Engineering, 84:237–248, 2014. doi:
10.1016/j.oceaneng.2014.04.007.

E. E. Bachynski, M. Etemaddar, M. I. Kvittem, C. Luan, and T. Moan.

Dynamic analysis of floating wind turbines during pitch actuator fault,

83
84 BIBLIOGRAPHY

grid loss, and shutdown. Energy Procedia, 35:210–222, 2013. doi:

10.1016/j.egypro.2013.07.174.

E. E. Bachynski, M. I. Kvittem, C. Luan, and T. Moan. Wind-wave mis-

alignment effects on floating wind turbines: motions and tower load ef-
fects. Journal of Offshore Mechanics and Arctic Engineering, 136, 2014.
doi: 10.1115/1.4028028.

E. E. Bachynski, M. Thys, T. Sauder, V. Chabaud, and L. O. Sæther.

Real-time hybrid model tests of a braceless semi-submersible wind tur-
bine. Part II: experimental results. In Proceedings of the ASME 2016
35th International Conference on Ocean, Offshore and Arctic Engineer-
ing (OMAE2016), Busan, South Korea, 2016. doi: 10.1115/OMAE2016-
54437.

C. Bak, F. Zahle, R. Bitsche, A. Yde, L. C. Henriksen, A. Natarajan, and

M. H. Hansen. Description of the DTU 10 MW reference wind turbine.
Technical Report DTU Wind Energy Report-I-0092, DTU Wind Energy,
2013.

J. G. Balchen, T. Andresen, and B. A. Foss. Reguleringsteknikk [Control

Engineering] (in Norwegian). Institutt for teknisk kybernetikk, NTNU,
2003.

L. Barj, S. Stewart, G. M. Stewart, M. Lackner, J. Jonkman, A. Robertson,

and D. Matha. Wind/wave misalignment in the loads analysis of a floating
offshore wind turbine. In 32nd ASME Wind Energy Symposium, National
Harbor, Maryland, 2014. doi: 10.2514/6.2014-0363.

I. Bayati, M. Belloli, and A. Facchinetti. Wind tunnel 2-DOF hybrid/HIL

tests on the OC5 floating offshore wind turbine. In Proceedings of the
ASME 2017 36th International Conference on Ocean, Offshore and Arctic
Engineering (OMAE2017), Trondheim, Norway, 2017.

I. Bayati, A. Facchinetti, A. Fontanella, and M. Belloli. 6-DOF hydro-

dynamic modelling for wind tunnel hybrid/HIL tests of FOWT: the real-
time challenge. In Proceedings of the ASME 2018 37th International Con-
ference on Ocean, Offshore and Arctic Engineering (OMAE2018), Mad-
rid, Spain, 2018. doi: 10.1115/OMAE2018-77804.

D. Benasciutti and R. Tovo. Spectral methods for lifetime prediction under

wide-band stationary random processes. International Journal of Fatigue,
27:867–877, 2005. doi: 10.1016/j.ijfatigue.2004.10.007.
 BIBLIOGRAPHY 85

G. Benassai, A. Campanile, V. Piscopo, and A. Scamardella. Ulti-

mate and accidental limit state design for mooring systems of float-
ing offshore wind turbines. Ocean Engineering, 92:64–74, 2014. doi:
10.1016/j.oceaneng.2014.09.036.
P. A. Berthelsen, E. E. Bachynski, M. Karimirad, and M. Thys. Real-time
hybrid model tests of a braceless semi-submersible wind turbine. Part
III: calibration of a numerical model. In Proceedings of the ASME 2016
35th International Conference on Ocean, Offshore and Arctic Engineer-
ing (OMAE2016), Busan, South Korea, 2016. doi: 10.1115/OMAE2016-
54640.
L. E. Borgman. Ocean wave simulation for engineering design. Journal of
the Waterways and Harbors Division, 95(4):557–586, 1969.
S. Boyd and L. Vandenberghe. Convex Optimization. Cambridge University
Press, 2004.
M. Brommundt, L. Krause, K. Merz, and M. Muskulus. Mooring system
optimization for floating wind turbines using frequency domain analysis.
Energy Procedia, 24:289–296, 2012. doi: 10.1016/j.egypro.2012.06.111.
M. L. Buhl. A new empirical relationship between thrust coefficient and
induction factor for the turbulent windmill state. Technical Report
NREL/TP-500-36834, National Renewable Energy Laboratory, 2005.
T. Burton, N. Jenkins, D. Sharpe, and E. Bossanyi. Wind energy handbook.
Wiley, second edition, 2011.
P. K. Chaviaropoulos, A. Natarajan, and P. H. Jensen. Key performance
indicators and target values for multi-megawatt offshore turbines. In Pro-
ceedings of European Wind Energy Association Conference and Exhibition
2014 (EWEA 2014), 2014.
C.-T. Chen. Linear system theory and design. Oxford University Press,
fourth edition, 2013.
K. H. Chew, K. Tai, E. Y. K. Ng, and M. Muskulus. Analytical gradient-
based optimization of offshore wind turbine substructures under fa-
tigue and extreme loads. Marine Structures, 47:23–41, 2016. doi:
10.1016/j.marstruc.2016.03.002.
G. F. Clauss and L. Birk. Hydrodynamic shape optimization of large off-
shore structures. Applied Ocean Research, 18(4):157–171, 1996. doi:
10.1016/S0141-1187(96)00028-4.
86 BIBLIOGRAPHY

T. Dirlik. Application of computers in fatigue analysis. PhD thesis, Univer-

sity of Warwick, 1985.

DNV GL. Loads and site conditions for wind turbines. Technical Report
DNVGL-ST-0437, DNV GL, 2016.

DNV GL. Modelling and analysis of marine operations. Technical Report

DNVGL-RP-N103, DNV GL, 2017.

DNV GL. Floating wind turbine structures. Technical Report DNVGL-ST-

0119, DNV GL, 2018a.

DNV GL. Support structures for wind turbines. Technical Report DNVGL-
ST-0126, DNV GL, 2018b.

DNV GL. Environmental conditions and environmental loads. Technical

Report DNVGL-RP-C205, DNV GL, 2019a.

DNV GL. Buckling Strength of Shells. Technical Report DNVGL-RP-C202,

DNV GL, 2019b.

DNV GL. Fatigue design of offshore steel structures. Technical Report

DNVGL-RP-C203, DNV GL, 2019c.

T. Duarte, A. J. Sarmento, and J. Jonkman. Effects of second-order hy-

drodynamic forces on floating offshore wind turbines. In 32nd ASME
Wind Energy Symposium, National Harbor, Maryland, 2014. doi:
10.2514/6.2014-0361.

O. M. Faltinsen. Sea loads on ships and offshore structures. Cambridge

University Press, 1990.

J. Farkas and K. Jármai. Optimum design of steel structures. Springer,

2013.

T. Fischer, P. Rainey, E. Bossanyi, and M. Kühn. Study on control con-

cepts suitable for mitigation of loads from misaligned wind and waves on
offshore wind turbines supported on monopiles. Wind Engineering, 35(5):
561–574, 2011. doi: 10.1260/0309-524X.35.5.561.

P. A. Fleming, I. Pineda, M. Rossetti, A. D. Wright, and D. Arora. Evalu-

ating methods for control of an offshore floating turbine. In Proceedings
of the ASME 2014 33rd International Conference on Ocean, Offshore and
Arctic Engineering (OMAE2014), San Francisco, California, USA, 2014.
doi: 10.1115/OMAE2014-24107.
 BIBLIOGRAPHY 87

P. A. Fleming, A. Peiffer, and D. Schlipf. Wind turbine controller to mitigate

structural loads on a floating wind turbine platform. Journal of Offshore
Mechanics and Arctic Engineering, 141, 2019. doi: 10.1115/1.4042938.

I. Fylling and P. A. Berthelsen. WINDOPT- an optimization tool for float-

ing support structures for deep water wind turbines. In Proceedings of the
ASME 2011 30th International Conference on Ocean, Offshore and Arc-
tic Engineering (OMAE2011), Rotterdam, The Netherlands, 2011. doi:
10.1115/OMAE2011-49985.

Z. Gao and T. Moan. Frequency-domain fatigue analysis of wide-

band stationary Gaussian processes using a trimodal spectral formula-
tion. International Journal of Fatigue, 30(10-11):1944–1955, 2008. doi:
10.1016/j.ijfatigue.2008.01.008.

A. J. Goupee, R. W. Kimball, and H. J. Dagher. Exper-

imental observations of active blade pitch and generator con-
trol influence on floating wind turbine response. Renewable En-
ergy, 104:9–19, 2017. doi: 10.1016/j.renene.2016.11.062. URL
http://dx.doi.org/10.1016/j.renene.2016.11.062.

J. S. Gray, J. T. Hwang, J. R. R. A. Martins, K. T. Moore, and B. A.

Naylor. OpenMDAO: an open-source framework for multidisciplinary
design, analysis, and optimization. Structural and Multidisciplinary Op-
timization, 59:1075–1104, 2019. doi: 10.1007/s00158-019-02211-z.

A. Halfpenny. Dynamic analysis of both on and offshore wind turbines in

the frequency domain. PhD thesis, University College London, 1998.

M. Hall, B. Buckham, and C. Crawford. Evolving offshore wind: a

genetic algorithm-based support structure optimization framework for
floating wind turbines. In OCEANS 2013 MTS/IEEE Bergen: The
Challenges of the Northern Dimension, 2013. doi: 10.1109/OCEANS-
Bergen.2013.6608173.

M. Hall, B. Buckham, and C. Crawford. Hydrodynamics-based floating

wind turbine support platform optimization: a basis function approach.
Renewable Energy, 66:559–569, 2014a. doi: 10.1016/j.renene.2013.12.035.

M. Hall, B. Buckham, and C. Crawford. Evaluating the importance of

mooring line model fidelity in floating offshore wind turbine simulations.
Wind Energy, 17:1835–1853, 2014b. doi: 10.1002/we.1669.
88 BIBLIOGRAPHY

H. A. Haslum and O. M. Faltinsen. Alternative shape of spar platforms

for use in hostile areas. In Offshore Technology Conference, 1999. doi:
10.4043/10953-ms.
J.-T. Horn, J. R. Krokstad, and B. J. Leira. Impact of model uncertainties
on the fatigue reliability of offshore wind turbines. Marine Structures, 64:
174–185, 2019. doi: 10.1016/j.marstruc.2018.11.004.
IEA. Offshore wind outlook 2019. Technical report, International Energy
Agency, 2019.
IEC. Wind turbines - part 1: design requirements. Technical Report IEC
61400-1, International Electrotechnical Commission, 2005.
IEC. Wind turbines - part 3: design requirements for offshore wind turbines.
Technical Report IEC 61400-3, International Electrotechnical Commis-
sion, 2009.
K. Johannessen, T. S. Meling, and S. Haver. Joint Distribution for Wind
and Waves in the Northern North Sea. International Journal of Offshore
and Polar Engineering, 12(1):1–8, 2002.
N. Johnson, J. Jonkman, A. Wright, G. Hayman, and A. Robertson. Verific-
ation of floating offshore wind linearization functionality in OpenFAST.
Journal of Physics: Conference Series, 1356, 2019. doi: 10.1088/1742-
6596/1356/1/012022.
J. Jonkman. Influence of control on the pitch damping of a floating wind
turbine. In 46th AIAA Aerospace Science Meeting and Exhibit, Reno,
Nevada, 2008. doi: 10.2514/6.2008-1306.
J. Jonkman and W. Musial. Offshore code comparison collaboration (OC3)
for IEA task 23 offshore wind technology and deployment. Technical
Report NREL/TP-5000-48191, National Renewable Energy Laboratory,
2010.
J. Jonkman, S. Butterfield, W. Musial, and G. Scott. Definition of a 5-MW
reference wind turbine for offshore system development. Technical Report
NREL/TP-500-38060, National Renewable Energy Laboratory, 2009.
J. M. Jonkman, A. D. Wright, G. J. Hayman, and A. N. Robertson. Full-
system linearization for floating offshore wind turbines in OpenFAST. In
Proceedings of the ASME 2018 1st International Offshore Wind Technical
Conference (IOWTC2018), San Francisco, California, USA, 2018. doi:
10.1115/IOWTC2018-1025.
 BIBLIOGRAPHY 89

A. M. P. Jurado, M. Borg, and H. Bredmose. An efficient frequency-domain

model for quick load analysis of floating offshore wind turbines. Wind
Energy Science, 3(2):693–712, 2018. doi: 10.5194/wes-3-693-2018.

M. Karimi, M. Hall, B. Buckham, and C. Crawford. A multi-objective

design optimization approach for floating offshore wind turbine support
structures. Journal of Ocean Engineering and Marine Energy, 3(1):69–87,
2017. doi: 10.1007/s40722-016-0072-4.

M. Karimirad and T. Moan. Extreme dynamic structural response analysis

of catenary moored spar wind turbine in harsh environmental conditions.
Journal of Offshore Mechanics and Arctic Engineering, 133, 2011. doi:
10.1115/1.4003393.

G. Katsouris and A. Marina. Cost modelling of floating wind farms. Tech-

nical Report ECN-E–15- 078, ECN, 2016.

J. M. Kluger, T. P. Sapsis, and A. H. Slocum. A reduced-order, statistical

linearization approach for estimating nonlinear floating wind turbine re-
sponse statistics. In Proceedings of the Twenty-sixth (2016) International
Ocean and Polar Engineering Conference (ISOPE2016), Rhodes, Greece,
2016.

B. J. Koo, M. H. Kim, and R. E. Randall. Mathieu instability of a spar plat-

form with mooring and risers. Ocean Engineering, 31:2175–2208, 2004.
doi: 10.1016/j.oceaneng.2004.04.005.

M. I. Kvittem and T. Moan. Frequency versus time domain fatigue analysis

of a semisubmersible wind turbine tower. Journal of Offshore Mechanics
and Arctic Engineering, 137, 2014. doi: 10.1115/1.4028340.

M. Lackner. Controlling platform motions and reducing blade loads for

floating wind turbines. Wind Engineering, 33(6):541–553, 2009. doi:
10.1260/0309-524X.33.6.541.

A. B. Lambe and J. R. R. A. Martins. Extensions to the design structure

matrix for the description of multidisciplinary design, analysis, and op-
timization processes. Structural and Multidisciplinary Optimization, 46:
273–284, 2012. doi: 10.1007/s00158-012-0763-y.

K. Larsen and P. C. Sandvik. Efficient methods for the calculation of dy-

namic mooring line tension. In Proceedings of the First (1990) European
Offshore Mechanics Symposium, Trondheim, Norway, 1990.
90 BIBLIOGRAPHY

T. J. Larsen and T. D. Hanson. A method to avoid negative damped

low frequent tower vibrations for a floating, pitch controlled wind tur-
bine. Journal of Physics: Conference Series, 75:012073, 2007. doi:
10.1088/1742-6596/75/1/012073.
C. F. Lee, E. E. Bachynski, and A. R. Nejad. Consequences of Load Mitiga-
tion Control Strategies for a Floating Wind Turbine. Journal of Physics:
Conference Series, 2020.
M. Leimeister, A. Kolios, and M. Collu. Critical review of floating sup-
port structures for offshore wind farm deployment. Journal of Physics:
Conference Series, 1104, 2018.
M. Leimeister, A. Kolios, M. Collu, and P. Thomas. Design optimization of
the OC3 phase IV floating spar-buoy, based on global limit states. Ocean
Engineering, 202, 2020. doi: 10.1016/j.oceaneng.2020.107186.
F. Lemmer, K. Müller, W. Yu, D. Schlipf, and P. W. Cheng. Optimiza-
tion of floating offshore wind turbine platforms with a self-tuning con-
troller. In Proceedings of the ASME 2017 36th International Conference
on Ocean, Offshore and Arctic Engineering (OMAE2017), Trondheim,
Norway, 2017. doi: 10.1115/OMAE2017-62038.
H. Lie and N. Sødahl. Simplified dynamic model for estimation of extreme
anchor line tension. In Offshore Australia, The 2nd Australian Oil, Gas
& Petrochemical Exhibition and Conference, Melbourne, Australia, 1993.
R. C. MacCamy and R. A. Fuchs. Wave forces on piles: a diffraction the-
ory. Technical Report Technical Memorandum 69, Beach Erosion Board;
Corps of Engineers, 1954.
H. O. Madsen, S. Krenk, and N. C. Lind. Methods of structural safety.
Prentice-Hall, 1986.
H. O. Madsen, J. D. Sørensen, and R. Olesen. Optimal inspection planning
for fatigue damage of offshore structures. In Proceedings of ICOSSAR’89,
the 5th International Conference on Structural Safety and Reliability, San
Francisco, USA, 1989.
J. F. Manwell, J. G. McGowan, and A. L. Rogers. Wind energy explained.
Wiley, second edition, 2009.
S. Márquez-Domínguez and J. D. Sørensen. Fatigue reliability and calib-
ration of fatigue design factors for offshore wind turbines. Energies, 5:
1816–1834, 2012. ISSN 19961073. doi: 10.3390/en5061816.
 BIBLIOGRAPHY 91

J. R. R. A. Martins. Perspectives on aerodynamic design optimization. In

AIAA Scitech 2020 Forum, Orlando, FL, 2020. doi: 10.2514/6.2020-0043.

J. R. R. A. Martins and A. Ning. Engineering design optimization. Unpub-

lished draft, 2020.

D. Matha, F. Sandner, C. Molins, A. Campos, and P. W. Cheng. Efficient

preliminary floating offshore wind turbine design and testing methodolo-
gies and application to a concrete spar design. Philosophical Transactions
of the Royal Society A, 373(20140350), 2015. doi: 10.1098/rsta.2014.0350.

K. Merz. A linear state-space model of an offshore wind turbine, implemen-

ted in the STAS wind power plant analysis program. Technical Report
TR A7474, SINTEF Energy Research, 2015.

P. J. Moriarty and A. C. Hansen. AeroDyn theory manual. Technical Report

NREL/TP-500-36881, National Renewable Energy Laboratory, 2005.

K. Müller, D. Matha, S. Tiedemann, R. Proskovics, and F. Lemmer.

LIFES50+ D7.5: guidance on platform and mooring line selection, install-
ation and marine operations. Technical report, University of Stuttgart,
2016.

K. Müller, F. Lemmer, and W. Yu. LIFES50+ D4.2: public definition of

the two LIFES50+ 10MW floater concepts. Technical report, University
of Stuttgart, 2018.

K. Müller, R. F. Guzmán, S. Zhou, M. Lerch, R. Proskovics, G. Pérez,

I. Mendikoa, D. Matha, F. Borisade, J. Bhat, R. Scheffler, M. Thys,
H. Bredmose, F. Madsen, and A. M. P. Jurado. LIFES50+ D7.11: design
practice for 10MW+ FOWT suport structures. Technical report, Univer-
sity of Stuttgart, 2019.

W. Musial, S. Butterfield, and B. Ram. Energy from offshore wind. In

Offshore Technology Conference, 2006. doi: 10.4043/18355-ms.

M. Muskulus. Simplified rotor load models and fatigue damage estimates for
offshore wind turbines. Philosophical Transactions of the Royal Society
A, 373(20140347), 2015. doi: 10.1098/rsta.2014.0347.

M. Muskulus and S. Schafhirt. Design optimization of wind turbine support

structures - A review. Journal of Ocean and Wind Energy, 1(1):12–22,
2014.
92 BIBLIOGRAPHY

A. Myhr and T. A. Nygaard. Load reductions and optimizations on tension-

leg-buoy offshore wind turbine platforms. In Proceedings of the Twenty-
second (2012) International Offshore and Polar Engineering Conference
(ISOPE2012), Rhodes, Greece, 2012.
A. Naess. Technical note: on a rational approach to extreme value ana-
lysis. Applied Ocean Research, 6(3):173–174, 1984. doi: 10.1016/0141-
1187(84)90007-5.
A. Naess and T. Moan. Stochastic dynamics of marine structures. Cam-
bridge University Press, 2013.
A. R. Nejad, Z. Gao, and T. Moan. On long-term fatigue damage and
reliability analysis of gears under wind loads in offshore wind turbine
drivetrains. International Journal of Fatigue, 61:116–128, 2014. doi:
10.1016/j.ijfatigue.2013.11.023.
A. R. Nejad, E. E. Bachynski, M. I. Kvittem, C. Luan, Z. Gao, and
T. Moan. Stochastic dynamic load effect and fatigue damage ana-
lysis of drivetrains in land-based and TLP, spar and semi-submersible
floating wind turbines. Marine Structures, 42:137–153, 2015. doi:
10.1016/j.marstruc.2015.03.006.
A. R. Nejad, E. E. Bachynski, and T. Moan. Effect of axial accel-
eration on drivetrain responses in a spar-type floating wind turbine.
Journal of Offshore Mechanics and Arctic Engineering, 141, 2019. doi:
10.1115/1.4041996.
J. N. Newman. Marine hydrodynamics. The MIT Press, 1977.
A. Ning and D. Petch. Integrated design of downwind land-based wind
turbines using analytic gradients. Wind Energy, 19:2137–2152, 2016. doi:
10.1002/we.1972.
S. A. Ning. A simple solution method for the blade element momentum
equations with guaranteed convergence. Wind Energy, 17(9):1327–1345,
2013. doi: 10.1002/we.1636.
J. Nocedal and S. J. Wright. Numerical optimization. Springer, second
edition, 2006.
J. Oest, R. Sørensen, L. C. T. Overgaard, and E. Lund. Structural optimiz-
ation with fatigue and ultimate limit constraints of jacket structures for
large offshore wind turbines. Structural and Multidisciplinary Optimiza-
tion, 55(3):779–793, 2017. doi: 10.1007/s00158-016-1527-x.
 BIBLIOGRAPHY 93

L. B. Pasamontes, F. G. Torres, D. Zwick, S. Schafhirt, and M. Muskulus.

Support structure optimization for offshore wind Turbines with a genetic
algorithm. In Proceedings of the ASME 2014 33rd International Con-
ference on Ocean, Offshore and Arctic Engineering (OMAE2014), San
Francisco, California, USA, 2014. doi: 10.1115/OMAE2014-24252.

M. Philippe, A. Babarit, and P. Ferrant. Comparison of time and frequency

domain simulations of an offshore floating wind turbine. In Proceedings
of the ASME 2011 30th International Conference on Ocean, Offshore
and Arctic Engineering (OMAE2011), Rotterdam, The Netherlands, 2011.
doi: 10.1115/OMAE2011-49722.

W. Popko, F. Vorpahl, A. Zuga, M. Kohlmeier, J. Jonkman, A. Robertson,

T. J. Larsen, A. Yde, K. Sætertrø, K. M. Okstad, J. Nichols, T. A.
Nygaard, Z. Gao, D. Manolas, K. Kim, Q. Yu, W. Shi, H. Park,
A. Vásquez-Rojas, J. Dubois, D. Kaufer, P. Thomassen, M. J. de Ruiter,
T. van der Zee, J. M. Peeringa, H. Zhiwen, and H. von Waaden. Offshore
code comparison collaboration continuation (OC4), phase I - results of
coupled simulations of an offshore wind turbine with jacket support struc-
ture. In Proceedings of the Twenty-second (2012) International Offshore
and Polar Engineering Conference (ISOPE2012), Rhodes, Greece, 2012.

W. Popko, M. L. Huhn, A. Robertson, J. Jonkman, F. Wendt, K. Müller,

M. Kretschmer, F. Vorpahl, T. R. Hagen, C. Galinos, J.-B. Le Dreff,
P. Gilbert, B. Auriac, F. N. Villora, P. Schünemann, I. Bayati, M. Bel-
loli, S. Oh, Y. Totsuka, J. Qvist, E. E. Bachynski, S. H. Sørum, P. E.
Thomassen, H. Shin, F. Vittori, J. Galvan, C. Molins, P. Bonnet,
T. van der Zee, R. Bergua, K. Wang, P. Fu, and J. Cai. Verification
of a numerical model of the offshore wind turbine from the Alpha Ventus
wind farm within OC5 phase III. In Proceedings of the ASME 2018
37th International Conference on Ocean, Offshore and Arctic Engineering
(OMAE2018), Madrid, Spain, 2018.

W. Popko, A. Robertson, J. Jonkman, F. Wendt, P. Thomas, K. Müller,

M. Kretschmer, T. R. Hagen, C. Galinos, J.-B. Le Dreff, P. Gilbert,
B. Auriac, S. Oh, J. Qvist, S. H. Sørum, L. Suja-Thauvin, H. Shin,
C. Molins, P. Trubat, P. Bonnet, R. Bergua, K. Wang, P. Fu, J. Cai,
Z. Cai, A. Alexandre, and R. Harries. Validation of numerical models of
the offshore wind turbine from the Alpha Ventus wind farm against full-
scale measurements within OC5 phase III. In Proceedings of the ASME
2019 38th International Conference on Ocean, Offshore and Arctic En-
gineering (OMAE2019), Glasgow, Scotland, UK, 2019.
94 BIBLIOGRAPHY

G. K. V. Ramachandran, L. Vita, A. Krieger, and K. Mueller. Design Basis

for the Feasibility Evaluation of Four Different Floater Designs. Energy
Procedia, 137:186–195, 2017. doi: 10.1016/j.egypro.2017.10.345.
L. Roald, J. Jonkman, A. Robertson, and N. Chokani. The effect of second-
order hydrodynamics on floating offshore wind turbines. Energy Procedia,
35:253–264, 2013. doi: 10.1016/j.egypro.2013.07.178.
A. Robertson, J. Jonkman, F. Vorpahl, W. Popko, J. Qvist, L. Froyd,
X. Chen, J. Azcona, E. Uzunoglu, C. G. Soares, C. Luan, H. Yutong,
F. Pengcheng, A. Yde, T. Larsen, J. Nichols, R. Buils, L. Lei, T. A.
Nygaard, D. Manolas, A. Heege, S. R. Vatne, H. Ormberg, T. Duarte,
C. Godreau, H. F. Hansen, A. W. Nielsen, H. Riber, C. L. Cunff, F. Beyer,
A. Yamaguchi, K. J. Jung, H. Shin, W. Shi, H. Park, M. Alves, and
M. Guérinel. Offshore code comparison collaboration continuation within
IEA wind task 30: phase II results regarding a floating semisubmersible
wind system. In Proceedings of the ASME 2014 33rd International Con-
ference on Ocean, Offshore and Arctic Engineering (OMAE2014), San
Francisco, California, USA, 2014. doi: 10.1115/OMAE2014-24040.
A. N. Robertson, F. Wendt, J. M. Jonkman, W. Popko, H. Dagher, S. Guey-
don, J. Qvist, F. Vittori, J. Azcona, E. Uzunoglu, C. G. Soares, R. Harries,
A. Yde, C. Galinos, K. Hermans, J. B. de Vaal, P. Bozonnet, L. Bouy,
I. Bayati, R. Bergua, J. Galvan, I. Mendikoa, C. B. Sanchez, H. Shin,
S. Oh, C. Molins, and Y. Debruyne. OC5 project phase II: validation
of global loads of the DeepCwind floating semisubmersible wind turbine.
Energy Procedia, 137:38–57, 2017. doi: 10.1016/j.egypro.2017.10.333.
D. Roddier, C. Cermelli, A. Aubault, and A. Peiffer. Summary and con-
clusions of the full life-cycle of the WindFloat FOWT prototype pro-
ject. Proceedings of the ASME 2017 36th International Conference on
Ocean, Offshore and Arctic Engineering (OMAE2017), Trondheim, Nor-
way, 2017. doi: 10.1115/OMAE2017-62561.
F. Sandner, D. Schlipf, D. Matha, and P. W. Cheng. Integrated optim-
ization of floating wind turbine systems. In Proceedings of the ASME
2014 33rd International Conference on Ocean, Offshore and Arctic En-
gineering (OMAE2014), San Francisco, California, USA, 2014. doi:
10.1115/OMAE2014-24244.
T. Sauder, V. Chabaud, M. Thys, E. E. Bachynski, and L. O. Sæther.
Real-time hybrid model tests of a braceless semi-submersible wind tur-
bine. Part I: the hybrid approach. In Proceedings of the ASME 2016
 BIBLIOGRAPHY 95

35th International Conference on Ocean, Offshore and Arctic Engineer-

ing (OMAE2016), Busan, South Korea, 2016.
P. Sclavounos, C. Tracy, and S. Lee. Floating offshore wind turbines: re-
sponses in a seastate Pareto optimal designs and economic assessment.
In Proceedings of the ASME 27th International Conference on Offshore
Mechanics and Arctic Engineering (OMAE2008), Estoril, Portugal, 2008.
doi: 10.1115/OMAE2008-57056.
SINTEF Ocean. RIFLEX user guide, 2016a.
SINTEF Ocean. SIMO user guide, 2016b.
B. Skaare, F. G. Nielsen, T. D. Hanson, R. Yttervik, O. Havmøller, and
A. Rekdal. Analysis of measurements and simulations from the Hywind
Demo floating wind turbine. Wind Energy, 18:1105–1122, 2015. doi:
10.1002/we.1750.
E. Smilden. Structural control of offshore wind turbines - Increasing the
role of control design in offshore wind farm development. PhD thesis,
Norwegian University of Science and Technology, 2019.
E. Smilden, E. E. Bachynski, A. J. Sørensen, and J. Amdahl. Site-specific
controller design for monopile offshore wind turbines. Marine Structures,
61:503–523, 2018. doi: 10.1016/j.marstruc.2018.03.002.
A. Smith, T. Stehly, and W. Musial. 2014–2015 offshore wind technolo-
gies market report. Technical Report NREL/TP-5000-64283, National
Renewable Energy Laboratory, 2015.
C. E. S. Souza and E. E. Bachynski. Changes in surge and pitch decay peri-
ods of floating wind turbines for varying wind speed. Ocean Engineering,
180:223–237, 2019. doi: 10.1016/j.oceaneng.2019.02.075.
C. E. S. Souza and E. E. Bachynski. Effects of hull flexibility on the
structural dynamics of a tension leg platform floating wind turbine.
Journal of Offshore Mechanics and Arctic Engineering, 142, 2020. doi:
10.1115/1.4044725.
A. Steinert, S. Ehlers, M. I. Kvittem, D. Merino, and M. Ebbesen. Cost as-
sessment for a semi-submersible floating wind turbine with respect to the
hydrodynamic response and tower base bending moments using particle
swarm optimisation. In Proceedings of the Twenty-sixth (2016) Interna-
tional Ocean and Polar Engineering Conference (ISOPE2016), Rhodes,
Greece, pages 419–426, 2016.
96 BIBLIOGRAPHY

M. Strach-Sonsalla and M. Muskulus. Dynamics and design of floating wind

turbines. In Proceedings of the Twenty-sixth (2016) International Ocean
and Polar Engineering Conference (ISOPE2016), Rhodes, Greece, 2016.

G. I. Taylor. The spectrum of turbulence. Proc. R. Soc. A, 164:476–490,

1938.

The Carbon Trust. Floating offshore wind: market and technology review.
Technical report, The Carbon Trust, 2015.

C. Tibaldi, M. H. Hansen, and L. C. Henriksen. Optimal tuning for a

classical wind turbine controller. Journal of Physics: Conference Series,
555(012099), 2014. doi: 10.1088/1742-6596/555/1/012099.

C. Tibaldi, L. C. Henriksen, M. H. Hansen, and C. Bak. Wind turbine

fatigue damage evaluation based on a linear model and a spectral method.
Wind Energy, 19:1289–1306, 2016. doi: 10.1002/we.1898.

K. Torsethaugen. Model for double peaked wave spectrum. Technical Report

STF22 A96204, SINTEF Civil and Envir. Engineering, 1996.

K. Torsethaugen and S. Haver. Simplified double peak spectral model for

ocean waves. In Proceedings of the Fourteenth (2004) International Ocean
and Polar Engineering Conference (ISOPE2004), Toulon, France, 2004.

C. Tracy. Parametric design of floating wind turbines. Master’s thesis,

Massachusetts Institute of Technology, 2007.

J. van der Tempel. Design of support structures for offshore wind turbines.
PhD thesis, TU Delft, 2006.

G. J. van der Veen, I. J. Couchman, and R. O. Bowyer. Control of

floating wind turbines. In 2012 American Control Conference, Fair-
mont Queen Elizabeth, Montreal, Canada, pages 3148–3153, 2012. doi:
10.1109/acc.2012.6315120.

J. Velarde, C. Kramhøft, J. D. Sørensen, and G. Zorzi. Fatigue reliability

of large monopiles for offshore wind turbines. International Journal of
Fatigue, 134, 2020. doi: 10.1016/j.ijfatigue.2020.105487.

F. Vorpahl, H. Schwarze, T. Fischer, M. Seidel, and J. Jonkman. Offshore

wind turbine environment, loads, simulation, and design. WIREs Energy
Environ, 2:548–570, 2013. doi: 10.1002/wene.52.
 BIBLIOGRAPHY 97

E. N. Wayman, P. Sclavounos, S. Butterfield, J. Jonkman, and W. Musial.

Coupled dynamic modeling of floating wind turbine systems. In Offshore
Technology Conference, 2006. doi: 10.4043/18287-ms.

WindEurope. Floating offshore wind energy - a policy blueprint for Europe.

Technical report, WindEurope, 2018.

WindEurope. Offshore wind in Europe - key trends and statistics 2019.

Technical report, WindEurope, 2019.

A. S. Wise and E. E. Bachynski. Wake meandering effects on floating

wind turbines. Wind Energy, 23:1266–1285, 2020. ISSN 10991824. doi:
10.1002/we.2485.

F. Zahle, C. Tibaldi, D. R. Verelst, C. Bak, R. Bitsche, and J. P. Blasques.

Aero-elastic optimization of a 10 MW wind turbine. In 33rd ASME Wind
Energy Symposium, American Institute of Aeronautics and Astronautics,
2015. doi: 10.2514/6.2015-0491.
98 BIBLIOGRAPHY
Appendix A

Appended Papers

99
100 Appended Papers
Paper 1

A semi-analytical frequency domain model for ef-

ficient design evaluation of spar floating wind tur-
bines
John Marius Hegseth and Erin E. Bachynski
Marine Structures, 2019

101
102 Appended Papers
 0DULQH6WUXFWXUHV  ²

Contents lists available at ScienceDirect

Marine Structures
journal homepage: www.elsevier.com/locate/marstruc

A semi-analytical frequency domain model for efficient design

evaluation of spar floating wind turbines
John Marius Hegseth∗, Erin E. Bachynski
Department of Marine Technology, NTNU, 7491 Trondheim, Norway

AR T IC L E I NFO A B S T R A C T

Keywords: A linear model for efficient design evaluation of spar floating wind turbines is presented, and
Offshore wind verified against a nonlinear time domain model with regards to long-term fatigue and short-term
Design analysis extreme response for two different spar designs. The model uses generalized displacements and a
Frequency domain semi-analytical approach to establish the equations of motion for the system, which are solved in
Fatigue
the frequency domain. The results show agreement within ± 30% for the long-term fatigue
Extreme response
considering operational conditions, however, the linear fatigue damage estimates are sensitive to
the accuracy of the estimated natural frequency of the first bending mode. The results also
suggest that a small number of environmental conditions can be simulated with a nonlinear time
domain model to verify and possibly tune the linear model, which then can be used to run the full
long-term analysis. Short-term extreme tower base bending moments and surge and pitch mo-
tions are observed to be nearly Gaussian above cut-out wind speed, as the response is dominated
by wave forces. Consequently, the linear model is able to accurately capture the upcrossing rates,
which are used to calculate the characteristic largest extreme response. For an operational case
near rated wind speed, the response is somewhat non-Gaussian, which gives larger discrepancies
between the linear and nonlinear models. However, due to large mean values in this condition,
the total error in the extreme response is reduced, and reasonable agreement is achieved.

1. Introduction

The offshore wind industry has had significant growth over the last decade, and as a majority of the global wind resources are
located in deeper waters, there has recently been an increased interest in floating wind turbines (FWTs). For floating wind farms to be
economically feasible, cost-effective and reliable designs are needed. Design optimization of FWTs is a complex task, and due to
interactions between aerodynamics, hydrodynamics, structural dynamics and control, coupled nonlinear time domain (TD) simu-
lations are usually applied. As numerous load cases need to be analysed, the design process becomes computationally very expensive.
It is therefore desirable to use simplified models, especially in preliminary design [1]. Frequency domain (FD) models provide an
efficient way of performing dynamic simulations and are frequently used for floating structures in the offshore industry, usually in
fatigue analyses [2]. Although less common in the offshore wind industry, design standards state that frequency domain analyses also
can be used to calculate the fatigue loads for FWTs, if validation against time domain analyses or experiments is performed [3].
Several studies have investigated the applicability of FD models for FWTs. Typically, these models have considered first order
hydrodynamic loads from potential theory and steady aerodynamic loads, and have been used to calculate response amplitude
operators (RAOs) for the rigid body motions of the platform [4–7].

∗
Corresponding author.
E-mail address: [email protected] (J.M. Hegseth).

https://doi.org/10.1016/j.marstruc.2018.10.015
Received 10 August 2018; Received in revised form 2 October 2018; Accepted 30 October 2018
‹(OVHYLHU/WG$OOULJKWVUHVHUYHG
J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Nomenclature J n Derivative of Bessel function of the first kind of

order n
2 Dirlik irregularity factor K Generalized stiffness matrix
β Stiffness-proportional Rayleigh damping coeffi- K Material property for SN curve
cient Kr Hydrostatic restoring in pitch
Gamma function Kt Linear horizontal mooring stiffness
p Expected rate of peaks k Wave number, thickness exponent for SN curve
ρ Water density L Spar FWT draft
σ Standard deviation M Generalized mass matrix
x Axial stress M Mass of spar including mooring system
ψ Shape function MRNA Mass of RNA
ω Angular frequency MT Rotor moment
A Cross sectional area My Bending moment about y-axis
a Hull radius m Mass per unit length, slope of SN curve
a11 Transverse added mass per unit length mn nth spectral moment
B Generalized damping matrix N Axial force
Baero Aerodynamic damping r Cross section outer radius
b visc Linearized viscous damping per unit length S Wave spectrum
Cd Drag coefficient SU Wind spectrum
D Hull diameter t Wall thickness
dFW Wave excitation force per unit length tref Reference thickness for SN curve
EI Bending stiffness V Hull displacement
F Generalized wave load vector xm Dirlik mean frequency
FT Thrust force Yn Derivative of Bessel function of the second kind of
FU Generalized wind load vector order n
g Gravitational acceleration zB Vertical centre of buoyancy
Hij Transfer function from i to j zG Vertical centre of gravity
h Water depth zmoor Vertical location of the fairleads
I Area moment of inertia zTB Vertical location of the tower base
IRNA Inertia of RNA about the tower top ztop Vertical location of the tower top
Iwp Waterplane area moment of inertia

Although useful information about the system may be found using only rigid body motions, the elasticity of the tower has been shown
to significantly affect the global response of FWTs. Bachynski and Moan [8] compared a linear FD model with three rigid body modes
to a nonlinear TD model for different tension leg platform wind turbine designs in both wave-only and combined wind-wave con-
ditions. The study was later expanded to include turbulent wind excitation and aerodynamic damping [9], and concluded that the
linear model was insufficient for design calculations, partly due to the rigid modelling of the tower. Kvittem and Moan [10] used a
similar procedure to calculate tower base bending moments for a semi-submersible wind turbine, where the first bending mode of the
tower was found from a free decay test and included using generalized coordinates. The contribution from the first bending mode on
the tower base bending moment response was expressed as a dynamic amplification factor. The model was found to perform rea-
sonably well in the 13 considered load cases, however, the fatigue damage was underestimated by up to 60% due to its exponential
nature. A somewhat different approach was applied by Kluger et al. [11], who used statistical linearization to develop a FD model for
the OC3 spar wind turbine [12]. The two first elastic tower modes were found for a fixed foundation and included in the analysis, and
equivalent fatigue stress (EFS) at the tower base due to wave excitation and steady wind was calculated for 11 different environ-
mental conditions. The results were compared to values from TD simulations reported by Matha [13], where the FD analysis was
found to underestimate the total EFS by 12%.
Although earlier work has assessed the fatigue damage based on FD analyses for FWTs, these calculations have only been per-
formed for a limited number of environmental conditions. Thus, little information exists regarding the accuracy of such models for a
full long-term analysis. In addition, the design process must also consider the extreme response of the system, which can be a
comprehensive task. Results from previous studies have shown that both the tower base bending moment [14] and global motions
[15] for spar FWTs can be quite Gaussian in harsh environmental conditions, which suggests that a linear model also may be used to
assess the extreme response in early stages of design. For a spar FWT, in addition to the ultimate stresses in the structure, the extreme
surge and pitch response of the platform may be of interest, as these motions are important for loads on the mooring system and
nacelle components, respectively.
In the present work, a semi-analytical FD approach is presented, where generalized degrees-of-freedom (DOFs) are used to
describe the dynamic behaviour of the FWT. The system is linearized, and due to the simple geometry of the hull, closed-form
expressions may be used for the hydrodynamic loads with good accuracy. This removes the need for a separate numerical hydro-
dynamic analysis, which significantly increases the computational efficiency. The aerodynamic loads are found numerically from TD
simulations, and the forces are superimposed to find the total response. The approach is then compared to fully coupled nonlinear TD
simulations for two different 10 MW spar FWT designs, considering long-term fatigue damage at the tower base, as well as short-term


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 1. Spar designs considered in the present work. Spar 1 (left) and spar 2 (right).

extreme values for the tower base bending moment and surge and pitch motions of the platform in selected conditions along the 50-
year environmental contour. The linear FD model presented in the present work is not applicable for detailed analyses, but may
complement state-of-the-art tools in early stages of design. The model can help the designer get a quick overview of the response of
the system, and give indications of how the system responds to changes in the design parameters. It may also be used to identify
critical load cases, which then can be analysed using coupled nonlinear TD analyses.

2. System description

2.1. Platform designs

The two spar buoys considered in the current work are shown in Fig. 1. The designs are based on the OC3-Hywind design [12] and
consist of two cylinders with different diameters, connected by a linearly tapered section. The hull extends to a height of 10 m above
still water level (SWL), where it is fixed to the tower. Spar 1 has the same draft as the original OC3 spar of 120 m, but with increased
diameters to provide enough buoyancy to support a 10 MW wind turbine, and to match the tower base diameter. The large draft of
the spar provides good stability and hydrodynamic performance, however, it limits the use to deep water, and complicates con-
struction and towing to site. The draft of the second design (spar 2) was therefore reduced by 25%, which makes it more suitable for
intermediate water depths. In addition to the draft reduction, the lower hull diameter was increased by 25% compared to spar 1, to
achieve a sufficient amount of buoyancy.
The two designs are presented in Table 1, where values for the vertical centre of gravity (CoG) and moment of inertia are
calculated by assuming that the hulls are partially filled with concrete ballast to achieve the correct draft, using a ballast density of
2600 kg/m3, while a constant wall thickness of 6 cm is assumed for the steel hull. The water depth for both platforms is 320 m.

Table 1
Platform properties.
Parameter Spar 1 Spar 2

Draft (m) 120.0 90.0

Elevation to tower base above SWL (m) 10.0 10.0
Depth to top of taper below SWL (m) 4.0 4.0
Depth to bottom of taper below SWL (m) 12.0 12.0
Diameter above taper (m) 8.3 8.3
Diameter below taper (m) 12.0 15.0
Mass including ballast (kg) 1.18E+7 1.33E+7
Displacement (m3) 1.31E+4 1.49E+4
Moment of inertia about CoG (kgm2) 6.53E+9 3.42E+9
Vertical CoG below SWL (m) 94.7 72.3
Vertical CoB below SWL (m) 62.0 47.8


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Table 2
Main characteristics of the DTU 10 MW reference wind turbine [16].
Parameter Value

Rated power 10 MW
Rotor orientation and configuration Upwind, three blades
Rotor, hub diameter 178.3 m, 5.6 m
Hub height 119.0 m
Cut-in, rated, cut-out wind speed 4.0 m/s, 11.4 m/s, 25.0 m/s
Cut-in, rated rotor speed 6.0 rpm, 9.6 rpm
Overhang, shaft tilt, pre-cone 7.1 m, 5.0°, −2.5°
Rotor, nacelle, tower massa 230.7 t, 446.0 t, 628.4 t

a
Mass for original land-based tower.

2.2. DTU 10 MW wind turbine with modified tower

The spar buoys support the DTU 10 MW reference wind turbine [16], with the tower shortened by 10 m to achieve the same hub
height as the onshore turbine at 119 m above SWL. The tower has a linearly tapered outer diameter and consists of ten sections with
constant wall thickness in each section. In the modified tower, each section is shortened by 1 m, while the top and bottom diameters
are kept unchanged. The key characteristics of the turbine are listed in Table 2.
The natural frequencies of the tower change when the turbine is placed on a floating substructure, and preliminary calculations
showed that the natural frequency of the first bending mode for spar 2 was very close to the blade passing frequency at rated rotor
speed. The wall thickness in all sections of the tower was therefore increased by 50% for spar 2, which moved the natural frequency
above the 3P range.

2.3. Mooring system

The mooring system, described in Table 3, consists of three catenary lines spread symmetrically about the vertical axis. As in the
OC3 project, a simplification is made in that the delta lines are removed, and lines with constant properties are instead used all the
way up to the fairleads. A rotational spring is added to the model to ensure that the yaw stiffness from the mooring system is included.
The fairleads are placed at a depth equal to the total CoG of the system including the wind turbine, in order to limit the coupling
between surge and pitch motion. This results in a slightly stiffer mooring system for spar 2.

3. Time domain model

The fully coupled nonlinear aero-hydro-servo-elastic analyses are carried out in the time domain using the simulation workbench
SIMA developed by SINTEF Ocean, which couples two computer codes: Riflex, a finite element solver developed for flexible beam
elements; and SIMO, which calculates large volume hydrodynamic loads [17,18]. The spar buoys are modelled as six-DOF rigid
bodies with first order wave forces found numerically from potential flow theory using WAMIT [19], combined with viscous forces
from the drag term in Morison's equation, which are integated up to the instantaneous wave elevation. Second order potential flow
forces are not considered in the model, as studies have shown that these loads have limited effect on the response for spar-type FWTs
[20,21].
Bar elements with only axial stiffness are used to model the mooring lines, together with hydrodynamic loads from Morison's
equation. The model thus includes the nonlinear restoring forces from the mooring system, as well as the dynamic behaviour of the
mooring lines. The tower and blades are modelled using flexible beam elements.
The aerodynamic loads are calculated using blade element/momentum (BEM) theory, including Glauert correction, Prandtl hub
and tip loss factors, dynamic stall, dynamic wake, skewed inflow and tower shadow. The code has previously been verified for FWTs
[22,23]. As wind drag forces on the tower may become important at high wind speeds, this effect is included for the extreme response
conditions using a drag force formulation with a drag coefficient of 0.7.
An external control system written in Java is used to modify the generator torque and blade pitch. In order to avoid pitch motion

Table 3
Mooring system properties.
Parameter Spar 1 Spar 2

Radius to anchors (m) 855.2 855.2

Unstretched mooring line length (m) 902.2 902.2
Equivalent mooring line mass density (kg/m) 155.4 155.4
Equivalent mooring line axial stiffness (MN) 3.84E+8 3.84E+8
Fairlead depth below SWL (m) 77.2 56.3
Yaw spring stiffness (Nm/rad) 1.48E+8 1.48E+8


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

instability above rated wind speed due to negative feedback [24,25], the proportional and integral gains for the blade pitch are
modified from the original controller. With the resulting gains, the controller has a natural frequency of 0.13 rad/s and a damping
ratio of 0.7.

4. Frequency domain model

In the frequency domain model, a linear representation of the system is created as shown in Fig. 2. For simplicity, only the
response in the xz-plane is considered in the present study. The platform is considered rigid, while the tower is modelled as a slender
flexible beam. The rotor and nacelle assembly (RNA) is replaced by a point mass and inertia at the top of the tower, with resultant
wind loads (thrust force, rotor pitching moment, and aerodynamic damping) acting on the tower top. The mooring system is re-
presented by a linear spring at the position of the fairleads. As the structure is statically determinate, internal loads in the hull can be
found by considering equilibrium between external, inertial, damping, restoring and internal reaction forces. In the present work, the
tower base bending moment is the only internal load which is considered (see also Appendix A), but the method can easily be applied
to other components.
Generalized displacements are used in combination with the principle of virtual work to establish the dynamic equations of
motion [26], which are solved in the FD. In this procedure, the total response is described by a weighted combination of an arbitrary
number of shape functions, ψ. The accuracy of the formulation thus depends on how well the actual displacement field is captured by
the shape functions, which typically are chosen to represent the most important eigenmodes of the system.
Three generalized DOFs are included in the model, namely surge, pitch, and the first bending mode, as illustrated in Fig. 3.
Orthogonal eigenmodes can be found numerically from the solution of the eigenvalue problem for a finite element (FE) model of the
linearized system. However, as the shapes of the rigid body modes are known a priori, they are instead chosen as analytical functions
for simplicity. In addition, to have global motions that are consistent with the TD model, the pitch mode is chosen as a rotation about
the SWL. Using non-orthogonal modes introduces off-diagonal terms in the system matrices, however, these coupling effects are
readily taken into account in the formulation.
The actual shape of the bending mode, on the other hand, is not known, and is therefore found from the eigenvalue solution. A
third-order spline function is then fitted to the nodal displacements to have continuous expressions for the first and second order
derivatives. The FE model used to solve the eigenvalue problem is based on the linearized system in Fig. 2 and does not include the
blades, however, added mass on the hull is included using strip theory.

Fig. 2. Linear representation of the FWT system.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 3. DOFs considered in the analysis.

4.1. Load formulation

In order to predict the dynamic behaviour of the system, the inertia, damping, restoring and excitation forces on the structure
need to be determined. Due to the simple and relatively slender geometry of the hull, the hydrodynamic loads on the floater can be
simplified without significant loss of accuracy. The transverse added mass per unit length is approximated using the analytical
expression for a 2D circular cylinder with diameter D in infinite fluid [27]:

a11 (z ) = D 2 (z )/4 (1)

The linear wave excitation forces are taken from MacCamy-Fuchs theory [28]. The force per unit length, dFW , for a regular wave
with unit amplitude is given by

4 g cosh k (z + h)
dFW (z , ) = G ei ( t )
k cosh kh (2)

Fig. 4. Aerodynamic damping ratio for the first fore-aft mode of the land-based DTU 10 MW turbine as a function of wind speed.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

where

1 J 1 (ka)
G= , tan = .
(J1 (ka))2 + (Y1 (ka))2 Y 1 (ka) (3)

For the aerodynamic loads, the BEM equations usually have to be solved in the TD [29], and there is no straight-forward way to
establish linearized frequency-dependent wind forces. In the present study, the turbulent thrust force (FT ) and rotor moment (MT ) for
different wind speeds are extracted from TD simulations on a fixed rotor in SIMA. The resulting time series are transformed to the FD
using fast Fourier transform (FFT), and normalized with the incoming wind spectrum to create force transfer functions. In addition,
the so-called aerodynamic damping, which arises due to the change in thrust force as a result of the nacelle's velocity, needs to be
explicitly calculated, which can be done in various ways [29,30]. Here, the linearized damping is estimated following the procedure
described by Bachynski [9]. Time domain simulations with a range of constant wind speeds are run on a fixed rotor, and the damping
values are found as the change in thrust force for small variations in wind speed while the blade pitch and rotor speed are kept fixed:

dFT
Baero = .
dU (4)

The resulting aerodynamic damping values are shown as a function of mean wind speed in Fig. 4, presented as critical damping
ratio for the first fore-aft tower mode of the original land-based DTU 10 MW turbine. The linearized aerodynamic damping applies to
all generalized modes with nonzero deflection at the tower top, and the coefficient Baero is varied with the mean wind speed. Damping
caused by pure rotation of the rotor is not considered. It is assumed that any changes in aerodynamic forces that arise due to the
motions of the FWT are captured by the aerodynamic damping term, which means that variations in control system outputs caused by
platform motions are not considered in the model.
Rayleigh damping is used to model structural damping in the tower, with a stiffness-proportional coefficient β = 0.007 for both
designs. For a given β, the damping ratio is proportional to the natural frequency, and using a constant value will thus reward stiffer
designs, such as spar 2, without any physical reason. In an actual design process, the coefficient should therefore be continuously
updated to keep the damping ratio constant. However, as the scope of the current work is not to perform design calculations, but
rather to compare different modelling techniques, a constant coefficient is used for simplicity. The applied coefficient corresponds to
a damping ratio of 0.9% and 1.2% at the first bending natural frequency for spar 1 and spar 2, respectively.
Viscous damping, which is important for the low-frequency surge response, is added based on stochastic linearization of the
quadratic drag term in Morison's equation [31]:

1 1 8
Cd D (z ) x |x | Cd D (z ) (x ) ( z ) x = b visc (z ) x
2 2 (5)

with a drag coefficient Cd = 0.7. The standard deviation of the velocity, (x ) , is found using an iteration scheme. Viscous wave
excitation was found to be small compared to the wind excitation forces, and is therefore not included in the model. Wave radiation
damping is also neglected. The aerodynamic drag forces on the tower, which only are included in the extreme response conditions,
are taken into account in the linear model as a combination of a mean force, which is added directly, and a frequency-dependent
force, which is found using stochastic linearization. Here, only the excitation forces are considered, meaning that the tower drag
forces arising from the movement of the turbine are neglected.
The analytical solution for quasi-static horizontal tension in elastic catenary lines can be found in e.g. Faltinsen [32]. In order to
find the linear mooring stiffness used in the model, Kt , this equation is differentiated numerically. The stiffness is calculated for the
zero offset position of the hull in all environmental conditions, as the offset was found to have little effect on the results. However,
updated stiffness values based on the actual mean position of the platform can easily be included. Mooring line excitation, damping
and inertia are neglected in the model.
The hydrostatic restoring in pitch, which is applied as a rotational spring at the SWL, is found from
Kr = gVzB Mgz G + gIwp. (6)

4.2. Establishing the equations of motion

Using the vector containing the shape functions and the terms derived in the previous section, the generalized mass (including
added mass), damping and stiffness matrices can be established [26]:
ztop 0
M= m (z ) (z ) (z )dz + [m (z ) + a11 (z )] (z ) (z )dz
0 L
+ MRNA (ztop) (ztop) + IRNA , z (ztop) ,z (ztop) (7)

ztop 0
B= EI (z ) , zz (z ) , zz (z )dz + b visc (z ) (z ) (z )dz + Baero (ztop) (ztop)
z TB L (8)


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

ztop ztop
K= EI (z ) , zz (z ) , zz (z )dz + N (z ) , z (z ) ,z (z )dz
z TB z TB

+ Kt (zmoor ) (zmoor ) + Kr , z (0) ,z (0), (9)

where N (z ) is the axial force in the tower. The generalized wind and wave load vectors are found from
FU ( ) = FT ( ) (ztop) + MT ( ) , z (z top) (10)
0
F( )= dFW (z , ) (z )dz.
L (11)

4.3. Frequency domain response

The transfer functions relating modal response to wave and wind input, H X ( ) and HUX ( ) respectively, are defined as
H X ( ) = HFX ( ) F ( ), HUX ( ) = HFX ( ) FU ( ) (12)
where
HFX ( ) = [ 2M + i B + K] 1
(13)
is the frequency response function matrix. Transfer functions between tower base bending moment and wind/wave input (H M ( )
and HUM ( ) ) are found by considering equilibrium between external, inertial, damping and internal forces. The complete equations
may be found in Appendix A. As both the wind speed and wave elevation are considered to be stationary Gaussian processes within
the short-term duration, the linear response will also be a stationary Gaussian process. The responses to wind and wave input are
assumed to be independent; that is, there is no interaction between the responses at different frequencies. This is an assumption
inherent in the linearization, and may not be equally applicable for all FWTs.
The response spectrum for an arbitrary response parameter ξ is then found by superimposing the wind and wave responses:
S ( ) = |H ( )|2 S ( ) + |HU ( )|2 SU ( ), (14)
while the variance is given as

2
( ) = S ( )d .
0 (15)

5. Dynamic simulations

5.1. Environmental conditions

The long-term fatigue assessment uses the joint probability distribution of metocean parameters given in Johannessen et al. [33].
Probability density functions (PDFs) for the parameters are shown in Fig. 5. IEC 61400-3 [34] prescribes a minimum bin size of 2 m/s
for the wind speed, 0.5 m for wave heights, and 0.5 s for wave periods. Kvittem and Moan [35] performed a long-term fatigue analysis
of a semi-submersible FWT, where damage sensitivities with regards to simulation length, bin size and number of samples were
addressed. Based on their findings, the bin sizes for wave heights and wave periods in the present work are increased to 1 m and 1 s,
respectively, and each condition is simulated using six 1-h realizations after removal of transients.
Long-term fatigue analyses used in design of FWTs should cover all relevant conditions over the lifetime of the structure, including
operational, fault, idling, and survival conditions. The fatigue analysis presented in this study is limited to operational cases, which
means that the mean wind speed is varied between 4 and 25 m/s, and the turbine is assumed to be operating normally. A total of 546
environmental conditions (ECs) are thus considered in the analysis, which corresponds to all ECs within 4–25 m/s with probability of
occurrence greater than 10-4. It is worth noting that, although not employed in the current fatigue comparison, the FD model also is
applicable for wind speeds below cut-in or above cut-out, as long as only steady-state conditions are considered.
Three-dimensional turbulent wind time series are generated using TurbSim [36], assuming a Kaimal spectrum for IEC61400-1
class B turbines and a normal turbulence model. A power law profile with exponent 0.14 is used to model the vertical wind shear
[34]. The same wind files are used both directly in the TD simulations, and to establish the wind thrust and rotor moment transfer
functions used in the FD calculations, which are derived by taking the average values over the six seeds. The calculated transfer
functions for each individual wind time series, as well as the average curves, are shown in Fig. 6 for 11 m/s mean wind speed, to
illustrate the degree of variability between the realizations. Wave time series are generated using JONSWAP spectra with a peak-
edness parameter of 3.3. Co-directional waves and wind travelling in the positive x-direction are applied in all simulations, and no
current is considered in the analysis.
Four ECs along the 50-year contour surface are selected to compare the short-term extreme response, as described in Table 4. In
EC 1, the mean wind speed is close to the rated speed of the turbine, which gives the maximum thrust force on the rotor. EC 2
considers an operational condition near cut-out, while ECs 3 and 4 represent wind speeds above cut-out, where the turbine is parked


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 5. Marginal distribution of 1-h wind speed, and conditional distributions of wave height and peak period. Based on Johannessen et al. [33].

Fig. 6. Calculated transfer functions for aerodynamic excitation loads at 11 m/s mean wind speed. Six different wind realizations (grey) and average
values (black).

and the blades are feathered. As for the fatigue analysis, 1-h simulations (excluding transients) are used in the extreme response
comparison, but with 20 different random seeds for each condition. No fault conditions are considered, however, previous studies on
blade pitch faults, grid loss and shutdown have shown that such events primarily affect the loads on the blades and shaft, and that the
tower base fore-aft bending moment and global motions tend to be smaller than during fault-free conditions for spar-type FWTs
[37,38].

5.2. Fatigue damage assessment

The fore-aft axial stress at the outer radius of the tower is calculated from

N My
x = + r.
A I (16)

Stress variations due to fluctuations in the axial force are found to be negligible compared to the moment-induced stress

Table 4
Environmental conditions for extreme response calculation.
Condition 1 2 3 4

Mean wind speed at hub height, U (m/s) 11.0 25.0 36.0 50.0
Significant wave height, Hs (m) 6.9 10.3 12.9 15.1
Spectral peak period, Tp (s) 10.0 12.5 14.0 16.0
Turbulence intensity at hub height, I (−) 0.18 0.14 0.13 0.12


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

variations, and are thus not included in the fatigue calculations. In the TD, rainflow counting is used together with SN curves and the
Miner-Palmgren rule to estimate the fatigue damage, which is taken as the average value over the six realizations for each EC. In the
FD, two different methods are considered:

1. Inverse Fourier transform (IFT)

2. Dirlik method

In the first method, 1-h stress time series are generated by performing IFT on the response spectrum with random phase angles.
The fatigue damage is then calculated using rainflow counting and the Miner-Palmgren rule. The phase angles introduce randomness
to the damage estimates, and the method is therefore performed with six different random seeds, as for the TD model. The average
value is then used in the fatigue comparison.
The Dirlik method [39] uses the stress response spectrum and empirical factors to fit the PDF of the stress cycles to a combination
of an exponential and two Rayleigh distributions, and the accuracy of the method is therefore dependent on how well the rainflow-
count can be represented by these distributions. The expression for the fatigue damage during T seconds becomes [26]:
mk
pT m t
DDK = (2 )m G1 Qm (1 + m) + ( 2 )m (G2 Rm + G3) 1+
K 2 tref (17)

where

2(xm 2
2) 1 2 G1 + G12 1.25( 2 G3 G2 R) 2 xm G12
G1 = , G2 = , G3 = 1 G1 G2, Q = ,R=
1+ 2
2 1 R G1 1 2 G1 + G12 (18)

and
m2 m m2 1 m4
2 = , xm = 1 , p = .
m 0 m4 m0 m4 2 m2 (19)

Here, mn is the nth spectral moment, defined as:

mn = nS ( )d . (20)
0

The Dirlik method has previously been shown to give accurate results over a wide range of bandwidths for a stationary Gaussian
process [40], and the closed-form expression makes it well-suited for efficient design optimization. However, the accuracy of the
method deteriorates if the fatigue loads are dominated by a few frequency components [41].
The SN curve used for the tower base is the D curve in air taken from DNV-RP-C203 [42]. The curve is bilinear, i.e. the material
parameters, and thus also the slope of the curve, change at a certain stress range. For the rainflow procedure, this is easily taken into
account, however, the Dirlik method only allows for single-slope curves. Based on initial calculations, the Dirlik damage estimation is
performed with material parameters valid for stress ranges above 52.63 MPa (fewer than 107 cycles).

5.3. Short-term extreme response

Several methods have been used to estimate the short-term extreme response of FWTs, such as the Gumbel method and Weibull
tail method for global motions [15,43], and the Winterstein method for the tower base bending moment [14]. The present study uses
the average upcrossing rate (AUR) method, see e.g. Naess and Gaidai [44], which previously has been shown to perform well for a
bottom-fixed offshore wind turbine under combined wind and wave loading [45]. The AUR method is based on the assumption that
upcrossings of high levels are statistically independent, which means that the number of upcrossings during time T will be Poisson
distributed. The cumulative distribution function (CDF) of the extreme value Y may thus be written as
FY (y ) = exp{ + (y ) T }, (21)

where Y can represent an arbitrary response parameter, such as tower base bending moment, global motions of the platform, or
mooring line tension. + (y ) is the mean upcrossing rate of level y, which can be estimated empirically from j simulated time series of
length T0 :
j
1
ˆ+(y ) = ni+ (y; T0 )
jT0 i=1 (22)

where ni+ (y ;
T0) is the number of upcrossings in time series i. Assuming that the appropriate asymptotic extreme value distribution for
the considered response is the Gumbel distribution, the tail of the mean upcrossing rate may be written as
+ (y ) = q (y ) exp{ a (y b)c }, y y0 (23)

where q (y ) is slowly varying and can be approximated as a constant. The parameters a, b, c and q are then determined by fitting Eq.
(23) to the empirical data, as described in detail by Saha and Naess [46]. This is done by minimizing the mean square error function:


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

N
F (a , b , c , q ) = wi |ln ˆ+(y ) ln q + a (yi b) c |2
i=1 (24)
using the Levenberg-Marquardt least-squares optimization algorithm. wi is a weight factor which puts more emphasis on the more
reliable data points, and is found from the estimated 95% confidence interval.
The AUR method is also applicable for the linear model, where an analytical expression is available. For a stationary Gaussian
process with zero mean, the upcrossing rate can be written as [26]:

+ (y ) + (0) y2
= exp ,
2 2 (25)
where + (0) is the mean zero-upcrossing rate, which can be found from the zeroth and second order spectral moments:

+ (0) 1 m2
= .
2 m0 (26)
The characteristic largest extreme response y1h , i.e. the most probable maximum value in one hour, is used to compare the TD and
FD models. The value is found approximately from the following relation:
FY (y1h ) = e 1. (27)

6. Results

6.1. Natural periods

The natural periods of the linear system are estimated and compared to results from decay tests performed with the TD model in
still water, as shown in Table 5. The decay tests are performed by releasing the platform from an offset position in the considered
DOF, and the natural period is then found from the decaying motion. For the first bending mode, the hull is released together with the
tower top, such that the coupling between platform pitch and tower bending is included.
The FD model slightly overestimates the natural periods, which for the rigid modes is found to mainly be a consequence of the
simplified added mass formulation. The reason why the first bending natural frequency is better approximated for spar 1 than spar 2
can be understood by examining the eigenmodes. In Fig. 7, the bending mode from the linearized eigenvalue solution is compared to
the mode shape found from a decay test in TD, where a bandpass filter around the natural frequency is used to extract the shape
including one of the blades. Due to the increased stiffness of the tower in spar 2, the coupled blade-tower mode changes, and the
blades undergo a larger amount of bending. The simplification of replacing the RNA by a point mass and inertia is thus less accurate
than for spar 1, where the blades behave more like a rigid body in the first tower bending mode. The shape of the tower is, however,
seemingly not affected by the blade behaviour, and both designs show good agreement between the decay tests and the eigenvalue
solutions.

6.2. Fatigue damage

6.2.1. Linear and nonlinear estimates

Weighted 1-h fatigue damage is shown as a function of wind speed in Fig. 8. The linear predictions are seen to agree well with the
TD results for spar 1, especially at wind speeds from 11 m/s and higher. For spar 2, the linear model is seen to overpredict the fatigue
damage quite consistently by about 25–35%. The reasons for this overprediction are further explored in Section 6.2.2. The good
agreement for spar 1 can be explained by the fact that the tower base bending moment is close to Gaussian in most of the considered
environmental conditions, in particular at higher wind speeds, which are associated with larger waves. The response in these con-
ditions is dominated by wave forces and by resonant motions, which tend to be Gaussian even for non-Gaussian loads [47].
A notable difference between the linear and nonlinear results for spar 1 is seen at 9 m/s, where the 3P frequency is very close to
the natural frequency of the first bending mode, which leads to resonance and thus a significant increase in fatigue damage for the FD
model. In TD, the resonance has a rather limited effect on the damage estimates, possibly due to nonlinearities in the model.
With a few exceptions, the IFT and Dirlik results agree well for all conditions in both designs. Compared to the TD model, the error
in total fatigue damage is 10.6% (IFT) and 7.1% (Dirlik) for spar 1, and 30.4% (IFT) and 30.9% (Dirlik) for spar 2.

Table 5
Natural periods.
Mode Spar 1 Spar 2

TD FD TD FD

Surge (s) 120.55 122.22 116.80 121.67

Pitch (s) 35.48 36.39 38.29 39.81
1st bending (s) 2.41 2.42 1.79 1.85


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 7. Normalized bending mode shapes.

Scatter plots showing the calculated fatigue damage for each environmental condition in TD and FD are presented in Fig. 9, where
the Dirlik results are used for the linear model. The solid black line indicates perfect agreement, while conservative FD predictions are
found below the line.
The accuracy of the linear model is to a large extent insensitive to the amount of fatigue damage in the individual conditions, and
the data follow a fairly straight line, with most points located within ± 10% of the line of best fit for both designs. For spar 1, this line,
when neglecting the 9 m/s conditions, is very close to the diagonal. The 9 m/s results, although being highly overestimated by the
linear model, are also seen to be reasonably consistent, as the ratio between the linear and nonlinear estimates is more or less
constant with increasing damage.
For spar 2, all points are located below the 1:1 line, and as for spar 1, the conditions with the largest discrepancies between linear
and nonlinear predictions are in general found at lower wind speeds (5–9 m/s). These wind speeds are associated with low sea states,
and little excitation of the bending mode. The tower base bending moment is thus dominated by the low-frequency wind response,
which is less accurately predicted by the linear model than the wave- and bending-frequency response. The low-frequency response in
the linear model is also somewhat sensitive to how the thrust force spectrum is estimated, as reported by Kvittem and Moan [10]. It
should be noted that these conditions are associated with relatively small bending moments, and their contribution to the total
fatigue damage over the lifetime of the structure is thus not significant.
The limited scatter in the data for both designs suggest that it is possible to simulate a small number of environmental conditions
in the TD to verify and, if necessary, calibrate a linear model which can be used to perform the full long-term analysis. The ver-
ification cases should cover the entire wind speed range, as resonance effects may have a large impact on the accuracy of the model.
In addition, a 1 m/s interval for the wind speed bins may be needed to correctly capture the response around possible resonance wind
speeds [35].

6.2.2. Sensitivity to errors in natural frequency

From the tower base stress spectra in Fig. 10, the reason for the general overprediction in fatigue damage by the linear model for
spar 2 is obvious: the response in the region of the first bending mode is captured quite well by the linear model for spar 1, but is
significantly overestimated for spar 2. The wave-frequency response is accurately estimated for both designs, whereas the low-
frequency response is underpredicted at the pitch resonance frequency, mainly due to interaction between platform motion and the
wind turbine controller in the TD model. Due to the limited number of cycles, however, the low-frequency response has little
influence on the fatigue damage in the tower compared to the wave- and bending-frequency responses.
The overestimation of the response around the natural frequency of the first bending mode originates mainly from the small error
in estimated natural period for the bending mode discussed in Section 6.1, as the wind excitation loads vary significantly with
frequency in this region. In Fig. 11, the natural frequencies found from FD and TD are shown together with the transfer functions for
thrust force and rotor moment at 21 m/s mean wind speed. From the figure, one clearly sees that the loads are reduced when the


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 8. 1-h weighted fatigue damage arranged by wind speed.

natural frequency for spar 2 is increased. The effect is most prominent for the rotor moment, which is about 25% lower at the TD
natural frequency than at the natural frequency found from the linearized eigenvalue solution.
To quantify the importance of this error on the fatigue damage results, the natural period of the bending mode is shifted from
1.85 s to 1.79 s in the linear model by adding an artificial stiffness term, and the fatigue analysis is rerun. The results, presented in
Fig. 12, show a large improvement at most wind speeds, and the total fatigue damage error is reduced by approximately 50%. This
highlights the importance of having accurate estimates for the natural frequency and shape of the first bending mode, which is
dependent on how well the RNA can be approximated as a rigid body. In an optimization process, the accuracy of the model
simplification may vary as changes are made to the design, however, this can be taken into account by a straightforward calibration
of the natural frequency in the linear model.
The eigenmode in tower bending depends on wind speed, due to increased blade stiffness caused by centrifugal forces and
pitching of the blades, as well as changes in the mean platform pitch angle. This is illustrated in Fig. 13, where mode shapes from
decay tests are shown for calm air and 21 m/s uniform wind. As a consequence, the natural frequency of the bending mode is slightly
increased by 2–3% at higher wind speeds. In addition, the mode shape derivative at the tower top is reduced by up to 20%.
From Fig. 11, an increase in the natural frequency of the bending mode is seen to result in higher wind loads for spar 1. However,
this effect is counteracted by the decrease in the tower top derivative, which reduces the generalized wind load caused by the rotor
moment (see Eq. (10)). For spar 2, on the other hand, both effects reduce the wind excitation loads for the bending mode, and cause
the tuned linear model to somewhat overestimate the fatigue damage at higher wind speeds, as observed in Fig. 12.
Regarding the bending-frequency response, it should also be mentioned that, even though no validation has been performed, the


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 9. Time and frequency domain fatigue damage for individual ECs.

Fig. 10. Tower base stress response spectrum for U = 21 m/s, Hs = 7.5 m, Tp = 14.5 s.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 11. Wind excitation transfer functions around first bending natural frequency, U = 21 m/s.

Fig. 12. 1-h weighted fatigue damage arranged by wind speed, spar 2 with tuned natural frequency for first bending mode.

present study assumes that the behaviour of the physical system is described correctly by the TD model. Full-scale measurements
from the Hywind Demo FWT [48] have shown that the tower base bending moment response spectra contain energy around the first
tower bending mode and that the numerical models thus capture real effects, however, the actual level of the response is sensitive to
the modelling of the system, and may therefore not be correctly represented in the simulations.

6.3. Extreme value prediction

The short-term extreme response of the system in the four 50-year environmental conditions is calculated for the tower base
bending moment, as well as for the surge and pitch motions of the platform. The results are presented in the following subsections.

6.3.1. Tower base bending moment

Response statistics for the tower base bending moment in the 50-year conditions are listed in Table 6. For a Gaussian process, the
skewness is equal to zero, while the Pearson's kurtosis is equal to 3. The skewness and kurtosis values for the tower base bending
moment suggest that the responses in EC 2–4 are very close to Gaussian, which is expected, as the bending moments in conditions
above cut-out are completely dominated by wave loading, whereas the response in operational conditions with high wind speeds is
governed by a combination of wave forces and resonance. The response in EC 1 exhibits some non-Gaussian behaviour due to the
large aerodynamic thrust near rated wind speed, however, the degree of nonlinearity is small. In this context, it should also be
emphasized that the TD model only includes first order wave loads, with the exception of viscous forces arising from the drag term in
Morison's equation.
The standard deviations are accurately predicted by the linear model in all conditions, with relative errors between −2.9% and


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 13. Normalized bending mode shapes in calm air and for U = 21 m/s. Natural periods found from decay tests are 2.41 s (calm air) and 2.35 s
(21 m/s) for spar 1, and 1.79 s (calm air) and 1.75 s (21 m/s) for spar 2.

Table 6
Tower base bending moment statistics, 50-year environmental conditions.
Model Statistical quantitya Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain Mean (MNm) 247.2 123.0 50.3 105.2 305.7 152.9 63.6 132.8
Std.dev. (MNm) 89.1 101.0 106.5 118.8 103.8 106.2 115.5 129.7
Skewness −0.20 −0.03 −0.01 0.01 −0.30 −0.04 −0.01 0.01
Kurtosis 3.03 3.00 3.00 3.01 3.13 3.00 3.01 3.01

Frequency domain Mean (MNm) 260.8 127.8 52.3 101.3 329.4 161.4 66.5 128.9
Std.dev. (MNm) 89.4 100.8 106.3 115.1 104.8 110.6 118.4 128.5

a
Time domain results are averaged over 20 seeds.

4.1%. The mean moments are somewhat less accurate, with the largest errors of 5.5% (spar 1) and 7.8% (spar 2) occurring in EC 1,
near rated wind speed, where also the largest mean values are found.
Empirical upcrossing rates are plotted together with optimized curve fits and Gaussian upcrossing rates based on the FD results in
Figs. 14 and 15. The plots show the multiplication factor κ, which is related to the extreme response, y, through the relation

y=μ+ , (28)

where μ and σ are the mean and standard deviation of the process. The characteristic largest 1-h extreme for a given environmental
condition, which is used in the comparison, corresponds to an upcrossing rate of 2.78 × 10 4 .
The Gaussian upcrossing rates are seen to be conservative for EC 1, due to the slightly nonlinear behaviour which becomes more
prominent as the bending moments increase. In the parked conditions, on the other hand, the response is linear also at lower
upcrossing rates, and the empirical data are well-described by the linear model.
For spar 1, the FD solution is seen to follow the optimized curve fit closely in EC 2, however, larger discrepancies are observed for spar 2.
The fact that the shape of the fitted curve is captured quite well, suggests that the error is related to the overprediction of the bending
response discussed in the previous section. As the bending mode is located at a relatively high frequency, the response level has significant
influence on the second order spectral moment. Consequently, the mean zero-upcrossing rate (see Eq. (26)) is overestimated by 29%
compared to the TD model, which causes a right shift in the linear upcrossing rate curve as observed in Fig. 15b.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 14. Tower base bending moment upcrossing rates for spar 1.

The calculated 1-h extreme bending moments are listed in Table 7. The linear estimates give reasonably good approximations of
the extreme response, particularly for EC 2–4, as indicated by the upcrossing results.
Due to the high mean value caused by the thrust force, the largest extreme response is found in EC 1. However, this may not be the
case in general, and previous studies comparing short-term extremes in selected operational and survival conditions have shown that
the largest tower base bending moments for a spar FWT can occur in storm conditions with extreme wind speeds [49,50]. The critical
load cases may vary with floater design and turbine size, as it alters the ratio between wind and wave loads, and for long-term
extremes, the probability of the conditions must be taken into account. The most critical operational condition is also not necessarily
associated with maximum thrust force on the turbine. In a recent study by Sultania and Manuel [51], the largest 50-year tower base
bending moment for a 5 MW spar FWT, considering only operational cases, was found for a mean wind speed of 21.7 m/s. This
suggests that a wider range of environmental conditions should be considered in an actual design process.

6.3.2. Global motions

Tables 8 and 9 list extreme response statistics for surge and pitch, respectively. The global motions show similar behaviour to the
tower base bending moments, but as a larger part of the response is located in the low-frequency range, there is less agreement
between the TD and FD results for the standard deviations. The values are in general found to be underestimated by the linear model,
with errors up to about 20% for both surge and pitch.
As for the bending moments, the upcrossing rates for the global motions (Appendix B) are accurately described by the Gaussian
curves in EC 2–4, while the nonlinear behaviour in EC 1 results in conservative predictions by the linear model. For both spar designs,
some slight nonlinearities are also seen at higher wind speeds, which causes the linear model to underestimate κ at lower upcrossing
rates. The discrepancies are likely caused by the aerodynamic drag forces on the tower, which, although they mostly affect the mean
position of the FWT, also become more important for the dynamic response as the wind speed increases. The good agreement for the
surge motions in the parked conditions suggests that the viscous hydrodynamic forces on the hull contribute little to the motion
response.
The resulting characteristic largest 1-h extreme values are listed in Tables 10 and 11. As expected, the most severe global motions
are also found in EC 1. The pitch response, which shows the most pronounced non-Gaussian behaviour near rated wind speed, has the


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Fig. 15. Tower base bending moment upcrossing rates for spar 2.

Table 7
Characteristic largest extreme tower base bending moment y1h , 50-year environmental conditions.

Model Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain (MNm) 534.2 480.6 418.7 509.8 624.1 529.6 462.6 577.4
Frequency domain (MNm) 575.7 491.7 415.4 491.3 699.3 562.0 469.9 563.3

Relative error (%) 7.8 2.3 −0.8 −3.6 12.0 6.1 1.6 −2.4

least accurate linear predictions in this condition. Large discrepancies are also seen in the most extreme wave condition (EC 4),
mostly due to underestimation of the standard deviations in FD. For the surge motions, the extremes found from FD generally agree
well with the TD simulations. Most notable are the results in EC 1, especially for spar 1, where the relative error between FD and TD is
only −1%. However, this result is somewhat coincidental, as it is a consequence of the error in κ being almost exactly balanced out by
the errors in μ and σ. Nonetheless, the results suggest that the extreme surge motions may be predicted with roughly the same
accuracy as the tower base bending moments, despite larger discrepancies in the standard deviations, as the mean values account for
a larger share of the total extreme response.
The accuracy of the linear model depends on the importance of nonlinearities in the system, which will vary with location, floater
geometry, and different operational conditions. Therefore, the results presented here cannot be generalized to any FWT, and the FD
model should be continuously verified against state-of-the-art tools during a design process. The agreement with nonlinear analyses
also depends on how the system is represented in the TD model, and care should be taken to make sure that relevant nonlinearities
are included.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Table 8
Platform surge response statistics, 50-year environmental conditions.
Model Statistical quantitya Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain Mean (m) 27.3 13.8 8.0 15.1 23.7 12.0 6.6 13.7
Std.dev. (m) 6.9 2.4 2.9 4.6 6.2 2.2 2.7 3.9
Skewness −0.14 −0.02 0.05 −0.03 −0.13 −0.01 0.01 0.06
Kurtosis 2.49 2.90 3.03 3.06 2.57 2.95 3.02 2.97

Frequency domain Mean (m) 28.6 14.1 7.9 15.3 26.1 12.8 7.0 13.6
Std.dev. (m) 5.4 2.3 2.5 3.5 5.2 2.2 2.5 3.5

a
Time domain results are averaged over 20 seeds.

Table 9
Platform pitch response statistics, 50-year environmental conditions.
Model Statistical quantitya Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain Mean (deg) 6.7 3.4 1.7 3.4 9.2 4.6 2.2 4.4
Std.dev. (deg) 1.8 1.0 1.2 1.7 2.4 1.3 1.3 2.1
Skewness −0.62 0.03 0.04 0.10 −0.72 0.03 0.03 0.10
Kurtosis 3.37 2.90 3.02 2.99 3.54 2.92 2.99 3.02

Frequency domain Mean (deg) 7.2 3.5 1.7 3.2 10.0 4.9 2.2 4.3
Std.dev. (deg) 1.7 1.0 1.0 1.4 2.3 1.2 1.2 1.7

a
Time domain results are averaged over 20 seeds.

Table 10
Characteristic largest extreme surge response y1h , 50-year environmental conditions.

Model Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain (m) 44.2 21.4 17.8 29.4 38.6 19.0 15.7 26.4
Frequency domain (m) 43.8 21.4 16.1 26.6 40.7 20.0 15.1 24.9

Relative error (%) −1.0 0.0 −9.5 −9.4 5.4 5.4 −3.6 −5.7

Table 11
Characteristic largest extreme pitch response y1h , 50-year environmental conditions.

Model Spar 1 Spar 2

Condition Condition

1 2 3 4 1 2 3 4

Time domain (deg) 10.7 6.7 5.5 9.1 14.4 8.6 6.6 11.3
Frequency domain (deg) 12.3 6.6 5.0 7.7 16.7 8.7 6.0 9.8

Relative error (%) 14.8 −0.9 −9.4 −14.9 16.2 −4.9 −8.9 −13.2

7. Conclusions

A semi-analytical FD model has been developed, which allows for efficient preliminary design optimization of spar FWTs. The
model has been verified against a fully coupled nonlinear TD model for two different 10 MW spar designs, with regards to fatigue
damage in the tower base and short-term extreme response in both operational and survival conditions. The fatigue results show that
the FD model is accurate in most environmental conditions, especially for spar 1, where the long-term fatigue damage is


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

overestimated by 7%. For spar 2, larger discrepancies are observed, and the linear model overpredicts the damage by approximately
30%. This is found to primarily be caused by errors in the response around the first bending mode, which is highly sensitive to the
estimated natural frequency. The accuracy depends on how well the RNA can be approximated as a rigid body for the first bending
mode, however, tuning the frequency by adding an artificial stiffness term can be performed in the case where the approximation
leads to unacceptable errors.
For 9 m/s mean wind speed, the 3P frequency of the rotor coincides with the natural frequency of the first bending mode for spar
1, which leads to increased damage values and causes the linear model to significantly overpredict the response. The sensitivity with
regards to wind speed also suggests that a finer discretization of the wind bins should be applied in a design process, to accurately
capture the response around possible resonances. The accuracy of the linear model is found to be fairly independent of wave height
and wave period, also for the wind speed which induces 3P resonance with the first tower bending frequency. This indicates that a
small number of environmental conditions, which cover the full wind speed range, can be simulated with a TD model to verify and
possibly tune the linear model, which then can be used to carry out the full long-term analysis.
The extreme response, both with regards to tower base bending moment and rigid body motions in surge and pitch, is seen to be
nearly Gaussian near cut-out and in parked conditions, as it is dominated by wave forces. This results in good agreement for the linear
model in these conditions. The largest extremes values are found in EC 1, near rated wind speed, which indicates that a wider range of
environmental conditions may be important for the extreme response. The large aerodynamic forces in this condition also introduce
some non-Gaussian behaviour, which causes the linear model to overestimate the multiplication factors (κ). However, as the mean
values in this condition are relatively large, the inaccuracies in the total extreme response are reduced, and the total error in y1h is
within 16% for all considered responses in this condition. Regarding the standard deviations, good agreement between the linear and
nonlinear models is obtained for the tower base bending moments, while larger discrepancies are seen for surge and pitch. The reason
for this is that the global motions are more dependent on the low-frequency (wind-induced) response, which is less accurately
captured by the linear model than the wave- and bending-frequency response. The TD model in the present study only includes first
order wave forces, and could thus be improved by also considering higher order loads, which may be important for certain FWT
concepts. The model should also be validated through comparisons with physical measurements.
The developed FD model is shown to be well-suited for preliminary design. It gives reasonable agreement with a fully coupled
nonlinear TD model, both with regards to fatigue and extreme response, and the semi-analytical approach provides an efficient way
to explore the design space. In addition, the closed-form expressions offers the possibility of using analytic gradients in a computer-
aided optimization process. However, the efficiency of the model is partly due to the simple geometry of the platform, and the
formulations are thus not valid for more complex platform designs, where more comprehensive numerical calculations may be
necessary. Also, the approach is only applicable for steady-state conditions, and transient events, which also must be considered,
cannot be analysed. Nonetheless, it can serve as a useful complement to more sophisticated models in the design process of spar-type
FWTs, and significantly reduce the needed computational effort.

A Transfer functions

A.1 Wave elevation to tower base bending moment

The transfer function from wave elevation to tower base bending moment is given in Eq. (A.1). The equation considers bending
moments due to inertial forces in the tower and RNA, aerodynamic damping forces, and gravitational forces.
ztop
H M( ) = m (z ) H X¨ ( ) (z )(z zTB)dz MRNA H X¨ ( ) (ztop)(ztop zTB)
z TB
IRNA H X¨ ( ) , z (ztop) Baero H X ( ) (ztop)(ztop zTB)
ztop
+ m (z ) g H X ( )[ (z ) (zTB)]dz + MRNA g H X ( )[ (ztop) (zTB)]
z TB (A.1)

A.2 Wind speed to tower base bending moment

The transfer function from wind speed to tower base bending moment is given in Eq. (A.2). The equation considers bending
moments due to inertial forces in the tower and RNA, aerodynamic damping forces, gravitational forces, and wind excitation forces.
ztop
HUM ( ) = m (z ) HUX¨ ( ) (z )(z zTB)dz MRNA HUX¨ ( ) (ztop)(ztop zTB)
z TB
IRNA HUX¨ ( ) , z (z top ) Baero H X ( ) (ztop)(ztop zTB)
ztop
+ m (z ) g HUX ( )[ (z ) (zTB)]dz + MRNA g HUX ( )[ (ztop) (zTB)]
z TB
+ FT ( )(ztop zTB) + MT ( ) (A.2)


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

B Upcrossing rates for global motions

B.1 Surge

Figure B.1. Platform surge upcrossing rates for spar 1.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Figure B.2. Platform surge upcrossing rates for spar 2.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

B.2 Pitch

Figure B.3. Platform pitch upcrossing rates for spar 1.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

Figure B.4. Platform pitch upcrossing rates for spar 2.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

References

[1] Strach-Sonsalla M, Muskulus M. Dynamics and design of floating wind turbines. Proceedings of the twenty-sixth (2016) international ocean and polar en-
gineering conference (ISOPE2016), rhodes, Greece. 2016.
[2] DNV. Environmental Conditions and Environmental Loads. 2014. Tech. rep. DNV-RP-C205.
[3] DNV. Design of Floating Wind Turbine Structures. 2013. Tech. rep. OS-J103.
[4] Wayman EN, et al. Coupled dynamic modeling of floating wind turbine systems. Offshore technology conference. 2006.
[5] Sclavounos P, Tracy C, Lee S. Floating offshore wind turbines: responses in a seastate pareto optimal designs and economic assessment”. proceedings of the ASME
27th international conference on offshore mechanics and arctic engineering (OMAE2008), estoril, Portugal. 2008.
[6] Philippe M, Babarit A, Ferrant P. Comparison of time and frequency domain simulations of an offshore floating wind turbine. Proceedings of the ASME 2011 30th
international conference on ocean, offshore and arctic engineering (OMAE2011), rotterdam, The Netherlands. 2011.
[7] Wang K, et al. Frequency domain approach for the coupled analysis of floating wind turbine system. Ships Offshore Struct 2017;12(6):767–74.
[8] Bachynski EE, Moan T. Linear and nonlinear analysis of tension leg platform wind turbines. Proceedings of the twenty-second (2012) international offshore and
polar engineering conference (ISOPE2012), rhodes, Greece. 2012.
[9] Bachynski EE. Design and dynamic analysis of tension leg platform wind turbines PhD thesis Norwegian University of Science and Technology; 2014
[10] Kvittem MI, Moan T. Frequency versus time domain fatigue analysis of a semisubmersible wind turbine tower. J Offshore Mech Arctic Eng 2014;137:1.
[11] Kluger JM, Sapsis TP, Slocum AH. A reduced-order, statistical linearization approach for estimating nonlinear floating wind turbine response statistics”.
Proceedings of the twenty- sixth (2016) international ocean and polar engineering conference (ISOPE2016), rhodes, Greece. 2016.
[12] Jonkman J. Definition of the Floating System for Phase IV of OC3. National Renewable Energy Laboratory; 2010. Tech. rep. NREL/TP-500-47535.
[13] Matha D. Model development and loads analysis of an offshore wind turbine on a tension leg platform, with a comparison to other floating turbine concepts MA
thesis University of Colorado - Boulder; 2009
[14] Karimirad M, Moan T. Extreme dynamic structural response analysis of catenary moored spar wind turbine in harsh environmental conditions. J Offshore Mech
Arctic Eng 2011;133.
[15] Aggarwal N, Manikandan R, Saha N. Nonlinear short term extreme response of spar type floating offshore wind turbines”. Ocean Eng 2017;130:199–209.
[16] Bak C, et al. Description of the DTU 10 MW Reference Wind Turbine. 2013. Tech. rep. DTU Wind Energy Report-I-0092. DTU Wind Energy.


J.M. Hegseth, E.E. Bachynski 0DULQH6WUXFWXUHV  ²

[17] MARINTEK. RIFLEX user guide. 2016.

[18] MARINTEK. SIMO user guide. 2016.
[19] Wamit Inc. WAMIT user manual. 2014.
[20] Roald L, et al. The effect of second-order hydrodynamics on floating offshore wind turbines. Energy Procedia 2013;35:253–64.
[21] Duarte T, Sarmento AJ, Jonkman J. Effects of second-order hydrodynamic forces on floating offshore wind turbines”. 32nd ASME wind energy symposium,
national harbor, Maryland. 2014.
[22] Ormberg H, Bachynski EE. Global analysis of floating wind turbines: code development, model sensitivity and benchmark study. Proceedings of the twenty-
second (2012) international offshore and polar engineering conference (ISQPE2012), rhodes, Greece. 2012.
[23] Robertson A, et al. Offshore code comparison collaboration continuation within IEA wind task 30: phase II results regarding a floating semisubmersible wind
system. Proceedings of the ASME 2014 33rd international conference on ocean, offshore and arctic engineering (OMAE2014), San Francisco, California, USA.
2014.
[24] Nielsen FG, Hanson TD, Skaare B. Integrated dynamic analysis of floating offshore wind turbines. Proceedings of OMAE 2006 25th international conference on
offshore mechanics and arctic engineering, hamburg, Germany. 2006.
[25] Larsen TJ, Hanson TD. A method to avoid negative damped low frequent tower vibrations for a floating, pitch controlled wind turbine. J Phys Conf 2007;75.
012073.
[26] Naess A, Moan T. Stochastic dynamics of marine structures. Cambridge University Press; 2013.
[27] Newman JN. Marine hydrodynamics. The MIT Press; 1977.
[28] MacCamy RC, Fuchs RA. Wave Forces on Piles: a diffraction Theory Tech rep. Technical memorandum 69 Beach erosion board; Corps of Engineers; 1954.
[29] van der Tempel J. Design of support structures for offshore wind turbines” PhD thesis TU Delft; 2006
[30] Salzmann DJC, van der Tempel J. Aerodynamic damping in the design of support structures for offshore wind turbines”. Proceedings of the offshore wind energy
conference, copenhagen, Denmark. 2005.
[31] Borgman LE. Ocean wave simulation for engineering design. J Waterw Harb Div 1969;95(4):557–86.
[32] Faltinsen OM. Sea loads on ships and offshore structures. Cambridge University Press; 1990.
[33] Johannessen K, Meling TS, Haver S. Joint distribution for wind and waves in the Northern North sea. Int J Offshore Polar Eng 2002;12(1):1–8.
[34] IEC. Design requirements for Offshore Wind Turbines. 2009. Tech. rep. IEC 61400-3.
[35] Kvittem MI, Moan T. Time domain analysis procedures for fatigue assessment of a semi- submersible wind turbine. Mar Struct 2015;40:38–59.
[36] Jonkman BJ, Kilcher L. TurbSim user's guide: version 1.50 Tech. rep. NREL/TP-500-46198 National Renewable Energy Laboratory; 2009.
[37] Bachynski EE, et al. Dynamic analysis of floating wind turbines during pitch actuator fault, grid loss, and shutdown. Energy Procedia 2013;35:210–22.
[38] Jiang Z, Karimirad M, Moan T. Dynamic response analysis of wind turbines under blade pitch system fault, grid loss, and shutdown events. Wind Energy
2014;17:1385–409.
[39] Dirlik T. Application of computers in fatigue analysis” PhD thesis University of Warwick; 1985
[40] Gao Z, Moan T. Frequency-domain fatigue analysis of wide-band stationary Gaussian processes using a trimodal spectral formulation. Int J Fatig 2008;30:10–1.
1944–1955.
[41] Ragan P, Manuel L. Comparing estimates of wind turbine fatigue loads using time-domain and spectral methods”. Wind Eng 2007;31(2):83–99.
[42] DNV. Fatigue Design of Offshore Steel Structures. 2010. Tech. rep. DNV-RP-C203.
[43] Aggarwal N, Manikandan R, Saha N. Predicting short term extreme response of spar offshore floating wind turbine”. Procedia Eng 2015;116(1):47–55.
[44] Naess A, Gaidai O. Monte Carlo methods for estimating the extreme response of dynamical systems. J Eng Mech 2008;134(8):628–36.
[45] Saha N, et al. Short-term extreme response analysis of a jacket supporting an offshore wind turbine. Wind Energy 2014;17:87–104.
[46] Saha N, Naess A. Monte Carlo-based method for predicting extreme value statistics of uncertain structures. J Eng Mech 2010;136(12):1491–501.
[47] Kvittem MI, et al. Short-term fatigue analysis of semi-submersible wind turbine tower. Proceedings of the ASME 2011 30th international conference on ocean,
offshore and arctic engineering (OMAE2011), rotterdam, The Netherlands. 2011.
[48] Skaare B, et al. Analysis of measurements and simulations from the Hywind Demo floating wind turbine. Wind Energy 2015;18:1105–22.
[49] Karimirad M, Moan T. Wave- and wind-induced dynamic response of a spar-type offshore wind turbine. J Waterw Port, Coast Ocean Eng 2012;138(1):920.
[50] Muliawan MJ, et al. Extreme responses of a combined spar-type floating wind turbine and floating wave energy converter (STC) system with survival modes.
Ocean Eng 2013;65:7182.
[51] Sultania A, Manuel L. Reliability analysis for a spar-supported floating offshore wind turbine. Wind Eng 2018;42(1):51–65.


128 Appended Papers
Paper 2

Frequency-dependent aerodynamic damping and in-

ertia in linearized dynamic analysis of floating wind
turbines
Carlos Eduardo S. Souza, John Marius Hegseth and Erin E. Bachynski
Journal of Physics: Conference Series, 2020

129
130 Appended Papers
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

Frequency-Dependent Aerodynamic Damping and

Inertia in Linearized Dynamic Analysis of Floating
Wind Turbines
Carlos Eduardo S Souza, John Marius Hegseth and Erin E Bachynski
Department of Marine Technology, Norwegian University of Science and Technology, 7491
Trondheim, Norway
E-mail: [email protected]

Abstract. In frequency-domain (FD) models of floating wind turbines (FWT), it is common to

regard the interaction between nacelle motions and thrust by means of a constant aerodynamic
damping coefficient. This approach is effective at higher motions frequencies, but does not
consider interactions between nacelle motions and the blade pitch control system. As a result,
the motions and loads at frequencies closer to the controller bandwidth may be underpredicted.
A remedy for this problem is to include the linearized thrust expression in the FD model,
such that the dynamic effects related to control are considered. In this paper, these dynamic
effects are related to frequency-dependent damping and inertia terms. Expressions for damping
and inertia coefficients are obtained with two different methods, and then included in the FD
model. The resulting responses are compared to those obtained with the constant damping
coefficient method, and also with coupled time-domain simulations of a semi-submersible 10
MW FWT. The better performance of the FD model with frequency-dependent inertia and
damping coefficients encourages the adoption of the linearized thrust approach for representing
the interaction between nacelle motions, thrust and control system.

1. Introduction
Frequency-domain (FD) methods can be helpful in the design of floating wind turbines (FWTs),
providing responses for a large number of loading conditions with relatively low computational
time. Previous work has indicated that the response of FWTs in moderate environmental
conditions may be estimated reasonably well using FD models [1, 2, 3]. However, their reliability
depends on an accurate prediction of loads and interactions with linearized models.
An especially interesting interaction takes place between nacelle motions and the rotor thrust.
Fluctuations in the nacelle’s horizontal velocity provoke changes in the flow through the rotor,
leading to oscillations in the rotor speed and thrust. An aerodynamic damping effect results from
this interaction, and is normally included in FD models by means of a constant aerodynamic
damping coefficient. This coefficient may be obtained from the thrust derivative w.r.t. the
relative wind speed [1, 3, 4], or by means of decay simulations of the FWT under different
incident wind speeds [2].
This method is convenient to implement due to its relative simplicity, and is normally adopted
in combination with the frequency-dependent thrust obtained for a fixed wind turbine, installed
on the top of the tower. A disadvantage of this approach is that the interactions of the nacelle
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

motions with the blade pitch control system are not considered, an effect that is known to reduce
the damping or even render it negative [5, 6, 7], besides introducing an inertia effect [8]. As a
consequence, the damping effect can be significantly overpredicted at frequencies close to the
controller bandwidth, if the constant coefficient is used.
Control system effects can be included in an FD model by means of linearization of the
thrust and torque. The rotor speed is then included as an additional degree of freedom, and the
blade-pitch controller can be written in terms of the system states [9].
The objective of the present work is to relate the linearized thrust equations to the above-
mentioned inertia and damping effects, providing a visualization of how those effects vary with
wind speed and frequency for a given control system.
The aerodynamic inertia and damping coefficients are calculated both from the linearized
thrust equations and from simulations where the nacelle is forced to oscillate with different
frequencies, under uniform wind. The obtained coefficients are then included in an FD model
of a semi-submersible 10 MW FWT. The responses predicted with the different methods for
obtaining the coefficients are compared to coupled time-domain simulations, where the thrust
is calculated with the blade element momentum (BEM) theory.
Section 2 presents the frequency domain model for the FWT; Section 3 introduces the thrust
linearization procedure and provides formulations for the frequency-dependent aerodynamic
inertia and damping coefficients; the time-domain simulations are explained in Section 4, and
the results are presented in Section 5; final discussions and conclusions are provided in Section 6.

2. Frequency-domain analysis of FWT

The frequency domain model used in the present work consists of three degrees-of-freedom
(DOFs), namely surge, heave, and pitch. The potential-theory hydrodynamic model is the same
as the one adopted in the time-domain simulations, presented in Section 4. Viscous excitation
is neglected, but viscous damping on the platform is added using stochastic linearization of the
drag term in Morison’s equation, where the standard deviation of the velocity along the length
of the columns and pontoons is found using an iteration scheme.
The stiffness matrix includes both hydrostatic restoring and linearized mooring forces. The
mooring stiffness matrix is a function of the mean offset, based on static equilibrium between
mooring forces and mean thrust.
Transfer functions from wind speed to thrust force, FU T (ω), are obtained for each mean wind
speed as the squared root of the ratio between the spectrum of thrust time series, obtained from
simulations with turbulent wind on a fixed turbine, and the incoming wind spectrum. The wind
force vector is then found as

FU (ω) = [FU T (ω) 0 zhub FU T (ω)] , (1)

where zhub is the hub height above still water level (SWL). Transfer functions from wave and
wind input to platform response, HζX (ω) and HU X (ω) respectively, are expressed as

HζX (ω) = HF X (ω)Fζ (ω), HU X (ω) = HF X (ω)FU (ω) , (2)

where  −1
HF X (ω) = −ω 2 (M + A(ω)) + iωB(ω) + K (3)
is the frequency response function matrix. After solving for the platform motions, the tower
base bending moment response is found by considering dynamic equilibrium.
Using the wind and wave spectra, the response spectrum for an arbitrary response parameter
ξ is then expressed as

Sξ (ω) = |Hζξ (ω)|2 Sζ (ω) + |HU ξ (ω)|2 SU (ω) . (4)

2
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

3. Aerodynamic damping and inertia effects

The FD model described in Section 2 neglects the variations in the thrust caused by the FWT
motions. The interactions between thrust and FWT motions are often represented by a constant
damping coefficient, which is added to the matrix B in equation (3). This approach neglects
the controller action. When the controller is included, its effect results in frequency-dependent
damping and inertia coefficients, as shown further below.

3.1. Thrust linearization without controller

A common way to calculate the aerodynamic damping in FD models [1, 4, 3] is to linearize the
thrust with respect to the relative wind speed at hub height, v, at different mean incoming wind
speeds, Uw , considering no change in the blade pitch angle or rotor speed. Using a first order
Taylor series expansion, the thrust is then expressed as

∂T
T = T0 + Δv . (5)
∂v
Let the nacelle dynamics be represented by an 1-DOF, 1st -order system:

mẍ + cẋ + kx = T . (6)

Uniform, non-turbulent wind is assumed for now. Recalling that v = Uw − ẋ, it is noted that
Δv = v − Uw = −ẋ. Replacing (5) in (6), the damping effect becomes clear:
 
∂T
mẍ + c + ẋ + kx = T0 , (7)
∂v

and the coefficient, baer , can be expressed as

∂T
baer = . (8)
∂v
The damping effect estimated with this method considers the change in the steady-state
thrust for a small perturbation in the uniform wind speed seen by the rotor, while neglecting
the effect of the control system and rotor dynamics. Different coefficients are obtained for
each incident wind velocity, but they are constant with respect to the nacelle frequency of
oscillation. This method is relatively simple to use and normally provides satisfactory results
when the turbine oscillates in a frequency range distant from the controller bandwidth. For
lower frequencies of oscillation, interaction with the control system takes place and the damping
coefficient dependence on frequency becomes more important.

3.2. Thrust linearization including controller

A more comprehensive method, which captures effects from the wind turbine controller, may be
used by including the rotor speed DOF and control system in the FD model, and also linearizing
the thrust with respect to rotor speed, Ω, and blade pitch angle, β:

T = T0 + Tv Δv + TΩ ΔΩ + Tβ Δβ , (9)
∂A
where the index indicates partial derivative w.r.t. the indicated variable, i.e., AB = ∂B . In order
to find the damping coefficient, Equation (9) must be written in terms of ẋ only. First, the
aerodynamic torque, Q, is also given in its linearized version:

Q = Q0 + Qv Δv + QΩ ΔΩ + Qβ Δβ , (10)

3
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

In addition, the rotor dynamics are given by

Id Ω̇ = Q − Ng E , (11)

where Id is the drivetrain inertia, Ng is the gear ratio and E is the electrical (generator) torque,
here assumed as constant above rated wind speed. In equilibrium, Q0 = Ng E, so replacing (10)
in (11):
Id Ω̇ = Qv Δv + QΩ ΔΩ + Qβ Δβ . (12)
It is assumed that a PI controller commands the blade pitch angle, i.e.,

Δβ = Kp ΔΩ + Ki ΔΩdt , (13)

where Kp and Ki are the proportional and integral controller gains. Equations (9) and (12) are
now rewritten replacing Δβ as given in (13), and noting that ΔΩ = φ̇, where φ is the rotor
azimuth angle:
T = T0 − Tv ẋ + (TΩ + Kp Tβ ) φ̇ + Ki Tβ φ , (14)
Id φ̈ − (QΩ + Kp Qβ ) φ̇ − Ki Qβ φ = −Qv ẋ . (15)
Assuming harmonic oscillation, Equation (15) can be written in the frequency domain, and the
following transfer function between x(ω) and φ(ω) is obtained:

iωQv
φ(ω) = x(ω) = C(ω)x(ω) . (16)
Id ω 2 + (QΩ + Kp Qβ ) iω + Ki Qβ

The nacelle dynamics (Equation 6) are now written in terms of T as given in (14), and also
expressed in the frequency domain, with φ(ω) as given in (16):
⎧ ⎫
⎪
⎨ ⎪
⎬
−mω 2 + iωc + k + iωTv − [(TΩ + Kp Tβ ) iω + Ki Tβ ] C(ω) x(ω) = T0 . (17)
⎪
⎩ ⎪
⎭
−ω 2 aaer (ω)+iωbaer (ω)

Based on the assumption that T responds harmonically to harmonic oscillations of the nacelle,
the thrust can be written as a combination of terms proportional to the nacelle acceleration and
velocity [8]. This assumption is further discussed in Section 3.3, but an immediate consequence
is the definition of frequency-dependent, aerodynamic inertia and damping coefficients:
1
aaer (ω) = − Re {iωTv − [(TΩ + Kp Tβ ) iω + Ki Tβ ] C(ω)} (18)
ω2
1
baer (ω) = Im {iωTv − [(TΩ + Kp Tβ ) iω + Ki Tβ ] C(ω)} (19)
ω

3.3. Forced oscillations

The thrust and torque derivatives in Section 3 can be calculated analytically, based on the BEM
equations, or numerically, using e.g. the central differences method. Both techniques assume
that the thrust and torque respond immediately to changes in v, Ω and β, while in reality
the time constants associated with rotor dynamics, aerodynamics, and control system, can be
relevant when the nacelle moves at frequencies close to the controller bandwidth.
An alternative method to find the aaer (ω) and baer (ω) coefficients is through simulations
of forced nacelle oscillations, with aerodynamics calculated nonlinearly and under influence of

4
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

0.8 1080

0.6 1060

1040
0.4

Nacelle velocity (m/s)

1020
0.2

Thrust (kN)
1000
0
980
-0.2
960
-0.4
940

-0.6 920

-0.8 900
100 150 200 250 300 350 400 450 500
Time (s)

Figure 1: Thrust response to oscillations of the DTU 10 MW wind turbine, under constant
incident wind speed of 13.0 m/s and period of oscillation of 100.0 s.

the control system. When the turbine oscillates harmonically, the thrust also oscillates nearly
harmonically, but with a phase relative to the nacelle velocity. This can be seen in Figure 1,
where the DTU 10 MW wind turbine is forced to oscillate with period 100.0 s and under constant
incident wind speed of 13.0 m/s. The thrust can then be assumed to be composed of a mean
plus an oscillating component:

T = T0 + Tosc (Uw , ω) . (20)

Writing the nacelle velocity ẋ as

ẋ = v0 cos(ωt) , (21)
the oscillating part of the thrust can then be assumed to be given by [8]

Tosc = f0 v0 cos(ωt + α)
= f0 [v0 cos(ωt) cos(α) − v0 sin(ωt) sin(α)]
(22)
f0 sin(α)
= f0 cos(α)ẋ + ẍ .
ω
The amplifying factor f0 and the phase α between nacelle velocity and thrust depend on
the dynamic effects mentioned above, and are not straightforward to determine analytically.
However, for a given turbine and control system they vary with the incident wind speed and
frequency of oscillation, while motion amplitude does not seem to have a significant influence.
They can therefore be obtained from forced oscillations covering the ranges of interest for both
parameters. Fast Fourier transforms (FFTs) of the nacelle velocity and thrust are calculated,
and f0 and α can be directly obtained from the ratio between amplitudes or phases of both FFTs
at the period of interest. Formulations for the aerodynamic inertia and damping coefficients are
then directly determined from equation (23):

f0 sin(α)
aaer (Uw , ω) = − , baer (Uw , ω) = −f0 cos(α) , (23)
ω

3.4. Coefficients calculation and comparison

The aerodynamic inertia and damping coefficients were obtained both from the linearized thrust
and the forced oscillations method. For the former case, the thrust and torque derivatives were

5
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

obtained from the linearized BEM equations. For the latter, forced oscillations of the DTU
10 MW wind turbine were performed with a simulator which obtains aerodynamic loads from
AeroDyn [10], updates the blade pitch angle based on the same controller strategy presented
in Section 4, and updates the rotor speed based on the rotor dynamics as in Equation (11).
Oscillation periods varied from 20.0 s to 130.0 s, and uniform wind with the three velocities in
Table 3 was considered. The BEM formulation with dynamic stall is adopted for the AeroDyn
calculations.
The coefficients obtained with both approaches are shown in Figures 2 and 3. In spite of the
oscillatory character of the curves obtained with the oscillation method, the agreement between
both methods looks satisfactory, especially for higher wind velocities. The inertia effect is of
the order of 1% of the mass of typical FWTs, resulting in negligible consequences for the surge
dynamics. The contribution to the moment of inertia in pitch is however considerable, due to
the nacelle height. Important changes in the pitch natural period can then result, as shown
in [8]. The constant damping coefficients as obtained in equation (5) are also plotted in Fig.
3, showing that the discrepancy w.r.t. the frequency-dependent coefficients is larger for lower
frequencies and higher wind velocities.

3.5. The aerodynamic inertia and damping effects on the FWT dynamics
It was already shown that the fluctuations in the aerodynamic thrust resulting from the nacelle
motions may be treated as frequency-dependent inertia and damping effects. The derivation
assumed non-turbulent wind, under the hypothesis that the fluctuations in the apparent wind
flow in the rotor due to turbulence can be decoupled from those caused by the nacelle motions.
Under the same assumption, the fluctuations in the thrust due to nacelle oscillations can now
be added to the FD model presented in Section 2. This is done by noting that, from Equations
(22) and (23), the oscillating component of the thrust can be written as

Tosc = −aaer ẍ − baer ẋ . (24)

The following aerodynamic and damping matrices, Aaer and Baer , are then defined:
⎡ ⎤ ⎡ ⎤
aaer (ω) 0 aaer (ω)zhub baer (ω) 0 baer (ω)zhub
Aaer (ω) = ⎣ 0 0 0 ⎦ , Baer (ω) = ⎣ 0 0 0 ⎦.
aaer (ω)zhub 2
0 aaer (ω)zhub 2
baer (ω)zhub 0 baer (ω)zhub
(25)
Aaer and Baer are now summed to A and B in equation (3), which includes the effect of nacelle
motion in the FD analysis.

4. Simulations
The platform considered is the OO-Star 10 MW [11], a semi-submersible concept designed for
the LIFES50+ project. The main properties are summarized in Table 1. The potential-theory
hydrodynamic loads are generated with WADAM, and viscosity is added in form of the Morison
drag term. The platform is installed at a water depth of 130.0 m, and the mooring system
consists on a simplified 3-catenary line arrangement, which differs from the original of taut lines
with clump weights.
The DTU 10 MW turbine (Table 2) is installed at the top of the tower, at a height of 118.4 m.
The blade-pitch controller corresponds to the PI formulation of Equation (13), i.e., the power
error contribution of the original controller [12] is not included. The controller gains are tuned
such that its bandwidth is below the pitch natural period, and a gain scheduling strategy corrects
the gains according to the current blade pitch angle.
The time-domain simulations were performed with SIMA, a workbench which allows for
coupled analyses between floating bodies and slender elements [14, 15]. The platform, nacelle

6
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

106 105
3 8
Calculated Calculated
Forced Forced
2.5 7

2 Tn,5 6 Tn,5
aaer (kg)

aaer (kg)
1.5 5

1 4

0.5 3

0 2
0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Frequency (Hz) Frequency (Hz)
(a) Uw = 13.0 m/s. (b) Uw = 19.0 m/s.

105
4.5
Calculated
Forced
4

3.5 Tn,5
aaer (kg)

2.5

1.5
0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Frequency (Hz)
(c) Uw = 25.0 m/s.

Figure 2: Aerodynamic inertia coefficients, calculated with the linearized approach or obtained
from the forced oscillations method.

Draft, D (m) 22.0

Total mass, mtot (kg) 2.36 × 107
Center of gravity, V CG (m) -7.9
Surge nat. per., T1 (s) 181.8
Pitch nat. per., T5 (s) 31.3

Table 1: Main properties of the OO-Star 10 MW FWT [11].

and hub are modeled as rigid bodies, while the blades, tower and mooring lines are modeled as
flexible structures, with finite elements. The aerodynamics are based on the BEM theory, with
dynamic stall and dynamic wake effects based on Øye’s models [16]. Hub and tip losses are
modeled with the Prandtl factor, and Glauerts correction is applied for high induction factors.
The environmental conditions are listed in Table 3. Turbulent wind was generated with TurbSim
[17], using the IEC Kaimal spectrum with turbulence characteristic B (moderate). A JONSWAP

7
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

104 104
16 16

14
14
12
12 Tn,5
10
10
baer (N.s/m)

baer (N.s/m)
8 Tn,5

6 8

4
6
2
4
0
Calculated Calculated
Forced 2 Forced
-2
Constant Constant
-4 0
0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Frequency (Hz) Frequency (Hz)
(a) Uw = 13.0 m/s. (b) Uw = 19.0 m/s.

104
16

14

12 Tn,5

10
baer (N.s/m)

4
Calculated
2 Forced
Constant
0
0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Frequency (Hz)
(c) Uw = 25.0 m/s.

Figure 3: Aerodynamic damping coefficients, calculated with the linearized approach (both
constant and frequency-dependent) or obtained from the forced oscillations method.

Rotor radius, R (m) 89.2

Rated rotor speed, Ω0 (rad/s) 1.0
Drivetrain inertia, Id (kg.m2 ) 1.6 × 108
Gear ratio, Ng (-) 50
Rotor-nacelle assemble mass, mRN A (kg) 6.7 × 105
Proportional controller gain, Kp (s) 0.1794
Integral controller gain, Ki (-) 0.0165

Table 2: Main properties of the DTU 10MW wind turbine [13]. The controller gains refer to
β = 0◦ .

wave spectrum with a γ factor of 3.3 was used to describe the sea state. The simulations lasted
for one hour, which showed to provide a sufficient number of LF oscillations for the analysis.
Only one realization of each condition was considered.

8
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

Uw (m/s) Hs (m) Tp (s)

C1 13.0 2.7 10.3
C2 19.0 4.0 11.1
C3 25.0 5.8 12.1

Table 3: Environmental conditions.

5. Results
Figure 4 shows the frequency-domain pitch response for the three conditions in Table 3, obtained
with time-domain simulations in SIMA; and with the FD model from Section 2, using the
different approaches for including the aerodynamic damping and inertia effects, as presented in
Section 3. When the constant damping coefficient from Equation (8) is adopted, the method
is referred to as FD constant; with the frequency-dependent inertia and damping coefficients
as given by Equations (18) and (19), the approach is named FD calculated ; and when the
coefficients are obtained from the forced harmonic oscillations, as in Equation (23), the FD
forced identifier is adopted.
The curves indicate that the adoption of a constant damping coefficient overestimates the
damping effect at lower frequencies, resulting in significantly underpredicted resonant response
for frequencies below 0.05 Hz. When frequency-dependent coefficients are adopted, however,
the situation is the opposite: the damping effect is underestimated, resulting in higher response
for lower frequencies. Using the coefficients based on the forced oscillations leads to improved
results in comparison with those obtained from the linearized thrust equations, in terms of
the pitch motion standard deviation (Table 4). All three methods perform equivalently in the
wave-frequency range (0.05-0.15 Hz), where the response is underpredicted due to the absence
of viscous excitation in the FD model.
Similar conclusions can be obtained regarding the tower-base bending moment (Figure 5).
The most remarkable difference in comparison with the plots for the pitch response is for Uw =
13.0 m/s (Figure 5a), which shows a persistent underprediction for all the three FD models
for frequencies above 0.04 Hz. The errors in predicted tower base bending moment, shown in
Table 4, are in general lower than for the pitch motion, since wave-frequency and 3p loads also
influence the tower-base bending moment, and are less affected by the damping dependence of
frequency. The accuracy is satisfactory for calculated and, especially, forced methods.
Figure 6 compares the low-frequency responses in pitch, for Uw = 19.0 m/s, when the inertia
effect is included or disregarded. The FD model with damping coefficient obtained from forced
oscillations is adopted, but only the blue curve includes the inertia coefficient – resulting in a
peak period about 2.5 s longer then when the inertia effect is not considered.

6. Conclusion
A method was developed to illustrate the importance of including the interaction between
the nacelle motions and thrust in the frequency-domain representation of a FWT. The thrust
fluctuations resulting from nacelle motions can be interpreted as frequency-dependent inertia and
damping effects, which challenges the traditional approach of adopting a constant aerodynamic
damping coefficient.
Expressions were derived for the respective coefficients based both on a linearized expression
for the thrust and on forced oscillations of a wind turbine in the time domain. The obtained
coefficients were included in a 3 DOF FD model of the OOstar semi-submersible FWT.
Comparisons were made between the models using a constant aerodynamic damping coefficient,
the frequency-dependent inertia and damping coefficients obtained in two different ways, and

9
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

10-1 10-1
SIMA SIMA
FD constant FD constant
-2 -2
10 FD calculated 10 FD calculated
FD forced FD forced

10-3 10-3
Sx5 (rad2.s)

Sx5 (rad2.s)
10-4 10-4

10-5 10-5

10-6 10-6

10-7 10-7
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Frequency (Hz) Frequency (Hz)
(a) Uw = 13.0 m/s. (b) Uw = 19.0 m/s.

SIMA
10-2
FD constant
FD calculated
FD forced

10-4
Sx5 (rad2.s)

10-6

10-8
0 0.05 0.1 0.15 0.2
Frequency (Hz)
(c) Uw = 25.0 m/s.

Figure 4: Frequency-domain platform pitch response – comparison between SIMA simulations

and frequency-domain analyses with different approaches for calculating the aerodynamic inertia
and damping.

time-domain simulations. Both platform pitch motion and tower base bending moment were
analyzed. It was observed that the constant damping coefficient underestimates the responses
in the frequency range of the FWT surge and pitch natural frequencies.
The frequency-dependent coefficients, on the other hand, overestimated the response in the
same frequency range, but are closer to the time-domain predictions. The aerodynamic inertia
was shown to increase the pitch natural period by about 2.5 s, for a mean incident wind speed
of 19.0 m/s. The differences between using coefficients based on the linearized thrust or on the
forced oscillations were not very significant, but a slightly better agreement of the LF responses
with the time-domain simulations was attained with the latter.
The introduction of inertia and damping coefficients was chosen for its didactic interest, but
the adoption of a linearized thrust as in Equation (9) in a FD model is a simpler and equivalent
approach to include the interaction between nacelle motion and thrust, with due consideration
of controller and rotor dynamic effects.
In any case, a better agreement of the LF response should still be pursued. Other linearization

10
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

Condition FD constant (%) FD calculated (%) FD forced (%)

C1 -0.2 +23.7 +21.8
C2 -24.8 +38.0 +20.0
C3 -29.6 +55.0 +30.6

Table 4: Error in pitch standard deviation compared to time domain simulations.

1017 1017
SIMA SIMA
FD constant FD constant
FD calculated FD calculated
FD forced FD forced

1016 1016
Sbm (N2m2.s)

Sbm (N2m2.s)

1015 1015

1014 1014
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Frequency (Hz) Frequency (Hz)
(a) Uw = 13.0 m/s. (b) Uw = 19.0 m/s.

1017
SIMA
FD constant
FD calculated
FD forced

1016
Sbm (N2m2.s)

1015

1014
0 0.05 0.1 0.15 0.2
Frequency (Hz)
(c) Uw = 25.0 m/s.

Figure 5: Frequency-domain tower base bending moment – comparison between SIMA

simulations and frequency-domain analyses with different approaches for calculating the
aerodynamic inertia and damping.

methods could be attempted, trying to preserve effects other than those obtained with the partial
derivatives of thrust and torque w.r.t. relative wind velocity, rotor speed and blade-pitch angle.

11
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

Condition FD constant (%) FD calculated (%) FD forced (%)

C1 -7.1 +13.0 +10.7
C2 -29.8 +15.1 +2.9
C3 -33.0 +19.2 +5.3

Table 5: Error in tower base bending moment standard deviation compared to time domain
simulations.

0.02
FD forced - w. inertia
FD forced - wo. inertia

0.015
Sx5 (rad2.s)

0.01

0.005

0
20 25 30 35 40 45 50
Period (s)

Figure 6: Pitch response estimated in the frequency domain for Uw = 19.0 m/s, with
aerodynamic damping estimated using the forced oscillations method and with/without
considering the aerodynamic inertia effect.

References
[1] Kvittem MI, Moan T. Frequency Versus Time Domain Fatigue Analysis of a Semisubmersible Wind
Turbine Tower. Journal of Offshore Mechanics and Arctic Engineering. 2015 Feb;137(1):011901–011901–
11. Available from: http://dx.doi.org/10.1115/1.4028340.
[2] Pegalajar-Jurado A, Borg M, Bredmose H. An efficient frequency-domain model for quick load analysis
of floating offshore wind turbines. Wind Energy Science. 2018;3(2):693–712. Available from: https:
//www.wind-energ-sci.net/3/693/2018/.
[3] Hegseth JM, Bachynski EE. A semi-analytical frequency domain model for efficient design evaluation
of spar floating wind turbines. Marine Structures. 2019;64:186 – 210. Available from: http://www.
sciencedirect.com/science/article/pii/S0951833918303149.
[4] Bachynski EE. Design and dynamic analysis of tension leg platform wind turbines [Ph.D. thesis]. Norwegian
University of Science and Technology. Trondheim, Norway; 2014.
[5] Nielsen F, Hanson T, Skaare B. Integrated dynamic analysis of floating offshore wind turbines. In:
Proceedings of the ASME 2016 25th International Conference on Offshore Mechanics and Arctic
Engineering; 2006. .
[6] Larsen TJ, Hanson TD. A method to avoid negative damped low frequent tower vibrations for a floating,
pitch controlled wind turbine. Journal of Physics: Conference Series. 2007 jul;75:012073. Available from:
https://doi.org/10.1088%2F1742-6596%2F75%2F1%2F012073.
[7] Jonkman JM. Influence of control on the pitch damping of a floating wind turbine. In: Proceedings of the
ASME Wind Energy Symposium; 2008. .
[8] Souza CES, Bachynski EE. Changes in surge and pitch decay periods of floating wind turbines for varying
wind speed. Ocean Engineering. 2019;180:223 – 237. Available from: http://www.sciencedirect.com/
science/article/pii/S0029801818315932.
[9] Goupee AJ, Kimball RW, Dagher HJ. Experimental observations of active blade pitch and generator control
influence on floating wind turbine response. Renewable Energy. 2017;104:9 – 19.
[10] Moriarty PJ, Hansen AC. AeroDyn Theory Manual. Denver, Colorado: National Renewable Energy

12
NAWEA WindTech 2019 IOP Publishing
Journal of Physics: Conference Series 1452 (2020) 012040 doi:10.1088/1742-6596/1452/1/012040

Laboratory; 2005. NREL/EL-500-36881.

[11] Müller K, Lemmer F, Yu W. LIFES50+ - D4.2 Public Definition of the Two LIFES50+ 10MW Floater
Concepts. University of Stuttgart; 2018.
[12] Hansen MH, Henriksen LC. Basic DTU Wind Energy controller. Technical University of Denmark; 2019.
DTU Wind Energy E No. 0028.
[13] Bak C, Zahle F, Bitsche R, Kim T, Yde A, Henriksen LC, et al. Description of the DTU 10 MW Reference
Wind Turbine. DTU Wind Energy; 2013. DTU Wind Energy Report-I-0092.
[14] SINTEF OCEAN. RIFLEX - Theory manual; 2016.
[15] SINTEF OCEAN. SIMO - Theory manual; 2016.
[16] Hansen MOL. Aerodynamics of Wind Turbines. Earthscan; 2013.
[17] Jonkman BJ, Kilcher L. TurbSim User’s Guide: version 1.06.00. National Renewable Energy Laboratory;
2012.

13
144 Appended Papers
Paper 3

Integrated design optimization of spar floating wind

turbines
John Marius Hegseth, Erin E. Bachynski and Joaquim R. R. A. Martins
Marine Structures, 2020

145
146 Appended Papers
 0DULQH 6WUXFWXUHV   

%  *  . **  - 7 

  -
journal homepage: http://www.elsevier.com/locate/marstruc

               

            
  !  ""#   



            


            !  " #$ %

# " $
% & 
' ( ) #  - $ " # % $

& ' # *    , , . ,*           *      
)                 *          *, 
(*         **    /0 1         )  *   *      

  
 ,        *   . .           2
+ ,     
          3 .               
 !*  )      *    * *    2      ,     ,
      4            ,        
  *    4         $    *    * .*
**     .    *    . *        *  
 *   *  $             * !  
  *    . .    * ,!  $    
        *        *,    *     
**    *      *  -, ,,  *   ,
   *      *    *   . .  *   * ,
                *   *   *  
   **  

   

(*       5(1$ 6    *2 *   *                **    
      7               *     *          **
   *   ! * 8  .       *      *     *    *
*    .**   *.*
    *                  
  *  *  *     ** *.   *   *   *        57)6  !  
   *.   * 9/:
# *    *  2       (1$              .     
$ 9;:    * 4 !,   *   .*  * .   (1$        9<:  
  **      **   *  ,   *   *         4.  
* *  5$&=6      + ** 2     9>: 2       * ,   (1$    
   
# *                     ,          

 %     
(  ' 3  ? 5  6

 @AA  A/0/0/BA3 ;0;0/0;CC/

" . D ' . ;0/E8 " .  .    ;E  ;0;08 #  <0  ;0;0
‹  7KH $XWKRU V  3XEOLVKHG E\ (OVHYLHU /WG 7KLV LV DQ RSHQ DFFHVV DUWLFOH XQGHU WKH && %< OLFHQVH
KWWSFUHDWLYHFRPPRQVRUJOLFHQVHVE\ 
+, , -  , 0DULQH 6WUXFWXUHV   

%*    9D:    *    ! . *   *       (1$       * 9B:
              ,  * (1$   !,   *   $  * * 
  *    .  * *    *     *       *   **  * 9C:   *, 
* , 3 . **       *          *       **  *  *     
 =        * .*   *2 *    .* *      *   $    * 2
 F   * 9G:         *        *      4*    .
 * *     ' 9E:    ,   *    2 *         *  
    *        .    *   4*    *  * *      .  
  *       *  -  9/0:      . .         
   
7   ,         *    4    **   *    9/0:      
 (1$              ,    $ * *     4*      
    ) 2   (**   *  9//:      **      *    *   
 (1$     ,        *       4    
$              *                    *  
    42                            *    
 (1$      *    * *           9/;: ' *      
*,        **      $ *  * 9/<:  * * 2    *     (1$  # 4 
         &  * 9/>:              , *  ,  * (1$
    **    **           *  !  *  5&H"6 $    
             *    ,! .* *        .  *        
         *    
# *     .     (1$         *,. * *           
*  
          **               -       * 
2  *    *      *                 * * .  
   $ *      *          * !   ’       
      
$          .*                 *   
*  *       *          # 2 * **          ,
     * *  *     *       * *     *  #    **
4  * ,4*   *    *   *    , , . ,*       /0 1  (1$   !
   $  *  *   ) 7#)   9/D:   * 4   ,       
*     $  *            *             
         2                        !* 
 *    3 .   ( **   *  . 4     *   ,   *   $       
  4           *      *   * * *    $         
                            * 9/B–/G:
-             *       * 4    *       *  
  *       7      * 4     *   *     !  *     
  .                      *    *  
7 *    * !      *  *          *  
$       **   $ *  (1$  *          *     -  ;

-  <       *       . *        3 .         *
$  *                 -  >        *   . 4  
  ,4*  *   * ( **  4      -  D

  

$   *    (1$   ** *     *   ,   *          *2 ,
           *       * (            ! 
.*     .  . *.   *   *             **     .
-  * 4  *       **     *      *        2 *  
  * *  
$                     7$I /0 1      9/E:   
   <;0  $ * **  ** 4**     **    .        **     ;B00 A<
$       *  /0   .   **  *  5-1&6  *     //E   .  -1& # 
           *    **    . * 2      ,   (   *   
,**   5  * *  6   .  *        ,  *       $      
*     *   *    *
)*       ),*, *            ,  * .    .**     . ),
     *   ** *   # ..    (1$   *        *     (  /


+, , -  , 0DULQH 6WUXFWXUHV   

.,, /  *  0 

$   *       !         *.   !    $  * 2   
,          *      *               * * ,*
 , , . ,*    *  -  ;> $ *                . *  x  u
  .*     * @
x ¼ x0 þ Δx; u ¼ u0 þ Δu: 5/6
$   !        2   
Δx_ ¼ AΔx þ BΔu; 5;6

 A    2  B     2
          Δ      . 

.,., %00  

.,.,, 1   0! 

$ *  *    *           , ,  57)( 6     (  ; $ !  
          7)( 5     4           * 6   .   
*   *  *      9;0:     2 * **    2   *    
  %–(    *       * * ;7  4   "     *  *
.               *           ’ !  $     7)(
    .  ξ ¼ ½ξ/ ξD ξC >  $   * !         ϕ_    *        
 *    * 7)(   @
€ ¼ QA  Ngear QG :
ID ϕ 5<6

 "
      .    2      !       23     !

 
 )..    (1$  


+, , -  , 0DULQH 6WUXFWXUHV   

   -* 7)(       *

(    ,.      *                ! *       3
   .  * *  
          *                 
!         *
#*   *    *      *  *        * *       
*      *          *   *      ** . *.  *  *     
   ) 2    9;/:      * *      **,*          * 
!,   *           *        *         ,
 .  * *                *   *    * !   4
       *  *                
   
.     (1$ .      *           *         * 
    *   2 * *  9;;;<: $ *              *8      
    *    * *       
1   * 7)(     .   .  4 
2 3
ξ
xs ¼ 4 ξ_ 5: 5>6
ϕ_

$     *           *          .  * *     
      .  u  u  $   *   .    
 
QG
usc ¼ ; 5D6
θ

 θ   ** . *  *          *    *         ! $
  *  *                    !  * 
     *      .*      
$    .  .      8  .   *   *    *           
 *    *  2          * * . * $  .  2        .  
        *    (  / $     .      **         * 
  *          **   4*             .      
    * *  $   .  2   
 >
usd ¼ vFT vMT vQA FW;1 FW;5 FW;7 ; 5B6

 4    2    .         *         !   .*     .
 -  ;D; 45;     *  . 2         7)(      %–(  
$  *       @


+, , -  , 0DULQH 6WUXFWXUHV   

x_ s ¼ As xs þ Bsc usc þ Bsd usd ; 5C6

ys ¼ Cs xs ; 5C6

    2       *  5M *       6   5K6    5D6  
    
 
0 I
As ¼ 1 1 5G6
M K M D

$     B  B *   *           * 7)(   *     2  
 *  * C            *        *    @
Cs ¼ ½0 0 0 0 0 0 1: 5E6

.,.,.,   
$         *  **     * 4   !,    *     &  
-. 9;>:  * 2  &   -ø* 9;D:   *    **   * 7)(       (  <
$       *      . 

TD ¼ kE ½xðtÞ  uðtÞ ¼ c juðtÞj

_ uðtÞ_ þ kG uðtÞ þ ω2 m xðtÞ; 5/06

 0ðÞ  *   *   *       ! ,   *       *    
 .     *        ! *     5  4   6 $      563 6   
      *   .   ! *         4    $ *     4 
5 6     4  5 6       ! ,   *   4        $ *    
 4  5/ 6           *   
rffiffiffi
8
c juðtÞj
_ uðtÞ_  c _ ¼ cL uðtÞ;
σ ðuÞ_ uðtÞ _ 5//6
π

 σ ð0Þ_     .    0 _   ! 5/06         .     ’ !   ,+ 
2         *            *  2              
     "*        *    *  *.*   2  *           *  *.* 9;D:@
8  
  y2
>
> 1  exp  3k2 þ 1 ; 0  y  y0
>
< 2
FY ðyÞ ¼ " pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi # 5/;6
>
> 3k2 þ 1 y0
>
: 1  exp  y ; y  y0 ;
2k 2

  ¼ 
=σ ð
Þ   *    

1
y0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi: 5/<6
2k 3k2 þ 1
$  6      * .          

   -  * 4      *   *  **   &   -ø* 9;D:


+, , -  , 0DULQH 6WUXFWXUHV   

.,7, 8     

$  *  *    *    **          . *,  ,  **     9;B: $
          ** @   , !   **   * ,       ** . *, 
  **    .,     $ **        ** * *          
  .    8  .    *      *     * 
     *    
  *          **   4 ,   
x_ c ¼ Ac xc þ Bc uc ; 5/>6

yc ¼ Cc xc ; 5/>6

       *            !   *  *@
 
_ QG
uc ¼ ½ϕ; yc ¼ : 5/D6
θ

(        4*    4 ,  * ,  4*     ! ω  .   ,! 2   
   *  
    " #
0 1 0 ϕlp
Ac ¼ ; Bc ¼ ; xc ¼ ; 5/B6
0 ωlp ωlp ϕ_ lp

 ϕ_   * ,  4*     *           !      *   ! 
      *      !        *  ,      2      
# .          !          !   =
  **         *,
* . *  *@
 
0 2K ϕ_ 0
Cc ¼ ðbelow
ratedÞ; 5/C6
0 0
 
0 0
Cc ¼ ðabove
ratedÞ:
ηk k i ηk kp
 &     !    η6   , *    * 6  6   *    *  
  =
  **   .*

.,, 8 (  

$ *    *    *       * * ,*      ys ¼ uc  u ¼ yc @
x_ ¼ Ax þ Busd ; 5/G6

y ¼ Cx; 5/G6


     
As Bsc Cc Bsd xs
A¼ ; B¼ ; x¼ ; 5/E6
Bc Cs Ac 0 xc

 C       2 $ * ,*         !    $    2 
        ! ω 4  @

HðωÞ ¼ CðiωI  AÞ1 B: 5;06

$  , *    2      .  y   **  

Sy ðωÞ ¼ HðωÞSu ðωÞHðωÞH ; 5;/6

 ðÞ-     3   9;C: $ .    y     *     *  Sy ðωÞ
$  *    2 5Su ðωÞ6      *        **   
 
Swind ðωÞ 0
Su ðωÞ ¼ 5;;6
0 Swave ðωÞ

$ ,  *   .         .          * 9;G:
$  *        *        ! *    4         *  


+, , -  , 0DULQH 6WUXFWXUHV   

   *  $     **  * *      7 *   9;E:
$ 2               .     5#I"6   9<0: #   
         +       * .       5%7(6   2 .* 9  
            9;C:

y2
FY ðyÞ ¼ exp  νþ ð0Þ T exp  2 ; 5;<6
2σ

 νþ ð0Þ    ,              ,   ,  *   @
rffiffiffiffiffiffi
1 m2
νþ ð0Þ ¼ : 5;>6
2π m0

$  *   4  @

Z ∞
mn ¼ ωn SðωÞdω: 5;D6
0

$    * 2  .*  /  /            *  2    $ .*
  2   @

FY ðy1h Þ ¼ e1 : 5;B6

$  .*           /      * 2 .* 1 .      *  
             **  2                  /0 1
 (1$    .    9;0: 1      *    *       <0J   * ,   
/DJ    , 2       *   ,   *  

.,:,    

.,:,, /  * ;  

$        *       *,*   56   9</<;: # ! , 
 *              **   * $ +*       
*    *    * $  *   *   .*       *        4  @
Vn ¼ v ð1  an Þ; 5;C6

Vt ¼ ϕ_ r ð1 þ at Þ; 5;C6

  * .           *   *        .*
  *   ,
         *        ,** *  *     *  *  
   $   .*   5    * φ    * *   2   
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
W ¼ V 2n þ V 2t ; 5;G6

Vn
φ ¼ tan1 : 5;G6
Vt
$ *          *   *’        *@
α ¼ φ  ðθ þ βÞ: 5;E6
 β    * $ *              * *     4 
1
L ¼ ρcW 2 Cl ðαÞ; 5<06
2

1
D ¼ ρcW 2 Cd ðαÞ: 5<06
2
$  *   *   * .      *               
Fn ¼ L cosφ þ D sinφ 5</6

Ft ¼ L sinφ  D cosφ: 5</6

$ *   *             4          *      
           *  .  *   *


+, , -  , 0DULQH 6WUXFWXUHV   

Z R
Fy ¼ Fn dr; 5<;6
0

Z R
Mz ¼ r Fn dr; 5<;6
0

Z R
My ¼ r Ft dr: 5<;6
0

#   !  5;C6–5<;6         *             *  * $
  *        *     * 4 (1$  *  *      *      4   $* 
2  @

Fy  Fy0 þ Fy;v Δv þ Fy;ϕ_ Δϕ_ þ Fy;θ Δθ; 5<<6

Mz  Mz0 þ Mz;v Δv þ Mz;ϕ_ Δϕ_ þ Mz;θ Δθ; 5<<6

My  My0 þ My;v Δv þ My;ϕ_ Δϕ_ þ My;θ Δθ; 5<<6

            *  . .       . * 
∂Ax
Ax;y ¼ : 5<>6
∂y

.,:,., <        

#     .                4* $  .          
 * *  5     *         !6  **      
#    *  9<<:  5 * 6  , *       **  *       * *
=  6      @
X
∞
SvðjkÞ ðωÞ ¼ _ SU ðjω  nϕjÞ;
einψ KnðjkÞ ðjω  nϕjÞ _ 5<D6
n¼∞

 % ðωÞ        ψ    *   *   &   (    4   
    γ   (   2       *@
Z
1 π  
KnðjkÞ ðωÞ ¼ γ ω; djk cosðnθÞ dθ: 5<B6
π 0

$                  ! 5  6    
% 9<>:  2   
 !
2 2 0:5
fd 0:12d
γðf ; dÞ ¼ exp  12 þ ; 5<C6
Uhub Lc

    ,      /    * 
$     * *           **  *         *   
      * $         *  . .   * *    ! 5</6     
   1        * *      *    *   **       * 
  * *        ! 5  <= B= E= 6     *   
        
   <=  * $           *
$      .     **     *       *     .    @
SF ðωÞ
SvFT ðωÞ ¼  T 2 ; 5<G6
3 Fy;v

SMT ðωÞ
SvMT ðωÞ ¼ 2
; 5<G6
3
2
Mz;v

SQ ðωÞ
SvQA ðωÞ ¼  A 2 : 5<G6
3 My;v

$   *   2  ! 5;;6 S  ðωÞ       *   .             


+, , -  , 0DULQH 6WUXFWXUHV   

   *      #    *  9<<:           !      *
 *    **        **  *     *    $    .  ,
      

.,:,7, <0    

( **     .         .     *    . .   *     
 *          . @
FT;v ¼ 3 Fy;v ; 5<E6

3
MT;v ¼ Mz;v ; 5<E6
2

   %         *    *             C A 5*6  /D A 5 6


+, , -  , 0DULQH 6WUXFWXUHV   

QA;v ¼ 3 My;v : 5<E6

$  *   *     . .            *  *8  .       
  ** . *  *  **       * !**       . *     
      *    $  *  *        2    @

FT  FT0 þ FT;v Δv þ FT;ϕ_ Δϕ_ þ FT;θ Δθ; 5>06

MT  MT0 þ MT;v Δv; 5>06

QA  QA0 þ QA;v Δv þ QA;ϕ_ Δϕ_ þ QA;θ Δθ: 5>06

.,:,, > ? 

$ .   *       *    *    *        *    *   ,
   *       -
#  .*   -
'$( )     *  "
(&K 9<D:  -
)
9<B:      *   **    *   *       *    *    
        +*       *          **      $
,  * *   4*    $-  9<C:   . *     **    *  
 2   0/> 9<G:
(  >           42        *    .        C  /D A  $  *
  .**       2    !  *    <= !   *   * .   
*   $                        .     *      
*   *  .         /      5        6   * ,
! *    *   !  .    2    4*    *       
    * * *       **  .     *     4    !     <= 

.,@,       

$      *    ** *. .  *     5% 6 .  *     8  .  *  
    *   <0 %     .*  * ,            $    
   /  <0 A   / A   (           * .*     4 .   
*       $  .*   **     *                  3 
  *             * 9<E:     (  D $   *     *    .   *
  *     
X
NEC
pi ¼ 1; 5>/6
i¼1

 %    %      *       *       

   -  4 .    *       *    *                *   
  **  


+, , -  , 0DULQH 6WUXFWXUHV   

$  %     $* /    *    D0,      *  .*  2   

 % /      *             *  2           % ;
     *        .,     * % <         . ,     
    *  
(  ** %   F *   
% *       * *  * 5'$6 9<>:        
    

     

$                      *  *@

/ -  * 
; 1     
<     
> *,    *  

$  *         *       **      

7,, 4 6

$ (1$  *  *  ) 7#) 9/D:    ,      *   *    * 
    #  *     *   **      5   6        *   
  *2 *   *   *
        *  . .         .       
*                *   **   *    . . 
     9>0: $ *   *   *      4      ***    =

$          ,      *   . .          *  * #* 
      !   4  *           * 4      
 4 ,    2    $  *  . .        3           
    . * 9>/: 1     *    G0    /G            
     * >  *     *  ,.         2 * D    *   $
 * *        **     2     2 5K7-6 9>;:  (  /  #  2 
$ -')=$ *   9><:    *.       * -')=$    ! * !     5-H=6
    *.  *           *      ) 7#)    ) -    9>>:
   ,    .   *   **     * *      3 .   * ,
 *           .*   *    

7,., A!= 

**  (1$          *        *.*      5&%)6    *    
     .     *      .    &%)          
   *  *           *  * 3 .   #         
     *                               
*      *   9/0:
                *  * $  
     **  *          * *  .         (1$   # 
*.      !*     * .                
   ,!*      *  .  *       *   .*     .     * 
 .        $ 3 .                      @ 
     8 *     .  *        .        .   σ ðϕÞ
_ 

        !* @

f ¼ w1 Ctotal þ w2 σ ðϕÞ
_ : 5>;6



.  *       2    **  
%    / ; <

         5A 6 //0 ;/0 D00

-  4 .   - 56 CD EE /D/
- *     5 6 /;0 />0 /B0
$*       " 56 0/G 0/> 0/;


+, , -  , 0DULQH 6WUXFWXUHV   

$ * .     **       /  ;   / þ ; ¼ /; / ;; 2 ½0; /       
   !*   4*  !            .*             ,
    , 3 . 
$  *      
Ctotal ¼ Cspar þ Ctower þ Cmoor ; 5><6

 8   8   8        *            .* $ *        
        *         *   (     9>D: %  *   ** 
          * $     *  5  **     6 2   
X
Cspar ¼ km Mspar þ kf Ti ; 5>>6
i

 6  *       *    **  6                 
     2           $     *     * *  ** 
 * *      $     *          6 =6      *     
.        
         /0       * .*   1    *  9>D:
      $ **  * 9>B:     * *   G0         **  5  6 $   
**    *      
            *               *      /0 €A 
5 *  * 9>C:6  >D €A  5 *      * 9/>:6   *     6     . 
.*  ;C €A      2 *     .*    (**   *  9//: $    *    
2                <–>D €A        9>G:
       *  ">   
      <D €A       *     *        D–/0J         $
* .                           *  
,   8  . *   *         *    .        
$   .        .      ! 5>;6        .*     
<0 (&-         -  ;B       *       @
X
NEC
_ ¼
σðϕÞ _ ;
pi σðϕÞ;i 5>D6
i¼1

 σðϕÞ;
_       .         

   -*     


+, , -  , 0DULQH 6WUXFWXUHV   

7,7,
   !

7,7,, %  

        **           . *    $ *      **  
   **            *    . *       $    **
       . * *     * $   *     *       
    !*     **            * 

          *      ** $  *      !      *
$,            4. . * @      *           
   $ **   (  B $     . *                 **  
 *    CGD0 A<     **    

7,7,.,  
-  **   *             * *  *   *     $
  *    42          .          * 7$I /0 1    //E 
 . -1&       *  /0D  2              *  ///<  $  
. *      **                        ,
          **   *      GD00 A<   .    *     
9/E:

7,7,7,   
$          *     . * @     *          * *  
-1&   * *   *        *     *       *  *     ;7
          ,  *    .   *              * *     
  
$    *                 *   $ !        
2 *           .*      * 9>E:@

wmoor ¼ 0:1875 D2moor ½N = m 5>B6



EAmoor ¼ 90000 D2moor ½N; 5>C6


  2    


7  . *     
7    - * '  . * I 

-   
 // 
-  **      // 
-  * /  /0 
"        ;   > 
"        ;   > 
"       !   > 
"          > 
"         > 

$   
  // 

$  **      // 

   *   
  / 
( *   *  / 
   *  * /  / 
   *      *    )  / 

=  *    *,    ** 6 /  A

*    *,    ** 6 / A


+, , -  , 0DULQH 6WUXFWXUHV   

7,7,, 8   

$   *             *  *     *,  =
  ** #*   
.   *       *           *   *         
  . *           *         * *      *    *
 * # *. 2 *  **,   “ .    *” *   *,    **  .   
       *     *         9D0: #      *.      
    *   ** .*   9D/:  *   .*   9D;:   *        $
.*       *          . *         

7,7,:, %0
%      . *              .     *  *    *  //0
  . *  $   !     *        . *       , *   
       , *     ,4     *         ¼ >       ,
        $         . * .          *       
   *        . $        *     G0   . *  * 
$* ;

7,, 8  

$        *     *      5(&-6  *  5I&-6 *        

             .   ,  *   *   *     * *  $     
 2 *  *                       3

7,,, %  

$      ** **       -' .   7 *   $ 7 .     
   .*   7'L +& 9D<:  *             57((6  <0 $ 7((    
   !  .*  .  7'L +& 9D>:       7((     *  *   .  7'L 
9DD: $ *     (1$       ;0                  4 
2   
1:0
Dtot  ; 5>G6
DFF


    *     ;0  @

X
NEC
Dtot ¼ N20 pi Di : 5>E6
i¼1

 ;0            ;0   
         
**  *              7'L +& 9DB:    **    *       @

/ -**  * @  *   ** *      

; =*    * @  *   ** *  *     
< % *  * @  *    *     *

(  **    *  !   . 

σj;Sd  fksd ; 5D06

 σ=;-    ! .* .           **  *   $ .  *    *  .*
!      $   *              
2
Astiff  þ 0:06 lstiff tspar ; 5D/6
Zl2

 B  .  #*    .           "       
Istiff  Ix þ Ixh þ Ih ; 5D;6

              !         *  * ** 3  2 *  
   5") 6        5") 6  2*      5" 6    **     *     
*      *      **       2*            *  *,
*  9D>: $ .  * *  *           *   


+, , -  , 0DULQH 6WUXFWXUHV   

sffiffiffiffi
E
hstiff  1:35 tw;stiff : 5D<6
fy

$   *  *              

7hstiff
bstiff  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi; 5D>6
10 þ fEy hstiff r

   M ’  *      , * **    $  *       <DD = $ .      *
 *    **   !   *   4
2
kLC E
 2:5 ; 5DD6
iC fy

 6   . *   5   /06 /8   * *   *   8          * 
  (   **  *   *    γ 4 ¼ /:<D  *    .  * *  9D>:  *   *   γ   .
   **   !  9D>:@
8
< 1:10; λs < 0:5
γM ¼ 0:80 þ 0:60 λs ; 0:5  λs  1:0 5DB6
:
1:40; λs > 1:0;

 λ   ** * 4  7'L +& 9DB:

$      *  !*    *          .  4**       *   
  **         * ! *   
     *  *    .  2   *  
   /D     2 %  # 2    !   <;        /0J      *
 *   .       * $                   *  . 4  
      *     
#   **          ** *      !*      
     (1$          9//DC:      3 .   9CG: $  *      *   
 2  *    **  ** *   *     . 8  .     '3  *
9DG:     *                    2     *       
   2 $               
$ .      **    *            *    . *   
   ;D  .    *   .     . !       .   2 
  .*    <7  *    .   7'L +& 9DE:              * 
#   *.    *             *        
      .    **   * .    9B0:  *  (*   9B/:     * * 
      .       * ! 0D /0 /D  ;0     *  .  
     *            *    .    .       .*
*        *  
$    * *  * .*    **     . * **   .    .     * 
                  $        *  *  $
.  *              *    *  2    *  /0  *   
        
$ *       *            4 *   *   *       
 *           F * –-   5F-6   9B;B<: $ *           
  4     D0,     

7,,.,  
$                      *      7 .     7(( ¼ ;:0
9DD:
$   *          < 9B>:                  * 
* (         **   !      4
σcr
σx  ; 5DC6
γM γF

 σ )  2 *           σ     *  *     γ4  γ  *    *   
  .* $    *        γ ¼ // 9BD:  γ4 ¼ /<D 9D>: $          
*             **      !*      **      *    #
   ** F-             


+, , -  , 0DULQH 6WUXFWXUHV   

7,,7,   
$       *        @  2     *     *   2     
            *  *    * $ 4      2    9D>:
Td  0:95 Smbs ; 5DG6

 %              

Td ¼ γmean Tc;mean þ γdyn Tc;dyn : 5DE6

 ;  ;                    D0,      ,
 .*  γ  ¼ /<  γ  ¼ /CD   *        *  * 9D>: $         
 

Smbs ¼ c D2moor ð44  0:08Dmoor Þ; 5B06

                9BB:

$                           . *      !  
   *   . 4  *  .   *       ** *. I&-     9BC: $     *   
2  2        *   *     * .*              
   *   . *      *  !*   * *  *

7,,, 8   

(          =
  **     (1$       * $ .      ! 
  * ,* *      * .  . *        *  9BG: #*   *  
*        2          * 8 -  * 9BE:      (1$      .
*     * *  **    *     *  *    .  *  .    
       *       *  

  

$  *             **         .   2   (      
*  *          * *    $,      , 3 .  ! 5>;6  
            *   -  <          - !*  *     
    *    *                    **    
 * ( **    . 4    *       *  

,, % (   (    

$            4*   * !   4         ,
     *   !       **      .          <=  $
        .         **       *    **   (  C 7   “ ”
   *   !      **    *   . .    ,        *     ,
     .  . 

  *                    *    ,          
 *   .    * *         *  4      4         

  
     *     * !       * 


+, , -  , 0DULQH 6WUXFWXUHV   

,    #*     *     **       (1$      ,       *   
      *      2  $   *        * 4  *      %    * ,
   **       .   *           
      .
  *        2*     .       * ,    *   .   
.     *     *         *    * ,      
*

,., (  !       C0 

$    ,       !*                  .*   
     ! 5>;6        .    *    .*  .    *,  7$I /0
1         *   ** 9C0:      *       *    -
# $ *,   
.      .       00>< A                    
 * 4   **      .    2    *   *,   .*
$  *     (  G      4        .      .   * .* **
     *     .    *      .  *       .  2   
$        .     *  .       *,    **         * 
     *  *        * , D0,     5% /6  *  *     $ 
 *,    **       2          ,    *         
 .,              ** 5        6 *         * 
   !*       *   

   **        *              *  *  (  G       / ¼ C
/0G  ; ¼ ð/ C /0G Þ  ! 5>;6

,7, 1    

$       *    *      *   *      (  E (   **  .*  
     *  *     
$  *      * .* ** *      .      . *    
   **      2 *      .           *      
  *  * **            (&-      (     *  *  
         2 * /0       *                 .    
*   .    ** 2     *     2      .*        2 * <;
  DB  *   -1&   .* $   *        2       
         *   2    *        .       *     $
       !    * 4     *  *     .*   **    
3 $  *         *       .       *2    
!        .  *  *  *            *     
    
$  *   . .*     4 *    *     $ *   3  *   .

  ! $,       .          * 3 .    


+, , -  , 0DULQH 6WUXFWXUHV   

  " )       *         *   5*6   *   *   5 6 $ **    *     
>0 * .       . *    


+, , -  , 0DULQH 6WUXFWXUHV   

         5% 6     *   .              **     
    %     *   .  5% +6   .*      *      
$                2 *       (      ** 4**    
**         * ! *             !        
*    !  **    *        *   *  % + -       
            *    .         *   .     
.,!  )              *      #   ,
!    *      5* *6   * 2      * !  **
$ *              D 1 )%<,    9C/: .      *   
 .             *   !    ,**    #*    *   **  
      3 .          **   * . *       *    
       ** *     
   2  *        /D   2         <;    D0,       .
        $   *             % /       *    $  
 *             CC  $            *   *   $* <
 * *         *    *   $* > 7            *     
 .  $* /0  //  #  2 #
-  *                  2       ;0D   .  -1& $
               **             **    ** $    
  . *   *                        
    *  *  *
 *     *       *           , .  
 .  !          .               **    
 .          *            *     *    *   
. *     $  *                    **,,  *
         !* *,,  * **    5     .    *  6   
      **     4* 
$  *   **         .       *  *      *
      .             *    I*    *         
          %  !*     *  4          (&-  I&-    
 *    *   *    * * $* D *               *.   
         42       

 (  /0                  )),- 1  (*  - /0 1      
&
(-D0þ  3 9C;: $              *   *     ** **   
  )),-   %     .**                        2 *
;>J *    )),-        * * *  .    5GD00 .  G;>< A<6 $ * !
  4   ,       5  2 *  6 0G   0DE            )),-
    .*   *    **      ,    )     *      
                     *        $    * 
  *    *    <= 2          *   *       
$    *         (  // #** 4. . *  **       
      *  ** !    
 *      *  *     .    
      *    ** * 

,, 8   

$      *             *       *   !*   (  /;

      *       ** *   *  .        
$            **   (1$   *             
  *  6  6 42      * .*      $* B $ .*          
9;0:    *                 * 7$I /0 1   ** 9C0:   *

 
)    *     5       6
= L* I 

7 G0> 
 *  **  /:DC /0C 
7 * /:C; /0> <
       % + >:BB /0E ;
L * % + *  -1& BB0 
L * %  *  -1& >;E 


+, , -  , 0DULQH 6WUXFWXUHV   

 
'*         (1$  
  L* I 

- />>C
=  <>>
. ;D0
/    /;D

 
)         
= L* I 

*.        . -1& /00 

*.       . -1& //DB< 
$ *  E:D /0D 
L * % +  . -1& >E/ 
       % + C:>B /0G ;

 
# %              &
(-D0þ )),- 1  (* 

 *   *   * !  *          
$     !*        *           ** 5    3 .     6 
   $* C I     *    **  *    * *        *   *      
*     .   #*   *      .*           * .*    
 * 2 *     *     **   *      *    2  . *   
*   


+, , -  , 0DULQH 6WUXFWXUHV   

 

 )          $ *     *   * 

 
 )    *,    **   *      *      .  

 
( 2   *    
= L* I 

6 0/CE>  A

6 00/BD A


)3 .   .*         ** *      
-    9/0B €: ' *      . 9:

1    ** /;/C /BD

1     ** /;/E /G;

7  0;J /0EJ


+, , -  , 0DULQH 6WUXFWXUHV   

,:,   

$           *   $* G $     * I&-               ,
      % /     2      *      -  >< $    *     
     $    *      * .* *     *     !     *       $
*  *     *        *   4*   . ** *      8  .
          2  .   ** 9C<:          *          * 
 **       *          , . * 
     *      4           *  *    *    
*          *   #*      *       *       ,
!*  *.*  2             *  .              
  *              $       *      
        *   *      *      ,    *     
     *  $  *   *        **             
* *    *  #    *    * ,!    *          
*       **       * ,! *     9C>:  .    *       
      
$ *                    *         *  *
  *  -                   (1$  *       9CD:       
      .     4            *    
      
   *         .    .       *         **   
   *        

,@,
   ? 

$            . 4  *    *          *   , ,
. ,*    ,   *   -
#     **  *   **    *   *  ,
   *       *      *    *    %–(       .    
        ’ ! 
 .     *   *           
          *     *     ( –F* .       
* -   .           *    *   4   . 2    . ** 
  *        
 *   * 2 *        *     *       *    *       
  ’ !  #   *       *                 *
        /:>G /0G 'A $    *                    
-  ;D>  .           *
$          **     *   <0 (&-       2 /, *    "  
      –=* *                  . .* .  2  
$ 2         * 2  .*  /    D0,           #I"   
            4     *  9CB: /, *    ;0        
   D0,    
$   *       “-  /”          9;0:    B0J    **   
 >0J      *   **     *  4         ,    $ (1$     
                   *   *  $    *     . 4    
                         *  *    *  
(    *   *         *   ,   *      (  /<  (  />      * 
 *      .* (     *       .*  .     *   *   .  
 *         *   *        $   *   *      7 * *,
*             *, *  -' . .*     ,*   $         *
       *  .*    *   4       * ,*   $  * 
 *      . **      2          * .* * 
(               *   *           * **     *
   *    *    **   . . $   *    .      *     
 *   *           2                  *    *
 **        *       *              
     *    *  (  /<
$   *     *    *   *             (  />  
*      ,   *   -
#        1 * .       ! 
  ,+                       7 *      ( 
           *   +    *        *       7 *
            (1$ 9;0:


+, , -  , 0DULQH 6WUXFWXUHV   

 !
)          
= L* I 

  00G/ 
*   /0:C 
/  /><<0 
)  />EC; 

 
 %         *   *        *  

#*     .    **  *     .            *   *
$                *             *      
 *    ** *    **
$ 2 .*       *        *   $* E $ *    * .   
*       *       *   *        *    * $   
           *   *   *    >J     2 .*     *
          .    *   *

 $   

# *    , , . ,*   (1$  *   .*              *   
      *,    **    /0 1  (1$ $  *                  
           .              !* 
$  *             *        * *    2      ,     ,  
    4     
  * *      * ,         **    .    
   (1$      *    4  (&-       4       . *   
.   *    *   .    .    ,      **  *     * *
       *


+, , -  , 0DULQH 6WUXFWXUHV   

 
 %         *   *          

 "
%     2        
 *
  *   )     

- =       - =      

&  <D/  C0  B:/D /0B ' <;0  /D0  D:E/ /0B '
' *  ;E>  DE  B:BG /0B ' ;C/  //<  B:BG /0B '

$ *      *   *   .         **  *    . *   
 $  *   2        %    % +            
* !      !*   .  .    * ,!     *      $
*      *       .   * *  * !    .,!

$       .       *        *,    *    
     *  **     (            *    . *  
                                  
 *  L         * , 3 .       *     *   . *   
 .  ,       .            5    *   6
)       42       *      *  0;J      /0DJ *     
.         * 
$ *     .                     *      
 .   2      .          * 
     4   * 
  *   * , 3 .         *      . *  .     #    
-  <; 4             , 3 . **     .     *   
  ,    . *  $  *         *  
$    *  *    !        *             *  


+, , -  , 0DULQH 6WUXFWXUHV   

        $     *     *  *             
           #*       *  *  * *    *     *   
       #   *       *      !       .  *   
    *     * **        2 *          *  
   , . *  
$   *         . 4   , ,, *          * 4  *  .
 *            2   
 *  *    *   . .   **  
 *               *    * *       $       
.    * *    *         *  
$     *       *   *         .         *   
 *      $    .*      *         **    * 
  .  .**   .  $   *    2        *        * 
      *    .*   *  

%   &   

N $   *   .        4 *     * *     * .   
         

' ()

(  *      "  %  *  '  5'("6    3  ;C>G;C , “+   -” **
  * $ 4    *        * *       ) 7#)  * 

' %    *  & +    *&  )


#
7   *     

*.   . -1& 9: )   9: 1**    9:

 G0:>;C ;/<BD 00G>

 CD:0B> /ED0D 00CD
 BG://C /C0DD 00B>
 B;:;>E /D00< 00DB
 DB:;>; /;E>0 00D/
 D0:BCC />0C0 00D>
 >>:</; /B</> 00B/
 </:DG> ;0BG/ 00BG
 /G:G;E /B;/; 00B0
 G:C>/ /;CG> 00GD
/0000 /;BB0 00B/




7        

*.   . -1& 9: )   9: 1**    9:

/0000 /;BB0 00B/

;0D00 /><CC 00<C
</000 />0EB 00<<
>/D00 /<G;C 00<0
D;000 /;C>; 00;E
B;D00 //>;; 00;E
C<000 /0>C< 00;C
G<D00 EGE; 00;>
E>000 GE0> 00;/
/0>D00 CCGE 00/G
//DB<0 BCC> 00/D


+, , -  , 0DULQH 6WUXFWXUHV   

, -%./ % 

  ,
 K7-        * *        "  *  2       2  *  * 
 ***      ***   42       *          $   .   
     *         *              *   *   *  - &   
9>;:      *        4

& 

9/:   ""# & # *   *       @  .     #
##  ;0/<8D/5E6@;0>E–CD  @AA  A/0;D/>A/0D/GED
9;: $ % =              ’     
    $ * 8 ;00C
9<:     $ 7          * *        - ;0/;8;E@GE–//>  @AA  A/0/0/BA3
 ;0/;0E00/
9>: + ** 2 ,%    = =  *   * ,*       *     * ,    
@ =    
,  5;0/>6  *   *      5
-)=;0/>6 F @  8 ;0/>
9D: %* +(   &           *      # * ) " /EEB8/G5>6@/DC–C/  @AA  A/0/0/BA-0/>/,//GC5EB6
000;G,>
9B:    F  &  F  *                      !    *   =   ;0/;8
;>@;GE–EB  @AA  A/0/0/BA3  ;0/;0B///
9C: **     %  % . *.      @    *  ,                    
@
)%#'- ;0/< $-A
 @  **   '    8 ;0/<  @AA  A/0//0EA)%#'-,;0/<BB0G/C<
9G: F   **     %  % # * , 3 .                           ) 
  ;0/C8<5/6@BE–GC  @AA  A/0/00CA >0C;;,0/B,00C;,>
9E:  # ' $# &           ,*,         *  
@ =     ,   5;0/;6
 *     *      5
-)=;0/;6 "   +8 ;0/;
9/0:  *  -  - 7              ,  .   ) 1   ;0/>8/5/6@/;–;;
9//: (** 
 *  =# 1
'7)=$,       *                 
@ =     #- ;0// <0
 *         #     5)#;0//6 "  $ '* 8 ;0//  @AA  A/0///DA)#;0//,
>EEGD
9/;: -,-  **   *  7             
@ =     , 2 5;0/B6  *   *
     5
-)=;0/B6 "   +8 ;0/B
9/<: $ * %      &% )  *      * *       **  = %  ;0/>8DDD5/;0EE6  @AA  A/0/0GGA/C>;,
BDEBADDDA/A0/;0EE
9/>: & ( O** F M 1 -*  7 % =1 )                *     *,    **
@ =    
#- ;0/C <B  *         #     5)#;0/C6    ' 8 ;0/C  @AA  A/0///DA
)#;0/C,B;0<G
9/D: + -  $   ""#   F$ '*  # ) 7#)@  ,      *   *    *      
- *   )   ;0/E8DE@/0CD–/0>  @AA  A/0/00CA 00/DG,0/E,0;;//,
9/B: #  $ P 3    ""# .  * +1 . F +# *   *                     *.* 
    "  ;0/>8BG@GE<–E0D  @AA  A/0/0/BA3;0/>0;0>D
9/C: #  $   ""# P 3  . F +# .  * +1 # . *     4     ;0 1           *
1   ;0/B8/E@;0C/–GC  @AA  A/0/00;A/EC0
9/G:    # P* ( -ø  ''   ""# *   ,4*  %(7,             /0 1     1 
 - ;0/E8>@/B<–E;  @AA  A/0D/E>A ,>,/B<,;0/E
--' ;<BBC>D/


+, , -  , 0DULQH 6WUXFWXUHV   

9/E:  % P* (    " M #   &% '3 #    7      7$I /0 1      $ * "  7$I 1 
 " ,
,00E; 7$I 1  8 ;0/<
9;0:      #  ,* * !     *   4    .*             - ;0/E8B>@
/GB–;/0  @AA  A/0/0/BA3 ;0/G/00/D
9;/:  F %  *     **,*                =7   '   I .   -   $ * 8 ;0//
9;;:     F.  
 & %   $ 7  *              *   *    
 =   ;0/<8<D@;/0–;;  @AA  A/0/0/BA3  ;0/<0C/C>
9;<:   P F      $ 7     *       *    *   *     .  1   ;0/>8
/C@/<GD–>0E  @AA  A/0/00;A/B<E
9;>: &  F -. =% 4       **        *   
@ =     4  5/EE06       
     ' 8 /EE0
9;D: &   -ø* ' -  * 4    *        2   *   
@ )   # *   ; # *  *     *
2       *  # * 8 /EE<
9;B:     4* -  * 1 -  + 74     D,1             .*  $ * "  '"&A$=,D00,
<G0B0 ' * "*  &  8 ;00E
9;C: ' #   $ -           %  I .  = 8 ;0/<
9;G: 7'L +& &               $ * "  7'L+&,-$,0><C 7'L +&8 ;0/B
9;E: 7 * $ # *          *  =7   I .   1  8 /EGD
9<0: ' # +  )   %*          2      *       ;00G8/<>5G6@B;G–<B  @AA  A/0/0B/A
5#-%60C<<,E<EE 5;00G6/<>@G5B;G6
9</:   $   ' -  7    1        1 *8 ;0//
9<;: ** ( +  + "  #& 1   2 *     1 *8 ;00E
9<<: *  # 7  *               !    =7   I .  % ** &  8 /EEG
9<>:
% 1    ,  /@   !   $ * " 
% B/>00,/
 * *  * %  8 ;00D
9<D: -
'$( ) "
(&K     ;0/B
9<B: -
'$( ) -
)     ;0/B
9<C:     F * & $-   ’  @ .  /D0 $ * "  '"&A$=,D00,>B/EG ' * "*  &  8 ;00E
9<G:
% 1    ,  <@   !            $ * " 
% B/>00,<
 * *  * %  8 ;00E
9<E:    F *  $- . -            .   '  '  -
  )   = *  ;00;8/;5/6@/–G
9>0:  $   ""# #    *       *      *  *        *  . .  #% $ 
-  ;0/G8>>5>6  @AA  A/0//>DA</G;<E<
9>/:   ""#  $ ".    4            . .  *   *     *  *  #
##  ;0/<8D/5//6@
;DG;–EE  @AA  A/0;D/>A/0D;/G>
--' 000//>D;
9>;: & #   ""# 2        2         *   *    *           -
*   )   ;0/;8>B@;C<–G>  @AA  A/0/00CA 00/DG,0/;,0CB<,
9><: + ** =  1 - # -')=$@  -H= *     *, *         -
#  )   ;00;8/;5>6@ECE–/00B  @AA  A
/0//<CA-00<B/>>D0>>>B0EB
9>>: = "   =1   ""# ) @ # = ,  3,        *          - *   )   ;0/;8
>D5/6@/0/–/G  @AA  A/0/00CA 00/DG,0//,0BBB,<
9>D: (     F )      *   -  8 ;0/<
9>B: $ **  FO* F + *  L   7 M * ( +   " . - ' ( ' * L -  " +  "  ! 
+ & %    # ' #     7  * ,% = *     + (   1$($ .        . *  
     *  7;<@    ** *  ** *  *  *   *          1.%
)   "* 8 ;0/B $ *  
9>C:  # 3  % Q  # ' $# &.*                    *  *   . "  ;0/>8BB@
C/>–;G  @AA  A/0/0/BA3;0/>0/0/C
9>G: &  F = *     ;0/E
9>E: * ' (*    @         *  ) *4* =*  
8 /EEG
9D0: &  $   $7 #    .   .   *  !   .            **      = %  - ;00C8CD@
0/;0C<  @AA  A/0/0GGA/C>;,BDEBACDA/A0/;0C<
9D/: .  L + % 
   ") %  *         
@ ;0/; #    *    * %@ (   H
* 8 ;0/;  </>G–D<  @AA  A/0//0EA;0/;B</D/;0
9D;: &   %  **  *        * *            1   ;00E8<<5B6@D>/–D<  @AA  A/0/;B0A0<0E,
D;>K<<BD>/
9D<: 7'L +& (        *   $ * "  7'L+&,"=,%;0< 7'L +&8 ;0/E
9D>: 7'L +& (*         $ * "  7'L+&,-$,0//E 7'L +&8 ;0/G
9DD: 7'L 7            $ * "  7'L,)-,/0< 7'L8 ;0/<
9DB: 7'L +&  *    **  $ * "  7'L+&,"=,%;0; 7'L +&8 ;0/E
9DC: -  # * - F.  
   7    %         ,  *                  
            *     
@ =     , 2 5;0/B6  *   *   
  5
-)=;0/B6 "   +8 ;0/B  >/E–;B
9DG: '3 #"     $   2 * *    .       ,          )    #   ;0/E8/>/
 @AA  A/0///DA/>0>/EEB
9DE: 7'L +&  **   *       $ * "  7'L+&,"=,'/0< 7'L +&8 ;0/C
9B0: F  F   "** "     *     *           )  ;00>8</@;/CD–;0G  @AA  A/0/0/BA3
;00>0>00D
9B/:  * # (*   ) #* .     *         *  
@ )    *   8 /EEE  @AA  A/0>0><A
/0ED<, 
9B;: F *  + -   " -     *         .    2
@
 *     .   *  
   ,       *    P  - *8 /ECE  @AA  A/0/0/BA />C>,BBC05/C6BDDG>,G
9B<: & # F +   ""# # .*                2       - *   )   ;0/C8DD@
;DC–CC  @AA  A/0/00CA 00/DG,0/B,/>ED,/
9B>:   %     -      <@    *    /,B@    *   **   $ * "  ' /EE<,/,
B@ ;00C   %     -  8 ;00C
9BD: 7'L +& -          $ * "  7'L+&,-$,0/;B 7'L +&8 ;0/G
9BB: 7'L +& )        $ * "  7'L+&,)-,<0; 7'L +&8 ;0/G
9BC: 7'L +& =      $ * "  7'L+&,)-,<0/ 7'L +&8 ;0/G
9BG: % %,$ &            )2  I .  = 8 ;0/<


+, , -  , 0DULQH 6WUXFWXUHV   

9BE: - ( -*  7  7 % =1
               
@ =     #- ;0/> <<  *
        #     5)#;0/>6  (   %*    I-#8 ;0/>  @AA  A/0///DA)#;0/>,;>;>>
9C0:      &%   7$I      ** $ * "  7$I 1   " ,,00/G 7$I 1  8 ;0/<
9C/:     74             
L  )%< $ * "  '"&A$=,D00,>CD<D ' * "*  &  8 ;0/0
9C;: O** F & ( M 1 &
(-D0þ 7>;@ *  4      &
(-D0þ /01       $ *   I .   -8 ;0/G
9C<:  * 1 4* -   # (  *      *          
@ ;< #-      "  '. I-#8 ;00>
 @AA  A/0;D/>AB;00>,/00C
9C>: &   + P   $    *           * 4    *
@ =    )# ;00C ;B  *      
   #      7   %*    I-#8 ;00C  @AA  A/0///DA)#;00C,;E/DD
9CD: -   ' *  (+   $7 M. " .ø** ) " * # #*     *       7       
1   ;0/D8/G@//0D–;;  @AA  A/0/00;A/CD0
9CB: - ' ' #   %* ,         2 .*            ;0/08/<B5/;6@/>E/–D0/  @AA 
A/0/0B/A5#-%6/E><,CGGE0000/E>


Paper 4

Design optimization of spar floating wind turbines

considering different control strategies
John Marius Hegseth, Erin E. Bachynski and Joaquim R. R. A. Martins
Accepted to: Journal of Physics: Conference Series, 2020

175
176 Appended Papers
Design Optimization of Spar Floating Wind Turbines
Considering Different Control Strategies
John Marius Hegseth1 , Erin E. Bachynski1 , Joaquim R. R. A. Martins2
1
Department of Marine Technology, NTNU, 7491 Trondheim, Norway
2
Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI 48109, USA
E-mail: [email protected]

Abstract. One of the challenges related to the design of floating wind turbines (FWTs) is the strong
interactions between the controller and the support structure, which may result in an unstable system.
Several control strategies have been proposed to improve the dynamic behaviour, all of which result in
trade-offs between structural loads, rotor speed variation, and blade pitch actuator use, which makes
controller design a challenging task. Due to the interactions, simultaneous design of the controller and
support structure should be performed to properly identify and compare different solutions. In the present
work, integrated design optimization of the blade-pitch controller and support structure is performed for a
10 MW spar FWT, considering four different control strategies, to evaluate the effect of the controller on the
structural design and associated costs. The introduction of velocity feedback control reduces the platform
pitch response and consequently the fatigue loads in the tower, which leads to a decrease in the tower
costs compared to a simple PI controller. Low-pass filtering of the nacelle velocity signal to remove the
wave-frequency components results in reduced rotor speed variation, but offers only small improvements
in costs, likely due to the limited wave-frequency response for the considered designs. Comparisons with
nonlinear time-domain simulations show that the linearized model is able to capture trends with acceptable
accuracy, but that significant overpredictions may occur for the platform pitch response.

1. Introduction
For floating wind turbines (FWTs), the performance of the control system generally depends on the
support structure design and vice versa. A well-known interaction between the blade pitch controller and
platform motions is the introduction of negative damping above rated wind speed [1]. While detuning
the controller gains such that the bandwidth is reduced achieves stability, this results in poorer rotor
speed tracking performance [1, 2]. Several alternative methods have also been suggested to resolve the
issue, such as introducing a feedback term proportional to the pitch velocity [3] or nacelle velocity [4, 5]
to manipulate the generator speed reference. For these types of feedback control, Fleming et al. [6] also
suggested to remove the wave-frequency components from the velocity signal, as this reduced the tower
loads. All of these control strategies result in trade-offs between structural loads, rotor speed variation,
and blade pitch actuator use, which vary with different environmental conditions. Therefore, identifying
optimal control parameters is a challenging task.
Due to the strong interactions, simultaneous design of the controller and support structure should be
performed to have a fair comparison between different solutions. A step toward simultaneous design was
made by Lemmer et al. [7], who optimized the main dimensions of a three-column semi-submersible
FWT. They minimized a combination of material costs and damage-equivalent loads in the tower for
seven operational conditions, with a constraint on the static pitch angle at rated thrust. The controller
was tuned at each design iteration using a linear quadratic regulator (LQR) approach. To properly
identify and compare optimal solutions, the integrated control and structural designs should, however,
be evaluated over the lifetime of the system, considering actual design limits. The purpose of the present
study is to perform integrated design optimization of the blade-pitch controller and support structure for
a 10 MW FWT, considering long-term fatigue damage and extreme response constraints, to evaluate the
effect of different control strategies on the structural design and associated costs.

2. Linearized FWT dynamics

2.1. FWT definition
The present study perfoms the design optimization of a spar buoy that supports the DTU 10
MW reference wind turbine [8] at a water depth of 320 m. The steel hull is partially filled
with concrete ballast to achieve the correct draft, using a ballast density of 2600 kg/m3 . The
interface with the tower is located 10 m above the still water line (SWL), while the hub
height is 119 m above the SWL. A catenary mooring system consisting of three lines spread
symmetrically about the vertical axis is used for station-keeping. Only the response in the xz-
plane is considered in the current work, and co-directional waves and wind travelling in the positive
x-direction are applied in all simulations. An overview of the FWT system is shown in Fig. 1.

2.2. Linearized equations of motion

The system is linearized and expressed in state-
MT QA space form, and consists of a structural part
and a control system part, which are connected
to obtain the complete closed-loop aero-hydro-
FT servo-elastic model as described by Hegseth et
al. [9]. For each wind-wave condition, the
operational point is found from static equilibrium
z when the system is subjected to the mean
wind loads. The linearized system considers
dFW x perturbations in the state and input variables, x
and u respectively, about the operational point:

x = x0 + Δx, u = u0 + Δu. (1)

The dynamic equations of motion are then

expressed as

Δẋ = AΔx + BΔu, (2)

Figure 1: Overview of the FWT system.
where A is the state matrix and B is the input
matrix.

2.3. Structural model

The structural model for the platform and turbine considers four degrees-of-freedom (DOFs), namely
surge, pitch, first tower/platform bending mode, and rotor speed. The equations of motions for the
former three are found from generalized displacements similar to Hegseth and Bachynski [10], but using
a flexible hull to ensure a correct natural frequency for the first bending mode, while a rigid drivetrain
and rotor is assumed for the rotor dynamics. The structural state vector is thus written as
 
x s = ξ1 ξ5 ξ7 ξ˙1 ξ˙5 ξ˙7 ϕ̇ , (3)

where ξn represents generalized support structure DOF n, and ϕ̇ is the rotor speed.
Hydrodynamic excitation loads on the hull are described by MacCamy–Fuchs theory, and transverse
added mass is based on analytical 2D coefficients. Radiation damping is neglected, while viscous
damping is found from stochastic linearization of the drag term in Morison’s equation.
Wind loads on the rotor are derived from linearized BEM theory with the incoming wind field
described by the Kaimal spectrum and an exponential coherence function for the longitudinal wind
2
velocity component [11]. The blades are considered rigid in the model, and the aerodynamic forces
on the rotor are applied as resultant loads at the tower top. In addition, the static component of the
aerodynamic quadratic drag force on the tower is included.
The inputs to the structural system consist of both control system outputs and disturbances due to
environmental loads. The control input vector contains the references for the generator torque (QG ) and
the collective blade pitch angle (θ), and is defined as

usc = [QG θ] . (4)

The disturbance vector contains rotor-effective wind speeds for thrust, tilting moment and aerodynamic
torque, and generalized wave excitation forces for each support structure DOF, i.e.

usd = [vFT v MT v QA FW,1 FW,5 FW,7 ] . (5)

2.4. Controller description

The baseline linear control system consists of a generator-torque controller and a collective blade-pitch
controller, which work independently in below-rated and above-rated wind speeds, respectively. Below
rated wind speed, the generator torque is set to be proportional to the square of the rotor speed to
maintain the optimal tip-speed ratio. Above rated wind speed, the generator torque is kept constant, and
four different strategies are considered for the blade-pitch controller:
• CS1: Gain-scheduled PI controller
• CS2: Gain-scheduled PI controller + platform pitch velocity feedback
• CS3: Gain-scheduled PI controller + nacelle velocity feedback
• CS4: Gain-scheduled PI controller + nacelle velocity feedback + low-pass filter
For feedback control using platform pitch or nacelle velocity, we use the modified rotor speed
reference, ϕ̇0 , defined by Lackner [3]:

ϕ̇0 = ϕ̇0 (1 + kf ẋf ) (6)

where ϕ̇0 is the nominal rotor speed reference, kf is the velocity feedback gain, and ẋf is either the
platform pitch velocity or the nacelle velocity. An updated expression for the rotor speed error can then
be established as
Δϕ̇ = ϕ̇ − ϕ̇0 = Δϕ̇ − ϕ̇0 kf ẋf , (7)
where Δϕ̇ is the nominal rotor speed error. In CS4, the nacelle velocity signal is passed through a first
order low-pass filter to remove the wave-frequency components before it is fed back to the blade-pitch
controller.

2.5. Response calculations

The structural and control system models are written as a single closed-loop system, which is solved
in the frequency domain. Dynamic force equilibrium is then used together with the response spectra to
calculate the bending moment response along the tower.
The fatigue damage is calculated at selected locations in the tower using the Dirlik method [12],
while the extreme response of the support structure is found using the AUR method, where the most
probable maximum value in one hour is used in the design constraints. The model has earlier been
verified against fully coupled nonlinear time domain simulations in SIMA [9].
The rotor speed tracking performance is evaluated using the weighted average of the rotor speed
standard deviation, which is found by summing the values from each short-term condition that is
considered, weighted by their associated probabilities:


N EC

σ(ϕ̇) = pi σ(ϕ̇),i . (8)

i=1
3
Here, NEC is the number of short-term conditions, σ(ϕ̇),i is the rotor speed standard deviation in
condition i, and pi is the probability of the condition.
To evaluate blade-pitch actuator fatigue, the actuator duty cycle (ADC) was defined in Kendall et al.
[13] as the total number of degrees pitched divided by the total simulation time. Although ADC cannot
be used as an absolute measure of the actuator fatigue damage, it is suitable for comparison of different
control strategies. The normalized ADC, which is used in the current work, is defined by Bottasso et al.
[14] as 
1 T |θ̇i (t)|
ADCi = dt, (9)
T 0 θ̇max
where θ̇i is the blade pitch rate in condition i, T is the total simulation time, and θ̇max is the maximum
allowable blade pitch rate, which for the DTU 10 MW is equal to 10 deg/s.
If the process is ergodic, the ADC can be expressed using the expected value:
1  
ADCi = E |θ̇i (t)| . (10)
θ̇max
Assuming that the blade pitch rate is Gaussian, the expected value can be calculated as
  2
E |θ̇i (t)| = σ . (11)
π (θ̇),i
For the linearized model, the ADC can thus be derived from the blade pitch rate standard deviation. The
weighted average ADC used in the design optimization is then found similarly as in Eq. (8).

2.6. Environmental conditions

Fifteen different ECs are used to evaluate the long-term fatigue performance in the present
work. The conditions span mean wind speeds from 1-30 m/s with 2 m/s step, and for each
mean wind speed, the most probable values for the significant wave height and spectral peak
period are used. These values, as well as the associated probabilities of occurrence, are found
from the joint probability distribution derived by Johannessen et al. [15], and shown in Fig. 2.

Three ECs, described in Table 1, along the

50-year contour surface, are selected to evaluate
the extreme response. EC 1 and 2 represent
operational wind speeds above rated, while EC 3
considers an extreme storm condition, where the
turbine is parked with feathered blades. For all
ECs, the IEC Class B normal turbulence model
(NTM) [11] is used for the incoming wind, while
the waves are described by a JONSWAP spectrum
with a peakedness parameter of 3.3.
Figure 2: Significant wave heights, spectral peak
periods, and normalized probabilities applied in the 3. Optimization problem
fatigue calculations. The FWT model is implemented in OpenM-
DAO [16], which is an open-source framework
for multidisciplinary design, analysis, and optimization. The design is optimized using a gradient-
based approach, where the derivatives of the model are computed analytically using coupled adjoints.
The optimization problem is solved using the SNOPT algorithm [17], which uses a sequential quadratic
programming (SQP) approach, through the pyOptSparse Python interface [18].

3.1. Objective function

The objective function used in the present work is the combined cost of the platform and tower, Cspar
and Ctower respectively:
f = Cspar + Ctower . (12)
4
 Table 1: Environmental conditions for extreme response calculations.

Condition 1 2 3
Mean wind speed at hub height, U (m/s) 13.0 21.0 50.0
Significant wave height, Hs (m) 8.1 9.9 15.1
Spectral peak period, Tp (s) 14.0 15.0 16.0
Turbulence intensity at hub height, I (-) 0.17 0.14 0.12

The costs consider both material and manufacturing, using the cost models developed by Farkas and
Jármai [19]. Costs related to installation, maintenance, and decommissioning are not included. The cost
of the platform (and similarly of the tower) is expressed as

Cspar = km Mspar + kf Ti , (13)
i

where km is the steel cost factor, Mspar is the steel mass of the hull, and kf is the fabrication cost per unit
time. Ti is the time spent at fabrication stage i, expressed as a function of the geometry. The steel cost
factor, km , is assumed to have a value of 2.7 e/kg, while the ratio between the material and fabrication
cost factors, km /kf , is set to 1.0, which is a typical value for West European labour [19]. The cost of
the concrete ballast is neglected in the current work.

3.2. Design variables

Both the platform and tower are discretized into ten sections along the length. For the tower, the diameter
and wall thickness at the nodes connecting the sections are set as design variables. The length of the
tower sections is kept fixed during the optimization, to maintain the original hub height.
The structural design of the spar platform is primarily governed by buckling loads, which for most
parts of the hull are dominated by the hydrostatic pressure. The structural design of the platform is
therefore not considered in the study, and only the diameter at the platform nodes and the length of each
section are included in the design optimization. The wall thickness is expressed as a function of depth,
based on the optimized structural design from Hegseth et al. [9]. This ensures a proper mass distribution,
and penalizes designs with large drafts and consequently high external pressure loads, which require
increased use of material. The mooring system design is kept fixed during the optimization.
For the control system, the optimization considers the proportional (kp ) and integral (ki ) gains for
the blade-pitch controller, as well as the velocity feedback gain (kf ) for the pitch and nacelle velocity
feedback control. For CS4, the corner frequency of the nacelle velocity low-pass filter (ωf ) is also
included.

3.3. Constraints
The fatigue damage at each tower node is evaluated using an SN curve approach, where the D curve in
air from DNV-RP-C203 [20] is applied together with a design fatigue factor (DFF) of 2.0 [21], and the
lifetime of the FWT system is chosen to be 20 years. The fatigue design constraints are thus expressed
as

N EC
1.0
Dtot = N20 pi D i ≤ , (14)
DFF
i=1

where Dtot is the total fatigue damage in 20 years, N20 is the number of short term conditions in 20
years, and Di is the fatigue damage in condition i.
Tower buckling is assessed using Eurocode 3 [22], assuming that the tower is stiffened between each
section to reduce the buckling length. To ensure a smooth transition between the platform and tower,
the tower base diameter is set to be equal to the diameter at the platform top. Both fatigue and buckling
constraints are aggregated using Kreisselmeier–Steinhauser (KS) functions [23].
5
 The maximum platform pitch angle in the considered 50-year conditions is limited to 15◦ . Although
the heave response is not included in the model, heave resonance in the wave frequency range is avoided
by placing a lower limit of 25 s on the heave natural period. The added mass in heave is approximated
as the value for a 3D circular disc with the same diameter as the platform bottom [24].
The presented model is valid strictly for hull sections with vertical walls, and a maximum taper angle
of 10◦ is therefore applied as a constraint for each section of the platform, to avoid shapes where the
physics are not captured correctly. Offset constraints are not considered, as the surge response is mostly
governed by the rotor and mooring system design.
Appropriate upper limits for the rotor speed variation and blade-pitch actuator use are difficult to
quantify. The constraints are therefore based on values taken from the land-based DTU 10 MW wind
turbine with the original controller [25], where the weighted average rotor speed standard deviation and
ADC are found from nonlinear time domain analyses using the simulation tool SIMA. Initial analyses
found that the rotor speed variation obtained with the land-based turbine was unrealistic for the floating
system with the simplified controllers considered in the present work. To enlarge the feasible region of
the design space, the constraints for both the rotor speed variation and the ADC are scaled by a factor of
1.5 compared to the land-based values, as shown in Table 2.

4. Results
4.1. Optimized designs
The optimized support structure design for CS1 is illustrated in Fig. 3. The hourglass shape
taken by the platform below the wave zone increases the
distance between the center of buoyancy and the center of
gravity, which leads to increased pitch restoring stiffness, while
the relatively large diameter at the bottom results in larger
added mass and consequently longer natural period in heave.
For the upper part of the platform and intersection with the
tower, the optimizer finds a balance between a small diameter,
which is desirable with regards to hydrodynamic loads, and a
large diameter, which (together with a small wall thickness) is
the most cost-effective way to achieve the required fatigue life.
The optimized tower diameter and wall thickness distribu-
tions for the different control strategies are plotted in Fig. 4. All
four solutions follow the same trends, and the effect of velocity
feedback control is most visible for the wall thickness, where
the values for CS2-4 are approximately 20 % lower than for
the simple PI controller (CS1) along most of the tower length.
The reduced wall thickness is enabled by a decrease in the fa-
tigue loads for these controllers, which is the design-driving
constraint for the tower. For all four control strategies, the 15◦
pitch angle constraint is also active; however, as the 50-year
storm condition with parked turbine (EC3) is found to be the
critical load case for extreme response, this constraint is not
affected by the controller.
The improved fatigue performance can be understood by
examining the response spectra for the optimized designs. In
Fig. 5a, the tower base bending moment spectra are shown for a
mean wind speed of 15 m/s. Large differences are seen around
the pitch natural frequency at 0.15 rad/s, where the velocity
feedback controllers increase the aerodynamic damping and
Figure 3: Optimized support structure thus reduce the response. The low-pass filtering of the nacelle
design for CS1. The wall thickness in velocity removes the wave-frequency range from the signal,
the tower is scaled by a factor of 40 which results in a higher optimal velocity feedback gain and
relative to the diameter for illustration nearly eliminates the tower base bending moment response
purposes.
6
Table 2: Land-based and applied constraint values for the rotor speed variation and blade-pitch actuator
use.

Variable Land-based value Constraint value

σ(ϕ̇) [rad/s] 4.22E-2 6.33E-2
ADC [-] 5.10E-3 7.65E-3

Figure 4: Tower diameter and wall thickness for optimized designs.

arising from resonant pitch motions. The wave-frequency bending moments, on the other hand, is
unaffected by the control strategy. The blade pitch spectra in Fig. 5b show how the larger PI gains in
the velocity feedback controllers result in overall increased actuator use, with the exception of the pitch
natural frequency, as well as the wave-frequency band for CS4.
The optimized costs of the tower, platform, and tower plus platform for CS2-4 are shown in Fig. 6a,
compared to the optimized costs for CS1. The majority of the cost reductions come from the tower, due
to the improved fatigue behaviour, whereas the platform is less affected due to the fixed costs related
to buckling resistance and the 15◦ pitch angle constraint. However, because a lighter tower results in a
lower overall center of gravity, the platform pitch response is somewhat improved, and a small reduction
in platform costs of about 2 % is also observed. Because the platform accounts for 70-75 % of the total
costs for the considered designs, the resulting total cost reduction is approximately 6 %.
The resulting rotor speed standard deviations and ADCs, normalized by their maximum allowable
values from Table 2, are shown in Fig. 6b. For each control strategy, there exists a limit where no further
reduction in cost can be achieved by increasing the actuator use. This limit is higher for the velocity
feedback controllers than for a controller using only the rotor speed error as input, which causes the
ADC constraint to be inactive at the optimum for CS1. A larger ADC may result in higher probability
of fatigue failure for the actuator bearings, and therefore more detailed design considering the lifetime
of the system should be performed to determine appropriate values for this constraint.
For the rotor speed variation, better performance is achieved with CS4 than with the other control
strategies. Since there is a trade-off between rotor speed variation and structural loads (and thus costs)
as previously discussed, it is expected that larger cost reductions can be achieved with CS4 if the rotor
speed constraint is tightened.
Some limitations to this work should be noted. The standard deviation in steady-state conditions is
used as the only measure of the rotor speed tracking performance, and extreme rotor speed excursions
due to gusts are not considered. The effect of the controllers on surge motions, drivetrain response,
mooring line tension, or blade response has also not been studied. In addition, the performance of the
FWT system could likely be further improved by also adding individual pitch control or modifications
to the torque controller, which have not been examined in the present work.

7
 (a) Tower base bending moment. (b) Blade pitch.

Figure 5: Response spectra, 15 m/s mean wind speed.

(a) Costs relative to CS1. (b) Rotor speed variation and ADC values.

Figure 6: Objective and constraint function values for optimized designs.

4.2. Sequential versus multidisciplinary optimization

In the presented methodology, the platform, tower, and blade-pitch controller are opti-
mized simultaneously. This approach, commonly known as multidisciplinary design op-
timization (MDO) [26], is preferred for coupled systems, where a sequential optimiza-
tion process in general leads to suboptimal solutions on the overall system level [27].
To assess the importance of integrated design, results
using MDO are compared to a sequential optimization
for CS4, where the controller and support structure
are optimized separately. For the sequential approach,
the control parameters are first optimized to minimize
the rotor speed variation, with constraints on system
stability and ADC. All support structure parameters
are kept constant and structural constraints are not
considered. The support structure is then optimized
for minimum cost, keeping the controller parameters
fixed at their optimized values from the previous
step. This procedure is repeated until the system
has converged. Since this method results in different
objective functions for the two sequential problems, a
Figure 7: MDO vs. sequential optimization. multi-objective approach is utilized for the MDO study,
where a combination of costs and rotor speed variation
is minimized with different relative weighting.
Figure 7 shows the MDO Pareto front together with the optimal solution from the sequential
approach. The results confirm that improved designs can be achieved with integrated optimization;
however, the differences are small. For the same rotor speed variation, using MDO results in
8
 (a) Weighted rotor speed variation. (b) Weighted ADC.

(c) Weighted tower base 1-h fatigue damage.

Figure 8: Comparison of response parameters with SIMA.

approximately 0.5 % reduction in the costs of the tower, or 0.2 % in total costs. Although small coupling
effects are seen for the considered optimization problem, larger differences are expected in cases where
the controller has a greater effect on the structural response.

4.3. Verification
The optimized controller designs are verified through fully coupled nonlinear time-domain simulations
using SIMA, where two different ECs above rated wind speed are simulated with each control strategy
for a specified support structure design. Comparisons with the linearized model for different response
parameters, weighted by the probability of each condition, are shown in Fig. 8.
The linearized model is seen to mostly follow the trends observed in the nonlinear simulations, but
some errors are present. The largest errors are observed for the tower base fatigue damage, which
is significantly overestimated for CS1, whereas good agreement is obtained with the other control
strategies. The reason for the poor agreement is that the aerodynamic (and thus the overall) damping
for the platform pitch mode is much lower with this control strategy. Consequently, the resonant pitch
response becomes very sensitive to the presence and amount of additional damping in the system, which
is either not considered or underpredicted in the linear model. This disagreement was also observed for
a linearized model of the 10 MW OO-Star semisubmersible in Souza et al. [28], which used a control
strategy similar to CS1. The overestimation of fatigue damage means that the optimized tower design
for CS1 is more conservative than for the other control strategies, suggesting that the cost reductions
in Fig. 6a are highly optimistic, and that CS1 may yield a fatigue design similar to that with a nacelle
velocity feedback controller. It also suggests that future optimization studies using the linearized model
should consider more advanced control strategies than CS1, to limit the pitch response error.
In addition, the rotor speed standard deviation is consistently underestimated by about 20 % for all
four control strategies. This disagreement could be taken into consideration in the optimization process
by adjusting the constraint value.

5. Concluding remarks
The design of the platform, tower, and blade-pitch control system for a 10 MW spar FWT was optimized
simultaneously using a linearized aero-hydro-servo-elastic model and gradient-based optimization with
9
analytic derivatives. The goal has been to minimize the material and manufacturing costs of the support
structure, with constraints on tower fatigue damage and buckling, extreme platform pitch motions, rotor
speed variation, and blade-pitch actuator use, considering four different strategies for the blade-pitch
controller.
The effect of the controller on the structural response was limited to the fatigue damage in the tower,
since the storm condition with parked turbine was found to govern the extreme responses of the system
considered here. The reduction in tower loads for the velocity feedback controllers compared to the
simple PI control system was a consequence of lower platform pitch response, which led to a reduction
in the wall thickness required to satisfy the long-term fatigue damage constraint. Consequently, the
tower costs were reduced, and also the platform costs due to better dynamic performance. Although
low-pass filtering of the nacelle velocity signal did not offer significant cost reductions, this control
strategy also saw a reduction in rotor speed variation, since the constraint was inactive at the optimum.
It is also expected that the effect of this filter will be more prominent for FWT concepts with larger
wave-frequency response.
Comparisons with nonlinear time-domain simulations showed that the linearized model in general is
able to capture trends with acceptable accuracy, but that the platform pitch response can be significantly
overpredicted for designs with low aerodynamic damping if contributions from other sources of damping
are small. This was the case for CS1, which indicates that the cost reductions achieved for CS2-
4 are considerably overestimated. For the velocity feedback controllers, which increases the amount
of aerodynamic damping induced by the control system, this problem was not observed, and good
agreement was achieved.
The presented approach is useful for conceptual FWT design, where it can be used to quickly explore
the design space before resorting to higher fidelity tools for detailed subsystem analysis and design.
The model captures important interactions between the controller and support structure, and enables
assessment of trade-off effects in a lifetime perspective. Further, the methodology can be extended to
account for additional design parameters and load cases, which may help identify novel design solutions.

References
[1] T. J. Larsen and T. D. Hanson. “A Method to Avoid Negative Damped Low Frequent Tower
Vibrations for a Floating, Pitch Controlled Wind Turbine”. In: Journal of Physics: Conference
Series 75 (2007), p. 012073.
[2] J. Jonkman. “Influence of Control on the Pitch Damping of a Floating Wind Turbine”. In: 46th
AIAA Aerospace Science Meeting and Exhibit, Reno, Nevada. 2008.
[3] M. Lackner. “Controlling Platform Motions and Reducing Blade Loads for Floating Wind
Turbines”. In: Wind Engineering 33.6 (2009), pp. 541–553.
[4] G. J. van der Veen, I. J. Couchman, and R. O. Bowyer. “Control of floating wind turbines”.
In: 2012 American Control Conference, Fairmont Queen Elizabeth, Montreal, Canada. 2012,
pp. 3148–3153.
[5] P. A. Fleming et al. “Evaluating Methods for Control of an Offshore Floating Turbine”. In:
Proceedings of the ASME 2014 33rd International Conference on Ocean, Offshore and Arctic
Engineering (OMAE2014), San Francisco, California, USA. 2014.
[6] P. A. Fleming, A. Peiffer, and D. Schlipf. “Wind Turbine Controller to Mitigate Structural Loads
on a Floating Wind Turbine Platform”. In: Journal of Offshore Mechanics and Arctic Engineering
141 (2019).
[7] F. Lemmer et al. “Optimization of Floating Offshore Wind Turbine Platforms With a Self-Tuning
Controller”. In: Proceedings of the ASME 2017 36th International Conference on Ocean, Offshore
and Arctic Engineering (OMAE2017), Trondheim, Norway. 2017.
[8] C. Bak et al. Description of the DTU 10 MW Reference Wind Turbine. Tech. rep. DTU Wind
Energy Report-I-0092. DTU Wind Energy, 2013.
[9] J. M. Hegseth, E. E. Bachynski, and J. R. R. A. Martins. “Integrated design optimization of spar
floating wind turbines”. In: Marine Structures 72 (2020).
10
[10] J. M. Hegseth and E. E. Bachynski. “A semi-analytical frequency domain model for efficient
design evaluation of spar floating wind turbines”. In: Marine Structures 64 (2019), pp. 186–210.
[11] IEC. Wind Turbines - Part 1: Design Requirements. Tech. rep. IEC 61400-1. 2005.
[12] T. Dirlik. “Application of Computers in Fatigue Analysis”. PhD thesis. University of Warwick,
1985.
[13] L. Kendall et al. “Application of Proportional-Integral and Disturbance Accommodating Control
to Variable Speed Variable Pitch Horizontal Axis Wind Turbines”. In: Wind Engineering 21.1
(1997), pp. 21–38.
[14] C. L. Bottasso et al. “Optimization-based study of bend-twist coupled rotor blades for passive and
integrated passive/active load alleviation”. In: Wind Energy 16 (2013), pp. 1149–1166.
[15] K. Johannessen, T. S. Meling, and S. Haver. “Joint Distribution for Wind and Waves in the
Northern North Sea”. In: International Journal of Offshore and Polar Engineering 12.1 (2002),
pp. 1–8.
[16] J. S. Gray et al. “OpenMDAO: An open-source framework for multidisciplinary design, analysis,
and optimization”. In: Structural and Multidisciplinary Optimization 59 (2019), pp. 1075–1104.
[17] P. E. Gill, W. Murray, and M. A. Saunders. “SNOPT: an SQP Algorithm for Large-Scale
Constrained Optimization”. In: SIAM Journal on Optimization 12.4 (2002), pp. 979–1006.
[18] R. E. Perez, P. W. Jansen, and J. R. R. A. Martins. “pyOpt: A Python-Based Object-Oriented
Framework for Nonlinear Constrained Optimization”. In: Structural and Multidisciplinary
Optimization 45.1 (2012), pp. 101–118.
[19] J. Farkas and K. Jármai. Optimum Design of Steel Structures. Springer, 2013.
[20] DNV. Fatigue Design of Offshore Steel Structures. Tech. rep. DNV-RP-C203. 2010.
[21] DNV. Design of Floating Wind Turbine Structures. Tech. rep. DNV-OS-J103. 2013.
[22] European Committee for Standardization. Eurocode 3: Design of Steel Structures, Part 1-6:
Strength and Stability of Shell Structures. Tech. rep. EN 1993-1-6: 2007. 2007.
[23] G. Kreisselmeier and R. Steinhauser. “Systematic Control Design by Optimizing a Vector
Performance Index”. In: International Federation of Active Controls Symposium on Computer-
Aided Design of Control Systems, Zurich, Switzerland. 1979.
[24] DNV. Modelling and Analysis of Marine Operations. Tech. rep. DNV-RP-H103. 2011.
[25] M. H. Hansen and L. C. Henriksen. Basic DTU Wind Energy controller. Tech. rep. DTU Wind
Energy Report-E-0018. 2013.
[26] J. R. R. A. Martins and A. B. Lambe. “Multidisciplinary Design Optimization: A Survey of
Architectures”. In: AIAA Journal 51.9 (2013), pp. 2049–2075.
[27] M. Muskulus and S. Schafhirt. “Design optimization of wind turbine support structures-A
review”. In: Journal of Ocean and Wind Energy 1.1 (2014), pp. 12–22.
[28] C. E. S. Souza, J. M. Hegseth, and E. E. Bachynski. “Frequency-dependent aerodynamic damping
and inertia in linearized dynamic analysis of floating wind turbines”. In: Journal of Physics:
Conference Series 1452.012040 (2020).

11
188 Appended Papers
Paper 5

Effect of environmental modelling and inspection

strategy on the optimal design of floating wind tur-
bines
John Marius Hegseth, Erin E. Bachynski and Bernt J. Leira
Submitted to: Reliability Engineering and System Safety, 2020

This article is awaiting submission and is therefore not included

189
220 Appended Papers
Appendix B

Formulation of Rotor
Effective Wind Speed

As shown by Halfpenny (1998), the (single sided) cross spectral density of

the rotationally sampled wind speed for two blade elements j and k may be
written:
∞

Sv(jk) (ω) = einψ Kn(jk) (|ω − nϕ̇|) SU (|ω − nϕ̇|), (B.1)
n=−∞

where SU (ω) is the incoming wind spectrum, ψ is the azimuth angle between
the elements, and Kn is the nth Fourier coefficient of the coherence function
γ, which is Fourier expanded in the rotor plane:
 π
1
Kn(jk) (ω) = γ(ω, djk ) cos(nθ) dθ. (B.2)
π 0

The coherence function for two points with separation distance d and fre-
quency f in Hz is taken from IEC (2005) and expressed as
⎛ ! ⎞
"2 ! "2 0.5
fd 0.12d
γ(f, d) = exp ⎝−12 + ⎠, (B.3)
Uhub Lc

where Uhub is the mean hub-height wind speed and Lc is a coherence scale
parameter.
Using Eq. (B.1), the wind speed seen by each blade element can be found,
and used together with aerodynamic derivatives to calculate blade root

221
222 Formulation of Rotor Effective Wind Speed

forces and subsequently resultant rotor loads. For a three-bladed turbine

with nb blade elements on each blade, Sv (ω) is a 3nb x 3nb matrix with the
following structure:
⎡ ⎤
Sv,11 (ω) Sv,12 (ω) Sv,13 (ω)
⎢ ⎥
Sv (ω) = ⎣Sv,21 (ω) Sv,22 (ω) Sv,23 (ω)⎦ (B.4)
Sv,31 (ω) Sv,32 (ω) Sv,33 (ω)

However, if only the resultant loads are of interest, and the blades are
identical, the calculations can be simplified by defining a single ‘effective’
blade, and the cross spectral density matrix for the effective blade can be
found from the auto-spectral density for one of the blades in Eq. (B.4). For
the thrust force and aerodynamic torque, which is the sum of the blade root
shear force Fy and edgewise bending moment My respectively, the same
formulation can be used:
∞

Sv (ω) = 32 Sv,n (ω), (B.5)
n=−∞

where
⎧
⎪
⎨Kn (|ω − nϕ̇|) SU (|ω − nϕ̇|), n ∈ {..., −6, −3, 0, 3, 6, ...}
Sv,n (ω) =
⎪
⎩0, otherwise,
(B.6)
and Kn only considers elements on the same blade. For the tilting moment,
the effective blade formulation is somewhat different, as the azimuth angle
of the blade affects the contribution from the blade root moment on the
resultant load. The contribution from the flapwise blade root moment on
the rotor titling moment is proportional to cos(ϕ̇t), and the cross spectral
density matrix thus becomes:
! "2 
∞
3
Sv (ω) = Sv,n (ω), (B.7)
2 n=−∞

where
⎧
⎪
⎪K (|ω − (n + 1)ϕ̇|) SU (|ω − (n + 1)ϕ̇|), n ∈ {..., −7, −4, −1, 2, 5, ...}
⎪ n
⎪
⎨
Sv,n (ω) = Kn (|ω − (n − 1)ϕ̇|) SU (|ω − (n − 1)ϕ̇|), n ∈ {..., −5, −2, 1, 4, 7, ...}
⎪
⎪
⎪
⎪
⎩
0, otherwise
(B.8)
 223

As seen from the non-zero terms in Eqs. (B.6) and (B.8), the cancellation of
harmonics with a three-bladed turbine results in peaks at multiples of three
of the rotor frequency (i.e. 3P, 6P, 9P, etc.) for the resultant loads. The
tower top load spectra can be found by pre- and post-multiplying Sv (ω)
with the weight factors for each element, i.e. factors that relate wind speed
to the aerodynamic loads on the element:

SFT (ω) = WFT Sv (ω) WFT (B.9a)


SMT (ω) = WM T
Sv (ω) WMT (B.9b)

SQA (ω) = WQ A
Sv (ω) WQA . (B.9c)

Here,
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
Fn,v (r1 )Δr r1 Fn,v (r1 )Δr r1 Ft,v (r1 )Δr
⎢ F (r )Δr ⎥ ⎢ r F (r )Δr ⎥ ⎢ r F (r )Δr ⎥
⎢ n,v 2 ⎥ ⎢ 2 n,v 2 ⎥ ⎢ 2 t,v 2 ⎥
WFT =⎢ ⎥ , WMT = ⎢ ⎥ , WQA = ⎢ ⎥.
⎣ · ⎦ ⎣ · ⎦ ⎣ · ⎦
Fn,v (rN )Δr rN Fn,v (rN )Δr rN Ft,v (rN )Δr
(B.10)
The spectra for the effective wind speeds can be calculated from the tower
top load spectra and the total weight factors over the rotor:

SFT (ω)
SvFT (ω) = (B.11a)
(3 Fy,v )2
SMT (ω)
SvMT (ω) =  2 (B.11b)
3
2 Mz,v
SQA (ω)
SvQA (ω) = . (B.11c)
(3 My,v )2

The cross spectral density matrix for the wind loads can be found by ob-
serving which harmonics of the wind are shared by the different load com-
ponents. As shown by Halfpenny (1998), the thrust force and aerodynamic
torque on the turbine are perfectly correlated, as they share all harmon-
ics, and completely uncorrelated from the tilting moment, as they have no
harmonics in common.
224 Formulation of Rotor Effective Wind Speed
Previous PhD theses published at the Department of Marine Technology
(earlier: Faculty of Marine Technology)
NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY

Report Author Title

No.
Kavlie, Dag Optimization of Plane Elastic Grillages, 1967

Hansen, Hans R. Man-Machine Communication and Data-Storage

Methods in Ship Structural Design, 1971

Gisvold, Kaare M. A Method for non-linear mixed -integer

programming and its Application to Design
Problems, 1971

Lund, Sverre Tanker Frame Optimalization by means of SUMT-

Transformation and Behaviour Models, 1971

Vinje, Tor On Vibration of Spherical Shells Interacting with

Fluid, 1972

Lorentz, Jan D. Tank Arrangement for Crude Oil Carriers in

Accordance with the new Anti-Pollution
Regulations, 1975

Carlsen, Carl A. Computer-Aided Design of Tanker Structures, 1975

Larsen, Carl M. Static and Dynamic Analysis of Offshore Pipelines

during Installation, 1976

UR-79-01 Brigt Hatlestad, MK The finite element method used in a fatigue

evaluation of fixed offshore platforms. (Dr.Ing.
Thesis)

UR-79-02 Erik Pettersen, MK Analysis and design of cellular structures. (Dr.Ing.

Thesis)

UR-79-03 Sverre Valsgård, MK Finite difference and finite element methods

applied to nonlinear analysis of plated structures.
(Dr.Ing. Thesis)

UR-79-04 Nils T. Nordsve, MK Finite element collapse analysis of structural

members considering imperfections and stresses
due to fabrication. (Dr.Ing. Thesis)

UR-79-05 Ivar J. Fylling, MK Analysis of towline forces in ocean towing systems.

(Dr.Ing. Thesis)

UR-80-06 Nils Sandsmark, MM Analysis of Stationary and Transient Heat

Conduction by the Use of the Finite Element
Method. (Dr.Ing. Thesis)

UR-80-09 Sverre Haver, MK Analysis of uncertainties related to the stochastic

modeling of ocean waves. (Dr.Ing. Thesis)

UR-81-15 Odland, Jonas On the Strength of welded Ring stiffened

cylindrical Shells primarily subjected to axial
Compression

UR-82-17 Engesvik, Knut Analysis of Uncertainties in the fatigue Capacity of

 Welded Joints

UR-82-18 Rye, Henrik Ocean wave groups

UR-83-30 Eide, Oddvar Inge On Cumulative Fatigue Damage in Steel Welded

Joints

UR-83-33 Mo, Olav Stochastic Time Domain Analysis of Slender

Offshore Structures

UR-83-34 Amdahl, Jørgen Energy absorption in Ship-platform impacts

UR-84-37 Mørch, Morten Motions and mooring forces of semi submersibles

as determined by full-scale measurements and
theoretical analysis

UR-84-38 Soares, C. Guedes Probabilistic models for load effects in ship

structures

UR-84-39 Aarsnes, Jan V. Current forces on ships

UR-84-40 Czujko, Jerzy Collapse Analysis of Plates subjected to Biaxial

Compression and Lateral Load

UR-85-46 Alf G. Engseth, MK Finite element collapse analysis of tubular steel

offshore structures. (Dr.Ing. Thesis)

UR-86-47 Dengody Sheshappa, MP A Computer Design Model for Optimizing Fishing

Vessel Designs Based on Techno-Economic
Analysis. (Dr.Ing. Thesis)

UR-86-48 Vidar Aanesland, MH A Theoretical and Numerical Study of Ship Wave

Resistance. (Dr.Ing. Thesis)

UR-86-49 Heinz-Joachim Wessel, MK Fracture Mechanics Analysis of Crack Growth in

Plate Girders. (Dr.Ing. Thesis)

UR-86-50 Jon Taby, MK Ultimate and Post-ultimate Strength of Dented

Tubular Members. (Dr.Ing. Thesis)

UR-86-51 Walter Lian, MH A Numerical Study of Two-Dimensional Separated

Flow Past Bluff Bodies at Moderate KC-Numbers.
(Dr.Ing. Thesis)

UR-86-52 Bjørn Sortland, MH Force Measurements in Oscillating Flow on Ship

Sections and Circular Cylinders in a U-Tube Water
Tank. (Dr.Ing. Thesis)

UR-86-53 Kurt Strand, MM A System Dynamic Approach to One-dimensional

Fluid Flow. (Dr.Ing. Thesis)

UR-86-54 Arne Edvin Løken, MH Three Dimensional Second Order Hydrodynamic

Effects on Ocean Structures in Waves. (Dr.Ing.
Thesis)

UR-86-55 Sigurd Falch, MH A Numerical Study of Slamming of Two-

Dimensional Bodies. (Dr.Ing. Thesis)

UR-87-56 Arne Braathen, MH Application of a Vortex Tracking Method to the

Prediction of Roll Damping of a Two-Dimension
Floating Body. (Dr.Ing. Thesis)
UR-87-57 Bernt Leira, MK Gaussian Vector Processes for Reliability Analysis
involving Wave-Induced Load Effects. (Dr.Ing.
Thesis)

UR-87-58 Magnus Småvik, MM Thermal Load and Process Characteristics in a

Two-Stroke Diesel Engine with Thermal Barriers
(in Norwegian). (Dr.Ing. Thesis)

MTA-88- Bernt Arild Bremdal, MP An Investigation of Marine Installation Processes –

59 A Knowledge - Based Planning Approach. (Dr.Ing.
Thesis)

MTA-88- Xu Jun, MK Non-linear Dynamic Analysis of Space-framed

60 Offshore Structures. (Dr.Ing. Thesis)

MTA-89- Gang Miao, MH Hydrodynamic Forces and Dynamic Responses of

61 Circular Cylinders in Wave Zones. (Dr.Ing. Thesis)

MTA-89- Martin Greenhow, MH Linear and Non-Linear Studies of Waves and

62 Floating Bodies. Part I and Part II. (Dr.Techn.
Thesis)

MTA-89- Chang Li, MH Force Coefficients of Spheres and Cubes in

63 Oscillatory Flow with and without Current. (Dr.Ing.
Thesis

MTA-89- Hu Ying, MP A Study of Marketing and Design in Development

64 of Marine Transport Systems. (Dr.Ing. Thesis)

MTA-89- Arild Jæger, MH Seakeeping, Dynamic Stability and Performance of

65 a Wedge Shaped Planing Hull. (Dr.Ing. Thesis)

MTA-89- Chan Siu Hung, MM The dynamic characteristics of tilting-pad bearings

66
MTA-89- Kim Wikstrøm, MP Analysis av projekteringen for ett offshore projekt.
67 (Licenciat-avhandling)

MTA-89- Jiao Guoyang, MK Reliability Analysis of Crack Growth under

68 Random Loading, considering Model Updating.
(Dr.Ing. Thesis)

MTA-89- Arnt Olufsen, MK Uncertainty and Reliability Analysis of Fixed

69 Offshore Structures. (Dr.Ing. Thesis)

MTA-89- Wu Yu-Lin, MR System Reliability Analyses of Offshore Structures

70 using improved Truss and Beam Models. (Dr.Ing.
Thesis)

MTA-90- Jan Roger Hoff, MH Three-dimensional Green function of a vessel with

71 forward speed in waves. (Dr.Ing. Thesis)

MTA-90- Rong Zhao, MH Slow-Drift Motions of a Moored Two-Dimensional

72 Body in Irregular Waves. (Dr.Ing. Thesis)

MTA-90- Atle Minsaas, MP Economical Risk Analysis. (Dr.Ing. Thesis)

73
MTA-90- Knut-Aril Farnes, MK Long-term Statistics of Response in Non-linear
74 Marine Structures. (Dr.Ing. Thesis)

MTA-90- Torbjørn Sotberg, MK Application of Reliability Methods for Safety

75 Assessment of Submarine Pipelines. (Dr.Ing.
 Thesis)

MTA-90- Zeuthen, Steffen, MP SEAMAID. A computational model of the design

76 process in a constraint-based logic programming
environment. An example from the offshore
domain. (Dr.Ing. Thesis)

MTA-91- Haagensen, Sven, MM Fuel Dependant Cyclic Variability in a Spark

77 Ignition Engine - An Optical Approach. (Dr.Ing.
Thesis)

MTA-91- Løland, Geir, MH Current forces on and flow through fish farms.
78 (Dr.Ing. Thesis)

MTA-91- Hoen, Christopher, MK System Identification of Structures Excited by

79 Stochastic Load Processes. (Dr.Ing. Thesis)

MTA-91- Haugen, Stein, MK Probabilistic Evaluation of Frequency of Collision

80 between Ships and Offshore Platforms. (Dr.Ing.
Thesis)

MTA-91- Sødahl, Nils, MK Methods for Design and Analysis of Flexible

81 Risers. (Dr.Ing. Thesis)

MTA-91- Ormberg, Harald, MK Non-linear Response Analysis of Floating Fish

82 Farm Systems. (Dr.Ing. Thesis)

MTA-91- Marley, Mark J., MK Time Variant Reliability under Fatigue

83 Degradation. (Dr.Ing. Thesis)

MTA-91- Krokstad, Jørgen R., MH Second-order Loads in Multidirectional Seas.

84 (Dr.Ing. Thesis)

MTA-91- Molteberg, Gunnar A., MM The Application of System Identification

85 Techniques to Performance Monitoring of Four
Stroke Turbocharged Diesel Engines. (Dr.Ing.
Thesis)

MTA-92- Mørch, Hans Jørgen Bjelke, MH Aspects of Hydrofoil Design: with Emphasis on
86 Hydrofoil Interaction in Calm Water. (Dr.Ing.
Thesis)

MTA-92- Chan Siu Hung, MM Nonlinear Analysis of Rotordynamic Instabilities in

87 Highspeed Turbomachinery. (Dr.Ing. Thesis)

MTA-92- Bessason, Bjarni, MK Assessment of Earthquake Loading and Response

88 of Seismically Isolated Bridges. (Dr.Ing. Thesis)

MTA-92- Langli, Geir, MP Improving Operational Safety through exploitation

89 of Design Knowledge - an investigation of offshore
platform safety. (Dr.Ing. Thesis)

MTA-92- Sævik, Svein, MK On Stresses and Fatigue in Flexible Pipes. (Dr.Ing.

90 Thesis)

MTA-92- Ask, Tor Ø., MM Ignition and Flame Growth in Lean Gas-Air
91 Mixtures. An Experimental Study with a Schlieren
System. (Dr.Ing. Thesis)

MTA-86- Hessen, Gunnar, MK Fracture Mechanics Analysis of Stiffened Tubular

92 Members. (Dr.Ing. Thesis)
MTA-93- Steinebach, Christian, MM Knowledge Based Systems for Diagnosis of
93 Rotating Machinery. (Dr.Ing. Thesis)

MTA-93- Dalane, Jan Inge, MK System Reliability in Design and Maintenance of

94 Fixed Offshore Structures. (Dr.Ing. Thesis)

MTA-93- Steen, Sverre, MH Cobblestone Effect on SES. (Dr.Ing. Thesis)

95
MTA-93- Karunakaran, Daniel, MK Nonlinear Dynamic Response and Reliability
96 Analysis of Drag-dominated Offshore Platforms.
(Dr.Ing. Thesis)

MTA-93- Hagen, Arnulf, MP The Framework of a Design Process Language.

97 (Dr.Ing. Thesis)

MTA-93- Nordrik, Rune, MM Investigation of Spark Ignition and Autoignition in

98 Methane and Air Using Computational Fluid
Dynamics and Chemical Reaction Kinetics. A
Numerical Study of Ignition Processes in Internal
Combustion Engines. (Dr.Ing. Thesis)

MTA-94- Passano, Elizabeth, MK Efficient Analysis of Nonlinear Slender Marine

99 Structures. (Dr.Ing. Thesis)

MTA-94- Kvålsvold, Jan, MH Hydroelastic Modelling of Wetdeck Slamming on

100 Multihull Vessels. (Dr.Ing. Thesis)

MTA-94- Bech, Sidsel M., MK Experimental and Numerical Determination of

102 Stiffness and Strength of GRP/PVC Sandwich
Structures. (Dr.Ing. Thesis)

MTA-95- Paulsen, Hallvard, MM A Study of Transient Jet and Spray using a

103 Schlieren Method and Digital Image Processing.
(Dr.Ing. Thesis)

MTA-95- Hovde, Geir Olav, MK Fatigue and Overload Reliability of Offshore

104 Structural Systems, Considering the Effect of
Inspection and Repair. (Dr.Ing. Thesis)

MTA-95- Wang, Xiaozhi, MK Reliability Analysis of Production Ships with

105 Emphasis on Load Combination and Ultimate
Strength. (Dr.Ing. Thesis)

MTA-95- Ulstein, Tore, MH Nonlinear Effects of a Flexible Stern Seal Bag on

106 Cobblestone Oscillations of an SES. (Dr.Ing.
Thesis)

MTA-95- Solaas, Frøydis, MH Analytical and Numerical Studies of Sloshing in

107 Tanks. (Dr.Ing. Thesis)

MTA-95- Hellan, Øyvind, MK Nonlinear Pushover and Cyclic Analyses in

108 Ultimate Limit State Design and Reassessment of
Tubular Steel Offshore Structures. (Dr.Ing. Thesis)

MTA-95- Hermundstad, Ole A., MK Theoretical and Experimental Hydroelastic

109 Analysis of High Speed Vessels. (Dr.Ing. Thesis)

MTA-96- Bratland, Anne K., MH Wave-Current Interaction Effects on Large-Volume

110 Bodies in Water of Finite Depth. (Dr.Ing. Thesis)

MTA-96- Herfjord, Kjell, MH A Study of Two-dimensional Separated Flow by a

111 Combination of the Finite Element Method and
 Navier-Stokes Equations. (Dr.Ing. Thesis)

MTA-96- Æsøy, Vilmar, MM Hot Surface Assisted Compression Ignition in a

112 Direct Injection Natural Gas Engine. (Dr.Ing.
Thesis)

MTA-96- Eknes, Monika L., MK Escalation Scenarios Initiated by Gas Explosions on

113 Offshore Installations. (Dr.Ing. Thesis)

MTA-96- Erikstad, Stein O., MP A Decision Support Model for Preliminary Ship
114 Design. (Dr.Ing. Thesis)

MTA-96- Pedersen, Egil, MH A Nautical Study of Towed Marine Seismic

115 Streamer Cable Configurations. (Dr.Ing. Thesis)

MTA-97- Moksnes, Paul O., MM Modelling Two-Phase Thermo-Fluid Systems

116 Using Bond Graphs. (Dr.Ing. Thesis)

MTA-97- Halse, Karl H., MK On Vortex Shedding and Prediction of Vortex-

117 Induced Vibrations of Circular Cylinders. (Dr.Ing.
Thesis)

MTA-97- Igland, Ragnar T., MK Reliability Analysis of Pipelines during Laying,

118 considering Ultimate Strength under Combined
Loads. (Dr.Ing. Thesis)

MTA-97- Pedersen, Hans-P., MP Levendefiskteknologi for fiskefartøy. (Dr.Ing.

119 Thesis)

MTA-98- Vikestad, Kyrre, MK Multi-Frequency Response of a Cylinder Subjected

120 to Vortex Shedding and Support Motions. (Dr.Ing.
Thesis)

MTA-98- Azadi, Mohammad R. E., MK Analysis of Static and Dynamic Pile-Soil-Jacket

121 Behaviour. (Dr.Ing. Thesis)

MTA-98- Ulltang, Terje, MP A Communication Model for Product Information.

122 (Dr.Ing. Thesis)

MTA-98- Torbergsen, Erik, MM Impeller/Diffuser Interaction Forces in Centrifugal

123 Pumps. (Dr.Ing. Thesis)

MTA-98- Hansen, Edmond, MH A Discrete Element Model to Study Marginal Ice

124 Zone Dynamics and the Behaviour of Vessels
Moored in Broken Ice. (Dr.Ing. Thesis)

MTA-98- Videiro, Paulo M., MK Reliability Based Design of Marine Structures.

125 (Dr.Ing. Thesis)

MTA-99- Mainçon, Philippe, MK Fatigue Reliability of Long Welds Application to

126 Titanium Risers. (Dr.Ing. Thesis)

MTA-99- Haugen, Elin M., MH Hydroelastic Analysis of Slamming on Stiffened

127 Plates with Application to Catamaran Wetdecks.
(Dr.Ing. Thesis)

MTA-99- Langhelle, Nina K., MK Experimental Validation and Calibration of

128 Nonlinear Finite Element Models for Use in Design
of Aluminium Structures Exposed to Fire. (Dr.Ing.
Thesis)

MTA-99- Berstad, Are J., MK Calculation of Fatigue Damage in Ship Structures.

129 (Dr.Ing. Thesis)

MTA-99- Andersen, Trond M., MM Short Term Maintenance Planning. (Dr.Ing. Thesis)
130
MTA-99- Tveiten, Bård Wathne, MK Fatigue Assessment of Welded Aluminium Ship
131 Details. (Dr.Ing. Thesis)

MTA-99- Søreide, Fredrik, MP Applications of underwater technology in deep

132 water archaeology. Principles and practice. (Dr.Ing.
Thesis)

MTA-99- Tønnessen, Rune, MH A Finite Element Method Applied to Unsteady

133 Viscous Flow Around 2D Blunt Bodies With Sharp
Corners. (Dr.Ing. Thesis)

MTA-99- Elvekrok, Dag R., MP Engineering Integration in Field Development

134 Projects in the Norwegian Oil and Gas Industry.
The Supplier Management of Norne. (Dr.Ing.
Thesis)

MTA-99- Fagerholt, Kjetil, MP Optimeringsbaserte Metoder for Ruteplanlegging

135 innen skipsfart. (Dr.Ing. Thesis)

MTA-99- Bysveen, Marie, MM Visualization in Two Directions on a Dynamic

136 Combustion Rig for Studies of Fuel Quality.
(Dr.Ing. Thesis)

MTA- Storteig, Eskild, MM Dynamic characteristics and leakage performance

2000-137 of liquid annular seals in centrifugal pumps.
(Dr.Ing. Thesis)

MTA- Sagli, Gro, MK Model uncertainty and simplified estimates of long

2000-138 term extremes of hull girder loads in ships. (Dr.Ing.
Thesis)

MTA- Tronstad, Harald, MK Nonlinear analysis and design of cable net

2000-139 structures like fishing gear based on the finite
element method. (Dr.Ing. Thesis)

MTA- Kroneberg, André, MP Innovation in shipping by using scenarios. (Dr.Ing.

2000-140 Thesis)

MTA- Haslum, Herbjørn Alf, MH Simplified methods applied to nonlinear motion of

2000-141 spar platforms. (Dr.Ing. Thesis)

MTA- Samdal, Ole Johan, MM Modelling of Degradation Mechanisms and

2001-142 Stressor Interaction on Static Mechanical
Equipment Residual Lifetime. (Dr.Ing. Thesis)

MTA- Baarholm, Rolf Jarle, MH Theoretical and experimental studies of wave

2001-143 impact underneath decks of offshore platforms.
(Dr.Ing. Thesis)

MTA- Wang, Lihua, MK Probabilistic Analysis of Nonlinear Wave-induced

2001-144 Loads on Ships. (Dr.Ing. Thesis)

MTA- Kristensen, Odd H. Holt, MK Ultimate Capacity of Aluminium Plates under

2001-145 Multiple Loads, Considering HAZ Properties.
(Dr.Ing. Thesis)

MTA- Greco, Marilena, MH A Two-Dimensional Study of Green-Water

2001-146
 Loading. (Dr.Ing. Thesis)

MTA- Heggelund, Svein E., MK Calculation of Global Design Loads and Load
2001-147 Effects in Large High Speed Catamarans. (Dr.Ing.
Thesis)

MTA- Babalola, Olusegun T., MK Fatigue Strength of Titanium Risers – Defect

2001-148 Sensitivity. (Dr.Ing. Thesis)

MTA- Mohammed, Abuu K., MK Nonlinear Shell Finite Elements for Ultimate
2001-149 Strength and Collapse Analysis of Ship Structures.
(Dr.Ing. Thesis)

MTA- Holmedal, Lars E., MH Wave-current interactions in the vicinity of the sea
2002-150 bed. (Dr.Ing. Thesis)

MTA- Rognebakke, Olav F., MH Sloshing in rectangular tanks and interaction with
2002-151 ship motions. (Dr.Ing. Thesis)

MTA- Lader, Pål Furset, MH Geometry and Kinematics of Breaking Waves.

2002-152 (Dr.Ing. Thesis)

MTA- Yang, Qinzheng, MH Wash and wave resistance of ships in finite water
2002-153 depth. (Dr.Ing. Thesis)

MTA- Melhus, Øyvin, MM Utilization of VOC in Diesel Engines. Ignition and

2002-154 combustion of VOC released by crude oil tankers.
(Dr.Ing. Thesis)

MTA- Ronæss, Marit, MH Wave Induced Motions of Two Ships Advancing

2002-155 on Parallel Course. (Dr.Ing. Thesis)

MTA- Økland, Ole D., MK Numerical and experimental investigation of

2002-156 whipping in twin hull vessels exposed to severe wet
deck slamming. (Dr.Ing. Thesis)

MTA- Ge, Chunhua, MK Global Hydroelastic Response of Catamarans due

2002-157 to Wet Deck Slamming. (Dr.Ing. Thesis)

MTA- Byklum, Eirik, MK Nonlinear Shell Finite Elements for Ultimate

2002-158 Strength and Collapse Analysis of Ship Structures.
(Dr.Ing. Thesis)

IMT- Chen, Haibo, MK Probabilistic Evaluation of FPSO-Tanker Collision

2003-1 in Tandem Offloading Operation. (Dr.Ing. Thesis)

IMT- Skaugset, Kjetil Bjørn, MK On the Suppression of Vortex Induced Vibrations

2003-2 of Circular Cylinders by Radial Water Jets. (Dr.Ing.
Thesis)

IMT- Chezhian, Muthu Three-Dimensional Analysis of Slamming. (Dr.Ing.

2003-3 Thesis)

IMT- Buhaug, Øyvind Deposit Formation on Cylinder Liner Surfaces in

2003-4 Medium Speed Engines. (Dr.Ing. Thesis)

IMT- Tregde, Vidar Aspects of Ship Design: Optimization of Aft Hull

2003-5 with Inverse Geometry Design. (Dr.Ing. Thesis)

IMT- Wist, Hanne Therese Statistical Properties of Successive Ocean Wave

2003-6 Parameters. (Dr.Ing. Thesis)

IMT- Ransau, Samuel Numerical Methods for Flows with Evolving

2004-7 Interfaces. (Dr.Ing. Thesis)

IMT- Soma, Torkel Blue-Chip or Sub-Standard. A data interrogation

2004-8 approach of identity safety characteristics of
shipping organization. (Dr.Ing. Thesis)

IMT- Ersdal, Svein An experimental study of hydrodynamic forces on

2004-9 cylinders and cables in near axial flow. (Dr.Ing.
Thesis)

IMT- Brodtkorb, Per Andreas The Probability of Occurrence of Dangerous Wave

2005-10 Situations at Sea. (Dr.Ing. Thesis)

IMT- Yttervik, Rune Ocean current variability in relation to offshore

2005-11 engineering. (Dr.Ing. Thesis)

IMT- Fredheim, Arne Current Forces on Net-Structures. (Dr.Ing. Thesis)

2005-12
IMT- Heggernes, Kjetil Flow around marine structures. (Dr.Ing. Thesis
2005-13
IMT- Fouques, Sebastien Lagrangian Modelling of Ocean Surface Waves and
2005-14 Synthetic Aperture Radar Wave Measurements.
(Dr.Ing. Thesis)

IMT- Holm, Håvard Numerical calculation of viscous free surface flow

2006-15 around marine structures. (Dr.Ing. Thesis)

IMT- Bjørheim, Lars G. Failure Assessment of Long Through Thickness

2006-16 Fatigue Cracks in Ship Hulls. (Dr.Ing. Thesis)

IMT- Hansson, Lisbeth Safety Management for Prevention of Occupational

2006-17 Accidents. (Dr.Ing. Thesis)

IMT- Zhu, Xinying Application of the CIP Method to Strongly

2006-18 Nonlinear Wave-Body Interaction Problems.
(Dr.Ing. Thesis)

IMT- Reite, Karl Johan Modelling and Control of Trawl Systems. (Dr.Ing.
2006-19 Thesis)

IMT- Smogeli, Øyvind Notland Control of Marine Propellers. From Normal to

2006-20 Extreme Conditions. (Dr.Ing. Thesis)

IMT- Storhaug, Gaute Experimental Investigation of Wave Induced

2007-21 Vibrations and Their Effect on the Fatigue Loading
of Ships. (Dr.Ing. Thesis)

IMT- Sun, Hui A Boundary Element Method Applied to Strongly

2007-22 Nonlinear Wave-Body Interaction Problems. (PhD
Thesis, CeSOS)

IMT- Rustad, Anne Marthine Modelling and Control of Top Tensioned Risers.
2007-23 (PhD Thesis, CeSOS)

IMT- Johansen, Vegar Modelling flexible slender system for real-time

2007-24 simulations and control applications

IMT- Wroldsen, Anders Sunde Modelling and control of tensegrity structures.

2007-25
 (PhD Thesis, CeSOS)

IMT- Aronsen, Kristoffer Høye An experimental investigation of in-line and

2007-26 combined inline and cross flow vortex induced
vibrations. (Dr. avhandling, IMT)

IMT- Gao, Zhen Stochastic Response Analysis of Mooring Systems

2007-27 with Emphasis on Frequency-domain Analysis of
Fatigue due to Wide-band Response Processes
(PhD Thesis, CeSOS)

IMT- Thorstensen, Tom Anders Lifetime Profit Modelling of Ageing Systems

2007-28 Utilizing Information about Technical Condition.
(Dr.ing. thesis, IMT)

IMT- Refsnes, Jon Erling Gorset Nonlinear Model-Based Control of Slender Body
2008-29 AUVs (PhD Thesis, IMT)

IMT- Berntsen, Per Ivar B. Structural Reliability Based Position Mooring.

2008-30 (PhD-Thesis, IMT)

IMT- Ye, Naiquan Fatigue Assessment of Aluminium Welded Box-

2008-31 stiffener Joints in Ships (Dr.ing. thesis, IMT)

IMT- Radan, Damir Integrated Control of Marine Electrical Power

2008-32 Systems. (PhD-Thesis, IMT)

IMT- Thomassen, Paul Methods for Dynamic Response Analysis and

2008-33 Fatigue Life Estimation of Floating Fish Cages.
(Dr.ing. thesis, IMT)

IMT- Pákozdi, Csaba A Smoothed Particle Hydrodynamics Study of

2008-34 Two-dimensional Nonlinear Sloshing in
Rectangular Tanks. (Dr.ing.thesis, IMT/ CeSOS)

IMT- Grytøyr, Guttorm A Higher-Order Boundary Element Method and

2007-35 Applications to Marine Hydrodynamics.
(Dr.ing.thesis, IMT)

IMT- Drummen, Ingo Experimental and Numerical Investigation of

2008-36 Nonlinear Wave-Induced Load Effects in
Containerships considering Hydroelasticity. (PhD
thesis, CeSOS)

IMT- Skejic, Renato Maneuvering and Seakeeping of a Singel Ship and

2008-37 of Two Ships in Interaction. (PhD-Thesis, CeSOS)

IMT- Harlem, Alf An Age-Based Replacement Model for Repairable

2008-38 Systems with Attention to High-Speed Marine
Diesel Engines. (PhD-Thesis, IMT)

IMT- Alsos, Hagbart S. Ship Grounding. Analysis of Ductile Fracture,

2008-39 Bottom Damage and Hull Girder Response. (PhD-
thesis, IMT)

IMT- Graczyk, Mateusz Experimental Investigation of Sloshing Loading

2008-40 and Load Effects in Membrane LNG Tanks
Subjected to Random Excitation. (PhD-thesis,
CeSOS)

IMT- Taghipour, Reza Efficient Prediction of Dynamic Response for

2008-41 Flexible amd Multi-body Marine Structures. (PhD-
 thesis, CeSOS)

IMT- Ruth, Eivind Propulsion control and thrust allocation on marine

2008-42 vessels. (PhD thesis, CeSOS)

IMT- Nystad, Bent Helge Technical Condition Indexes and Remaining Useful
2008-43 Life of Aggregated Systems. PhD thesis, IMT

IMT- Soni, Prashant Kumar Hydrodynamic Coefficients for Vortex Induced

2008-44 Vibrations of Flexible Beams, PhD
thesis, CeSOS

IMT- Amlashi, Hadi K.K. Ultimate Strength and Reliability-based Design of

2009-45 Ship Hulls with Emphasis on Combined Global and
Local Loads. PhD Thesis, IMT

IMT- Pedersen, Tom Arne Bond Graph Modelling of Marine Power Systems.
2009-46 PhD Thesis, IMT

IMT- Kristiansen, Trygve Two-Dimensional Numerical and Experimental

2009-47 Studies of Piston-Mode Resonance. PhD-Thesis,
CeSOS

IMT- Ong, Muk Chen Applications of a Standard High Reynolds Number

2009-48 Model and a Stochastic Scour Prediction Model for
Marine Structures. PhD-thesis, IMT

IMT- Hong, Lin Simplified Analysis and Design of Ships subjected

2009-49 to Collision and Grounding. PhD-thesis, IMT

IMT- Koushan, Kamran Vortex Induced Vibrations of Free Span Pipelines,

2009-50 PhD thesis, IMT

IMT- Korsvik, Jarl Eirik Heuristic Methods for Ship Routing and
2009-51 Scheduling. PhD-thesis, IMT

IMT- Lee, Jihoon Experimental Investigation and Numerical in

2009-52 Analyzing the Ocean Current Displacement of
Longlines. Ph.d.-Thesis, IMT.

IMT- Vestbøstad, Tone Gran A Numerical Study of Wave-in-Deck Impact usin a

2009-53 Two-Dimensional Constrained Interpolation Profile
Method, Ph.d.thesis, CeSOS.

IMT- Bruun, Kristine Bond Graph Modelling of Fuel Cells for Marine
2009-54 Power Plants. Ph.d.-thesis, IMT

IMT Holstad, Anders Numerical Investigation of Turbulence in a Sekwed

2009-55 Three-Dimensional Channel Flow, Ph.d.-thesis,
IMT.

IMT Ayala-Uraga, Efren Reliability-Based Assessment of Deteriorating

2009-56 Ship-shaped Offshore Structures, Ph.d.-thesis, IMT

IMT Kong, Xiangjun A Numerical Study of a Damaged Ship in Beam

2009-57 Sea Waves. Ph.d.-thesis, IMT/CeSOS.

IMT Kristiansen, David Wave Induced Effects on Floaters of Aquaculture

2010-58 Plants, Ph.d.-thesis, CeSOS.
IMT Ludvigsen, Martin An ROV-Toolbox for Optical and Acoustic
2010-59 Scientific Seabed Investigation. Ph.d.-thesis IMT.

IMT Hals, Jørgen Modelling and Phase Control of Wave-Energy

2010-60 Converters. Ph.d.thesis, CeSOS.

IMT Shu, Zhi Uncertainty Assessment of Wave Loads and

2010- 61 Ultimate Strength of Tankers and Bulk Carriers in a
Reliability Framework. Ph.d. Thesis, IMT/ CeSOS
IMT Shao, Yanlin Numerical Potential-Flow Studies on Weakly-
2010-62 Nonlinear Wave-Body Interactions with/without
Small Forward Speed, Ph.d.thesis,CeSOS.

IMT Califano, Andrea Dynamic Loads on Marine Propellers due to

2010-63 Intermittent Ventilation. Ph.d.thesis, IMT.
IMT El Khoury, George Numerical Simulations of Massively Separated
2010-64 Turbulent Flows, Ph.d.-thesis, IMT

IMT Seim, Knut Sponheim Mixing Process in Dense Overflows with Emphasis
2010-65 on the Faroe Bank Channel Overflow. Ph.d.thesis,
IMT

IMT Jia, Huirong Structural Analysis of Intect and Damaged Ships in

2010-66 a Collission Risk Analysis Perspective. Ph.d.thesis
CeSoS.

IMT Jiao, Linlin Wave-Induced Effects on a Pontoon-type Very

2010-67 Large Floating Structures (VLFS). Ph.D.-thesis,
CeSOS.

IMT Abrahamsen, Bjørn Christian Sloshing Induced Tank Roof with Entrapped Air
2010-68 Pocket. Ph.d.thesis, CeSOS.

IMT Karimirad, Madjid Stochastic Dynamic Response Analysis of Spar-

2011-69 Type Wind Turbines with Catenary or Taut
Mooring Systems. Ph.d.-thesis, CeSOS.

IMT - Erlend Meland Condition Monitoring of Safety Critical Valves.

2011-70 Ph.d.-thesis, IMT.

IMT – Yang, Limin Stochastic Dynamic System Analysis of Wave

2011-71 Energy Converter with Hydraulic Power Take-Off,
with Particular Reference to Wear Damage
Analysis, Ph.d. Thesis, CeSOS.

IMT – Visscher, Jan Application of Particla Image Velocimetry on

2011-72 Turbulent Marine Flows, Ph.d.Thesis, IMT.

IMT – Su, Biao Numerical Predictions of Global and Local Ice

2011-73 Loads on Ships. Ph.d.Thesis, CeSOS.

IMT – Liu, Zhenhui Analytical and Numerical Analysis of Iceberg

2011-74 Collision with Ship Structures. Ph.d.Thesis, IMT.

IMT – Aarsæther, Karl Gunnar Modeling and Analysis of Ship Traffic by

2011-75 Observation and Numerical Simulation.
Ph.d.Thesis, IMT.
Imt – Wu, Jie Hydrodynamic Force Identification from Stochastic
2011-76 Vortex Induced Vibration Experiments with
Slender Beams. Ph.d.Thesis, IMT.

Imt – Amini, Hamid Azimuth Propulsors in Off-design Conditions.

2011-77 Ph.d.Thesis, IMT.

IMT – Nguyen, Tan-Hoi Toward a System of Real-Time Prediction and

2011-78 Monitoring of Bottom Damage Conditions During
Ship Grounding. Ph.d.thesis, IMT.

IMT- Tavakoli, Mohammad T. Assessment of Oil Spill in Ship Collision and

2011-79 Grounding, Ph.d.thesis, IMT.

IMT- Guo, Bingjie Numerical and Experimental Investigation of

2011-80 Added Resistance in Waves. Ph.d.Thesis, IMT.

IMT- Chen, Qiaofeng Ultimate Strength of Aluminium Panels,

2011-81 considering HAZ Effects, IMT

IMT- Kota, Ravikiran S. Wave Loads on Decks of Offshore Structures in

2012-82 Random Seas, CeSOS.

IMT- Sten, Ronny Dynamic Simulation of Deep Water Drilling Risers

2012-83 with Heave Compensating System, IMT.

IMT- Berle, Øyvind Risk and resilience in global maritime supply

2012-84 chains, IMT.

IMT- Fang, Shaoji Fault Tolerant Position Mooring Control Based on

2012-85 Structural Reliability, CeSOS.

IMT- You, Jikun Numerical studies on wave forces and moored ship
2012-86 motions in intermediate and shallow water, CeSOS.

IMT- Xiang ,Xu Maneuvering of two interacting ships in waves,

2012-87 CeSOS

IMT- Dong, Wenbin Time-domain fatigue response and reliability

2012-88 analysis of offshore wind turbines with emphasis on
welded tubular joints and gear components, CeSOS

IMT- Zhu, Suji Investigation of Wave-Induced Nonlinear Load

2012-89 Effects in Open Ships considering Hull Girder
Vibrations in Bending and Torsion, CeSOS

IMT- Zhou, Li Numerical and Experimental Investigation of

2012-90 Station-keeping in Level Ice, CeSOS

IMT- Ushakov, Sergey Particulate matter emission characteristics from

2012-91 diesel enignes operating on conventional and
alternative marine fuels, IMT

IMT- Yin, Decao Experimental and Numerical Analysis of Combined

2013-1 In-line and Cross-flow Vortex Induced Vibrations,
CeSOS
IMT- Kurniawan, Adi Modelling and geometry optimisation of wave
2013-2 energy converters, CeSOS

IMT- Al Ryati, Nabil Technical condition indexes doe auxiliary marine

2013-3 diesel engines, IMT

IMT- Firoozkoohi, Reza Experimental, numerical and analytical

2013-4 investigation of the effect of screens on sloshing,
CeSOS

IMT- Ommani, Babak Potential-Flow Predictions of a Semi-Displacement

2013-5 Vessel Including Applications to Calm Water
Broaching, CeSOS

IMT- Xing, Yihan Modelling and analysis of the gearbox in a floating

2013-6 spar-type wind turbine, CeSOS

IMT-7- Balland, Océane Optimization models for reducing air emissions

2013 from ships, IMT

IMT-8- Yang, Dan Transitional wake flow behind an inclined flat

2013 plate-----Computation and analysis, IMT

IMT-9- Abdillah, Suyuthi Prediction of Extreme Loads and Fatigue Damage

2013 for a Ship Hull due to Ice Action, IMT

IMT-10- Ramìrez, Pedro Agustìn Pèrez Ageing management and life extension of technical
2013 systems-
Concepts and methods applied to oil and gas
facilities, IMT

IMT-11- Chuang, Zhenju Experimental and Numerical Investigation of Speed

2013 Loss due to Seakeeping and Maneuvering. IMT

IMT-12- Etemaddar, Mahmoud Load and Response Analysis of Wind Turbines

2013 under Atmospheric Icing and Controller System
Faults with Emphasis on Spar Type Floating Wind
Turbines, IMT

IMT-13- Lindstad, Haakon Strategies and measures for reducing maritime CO2
2013 emissons, IMT

IMT-14- Haris, Sabril Damage interaction analysis of ship collisions, IMT

2013
IMT-15- Shainee, Mohamed Conceptual Design, Numerical and Experimental
2013 Investigation of a SPM Cage Concept for Offshore
Mariculture, IMT

IMT-16- Gansel, Lars Flow past porous cylinders and effects of

2013 biofouling and fish behavior on the flow in and
around Atlantic salmon net cages, IMT

IMT-17- Gaspar, Henrique Handling Aspects of Complexity in Conceptual

2013 Ship Design, IMT

IMT-18- Thys, Maxime Theoretical and Experimental Investigation of a

2013 Free Running Fishing Vessel at Small Frequency of
Encounter, CeSOS

IMT-19- Aglen, Ida VIV in Free Spanning Pipelines, CeSOS

2013
IMT-1- Song, An Theoretical and experimental studies of wave
2014 diffraction and radiation loads on a horizontally
submerged perforated plate, CeSOS

IMT-2- Rogne, Øyvind Ygre Numerical and Experimental Investigation of a

2014 Hinged 5-body Wave Energy Converter, CeSOS

IMT-3- Dai, Lijuan Safe and efficient operation and maintenance of

2014 offshore wind farms ,IMT

IMT-4- Bachynski, Erin Elizabeth Design and Dynamic Analysis of Tension Leg
2014 Platform Wind Turbines, CeSOS

IMT-5- Wang, Jingbo Water Entry of Freefall Wedged – Wedge motions

2014 and Cavity Dynamics, CeSOS

IMT-6- Kim, Ekaterina Experimental and numerical studies related to the

2014 coupled behavior of ice mass and steel structures
during accidental collisions, IMT

IMT-7- Tan, Xiang Numerical investigation of ship’s continuous- mode

2014 icebreaking in leverl ice, CeSOS

IMT-8- Muliawan, Made Jaya Design and Analysis of Combined Floating Wave
2014 and Wind Power Facilities, with Emphasis on
Extreme Load Effects of the Mooring System,
CeSOS

IMT-9- Jiang, Zhiyu Long-term response analysis of wind turbines with

2014 an emphasis on fault and shutdown conditions, IMT

IMT-10- Dukan, Fredrik ROV Motion Control Systems, IMT

2014
IMT-11- Grimsmo, Nils I. Dynamic simulations of hydraulic cylinder for
2014 heave compensation of deep water drilling risers,
IMT

IMT-12- Kvittem, Marit I. Modelling and response analysis for fatigue design
2014 of a semisubmersible wind turbine, CeSOS

IMT-13- Akhtar, Juned The Effects of Human Fatigue on Risk at Sea, IMT
2014
IMT-14- Syahroni, Nur Fatigue Assessment of Welded Joints Taking into
2014 Account Effects of Residual Stress, IMT

IMT-1- Bøckmann, Eirik Wave Propulsion of ships, IMT

2015

IMT-2- Wang, Kai Modelling and dynamic analysis of a semi-

2015 submersible floating vertical axis wind turbine,
CeSOS

IMT-3- Fredriksen, Arnt Gunvald A numerical and experimental study of a two-

2015 dimensional body with moonpool in waves and
current, CeSOS

IMT-4- Jose Patricio Gallardo Canabes Numerical studies of viscous flow around bluff
2015 bodies, IMT
IMT-5- Vegard Longva Formulation and application of finite element
2015 techniques for slender marine structures subjected
to contact interactions, IMT

IMT-6- Jacobus De Vaal Aerodynamic modelling of floating wind turbines,

2015 CeSOS

IMT-7- Fachri Nasution Fatigue Performance of Copper Power Conductors,

2015 IMT

IMT-8- Oleh I Karpa Development of bivariate extreme value

2015 distributions for applications in marine
technology,CeSOS

IMT-9- Daniel de Almeida Fernandes An output feedback motion control system for
2015 ROVs, AMOS

IMT-10- Bo Zhao Particle Filter for Fault Diagnosis: Application to

2015 Dynamic Positioning Vessel and Underwater
Robotics, CeSOS

IMT-11- Wenting Zhu Impact of emission allocation in maritime

2015 transportation, IMT

IMT-12- Amir Rasekhi Nejad Dynamic Analysis and Design of Gearboxes in

2015 Offshore Wind Turbines in a Structural Reliability
Perspective, CeSOS

IMT-13- Arturo Jesùs Ortega Malca Dynamic Response of Flexibles Risers due to
2015 Unsteady Slug Flow, CeSOS

IMT-14- Dagfinn Husjord Guidance and decision-support system for safe

2015 navigation of ships operating in close proximity,
IMT

IMT-15- Anirban Bhattacharyya Ducted Propellers: Behaviour in Waves and Scale

2015 Effects, IMT

IMT-16- Qin Zhang Image Processing for Ice Parameter Identification

2015 in Ice Management, IMT

IMT-1- Vincentius Rumawas Human Factors in Ship Design and Operation: An

2016 Experiential Learning, IMT

IMT-2- Martin Storheim Structural response in ship-platform and ship-ice

2016 collisions, IMT

IMT-3- Mia Abrahamsen Prsic Numerical Simulations of the Flow around single
2016 and Tandem Circular Cylinders Close to a Plane
Wall, IMT

IMT-4- Tufan Arslan Large-eddy simulations of cross-flow around ship

2016 sections, IMT
IMT-5- Pierre Yves-Henry Parametrisation of aquatic vegetation in hydraulic
2016 and coastal research,IMT

IMT-6- Lin Li Dynamic Analysis of the Instalation of Monopiles

2016 for Offshore Wind Turbines, CeSOS

IMT-7- Øivind Kåre Kjerstad Dynamic Positioning of Marine Vessels in Ice, IMT
2016

IMT-8- Xiaopeng Wu Numerical Analysis of Anchor Handling and Fish

2016 Trawling Operations in a Safety Perspective,
CeSOS

IMT-9- Zhengshun Cheng Integrated Dynamic Analysis of Floating Vertical

2016 Axis Wind Turbines, CeSOS

IMT-10- Ling Wan Experimental and Numerical Study of a Combined

2016 Offshore Wind and Wave Energy Converter
Concept

IMT-11- Wei Chai Stochastic dynamic analysis and reliability

2016 evaluation of the roll motion for ships in random
seas, CeSOS

IMT-12- Øyvind Selnes Patricksson Decision support for conceptual ship design with
2016 focus on a changing life cycle and future
uncertainty, IMT

IMT-13- Mats Jørgen Thorsen Time domain analysis of vortex-induced vibrations,

2016 IMT

IMT-14- Edgar McGuinness Safety in the Norwegian Fishing Fleet – Analysis

2016 and measures for improvement, IMT

IMT-15- Sepideh Jafarzadeh Energy effiency and emission abatement in the

2016 fishing fleet, IMT

IMT-16- Wilson Ivan Guachamin Acero Assessment of marine operations for offshore wind
2016 turbine installation with emphasis on response-
based operational limits, IMT

IMT-17- Mauro Candeloro Tools and Methods for Autonomous Operations on

2016 Seabed and Water Coumn using Underwater
Vehicles, IMT

IMT-18- Valentin Chabaud Real-Time Hybrid Model Testing of Floating Wind

2016 Tubines, IMT

IMT-1- Mohammad Saud Afzal Three-dimensional streaming in a sea bed boundary

2017 layer

IMT-2- Peng Li A Theoretical and Experimental Study of Wave-

2017 induced Hydroelastic Response of a Circular
Floating Collar

IMT-3- Martin Bergström A simulation-based design method for arctic

2017 maritime transport systems
IMT-4- Bhushan Taskar The effect of waves on marine propellers and
2017 propulsion

IMT-5- Mohsen Bardestani A two-dimensional numerical and experimental

2017 study of a floater with net and sinker tube in waves
and current

IMT-6- Fatemeh Hoseini Dadmarzi Direct Numerical Simualtion of turbulent wakes

2017 behind different plate configurations

IMT-7- Michel R. Miyazaki Modeling and control of hybrid marine power

2017 plants

IMT-8- Giri Rajasekhar Gunnu Safety and effiency enhancement of anchor

2017 handling operations with particular emphasis on the
stability of anchor handling vessels

IMT-9- Kevin Koosup Yum Transient Performance and Emissions of a

2017 Turbocharged Diesel Engine for Marine Power
Plants

IMT-10- Zhaolong Yu Hydrodynamic and structural aspects of ship

2017 collisions

IMT-11- Martin Hassel Risk Analysis and Modelling of Allisions between

2017 Passing Vessels and Offshore Installations

IMT-12- Astrid H. Brodtkorb Hybrid Control of Marine Vessels – Dynamic

2017 Positioning in Varying Conditions

IMT-13- Kjersti Bruserud Simultaneous stochastic model of waves and

2017 current for prediction of structural design loads

IMT-14- Finn-Idar Grøtta Giske Long-Term Extreme Response Analysis of Marine

2017 Structures Using Inverse Reliability Methods

IMT-15- Stian Skjong Modeling and Simulation of Maritime Systems and

2017 Operations for Virtual Prototyping using co-
Simulations

IMT-1- Yingguang Chu Virtual Prototyping for Marine Crane Design and
2018 Operations

IMT-2- Sergey Gavrilin Validation of ship manoeuvring simulation models

2018

IMT-3- Jeevith Hegde Tools and methods to manage risk in autonomous

2018 subsea inspection,maintenance and repair
operations

IMT-4- Ida M. Strand Sea Loads on Closed Flexible Fish Cages

2018

IMT-5- Erlend Kvinge Jørgensen Navigation and Control of Underwater Robotic

2018 Vehicles
IMT-6- Bård Stovner Aided Intertial Navigation of Underwater Vehicles
2018

IMT-7- Erlend Liavåg Grotle Thermodynamic Response Enhanced by Sloshing

2018 in Marine LNG Fuel Tanks

IMT-8- Børge Rokseth Safety and Verification of Advanced Maritime

2018 Vessels

IMT-9- Jan Vidar Ulveseter Advances in Semi-Empirical Time Domain

2018 Modelling of Vortex-Induced Vibrations

IMT-10- Chenyu Luan Design and analysis for a steel braceless semi-
2018 submersible hull for supporting a 5-MW horizontal
axis wind turbine

IMT-11- Carl Fredrik Rehn Ship Design under Uncertainty

2018

IMT-12- Øyvind Ødegård Towards Autonomous Operations and Systems in

2018 Marine Archaeology

IMT-13- Stein Melvær Nornes Guidance and Control of Marine Robotics for
2018 Ocean Mapping and Monitoring

IMT-14- Petter Norgren Autonomous Underwater Vehicles in Arctic Marine

2018 Operations: Arctic marine research and ice
monitoring
IMT-15- Minjoo Choi Modular Adaptable Ship Design for Handling
2018 Uncertainty in the Future Operating Context

MT-16- Ole Alexander Eidsvik Dynamics of Remotely Operated Underwater

2018 Vehicle Systems

IMT-17- Mahdi Ghane Fault Diagnosis of Floating Wind Turbine

2018 Drivetrain- Methodologies and Applications

IMT-18- Christoph Alexander Thieme Risk Analysis and Modelling of Autonomous

2018 Marine Systems

IMT-19- Yugao Shen Operational limits for floating-collar fish farms in

2018 waves and current, without and with well-boat
presence
IMT-20- Tianjiao Dai Investigations of Shear Interaction and Stresses in
2018 Flexible Pipes and Umbilicals

IMT-21- Sigurd Solheim Pettersen Resilience by Latent Capabilities in Marine

2018 Systems

IMT-22- Thomas Sauder Fidelity of Cyber-physical Empirical Methods.

2018 Application to the Active Truncation of Slender
Marine Structures

IMT-23- Jan-Tore Horn Statistical and Modelling Uncertainties in the

2018 Design of Offshore Wind Turbines
IMT-24- Anna Swider Data Mining Methods for the Analysis of Power
2018 Systems of Vessels

IMT-1- Zhao He Hydrodynamic study of a moored fish farming cage

2019 with fish influence

IMT-2- Isar Ghamari Numerical and Experimental Study on the Ship

2019 Parametric Roll Resonance and the Effect of Anti-
Roll Tank

IMT-3- Håkon Strandenes Turbulent Flow Simulations at Higher Reynolds

2019 Numbers

IMT-4- Siri Mariane Holen Safety in Norwegian Fish Farming – Concepts and
2019 Methods for Improvement

IMT-5- Ping Fu Reliability Analysis of Wake-Induced Riser

2019 Collision

IMT-6- Vladimir Krivopolianskii Experimental Investigation of Injection and

2019 Combustion Processes in Marine Gas Engines using
Constant Volume Rig

IMT-7- Anna Maria Kozlowska Hydrodynamic Loads on Marine Propellers Subject

2019 to Ventilation and out of Water Condition.

IMT-8- Hans-Martin Heyn Motion Sensing on Vessels Operating in Sea Ice: A

2019 Local Ice Monitoring System for Transit and
Stationkeeping Operations under the Influence of
Sea Ice
IMT-9- Stefan Vilsen Method for Real-Time Hybrid Model Testing of
2019| Ocean Structures – Case on Slender Marine
Systems
IMT-10- Finn-Christian W. Hanssen Non-Linear Wave-Body Interaction in Severe
2019 Waves

IMT-11- Trygve Olav Fossum Adaptive Sampling for Marine Robotics

2019

IMT-12- Jørgen Bremnes Nielsen Modeling and Simulation for Design Evaluation
2019

IMT-13- Yuna Zhao Numerical modelling and dyncamic analysis of

2019 offshore wind turbine blade installation

IMT-14- Daniela Myland Experimental and Theoretical Investigations on the

2019 Ship Resistance in Level Ice

IMT-15- Zhengru Ren Advanced control algorithms to support automated

2019 offshore wind turbine installation

IMT-16- Drazen Polic Ice-propeller impact analysis using an inverse

2019 propulsion machinery simulation approach

IMT-17- Endre Sandvik Sea passage scenario simulation for ship system
2019 performance evaluation
IMT-18- Loup Suja-Thauvin Response of Monopile Wind Turbines to Higher
2019 Order Wave Loads

IMT-19- Emil Smilden Structural control of offshore wind turbines –

2019 Increasing the role of control design in offshore
wind farm development
IMT-20- Aleksandar-Sasa Milakovic On equivalent ice thickness and machine learning
2019 in ship ice transit simulations

IMT-1- Amrit Shankar Verma Modelling, Analysis and Response-based

2020 Operability Assessment of Offshore Wind Turbine
Blade Installation with Emphasis on Impact
Damages

IMT-2- Bent Oddvar Arnesen Autonomous Technology for Inspection,

2020 Haugaløkken Maintenance and Repair Operations in the
Norwegian Aquaculture
IMT-3- Seongpil Cho Model-based fault detection and diagnosis of a
2020 blade pitch system in floating wind turbines

IMT-4- Jose Jorge Garcia Agis Effectiveness in Decision-Making in Ship Design

2020 under Uncertainty

IMT-5- Thomas H. Viuff Uncertainty Assessment of Wave-and Current-

2020 induced Global Response of Floating Bridges

IMT-6- Fredrik Mentzoni Hydrodynamic Loads on Complex Structures in the

2020 Wave Zone

IMT-7- Senthuran Ravinthrakumar Numerical and Experimental Studies of Resonant

2020 Flow in Moonpools in Operational Conditions

IMT-8- Stian Skaalvik Sandøy Acoustic-based Probabilistic Localization and

2020 Mapping using Unmanned Underwater Vehicles for
Aquaculture Operations

IMT-9- Kun Xu Design and Analysis of Mooring System for Semi-

2020 submersible Floating Wind Turbine in Shallow
Water

IMT-10- Jianxun Zhu Cavity Flows and Wake Behind an Elliptic

2020 Cylinder Translating Above the Wall

IMT-11- Sandra Hogenboom Decision-making within Dynamic Positioning

2020 Operations in the Offshore Industry – A Human
Factors based Approach

IMT-12- Woongshik Nam Structural Resistance of Ship and Offshore

2020 Structures Exposed to the Risk of Brittle Failure

IMT-13- Svenn Are Tutturen Værnø Transient Performance in Dynamic Positioning of

2020 Ships: Investigation of Residual Load Models and
Control Methods for Effective Compensation
IMT-14- Mohd Atif Siddiqui Behaviour of a Damaged Ship in Waves:
2020 Experimental and Numerical Hydrodynamic
Analysis of a Damaged Ship Section
IMT-15- John Marius Hegseth Efficient Modelling and Design Optimization of
2020 Large Floating Wind Turbines