University of Massachusetts Amherst

ScholarWorks@UMass Amherst

Masters Theses 1911 - February 2014

2012

Load Reduction of Floating Wind Turbines using Tuned Mass

Dampers
Gordon M. Stewart
University of Massachusetts Amherst

Follow this and additional works at: https://scholarworks.umass.edu/theses

Part of the Acoustics, Dynamics, and Controls Commons, Energy Systems Commons, Ocean
Engineering Commons, and the Structural Engineering Commons

Stewart, Gordon M., "Load Reduction of Floating Wind Turbines using Tuned Mass Dampers" (2012).
Masters Theses 1911 - February 2014. 781.
Retrieved from https://scholarworks.umass.edu/theses/781

This thesis is brought to you for free and open access by ScholarWorks@UMass Amherst. It has been accepted for
inclusion in Masters Theses 1911 - February 2014 by an authorized administrator of ScholarWorks@UMass
Amherst. For more information, please contact [email protected].
LOAD REDUCTION OF FLOATING WIND TURBINES
USING TUNED MASS DAMPERS

A Thesis Presented
by

GORDON M. STEWART

Submitted to the Graduate School of the

University of Massachusetts Amherst in partial fulfillment
of the requirements for the degree of

MASTER OF SCIENCE IN MECHANICAL ENGINEERING

February 2012

Mechanical Engineering
c Copyright by Gordon M. Stewart 2012

All Rights Reserved

 LOAD REDUCTION OF FLOATING WIND TURBINES
USING TUNED MASS DAMPERS

A Thesis Presented

by

GORDON M. STEWART

Approved as to style and content by:

Matthew Lackner, Chair

Sanjay Arwade, Member

James Manwell, Member

Donald Fisher, Department Chair

Mechanical Engineering
 ABSTRACT

LOAD REDUCTION OF FLOATING WIND TURBINES

USING TUNED MASS DAMPERS

FEBRUARY 2012

GORDON M. STEWART

B.Sc., UNIVERSITY OF MASSACHUSETTS AMHERST

M.S.M.E., UNIVERSITY OF MASSACHUSETTS AMHERST

Directed by: Professor Matthew Lackner

Offshore wind turbines have the potential to be an important part of the United

States’ energy production profile in the coming years. In order to accomplish this wind
integration, offshore wind turbines need to be made more reliable and cost efficient

to be competitive with other sources of energy. To capitalize on high speed and high

quality winds over deep water, floating platforms for offshore wind turbines have been
developed, but they suffer from greatly increased loading. One method to reduce loads

in offshore wind turbines is the application of structural control techniques usually

used in skyscrapers and bridges. Tuned mass dampers are one structural control
system that have been used to reduce loads in simulations of offshore wind turbines.

This thesis adds to the state of the art of offshore wind energy by developing a set of

optimum passive tuned mass dampers for four offshore wind turbine platforms and by

quantifying the effects of actuator dynamics on an active tuned mass damper design.
The set of optimum tuned mass dampers are developed by creating a limited

degree-of-freedom model for each of the four offshore wind platforms. These models

iv
are then integrated into an optimization function utilizing a genetic algorithm to

find a globally optimum design for the tuned mass damper. The tuned mass damper

parameters determined by the optimization are integrated into a series of wind turbine
design code simulations using FAST. From these simulations, tower fatigue damage

reductions of between 5 and 20% are achieved for the various TMD configurations.

A previous study developed a set of active tuned mass damper controllers for an

offshore wind turbine mounted on a barge. The design of the controller used an ideal
actuator in which the commanded force equaled the applied force with no time lag.

This thesis develops an actuator model and conducts a frequency analysis on a limited

degree-of-freedom model of the barge including this actuator model. Simulations of

the barge with the active controller and the actuator model are conducted with FAST,

and the results are compared with the ideal actuator case. The realistic actuator

model causes the active mass damper power requirements to increase drastically, by

as much as 1000%, which confirms the importance of considering an actuator model

in controller design.

v
 TABLE OF CONTENTS

Page

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

CHAPTER

1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2. LITERATURE REVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Offshore Wind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1.1 Properties of Offshore Wind Turbines . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1.2 Fixed Bottom Offshore Wind Turbines . . . . . . . . . . . . . . . . . . . . . . . . 8

2.1.2.1 Shallow Water Foundations . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1.2.2 Transitional Depth Foundations . . . . . . . . . . . . . . . . . . . . . 11

2.1.3 Floating Wind Turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.1.3.1 ITI Energy Barge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.1.3.2 OC3-Hywind Spar Buoy . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.3.3 MIT/NREL Tension-Leg Platform . . . . . . . . . . . . . . . . . . . 16
2.1.3.4 Floating Platform Comparison . . . . . . . . . . . . . . . . . . . . . . 17

2.2 Structural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.1 Passive Structural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.1.1 Passive Tuned Mass Dampers . . . . . . . . . . . . . . . . . . . . . . . 19

2.2.1.2 Other Passive Structural Control Designs . . . . . . . . . . . . . 20

2.2.2 Semi-Active Structural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2.3 Active Structural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

vi
 2.2.3.1 Active Mass Damper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.3.2 Hybrid Mass Damper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.3.3 Control Structure Interaction . . . . . . . . . . . . . . . . . . . . . . . 23

2.3 Structural Control in Offshore Wind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3.1 Passive Structural Control in Offshore Wind Turbines . . . . . . . . . . 24

2.3.2 Active Structural Control in Offshore Wind Turbines . . . . . . . . . . . 26

2.4 Problems Addressed and Contributions to the State of the Art . . . . . . . . . 26

3. PASSIVE TUNED MASS DAMPER OPTIMIZATION . . . . . . . . . . . 28

3.1 Limited Degree of Freedom Offshore Turbine Models . . . . . . . . . . . . . . . . . 28

3.1.1 Monopile Limited DOF Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.1.2 Barge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.3 Spar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.4 TLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.2 Model Implementation and Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.2.1 Realistic Model Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2.2 Modeling TMD Position Constraints . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.3 Initial Optimization Attempts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.3.1 Surface Response Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.3.2 Sequential Quadratic Programming Method . . . . . . . . . . . . . . . . . . . 44

3.4 Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.4.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4.2 Results of GA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.3 TLP Optimization Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.5 FAST-SC Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.5.1 Results with Monopile with Passive TMD . . . . . . . . . . . . . . . . . . . . 49

3.5.2 Barge Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.3 Spar Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.4 TLP Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.5.5 Sensitivity Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4. CONTROL STRUCTURE INTERACTION . . . . . . . . . . . . . . . . . . . . . . . 60

4.1 Limited Degree of Freedom Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

vii
 4.2 Frequency Domain Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.2.1 Effect of Gear Ratio on CSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.3 FAST-SC Simulation of Active Control of Wind Turbines with CSI . . . . . 66

4.3.1 Pseudo-Passive Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.3.2 HMD Anlysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5. CONCLUSIONS AND FUTURE WORK . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.1 Other TMD Research Topics for Investigation . . . . . . . . . . . . . . . . . . . . . . . 76

5.1.1 TLCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.1.2 Semi-Active . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.1.3 Redesigned HMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.2 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

APPENDIX: GENETIC ALGORITHM CODE . . . . . . . . . . . . . . . . . . . . . . 78

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

viii
 LIST OF TABLES

Table Page

2.1 Physical Parameters of NREL 5MW Baseline Turbine [20] . . . . . . . . . . . . . 9

2.2 Table showing the parameters of the three floating platforms [21]. . . . . . . 15

3.1 Table showing the results of the genetic algorithm . . . . . . . . . . . . . . . . . . . . 47

3.2 Table showing the results of the monopile simulations . . . . . . . . . . . . . . . . 49

3.3 Table showing the results of the barge simulations . . . . . . . . . . . . . . . . . . . 53

3.4 Table showing the results of the spar buoy simulations with the TMD
in the platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.5 Table showing the results of the spar buoy simulations with the TMD
in the Nacelle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.6 Table showing the results of the TLP simulations with the TMD in
the nacelle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.7 Table showing the results of the TLP simulations with the TMD in
the nacelle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.8 Table showing the results of the TLP simulations with the TMD in
the platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.9 Table showing the results of the TLP simulations with the TMD in
the platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.10 Table showing the results sensitivity study using the barge floating
platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.1 Motor Constants for 505 Frame [25] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.2 Results of simulation with 18m/s wind and pseudo-passive

controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

ix
4.3 Results of simulation with 10m/s wind and pseudo-passive
controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.4 Results of simulation with 18m/s wind and high authority

controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.5 Results of simulation with 18m/s wind and low authority

controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.6 Results of simulation with 10m/s wind and high authority

controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.7 Results of simulation with 10m/s wind and low authority

controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

x
 LIST OF FIGURES

Figure Page

2.1 Onshore wind resource at 80m height for the United States . . . . . . . . . . . . . 5

2.2 Offshore wind resource at 90m height for the United States . . . . . . . . . . . . . 6

2.3 Advances in wind turbine size [11]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.4 Offshore wind turbine terminology [15] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.5 Depth ranges for proposed and existing offshore wind turbine
foundation designs [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.6 Three shallow water wind turbine foundations; the monopile, the
gravity base, and the suction bucket from right to left [31]. . . . . . . . . . 10

2.7 Transitional water depth foundation designs. [31]. . . . . . . . . . . . . . . . . . . . . 11

2.8 Cost comparison of various offshore wind turbine platform designs

[31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.9 Naming and sign conventions for the 6 platform DOFs [17]. . . . . . . . . . . . 13

2.10 Scatter plot showing how different platform designs acheive stability
[5]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.11 Graphic depicting the ITI Energy Barge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.12 Graphic depicting the OC3-Hywind Spar Buoy . . . . . . . . . . . . . . . . . . . . . . 16

2.13 Graphic depicting the MIT/NREL TLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.14 Fatigue damage on various components of the three floating platform

designs [21]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.15 Schematic of a Passive TMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

xi
2.16 Schematic tuned liquid column damper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.17 Pendulum damper from the Taipei101 tower. . . . . . . . . . . . . . . . . . . . . . . . . 21

2.18 Diagram of controllable valve damper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.19 Schematic of electrorheological damper. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.20 Block diagram showing CSI [8] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.21 Diagram showing direction of Fore-Aft and Side-Side TMDs in a

nacelle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1 Diagram of the limited degree-of-freedom model for the monopile. . . . . . . 30

3.2 Diagram of the limited degree-of-freedom model for the barge. . . . . . . . . . 31

3.3 Diagram of the limited degree-of-freedom model for the spar with the
TMD in the nacelle and in the spar. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.4 Diagram of the limited degree-of-freedom model for the surging TLP,
the pitching TLP with the TMD in the platform, and the pitching
TLP with the TMD in the nacelle, from left to right. . . . . . . . . . . . . . . 35

3.5 Diagram showing surge/pitch/heave coupling in TLP due to mooring

lines (Exaggerated for effect) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.6 FAST-SC simulation output showing surge/pitch/heave coupling . . . . . . . 37

3.7 Simulink Model of Limited Degree of Freedom Model . . . . . . . . . . . . . . . . . 38

3.8 Figure showing agreement between limited DOF model and FAST
with no TMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.9 Figure showing an example of a barge limited DOF model

simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.10 Surface plot of standard deviation of tower top displacement vs. TMD
spring and damping constants for the barge with no stops . . . . . . . . . . 42

3.11 Surface Plot of standard deviation of tower top displacement vs.

TMD spring and damping constants for the barge with stops at
±8m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

xii
3.12 Surface plot showing the difference between surfaces with stops and
no stops. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.13 Convergence of fitness through generations . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.14 Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the monopile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3.15 Plot of side-side damage reduction from a side-side TMD in the

nacelle of the monopile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.16 Plot comparing the power spectrum of fore-aft and side-side bending
moments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.17 Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the barge turbine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.18 Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the barge turbine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.19 Plot of fore-aft TMD displacement showing the effect of constant

thrust. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.1 Block diagram showing feedback path [4] . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.2 Figure showing the three transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.3 Bode Plot of motors with different gear ratios . . . . . . . . . . . . . . . . . . . . . . . 66

4.4 Actuator power over time for 18 m/s wind and high control
authority. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

xiii
 CHAPTER 1

INTRODUCTION

Offshore wind turbines have the potential to be a significant contributor to global

energy production, due to the proximity of the high quality wind resource to coastal

energy loads. However, due to the addition of wave and current loads, offshore struc-
tures must be made stronger, and thus more expensive than their land based coun-

terparts. The reliability of offshore turbines suffers due to the higher loading, and the

inaccessibility of the turbines for maintenance compounds this problem. The abil-
ity to reduce loads is therefore extremely important for offshore wind turbines, as it

allows for increased reliability and possibly lighter and cheaper structures [31].

In order to access offshore winds far offshore over deeper water, floating platforms
for wind turbines are being designed and studied. With few water depth and sea floor

restrictions, these platforms could be placed anywhere in the oceans with suitable

electricity transmission. Also, since the platforms can be towed by boats, the wind

turbines could be moved or brought to shore for maintenance. Floating wind turbines,
however, have been shown to experience much higher fatigue and ultimate loading

than onshore or fixed bottom offshore turbines, and could therefore benefit greatly

from load reduction techniques.

One method to reduce loading is to utilize structural control systems, which have

been used successfully in civil structures to achieve improved structural response [32,

40]. For civil structures, the main purpose of structural control systems is increased

inhabitant comfort by reducing building accelerations, but this reduced acceleration

also leads to reduced fatigue loading. In wind turbine applications, fatigue is a design

1
driver. While fixed bottom offshore platforms can benefit from structural control

due to reduced fatigue damage, floating platforms experience increased motion due

to relative platform flexibility and any reduction in these motions should result in
a significant decrease in fatigue. The application of these systems to offshore wind

turbines is the subject of this Master’s research. Specifically, the thesis develops

models and tools to design, analyze and optimize both passive and active structural

control systems for floating and fixed bottom offshore wind turbines.
In the first chapter, a review of the literature explains the background of the

problem. The state of offshore wind research is discussed which will review the offshore

foundations and platforms used in the studies in this thesis. The wind turbine design
code, FAST, is introduced. Civil engineering applications of structural control are

outlined, reviewing the different types of structural control. Finally, the literature on

the integration of structural control and offshore wind turbines is presented.

The second chapter investigates the work on the optimization of a passive tuned
mass damper (TMD) for four offshore platforms. The goal of this chapter is to identify

the best passive system for load reduction. First a limited degree of freedom structural

model of each platform is created, then this model is used as an objective function
for a genetic algorithm. The genetic algorithm is used to find the TMD spring and

damper that provides either the largest tower fatigue reduction, or the largest mooring

line fatigue reduction, depending on the specific design drivers of the platform used.
Once the optimum TMDs are found, a series of high fidelity simulations is run for each

platform using a modification to the National Renewable Energy Laboratory (NREL)

wind turbine design code, FAST. These simulations quantify the improvement that

each TMD design has over the baseline system.

Chapter 3 presents research on the active tuned mass damper. A previous study

had developed an active controller for the barge floating platform, but an actuator

model was excluded from the system identification and simulations [33]. The active

2
structural control literature stresses that inclusion of an actuator model is important

in obtaining a robust controller because the actuator adds dynamics to the system.

Without the actuator model, the original active controller may not be viable, so this
chapter seeks to determine the effect of adding a realistic actuator. After developing a

mathematic model of an electric motor to be used as the actuator, a frequency analysis

is conducted on the system. The frequency analysis helps to provide a thorough

understanding of the problem of control structure interaction. Finally, full degree of

freedom FAST simulations are run for cases with and without the actuator model for

comparison.

3
 CHAPTER 2

LITERATURE REVIEW

This chapter will take an in-depth look at the literature surrounding offshore wind

turbines as well as structural control. Throughout the review, problems or gaps in

the state of the art will be identified and used as motivation for the work done in this
thesis.

2.1 Offshore Wind

As onshore wind reaches a state of relative technical maturity, more offshore wind
farms are being built. The low turbulence, high speed wind resource offshore is

another benefit. Figure 2.1 shows the onshore wind resource map for the United

States. The wind speeds are taken at a height of 80 meters. It can be seen from this

map that most of the onshore wind resource is in the interior of the country, in areas
of low population density and far from many of the major load centers on the coasts.

Figure 2.2 shows the offshore wind resource. Not only are the wind speeds higher, but

the high wind speeds regions are larger and more uniform. Also, the offshore wind
resource is closer to the coastal population, which reduces electricity transmission

distances. Offshore wind turbines also may be able to achieve more efficient designs

due to a higher noise tolerance. Without the stringent noise requirements of onshore
turbines, turbines can have higher tip speed ratios, which in general leads to more

efficient turbine designs.

4
 Figure 2.1. Onshore wind resource at 80m height for the United States

2.1.1 Properties of Offshore Wind Turbines

Wind turbine sizes have been changing dramatically over the past 40 years. Figure
2.3 from the European Wind Energy Association (EWEA) shows the increase in the

size of the diameters of installed wind turbines.

Figure 2.3 covers both on land and offshore wind turbines, and the trend toward
large turbines is more pronounced for offshore turbines. This is due to the costs

associated with constructing and installing the foundations or platforms for offshore

turbines. Figure 2.4 shows the construction of an offshore wind turbine.

This research uses a representative 5 MW wind turbine model developed by NREL.

This is a three bladed upwind machine with a 90m hub height and a 126m rotor

5
 Figure 2.2. Offshore wind resource at 90m height for the United States

diameter [20]. Table 2.1 outlines other properties of the turbine. The baseline control

includes variable speed operation and collective blade pitch control. This turbine is

used for many research efforts as it provides a common model for comparison between
studies.

There are a few additional engineering challenges that must be addressed to make

offshore wind turbines (OWTs) more viable. Both floating and fixed bottom OWTs

require a more expensive foundation than land based turbines. The addition of wave
and current loading on the turbine coupled with the added flexibility of the offshore

floating platform or fixed bottom foundation is an important engineering problem.

With the additional loading and flexibility, the components of OWTs, particularly

6
 Figure 2.3. Advances in wind turbine size [11].

the towers and blades, must be made stronger and heavier, which increases the cost.

Boats are required in order to maintain the components of OWTs, which increases

operation and maintenance costs as well. All of these factors combine to make fixed
bottom offshore wind turbines up to 2-3 times more expensive than onshore wind

turbines, and floating turbines even more expensive. Reducing the effects of the

additional loading on OWTs has the potential to reduce these elevated costs.
For the purposes of this thesis, there are two broad categories of offshore wind

turbine platforms, fixed bottom and floating. Both categories are further divided into

foundation types. Fixed bottom structures include monopiles, jacket foundations,

gravity foundations, and vacuum piles. For floating platforms, the designs include
buoyancy stabilized barges, ballast stabilized spar buoys, and mooring line stabilized

tension-leg platforms. See Figure 2.5 for a depiction of feasible depth ranges for the

7
 Figure 2.4. Offshore wind turbine terminology [15]

various platforms designs [31]. The features of all of these designs will be discussed

in the following sections.

2.1.2 Fixed Bottom Offshore Wind Turbines

All of the power producing offshore wind turbines in the world today use fixed

bottom foundations. These structures are highly dependent on ocean floor conditions,
as well as water depth. Different types of fixed bottom foundations have been devel-

oped in order to expand the locations suitable for installation, which are categorized

8
 Rating 5 MW
Rotor Orientation, Configuration Upwind, 3 Blades
Control Variable Speed, Collective Pitch
Drivetrain High Speed, Multiple-Stage Gearbox
Rotor, Hub Diameter 126 m, 3 m
Hub Height 90 m
Cut-In, Rated, Cut-Out Wind Speed 3 m/s, 11.4 m/s, 25 m/s
Cut-In, Rated Rotor Speed 6.9 rpm, 12.1 rpm
Rated Tip Speed 80 m/s
Overhang, Shaft Tilt, Precone 5 m, 5, 2.5
Rotor Mass 110,000 kg
Nacelle Mass 240,000 kg
Tower Mass 347,460 kg
Coordinate Location of Overall CM (-0.2 m, 0.0 m, 64.0 m)
Nacelle Dimensions 18 m x 6 m x 6 m

Table 2.1. Physical Parameters of NREL 5MW Baseline Turbine [20]

Figure 2.5. Depth ranges for proposed and existing offshore wind turbine foundation
designs [31].

by water depth in this thesis. The following sections will define and give examples

for both shallow water and transitional depth foundations.

9
2.1.2.1 Shallow Water Foundations

Wind turbines that are installed in up to approximately 30 meters of water are con-
sidered shallow water foundations [31]. The main three foundation types for shallow

water are the monopile, the gravity base, and the suction bucket. These foundations

are depicted in Figure 2.6 The monopile consists of a steel hollow tube that is driven

Figure 2.6. Three shallow water wind turbine foundations; the monopile, the gravity
base, and the suction bucket from right to left [31].

down into the seabed, with a transitional piece attaching the pile to the tower of the

turbine. The gravity base uses a large heavy slab, usually concrete, on the ocean
floor for its support. The suction bucket, also known as a suction caisson, uses a

shorter but wider tube than the monopile, but the tube is evacuated of all water after

installation, providing a suction force that gives stability to the turbine. All three of
these foundations are highly dependent on seabed conditions. The monopile needs

the ocean floor to be soft enough to allow the pile to go deep enough, but also firm

enough to provide lateral support under tower loading. Gravity bases have problems

with non-homogeneous soil settling, which could cause the turbine to angle over. The

10
suction caisson needs certain kinds of soil to maintain the partial vacuum that gives

the structure its support.

The most common shallow water foundation that is used today is the monopile due
to its relative simplicity and small environmental footprint. Therefore, a monopile

will be used in this research as the representative fixed bottom foundation. The

specific monopile used is a standard design from NREL that can be simulated by the

FAST turbine design code. This monopile is a 6m diameter hollow steel tube that is
simulated in 20m of water.

2.1.2.2 Transitional Depth Foundations

In transition water depths between 30 and 60 m, a different class of fixed bottom

foundations are used, as the shallow water foundations just discussed are no longer
feasible. There are a multitude of proposed foundations for this depth, which can be

seen in Figure 2.7. These foundations are as follows (from left to right): tripod tower,

Figure 2.7. Transitional water depth foundation designs. [31].

guyed monopole, full-height jacket, submerged jacket with transition to tube tower,
and enhanced suction bucket or gravity base [31]. This research will not utilize any

transitional depth turbines, but they are included here for reference.

11
2.1.3 Floating Wind Turbines

Floating wind turbines have the potential to be placed anywhere in the ocean from
60 meters to upwards of 900 m or beyond. This is a great benefit, because floating

platforms allow offshore wind penetration into places where it may be prohibitive for

fixed bottom offshore turbines. These places include the Great Lakes and the west
coast of the United States where there is little shallow water. Floating platforms are

also much less dependent on seabed conditions than fixed bottom structures because

they do not rely on the ocean floor for support, with mooring line anchors being a

notable exception. Many of the floating platform designs are able to be towed by
boats in order to be moved relatively easily. This may reduce costs associated with

construction and maintenance. See Figure 2.8 for a qualitative chart comparing costs

from different offshore wind turbines.

Figure 2.8. Cost comparison of various offshore wind turbine platform designs [31].

Floating platforms lose the stiffness associated with the fixed ground foundations,
and gain new degrees of freedom. The naming convention for the floating platforms’

degrees of freedom used in this thesis can be seen in Figure 2.9. This thesis uses

three different floating platforms designs. The three major sources of stability for
floating platforms are buoyancy, ballast, and mooring line tension. Each platform uses

12
 Figure 2.9. Naming and sign conventions for the 6 platform DOFs [17].

some combination of these three stability sources, with one source being dominant.

Figure 2.10 shows a diagram of how different platform designs achieve stability. The

Figure 2.10. Scatter plot showing how different platform designs acheive stability
[5].

platforms used in this research are the ITI Energy Barge, the OC3-Hywind Spar Buoy,

13
and the NREL Tension-Leg Platform. As Figure 2.10 shows, the barge depends mostly

on buoyancy, the spar buoy mostly on ballast, and the tension-leg platform mostly

on mooring line tension.

2.1.3.1 ITI Energy Barge

The ITI Energy barge is a floating platform designed by the Department of Naval
Architecture and Marine Engineering at the Universities of Glasgow and Strathclyde

through a contract with ITI Energy [17]. This platform was originally designed to

be used with the NREL 5 MW turbine. The barge has eight catenary mooring lines,
two coming off each corner. The lines are added to tether the barge in place, but also

provide some stiffness. A graphic of the barge with the NREL 5MW mounted on it

can be seen in Figure 2.11 [21]. A table referencing important physical parameters of
the barge is in Table 2.2.

Figure 2.11. Graphic depicting the ITI Energy Barge

There are a few problems with the design of the ITI Energy Barge. The pitching

mode has a natural frequency close to the peak frequency of a typical wave spectrum,

and thus is significantly excited by wave loading. This excitation causes high loading

14
on the tower in the fore-aft direction, which is discussed more in Section 2.1.3.4.

The barge is one of the simplest of the floating platforms to construct, and it is also

relatively easy to tow from a construction site near the shore to the turbine site.
These attributes make it an attractive platform, but the high tower loading must be

addressed.
MIT/NREL TLP OC3-Hywind Spar Buoy ITI Energy Barge
Diameter or width × length 18m 6.5to9.4m 40m × 40m
(is tapered)
Draft 47.89m 120m 4m
Water displacement 12, 180m3 8, 029m3 6, 000m3
Mass, including ballast 8.6 × 106 kg 7.47 × 106 kg 5.45 × 106 kg
CM location below 40.61m 89.92m 0.2818m
still water level (SWL)
Roll inertia about CM 5.72 × 108 kg · m2 4.23 × 109 kg · m2 7.27 × 108 kg · m2
Pitch inertia about CM 5.72 × 108 kg · m2 4.23 × 109 kg · m2 7.27 × 108 kg · m2
Yaw inertia about CM 3.61 × 108 kg · m2 1.64 × 108 kg · m2 1.45 × 109 kg · m2
Number of mooring lines 8(4pairs) 3 8
Depth to fairleads, anchors 47.89m, 200m 70m, 320m 4m, 150m
Radius to fairleads, anchors 27m, 27m 5.2m, 853.9m 28.28m, 423.4m
Unstretched line length 151.7m 902.2m 473.3m
Line diameter 0.127m 0.09m 0.0809m
Line mass density 116kg/m 77.71kg/m 130.4kg/m
Line extensional stiffness 1.5 × 109 N 3.84 × 108 N 5.89 × 108 N

Table 2.2. Table showing the parameters of the three floating platforms [21].

2.1.3.2 OC3-Hywind Spar Buoy

A Norway based company, StatoilHydro, developed a spar buoy design that is

currently supporting a Siemens 2.3 MW turbine in a floating demonstration project,

the Hywind project. NREL modified this design to be compatible with the NREL

5MW turbine, and the result is called the OC3-Hywind Spar Buoy. Figure 2.12 shows
the design, and physical properties of the platform can be seen in Table 2.2. The spar

buoy uses a heavy counterbalance at the base of the spar to move the center of mass

below the center of buoyancy. This creates a restoring moment if the spar is pitched
or rolled. This design lacks stiffness in the yaw DOF however, which could cause

substantial off-axis wind flow.

15
 Figure 2.12. Graphic depicting the OC3-Hywind Spar Buoy

2.1.3.3 MIT/NREL Tension-Leg Platform

The MIT/NREL Tension-Leg Platform (TLP) is a joint design by MIT and NREL.

This platform uses four legs with 2 taut mooring lines on each leg to provide a

restoring force. The central spar part of the TLP is weighted on the bottom, which
adds stiffness and also makes it possible for the platform to be towed without the

lines or turbine attached. A depiction of the platform can be seen in Figure 2.13, and

properties are once again listed in Table 2.2. This platform is very stiff in pitch and
roll due to the tendons, but it lacks stiffness in surge and sway. Mooring line fatigue

and ultimate loads become a driving factor in the design of this platform due to the

potentially catastrophic situation that would result from a mooring line failure.

16
 Figure 2.13. Graphic depicting the MIT/NREL TLP

2.1.3.4 Floating Platform Comparison

All three platforms have relative advantages and disadvantages. The TLP has

better load reduction characteristics, it is more expensive to manufacture and harder

to site due to the anchors needed. The spar buoy has medium performance in loading,
has the largest mass of material involved in construction, but is fairly simple. The

barge has the worst load performance, but is relatively simple to manufacture.

In one study, fatigue loads were compared in identical wind and wave conditions
using FAST [21]. The results from this study can be seen in Figure 2.14. For all

of the platforms, tower fore-aft bending fatigue was at least 50% greater than the

land based turbine. The spar and barge had fatigue damage of up to 2.5 and 8 times

greater than the land turbine, respectively. The side-side tower fatigue loading is

17
Figure 2.14. Fatigue damage on various components of the three floating platform
designs [21].

also higher for the floating platforms. This increase in fatigue damage is the main

motivating factor of the entire thesis, which is aimed at trying to reduce this loading.

2.2 Structural Control

Structural control is the civil engineering discipline that uses dynamic systems to

reduce acceleration and loading in buildings and bridges due to wave and earthquake

forcing. There are many different designs for the systems used to accomplish this goal

ranging from massive pendulums to precisely controlled servomotor mass dampers.

For over twenty years, numerous large-scale active and passive structural control

systems have been implemented for civil structures [1, 2, 22, 34, 36, 37]. The following

section will outline passive, semi-active, and active structural control.

2.2.1 Passive Structural Control

The simplest type of structural control devices are passive, which use no power to

operate. As the structure vibrates, some of the vibrational energy is transfered to the

18
mass of the structural control device and dissapated by the damper. Sections 2.2.1.1

& 2.2.1.2 will outline the designs of passive structural control devices.

2.2.1.1 Passive Tuned Mass Dampers

The most common passive structural control device is the tuned mass damper

(TMD). This device utilizes a mass on an ideally frictionless track. The TMD mass
and the main structure are connected via a spring and dashpot. In the ideal form

of the TMD, both of these components are linear and have a constant spring and

damping constant. The mass and spring are tuned to a system frequency that causes
loading, which results in the TMD mass vibrating at this frequency. The damper

then dissipates energy from the whole system in the form of heat. The theory is

simple, but tuning the spring and damping constants optimally can be difficult. Even
for an idealized one degree of freedom structure, the optimal tuning for the spring

and damper is dictated by a complex function [23]. For structures with more degrees

of freedom and nonlinearities like an offshore wind turbine, there is no analytical

solution for the optimal tuning, and numerical approaches must be used. Figure 2.15
shows a diagram of a tuned mass damper.

Figure 2.15. Schematic of a Passive TMD

19
2.2.1.2 Other Passive Structural Control Designs

Alternative passive devices have been utilized besides the simple mass on a track
just discussed. These include tuned liquid dampers (TLDs), tuned liquid column

dampers (TLCDs), and pendulum dampers. Tuned liquid dampers use the sloshing

of a fluid to provide a force on the structure, while TLCDs improve upon this idea by
using two attached vertical columns of liquid with an orifice between them to provide

the damping force [13, 29]. The difference between the heights of the two liquid

columns provides an equivalent spring force, and the fluid passing through the orifice

provides a damping force. Pendulum dampers use the swinging of a large pendulum
tuned to a certain frequency to provide a counter-force to structural accelerations.

Figure 2.16 & 2.17 show these two dampers.

Figure 2.16. Schematic tuned liquid column damper

2.2.2 Semi-Active Structural Control

Semi-active mass damper (SAMD) utilize a damper that can change its damping

constant during operation. This ability can be used to tune the mass damper on the

fly and can result in better performance compared to passive TMDs with a minimal

energy investment when compared to active dampers (see Section 2.2.3) [34, 38]. The
damper in these systems can take the form of an electrorheological (ER), magne-

torheological (MR), or fluid viscous damper [35–37]. The ER and MR dampers use

20
 Figure 2.17. Pendulum damper from the Taipei101 tower.

either an electric or magnetic field respectively to change the viscosity of the fluid

in the damper. The fluid viscous damper uses a controlled valve to vary the viscous

resistance through the damper orifice [38]. Figure 2.18 shows a fluid viscous damper,
and Figure 2.19 shows an electrorheological damper.

Figure 2.18. Diagram of controllable valve damper

2.2.3 Active Structural Control

Active structural control devices use a controlled actuator in order to apply forces

to the mass and structural and potentially have an even greater impact on structural

acceleration than passive and semi-active systems. Active systems can operate over

21
 Figure 2.19. Schematic of electrorheological damper.

a wider frequency band, and can apply higher forces to the structure by way of the

actuator.

2.2.3.1 Active Mass Damper

An active mass damper (AMD) consists of a mass and an actuator, which can
be actively controlled to apply a force to the mass and an equal and opposite force

on the structure [3, 9, 14, 34]. Since there is no physical spring and damper in this

system, the actuator must provide all of the forces to the mass damper. There is also

the potential to destabilize the system and add energy to the structure if the control
scheme is not well designed.

2.2.3.2 Hybrid Mass Damper

The HMD combines the TMD and AMD, and features both a tuned mass, spring,

and damper system as well as an actuator [35, 37]. With the addition of an actuator,
the HMD gains the potential for improved performance over a passive system. Exam-

ples of installed HMDs utilizing servomotor and hydraulic actuators can be found in

the literature [12, 40]. Both the AMD and HMD can add energy to the system, thus

there is a potential for instability. The HMD, however, includes a passive system, so
it can still provide load reduction with no actuation power.

22
2.2.3.3 Control Structure Interaction

In practical applications of active structural control, it is critical to understand

and account for the dynamics of the actuator when modeling and designing the over-

all system. Control-structure interaction (CSI) refers to the dynamic interaction

between the structure and the actuator in active structural control applications, and
is an unavoidable result of using a real actuator for generating active control forces.

Control-structure interaction exists because there is a natural feedback path between

the structure and the actuator. This feedback can be seen in the block diagram

in Figure2.20 [8]. Note that in addition to the effect of the actuator on the struc-
ture (indicated by “f” in Figure2.20), there is also an influence on the actuator by

the structure. In the past, control systems for structures neglected CSI, which can

severely limit performance and robustness [4, 8].

Figure 2.20. Block diagram showing CSI [8]

23
2.3 Structural Control in Offshore Wind
2.3.1 Passive Structural Control in Offshore Wind Turbines

Research has been conducted on using passive TMDs for wind turbines, especially

for offshore structures due to the larger loading [6, 7, 10, 30, 39]. Earlier studies
focused on fixed bottom structures, but previous work also focused on floating struc-

tures [26, 28]. This research led to the development of FAST-SC, an updated version

of the NREL wind turbine aero-elastic design code, which has the capability to simu-
late both passive and active tuned mass dampers. More details on the capabilities of

the FAST-SC code are discussed below, and can be found in the literature [27, 28, 33].

FAST is a fully coupled aero-hydro-servo-elastic code that simulates the perfor-

mance of wind turbines [19]. It uses Blade Element-Momentum theory (BEM) or

generalized dynamic wake theory to calculate aerodynamic loads, a linear modal rep-

resentation for structural components, and a non-linear hydrodynamic subroutine

that calculates wave loading on the platform for offshore applications [16]. This code
is interfaced through Matlab/Simulink, and a controller can be implemented graphi-

cally with Simulink.

A modification to FAST to accommodate structural control (FAST-SC) was de-

veloped by Lackner and Rotea [28]. This code includes the capability to model two

independent TMDs, one in the fore-aft direction and one in the side-side direction (see

Figure 2.21). The TMDs can be located in the nacelle or the platform. The addition

of locating the TMD in platform is mainly for the spar buoy and TLP platforms, in
which it may be desired to move the TMD into the platform. This layout is attractive

because there is little room in the nacelle for extraneous systems like the TMD, and

since extra mass in the nacelle could create unwanted loading. It may also be feasible
to use a larger TMD in the platform than in the nacelle, which could increase per-

formance. In addition to the spring and damping forces, an active force provided by

an actuator can be applied to the mass [27, 33]. Position constraints known as stops

24
are imposed on the stroke of the TMDs. These constraints were introduced because

the nacelle has a limited amount of space, but the stops can be set to any distance.

Figure 2.21. Diagram showing direction of Fore-Aft and Side-Side TMDs in a

nacelle.

In the research of Lackner et al., a passive TMD was tuned for the ITI Energy
Barge and the monopile. A parametric study was used to find the optimum spring

and damping constants. The spring constant was chosen by finding the system nat-

ural frequency that was dominant in causing tower bending and tuning the TMD
spring constant for a given mass to match that frequency. The damping constant was

then found by running FAST-SC with a range of different values and determining the

decrease in tower bending for each value. From this, the near optimum damping con-
stant was found for that specific spring constant. This approach has a few problems

however. First, it was assumed that the spring constant that matches the system

natural frequency is optimum. Further research shows this is not necessarily the case,

especially in nonlinear systems like a wind turbine [23, 24]. Also, the optimization
scheme is limited to only the damping values selected in the parametric study.

25
2.3.2 Active Structural Control in Offshore Wind Turbines

Previous work [27, 33] developed a set of active controllers for an HMD system
in the NREL 5MW offshore wind turbine [20] using a barge for the platform model.

The HMD was located in the nacelle of the turbine in the fore-aft orientation (see

Fig 2.21), with a mass of 20, 000kg. Using 4% of the rated turbine power for the
actuator, the tower damage load was reduced by up to 20% over the optimal passive

system, and 28-33% compared to the baseline system. However, there are practical

limitations to the spring length, and the turbine HMD has a much longer stroke than

is common in systems in civil structures. For this reason, it may be necessary to use
an AMD configuration rather than an HMD. Also, the past research used an ideal

actuator model. With this model, the force that the controller demanded was applied

instantly to the mass and nacelle, and thus actuator dynamics were ignored.

2.4 Problems Addressed and Contributions to the State of

the Art
After a review of the literature surrounding the use of structural control in offshore
wind, this thesis will address the following important research topics:

• A more thorough analysis and design of a passive tuned mass damper will be

conducted for all four of the offshore platforms: the monopile, barge, spar buoy,
and tension leg platform, discussed above in Section 2.1.3. This research will

seek to identify the optimum configuration of a passive tuned mass damper for

each of respective platforms in an attempt to reduce tower fatigue and mooring

line fatigue. With a comprehensive set of simulations encompassing different

platform designs and TMD locations, masses and orientations, a quantitative

assessment of the relative performance of passive tuned mass dampers will be

made.

26
• The effects of introducing a more realistic actuator model into the current active

control design for Rotea et al will be addressed in Chapter 4 [33]. According to

the literature on control structure interaction, the addition of a realistic actuator

changes how the controller should be designed, and if this is not considered it

could lead to a sub-optimum or worse, an unstable system. This research will

develop and analyze a more realistic simulation of the active controller system

for the barge. The results for this simulation will be compared to previous
results from the active controller, and suggestions will be made for future work

for the controller.

27
 CHAPTER 3

PASSIVE TUNED MASS DAMPER OPTIMIZATION

In this chapter, the work on the optimization of a passive TMD for each of the

four platforms is discussed. First, a limited degree of freedom model is designed for

each platform configuration. Next, an optimization scheme is employed in order to

optimize the spring and damping constants for the TMD. Finally, the optimum TMD

configurations are run through a series of FAST-SC simulations in order to compare

their performance to the baseline case with no TMD.

3.1 Limited Degree of Freedom Offshore Turbine Models

The ideal way to optimize the TMD parameters would be to create a function that

would wrap an optimization scheme around FAST-SC, which would pick a spring and

damping constant, simulate the system in FAST-SC, and then modify the parameters
in order to get closer to the minimum of some output from FAST-SC, for example,

tower fatigue. However, the simulation time for a 10 minute FAST-SC simulation

is approximately 10-30 minutes, and the optimization scheme could potentially need
thousands of function calls to find an optimum. Also, in the case of the offshore

platforms used in this study, there is usually a single system degree of freedom that

is responsible for the most fatigue loading, so FAST-SC is overkill in terms of compu-
tations needed. Therefore, in order to quickly and efficiently find the optimum TMD

configuration for each platform, a limited degree of freedom model is constructed

from the basic equations of motion. For the monopile, barge and spar, minimization

of tower base fatigue loading is the objective function, and for the TLP, both tower

28
fatigue and mooring line fatigue are considered. These models are built to capture

the degrees of freedom for the specific platform that are the source of most of the

loading.

3.1.1 Monopile Limited DOF Model

The monopile is the simplest of the models; there are only 2 degrees of freedom
that are a concern, the tower bending DOF and the TMD DOF. Since the fore-aft

direction has the highest loading from wind and waves, this direction has the highest

tower fatigue damage. For this reason, the following models consider the for-aft
direction, but side-side modeling is possible with minor modifications. The tower is

modeled as an inverted pendulum with the structural stiffness and damping modeled

as a rotary spring and rotary damper at the base of the rigid body, and the TMD
is modeled as a simple mass on a linear track with a linear spring and damper. A

diagram of the model can be seen in Figure 3.1.

In this figure, the k terms are spring constants, the d terms are damping constants,

and the m terms are masses. The t subscripts represent the tower degree of freedom
and the tmd subscripts are for the TMD. The angle that the tower has bent from

vertical is denoted by θt , and the displacement of the TMD from is shown as xtmd .

After applying a simple dynamic analysis as well as small angle approximations to the
two degrees of freedom, Equations 3.1 & 3.2 are found. Small angle approximations

are appropriate because in simulations, none of the platforms exceed 10 degrees of

pitch, even in the heaviest wind and wave loadings.

It θ̈t = mt gRt θt − kt θt − dt θ̇t − ktmd Rtmd (Rtmd θt − xtmd )

− dtmd Rtmd (Rtmd θ̇t − ẋtmd ) − mtmd g(Rtmd θt − xtmd ) (3.1)

mT M D ẍT M D = kT M D (RT M D θt − xT M D ) + dT M D (RT M D θ̇t − ẋT M D )

+ mT M D gθt (3.2)

29
 Figure 3.1. Diagram of the limited degree-of-freedom model for the monopile.

The R terms are the distances from the tower hinge to the center of mass of the

degree of freedom indicated by the subscript. For example, Rtmd is the distance from

the tower hinge to the center of mass of the TMD. All of the degrees of freedoms in
the models are in a global reference frame. Thus, xtmd is not defined relative to the

position in the nacelle, but rather from the global zero which can be seen in Figure 3.1

as the z-axis. If the nacelle has moved 1 meter to the right, and the TMD has moved
1 meter to the right in the nacelle, then xtmd would equal 2 meters. This modeling

choice is made so there are no inertial terms from other degrees of freedom in any of

the equations, which simplifies the implementation of the model (see Section 3.2).

30
3.1.2 Barge

The barge must be modeled with an additional degree of freedom to account for
the compliance of the floating platform. It has been shown in other studies that

the pitching degree of freedom for the barge causes the most tower bending [18].

Therefore, the model includes the tower and TMD degrees of freedom, and also has a
pitching degree of freedom. This model can be seen in Figure 3.2. Equations 3.3-3.5

Figure 3.2. Diagram of the limited degree-of-freedom model for the barge.

show the equations for the platform, tower, and TMD.

Ip θ̈p = − dp θ̇p − kp θp − mp gRp θp + kt (θt − θp ) + dt (θ̇t − θ̇p ) (3.3)

It θ̈t = mt gRt θt − kt (θt − θp ) − dt (θ̇t − θ̇p ) − kT M D RT M D (RT M D θt − xT M D )

− dT M D RT M D (RT M D θ̇t − ẋT M D ) − mT M D g(RT M D θt − xT M D )

(3.4)

mT M D ẍT M D = kT M D (RT M D θt − xT M D ) + dT M D (RT M D θ̇t − ẋT M D ) + mT M D gθt

(3.5)

31
The subscripts and variables in these equations are the same as in the Equations

3.1 & 3.2, with the addition of the “p” subscript for the platform DOF. The spring

constant of the barge, kp , represents a summation of hydrostatic restoring moments

and mooring line stiffness. The barge damping constant, dp , includes many sources of

hydrodynamic damping, including wave radiation and viscous damping. These terms

are non-linear, so the assumption of a linear damping constant adds some inaccuracies

to the model. Once again, the angles and displacement are in absolute coordinates so
the equations include only one inertial term. The R parameters are from the hinge to

the center of mass of the corresponding DOF. In all of the models, the choice of the

hinge point as a reference is arbitrary, but it is convenient because the corresponding

distances from this point are straightforward to calculate from the given dimensions

of the platforms.

3.1.3 Spar

The spar buoy model is very similar to the barge, except with this platform, it

may make sense to put the TMD in the spar itself, which may have to be widened
for the TMD to fit. In order to analyze both the nacelle-based and platform-based

TMDs, two models are developed for the spar buoy. Once again, pitch is the platform

degree of freedom that causes the most tower bending, so it is the DOF included in
the model. Figure 3.3 shows the two models for the spar. There are different sets of

equations for the spar with the TMD in the nacelle and the TMD in the platform.

For the spar with the TMD in the nacelle, the equations are identical to the barge
equations, it is only the parameters themselves that change. With the TMD in the

spar, the equations can be seen in Equations 3.6-3.8.

32
Figure 3.3. Diagram of the limited degree-of-freedom model for the spar with the
TMD in the nacelle and in the spar.

Ip θ̈p = − dp θ̇p − kp θp − mp gRp θp + kt (θt − θp ) + dt (θ̇t − θ̇p )

− ktmd Rtmd (Rtmd θp − xtmd ) − dtmd Rtmd (Rtmd θ̇p − ẋtmd )

+ mtmd g(Rtmd θp − xtmd )) (3.6)

It θ̈t = mt gRt θt − kt (θt − θp ) − dt (θ̇t − θ̇p ) (3.7)

mT M D ẍT M D = kT M D (RT M D θp − xT M D ) + dT M D (RT M D θ̇p − ẋT M D ) + mT M D gθp

(3.8)

33
 Most of the spar buoy’s stiffness in pitch comes from the ballast term, which is

the mp gRp θp term. The spring constant term for the spar buoy represents the effect

of the mooring lines and the flotation. Since the flotation force is in the upward
direction, the spring constant in this case gives a destabilizing moment to the spar,

so it shows up as a negative constant in the equation. The terms in Equation 3.4 &

3.5 that describe the interaction between the tower and the TMD have switched to

terms in Equation 3.6 & 3.8 that describe the interaction between the TMD and the
platform.

3.1.4 TLP

For the tension leg platform, the dominant mode of the system in terms of plat-

form motion is surge; the translational motion into and out of the wind. The TLP
undergoes cycles on the order of a few meters in surge, and it is relatively stiff for

other degrees of freedom, so this platform motion was initially chosen as the modeling

degree of freedom. However, it was found that while the surge motion is the largest in

term of magnitude, it was not the cause of most of the tower bending or mooring line
fatigue. Instead, pitch was identified as the “problem” degree of freedom, so models

for both pitch and surge were built. Furthermore, both a nacelle and platform based

TMD are feasible for the pitch model, while only a platform based TMD makes sense
for the surge DOF. Figure 3.4 shows the three models used.

These three models show the configuration of the TLP model, but the mathemat-

ical model could not be as easily constructed due to a problem with the coupling of
the DOFs of the TLP. Due to the taut mooring lines, this platform is different than

the other platforms because the lines couple the degrees of freedom. For example,

assuming the lines do not stretch (a justifiable assumption since the lines stretch very

little relative to the platform motions), if the TLP surges, it must also pitch and

34
Figure 3.4. Diagram of the limited degree-of-freedom model for the surging TLP,
the pitching TLP with the TMD in the platform, and the pitching TLP with the
TMD in the nacelle, from left to right.

heave to maintain the length of the lines. This relationship can be seen in Figure 3.5,

although it is exaggerated in the figure.

This issue can also be seen in a FAST-SC simulation. In the following simulation,

only the platform degrees of freedom for the TLP were turned on, and there was no

wind or wave loading. The input was a 2 meter surge initial displacement. Figure 3.6
shows surge, pitch and heave from this simulation. The coupling is clear for surge and

pitch, as the surging causes pitching at the same frequency due to the effect described

in Figure 3.5.

When one tries to isolate one platform degree of freedom like in the monopile,
barge, and spar models, the result is an unrealistic model due to the missing coupling

effects. An important consequence of this coupling is that if the platform is forced to

only surge, the mooring lines are stretched more than they would if the platform could
also pitch and heave. This puts large forces on the platform and changes the effective

spring constant of the surging motion, which changes the natural frequency of the

35
Figure 3.5. Diagram showing surge/pitch/heave coupling in TLP due to mooring
lines (Exaggerated for effect)

motion. However, trying to include all of the relevant degrees of freedom results in a

much more complex model and loses the clarity of the single platform DOF models.
For these reasons, a different optimization strategy is used for the TLP that will be

explained in Section 3.4.3.

3.2 Model Implementation and Tuning

With the mathematical models determined, the equations are then implemented

in Simulink. After solving each of the Equations 3.1-3.8 for the acceleration terms,

the equations are put into embedded function blocks. The output of these block is

the degree of freedom acceleration, which can be integrated twice to get velocity and
position. The velocity and position are fed back into the acceleration equation as

inputs. The Simulink model showing these main blocks can be seen in Figure 3.7.

36
 TLP Surge
2

Surge (m)
0

−1

−2
0 100 200 300 400 500 600
time (s)
TLP Pitch
0.04

0.02

Pitch (deg)
0

−0.02

−0.04
0 100 200 300 400 500 600
time (s)
TLP Heave
0.1
Heave (m)

0.05

0
0 100 200 300 400 500 600
time (s)

Figure 3.6. FAST-SC simulation output showing surge/pitch/heave coupling

In order to use the models to optimize the TMD, the model parameters and the
loading must be determined. For all of the platforms, the platform inertia, tower

inertia, platform mass, and tower mass can be determined from the FAST-SC input

files. The distance terms, Rp and Rt are determined by the center of mass location in
relation to the base of the tower, and can be calculated from the platform and tower

input files as well. The platform and tower spring and damping constants, however,

need to be found by comparing the model to FAST-SC outputs. By applying the

same input conditions to the model and a FAST-SC simulation, the outputs can

be compared and a non-linear least squares algorithm can determine the spring and

damping constants that give the best fit for the model. For the simulations, a step

input of tower deflection was used for the monopile, and an initial platform pitch
displacement was used for the barge and spar buoy. No wave or wind loading was

applied, and the only degrees of freedom that were turned on in FAST-SC were the

two or three DOFs in the model. The displacements of the tower and platform degrees
of freedom were used as performance metrics, and the least squares algorithm was

37
 Figure 3.7. Simulink Model of Limited Degree of Freedom Model

used to fit the output of the model to the the output of FAST-SC. All platform models

had a good agreement with the FAST-SC output using this tuning technique. Figure

3.8 shows the a plot of the fit for the barge model as an example.
The barge pitch angle starts at 5◦ and oscillates at a frequency of approximately

0.08 Hz, and the damping attenuates the signal throughout the 100s simulation. The

second plot is the tower top displacement, which is the amount of tower bending in
meters measured at the top of the tower. Tower top displacement (TTD) is relative

to the platform coordinate system, so if the platform is pitched but the tower is not

bent, then the TTD is zero. In this graph, both the tower and barge pitch frequencies

are apparent. The fit between the FAST-SC output and the model output is close,
the differences can be attributed to non-linearities in the barge pitch damper and

spring in FAST-SC.

38
 Barge Pitch Angle
5
State Space Model
FAST Results

Pitch Angle (deg)

0

−5
0 20 40 60 80 100 120
Time (s)

Tower Top Displacement

2
State Space Model
1.5 FAST Results

1
TTD (m)

0.5

−0.5

−1

−1.5
0 20 40 60 80 100 120
Time (s)

Figure 3.8. Figure showing agreement between limited DOF model and FAST with
no TMD

3.2.1 Realistic Model Loading

With all of the parameters identified for the models, the loading must be chosen

for the optimization function. For the actual turbine, the loading is a complex com-
bination of stochastic wind and wave loading. Originally, the model loading for the

optimization was the simple step input used in the parameter identification. How-

ever, it was found that this gave inaccurate results due to the fact that the thrust

force caused by the wind has a non-zero mean. This essentially creates a constant
heeling over for some of the platforms, specifically the spar buoy, which changes the

parameters of the optimum TMD. The solution is to use a combination of a constant

thrust moment on the tower degree of freedom and a pitch step input. The magnitude

39
of the thrust force is the average of the thrust force from a FAST-SC simulation at

rated wind speed.

3.2.2 Modeling TMD Position Constraints

One difficulty in the modeling is how to model the stops. Since there is limited

space in the nacelle and platforms, stops are necessary to prevent the TMD from
exceeding these space requirements. For the NREL 5MW nacelle, the dimensions

dictate that the longest stroke for the TMD is ±8m in the fore-aft direction, and

±2.5m for the side-side TMD. The stops in FAST-SC are modeled as a large spring
and damper that come into contact with the mass at a certain set distance from

the undeflected TMD spring position. This makes the system highly non-linear. In

order to model this in Simulink, a series of if elsethen statements are used. These
statements are inserted in the acceleration equations such that if the mass is hitting

the stop, an additional force is applied to the mass and an equal and opposite force

is applied to the nacelle or tower. This system gave good agreement with FAST-SC.

In Figure 3.9, both the effect of the stops and the non-zero mean thrust loading can
be seen.

In this figure, the barge’s steady state pitch angle is approximately 3◦ because of

the steady thrust loading. Also, the bottom graph shows the TMD hitting the stop
at 8m.

3.3 Initial Optimization Attempts

Once the models for the platforms are built and tuned, the next step is to use the

models to optimize the TMD parameters. In order to view a graphical representation

of the function that was being optimized, a surface response method was employed.

This proved useful to visualize how to approach the problem, but did not prove to be

a viable optimization method.

40
 Barge Pitch
6

Pitch (deg)
4

0
0 10 20 30 40 50 60 70 80 90 100
time (s)
Tower Top Displacement
1.5

1
TTD (m)

0.5

−0.5
0 10 20 30 40 50 60 70 80 90 100
time (s)
TMD Displacement
10
Displacement (m)

−5

−10
0 10 20 30 40 50 60 70 80 90 100
time (s)

Figure 3.9. Figure showing an example of a barge limited DOF model simulation

3.3.1 Surface Response Plots

The surface response plots were created by inputing different combinations of

spring and damping constants and recording the standard deviation of the tower top

displacement, which is an easy to calculate indicator of fatigue. By inputing a range

of evenly spaced spring and damping constants, and plotting this array with the

standard deviation of tower top displacement on a surface plot, the optimum can be

found graphically. Figure 3.10 shows the response of the barge model with a 20,000

kg TMD for various spring and damping constants. For Figure 3.10, no stops were
applied, which makes the system linear. There is a very clear optimum in this plot

which can be seen as the bottom of the valley on the surface. Without stops, this

41
 Barge Optimum− No Stops

0.5

0.48

0.46
Standard Deviation of TTD (m)

0.44

0.42

0.4

0.38

X: 2700 6000
0.36 Y: 4600
Z: 0.3425
5000

0.34 4000
7000
6000 3000
5000
4000 2000
3000
2000
1000 1000
TMD Damping Constant (N−s/m)
TMD Spring Constant (N/m)

Figure 3.10. Surface plot of standard deviation of tower top displacement vs. TMD
spring and damping constants for the barge with no stops

curve is very smooth, however, as soon as stops are added, the surface appears as is
Figure 3.11.

All parameters and loading conditions are the same for these two plots, only the

existence of stops is different. The plot without stops has a clearly defined minimum
at kT M D = 4600N/m, dT M D = 2700N − s/m. The plot with stops enabled has several

local minima, and the global minimum is approximately kT M D = 2700N/m, dT M D =

4700N − s/m. (The swap of values is purely coincidental.) It should be noted that
for low spring and damping, the mass tends to hit the stops more, but for higher

spring and damping, the mass does not hit the stops and the system resembles the

linear case. Figure 3.12 shows the difference of Figure 3.10 & 3.11. For high spring

and damping constants, the difference in the two plots equals zero, which means the
responses are the same with stops or no stops for these cases.

42
 Barge Optimum− +/−8m Stops

0.54

0.52

0.5

0.48
Standard Deviation of TTD (m)

0.46

0.44

0.42

0.4

0.38

0.36

0.34
15000
X: 4700 15000
Y: 2700
Z: 0.3474
10000 10000

5000 5000

0 0
Spring Constant (N/m) Damping Constant (N−s/m)

Figure 3.11. Surface Plot of standard deviation of tower top displacement vs. TMD
spring and damping constants for the barge with stops at ±8m

Regardless of non-linearity, these surface plots could theoretically be created for

all of the models and TMD masses that are desired, and the optimums could be found.

However, this approach has a few problems. One is that it is very computationally

expensive. Each gridpoint on the surface is a function call, and a full surface may
need tens of thousands of gridpoints to be resolved. This could translate to many

hours of processor time. The more important reason, however, is that with the non-

linearity of the problem, it is hard to be sure that the optimum is captured within
the range of spring and damping parameters that are specified. With this in mind, a

more elegant optimization algorithm is described next.

43
 Difference between Surface with stops and without stops

0.14

0.12

0.1
Standard Deviation of TTD (m)

0.08

0.06

0.04

0.02

−0.02

−0.04
15000

15000
10000
10000
5000
5000

Spring Constant (N/m) 0 0

Damping Constant (N−s/m)

Figure 3.12. Surface plot showing the difference between surfaces with stops and
no stops.

3.3.2 Sequential Quadratic Programming Method

Sequential quadratic programming (SQP) is a common iterative steepest descent

method for finding an optimum. Matlab has a built in function that uses this algo-

rithm called fmincon. This algorithm, unfortunately, performs poorly for the cases

with stops. Due to the numerous local minima that can be seen in Figure 3.11, the
algorithm gets stuck in the closest minima to where it is started. This would be true

for all gradient based methods for this problem, so a global optimization technique is

needed.

44
3.4 Genetic Algorithm
The global optimization method that was selected for this research is known as a
genetic algorithm. It is an evolution inspired algorithm that includes mathematical

representations of populations, individual fitness, mutation, and even mating to find

an optimum to a problem. The basic theory is that an initial population of designs

is generated randomly, and the most fit designs are mated together to produce a

new population. This is accomplished by representing designs as binary strings that

contain the “genome” of the individual. Even though this sounds far fetched, the math

behind it is quite simple. However, there are a few caveats. This algorithm suffers
from the same constraint problem as the surface response method; if the optimum is

outside the initial range of values chosen, the algorithm will not find it. Compared to

the surface response method, however, there is much less computation cost associated
with increasing the range of values used, so the search range can be expanded such

that there is high confidence in the algorithm finding the global minimum.

3.4.1 Implementation

The algorithm used is a modification of the open source SpeedyGA written for

Matlab by Keki Burjorjee. This modified code can be seen in AppendixA. The
capability of using multiple design variables is introduced so both the spring and

damping constants can be specified in one individual. The algorithm starts by picking

a population of random binary strings for both the spring and damping constants.
These designs are kept in pairs throughout the algorithm. The binary strings are

simply binary encodings of the integer value of the corresponding spring or damping

constant. The length of these strings put bounds on the variables. For example, if
the length is 15, then the maximum value for the parameter is 215 − 1 = 32, 767.

Once the initial population is chosen, the first step is to check fitness values,

which are the reciprocal of the standard deviation of tower top displacement. The

45
reciprocal is used because genetic algorithms find maximum values, not minimums.

The checking of fitness values is where the majority of the computational time is

taken because the model must be run for each design. Next the algorithm saves the
elite and average fitness, and goes on to the scaling, crossover, and mutation steps.

The scaling used is called sigma-scaling, which normalizes the fitnesses by the

standard deviation, and removes individuals which are more than one standard devi-

ation below the mean of the population. A roulette wheel uniform crossover is used
with a Pcrossover = 0.7. Roulette wheel crossover can be visualized as a weighted

roulette wheel. The more fit individuals have a larger slice of the wheel, and thus

are more likely to be bred in the crossover. Uniform crossover takes random bits
from the strings of both of the parents and swaps them. A mutation probability,

Pmutation = 0.01, per bit is applied, where the algorithm switches 1% of the bits to

the opposite of what it was. The mutation step helps to maintain genetic diversity

in the population. The process is then repeated.

3.4.2 Results of GA

With a population size of 50 individuals, the algorithm converges within approx-

imately 10 generations, as can be seen in Figure3.13.

Maximum and Average Fitness

8500
Average Fitness
Max Fitness

8000

7500

7000
Fitness

6500

6000

5500

5000

4500
0 2 4 6 8 10 12 14 16 18 20
Generation

Figure 3.13. Convergence of fitness through generations

46
 This amounts to 500 function calls, which takes approximately 30 minutes. This

is much more feasible for use than the surface method, which took tens of thousands

of function calls. The genetic algorithm is run on all of the platform models and the
optimum values can be seen in Table 3.1.
Platform TMD Mass (kg) ktmd (N/m) dtmd (N − s/m) TMD Location Stop Dist
Barge 10000 1237 255 Nac ±8m
Barge 20000 2345 1235 Nac ±8m
Barge 40000 5274 10183 Nac ±8m
Monopile 10000 28805 2800 Nac ±8m
Monopile 20000 54274 7414 Nac ±8m
Monopile 40000 98641 19690 Nac ±8m
Spar 10000 54151 3759 Nac ±8m
Spar 20000 101426 10076 Nac ±8m
Spar 40000 183625 26747 Nac ±8m
Spar 100000 157 57395 Plat ±12m
Spar 200000 440 92506 Plat ±12m
Spar 400000 262081 131008 Plat ±12m
TLP 10000 12350 1300 Nac ±8m
TLP 20000 24500 4000 Nac ±8m
TLP 40000 41000 8700 Nac ±8m
TLP 100000 56500 100 Plat ±8m
TLP 200000 115000 200 Plat ±8m
TLP 400000 230000 400 Plat ±8m

Table 3.1. Table showing the results of the genetic algorithm

3.4.3 TLP Optimization Method

For the reasons discussed in Section 3.1.4, building a simple mathematical model

to represent the tension leg platform is very difficult, and it was decided that using
FAST-SC to optimize the TMD is the best option. This is accomplished by a sim-

ple parametric study in which a range of spring constants with a constant damping

constant is run in a full DOF simulation, and the best spring constant is found for

each TMD mass. Next, the spring constants are set to the optimum that was found,
and the damping constant is varied. This technique finds a rough optimum for the

system, but once again there are windowing issues. Since this is a non-linear system,

47
there could be an optimum outside of the range of the parameters that were chosen.

A more thorough (but time consuming) optimization could use FAST-SC itself as the

design function for a global optimization scheme, but for the purposes of this thesis,
the parametric study is used.

3.5 FAST-SC Simulations

In order to evaluate the effectiveness of the optimal TMD designs, a series of

simulations are run to compare baseline cases to the TMD cases. Each case in Table
3.1 was run in FAST-SC for four different wind and wave inputs, as well as a baseline

with no TMD and the same wind and wave inputs. A side-side TMD simulation was

also run for each case using the same parameters. While the optimal parameters found
with the genetic algorithm might not necessarily be optimal for a side-side TMD, it

will at least show what effect the side-side TMD has on the platform. Finally, all the

cases were run with identical side-side and fore-aft TMDs active. This amounts to

236 separate FAST-SC simulations.

The previous work by Lackner [28] considered a 20,000 kg mass for the TMD in the

nacelle of the barge and monopile. This mass was chosen because it is approximately

2% of the total mass of the monopile, which is a mass percentage that is commonly
used in civil structures. For the purposes of this thesis, 10,000 kg, 20,000 kg, and

40,000 kg were chosen for TMDs in the nacelle, which equates to approximately 1%,

2%, and 4% of the monopile based offshore wind turbine. For the simulations with
the TMD in the platform, 100,000 kg, 200,000 kg, and 400,000 kg are used as masses,

as this is approximately 1%, 2%, and 4% of the entire floating structure mass.

When running the simulations for the TLP and spar buoy with the TMD in the

platform, it was discovered that simply adding mass to the platform in the form of a
TMD has some negative effects. The added mass causes the platform to be positioned

lower in the water. This is especially a problem in the case of the TLP, because the

48
lines become more slack. Also, the natural frequencies of the platform change. For

these simulations, the mass and inertia was subtracted from the platform equivalent

to the mass and inertia added by the TMD.

3.5.1 Results with Monopile with Passive TMD

This section analyzes the results of the simulations of the wind turbine using the
monopile foundation and the previously determined optimum TMD constants. Fore-

aft and side-side tower fatigue damage are considered, as well as the 95th percentile

bending moment that the tower experienced, which can be extrapolated to estimate
ultimate loading on the tower. Table 3.2 shows these metrics, as well as the percent

improvement over the baseline simulations with no TMD. A negative value for percent

improvement indicates an increase in that output compared to the baseline.

Results of Monopile Simulations
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 6361 5997 5968 5817 6333 6328 6335 5997 5857 5724
Percent Improvement from Baseline 10 m/s N/A 5.7% 6.2% 8.5% 0.4% 0.5% 0.4% 6.1% 7.9% 10.0%
Fore-Aft Fatigue Damage (kNm) 18 m/s 10630 10100 9948 9769 10641 10662 10678 10100 9973 9821
Percent Improvement from Baseline 18 m/s N/A 5.0% 6.4% 8.1% -0.1% -0.3% -0.5% 4.9% 6.2% 7.6%
Side-Side Fatigue Damage (kNm) 10 m/s 2577 2625 2820 2943 1182 1107 1043 2625 1083 1020
Percent Improvement from Baseline 10 m/s N/A -1.9% -9.5% -14.2% 54.1% 57.0% 59.5% 55.4% 58.0% 60.4%
Side-Side Fatigue Damage (kNm) 18 m/s 7084 7101 5703 6320 2989 2732 2489 7101 2637 2404
Percent Improvement from Baseline 18 m/s N/A -0.2% 19.5% 10.8% 57.8% 61.4% 64.9% 58.9% 62.8% 66.1%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 43367 43060 43152 43133 43368 43390 43524 43060 43196 43384
Percent Improvement from Baseline 10 m/s N/A 0.7% 0.5% 0.5% -0.0% -0.1% -0.4% 0.4% 0.4% -0.0%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 32925 32836 32596 32742 32958 33000 33262 32836 32740 32934
Percent Improvement from Baseline 18 m/s N/A 0.3% 1.0% 0.6% -0.1% -0.2% -1.0% 0.0% 0.6% -0.0%
Side-Side 95th Percentile Load (kNm) 10 m/s 5972 5806 6070 6233 5042 4970 4946 5806 4984 4974
Percent Improvement from Baseline 10 m/s N/A 2.8% -1.6% -4.4% 15.6% 16.8% 17.2% 15.8% 16.5% 16.7%
Side-Side 95th Percentile Load (kNm) 18 m/s 11626 11479 10561 11095 8244 8045 7897 8234 8063 7976
Percent Improvement from Baseline 18 m/s N/A 1.3% 9.2% 4.6% 29.1% 30.8% 32.1% 29.2% 30.6% 31.4%

Table 3.2. Table showing the results of the monopile simulations

A number of interesting conclusions can be drawn from these data. As a general

rule, as the TMD mass increased, so did the load reduction. However, diminishing

returns can be seen in this trend. A plot of the fore-aft tower damage can be seen

in Figure 3.14. The four lines represent the 4 different turbulent wind files used.

Each line has identical wind and wave loading, the only difference is the TMD mass
and parameters. The plots show a sharp drop from the baseline simulation with no

TMD to the 10,000 kg TMD, and after that, the mass improves the amount of fatigue

49
 Fore−Aft Monopile Tower Damage Equivalent Loads with Fore−Aft TMD
11000
10 m/s Wind 1
10 m/s Wind 2
18 m/s Wind 1
18 m/s Wind 2
10000

9000

Damage (kNm) 8000

7000

6000

5000
0 0.5 1 1.5 2 2.5 3 3.5 4
TMD Mass (kg) x 10
4

Figure 3.14. Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the monopile.

damage only slightly. This trend can be seen more sharply in the side-side fatigue
graph with a side-side TMD, Figure 3.15. The addition of a 10,000 kg side-side

TMD reduces side-side damage by approximately 55%, and the 40,000 kg acheives

approximately a 60% reduction.

Although not specifically designed to reduce ultimate loads, the TMD in the
monopile reduces the 95th percentile bending moment by approximately 1% in the

fore-aft direction, and as much as 32% for side-side loads. This is due to the fact

that the TMD reduces the amplitude of the bending moment, which reduces ultimate
loads as well as fatigue.

It is also interesting that for both fatigue and ultimate loading, the TMD had a

better effect in the side-side direction than the fore-aft. This is most likely because

the excitation in the side-side direction is mostly caused by degree of freedom coupling
as there is no direct loading in this direction. This results in the side-side bending

to have most of its energy in the tower vibrational frequency (see Figure 3.16). In

50
 Side−Side Monopile Tower Damage Equivalent Loads with Side−Side TMD
8000
10 m/s Wind 1
10 m/s Wind 2
18 m/s Wind 1
7000 18 m/s Wind 2

6000

Damage (kNm)
5000

4000

3000

2000

1000
0 0.5 1 1.5 2 2.5 3 3.5 4
TMD Mass (kg) x 10
4

Figure 3.15. Plot of side-side damage reduction from a side-side TMD in the nacelle
of the monopile.

Figure 3.16 the peak at .28 Hz is the tower vibrational mode. Since the TMD is tuned

for this tower vibrational frequency, the TMD is able to damp out much more of the
energy in the side-side direction.

In FAST-SC, the wave forcing direction is in the same direction as the wind.

With realistic loading conditions, this may not always be the case, as there is some
lag between wind and wave forcing. Therefore, this large side-side reduction seen

in the FAST-SC simulations may not be a realistic reduction. Further study using

misaligned wind and wave forcing is needed to quantify this effect.

3.5.2 Barge Results

The wind and wave inputs that were used for the monopile simulation are also
used for the barge simulations so that comparisons can be drawn between the two.

The table showing both the fatigue damage and the 95th percentile load can be seen

in Table 3.3. The trend of diminishing returns is not as apparent for the barge as

51
 8 PSD of Fore−Aft Tower Bending Moment
x 10
7

Power
4

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Frequency (Hz)

9 PSD of Side−Side Tower Bending Moment

x 10
2

1.5
Power

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Frequency (Hz)

Figure 3.16. Plot comparing the power spectrum of fore-aft and side-side bending
moments.

the monopile. Figure 3.17 shows the fore-aft tower bending damage with a fore-aft

TMD. When compared with Figure 3.14, the slopes of the lines is roughly linear, and
does not show the diminishing returns effect seen in the monopile simulations. The

ultimate loads for the fore-aft direction, Figure 3.18, were more substantially reduced

for the barge than the monopile. Fore-aft 95th percentile loads were reduced by up

to 12% from the baseline. Once again, the side-side loads were substantially reduced
by the side-side TMD. The reduction in side-side fatigue loads is up to 30% by the

TMDs used in these simulations.

3.5.3 Spar Results

Two sets of simulations were run for the spar buoy. One uses a nacelle based
TMD like the monopile and barge simulations, and the other uses a TMD in the

spar itself. A few problems with the spar buoy TMDs were discovered in the course

of the research. One is that since the pitching mode of the spar buoy has a low

52
 Results of Barge Simulations
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 33684 31699 29984 28424 33881 34101 34647 31699 30338 28879
Percent Improvement from Baseline 10 m/s N/A 5.9% 11.0% 15.6% -0.6% -1.2% -2.9% 5.8% 9.9% 14.3%
Fore-Aft Fatigue Damage (kNm) 18 m/s 54230 52109 50354 49267 54473 54739 55131 52109 51689 46516
Percent Improvement from Baseline 18 m/s N/A 3.9% 7.1% 9.2% -0.4% -0.9% -1.7% 2.8% 4.7% 14.2%
Side-Side Fatigue Damage (kNm) 10 m/s 8111 8341 7869 8869 8664 8236 6997 8341 7408 6655
Percent Improvement from Baseline 10 m/s N/A -2.8% 3.0% -9.3% -6.8% -1.5% 13.7% -2.5% 8.7% 17.9%
Side-Side Fatigue Damage (kNm) 18 m/s 17001 15440 14728 15888 15601 15231 12887 15440 13349 11427
Percent Improvement from Baseline 18 m/s N/A 9.2% 13.4% 6.5% 8.2% 10.4% 24.2% 11.8% 21.5% 32.8%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 67974 66239 64078 63152 71640 72178 73531 66239 65403 65337
Percent Improvement from Baseline 10 m/s N/A 2.6% 5.7% 7.1% -5.4% -6.2% -8.2% 2.0% 3.8% 3.9%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 83259 80687 77836 73126 84515 84869 86480 80687 79398 74638
Percent Improvement from Baseline 18 m/s N/A 3.1% 6.5% 12.2% -1.5% -1.9% -3.9% 2.3% 4.6% 10.4%
Side-Side 95th Percentile Load (kNm) 10 m/s 11049 10829 10626 11702 11143 10550 9353 10829 9810 9156
Percent Improvement from Baseline 10 m/s N/A 2.0% 3.8% -5.9% -0.8% 4.5% 15.3% 4.1% 11.2% 17.1%
Side-Side 95th Percentile Load (kNm) 18 m/s 21821 20991 20011 21819 19487 18495 16491 18834 17246 15286
Percent Improvement from Baseline 18 m/s N/A 3.8% 8.3% 0.0% 10.7% 15.2% 24.4% 13.7% 21.0% 29.9%

Table 3.3. Table showing the results of the barge simulations

4 Fore−Aft Barge Tower Damage Equivalent Loads with Fore−Aft TMD

x 10
6
10 m/s Wind 1
10 m/s Wind 2
18 m/s Wind 1
5.5 18 m/s Wind 2

5
Damage (kNm)

4.5

3.5

2.5
0 0.5 1 1.5 2 2.5 3 3.5 4
TMD Mass (kg) x 10
4

Figure 3.17. Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the barge turbine.

frequency, requiring a low TMD spring constant. This, in turn, makes the TMD

more susceptible to gravity forcing, as there is not a strong enough spring. This
would not be a problem with a translational system like a civil structure, but for a

pitching system, the gravity forcing on the TMD can not be ignored. This problem

is exacerbated by the thrust forcing of the wind. The constant component of the

53
 4 Fore−Aft Barge Tower 95% Loads with Fore−Aft TMD
x 10
8.5
10 m/s Wind 1
10 m/s Wind 2
18 m/s Wind 1
18 m/s Wind 2

95% Bending Moment (kNm)

7.5

6.5

6
0 0.5 1 1.5 2 2.5 3 3.5 4
TMD Mass (kg) x 10
4

Figure 3.18. Plot of fore-aft damage reduction from a fore-aft TMD in the nacelle
of the barge turbine.

wind forcing causes the spar to be pitched backward, which in turn forces the TMD

against the downwind stop. Figure 3.19 shows the displacement of the TMD during
a representative simulation to show the effect of constant thrust.

Another effect of the TMD resting against the stop is that when the TMD is

displaced, it creates a gravity moment on the spar buoy. In Equation 3.6, the term
+mtmd g(Rtmd θp − xtmd ) encompasses this effect. The TMD mass exerts a vertical

gravity force on the spar, and when the TMD is off center, this force turns into

a moment, with the moment arm being the distance that the TMD has displaced
relative to the spar coordinate system. This term scales with the mass of the TMD

and the amount of TMD deflection, so when there is a heavy TMD that is deflected

as far as possible, which is the case in the spar, this term becomes a non-negligible

destabilizing moment.
The TMD parameters are selected with the gravitational moment integrated since

this term is included in the model that the genetic algorithm optimized. For the

54
 Displacement of Fore−Aft TMD
14

12

Displacement from neutral spring position (m)

10

−2
0 100 200 300 400 500 600
Time (s)

Figure 3.19. Plot of fore-aft TMD displacement showing the effect of constant
thrust.

400,000 kg mass, this term became too large, and the genetic algorithm chose to
maximize the spring and damping constants to force the TMD to stay near the center

of the spar, which essentially eliminates the moment arm, and thus, the destabilizing

moment. For this reason, the 400,000 kg TMD results are essentially the same as not
having a TMD at all, and can be disregarded.

Even with all of these problems, there was improvement in fatigue and ultimate

loads in some simulations, which can be seen in Table 3.4. The 18 m/s average wind
speed simulations in particular show reductions in tower fatigue and ultimate loads.

This is because there is less rotor thrust in these simulations, so the spar pitches less,

and the effect of the gravity moment is diminished.

Table 3.5 show the tower bending fatigue of the spar buoy with the TMD moved
to the nacelle. This TMD is tuned to the tower first bending mode, which has a

much higher frequency than the pitching mode, so the spring constant is higher. This

55
 Results of Spar Simulations with the TMD in the Platform
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 100,000 kg 200,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 15319 15217 15076 15360 15263 15318 15224 15217 14986 15483
Percent Improvement from Baseline 10 m/s N/A 0.7% 1.6% -0.3% 0.4% 0.0% 0.6% 1.9% 2.2% -1.1%
Fore-Aft Fatigue Damage (kNm) 18 m/s 25048 23277 22302 23586 23775 24974 25024 23277 23899 30169
Percent Improvement from Baseline 18 m/s N/A 7.1% 11.0% 5.8% 5.1% 0.3% 0.1% -2.4% 4.6% -20.4%
Side-Side Fatigue Damage (kNm) 10 m/s 4683 4683 4642 4381 4679 5069 4735 4683 4387 4206
Percent Improvement from Baseline 10 m/s N/A 0.0% 0.9% 6.5% 0.1% -8.2% -1.1% 4.2% 6.3% 10.2%
Side-Side Fatigue Damage (kNm) 18 m/s 7043 7118 6951 6156 6717 8785 6807 7118 7660 6672
Percent Improvement from Baseline 18 m/s N/A -1.1% 1.3% 12.6% 4.6% -24.7% 3.4% 1.3% -8.8% 5.3%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 53595 55060 56368 52718 53065 52484 51670 55060 51540 46554
Percent Improvement from Baseline 10 m/s N/A -2.7% -5.2% 1.6% 1.0% 2.1% 3.6% 2.5% 3.8% 13.1%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 62515 59749 59148 59850 57931 62462 59536 59749 59240 65129
Percent Improvement from Baseline 18 m/s N/A 4.4% 5.4% 4.3% 7.3% 0.1% 4.8% 0.6% 5.2% -4.2%
Side-Side 95th Percentile Load (kNm) 10 m/s 7563 7437 7332 7129 9970 13373 7373 7437 10433 6683
Percent Improvement from Baseline 10 m/s N/A 1.7% 3.1% 5.7% -31.8% -76.8% 2.5% -18.1% -37.9% 11.6%
Side-Side 95th Percentile Load (kNm) 18 m/s 12223 12078 11718 10961 15202 22487 11618 15243 17213 11426
Percent Improvement from Baseline 18 m/s N/A 1.2% 4.1% 10.3% -24.4% -84.0% 5.0% -24.7% -40.8% 6.5%

Table 3.4. Table showing the results of the spar buoy simulations with the TMD in
the platform

reduces the gravitational effect that were seen in the platform based TMD simulations.

Results of Spar Simulations with the TMD in the Nacelle

Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 15319 15026 14647 14034 15261 14985 14449 15026 14447 13670
Percent Improvement from Baseline 10 m/s N/A 1.9% 4.4% 8.4% 0.4% 2.2% 5.7% 2.2% 5.7% 10.8%
Fore-Aft Fatigue Damage (kNm) 18 m/s 25048 25290 24406 25859 24798 24758 24141 25290 24939 28254
Percent Improvement from Baseline 18 m/s N/A -1.0% 2.6% -3.2% 1.0% 1.2% 3.6% 4.0% 0.4% -12.8%
Side-Side Fatigue Damage (kNm) 10 m/s 4683 3946 3662 3055 4415 3893 3387 3946 3392 2902
Percent Improvement from Baseline 10 m/s N/A 15.7% 21.8% 34.8% 5.7% 16.9% 27.7% 15.1% 27.6% 38.0%
Side-Side Fatigue Damage (kNm) 18 m/s 7043 5809 5654 6011 6367 6124 5441 5809 5057 5136
Percent Improvement from Baseline 18 m/s N/A 17.5% 19.7% 14.6% 9.6% 13.0% 22.7% 17.3% 28.2% 27.1%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 53595 54334 55042 56255 54644 55686 58110 54334 57192 61412
Percent Improvement from Baseline 10 m/s N/A -1.4% -2.7% -5.0% -2.0% -3.9% -8.4% -3.5% -6.7% -14.6%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 62515 63727 62086 65036 63042 60855 63219 63727 63439 74262
Percent Improvement from Baseline 18 m/s N/A -1.9% 0.7% -4.0% -0.8% 2.7% -1.1% 1.9% -1.5% -18.8%
Side-Side 95th Percentile Load (kNm) 10 m/s 7563 7191 7118 6953 7366 7365 7545 7191 7359 7781
Percent Improvement from Baseline 10 m/s N/A 4.9% 5.9% 8.1% 2.6% 2.6% 0.2% 2.9% 2.7% -2.9%
Side-Side 95th Percentile Load (kNm) 18 m/s 12223 12054 11981 12511 12128 12681 12278 12201 11710 14007
Percent Improvement from Baseline 18 m/s N/A 1.4% 2.0% -2.4% 0.8% -3.7% -0.4% 0.2% 4.2% -14.6%

Table 3.5. Table showing the results of the spar buoy simulations with the TMD in
the Nacelle

These results show good fatigue reduction in both the fore-aft and side-side di-

rections, but little ultimate load reduction. In most simulations, the ultimate loads

were actually increased.

3.5.4 TLP Results

As discussed in Section 3.1.4, the TMD tuned to the surge degree of freedom
did reduce platform surge, but this had little effect on mooring line fatigue or tower

56
bending. Therefore, a TMD tuned to the faster pitching mode was developed. Like

the spar buoy, this platform is compatible with both a nacelle based and platform

based TMD. Table 3.6 shows the tower loading results from the simulations with the
TMD in the nacelle.
Results of TLP Simulations with the TMD in the Nacelle
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 13926 13425 13110 12837 14159 14515 15210 13425 13271 13411
Percent Improvement from Baseline 10 m/s N/A 3.6% 5.9% 7.8% -1.7% -4.2% -9.2% 3.8% 4.7% 3.7%
Fore-Aft Fatigue Damage (kNm) 18 m/s 23370 22923 22674 22566 23793 24278 25311 22923 23261 23769
Percent Improvement from Baseline 18 m/s N/A 1.9% 3.0% 3.4% -1.8% -3.9% -8.3% 1.3% 0.5% -1.7%
Side-Side Fatigue Damage (kNm) 10 m/s 3392 3481 3491 3398 3118 2903 2804 3481 2836 2675
Percent Improvement from Baseline 10 m/s N/A -2.6% -2.9% -0.2% 8.1% 14.4% 17.3% 8.9% 16.4% 21.1%
Side-Side Fatigue Damage (kNm) 18 m/s 7394 7384 7445 7958 7020 6759 6626 7384 6356 6110
Percent Improvement from Baseline 18 m/s N/A 0.1% -0.7% -7.6% 5.1% 8.6% 10.4% 8.3% 14.0% 17.4%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 68827 68606 68588 68739 69073 69425 70173 68606 68922 69392
Percent Improvement from Baseline 10 m/s N/A 0.3% 0.3% 0.1% -0.4% -0.9% -2.0% 0.1% -0.1% -0.8%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 55182 55032 54940 54964 55639 55982 57128 55032 55434 56282
Percent Improvement from Baseline 18 m/s N/A 0.3% 0.4% 0.4% -0.8% -1.5% -3.5% 0.3% -0.5% -2.0%
Side-Side 95th Percentile Load (kNm) 10 m/s 6592 6631 6703 6682 6381 6231 6181 6631 6185 6139
Percent Improvement from Baseline 10 m/s N/A -0.6% -1.7% -1.4% 3.2% 5.5% 6.2% 3.4% 6.2% 6.9%
Side-Side 95th Percentile Load (kNm) 18 m/s 11938 11892 12150 12865 11746 11486 11244 11524 11166 11012
Percent Improvement from Baseline 18 m/s N/A 0.4% -1.8% -7.8% 1.6% 3.8% 5.8% 3.5% 6.5% 7.8%

Table 3.6. Table showing the results of the TLP simulations with the TMD in the
nacelle

Fore-aft bending is reduced by up to 8%, and side-side bending up to 20%. For

the TLP, the side-side TMD negatively impacts the fore-aft bending, and the fore-aft

TMD causes more damage to the side-side tower degree of freedom. This effect was
seen for the other platforms, but not to the extent that it occurs for the TLP.

In addition to tower loads, this TMD is also targeted towards reducing mooring

line loading. Table 3.7 shows the fatigue damage and 95th percentile tensile load on

the mooring line with the highest loading. Mooring line fatigue damage is reduced
Results of TLP Simulations with the TMD in the Nacelle
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg 10,000 kg 20,000 kg 40,000 kg
Line Fatigue Damage (kN) 10 m/s 372 364 357 347 388 388 389 364 363 350
Percent Improvement from Baseline 10 m/s N/A 2.2% 4.0% 6.8% -4.3% -4.3% -4.6% -0.5% 2.3% 5.9%
Line Fatigue Damage (kN) 18 m/s 599 588 578 563 601 603 609 588 577 563
Percent Improvement from Baseline 18 m/s N/A 1.9% 3.4% 6.1% -0.4% -0.7% -1.7% 2.2% 3.6% 6.1%
Line 95th Percentile Load (kN) 10 m/s 4989 4983 4977 4975 5096 5083 5066 4983 5050 4997
Percent Improvement from Baseline 10 m/s N/A 0.1% 0.2% 0.3% -2.2% -1.9% -1.5% -1.8% -1.2% -0.2%
Line 95th Percentile Load (kN) 18 m/s 4916 4912 4903 4889 5019 5009 4988 4912 4977 4914
Percent Improvement from Baseline 18 m/s N/A 0.1% 0.3% 0.5% -2.1% -1.9% -1.5% -1.8% -1.2% 0.0%

Table 3.7. Table showing the results of the TLP simulations with the TMD in the
nacelle

57
by up to 7% with a fore-aft TMD, but the ultimate load is virtually unchanged.

Since the platform in the TLP has a 45 m draft, there is both space and a suitable

moment arm to encourage trying a TMD in the platform. The 45 m moment arm
is less than the nacelle’s 90 m distance to the water level, but the platform based

TMD can use a larger mass. Table 3.8 shows the effect of the platform-based TMD

on tower loading. This configuration performed poorly for reducing tower bending.

This is most likely because the TMD cannot directly apply forces to the tower like
the TMD in the nacelle can.
Results of TLP Simulations with the TMD in the Platform
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 100,000 kg 20,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 13926 13883 13797 13822 18083 24167 26605 13883 23907 26099
Percent Improvement from Baseline 10 m/s N/A 0.3% 0.9% 0.7% -29.9% -73.5% -91.1% -29.2% -71.7% -87.4%
Fore-Aft Fatigue Damage (kNm) 18 m/s 23370 23267 23293 23483 29095 35288 41007 23267 35010 41421
Percent Improvement from Baseline 18 m/s N/A 0.4% 0.3% -0.5% -24.5% -51.0% -75.5% -23.9% -49.8% -77.2%
Side-Side Fatigue Damage (kNm) 10 m/s 3392 3414 3467 3495 4444 4773 6460 3414 4780 6449
Percent Improvement from Baseline 10 m/s N/A -0.6% -2.2% -3.0% -31.0% -40.7% -90.4% -30.9% -40.9% -90.1%
Side-Side Fatigue Damage (kNm) 18 m/s 7394 7371 7252 7165 9387 10953 11308 7371 10990 11245
Percent Improvement from Baseline 18 m/s N/A 0.3% 1.9% 3.1% -27.0% -48.1% -52.9% -26.8% -48.6% -52.1%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 68827 68773 68800 68694 72656 78503 84382 68773 78420 83953
Percent Improvement from Baseline 10 m/s N/A 0.1% 0.0% 0.2% -5.6% -14.1% -22.6% -5.6% -13.9% -22.0%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 55182 55131 54916 55265 60421 67106 75552 55131 66838 76205
Percent Improvement from Baseline 18 m/s N/A 0.1% 0.5% -0.2% -9.5% -21.6% -36.9% -9.2% -21.1% -38.1%
Side-Side 95th Percentile Load (kNm) 10 m/s 6592 6570 6602 6683 7838 8429 11027 6570 8454 10996
Percent Improvement from Baseline 10 m/s N/A 0.3% -0.2% -1.4% -18.9% -27.9% -67.3% -19.0% -28.2% -66.8%
Side-Side 95th Percentile Load (kNm) 18 m/s 11938 11949 11892 11846 14113 16492 17147 14116 16530 17021
Percent Improvement from Baseline 18 m/s N/A -0.1% 0.4% 0.8% -18.2% -38.1% -43.6% -18.2% -38.5% -42.6%

Table 3.8. Table showing the results of the TLP simulations with the TMD in the
platform

Table 3.9 shows the effect of the TMD in the platform on mooring line loading.

The TMD in the nacelle had better performance, with the platform based TMD only

achieving a 1.3% reduction. This is due to the shorter moment arm combined with
the gravitational moment effect described in Section 3.5.3

3.5.5 Sensitivity Study

In order to investigate the robustness of the optimum found by the genetic algo-

rithm, a sensitivity study was conducted to quantify the change in fatigue reduction

with a change in the TMD parameters. The barge platform was used for this study.
FAST-SC was run for with a TMD spring constant that is 10% higher and 10% lower

58
 Results of TLP Simulations with the TMD in the Platform
Fore-Aft TMD Side-Side TMD Fore-Aft and Side-Side TMDs
Baseline 100,000 kg 200,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg 100,000 kg 200,000 kg 400,000 kg
Fore-Aft Fatigue Damage (kNm) 10 m/s 372 370 368 367 398 433 400 370 427 390
Percent Improvement from Baseline 10 m/s N/A 0.5% 0.9% 1.3% -7.1% -16.4% -7.6% -6.5% -14.9% -4.9%
Fore-Aft Fatigue Damage (kNm) 18 m/s 599 595 594 592 629 636 598 595 630 598
Percent Improvement from Baseline 18 m/s N/A 0.6% 0.9% 1.2% -5.0% -6.3% 0.1% -4.3% -5.1% 0.1%
Fore-Aft 95th Percentile Load (kNm) 10 m/s 4989 4988 4985 4983 5015 5062 5054 4988 5060 5052
Percent Improvement from Baseline 10 m/s N/A 0.0% 0.1% 0.1% -0.5% -1.5% -1.3% -0.5% -1.4% -1.3%
Fore-Aft 95th Percentile Load (kNm) 18 m/s 4916 4913 4907 4901 4933 4969 4928 4913 4955 4930
Percent Improvement from Baseline 18 m/s N/A 0.1% 0.2% 0.3% -0.4% -1.1% -0.2% -0.3% -0.8% -0.3%

Table 3.9. Table showing the results of the TLP simulations with the TMD in the
platform

than the optimum found with the genetic algorithm, and a TMD damping constant

that is 10% higher and lower than the optimum. The change in fatigue damage from

the optimum for these cases is shown in Table 3.10.

Percentage Increase in Fore-Aft Tower Fatigue Damage from Optimum

TMD Mass (kg) +10% Damping Constant -10% Damping Constant +10% Spring Constant -10% Spring Constant
10,000 0.91 0.55 1.03 -0.06
20,000 0.38 0.63 1.52 -1.16
40,000 0.17 0.77 1.24 -0.32

Table 3.10. Table showing the results sensitivity study using the barge floating
platform

In Table 3.10, it can be seen that in general, a 10% change in TMD parameters
results in a much smaller change in tower fatigue damage, which means that the

system is robust to inconsistancies in the TMD parameters. Changing the spring

constant seems to have more effect on the tower fatigue than changing the damping

constant, with changes of over 1%. This is most likely due to the fact that changing the
spring constant changes the TMD frequency tuning. One other important conclusion

of this sensitivity study is that the spring constant is mistuned for the barge, because

with a 10% reduction in TMD spring constant, there is a further reduction in tower
fatigue damage. However, this additional reduction is small, so the optimum found

by the genetic algorithm is close to the true optimum.

59
 CHAPTER 4

CONTROL STRUCTURE INTERACTION

The goal of the research on control-structure interaction (CSI) is to build on the

previous study that developed an active controller for the barge by adding actuator

dynamics, and investigating the effects. First, a limited degree of freedom model is
built to enable an analytical frequency domain analysis of the system. Using the

insight gained from the frequency domain analysis, a full DOF FAST-SC simulation

is conducted to test the original controllers with the new motor model. Finally, the
original analysis of the active controller is compared with the new data from the

updated actuator model simulations.

4.1 Limited Degree of Freedom Model

A limited degree of freedom model was needed for the barge in the CSI research.
The model that was developed in Section 3.1.2 is used for this research with a few

modifications. In order to model the actuator, a force term was added to the TMD

equation as well as the tower equation. The modified equations can be seen in Equa-
tions 4.1-4.3.

60
 Ip θ̈p = − dp θ̇p − kp θp − mp gRp θp + kt (θt − θp ) + dt (θ̇t − θ̇p ) (4.1)

It θ̈t = mt gRt θt − kt (θt − θp ) − dt (θ̇t − θ̇p ) − kT M D RT M D (RT M D θt − xT M D )

− dT M D RT M D (RT M D θ̇t − ẋT M D )

− mT M D g(RT M D θt − xT M D ) − RT M D fa (4.2)

mT M D ẍT M D = kT M D (RT M D θt − xT M D ) + dT M D (RT M D θ̇t − ẋT M D )

+ mT M D gθt + fa (4.3)

These equations are identical to Equations 3.3-3.5 except fa is added to represent

actuator force. The equations are tuned in the same way that is outlined in Section
3.2.

4.2 Frequency Domain Analysis

In the literature on control structure interaction, frequency analyses are used to

identify CSI, so this method is applied to the barge model. The frequency analysis
could have been applied to a FAST-SC model as well, but with a simple linear model,

the transfer functions can be solved analytically, which allows for greater insight into

the problem.
Using the limited degree of freedom model identified in Section 4.1, a frequency

domain analysis was preformed in order to investigate the effects of actuator force on

the response of the structure. Equations (4.4)-(4.6) are the Laplace transforms of the

equations of motion, Equations 4.1-4.3.

61
 Ib s2 Θb = − db s Θb − kb Θb − mb g Rb Θb + kt (Θt − Θb ) + dt s (Θt − Θb )
(4.4)

It s2 Θt = mt g Rt Θt − kt (Θt − Θb ) − dt s (Θt − Θb )

− kT M D RT M D (RT M D Θt − XT M D )

− dT M D RT M D s (RT M D , Θt − XT M D )

− mT M D g (RT M D Θt − XT M D ) − RT M D Fa (4.5)

mT M D s2 XT M D = kT M D (RT M D Θt − XT M D ) + dT M D s (RT M D Θt − XT M D )

+ mT M D g Θt + Fa (4.6)

In the HMD configurations, an electric motor is needed to generate the control

force. The motor is modeled as a permanent magnet DC motor, due to the lin-

ear characteristic equation (Equation 4.7), availability of motor constants, and clear

control-structure feedback path [4]. While actual large scale applications in civil
structures often use AC servomotors, CSI is a consideration for any actuator, and the

DC motor allows for a concise analysis [8]. Future work could analyze the effects of

different actuators on CSI.

Ra Kb Ki Kg Ki
T˙m = − Tm − ẋT M D + Vt (4.7)
La L a rm La

Ra is the armature resistance, La is the armature inductance, Tm is the motor

torque, ẋT M D is the speed of the HMD mass, Vt is the applied voltage, Kb is the back
electromotive force (EMF) constant, Ki is the motor torque constant, and Kg and rm

are the gear ratio and the lead of the ball screw, which converts the motor rotation

into linear motion. The values of these constants used in the analysis can be found

in Table 4.1 [25]. This motor was chosen from a handbook on direct current motors
based on the power requirements that were estimated in the previous work on the

active controller.

62
 Ki 1.2 Nm/A
Kb 1.27 V/s
Kg 1
Ra .0099 Ω
rm .15 m
La .00073 H

Table 4.1. Motor Constants for 505 Frame [25]

The force applied to the mass is defined in Equation (4.8).

T m Kg
fa = (4.8)
rm

An equation for force and its transform, Equation (4.9 and 4.10), can be made by

combining Equation (4.7) and Equation (4.8).

Ra Kb Ki Kg2 Ki Kg
f˙a = − fa − ẋT M D + vt (4.9)
La L a rm
2 L a rm

Ra Kb Ki Kg2 Ki Kg
s Fa = − Fa − s XT M D + Vt (4.10)
La L a rm
2 L a rm
Combining Equation (4.10) and Equations (4.4)-(4.6) leads to the system transfer

functions. These functions include GF V , the transfer function from voltage to motor
force, GθF , the transfer function from motor force to tower acceleration, and GθV

the transfer function from voltage to tower acceleration. Tower acceleration is used

because it is positively correlated to tower load. See Figure 4.1 for a block diagram of

the system. These transfer functions are very complex, with over a thousand terms,
however, using MatLab’s symbolic toolbox analytical equations can be obtained.

The Bode plot of the structure transfer function, GθF , can be seen in Figure 4.2,

and shows the poles of the structure at .574 rad/s and 3.4 rad/s, which correspond to
the barge and tower natural frequencies, respectively. The Bode plot of GF V , also in

63
 Figure 4.1. Block diagram showing feedback path [4]

Figure 4.2, shows the effects of CSI clearly. At the two structure poles, the actuator
has zeros (refer to Figure 4.3 for a clearer depiction of these zeros). Because of the

zeros, the actuator is least effective at the natural frequencies of the structure. The

drop in force magnitude at higher frequencies is due to the inductance of the motor,

but the system is seldom operated at these frequencies, so the effect of inductance
is negligible. The combined Bode plot, GθV , is simply the product of the structure

transfer function and the motor transfer function, which is the middle line in Figure

4.2.

4.2.1 Effect of Gear Ratio on CSI

Changing the motor gear ratio, which can be seen as Kg /rm in Equation (4.10),

can have a large effect on CSI. This term has units of rad/m, and is the number of

radians that the motor needs to turn to move the mass one meter. The X term in

Equation (4.10) has the gear ratio squared in the coeffecient, and the voltage term has
the gear ratio to the first power. Since control-structure interaction is caused by the

effect of the X term, by reducing the gear ratio, one can reduce the magnitude of the

X coefficient in relation to the voltage coefficient, and reduce the effect of CSI. This

64
 Bode Diagram

100

50

−50

Magnitude (dB)
−100

−150

−200

−250

−300
45
G
θF

0 G
FV
G
θV
−45
Phase (deg)

−90

−135

−180

−225

−270
−2 −1 0 1 2 3
10 10 10 10 10 10
Frequency (rad/sec)

Figure 4.2. Figure showing the three transfer functions

phenomenon can be seen in Figure 4.3, which compares the motor transfer function

of different gear ratios. These gear ratios were selected in order to give a reasonable
range of motor speeds for the motor size that is used.

As the gear ratio is reduced, the transfer function starts to resemble a low-pass

filter due to the inductance. Within the frequency of operation, the magnitude of the

force is flat, so the impact of CSI is diminished. This effect has not been shown in
previous investigations of CSI, and provides insight into how to design the mechanical

components of the actuator to minimize unwanted CSI effects.

65
 Bode Plot of G for different gear ratios
FV

80

70

60

Magnitude (dB)
50

40

30

20

10
45
K /r = 50 rad/m
g m
0 K /r = 20 rad/m
g m

−45 K /r = 10 rad/m
g m
K /r = 4 rad/m
g m
−90
Phase (deg)

−135

−180

−225

−270

−315

−360
−4 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10
Frequency (rad/sec)

Figure 4.3. Bode Plot of motors with different gear ratios

4.3 FAST-SC Simulation of Active Control of Wind Turbines

with CSI
Simulations are now performed in the time domain using FAST-SC, a state of the
art aero-elastic design code described in Section 3.1. These simulations will further

highlight the effect of actuator dynamics on the wind turbine system operating under

realistic loading conditions, and the impact on structural control approaches. The

simulations will also explore the effects of gear ratio and controller type in a realistic
setting. Simulations using a “pseudo-passive” controller and an H∞ are analyzed in

the following Sections 4.3.1-4.3.2.

66
4.3.1 Pseudo-Passive Analysis

In order to run FAST simulations using actuator models, an updated control

module is integrated with the FAST Simulink interface. Since the previous work on

active controller design used a complex H∞ controller, a “pseudo-passive” simulation

is run initially to isolate the effects of CSI on performance. In the pseudo-passive

simulations, the force that the passive spring and damper would apply to the mass is

used as the control force value for the actuator. This system is essentially mimicking

a simple PD controller, with the spring constant as the proportional constant, and

the damping value as the derivative constant.

Due to the effect of different gear ratios that was shown in the frequency analysis in

Figure 4.3, high and low gear ratio values are used in the simulations. The FAST-SC

simulations are run for 600 seconds with all turbine degrees of freedom turned on in
order to simulate a realistic senario. This includes all platform motions, all tower and

blade bending modes, rotor aerodynamic calculations and loading, and wave loading.

Two different wind input files are generated by TurbSim, a subroutine that produces
full-field turbulent wind data. The two wind files used have average speeds of 10 m/s

(below-rated) and 18 m/s (above-rated) and use the IEC Normal Turbulence Model.

The wave loading used has a significant wave height of 2.3 m for the 10 m/s wind

speed simulations and 3.7 m for the 18 m/s simulations. The peak spectral period
for these wave inputs were 14 s, and the waves are constructed using the Jonswap

spectrum. This wave loading represents relatively rough seas. It should also be noted

that in all simulations in this chapter, no position constraints on the HMD mass are
utilized. In some cases this results in unrealistically large displacements, but for the

purposes of analyzing the effects of CSI, this issue can be ignored.

In order to select the gear ratios to be used, preliminary testing is carried out, and

it is found that gear ratios higher than 10 rad/m resulted in controller instabilities

67
with the motor model that is used. Therefore, the high gear ratio is selected to be

6.67 rad/m, and the low gear ratio is 2.5 rad/m.

Comparisons of major structural performance metrics for the pseudo-passive sim-

ulation to the baseline passive system can be seen in Table 4.4. The metrics used for

the pseudo-passive simulation to quantify performance include: blade root flapwise

damage equivalent load (DEQL), tower fore-aft DEQL, low speed shaft (LSS) DEQL,

RMS platform pitch angle, RMS actuator power, RMS HMD displacement, and RMS
Fore-Aft Tower Top displacement. The damage equivalent loads are fatigue loads cal-

culated using a rainflow counting method. This method uses a one Hertz equivalent

load and uses a Wohler exponent of 10 for fiberglass and 3 for steel components. Min-
imizing the tower fatigue loads is the goal of this research, so tower fore-aft DEQL is

one of the more important metrics.

Performance Passive Low Gear Ratio High Gear Ratio
Index 18 m/s
Flap DEQL [kNm] 12326 12194 11348
% Improvement from passive - 1.1% 7.9%
Tower Fore-Aft DEQL [kNm] 53307 52674 48501
% Improvement from passive - 1.2% 9.0%
LSS DEQL [kNm] 990 979 927
% Improvement from passive - 1.2% 6.4%
RMS Platform Pitch Angle [deg] 2.50 2.46 2.19
% Improvement from passive - 1.7% 12.7%
Mean Actuator Power [kW] 0 28.7 36.7
% Improvement from low - - -27.9%
RMS AMD Displacement [m] 2.99 3.40 8.01
% Improvement from passive - -13.5% -167.6%
RMS Fore-Aft Tower Top Displacement [m] 0.507 .503 0.467
% Improvement from passive - 1.0% 8.02%

Table 4.2. Results of simulation with 18m/s wind and pseudo-passive controller

The results for the 10 m/s wind speed pseudo-passive simulation can be seen in Ta-
ble 4.3. All of the pseudo-passive simulations indicate that the models that included

CSI effects actually demonstrated better performance compared to the baseline pas-

sive system in all metrics except actuator power and AMD displacement. This is
due to the controller over-correcting the output force, which puts a higher force on

the tower top, and reduces accelerations and loads. However, the increased power

requirements for the actuator are extreme as the gear ratio is increased, with the

68
 Performance Passive Low Gear Ratio High Gear Ratio
Index 10 m/s
Flap DEQL [kNm] 6665 6520 7647
% Improvement from passive - 2.2% -14.7%
Tower Fore-Aft DEQL [kNm] 31868 31461 43751
% Improvement from passive - 1.3% -37.3%
LSS DEQL [kNm] 556 550 583.4
% Improvement from passive - 1.1% -5.0%
RMS Platform Pitch Angle [deg] 2.10 2.19 2.71
% Improvement from passive - 0.8% -23.4%
Mean Actuator Power [kW] 0 13.1 48.3
% Improvement from low - - -267.3%
RMS HMD Displacement [m] 2.03 2.32 9.22
% Improvement from passive - -14.1% -354.3%
RMS Fore-Aft Tower Top Displacement [m] 0.476 0.475 .538
% Improvement from passive - 0.3% -12.9%

Table 4.3. Results of simulation with 10m/s wind and pseudo-passive controller

high gear ratio actuator requiring about 2.5 times more power than the low gear ratio
system int the 10 m/s wind speed simulation. This trend becomes more apparent

when the complex controller is integrated in the next section.

4.3.2 HMD Anlysis

Next, simulations are run in order to compare previous work in active control using
an ideal actuator with the more realistic actuator model developed here. A family

of active controllers was developed by Rotea et al. [27, 33] for this application, but

was designed for an ideal actuator in which commanded force equals applied force. A

high authority and a low authority controller is selected from this previous work to
test with the motor model.

In this section of the results only one representative set of the simulations will be

analyzed in depth. The simulations selected have a mean wind speed of 18 m/s and
use the high authority controller. The wind and wave inputs used in these simulations

are the same as those used in the pseudo-passive analysis. Most of the trends that

will be presented with these simulations hold true for the other wind speeds and
controllers, however, for low authority controllers in below rated conditions, the high

gear ratio actuator performs worse in all performance indices. This can be seen in

Table 4.7.

69
 Table 4.4 shows the results of the four control schemes at 18 m/s windspeed using

the high authority controller. As in the pseudo-passive simulations, the actuator

models that include CSI preform better than the ideal controller in most metrics.
Once again, the extra force from the CSI reduces tower top acceleration, and thus

loading. However, this extra force requires substantially more actuation power, with

the high gear ratio motor requiring over 5 times as much power as the ideal motor.

A plot of a one minute section of actuation power can be seen in Figure 4.4. Tables
4.5-4.7 show the three other wind and controller combinations resulting from the

analysis.
Performance Passive Ideal Actuator Low Gear Ratio High Gear Ratio
Index 18 m/s High Authority High Authority High Authority
RMS Power Error [kW] 199 160 156 142
% Improvement from passive - 19.6% 21.5% 28.9%
RMS Rotor Speed Error [rpm] 1.25 1.10 1.09 1.02
% Improvement from passive - 12.3% 13.6% 18.8%
Flap DEQL [kNm] 12326 10949 10853 10586
% Improvement from passive - 11.1% 12.0% 14.1%
Tower Fore-Aft DEQL [kNm] 53307 44090 43557 40963
% Improvement from passive - 17.3% 18.3% 23.2%
LSS DEQL [kNm] 990 916 909 882
% Improvement from passive - 7.5% 8.2% 10.9%
RMS Platform Pitch Angle [deg] 2.50 2.12 2.08 1.98
% Improvement from passive - 15.5% 17.0% 20.8%
Mean Actuator Power [kW] 0 253.2 308.7 1517.2
% Improvement from ideal - - -21.9% -499.0%
RMS HMD Displacement [m] 2.99 6.54 7.07 13.8
% Improvement from passive - -118.4% -136.0% -360.1%
RMS Fore-Aft Tower Top Displacement [m] 0.507 0.455 0.450 0.434
% Improvement from passive - 10.3% 11.4% 14.5%

Table 4.4. Results of simulation with 18m/s wind and high authority controller

Performance Passive Ideal Actuator Low Gear Ratio High Gear Ratio
Index 18 m/s Low Authority Low Authority Low Authority
RMS Power Error [kW] 199 184 180 156
% Improvement from passive - 7.9% 9.6% 21.6%
RMS Rotor Speed Error [rpm] 1.25 1.19 1.18 1.06
% Improvement from passive - 5.1% 6.3% 15.4%
Flap DEQL [kNm] 12326 11755 11651 10815
% Improvement from passive - 4.6% 5.5% 12.3%
Tower Fore-Aft DEQL [kNm] 53307 48842 48305 44448
% Improvement from passive - 8.4% 9.4% 16.6%
LSS DEQL [kNm] 990 955 949 899
% Improvement from passive - 3.6% 4.2% 9.2%
RMS Platform Pitch Angle [deg] 2.50 2.34 2.30 2.08
% Improvement from passive - 6.4% 7.8% 17.0%
RMS Actuator Power [kW] 0 61.9 84.3 709.0
% Improvement from ideal - - -36.3% -1045.1%
RMS HMD Displacement [m] 2.99 4.28 4.65 9.81
% Improvement from passive - -42.9% -55.3% -227.6%
RMS Fore-Aft Tower Top Displacement [m] 0.507 0.485 0.480 0.449
% Improvement from passive - 4.5% 5.4% 11.6%

Table 4.5. Results of simulation with 18m/s wind and low authority controller

70
 Performance Passive Ideal Actuator Low Gear Ratio High Gear Ratio
Index 10 m/s Low Authority Low Authority Low Authority
Flap DEQL [kNm] 6665 6038 6041 5959
% Improvement from passive - 9.4% 9.4% 10.6%
Tower Fore-Aft DEQL [kNm] 31868 24811 24429 22346
% Improvement from passive - 22.1% 23.3% 29.9%
LSS DEQL [kNm] 556 514 520 519
% Improvement from passive - 7.5% 6.5% 6.7%
RMS Platform Pitch Angle [deg] 2.10 2.03 2.01 1.95
% Improvement from passive - 8.0% 8.7% 11.4%
Mean Actuator Power [kW] 0 126.1 154.1 769.3
% Improvement from ideal - - -22.1% -509.9%
RMS HMD Displacement [m] 2.03 4.56 4.93 9.75
% Improvement from passive - -124.4% -143.0% -380.4%
RMS Fore-Aft Tower Top Displacement [m] 0.476 0.457 0.455 0.449
% Improvement from passive - 4.1% 4.5% 5.8%

Table 4.6. Results of simulation with 10m/s wind and high authority controller

Performance Passive Ideal Actuator Low Gear Ratio High Gear Ratio
Index 10 m/s Low Authority Low Authority Low Authority
Flap DEQL [kNm] 6665 6377 6356 7343
% Improvement from passive - 4.3% 4.6% -10.2%
Tower Fore-Aft DEQL [kNm] 31868 28462 28245 40123
% Improvement from passive - 10.7% 11.4% -25.9%
LSS DEQL [kNm] 556 542 538 575
% Improvement from passive - 2.4% 3.2% -3.4%
RMS Platform Pitch Angle [deg] 2.10 2.14 2.13 2.61
% Improvement from baseline - 2.9% 3.5% -18.6%
Mean Actuator Power [kW] 0 29.3 40.2 982.8
% Improvement from ideal - - -37.1% -3247.8%
RMS HMD Displacement [m] 2.03 2.96 3.23 11.69
% Improvement from passive - -45.8% -59.0% -475.6%
RMS Fore-Aft Tower Top Displacement [m] 0.476 0.468 0.466 .522
% Improvement from passive - 1.8% 2.0% -9.5%

Table 4.7. Results of simulation with 10m/s wind and low authority controller

71
 Actuation Power as a Function of Time
8000
Ideal Controller
Low Gear Ratio
High Gear Ratio
7000

6000

5000
Actuation Power (kW)

4000

3000

2000

1000

−1000
120 130 140 150 160 170 180 190
Time (s)

Figure 4.4. Actuator power over time for 18 m/s wind and high control authority.

The large increase in power consumption results in a small improvement in tower

top displacement, which is evident when comparing Tower Fore-Aft DEQL for the
different control schemes. The high gear ratio actuator uses almost 500% more power

than the ideal actuator case while decreasing fatigue loading by only 6%.

One important result is that the low gear ratio simulation is similar in performance
to the ideal actuator case. This is due to the nearly flat gain in the motor transfer

function for low gear ratios (see Figure 4.3). The result is that when designing a

structural control system, there is a clear benefit to using low gear ratios, so as to
minimize the effect of CSI. The practical effect of changing the gear ratio is changing

the rotational speed and torque of the motor. Depending on the motor selected, gear

72
ratios as low as the ones used in these simulations may result in torque outputs that

are too high for continuous duty operation.

73
 CHAPTER 5

CONCLUSIONS AND FUTURE WORK

Tuned mass dampers show great promise for use in offshore wind turbines. Both

passive and active TMDs can reduce tower and mooring line loads significantly, which

leads to lighter and less expensive structures. This thesis contributes a set of opti-
mized passive TMDs for four offshore wind turbines, as well as an analysis of an active

controller for the barge platform.

The optimization of the passive tuned mass damper is completed for the monopile,
barge, spar buoy, and tension leg platform in Chapter 3. This chapter includes the

optimization of the spring and damping parameters using a genetic algorithm. The

results of this genetic algorithm are then simulated in FAST-SC and compared to a
baseline without a TMD. Each of the four platforms has a unique TMD configuration

that results in the best load reduction.

The monopile achieves the best load reduction with both a fore-aft and a side-side

TMD in the nacelle. With this configuration, fore-aft loads are reduced by up to
10%, and side-side fatigue loading is reduced by as much as 66%. Side-side ultimate

loads are effected as well, with the 95th percentile load being reduced by 32%. The

barge benefits from the same TMD configuration, with a 15% fore-aft fatigue damage
reduction and up to a 33% reduction in side-side fatigue loading. Even with the

load reduction, the best barge fatigue damage values are approximately 4 times the

baseline for the monopile.

For the spar buoy, the best configuration involves a nacelle-based TMD as well.
With the TMD in the platform, low value of the spring constant combined with

74
the thrust forcing on the turbine rotor caused the TMD to rest against the stop

and contribute very little load reduction. The nacelle-based TMD reduced fore-aft

damage by as much as 8%, and side-side fatigue loads by 38%. Ultimate tower loads,
however, were increased with the nacelle based TMD. This could be caused by the

extra tower top mass from the TMD causing more tower bending due to gravitational

and inertial loading in the extremes of the pitching motion.

The TLP was a complex structure to model; the genetic algorithm that was used
for the other platforms could not be used for the TLP. However, the parametric

study method that was used instead allowed a primary investigation into the effects

of a TMD. Since the surge motion of the TLP has the largest amplitude of the
platform DOFs, this was initially assumed to have the largest effect on tower loading.

Preliminary investigations showed that although a surge-tuned TMD reduces the

amplitude of this motion, it does little to effect tower or mooring line loading. The

platform pitching motion was found to influence both tower bending and line tension
the most, so a TMD was developed to reduce this motion for use in the platform and

nacelle. The nacelle based TMD once again proved to have a better effect on loading

than the TMD in the platform. The nacelle TMD reduced tower fore-aft loads by
almost 8%, and side-side loads by 21%. The TMD had little effect on fore-aft ultimate

loads, but reduced side-side peak loads by 7%. Line loading was also effected; the

TMD reduced line fatigue damage by 7%.

In Chapter 4, a design of an active control system for a mass damper in an offshore

wind turbine using the barge floating platform is modified to include an actuator

model. The addition of the actuator model creates a more realistic simulation, and

motivates a controller redesign. When the actuator is added, the AMD uses 5-10
times more power in some cases, while reducing loads by only a few percent. In order

to make an active controller viable for offshore wind turbines, actuator models must

be included in future work.

75
 These load reductions for all of the platforms could have a beneficial effect on

the cost of an offshore wind turbine as long as the TMD could be constructed at a

reasonable cost. Future work on this topic should consider a preliminary economic
feasibility study. This study should quantify the cost associated with the load reduc-

tion and compare this to the material and construction cost of a TMD, as well as the

power usage in the case of an AMD.

5.1 Other TMD Research Topics for Investigation

5.1.1 TLCD

All of the studies in this thesis focused on using an ideal tuned mass damper. In
practice, however, the mechanics of a tuned mass damper may be difficult to achieve.

For example, with a stroke of ±8m, a spring would need to stretch 16m and a spring

like this may not exist, or may prohibitively expensive. A tuned liquid column damper
may be a good alternative to the tuned mass damper for this application. The water

which supplies the mass in a TLCD is essentially free for an offshore turbine, and

there are no large scale springs and dampers involved. Further research should go

into developing a passive TLCD for use in an offshore wind turbine.

5.1.2 Semi-Active

Another topic that should be investigated is a semi-active damper. With a proper

control scheme, these devices have been shown to approach AMDs in terms of load

reduction with minimum power requirements. A controllable orifice TLCD is also a

promising alternative. This modification to the TLCD allows for semi-active control

in addition to the other benefits of the TLCD.

5.1.3 Redesigned HMD

Due to the findings from the CSI study and the passive optimization study, the

HMD that was developed for the barge should be altered, and HMDs should be

76
developed for the remaining platforms. The controller itself needs to be redesigned

taking actuator dynamics into consideration. Also, the original spring and damping

constants for the HMD were taken from the parametric study, and the improved
parameters should be used to achieve the optimum system.

5.2 Concluding Remarks

Tuned mass dampers have great potential to be used in offshore wind turbines. All

of the floating platforms and the fixed bottom platform analyzed in this thesis showed
improvement in fatigue damage with the application of both active and passive tuned

mass dampers. A few problems still need to be addressed to make mass dampers

practical for floating wind turbines. Due to the cost and manufacturing challenges
of constructing a tuned mass damper for this application, alternative forms of mass

damper systems including tuned liquid column dampers should be considered. Alter-

natively, a purely active mass damper should be considered, which would eliminate

the need for large displacement springs and dampers. Control-structure interaction
must be taken into account when designing this system. With these recommenda-

tions, tuned mass dampers could provide a cost effective means of load reduction in

offshore wind turbines.

77
 APPENDIX

GENETIC ALGORITHM CODE

lenk=13; % The length of the genomes

lend=15;
popSize=50; % The size of the population (must be an even number)
maxGens=20; % The maximum number of generations allowed in a run
probCrossover=.7; % The probability of crossing over.
probMutation=0.01; % The mutation probability (per bit)
sigmaScalingFlag=1; % Sigma Scaling is described on pg 168 of M. Mitchell’s
% GA book. It often improves GA performance.
sigmaScalingCoeff=.7; % Higher values => less fitness pressure

SUSFlag=1; % 1 => Use Stochastic Universal Sampling (pg 168 of

% M. Mitchell’s GA book)
% 0 => Do not use Stochastic Universal Sampling
% Stochastic Universal Sampling almost always
% improves performance

crossoverType=1; % 0 => no crossover

% 1 => 1pt crossover
% 2 => uniform crossover

visualizationFlag=0; % 0 => don’t visualize bit frequencies

% 1 => visualize bit frequencies

verboseFlag=1; % 1 => display details of each generation

% 0 => run quietly

useMaskRepositoriesFlag=1; % 1 => draw uniform crossover and mutation masks from

% a pregenerated repository of randomly generated bits.
% Significantly improves the speed of the code with
% no apparent changes in the behavior of
% the SGA
% 0 => generate uniform crossover and mutation
% masks on the fly. Slower.

% crossover masks to use if crossoverType==0.

%mutationOnlycrossmasks=false(popSize,len);

% pre-generate two repositories of random binary digits from which the

% the masks used in mutation and uniform crossover will be picked.
% maskReposFactor determines the size of these repositories.

maskReposFactor=5;
uniformCrossmaskReposk=rand(popSize/2,(lenk+1)*maskReposFactor)<0.5;
uniformCrossmaskReposd=rand(popSize/2,(lend+1)*maskReposFactor)<0.5;
uniformCrossmaskReposm=rand(popSize/2,(lenm+1)*maskReposFactor)<0.5;
mutmaskReposk=rand(popSize,(lenk+1)*maskReposFactor)<probMutation;
mutmaskReposd=rand(popSize,(lend+1)*maskReposFactor)<probMutation;
mutmaskReposm=rand(popSize,(lenm+1)*maskReposFactor)<probMutation;

% preallocate vectors for recording the average and maximum fitness in each
% generation
avgFitnessHist=zeros(1,maxGens+1);
maxFitnessHist=zeros(1,maxGens+1);

eliteIndiv=[];
eliteFitness=-realmax;

% the population is a popSize by len matrix of randomly generated boolean

% values
popk=rand(popSize,lenk)<.5;
popd=rand(popSize,lend)<.5;

for gen=0:maxGens

% evaluate the fitness of the population. The vector of fitness values

% returned must be of dimensions 1 x popSize.
for c = 1:popSize
fitnessVals(c)=1/(optimize_sim_bouy([bin2dec(num2str((popk(c,:)))),...
bin2dec(num2str((popd(c,:))))]));
end

78
[maxFitnessHist(1,gen+1),maxIndex]=max(fitnessVals);
avgFitnessHist(1,gen+1)=mean(fitnessVals);
if eliteFitness<maxFitnessHist(gen+1)
eliteFitness=maxFitnessHist(gen+1);
eliteIndivk=popk(maxIndex,:);
eliteIndivd=popd(maxIndex,:);
end

% display the generation number, the average Fitness of the population,

% and the maximum fitness of any individual in the population
if verboseFlag
display([’gen=’ num2str(gen,’%.3d’) ’ avgFitness=’ ...
num2str(avgFitnessHist(1,gen+1),’%3.3f’) ’ maxFitness=’ ...
num2str(maxFitnessHist(1,gen+1),’%3.3f’)]);
end

% Conditionally perform sigma scaling

if sigmaScalingFlag
sigma=std(fitnessVals);
if sigma~=0;
fitnessVals=1+(fitnessVals-mean(fitnessVals))/...
(sigmaScalingCoeff*sigma);
fitnessVals(fitnessVals<=0)=0;
else
fitnessVals=ones(popSize,1);
end
end

% Normalize the fitness values and then create an array with the
% cumulative normalized fitness values (the last value in this array
% will be 1)
cumNormFitnessVals=cumsum(fitnessVals/sum(fitnessVals));

% Use fitness proportional selection with Stochastic Universal or Roulette

% Wheel Sampling to determine the indices of the parents
% of all crossover operations
if SUSFlag
markers=rand(1,1)+[1:popSize]/popSize;
markers(markers>1)=markers(markers>1)-1;
else
markers=rand(1,popSize);
end
[temp parentIndices]=histc(markers,[0 cumNormFitnessVals]);
parentIndices=parentIndices(randperm(popSize));

% deterimine the first parents of each mating pair

firstParentsk=popk(parentIndices(1:popSize/2),:);
firstParentsd=popd(parentIndices(1:popSize/2),:);
% determine the second parents of each mating pair
secondParentsk=popk(parentIndices(popSize/2+1:end),:);
secondParentsd=popd(parentIndices(popSize/2+1:end),:);

% create crossover masks

if crossoverType==0
masks=mutationOnlycrossmasks;
elseif crossoverType==1
masksk=false(popSize/2, lenk);
masksd=false(popSize/2, lend);
tempk=ceil(rand(popSize/2,1)*(lenk-1));
tempd=ceil(rand(popSize/2,1)*(lend-1));
for i=1:popSize/2
masksk(i,1:tempk(i))=true;
masksd(i,1:tempd(i))=true;
end
else
if useMaskRepositoriesFlag
tempk=floor(rand*lenk*(maskReposFactor-1));
tempd=floor(rand*lend*(maskReposFactor-1));
masksk=uniformCrossmaskReposk(:,tempk+1:tempk+lenk);
masksd=uniformCrossmaskReposd(:,tempd+1:tempd+lend);
else
masksk=rand(popSize/2, lenk)<.5;
masksd=rand(popSize/2, lend)<.5;
end
end

% determine which parent pairs to leave uncrossed

reprodIndices=rand(popSize/2,1)<1-probCrossover;
masksk(reprodIndices,:)=false;
masksd(reprodIndices,:)=false;

% implement crossover
firstKidsk=firstParentsk;
firstKidsd=firstParentsd;
firstKidsk(masksk)=secondParentsk(masksk);
firstKidsd(masksd)=secondParentsd(masksd);
secondKidsk=secondParentsk;
secondKidsd=secondParentsd;
secondKidsk(masksk)=firstParentsk(masksk);
secondKidsd(masksd)=firstParentsd(masksd);
popk=[firstKidsk; secondKidsk];

79
 popd=[firstKidsd; secondKidsd];

% implement mutation
if useMaskRepositoriesFlag
tempk=floor(rand*lenk*(maskReposFactor-1));
tempd=floor(rand*lend*(maskReposFactor-1));
masksk=mutmaskReposk(:,tempk+1:tempk+lenk);
masksd=mutmaskReposd(:,tempd+1:tempd+lend);
else
masksk=rand(popSize, lenk)<probMutation;
masksd=rand(popSize, lend)<probMutation;
end
popk=xor(popk,masksk);
popd=xor(popd,masksd);
end
if verboseFlag
figure(2)
%set(gcf,’Color’,’w’);
hold off
plot([0:maxGens],avgFitnessHist,’k-’);
hold on
plot([0:maxGens],maxFitnessHist,’c-’);
title(’Maximum and Average Fitness’)
xlabel(’Generation’)
ylabel(’Fitness’)
end

80
 BIBLIOGRAPHY

[1] Adeli, H. Smart structures and building automation in the 21st century. In
The 25th International Symposium on Automation and Robotics in Construction
(June 2008).

[2] Adeli, H., and Saleh, A. Integrated structural/control optimization of large

adaptive/smart structures. International Journal of Solids and Structures 35,
28-29 (1998), 3815 – 3830. Active control systems;Parallel vector algorithms;.

[3] Alves, R.M., and Batista, R.C. Active/passive control of heave motion for tlp
type offshore platform. In International Conference on Offshore Mechanics and
Arctic Engineerings (1999), pp. 332–338.

[4] Battaini, M., Yang, G., and Spencer, B.F. Bench-scale experiment for structural
control. Journal of Engineering Mechanics 126, 2 (2000), 140–148.

[5] Butterfield, S., Musial, W., Jonkman, J., and Sclavounos, P. Engineering chal-
lenges for floating offshore wind turbines. In Copenhagen Offshore Wind Con-
ference (Copenhagen, Denmark, October 2005), pp. 377–382.

[6] Calderon, B.E. Design and optimization of a wind turbine tower by using a
damper device. Master’s thesis, Stuttgart University, 2009.

[7] Colwell, S., and Basu, B. Tuned liquid column dampers in offshore wind turbines
for structural control. Engineering Structures 31, * (2009), 358–368.

[8] Dyke, S.J., Spencer, B.F., Quast, P., and Sain, M.K. The role of control-structure
interaction in protective system design. Journal of Engineering Mechanics 121,
2 (1995), 322–338.

[9] Dyke, S.J., Spencer, B.F., Quast, P., Sain, M.K., Kaspari, D.C., and Soong, T.T.
Experimental verification of acceleration feedback control strategies for an active
tendon system. Tech. Rep. NCEER-94-0024, National Center for Earthquake
Engineering Research, 1994.

[10] Enevoldsen, I. Effects of a vibration mass damper in a wind turbine tower.

Mechanics of Structures and Machines 24, 2 (1996), 155–187.

[11] EWEA. Upwind: Design limits and solutions for very large wind turbines. Tech.
rep., EWEA, March 2011.

81
[12] Fujita, T. Application of hybrid mass damper with convertible active and pas-
sive modes using hydraulic actuator to high-rise building. In American Control
Conference (Baltimore, Maryland, June 1994).

[13] Gao, H., and Kwok, K.C.S. Optimization of tuned liquid column dampers.
Engineering Structures 19, 6 (1997), 476–486.

[14] Hirayama, T., and Ma, N. Dynamic response of a very large floating structure
with active pneumatic control. In 7th International Offshore and Polar Engi-
neering Conference (Honolulu, Hawaii, May 1997), pp. 377–382.

[15] International Electrotechnical Commission 614003:2008. Wind Turbines–Part 3:

Design Requirements for Offshore Wind Turbines. International Electrotechnical
Commission, Geneva, Switzerland.

[16] Jonkman, B.J., and Buhl, M. L. Turbsim user’s guide. Tech. Rep. NREL/EL-
500-36970, National Renewable Energy laboratory, Golden, Colorado, 2005.

[17] Jonkman, J. Dynamics Modeling and Loads Analysis of an Offshore Floating

Wind Turbine. PhD thesis, National Renewable Energy Laboratory, 2008.

[18] Jonkman, J. Influence of control on the pitch damping of a floating wind turbine.
In 46th AIAA Aerospace Science Meeting and Exhibit (Reno, Nevada, January
2008).

[19] Jonkman, J., and Buhl, M. L. Fast user’s guide. Tech. Rep. NREL/EL-500-
38230, National Renewable Energy laboratory, Golden, Colorado, 2006.

[20] Jonkman, J., Butterfield, S., Musial, W., and Scott, G. Definition of a 5-mw
reference wind turbine for offshore system development. TP 500-38060, National
Renewable Energy Laboratory, 2008.

[21] Jonkman, J., and Matha, D. A quantitative comparison of the responses of

three floating platform concepts. In European Offshore Wind 2009 Conference
and Exhibition (Stockholm, Sweden, September 2009).

[22] Kanegaonkar, H.B. Smart technology systems in offshore structures: Status and
needs. In 9th International Offshore and Polar Engineering Conference (Brest,
France, May 1999).

[23] Krenk, S. Frequency analysis of the tuned mass damper. ASME Journal of
Applied Mechanics 72 (November 2005), 936– 942.

[24] Krenk, S., and Høgsberg, Jan. Tuned mass absorbers on damped structures
under random load. Probabilistic Engineering Mechanics 23 (2008), 408– 415.

[25] Kusko, A. Solid-State DC Motor Drives. The M.I.T. Press, Cambridge, Mas-
sachusetts, 1969.

82
[26] Lackner, M.A. An investigation of the control and loads of floating wind turbines.
Under Review, Wind Energy, 4 (2009), 1118–1142.
[27] Lackner, M.A., and Rotea, M. Structural control of floating wind turbines.
Mechatronics (2010). In press, available online.
[28] Lackner, M.A., and Rotea, M.A. Passive structural control of offshore wind
turbines. Wind Energy (September 2010). DOI: 10.1002/we.426.
[29] Lee, H.H., Wong, S.H., and Lee, R.S. Response mitigation on the offshore floating
platform system with tuned liquid column damper. Ocean Engineering 33, 4
(2006), 1118–1142.
[30] Murtagh, P.J., Ghosh, A., Basu, B., and Broderick, B.M. Passive control of wind
turbine vibrations including blade/tower interaction and rotationally sampled
turbulence. Wind Energy 11, 4 (2007), 305–317.
[31] Musial, W., Butterfield, S., and Ram, B. Energy from offshore wind. In Offshore
Technology Conference (Houston, Texas, May 2006).
[32] Nagashima, I., Maseki, R., Asami, Y., Hirai, J., and Abiru, H. Performance of
hybrid mass damper system applied to a 36-storey high-rise building. Earthquake
Engineering and Structural Dynamics 30 (2001), 1615–1637.
[33] Rotea, M.A., Lackner, M.A., and Saheba, R. Active structural control of offshore
wind turbines. In 48th AIAA Aerospace Science Meeting and Exhibit (Orlando,
Florida, January 2010).
[34] Singh, M.P., and Matheu, E.E. Active and semi-active control of structures
under seismic excitation. Earthquake Engineering and Structural Dynamics 26
(1997), 193–213.
[35] Soong, T.T., and Spencer, B.F. Supplemental energy dissipation: state-of-the-
art and state-of-the practice. Engineering Structures 24 (2002), 243–259.
[36] Spencer, B.F., and Nagarajaiah, S. State of the art of structural control. Journal
of Structural Engineering 129, 7 (2003), 845–856.
[37] Spencer, B.F., and Sain, M.K. Controlling buildings - a new frontier in feedback.
The Shock and Vibration Digest 30, 4 (1998), 267–281.
[38] Symans, M.D., and Constantinou, M.C. Semi-active control systems for seismic
protection of structures: A state-of-the-art review. Engineering Structures 21, *
(1999), 469–487.
[39] Wilmink, A.J., and Hengeveld, J.F. Application of tuned liquid column dampers
in wind turbines. Tech. rep., Mecal Applied Mechanics, Netherlands, 2006.
[40] Yamanaka, M., and Okuda, H. Dentsu head office damping systems, tokyo,
japan. Structural Engineering International 15, 1 (2005).

83