You are on page 1of 336

KATHOLIEKE UNIVERSITEIT LEUVEN

FACULEIT INGENIEURSWETENSCHAPPEN
DEPARTEMENT WERKTUIGKUNDE
AFDELING PRODUCTIETECHNIEKEN
MACHINEBOUW EN AUTOMATISERING
Celestijnenlaan 300B B-3001 Leuven (Heverlee), Belgium

SIMULATION OF DYNAMIC DRIVE TRAIN


LOADS IN A WIND TURBINE

Promotoren: Proefschrift voorgedragen tot


Prof. Dr. Ir. D. Vandepitte het bekomen van de graad
Prof. Dr. Ir. P. Sas van Doctor in de
Ingenieurswetenschappen

door

Joris PEETERS

2006D07 Juni 2006


KATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT INGENIEURSWETENSCHAPPEN
DEPARTEMENT WERKTUIGKUNDE
AFDELING PRODUCTIETECHNIEKEN
MACHINEBOUW EN AUTOMATISERING
Celestijnenlaan 300B - B-3001 Leuven (Heverlee), Belgium

SIMULATION OF DYNAMIC DRIVE TRAIN


LOADS IN A WIND TURBINE

Jury: Proefschrift voorgedragen tot


Prof. Dr. Ir. P. Van Houtte, voorzitter het bekomen van de graad
Prof. Dr. Ir. D. Vandepitte, promotor van Doctor in de
Prof. Dr. Ir. P. Sas, promotor Ingenieurswetenschappen
Prof. Dr. Ir. J. De Schutter
Prof. Dr. Ir. W. Desmet door
Prof. Dr. Ir. W. D’haeseleer
Joris PEETERS
Prof. Dr. Ir. J. Driesen
Prof. Dr. Ir. J.C. Golinval (Université de Liège)
Dr. Ir. P. Flamang (Hansen Transmissions International)

UDC: 681.3∗D2 Juni 2006


© Katholieke Universiteit Leuven – Faculteit Ingenieurswetenschappen
Arenbergkasteel, B-3001 Heverlee (Belgium)
Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenig-
vuldigd en/of openbaar gemaakt door middel van druk, fotokopie, microfilm,
elektronisch of op welke andere wijze ook zonder voorafgaande schriftelijke
toestemming van de uitgever.

All rights reserved. No part of the publication may be reproduced in any form
by print, photoprint, microfilm or any other means without written permission
from the publisher.

D/2006/7515/62
ISBN 90-5682-728-6
UDC 681.3∗D2
Voorwoord

The significant problems we face cannot be solved


at the same level of thinking we were at
when we created them.
Albert Einstein

In 2001 besliste ik om na mijn ingenieursstudies te starten als onderzoeker aan


de KULeuven. Mijn keuze werd gemaakt op basis van de omschrijving van een
uitdagend onderzoeksproject in samenwerking met de firma Hansen Transmis-
sions International. Het project bleek achteraf gezien een echte opportuniteit,
waaraan ik met overgave en plezier gewerkt heb en waaruit dit proefschrift
ontstaan is. Het aanvatten van het onderzoek over windturbines en hun tand-
wielkasten vergde het inslaan van een nieuwe onderzoekspiste binnen de on-
derzoeksgroep Geluid & Trillingen aan het Departement Werktuigkunde. Tij-
dens mijn tocht op deze piste werd ik enerzijds geı̈nspireerd door input vanuit
de academische wereld en anderzijds door input vanuit “de industrie”. Het was
een unieke ervaring om als onderzoeker zo dicht bij de industriële toepassing
te staan en deze ervaring heeft zeker haar stempel gedrukt op het voorliggende
resultaat. Aan het begin van deze tekst wil ik een oprecht woord van dank
richten aan allen die mij hebben bijgestaan tijdens mijn onderzoek.

Vooreerst dank ik mijn hoofdpromotor en tevens vaste begeleider, professor


Dirk Vandepitte, voor de ontelbare interessante gesprekken, voor de continue
ondersteuning, voor het geven van richting en voor het inspireren en dit zonder
mijn vrijheid te beperken. Verder ook voor het in mij gestelde vertrouwen en
voor de verantwoordelijkheid die ik kreeg in het onderzoeksproject en tijdens
didactische taken. Ik voelde me geapprecieerd als onderzoeker, maar zeker
ook als mens.

I
II Voorwoord

Mijn co-promotor professor Paul Sas wil ik bedanken voor het verkregen ver-
trouwen en voor zijn nuttige commentaar, vooral bij het uitwerken van de ex-
perimentele analyses. Dank ook aan mijn assessoren professor Wim Desmet en
professor William D’haeseleer voor hun opmerkingen bij het nalezen van dit
proefschrift. Mijn promotoren en mijn assessoren wens ik bovendien, samen
met professor Joris De Schutter, professor Johan Driesen, Dr. Peter Flamang
en professor Jean-Claude Golinval, te bedanken om deel te willen uitmaken
van mijn jury.

Een woord van dank ook aan de firma Hansen Transmissions International,
kortweg Hansen, en de talrijke medewerkers, die hebben bijgedragen aan dit
onderzoek. In het begin was er Stefan Lammens: als ex-PMA-er de ideale
persoon om me te introduceren bij Hansen en in de problematiek van wind-
turbines, tandwielkasten en dynamica. Al vrij snel leerde ik ook de rest van
het R&D-team van Hansen kennen, dat onder leiding staat van Peter Flamang.
Bedankt aan hem en aan iedereen uit de groep van Roger Bogaert, van Marcel
De Wilde en van Dirk Leimann om mij te steunen in mijn onderzoek. Bedankt
om mijn resultaten - vaak met een kritische blik - te evalueren, om mij bij te
sturen, om mij te inspireren en om mij te betrekken in talloze boeiende dis-
cussies en in de unieke meetcampagne. Jullie “industriële” input vormde een
belangrijke bijdrage in het uitwerken van het onderzoeksproject. Ik ben blij
dat ik ondertussen zelf deel uitmaak van het Hansen-team en dat we samen
kunnen verder werken om de beste tandwielkasten te blijven produceren.

Dank aan de vele medewerkers van de afdeling PMA en van het departement,
die er altijd waren voor de praktische en morele ondersteuning en voor de
vele toffe babbels. Paul, bedankt voor je technische hulp bij het uitvoeren van
metingen, voor je eeuwige enthousiasme en voor de toffe samenwerking tij-
dens de meetcampagne in een windturbine. Ann, Carine, Karin, Lieve en Luc,
bedankt om mij te verlossen van de meeste administratieve zorgen. Dank aan
Jan en Ronny, die er steeds waren als helpdesk voor al mijn IT-problemen. Be-
dankt ook aan Dirk, Eddy en Viggo van de werkplaats, aan Raymond, Paul
en Luc van de dienst elektronica en aan Jean-Pierre om altijd opnieuw ter
beschikking te staan voor de meest uiteenlopende praktische vraagjes of als
luisterend oor.

Bedankt aan iedereen die zorgde voor de aangename werksfeer gedurende de


voorbije vijf jaren op PMA. Ik begon op een bureau met o.a. Bas en Filip. Bas
was onze peter in die beginperiode en het was een tof eerste jaar. Na dat jaar
werden Filip en ik getransfereerd naar “de Pick&Pay-bureau” in het nieuwe
Labo Voertuigtechnologie & Lichtgewichtconstructies aan den overkant. Het
aantal assistenten in dit labo nam snel toe en den overkant kreeg ook een eigen
Voorwoord III

sociaal leven, met o.a. de koffiekoeken op vrijdag. Het was een leuke werk-
plek! Andrea, it was a pleasure to share an office with you. Aan het einde
van 2005 beslisten Filip en ik om opnieuw te verhuizen om “rustig” te kun-
nen schrijven aan onze doctoraatstekst. Filip, bedankt voor de leuke tijd als
vaste bureau-collega en om steeds weer klaar te staan om te helpen. Greg,
bedankt voor alle bezoekjes aan den overkant en je hulp bij van-alles-en-nog-
wat. Bert, bedankt omdat ik altijd op je kon rekenen en om mijn wegwijzer te
zijn tijdens de laatste drukke maanden. Dank ook aan iedereen die meehielp
in de talloze leuke nevenactiviteiten, zoals het voetbal op donderdagmiddag en
op occasionele toernooien, het squashen op maandag, de PMA-weekends, de
MOD-activiteiten, de Happy-Hours, de BBQ’s, de kroegentochten, de mech-
Prono’s en aanverwanten, . . .

Een speciaal woord van dank aan Bert, David, Greg, Raymond en Wim voor de
oprichting van het BAD. Deze groep van vrienden-collega’s had een belangrijk
aandeel in de toffe werksfeer en vormt daarvan een geslaagd verlengstuk tot
buiten de muren van het departement. Heren, laat onze - nu al - legendarische
activiteiten en traditionele TWVM’s nog lang voortduren!

Last, but certainly not least, wens ik mijn familie te bedanken voor al hun en-
thousiasme en motivatie. Mama en papa, bedankt voor alle kansen die ik kreeg
in mijn leven en voor jullie onvoorwaardelijke steun! Els, bedankt voor alles
wat je voor mij betekent! Bedankt voor alle toffe momenten, de vele kleine
dingen en het eeuwige begrip. Ik hoop dat het ons voor de wind mag gaan!

Joris
Juni 2006
IV
Samenvatting

Dit proefschrift bespreekt de ontwikkeling van een consistente modellerings-


techniek voor het simuleren van de dynamische belasting in de aandrijflijn van
een windturbine met de focus op de tandwielkast. De motivatie voor dit onder-
zoek zit in de beperkingen van de simulatiecodes, die worden gebruikt in het
huidige ontwerp van een windturbine. De output van deze codes impliceert een
quasi-statisch ontwerp van alle componenten in de aandrijflijn en biedt boven-
dien onvoldoende inzicht in de optredende belastingswisselingen en lokale
spanningsniveaus. Een formulering op basis van flexibele meerlichamen- sys-
temen (MLS) wordt beschouwd als de beste methode voor de ontwikkeling
van een meer gedetailleerd model van de aandrijflijn. Deze studie beschrijft
een generieke methodologie gebaseerd op drie MLS-technieken.

De eerste methode is beperkt tot de analyse van torsietrillingen. De tweede


techniek laat toe van de lagers en de vertanding in de aandrijflijn op een realis-
tischere manier voor te stellen. De generieke implementatie van deze techniek
kan gebruikt worden voor de modellering van schuine en rechte vertanding in
zowel parallelle als planetaire tandwieltrappen. De derde methode is de flexi-
bele MLS-techniek, die naast de globale beweging van de componenten in de
aandrijflijn ook inzicht biedt in hun elastische vervorming.

Het proefschrift illustreert de toepassing van de drie simulatietechnieken voor


de analyse van een enkelvoudige tandwieltrap, van een volledige tandwielkast
en uiteindelijk van een aandrijflijn in een windturbine met inbegrip van de kop-
peling met de toren, de rotor en de generator. Deze analyses zijn gericht op het
laagfrequente en middenfrequente gebied en tonen aan hoe resonantie van de
aandrijflijn in dit gebied kan geı̈dentificeerd worden. De gesimuleerde resul-
taten worden uiteindelijk vergeleken met de resultaten van een unieke meet-
campagne op een moderne windturbine van de multi-MW range.

V
VI
Abstract

This dissertation describes the development of a consistent modelling approach


to correctly describe the dynamic behaviour of a complex drive train in a wind
turbine, focussing on the gearbox. This research is motivated by the limita-
tions in the traditional design codes for wind turbines, which imply a quasi-
static design of all drive train components and yield insufficient insight in load
variations and local stress levels. The flexible multibody system (MBS) formu-
lation is chosen as the best alternative for the development of a more detailed
drive train model. This study presents a generic methodology based on three
MBS modelling approaches.

The first approach is limited to the analysis of torsional vibrations only. The
second technique offers a more realistic representation of the bearings and the
gears in the drive train and its generic implementation can be used for both
helical and spur gears in parallel and planetary gear stages. The third method
is the extension to a flexible MBS analysis, which yields information about
the elastic deformation of the drive train components in addition to their large
overall rigid-body motion.

The dissertation demonstrates the application of all three simulation methods


for the analysis of a single gear stage, of a complete gearbox and finally of
a drive train in a wind turbine, including coupling effects with the tower, the
rotor and the generator. These analyses cover the low- and mid-frequency
range and indicate how possible drive train resonances can be identified in this
range. Finally, the simulated results are compared with the results of a unique
measurement campaign on a modern multi-MW wind turbine.

VII
VIII
Contents

Voorwoord I

Samenvatting V

Abstract VII

List of symbols XV

Table of contents XV

1 Introduction 1
1.1 From a quasi-static design towards a dynamic approach in the
design of a wind turbine’s drive train . . . . . . . . . . . . . . 1
1.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Overview of the dissertation . . . . . . . . . . . . . . . . . . 4

2 Introduction to modern wind turbines 7


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Historical evolution . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 State-of-the-art in modern wind turbines . . . . . . . . . . . . 11
2.3.1 Extracting energy from the wind . . . . . . . . . . . . 11
2.3.1.1 Lift and drag forces . . . . . . . . . . . . . 11
2.3.1.2 The wind resource . . . . . . . . . . . . . . 15
2.3.2 General description of a modern wind turbine . . . . . 20
2.3.2.1 Rotor blades . . . . . . . . . . . . . . . . . 21
2.3.2.2 Nacelle . . . . . . . . . . . . . . . . . . . . 24
2.3.2.3 Tower and its foundation . . . . . . . . . . 26
2.3.3 Drive train layout in a modern wind turbine . . . . . . 28
2.3.3.1 Generator . . . . . . . . . . . . . . . . . . 29
2.3.3.2 Drive train concepts . . . . . . . . . . . . . 33
2.3.4 Power control aspects . . . . . . . . . . . . . . . . . 35
2.3.4.1 Passive stall control . . . . . . . . . . . . . 36

IX
X Table of contents

2.3.4.2 Active stall control . . . . . . . . . . . . . . 36


2.3.4.3 Pitch control . . . . . . . . . . . . . . . . . 37
2.4 Current status of wind powered electricity . . . . . . . . . . . 38
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 State-of-the-art in the design of a wind turbine drive train 45


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Design specifications of a wind turbine . . . . . . . . . . . . . 45
3.3 Traditional wind turbine design codes . . . . . . . . . . . . . 49
3.3.1 Principle . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.2 Existing software codes and their validation . . . . . . 54
3.3.3 Structural model in the traditional codes . . . . . . . . 59
3.3.3.1 Clarification of the different DOFs . . . . . 60
3.3.3.2 Results of normal modes calculations . . . . 64
3.3.3.3 Consequences for a traditional drive train de-
sign . . . . . . . . . . . . . . . . . . . . . 67
3.4 Drive train design loads . . . . . . . . . . . . . . . . . . . . . 68
3.4.1 Limitations of the traditional wind turbine design codes 68
3.4.1.1 Dynamic loads . . . . . . . . . . . . . . . . 68
3.4.1.2 Detailed component loads . . . . . . . . . . 70
3.4.1.3 Safety factors . . . . . . . . . . . . . . . . 71
3.4.2 New approach: the MBS formulation . . . . . . . . . 71
3.4.3 Overview of similar research activities . . . . . . . . . 74
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4 Detailed modelling of the drive train in a wind turbine 81


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Load transfer in the drive train . . . . . . . . . . . . . . . . . 82
4.2.1 General drive train layout . . . . . . . . . . . . . . . 82
4.2.2 Gearbox loads . . . . . . . . . . . . . . . . . . . . . 85
4.3 Drive train modelling techniques . . . . . . . . . . . . . . . . 91
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 91
4.3.2 Review of the literature on gear dynamics . . . . . . . 94
4.3.2.1 Tooth contact . . . . . . . . . . . . . . . . 96
4.3.2.2 Bearings . . . . . . . . . . . . . . . . . . . 101
4.3.2.3 Component flexibilities . . . . . . . . . . . 102
4.3.2.4 Damping in the drive train . . . . . . . . . . 104
4.3.3 Summary and conclusions . . . . . . . . . . . . . . . 105
4.4 Purely torsional multibody models . . . . . . . . . . . . . . . 107
4.5 Rigid multibody models with discrete flexible elements . . . . 109
4.5.1 Modelling of the bearing flexibility . . . . . . . . . . 110
4.5.2 Modelling of the tooth contact forces . . . . . . . . . 110
Table of contents XI

4.6 Flexible multibody models . . . . . . . . . . . . . . . . . . . 116


4.6.1 Component mode synthesis . . . . . . . . . . . . . . 116
4.6.1.1 Selection of component mode sets . . . . . 117
4.6.1.2 Substructure coupling . . . . . . . . . . . . 122
4.6.2 Inclusion of a flexible body in DADS . . . . . . . . . 124
4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

5 Analysis of parallel and planetary gear stages 131


5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2 Parallel helical gear pair . . . . . . . . . . . . . . . . . . . . 132
5.2.1 Torsional model . . . . . . . . . . . . . . . . . . . . 133
5.2.2 Rigid multibody model . . . . . . . . . . . . . . . . . 134
5.2.2.1 Influence of the bearing flexibilities . . . . . 137
5.2.2.2 Influence of the helix angle . . . . . . . . . 137
5.3 Planetary gear stages . . . . . . . . . . . . . . . . . . . . . . 138
5.3.1 Torsional model . . . . . . . . . . . . . . . . . . . . 139
5.3.2 Rigid multibody model . . . . . . . . . . . . . . . . . 142
5.4 Gearbox with a parallel and a planetary gear stage . . . . . . . 147
5.5 Torsional stiffness of a gearbox . . . . . . . . . . . . . . . . . 149
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

6 Analysis of the drive train in a modern wind turbine 155


6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
6.2 Drive train layout . . . . . . . . . . . . . . . . . . . . . . . . 156
6.3 Individual gear stages . . . . . . . . . . . . . . . . . . . . . . 158
6.3.1 Parallel stage . . . . . . . . . . . . . . . . . . . . . . 158
6.3.1.1 Modelling . . . . . . . . . . . . . . . . . . 158
6.3.1.2 Analysis . . . . . . . . . . . . . . . . . . . 163
6.3.2 High speed planetary stage . . . . . . . . . . . . . . . 167
6.3.2.1 Modelling . . . . . . . . . . . . . . . . . . 167
6.3.2.2 Analysis . . . . . . . . . . . . . . . . . . . 171
6.3.3 Low speed planetary stage . . . . . . . . . . . . . . . 173
6.4 Complete gearbox . . . . . . . . . . . . . . . . . . . . . . . . 174
6.4.1 Static analysis . . . . . . . . . . . . . . . . . . . . . 175
6.4.2 Dynamic analysis . . . . . . . . . . . . . . . . . . . . 176
6.5 Drive train integrated in the wind turbine . . . . . . . . . . . . 180
6.5.1 Model without rotor and tower . . . . . . . . . . . . . 180
6.5.2 Model including rotor and tower . . . . . . . . . . . . 182
6.5.2.1 Normal modes analysis . . . . . . . . . . . 183
6.5.2.2 Frequency response analysis . . . . . . . . 185
6.5.2.3 Simulation of a transient load case . . . . . 191
6.5.3 Features of the new simulation approach . . . . . . . . 195
XII Table of contents

6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

7 Measurement campaign on a modern wind turbine 203


7.1 Overview of the measurement campaign . . . . . . . . . . . . 203

8 General conclusions 207


8.1 Overview and main contributions . . . . . . . . . . . . . . . . 207
8.2 Recommendations for future research . . . . . . . . . . . . . 214

Bibliography 217

Curriculum Vitae 237

List of publications 239


International peer reviewed journal articles . . . . . . . . . . . 239
Full papers in proceedings of international conferences . . . . 239
Full papers in proceedings of national conferences . . . . . . . 240

A Drive train modes in traditional wind turbine design codes 241


A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
A.2 FE model of the wind turbine . . . . . . . . . . . . . . . . . . 242
A.2.1 Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . 242
A.2.2 Tower . . . . . . . . . . . . . . . . . . . . . . . . . . 244
A.2.3 Wind turbine . . . . . . . . . . . . . . . . . . . . . . 245
A.3 An industrial approach to calculate the 1st drive train mode . . 249
A.4 Sensitivity analyses . . . . . . . . . . . . . . . . . . . . . . . 252
A.4.1 Sensitivity analyses for the FE model . . . . . . . . . 252
A.4.1.1 Rotor inertia and flexibility . . . . . . . . . 252
A.4.1.2 Drive train stiffness . . . . . . . . . . . . . 256
A.4.1.3 Generator inertia . . . . . . . . . . . . . . . 257
A.4.2 Flexible multibody model of the wind turbine . . . . . 258
A.4.3 Sensitivity analyses for the flexible multibody model . 260
A.4.3.1 Tower flexibility . . . . . . . . . . . . . . . 260
A.4.3.2 The coupling between drive train and tower 261
A.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 264

B Eigenmodes and eigenfrequencies of modern wind turbines 267

C Gear ratio of a planetary gear with a fixed ring wheel 273

D Tooth contact forces for a helical gear pair 275


Table of contents XIII

E Numerical calculations in DADS 277


E.1 Normal modes calculation . . . . . . . . . . . . . . . . . . . 277
E.2 Simulation of motion and loads . . . . . . . . . . . . . . . . . 278
E.2.1 Computational time . . . . . . . . . . . . . . . . . . . 279
E.3 Static analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 280

Nederlandse samenvatting I
1 Inleiding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I
1.1 Situatieschets en probleembeschrijving . . . . . . . . I
1.2 Doelstelling van het onderzoek . . . . . . . . . . . . . III
1.3 Overzicht van het proefschrift . . . . . . . . . . . . . III
2 Ontwerp van de aandrijflijn in een windturbine . . . . . . . . IV
3 Modelleren van de aandrijflijn in een windturbine . . . . . . . VII
3.1 Torsionele meerlichamen-systemen . . . . . . . . . . IX
3.2 Meerlichamen-systemen met discrete flexibele elementen XI
3.3 Flexibele meerlichamen-systemen . . . . . . . . . . . XII
4 Analyse van een planetaire tandwieltrap . . . . . . . . . . . . XIII
5 Analyse van de aandrijflijn in een moderne windturbine . . . . XVII
5.1 Een flexibel meerlichamen-model van de parallelle trap XVIII
5.2 “Out-of-plane” modes van een planetaire tandwieltrap XXI
5.3 Model van de volledige windturbine . . . . . . . . . . XXI
5.3.1 Identificatie van de eigenmodes . . . . . . . XXIII
5.3.2 Frequentie-respons analyse . . . . . . . . . XXIII
5.3.3 Simulatie van een transiënt belastingsgeval . XXVI
6 Meetcampagne op een moderne windturbine . . . . . . . . . . XXVII
6.1 Overzicht van de meetcampagne . . . . . . . . . . . . XXVII
7 Algemene conclusies . . . . . . . . . . . . . . . . . . . . . . XXVIII
XIV
Symbols and abbreviations

General symbols:
˙ First derivative of 
¨ Second derivative of 
{} Vector
[] Matrix
[]T Transpose of a matrix
[]−1 Inverse of a matrix
i j Element (i, j) of the matrix 
i Element i of the vector , or column i of the matrix 
diag(x1 , x2 , . . . , xn ) Diagonal n × n matrix with x1 , x2 , . . . and xn on the
diagonal
∑ Summation
∞ Infinity
◦ Degrees
µm Micrometer

Wind (turbines):
a Axial flow induction factor
a0 Tangential flow induction factor
A Rotor swept area
c Chord length
CL Lift coefficient
CD Drag coefficient
CM Moment coefficient
Cp Power coefficient
D Drag force
fgrid Frequency of the electricity grid
fW Weibull distribution
H Height above the ground

XV
XVI List of symbols

Hhub Hub height


L Lift force
N0 Hours per year
np Pull out speed of the generator
nr Rotational speed of the rotor in the generator
ns Synchronous speed of the generator
Pmax Maximum power in the wind
pp Number of pole pairs of the generator
Re Reynolds number
Rrot Radius of the rotor of the wind turbine
s Slip of the generator
Tp Pull out torque of the generator
V10 10-minute mean wind speed
Va Axial wind speed at the blade profile
Vcut−in Cut-in wind speed
Vcut−out Cut-out wind speed
Vmean Annual mean wind speed at hub height for a specific
site
Vo Undisturbed wind speed
Vrat Rated wind speed: start of rated power production
Vre f Reference wind speed used in the IEC61400-1 standard
Vrel Relative wind speed at the blade profile
Vrot Tangential wind speed at the blade profile

α Angle of attack
αh Exponent of the power law to describe a normal wind
profile
ηm Mechanical efficiency
ηe Electrical efficiency
θ Local pitch angle of the blade profile
θcontrol The setting of the blade pitch angle in active stall or
pitch controlled wind turbines
λ Tip speed ratio
ν Kinematic viscosity
ρ Air density
σV Standard deviation of the 10-minute mean wind speed
φ Flow angle
ωrot Rotational speed of the rotor of the wind turbine
List of symbols XVII

Modelling (general):
[C] Damping matrix
Cgenerator Generator damping
{f} Set of force vectors
[K] Stiffness matrix
[M] Mass matrix
{x} Vectors containing the DOFs of all nodes
Igenerator Inertia of the generator of the wind turbine
Irotor Inertia of the rotor of the wind turbine
KDT Drive train stiffness
Kshaft Torsional stiffness of a shaft

Gears:
as Distance between the axes of two gears
d Diameter pitch circle
d0 Diameter operating pitch circle
db Diameter base circle
db0 Diameter operating base circle
e(t) Transmission error
fgm Gear mesh frequency


Fb Load on the bearing
Fbn Tooth contact force
Ft Tangential component of the tooth contact force
Fr Radial component of the tooth contact force
cgear Gear mesh damping
I Mass moment of inertia
igear Gearbox ratio
J Polar moment of inertia
Kb Bearing stiffness
kaxial Axial bearing stiffness
Kgear Torsional stiffness of a gear stage
kradial Radial bearing stiffness
ktilt Tilt bearing stiffness
m Multiplicity of an eigenmode
m Gear module
ngear Rotational speed of the gear
ncarrier Rotational speed of the planet carrier
r Radius pitch circle
r0 Radius operating pitch circle
rb Radius base circle
rb0 Radius operating base circle
XVIII List of symbols

T Torque
z Number of teeth of the gear
z planet Number of teeth of the planet
zrw Number of teeth of the ring wheel
zsun Number of teeth of the sun

αa Pressure angle in the longitudinal plane


α0a Actual pressure angle in the longitudinal plane
αn Pressure angle in the normal plane
α0n Actual pressure angle in the normal plane
αt Pressure angle in the transversal plane
αt0 Actual pressure angle in the transversal plane
β Helix angle referred to the pitch circle
β0 Helix angle referred to the operating pitch circle
βb Helix angle referred to the base circle
δ Gear deformation along the contact line
ζ Modal damping
κgear Gear mesh stiffness
η Viscosity of the lubrication fluid
ηgb Gearbox efficiency
vt Peripheral speed at the pitch circle
ω Eigenfrequency [rad/sec]
ωin Rotational speed of the input shaft
ωout Rotational speed of the output shaft
ω pc Rotational speed of the planet carrier
ωsun Rotational speed of the sun

Component mode synthesis:


cc Set of generalised constraint coordinates
ck Set of generalised normal coordinates
cs Set of generalised coordinates
Ni Number of physical interior DOFs
Nj Number of physical interface DOFs
Ns Number of physical DOFs
qs Number of generalised coordinates
Ts Transformation matrix
u Unit vector
yi Set of physical interior coordinates
yj Set of physical interface coordinates
ys Set of physical coordinates
List of symbols XIX

Λnn Matrix of squared eigenvalues


Φk Kept component normal modes
Φki Kept component normal modes corresponding to the
interior DOFs
Φn Component normal modes
ψa,1 Attachment component mode at interface 1
ψa, j Attachment component mode at interface j
ψa,N j Attachment component mode at interface N j
ψc,1 Constrained component mode at interface 1
ψc, j Constrained component mode at interface j
ψc,N j Constrained component mode at interface N j
Ψa Matrix of attachment component modes
Ψc Matrix of constrained component modes
Ψci Matrix of constrained component modes correspond-
ing to the interior DOFs

Abbreviations:
2D Two-dimensional
3D Three-dimensional
AC Alternating Current
ADAMS/WT Automatic Dynamic Analysis of Mechanical Systems /
Wind Turbine
AEO Annual Energy Output
BC Base Circle
BEM method Blade Element Momentum method
CAD Computer-Aided Design
CADSI Computer-Aided Design Software, Inc.
CFD Computational Fluid Fynamics
CMS Component Mode Synthesis
CP Contact Point
CPU Central Processing Unit
CRES Centre for Renewable Energy Sources
DADS Dynamic Analysis and Design System
DAE Differential Algebraic Equation
DAQ Data-Acquisition
DC Direct Current
DFIG Doubly Fed Induction Generator
DHAT Dynamic analysis of Horizontal Axis Turbines
DOF(s) Degree(s) Of Freedom
DRESP DrehschwingungsSimulation-Programm
DTU Technical University of Denmark
XX List of symbols

DUWECS Delft University Wind Energy Converter Simulation


Program
DYLA Dynamische Lagerkräfte
ECN Energy Research Centre of the Netherlands
ECG Extreme Coherent Gust
EDC Extreme Direction Change
EE Eindige Elementen
EOG Extreme Operating Gust
EWS Extreme Wind Shear
FAST Fatigue, Aerodynamics, Structures
FE Finite Element
Flexlast Flexible Load Analyzing Simulation Tool
FVA Forschungsvereinigung Antriebstechnik e.V.
FRP Fibre Reinforced Plastic
FRF Frequency Response Function
GAST General Aerodynamic and Structural prediction Tool
for wind turbines
GH Garrad Hassan & Partners Ltd.
GL Germanischer Lloyd
GWEC Global Wind Energy Council
HAWC Horizontal Axis Wind turbine Code
HAWT Horizontal Axis Wind Turbine
HSS High Speed Shaft
IME Institut für Maschinenelemente und Maschinengestal-
tung
LMGK Lehrstuhl für Maschinenelemente, Getriebe und Kraft-
fahrzeuge
LOC Line Of Contact
LSS Low Speed Shaft
LTI Linear Time Invariant
LTV Linear Time-Varying
MAC Modal Assurance Criterion
MBS Multibody System
MBM Multibody Model
MLS Meerlichamen-Systeem
NREL National Renewable Energy Laboratory
NTM Normal Turbulence Model
NTUA National Technical University of Athens
NWP Normal Wind Profile
ODE Ordinary Differential Equation
List of symbols XXI

PHATAS Program for Horizontal Axis wind Turbine Analysis


and Simulation
PSD Power Spectral Density
RMS Root-Mean-Square
RPM Revolutions Per Minute
RWTH Rheinisch Westfälische Technische Hochschule
VAWT Vertical Axis Wind Turbine
VEWTDC Verification of European wind turbine design codes
WTGS Wind Turbine Generator System
XXII
1

Introduction

1.1 From a quasi-static design towards a dynamic ap-


proach in the design of a wind turbine’s drive train
In recent years wind energy has been a fast growing source of electrical power.
At the end of 2005, the global installed wind power capacity reached 59 GW,
of which about 20 % had been installed in that year. The wind turbine industry
is booming and is spending a lot on research and development for further im-
proving the performance of its present machines and increasing their capacity.
Modern wind turbines are already enormous with tower heights up to 100 m,
blade lengths of 60 m and a total capacity up to 5 MW. They operate further-
more in complex conditions, determined by a turbulent wind field, by possible
disturbances in the electricity grid and by the behaviour of sea waves for off-
shore turbines. Guaranteeing the structural integrity of these machines during
a lifetime of 20 years is an enormous challenge.

In their design calculations, the wind turbine manufacturers use dedicated sim-
ulation codes1 to predict the load levels and variations on all components in
their machines. The structural model of the wind turbine in these simulations
usually contains sufficient detail to accurately describe the dynamic loads on
the rotor and the tower. However, for the representation of the complete drive
train, only one degree of freedom (DOF) is considered in the traditional codes.
This imposes considerable limitations on the reliability of the drive train de-
sign.

1 The word “codes” in the present dissertation refers to the software codes used in the design

of a wind turbine for the simulation of loads as opposed to prescriptive design codes (e.g. from
ISO), which are further called “standards”.

1
2 1. Introduction

1. Since no load amplifications due to internal dynamics are possible in


these simulations, the design of all drive train components is quasi-static.
Therefore, the use of specific load application factors is recommended
in the traditional design process. A dynamic modelling approach is re-
quired in order to accurately predict the dynamic loads on all compo-
nents and to assess the sufficiency or overestimation of the application
factors.

2. The focus in this dissertation is on wind turbines including a gearbox


in the drive train. The design of gears and bearings in a gearbox is
determined by fatigue calculations. These calculations require insight in
local stresses, which is only possible with detailed models. This need
is further confirmed by a series of component failures in wind turbine
gearboxes, which indicate an insufficient understanding of the design
loads.

3. The exploration of more integrated drive train concepts in new wind


turbine designs further emphasises the need for more insight in the drive
train dynamics. The traditional drive train model is too limited to yield
this insight. In addition, gaining insight from experimental experience
is very difficult, expensive and even impossible during the design phase.
As a result, the use of more detailed drive train models is required.

In order to fulfil the identified needs, a dynamic simulation approach is re-


quired instead of the traditional quasi-static design methods. More insight
from this approach will lead to a better description of all load cases and to
a more accurate prediction of design loads. This will finally result in more
cost-efficient wind turbines, by economising on needlessly over-dimensioned
designs or by avoiding component failures as a result of load underestima-
tion. This dissertation describes the development of a consistent modelling
approach to correctly describe the dynamic behaviour of a complex drive train
in a wind turbine. This work includes also the results of a unique measurement
campaign on a modern multi-MW wind turbine.

1.2 Research objectives


The study in this dissertation starts from the state-of-the-art in the design of
a wind turbine’s drive train. Therefore, the first objective is describing the
limitations of the traditional wind turbine design codes, with respect to the
prediction of drive train loads and the drive train behaviour more generally.
This requires an investigation of the structural model used in these calculation
methods and, furthermore, a listing of those issues that considerably influence
1.2 Research objectives 3

the design specifications for the drive train. A combination of the respective
consequences yields a list of limitations in the traditional design codes.

The outcome of the initial phase of this work serves as the input to the main
objective in this dissertation, which is the development of additional simula-
tion methods. The key question to be addressed in the development of these
methods is: “Which tools are necessary to guarantee a robust and cost-efficient
drive train design?” The new method should permit to:

1. determine the dynamic loads on all drive train components more accu-
rately

2. identify those transient phenomena which can be harmful to the gearbox,


in order to avoid them or to consider them in an appropriate way during
the design process

3. determine the level and variation of local stresses in the drive train com-
ponents

4. assess the redundancy or insufficiency of the traditional safety factors

An optimal target in the search for more advanced calculation methods is a


combination of, on the one hand, accuracy and, on the other hand, workabil-
ity, such as regarding time-efficiency and user-friendliness. This target is often
aimed at in various applications and is well summarised by Einstein’s quote:
“Make everything as simple as possible, but not simpler”.

Solving the key question requires the knowledge of and experience with vari-
ous modelling techniques in combination with experience from measurements
on a real wind turbine. These objectives are elaborated separately in this dis-
sertation. Chapter 4 gives an overview of existing methods in the literature
for the simulation of drive trains and gear dynamics more particularly. The
multibody system (MBS) formulation is chosen as the best alternative for the
development of a more detailed drive train model. Subsequently, it presents a
generic methodology based on three MBS modelling approaches. Chapters 5
and 6 demonstrate the use of these simulation methods for the analysis of a
single gear stage, a complete gearbox and finally a drive train in a wind tur-
bine. This gives insight in their application and focusses on the respective
limitations and capabilities. Chapter 7 describes the measurement campaign
and discusses the analyses of the experimental data. The next section gives a
more elaborate overview of this dissertation.
4 1. Introduction

1.3 Overview of the dissertation


This dissertation comprises eight chapters. Chapter 2 gives an introduction
to modern wind turbines. It starts with a brief summary of their historical
evolution and continues with an explanation of their working principle. This
includes a description of the state-of-the-art in modern wind turbine technol-
ogy, with special attention to the drive train layout and power control aspects.
Additionally, it discusses the current status of wind powered electricity.

Chapter 3 describes the state-of-the-art in the design of a drive train in a wind


turbine. It gives an overview of various existing wind turbine design codes
and presents the structural model, which is used in these codes. In addition,
it indicates the limitations in this structural model for the design of a drive
train. These limitations imply the need for additional simulation methods.
This chapter justifies the choice of the MBS approach to develop these meth-
ods. It concludes with a brief overview of similar research activities.

Chapter 4 contains the essence of the dissertation, since it describes how the
limitations in the traditional design codes are tackled. It starts with a descrip-
tion of the load transfer in a drive train in general, and in a gearbox more
particularly. Subsequently, it gives an overview of the state-of-the-art in mod-
elling gear dynamics. This includes various mathematical models to describe
the load transfer, the flexibility and the damping in the tooth contact as well as
in the bearings and other drive train components. The remainder of this chap-
ter describes three modelling approaches with a gradual increase in the level
of complexity.

1. The first approach is based on the state-of-the-art and includes one DOF
for each individual body. The corresponding models are called purely
torsional multibody models. This approach is a logic extension of the
traditional drive train model, which has only one torsional DOF for the
complete drive train.

2. The second type of model includes six DOFs per body and is a so-called
rigid multibody model with discrete flexible elements. This approach
starts from existing techniques for the analysis of a helical parallel gear
system and a spur planetary gear system. These techniques are com-
bined and further developed to model helical planetary gear stages, mul-
tistage gearboxes and, finally, a complete drive train integrated in a wind
turbine. The additional developments and their formulation as a generic
methodology are important contributions to the state-of-the-art.
1.3 Overview of the dissertation 5

3. The third modelling approach is using a flexible multibody model. Such


a model includes the six DOFs per body, as in the second type of model,
and an additional set of DOFs to represent its elastic deformation. This
latter set is calculated by applying the component mode synthesis (CMS)
technique on FE models of the individual bodies. The combination of
the mathematical expressions to simulate gear systems and the CMS
technique is an important step forward in the state-of-the-art of mod-
elling gear dynamics, since it yields local stress levels in the flexible
bodies.

Chapter 5 demonstrates the application of the first and second modelling tech-
nique for the analysis of two gear systems, which originate from the literature.
The first example is a helical parallel gear system and the second a spur plan-
etary gear system. A comparison of the eigenfrequencies and eigenmodes for
these systems with the results from the literature aims at a numerical verifi-
cation of the modelling techniques. Moreover, a discussion of the calculated
results yields valuable insight in the use of these techniques. This chapter
discusses furthermore the analysis of a gearbox model, which consists of a
combination of the two gear systems. Finally, it demonstrates how the MBS
formulation can offer an alternative for large FE models in a static analysis.

Chapter 6 demonstrates the actual application of the three developed tech-


niques for the analysis of a wind turbine drive train. This drive train includes
a three stage gearbox and the discussion starts with the analyses of the indi-
vidual gear stages. These are a helical parallel gear stage, a helical planetary
gear stage and a spur planetary gear stage. The static and dynamic analysis
of a complete gearbox model is the subsequent step in this study. The further
extension of this model towards a complete wind turbine model consists of
two more steps. Firstly, the influence of adding the generator is investigated
and, afterwards, the effect of including the rotor and the tower. The dynamic
analysis of this comprehensive model includes:

1. a normal modes analysis for the determination of eigenmodes and eigen-


frequencies in the drive train

2. a response calculation for an external excitation at the generator side and


an internal excitation from the meshing gears

3. a Campbell analysis for the identification of possible resonance behaviour

4. a simulation of a transient load case


6 1. Introduction

Chapter 7 gives an overview of a measurement campaign on a multi-MW


wind turbine. The focus in the measurements is on the drive train behaviour
and the elaborateness of this campaign makes it extremely valuable. The col-
lection of all acquired experimental data is a unique database. This chapter
includes the analyses of the experimental data. Firstly, it discusses the char-
acteristics of the wind turbine, which are derived from measurements during
normal operation. Subsequently, it presents the drive train behaviour during
various transient load cases. Finally, this chapter describes an experimental
validation of the numerical models.

Chapter 8 summarises the general conclusions of the present research and


highlights the main contributions. In addition, it lists some suggestions for
future research.
2

Introduction to modern wind


turbines

2.1 Introduction
This chapter starts with a description of the historical evolution of modern wind
turbines. The overview goes back to the earliest publication on the predeces-
sors of wind turbines, which are the windmills. These wind driven machines
were mainly used for grinding grain and pumping water. Their evolution led
to the application of wind power for generating electricity near the end of the
19th century. The windmills used for producing electricity are further called
wind turbines and section 2.3 describes the current state-of-the-art for these
machines. This starts with an explanation of the aerodynamic principles of
extracting energy from the wind. Section 2.3.2 continues with a general de-
scription of a modern wind turbine. This elaborates on the rotor blades, the
nacelle and the tower of a wind turbine. The drive train in a wind turbine
converts the mechanical energy at the rotor hub into electrical energy in the
generator. Section 2.3.3 distinguishes between generators in a direct and an
indirect grid connection and describes how this leads to various drive train
concepts. In addition, three main types of control may be applied to regulate
the power production of a wind turbine. Section 2.3.4 explains the principles
of these controls. The final section in this chapter presents the current status of
wind powered electricity and gives an overview of the wind turbine market.

2.2 Historical evolution


Wind is a fascinating energy source which inspired already people from the an-
cient times to use it in their sailing boats on the River Nile [217]. Its first appli-
cation as a force driving a windmill is estimated in the 7th century A.D. and is

7
8 2. Introduction to modern wind turbines

described by Dennis G. Shepherd, “Historical Development of the Windmill”,


pp. 1-46 in “Wind Turbine Technology”, D.A. Spera (Editor) [203]. Shepherd
gives an overview of the historical development of the windmill, which starts
in the 7th century with a windmill in Persia. Windmills were used for grinding
grain and for pumping water. Their introduction in Europe in the 12th cen-
tury is supposedly done by traders and crusaders returning from the Far East.
Based on descriptions and drawings, the horizontal-axis windmill with a ro-
tor perpendicular to the wind direction, is believed to be the most widespread
type from those ages. In spite of the temporal absence of wind from time to
time, the many appropriate sites for these mills allowed much more flexibility
of application than did water mills. As a consequence, the usage of windmills
became a major source of power since the 14th century. Furthermore, the va-
riety of applications grew, such as for instance the use of windmills in The
Netherlands to drain the Dutch polders or for sawing long timbers into small
planks.

This evolution went hand in hand with a continuous progress in windmill tech-
nology, driven by inventive millers to improve the efficiency of the mills and to
ease their heavy work. The most important improvements occurred during the
Industrial Revolution. Firstly, the invention of the fantail in 1745 enabled the
windmill to follow the wind direction automatically. This turning movement,
which is called yawing, was done manually before. The patent sail from 1806
is a second invention which could keep the mill speed reasonably constant by
automatically adapting the sail setting. The sail is the cloth that was used as
wind catching material on the rotor blades. In addition, the patent sail could
be used as an aerodynamic brake.

The mature windmill design led to an increased use of wind energy in the
United States from the mid-19th century. The power of a windmill was mainly
used for pumping water to houses and on farms all over the country. Figure 2.1,
from Baker [10], shows some of the many different designs of the American
windmill. Their production peaked in the 1920s. Meanwhile, the introduction
of cheap fossil fueled engines and electricity lines was growing, which led to
a gradual transition to other water pumps and a decline of the use of windmills
in the 1930s. Although, windmills for pumping water kept existing and are
still used today in remote locations.
2.2 Historical evolution 9

Figure 2.1: Some of the many different designs of the American windmill,
which became popular water pumps in the United States at the end of the 19th
century, reproduced from Baker [10].

(a) The 12 kW DC Brush windmill (1888): a wind (b) A 4-bladed wind turbine erected
turbine with 144 blades and 17 m rotor diameter. by Poul La Cour for generating DC
electricity in Denmark (1897).

Figure 2.2: The first wind turbines, reproduced from [220].


10 2. Introduction to modern wind turbines

Near the end of the 19th century, electricity came into use and interest devel-
oped in using wind power for its generation. The windmills connected to an
electric generator are further called wind turbines and a thorough description
of their evolution is given in [220]. Charles F. Brush used the American wind-
mill concept in the first wind turbine ever built [37]. In 1888, he erected this
Brush windmill in Cleveland, Ohio: figure 2.2(a) shows this 12 kW DC tur-
bine, which was used to supply power for charging storage batteries.

Poul La Cour was another wind turbine pioneer living in Denmark, who used
less rotor blades at a higher rotation speed, which yielded a higher efficiency
for electricity generation. Figure 2.2(b) shows one of his test turbines from
1897. Despite the work of these two pioneers, wind powered electricity knew
only few applications during the first half of the 20th century. Whenever there
was fuel scarcity, such as during the two world wars, its application regained
interest. As a result of the collaboration between Palmer C. Putnam and the
Smith Company in the United States, the Smith-Putnam wind turbine was
erected in 1941 [203]. Figure 2.3 shows this 1.25 MW 2-bladed prototype
with a steel rotor, 53 m in diameter. This AC power supply remained the
largest wind turbine ever built for some 40 years.

Figure 2.3: The 1.25 MW Smith-Putnam wind turbine (1941), reproduced


from [203].
2.3 State-of-the-art in modern wind turbines 11

After the second world war, the engineer Johannes Juul, who was a former stu-
dent of Poul La Cour, made some valuable contributions to the wind turbine
technology. In the mid 1950s he introduced the innovative 200 kW Gedser
wind turbine. This three-bladed upwind stall-regulated wind turbine ran with a
constant rotational speed connected to an AC asynchronous generator and pro-
duced electricity for 11 years without maintenance [220]. It was a pioneering
design for modern wind turbines, which was later called the Danish concept, a
reference that is still in use.

The oil crisis of 1973 caused a stimulation of research and subsidiary pro-
grammes in countries, which wanted to be less dependent on oil imports [37].
Research was done on large prototypes, but the development of several 50 kW
wind turbine designs in the beginning of the 1980s is seen as the actual techno-
logical and industrial breakthrough and, thus, the start of the professional wind
turbine industry. During the Californian wind support programme, more than
1000 wind turbines were installed in Palm Springs only [220] and, at its end
in 1985, more than 1000 MW capacity was installed in the whole state. About
half of this capacity originated from Danish manufacturers, which still occupy
a prominent position on today’s market (cfr. section 2.4). Since the mid 1980s,
the wind turbine industry further evolved: the capacity of commercially avail-
able wind turbines increased up to 5 MW in 2005 and the cumulative installed
capacity worldwide at the end of 2005 reached 59.3 GW [81].

2.3 State-of-the-art in modern wind turbines


2.3.1 Extracting energy from the wind
2.3.1.1 Lift and drag forces
The extraction of energy from the wind is done by using the reaction forces
on a structure placed in the wind flow. Figure 2.4 shows the reaction forces
acting on an aerofoil in a two-dimensional (2D) flow. α represents the angle of
attack, which is the angle between the relative wind speed Vrel at the aerofoil
and the chord line. The consideration of the flow in only two directions is
valid, when the span wise velocity component is small compared to the stream
wise component [87]. Generally, this assumption is valid for wind turbines,
which have long and slender blades. The reaction force F on the aerofoil is
split up in a lift force L, which is perpendicular to the relative wind speed, and
a drag force D, which is parallel to this wind speed. These forces are written
as:
12 2. Introduction to modern wind turbines

Figure 2.4: Definition of the aerodynamic forces on an aerofoil.

1
L= · ρ · Vrel
2
· c ·CL (2.1)
2
1
D = · ρ · Vrel 2
· c ·CD (2.2)
2
where ρ is the air density and c is the chord length of the aerofoil. CL and CD
are called the lift and drag coefficients. In addition to the reaction force F, the
flow around the aerofoil also causes a moment M. This moment is positive as
indicated in figure 2.4 and is written as:
1
M= · ρ · Vrel
2
· c2 ·CM (2.3)
2
CM is called the moment coefficient and, like CL and CD , it is a function of:
• α, the angle of attack

• the aerofoil shape

• Re, the Reynolds number based on the chord and Vrel (Re = c · Vrel /ν,
where ν is the kinematic viscosity)

• the Mach number, which is the ratio between Vrel and the speed of sound
For low wind speeds, such as around wind turbine blades, the influence of the
Mach number is negligible. Consequently, the specifications of CL , CD and CM
for a given aerofoil can be given as a function of Re and α. The values for
these coefficients are often experimentally defined during wind tunnel tests. A
lot of aerofoil shapes have been analysed and reported in this way, mainly for
aircraft applications. Here, a high CL /CD is desired, yielding high lift forces for
lifting the aircraft and low drag forces, which counteract the aircraft motion.
2.3 State-of-the-art in modern wind turbines 13

Figure 2.5: A HAWT with an upwind three-bladed rotor: the cross section of
one blade at radius r shows the local wind flow and the resulting forces.

All modern electricity generating wind turbines demand for aerofoils with the
same characteristics, because they also use the lift forces on the blades to drive
the rotor. The drag forces are unwanted loads, which need to be as small as pos-
sible. Because of the appropriate specifications of aircraft aerofoils for wind
turbine applications, these aerofoils have been used frequently for wind tur-
bine blades. Especially at the start of the modern wind turbine era, their blade
profiles were copied from aerofoil shapes of aircrafts. Meanwhile, the wind
turbine industry has become more mature and has developed its own range of
aerofoils with more dedicated qualities.

Figure 2.5 sketches the lift principle for a typical modern wind turbine. It is a
horizontal axis wind turbine (HAWT) with a three-bladed upwind rotor, which
rotates clockwise when looking in the wind direction (cfr. section 2.3.2). The
flow around the blade profile at radius r is shown for a cross section of one of
the blades. The relative wind speed at the blade profile (Vrel ) is decomposed of
a component lying in the rotor plane (Vrot ) and one perpendicular to it (Va ). As
introduced above, the local angle of attack α is the angle between the chord
line and Vrel . The angle between the rotor plane and Vrel is the flow angle φ and
θ is the local pitch angle of the blade profile. These angles can be written as:

α = φ−θ (2.4)
 
Va
φ = tan−1 (2.5)
Vrot
14 2. Introduction to modern wind turbines

Equations (2.4) and (2.5) yield the local angle of attack for known values of Va
and Vrot . These wind speeds can be calculated based on the undisturbed wind
speed Vo at radius r and the rotational speed ωrot of the rotor:
Va = Vo − aVo = (1 − a) ·Vo (2.6)
Vrot = ωrot r + a ωrot r = (1 + a ) · ωrot r
0 0
(2.7)
a is called the axial flow induction factor: the wind speed component aVo
lies opposite to the direction of Vo and represents a drop in the wind speed
at the rotor, which can be derived based on the laws of conservation of mass
and conservation of energy, as described by Burton et al [37]. a0 is called the
tangential flow induction factor: the wind speed component a0 ωrot r is induced
by the vortex system [87] and lies opposite to the rotation of the rotor blades.
It is clear from equations (2.6) and (2.7) that the knowledge of a and a0 yield
the necessary values to calculate the local angle of attack α. In addition, the
knowledge of α and the lift and drag coefficients CL (α) and CD (α) for the
blade profile at radius r, permits to calculate the forces at the cross section,
according to equations (2.1) and (2.2). This procedure can be repeated at all
cross sections of a blade and an integration of the calculated force distribution
along the span, yields the global aerodynamic loads. Thus, the determination
of the loads starts with the calculation of the induction factors a and a0 , which
is the purpose of the blade element momentum (BEM) method. This method is
based on a theory of Glauert [77, 78] and is described in detail by Hansen [87]
and Burton [37], but is not further discussed here. Furthermore, an accurate
prediction of CL and CD is necessary for the load calculation and, therefore,
this is the subject for a lot of research. Figure 2.6 shows a typical example of
the lift and drag coefficients of an aerofoil as a function of α. The α domain is
split up in three parts.
1. The first part is called the pre-stall region. Here, CL increases quasi
linearly with α and CD is quasi constant. A prediction of these coefficient
values is mainly based on 2D wind tunnel measurements or on CFD
calculations.
2. Starting from a certain α value, the blade profile stalls. This is the stall
region, where CL further increases to a certain maximum and then drops.
The CD value increases monotonically after stall. The stall phenomenon
starts when the boundary layer of the flow starts to separate from the
upper side of the aerofoil. This phenomenon and its origin highly depend
on the aerofoil shape and the CL and CD values in the stall region are, in
general, results from computational fluid dynamics (CFD) calculations
with corrections for three-dimensional (3D) effects. The stall behaviour
of a wind turbine blade can be used as an automatic limitation for the
power production in certain wind turbines (cfr. section 2.3.4).
2.3 State-of-the-art in modern wind turbines 15

Figure 2.6: The lift and drag coefficient for an aerofoil, reproduced from [182].

3. At a certain α value, CL tends to stay constant, while CD increases fur-


ther. The determination of the coefficient values in this post-stall region
is typically based on measurements or estimation methods.
Note that the CL and CD values are heavily dependent on the aerofoil shape,
which can be influenced by the surface roughness of the blade. An incre-
sae in roughness has been noticed e.g. as insect contamination on the leading
edges of wind turbine blades, which yielded a considerable loss in power pro-
duction [42]. Therefore, blade manufacturers put a lot of effort in producing
smooth blade profiles, which are little affected by the environment.

2.3.1.2 The wind resource


The lift forces enable a wind turbine to transfer the kinetic energy in the wind
into mechanical energy at the rotor shaft. This energy is further converted into
electrical energy by a generator. The maximum available power flow in the
wind is Pmax :
1
Pmax = · ρ ·Vo3 · A (2.8)
2
which would be obtained when the undisturbed wind speed Vo would be re-
duced to zero by passing through the rotor. It is the product of the kinetic
energy in the wind 12 ρVo2 and the flow rate A · Vo through the rotor, where A
stands for the swept area of the rotor. Stopping the wind at the rotor would
imply that there is no more flow through the rotor and, thus, no energy extrac-
tion possible. The other extreme is no reduction of the wind speed at the rotor
16 2. Introduction to modern wind turbines

and, consequently, again no energy extraction. The German aerodynamicist


Albert Betz has proven in 1919 that there is a theoretical limit on the energy
extraction from the wind, using the concept of a wind turbine [21]. This Betz
limit is a theoretical maximum for the power coefficient Cp , which is the ratio
of the actual power obtained and the maximum available power. Cpmax equals
16/27 = 0.593 and the actual power produced by a wind turbine is written as:
1
P= · ρ ·Vo3 · A ·Cp (2.9)
2
Equation (2.9) shows that the actual power P varies linearly with the air den-
sity ρ, which depends on the air pressure, the air temperature and the humidity.

Moreover, P varies with the third power of the wind speed, which is there-
fore the most determining parameter in the evaluation of a future wind site. To
guarantee a good estimation of this wind speed, it is often measured during a
long period before a project is started up. The summary of such measurements
leads to a wind climate description, which is used afterwards in load calcu-
lations. In general, two statistical values represent the wind speed in such a
description:
1. the 10-minute mean wind speed V10 at the site
2. the standard deviation of V10 , which is denoted σV
During stationary wind climate conditions, these factors are assumed to re-
main constant for a 10-minute period [182]. The mean wind speed V10 will
vary from period to period. The long-term variation of this natural variabil-
ity can be represented by a probability density function. Observation of the
variation has shown that the Weibull distribution fW (V10 (H)) gives a good es-
timation for most wind climates. [182] describes how this function depends on
terrain conditions and on the height H above the ground.

The annual mean wind speed at hub height Hhub is denoted Vmean . This value
can be calculated as:
Z
Vmean = fW (V10 ) ·V10 · dV10 (2.10)
year

The IEC61400-1 standard [99], which describes the safety requirements for
wind turbine generator systems (WTGS), introduces four normal wind classes
based on the annual mean wind speed at hub height. Table 2.1 summarises
these so-called WTGS classes. In addition to the values for Vmean , a value Vre f
is included for each class, which represents a reference wind speed for the de-
scription of extreme wind conditions.
2.3 State-of-the-art in modern wind turbines 17

WTGS class
I II III IV
Vmean (m/s) 10 8.5 7.5 6
Vre f (m/s) 50 42.5 37.5 30

Table 2.1: Two parameters for WTGS classes, reproduced from the IEC61400-
1 standard [99]:
Vre f is a reference wind speed used to describe extreme wind conditions
Vmean is the annual mean wind speed at hub height

The wind shear is the variation of the wind speed with the height H above the
ground. This height dependence during normal wind conditions is given for
the mean wind speed V10 in the IEC61400-1 standard. It is characterised as a
“normal wind profile (NWP)”, which is assumed to follow the power law1 :

V10 (H) = V10 (Hhub ) · (H/Hhub )αh (2.11)

Another characteristic variable of the wind speed is the standard deviation σV


for a 10-minute period. This parameter varies also from period to period and a
log-normal distribution appears to be a good representation for this variation.
Again, [182] describes how this function should be evaluated for specific site
dependent coefficients. This function is used to predict the standard deviation
of the wind speed during a 10-minute period. Consequently, it gives an idea
about the short-term variability of the wind speed, which is known as the tur-
bulence and plays an important role as a design driver in fatigue calculations.
The IEC61400-1 standard [99] distinguishes between two types of turbulence
classes: the A type denotes the category of high turbulence and the B type de-
notes the category of low turbulence. Consequently, a site of e.g. the class IB
is characterised by a high mean wind speed and a low turbulence, which may
correspond to a typical offshore site.

Note that a separate consideration is required for the description of the turbu-
lence in the flow behind a wind turbine, since the presence of a wind turbine
influences the wind flow locally. This influence is visible as an increased tur-
bulence intensity behind the wind turbine and this phenomenon is called “the
wake effect”. The increased turbulence in the wake needs to be considered for
a wind turbine, which is installed at a distance of less than 20 rotor diameters
behind another wind turbine [182]. This consideration is of particular inter-
est for large wind turbine parks and has been the subject of several research
projects [13, 120, 126, 141, 206].

1 The IEC61400-1 standard [99] prescribes to use a value αh = 0.2.


18 2. Introduction to modern wind turbines

Rotor diameter (m) Rated power (kW)


12.5 40
19.4 120
25 300
36.4 400
39 500
43 600
48 750
54 1000
58 1500
71 2000
80 2500
90 3000
120 4500
126 5000

Table 2.2: Pairs of a rotor diameter and the corresponding rated power for wind
turbines found in the literature [182, 203].

Besides the wind speed and the turbulence, other wind related conditions need
to be taken into account for the design of a wind turbine. Among those are the
wind direction, certain transient wind phenomena often in combination with
extreme wind conditions. Extreme wind conditions are defined according to
IEC61400-1 as having a recurrence period of 1 to 50 years. Examples of these
conditions are an extreme operating gust (EOG), an extreme direction change
(EDC), an extreme coherent gust (ECG) and an extreme wind shear (EWS).

The third parameter in equation (2.9) is the rotor swept area A. This increases
with the rotor diameter squared and table 2.2 gives an overview of typical wind
turbine sizes found in the literature [182, 203]. The rotor diameter puts a limit
on the rotor speed, since the tip speed of a blade is limited for reasons of
radiated noise. Neglecting other parameters, the sound pressure increases with
the 5th power of the wind speed relative to the blade and the effect of the rotor
speed on this relative speed is shown in figure 2.5. In general, the limitation
for the tip speed is taken at about 76 m/s for onshore wind sites.

The last parameter in equation (2.9) is the power coefficient Cp , which is a


measure for the efficiency of the energy conversion. It has a theoretical max-
imum of 16/27 at the Betz limit. However, the actual value is even lower be-
cause of the losses in the mechanical and electrical energy conversion. These
losses are characterised by the mechanical efficiency ηm and the electrical ef-
2.3 State-of-the-art in modern wind turbines 19

[kW]

Wind speed [m/s]


(a) Power in the wind and actual power
[-]

Wind speed [m/s]


(b) Power coefficient C p

Figure 2.7: Power and efficiency curve of a 2 MW wind turbine.

ficiency ηe respectively. Figure 2.7 shows a typical power curve for a mod-
ern 2 MW wind turbine as a function of wind speed. In addition, the power
in the wind and the power coefficient are given. Note that this wind turbine
starts operating at Vcut−in = 4 m/s, which is the cut-in wind speed, and stops at
Vcut−out = 25 m/s, which is called the cut-out wind speed. The power produc-
tion increases with the wind speed till it reaches rated power. This typically
occurs at a wind speed of approximately 10 to 15 m/s. The power is kept con-
stant at higher wind speeds by a power control system (cfr. section 2.3.4). The
Cp coefficient varies with the tip speed ratio λ, which is the ratio of the rotor
tip speed to the undisturbed wind speed, λ=Rrot ωrot /Vo . Maximising the rela-
tion Cp (λ) leads to an optimised energy extraction from the wind. This design
driver leads often to small design changes, for which the advantages and dis-
20 2. Introduction to modern wind turbines

advantages are weighed against each other. In general, the extra costs of the
design changes are the disadvantages. However, the design changes can also
lead to a reduction in costs (e.g. reducing the cost of a gearbox by lowering
the input torque), but the advantages are mostly expressed as an increase of the
annual energy output (AEO). The potential AEO is based on the assumption
of a 100% availability of the wind turbine and can be calculated as:
Z
AEO = P(Vo ) · dt (2.12)
year

Taking into account a probability density function f (Vo ) for the wind speed,
equation (2.12) turns into:
Z Vcut−out
AEO = N0 · P(Vo ) · f (Vo ) · dVo [kW h/year] (2.13)
Vcut−in

where N0 = 8760 hours/year. Based on the AEO, the actual average power of a
wind turbine can be calculated as the ratio of the AEO and N0 . Furthermore,
the ratio of the actual average power and the rated power of a wind turbine is
“the capacity factor”:

AEO [kW h/year]


capacity factor [%] = (2.14)
N0 [h/year] · rated power [kW ]

This coefficient expresses how much electrical energy a wind turbine produces
during one year, divided by the energy that would have been produced if the
wind turbine had been running continually at rated power. Therefore, it is a
practical tool in the assessment of specific wind turbines for the installation
at certain wind sites. Reasonable values for the capacity factor at moderate
wind sites are about 25-30% [8]. Estimates for Belgium are about 30% for
offshore wind turbines and about 20% for conveniently sited onshore installa-
tions [197].

2.3.2 General description of a modern wind turbine


Section 2.2 describes the historical evolution of wind turbines and introduces
the Danish concept as a reference for modern wind turbines. This concept has
an upwind rotor facing the wind on a horizontal axis, such as the vast major-
ity of modern wind turbines. Vertical axis wind turbines (VAWT’s) only have
some historical value and, therefore, they fall outside the scope of this survey.
On the other hand, the application of downwind rotors is still existing in some
rather old and small designs. Downwind turbines have the rotor at the lee side
of the tower and, as a result, the blades pass once per revolution through the
wake behind the tower. This causes more fluctuations in the blade loads and
2.3 State-of-the-art in modern wind turbines 21

the power production than in an upwind turbine, because the wake effect is ex-
isting, but much smaller in front of the tower. These variations have a negative
impact on the power quality and can lead to more fatigue loads on the turbine.
Moreover, the passage of each blade through the “tower shadow” produces a
low-frequency noise, which makes downwind turbines noisier. However, they
do have the advantage that the rotor can be made more flexible, because the
bending of the blades is away from the tower, avoiding the risk of hitting the
tower. More flexibility implies less weight and, consequently, a reduction in
costs.

On the other hand, the blades of an upwind turbine require sufficient stiffness,
which is assessed by analysing the deflection of the blade tip. This tip deflec-
tion and the corresponding clearance between the tip and the tower is deter-
mined for the most unfavourable load condition. This analysis requires the use
of a safety factor for the characteristic extreme load, which is 1.35 according
to the IEC61400-1 standard [99] and 1.5 according to the Dutch NVN11400-0
standard [157]. The blade may never hit the tower and, in order to fulfil this
requirement, several design measures are in use. First of all, the rotor plane is
usually placed with a tilt angle of about 5◦ between the rotor axis and the hori-
zontal plane. Furthermore, the blades are often produced with a pre-deflection
or put on the hub under a certain cone angle, pointing away from the tower.
The design process and tip deflection calculations for the blades are verified
during experimental tests of the blades. Figure 2.8 shows a static deformation
test of a blade, performed at LM Glasfiber in Lunderskov, Denmark.

2.3.2.1 Rotor blades

Similar to the Danish concept, the majority of modern wind turbines has a
three-bladed rotor. They have the advantage of yielding the same power out-
put as two-bladed or one-bladed rotors with the same diameter, but at a lower
rotational speed. This makes them less disturbing in a landscape and less prob-
lematic with respect to noise radiation. In addition, the two-bladed rotors suf-
fer more from load fluctuations, since the lowermost blade passes through the
wake in front of the tower when the uppermost blade sees the highest wind
speed. In order to avoid too heavy shocks, these designs require a more com-
plex “teetering hub”, which means that the rotor is hinged to the main shaft.
One-bladed rotors require also the teetering design and, in addition, a coun-
terweight to balance the rotor. In comparison, the three-bladed rotors have the
costs and the weight of the extra blade(s) as main drawbacks. Since the costs
of the rotor represent a considerable part of the complete wind turbine, the ap-
plication of the cheaper two-bladed rotor might revive, when the wind turbines
are further scaled up. In addition, the assembly of a complete two-bladed ro-
22 2. Introduction to modern wind turbines

Figure 2.8: Static deformation test of a wind turbine blade, reproduced


from [137].

tor on the ground is more straightforward and, because of its higher rotational
speed, the drive train torque is lower, which can lead to a further reduction of
the costs, e.g. for the gearbox.

The blade profiles applied along the span of a wind turbine blade all have their
CL and CD characteristic as shown in figure 2.6. Typically, the maximum of
the CL curve, which is often the target in the optimisation of Cp , lies around
the same α values for the different profiles. This means that a quasi constant
α value is desired at all positions along the blade. Because of the difference
in tangential wind speed along the blade (cfr. Vrot in figure 2.5), this requires
a local twist of the blade profiles. The twist varies typically from 0◦ at the tip
up to 30◦ at the hub. Near the maximum of CL , the blade can come into stall.
Especially at low wind speeds, this is an undesirable effect causing a drop
in the electrical energy production. Therefore, it is sometimes counteracted by
installing vortex generators on the low pressure side of the blade [41,195,203].
These strips shift the separation angle of the blade to higher α values and,
thus, delay the stall effect. The use of vortex generators can yield an increase
in the annual energy production up to 4-6% according to the Danish blade
manufacturer LM Glasfiber [136]. This company is a supplier of wind turbine
blades ranging from 13 m up to 61.5 m length. Figure 2.9 shows their 61.5 m
blade. This is the world’s largest wind turbine blade of the year 2005, which
weighs less than 18 tons and is installed in wind turbines with a capacity of
5 MW.
2.3 State-of-the-art in modern wind turbines 23

Figure 2.9: The 61.5 m long wind turbine blade of LM Glasfiber weighs less
than 18 tons, reproduced from [137].

In the search for an α yielding an optimal Cp , the difference in tangential speed


along the blade is taken into account by applying the twist angle. However, the
influence of the axial wind speed variation (cfr. Va in figure 2.5), which goes
together with the undisturbed wind speed Vo , requires a wind speed dependent
adaptation of α. This adaptation is possible by changing the pitch angle of
the wind turbine blade through a rotation around its longitudinal axis. Certain
wind turbine types have a fixed pitch angle and their optimisation of Cp occurs
for the wind speed with the highest energy probability, i.e. the energy weighed
with the probability density function. Other wind turbines adapt the blade pitch
angle to the wind speed based on a control strategy discussed in section 2.3.4.
The pitch mechanism drives the pitch angle of a blade and is generally oper-
ated using hydraulics. This mechanism enables the wind turbine furthermore
to stop by pitching the blades “out of the wind” (cfr. section 2.3.4). Fixed pitch
wind turbines also use an aerodynamic brake principle to stop, based on a ro-
tation of the blade tips only. This tip rotation is often driven by hydraulics and
places the tips “out of the wind”. This causes a braking torque on the rotor
which is sufficient to stop its rotation.

A wind turbine blade consists of lightweight material and is built up of exter-


nal panels connected to internal longitudinal webs. Figure 2.10 shows how the
upper and lower shells are supported by the internal webs. The external shells
form the aerodynamic shape and carry part of the bending load. The internal
structure carries also part of the bending load and the shear load. Furthermore,
it restrains the cross section from deformation and the external panels from
buckling. For a lightweight material with a high load carrying capacity, blade
designers find a favourable option in fibre reinforced plastics (FRPs). This
material is used for the production of the external shells as well as for the in-
24 2. Introduction to modern wind turbines

ternal webs. Usually, the blades are made of fibreglass mats impregnated with
a polyester resin. The use of the more expensive epoxy resin yields a weight
reduction for the same strength of the blade. Likewise, a more expensive, but
lighter alternative for the fibres is carbon. This material has a high strength and
is already produced cost-effectively for parts of certain wind turbine blades.

Figure 2.10: A typical wind turbine blade is built up of an upper and a lower
shell on internal longitudinal webs, reproduced from [182].

An additional element in a wind turbine blade is lightning protection, which is


optional on most types of blades. Because of its height and its separate position
in the landscape, wind turbines are vulnerable to lightning strokes and light-
ning protection is a valuable option in areas with a high lightning frequency.
The protection system consists of receptors connected to conductors mounted
on the blade surface or inside the blade. These conductors prevent the lightning
from harming the blade by guiding it to the earth. Typical lightning protection
systems are discussed in [137, 142, 199].

2.3.2.2 Nacelle
Figure 2.11 shows a picture of a modern 1500 kW wind turbine and a cross
section through its nacelle. The nacelle of a wind turbine refers in general to
all the components installed on top of the tower, except for the rotor blades
and the rotor hub. The nacelle enclosure protects these components against
rain, dust, salt and other harmful particles or objects in the air. All components
are connected to a bed plate. The rotation of this support frame around the
longitudinal axis of the tower is called “yawing” and is driven by the yaw
drives. These drives turn the complete nacelle in the yaw bearing and keep the
rotor always directed towards the wind. The yaw system enables furthermore
to untwist the cables in the tower that transfer the power from the generator
towards the ground. The generator is part of the drive train in a wind turbine
and generates usually 690 V three-phase alternating current (AC), at 60 Hz
in America and at 50 Hz in most other places. Subsequently, this current goes
through a transformer which raises the voltage to the level required by the local
electrical grid, e.g. at 10 kV or 50 kV. The location of this transformer varies
with the turbine type. It can be placed next to the tower, in the tower at the
bottom or in the nacelle. The generator and all other drive train components,
2.3 State-of-the-art in modern wind turbines 25

such as the main bearing, the main shaft, the gearbox, the brake disk and the
coupling are discussed separately in section 2.3.3.

(a) Picture of a modern wind turbine.

(b) The nacelle in a modern wind turbine.

1 pitch drive 8 bed plate


2 main bearing 9 coupling
3 main shaft 10 yaw bearing
4 gearbox 11 tower
5 brake disk 12 yaw drives
6 generator 13 rotor hub
7 nacelle enclosure

Figure 2.11: The NORDEX S77/1500 kW wind turbine, reproduced


from [156].
26 2. Introduction to modern wind turbines

2.3.2.3 Tower and its foundation

(a) Tubular steel tower. (b) Lattice tower.

Figure 2.12: Typical tower structures for modern wind turbines, reproduced
from [156].

Figure 2.12 shows two tower structures in use for modern wind turbines. The
tubular steel tower is the most popular design. It is manufactured in differ-
ent sections with flanges at both ends. The sections are limited in length to
20-30 m, because of the requirements for transport. These requirements also
put a maximum limit on the diameter, since e.g. the maximum clearance under
highway bridges in Denmark is 4.2 m [182]. The diameter has its maximum
at the bottom where the bending moment, caused by the wind forces on the
rotor, is maximum. The diameter further decreases towards the top to save
material and weight, which gives the tubular steel tower a conical shape. Fur-
thermore, the inner tube often gives access to the nacelle by dedicated ladders
or elevators. This is a more comfortable and safer way compared to the lattice
towers. However, these towers require less material and are therefore cheaper.
Moreover, they cause less tower shadow than a massive tubular steel tower.
Nevertheless, subjective opinions about the aesthetics of lattice towers have
made them extremely rare objects in the landscape. Other tower structures,
such as guy-wired towers, three-legged towers and tubular concrete towers,
have some historical value, but have disappeared almost completely from the
scene.
2.3 State-of-the-art in modern wind turbines 27

Choosing an appropriate tower height for a wind turbine is a trade-off be-


tween the extra costs of a higher tower and the extra energy output for a higher
hub level. Equation (2.11) describes the increase in wind speed above the
hub height Hhub . This relation is furthermore site dependent and can be used
for choosing an optimal tower height at a specific site. In general, the tower
height and rotor diameter of a modern wind turbine are close to be equal:
2Rrot /Hhub ≈ 1. This ratio seems to correspond with what many people expe-
rience as more pleasant to look at.

In addition, the tower height is closely linked with the dynamic response of
the complete wind turbine. The turbine’s 1st eigenfrequency is typically the
first bending mode of the tower (cfr. chapter 3). This frequency is mainly de-
termined by the tower stiffness and the “tower head mass” which is the sum
of the nacelle mass and the rotor mass. Given a specific wind turbine with
a fixed tower head mass, the tower is the only tunable variable for avoiding
resonance at the 1st eigenfrequency. Avoiding resonance means tuning the
eigenfrequency away from the excitation frequencies. Typically, the strongest
excitations are the rotational frequency (1P) and the blade-passing frequency,
which is a multiple of the former one depending on the number of blades. For
a traditional three-bladed wind turbine, the main excitations are at 1P and all
multiples of 3P (cfr. section 3.4).

Avoiding resonance by tuning the 1st eigenfrequency below 1P, leads to a soft-
soft design [37]. A more common design leads to a 1st eigenfrequency be-
tween 1P and 3P and is called soft. The hard design is the stiffest alternative
yielding a 1st eigenfrequency above 3P. Even though resonance is avoided
in these designs, it may still be necessary to damp the tower top motion to
avoid excessive tower loads. The German company ESM GmbH [70] pro-
duces therefore a tower damper, which acts as a tuned damper. Figure 2.13
shows a sketch and a picture of this concept. A huge mass is hanging below
the nacelle. As a result, a tower top motion will cause an oscillation of this
mass. Since this mass is moving in an oil bath, the overall motion is damped.

Tower foundation types differ depending on the soil conditions of the site.
Figure 2.14(a) shows a frequently used foundation for onshore wind turbines.
This slab foundation is normally preferred when the top soil is strong enough
to support the loads from the wind turbine. An alternative for softer top soils is
the pile foundation where the loads are transferred to larger depths. Offshore
wind turbines are installed at sandbanks in the sea and need special foundation
structures to transfer the loads from the tower bottom through the water to
the supporting soil. Figure 2.14(b) shows a gravity foundation for an offshore
wind turbine. This is usually a huge mass made from concrete and steel which
28 2. Introduction to modern wind turbines

(a) Sketch of the concept. (b) Tower damper mounted below the nacelle of a 1 MW
wind turbine.

Figure 2.13: A tower damper to damp the tower top motion and, consequently,
reduce the loads on a wind turbine, reproduced from [70].

rests on the bottom of the sea. A monopile foundation, on the other hand, is
clamped into the ground by driving a long pipe into the soil. Finally, the tripod
foundation is a support structure standing on three legs that are clamped into
the bottom of the sea.

(a) A slab foundation for an onshore wind (b) A gravity foundation for an offshore
turbine. wind turbine at the Middelgrunden wind
farm [148].

Figure 2.14: Examples of tower foundation types.

2.3.3 Drive train layout in a modern wind turbine


The conversion of the mechanical energy at the rotor hub into electrical energy
at the generator side occurs in the drive train. Furthermore, the drive train
has to limit the rotor speed, mainly because of the limitation on the blade
2.3 State-of-the-art in modern wind turbines 29

tip speed for noise issues. During operation, this limitation is controlled by
the generator. In fault conditions where the generator is disconnected, the
brake system stops the wind turbine. Figure 2.11 shows the most popular drive
train design for modern wind turbines. Here, the generator rotates at a higher
speed than the rotor of the wind turbine and, therefore, a gearbox is necessary
to increase the speed. In an other existing drive train design, the gearbox is
omitted because the generator can produce electricity at the low speed of the
wind turbine’s rotor. It is clear that the generator type plays a determining role
in the layout of a drive train.

2.3.3.1 Generator
Modern wind turbines produce electricity for a three-phase AC grid with a fre-
quency of 50 Hz in Europe. This puts a lot of requirements on the wind turbine
operation and electricity transfer towards the grid. These requirements are pub-
lished in the relevant standards and guidelines and are not further elaborated
in this text. Only the requirement of the fixed frequency of the AC electricity
transferred to the grid, is used here to introduce the distinction between two
generator connection types. The first type is a direct grid connection with-
out a frequency converter and, therefore, it can only produce electricity at the
grid frequency. The second type is an indirect connection to the grid, namely
through a frequency converter.

Direct grid connection

Both an asynchronous generator, also called induction generator, and a syn-


chronous generator can produce electricity through a direct grid connection.
These generators have a rotor which rotates in a stator. The space between
the rotor and stator is known as the “air gap”. Figure 2.15 shows the torque
characteristic of these generators. The synchronous speed ns depends on the
grid frequency and on the number of pole pairs in the generator:
ns = 60 · fgrid /p p (2.15)
ns synchronous rotational speed [rpm]
fgrid grid frequency [Hz]
pp number of pole pairs
The vertical torque characteristic of the synchronous generator indicates that
it works at a constant speed. This means that the load variations common in
wind turbines are translated into heavy torque fluctuations and high peak loads
in the drive train. Therefore, this type of generator is rarely used in a direct
grid connection.
30 2. Introduction to modern wind turbines

(a) Synchronous generator. (b) Asynchronous generator.

Figure 2.15: Torque characteristics.


(*) Tp and n p are the pull-out torque and speed of the asynchronous generator.

The torque of the asynchronous generator is a function of the slip s:

ns − nr
s= (2.16)
ns

nr speed of the rotor in the asynchronous generator [rpm]

In general, the slip is expressed in percentages and negative in generator mode.


Moreover, it is small for a short-circuited rotor such as the popular squirrel-
cage rotor. This means that the speed variation of an asynchronous generator
is limited (e.g. 1%). Within this limited speed range, the torque characteristic
can be linearised with a reasonable accuracy. The equation obtained in this
way, indicates a linear relation between the torque and the slip of the gen-
erator. Since the slip is a linear function of the generator speed, the torque
produced by the asynchronous generator varies linearly with its speed which
corresponds to a damper characteristic. This characteristic and its ability to
allow small speed variations, make the asynchronous generator appropriate for
wind turbines and their fluctuating loads. In addition, its robustness, its me-
chanical simplicity and its relatively low price are advantages that contribute to
the success of the asynchronous generator in wind turbine applications. Since
the speed range in which they produce electricity is very small, their applica-
tion is still referred to as “fixed speed wind turbines”.

The number of pole pairs p p in these applications is usually two, resulting in


a synchronous generator speed of 1500 rpm. A higher number of pole pairs
is rarely used, because the generator efficiency decreases with increasing p p .
2.3 State-of-the-art in modern wind turbines 31

A generator speed of 1500 rpm implies the need for a gearbox in these ap-
plications to reduce it to the desired speed of the wind turbine’s rotor. The
generator is typically located in the back of the nacelle (cfr. figure 2.11) where
it is mounted with dampers on the bed plate.

Rotating at a fixed speed for all wind speeds is disadvantageous for the aerody-
namic efficiency [220]. A lower rotational speed at low wind speeds increases
this efficiency and, furthermore, reduces the noise from the rotor blades which
is mainly a problem at low wind speeds. In addition, the fixed speed opera-
tion gives only a limited power quality control and yields higher loads on the
drive train components. Therefore, designers have constantly been looking
for cost-effective design changes which could yield a larger speed range for
an asynchronous generator. Their alternatives are based on two concepts: (1)
changing the number of pole pairs or (2) changing the slip in the generator.

1. Changing the number of pole pairs can be considered as switching to


another generator, although it is normally applied in one single generator
by changing the connection of the stator windings [220]. At low wind
speeds, the generator operates with a higher number of pole pairs (e.g.
p p =3), but a limited capacity. At higher wind speeds, it switches to
a lower number of pole pairs (e.g. p p =2) and operates, consequently,
at a higher rotational speed. The extra hardware costs of operating at
different speeds are weighed against the advantages: a lower rotational
speed at low wind speeds yields a higher aerodynamic efficiency and a
lower noise emission.

2. An alternative for the popular squirrel-cage rotor in an asynchronous


generator is a wound rotor [37]. This rotor can be connected through slip
rings and brushes with an external variable resistance. By controlling
the rotor resistance in this way, the slip of the generator can be affected
and, thus, the rotor speed changed. An increase of the resistance causes
an increase in speed and application of this technique leads typically to
a speed variation up to 10% above synchronous speed. This design is
vulnerable to wear at the slip rings and requires sufficient maintenance.
However, an integration of the variable resistance in the rotor itself can
avoid these problems. The wind turbine manufacturer Vestas [212] ap-
plies this interesting alternative in their OptiSlipr system. Here, the
communication between the variable rotor resistance and its controller
is contact-less by means of an optical fibre.
An alternative for the energy dissipation in the generator’s rotor resis-
tance is an energy exchange between the rotor and the electricity grid.
This concept is implemented in a so-called “doubly fed induction gen-
32 2. Introduction to modern wind turbines

erator” (DFIG). By controlling the energy exchange between the rotor


and the grid, the slip in the generator is controlled and the rotor speed
can be varied. The link between the rotor and the grid is made through a
frequency converter and, in this way, the DFIG is only partly in “direct
grid connection”. The implementation of this concept implies specific
control requirements and design measures which are described exten-
sively in the literature [172, 197]. The DFIG became popular in modern
wind turbines because of its ability to operate with large speed varia-
tions. Their speed range depends on the amount of power the frequency
converter can take. The ratio between this amount and the generator’s
rated power determines the costs of the system and the benefit, namely
the possible speed variation. This control concept has become very pop-
ular in modern multi-MW wind turbines. A typical application of the
DFIG is included in the OptiSpeedr system of Vestas [212] with speed
variations up to 60%.

Indirect grid connection

When all the power produced by the generator goes through a frequency con-
verter towards the grid, this generator is indirectly connected to the grid. In
this concept the speed range of operation can vary, in principle, from zero to
the maximum which the wind turbine can handle. Asynchronous generators
are rarely used in combination with full-scale converters, because the alterna-
tive of the DFIG and a limited speed variation is more cost-effective. On the
other hand, synchronous generators and full-scale converters do make a pop-
ular combination for modern wind turbines. These generators can be made
with a high number of pole pairs yielding a low synchronous speed. When this
speed is as low as the desired speed for the wind turbine’s rotor, a gearbox is
no longer needed.

The application of this concept for wind turbines is known as “direct drive
wind turbines”. Note that the required diameter of a synchronous generator
increases with the torque density. This means that modern multi-MW wind
turbines with a rotor speed at about 10 rpm, require generators with large di-
ameters. Figure 2.16 shows the E112 ENERCON wind turbine. This turbine
has a power capacity of 4.5 MW, a tower head mass of 500 tons and a generator
diameter of about 10 m [68]. These large generator diameters cause problems
for the transport of the nacelle. Some wind turbine manufacturers equip their
synchronous generator therefore with a smaller number of pole pairs, yield-
ing a higher rotational speed and, consequently, a lower torque density. These
generators are smaller in diameter, but need a gearbox to reduce their speed
2.3 State-of-the-art in modern wind turbines 33

towards the wind turbine’s rotor. This combination of a gearbox and a syn-
chronous generator is applied in the Multibridr Technology as used in the
M5000 5 MW Multibrid wind turbine [152] and in the WWD-3 3 MW Win-
WinD wind turbine [221].

(a) Nacelle cross section. (b) Picture of the E112.

Figure 2.16: The E112 ENERCON 4.5 MW direct drive wind turbine, repro-
duced from [68].

2.3.3.2 Drive train concepts

The rotor hub introduces the mechanical energy from the wind into the drive
train as a load vector with six components. Only the torque component is
needed in the generator to produce electricity. The other loads are transferred
through the drive train towards the tower. Different drive train concepts yield
the desired torque separation towards the generator. Figure 2.17(a) shows a
first concept with two separate bearings supporting the main shaft. The bear-
ing near the wind turbine’s rotor typically carries axial loads. Both bearings
carry radial loads and hence transfer bending moments towards the tower. As
a result, the main shaft introduces only torque into the gearbox. The gearbox
suspension may only carry the reaction torque towards the bed plate. There-
fore, specific designs of torque arms are in use which give the gearbox suffi-
cient freedom to move, but still carry the torque. These torque arms can be
34 2. Introduction to modern wind turbines

equipped with vibration dampers and need specific design considerations with
respect to their flexibility.

Figure 2.17(b) shows a concept similar to the one described above. Again, one
axial bearing supports the main shaft near the rotor and one radial bearing is
located on the other side. Here, the radial bearing is integrated in the gear-
box. Therefore, the gearbox suspension is mounted on the bed plate. Since
this suspension consists typically of two torque arms, this drive train concept
is often referred to as a “three-point-suspension” system. The stiffness and
damping characteristics of the suspension play an important role in the dy-
namic response of the drive train.

(a) Two bearings outside the gearbox supporting the


main shaft, reproduced from [89].

(b) Three-point-suspension: one bear- (c) Main bearing integrated in the


ing integrated in the gearbox, reproduced gearbox, reproduced from [88]
from [89].

Figure 2.17: Different drive train concepts for modern wind turbines.

Figure 2.17(c) shows a third drive train concept used in modern wind turbines.
Here, all the loads enter the gearbox and, thus, all the bearings supporting the
rotor hub are integrated in this component. Since there is no longer space for
2.3 State-of-the-art in modern wind turbines 35

a main shaft, nor for two bearings supporting it, the rotor hub is hanging in
one large bearing carrying radial and axial loads as well as all bending mo-
ments. The gearbox is an integrated part of the whole nacelle in this concept.
This design can lead to a considerable weight reduction of the nacelle, but it
demands for a close cooperation between the gearbox designer and the wind
turbine manufacturer.

Under specific circumstances a wind turbine requires a complete stop. De-


pending on the wind turbine type, the braking forces are aerodynamic loads
at the rotor (e.g. by using tip brakes or by pitching the blades) or a combi-
nation of these loads and a mechanical braking system. In any type of wind
turbine, a mechanical emergency brake is required for safety reasons. This is
typically a brake disk with corresponding callipers mounted at the output of
the gearbox. For the coupling of the gearbox and the generator several alterna-
tives are in use. This coupling should be able to deal with misalignments and
structural deformations caused by temperature variations. Some couplings are
equipped with an integrated torque limiter, such as the escowind couplings of
ESCO [69].

2.3.4 Power control aspects


Section 2.3.1.2 and figure 2.7 describe how the actual power of a wind tur-
bine varies with the wind speed. The power curve is a result of several design
trade-off’s, all aiming at the most cost-effective solution. Choosing a higher
maximum output for a specific wind turbine would not be beneficial, because
this output is only available at higher wind speeds which have a low proba-
bility. Likewise, installing a cheaper generator with a lower capacity in this
wind turbine would decrease the rated wind speed Vrat , i.e. the speed at which
rated power is produced. This wind turbine will operate more often at rated
power, but the increased amount of wasted energy above rated wind speed will
make it less cost-effective. Typical values for Vrat are about 10 to 15 m/s. This
characteristic divides the wind turbine operation in two regimes with respect
to power control.

1. [Vcut−in Vrat ]: the power production is optimised at low wind speeds to


come as close as possible near the power in the wind.

2. [Vrat Vcut−out ]: the power production is limited at high wind speeds to


yield a constant power equal to the rated power of the wind turbine.

Three different control concepts are in use to achieve the desired optimisation
and respective limitation in these wind speed ranges. These concepts are the
passive stall control, the active stall control and the pitch control [37, 87, 182].
36 2. Introduction to modern wind turbines

2.3.4.1 Passive stall control


Passive stall control is used in wind turbines with a fixed pitch angle. For these
wind turbines, an increase in wind speed will result in an increase of the an-
gle of attack α. A wind turbine blade starts to stall at a certain value for α
(cfr. figure 2.6), which corresponds to a certain wind speed. Starting from this
wind speed, the wind flow separates at the downwind side of the blade and
the lift coefficient CL drops. This decrease in CL leads in wind turbines with
a fixed pitch angle to a passive limitation of the power at high wind speeds.
Since CL (α) has a negative slope in the stall region, an increase in wind speed
implies a decrease in lift forces and as a result a decrease in power. This is
a stable and self-regulating system. Since CD (α) increases rapidly after stall,
stall-controlled wind turbines experience large thrust loads and require in gen-
eral a rather stiff tower design. Moreover, these wind turbines require a correct
setting of the blade angle relative to the rotor plane. The major drawbacks of
the passive stall control are the lower efficiency at low wind speeds and the
higher power fluctuations at high wind speeds. Furthermore, the stall region is
less-known and the corresponding load simulations less accurate, the aerody-
namic damping in this region is lower [171] and the control concept does not
allow an assisted start of the wind turbine.

2.3.4.2 Active stall control


The pitch mechanism in a wind turbine enables the rotation of a blade to a dif-
ferent pitch angle and consequently different α. Figure 2.18 describes the set-
ting of the blade pitch angle θcontrol for an arbitrary cross section of the blade.
The control of this angle permits to control the power production. At low
wind speeds, the blade is turned “out of the wind” towards a negative θcontrol ,
which yields a higher α. This results in higher lift forces (cfr. figure 2.6) and,
consequently, a better efficiency. The power limitation at high wind speeds
occurs also by pitching the blade with a negative pitch angle. Here, the result-
ing higher α causes the blade to stall and, thus, waste the excess energy. This
control concept is known as “active stall” or “pitch to stall” which refers to the
power limitation at high wind speeds. It allows a more accurate control at high
wind speeds resulting in a more constant power curve at rated power. Further-
more, the negative slope of CL (α) in the stall region makes it a stable system
and, since the slope is quite moderate, rather small load variations occur during
heavy wind gusts.
2.3 State-of-the-art in modern wind turbines 37

Figure 2.18: Definition of the setting of the blade pitch angle θcontrol for an
arbitrary cross section of the blade. This angle is regulated to control the power
production in the active stall concept and the pitch control concept.

2.3.4.3 Pitch control

At low wind speeds, the blades are pitched with a negative θcontrol to yield
higher α values for power optimisation. This is applied both in “pitch con-
trol” and in “active stall control”. They differ in the way the power is limited
at high wind speeds. Pitch controlled wind turbines are pitched “towards the
wind” with a positive θcontrol , which yields a smaller α. This results in a reduc-
tion of the lift forces, while the flow around the blade remains attached and no
stall occurs. Therefore, pitch control is also referred to as “pitch to feather” in
contrast with “pitch to stall”. The power is controlled on the rather steep pos-
itive slope of CL (α) in the pre-stall region of figure 2.6. An increase in wind
speed causes higher loads which can make this control unstable. Therefore,
the reaction time of the pitch mechanism is critical in order to follow the wind
speed variations to prevent excessive peak loads. However, in gusty conditions
large pitch excursions are needed to maintain constant power and the inertia of
the blades will limit the speed of the control system’s response. Therefore, in
practice pitch control requires a generator with variable speed allowing a slight
acceleration of the wind turbine’s rotor at wind gusts. The pitch mechanism is
usually operated using hydraulics.

By staying out of the stall region, the aerodynamics are better understood and
the loads can be predicted more accurately. Moreover, the blades stay well
damped. Other advantages of the pitch control are the possibilities of feathered
blade parking and assisted starting. Furthermore, an individual pitch control
of the blades can be used to actively reduce the loads on the wind turbine or to
control drive train vibrations and tower vibrations [28–30, 57, 123].
38 2. Introduction to modern wind turbines

2.4 Current status of wind powered electricity


D’haeseleer et al [53] describe the problem of securing a reliable, sustainable,
clean and affordable energy supply as one of the biggest challenges to mankind
in the 21st century. In their study, they give an extensive overview of various
aspects related to this topic, including the so-called energy mix for electricity
generation. The actual composition of this mix varies for different locations
and depends on many technological, economic, environmental and social cri-
teria. During the last decades, the interest for replacing fossil fuel in this mix
partly by renewable energy sources, such as wind energy, has increased, espe-
cially in Europe. This trend is mainly driven by the following motives.
1. The global oil crisis in 1973 demonstrated clearly the dependence of
the European economy on the import of fossil fuel. Replacing fossil
fuel by renewable energy sources can help in reducing this import and,
consequently, the dependence.
2. Neither the enhanced greenhouse effect, as a result of the increase in
CO2 emission during the last centuries, nor its link with a climate change
is completely proven yet [53]. However, the probability of these as-
sumptions is so high that a neglect is considered as irresponsible. Using
renewable energy instead of fossil fuel can contribute to the advisable
reduction in CO2 emission.
3. Renewable energy is furthermore considered as a long-term sustainable
alternative for fossil fuel, which has limited resources on earth.
These motives are translated in numerous initiatives and political incentives
for the application of wind energy for electricity generation. As a result, the
wind turbine industry boomed during recent years. Figure 2.19 shows the evo-
lution of the global installed wind power capacity from 1995 to 2005. This
capacity reached 59.3 GW at the end of 2005 according to the Global Wind
Energy Council (GWEC) [81]. Figure 2.20 shows the evolution of the annual
installed capacity for this period and indicates an impressive increase in 2005
of 11.8 GW, which is about 20% of the total.

Europe is the market leader with 40.9 GW of installed wind power capacity,
representing a share of almost 70%. Figure 2.21 shows the top 10 of countries
with the highest total installed capacity. Germany is on top of this list with
18.4 GW, which is 31% of the global total. The value for the installed capacity
of wind power equals the sum of the maximal power output for all installed
wind turbines. In terms of annual energy output, the global total installed
capacity corresponds to 130 TWh for an average capacity factor world wide of
25% (cfr. equation 2.14).
2.4 Current status of wind powered electricity 39

[MW]

Figure 2.19: The evolution of the global cumulative installed wind power ca-
pacity from 1995 to 2005 [81].
[MW]

Figure 2.20: The evolution of the global annually installed wind power capac-
ity from 1995 to 2005 [81].
40 2. Introduction to modern wind turbines

Total capacity MW %
Germany 18428 31.0
Spain 10027 16.9
US 9149 15.4
India 4430 7.5
Denmark 3122 5.3
Italy 1717 2.9
UK 1353 2.3
China 1260 2.1
Japan 1231 2.1
NL 1219 2.1
Rest of the world 7368 12.5
World total 59322 100

Figure 2.21: The top 10 of cumulative installed wind power capacity per coun-
try (December 2005) [81].

Various political support mechanisms promote the installation of wind tur-


bines [197] and, under these conditions, the wind energy industry is expected
to grow further rapidly during the next years. Meanwhile, new technological
improvements will enable a further cost reduction of wind turbine installa-
tions. Current costs are estimated at 0.75 to 1.5 e/W for onshore turbines and
about 2 e/W for offshore installations, corresponding to an energy price of
0.03 to 0.10 e/kWh, according to J. Willems2 , “Hernieuwbare energie: mo-
gelijkheden en beperkingen”, pp. 215-233 in “Energie vandaag en morgen”,
W. D’haeseleer (Editor) [53]. By its contribution to the state-of-the-art in wind
turbine technology, the present dissertation can contribute indirectly to a fur-
ther cost optimisation of wind turbines and, consequently, a reduction of the
wind energy price.

2.5 Conclusions
Modern wind turbines originate from the application of wind power in wind-
mills, which were used for grinding grain and pumping water. The history of
this application goes back to as early as the 7th century A.D. and its evolu-
tion led to the first electricity generating windmill, or wind turbine, installed
in 1888 by Charles F. Brush. Another wind turbine pioneer from that period
is Poul La Cour, who lived in Denmark. It was his student Johannes Juul who
introduced an innovative wind turbine concept in the mid 1950s. His wind tur-
bine had a three-bladed upwind rotor, which was passively stall-regulated and
2 Willemsnotes furthermore that, since the power production heavily depends on the wind
conditions, some projects at poor wind locations lead to higher prices.
2.5 Conclusions 41

connected to an AC asynchronous generator. This concept, known as the Dan-


ish concept, is still considered as the basis of most modern wind turbines. The
actual breakthrough of the professional wind turbine industry dates from the
beginning of the 1980s. Since then, the industry rapidly expanded and wind
turbines evolved to what they are today.

The basic aerodynamic principle behind a wind turbine’s operation is the ex-
traction of energy from the wind by using the reaction forces on the rotor
blades, which are rotating in the wind flow. These reaction forces are split
up in lift forces, drag forces and pitch moments, where the former component
is responsible for driving the rotor. A determining parameter for the magni-
tude of these forces, is the local angle of attack α. This angle defines whether
a rotor blade operates in pre-stall, in stall or in post-stall conditions. α changes
with varying wind speed and it can furthermore be controlled by pitching the
blades around their longitudinal axis. An expression for the actual power, in-
duced by the reaction forces and transferred into electricity towards the grid,
is characterised by:
1
P = · ρ ·Vo3 · A ·Cp (2.17)
2
1. ρ is the air density.

2. Vo is the wind speed, which is in the third power and has consequently a
major impact on the total power production. The wind speed distribution
is site dependent and is classified according to the IEC61400-1 standard.

3. A is the rotor swept area, which is in addition determining for the rotor
speed, since the blade tip speed is usually limited to 76 m/s to keep the
radiated noise below an acceptable level.

4. Cp is the power coefficient, which is a measure for the efficiency of the


energy conversion with a theoretical maximum of 16/27.

A modern wind turbine has typically a horizontal axis and a three-bladed up-
wind rotor. This latter component is manufactured in lightweight material and
often equipped with dedicated features, such as a pitch mechanism, vortex gen-
erators or lightning protection. The tower is mostly a tubular steel structure,
which is designed to avoid resonance at the main excitation frequencies, being
1P and 3P. The drive train converts the mechanical energy at the rotor hub into
electrical energy at the generator. Depending on the type of generator and its
grid connection, the drive train may differ considerably. The grid connection
can be direct or indirect, where the latter type implies that all produced electric
power goes through a frequency converter towards the grid.
42 2. Introduction to modern wind turbines

1. The asynchronous generator in a direct grid connection is a popular de-


sign concept. Here, the relatively stable grid frequency (e.g. 50 Hz in
Europe) determines the synchronous speed of this generator for a par-
ticular number of pole pairs. This number equals typically two, which
results in a synchronous speed of 1500 rpm and which implies the need
for a gearbox in these applications. Since the slip in this generator is
small, the possible speed variations are very limited. Dedicated design
optimisations have been performed, in order to extend the speed range
in this concept, aiming at an increased aerodynamic efficiency. One of
the alternatives is changing the number of pole pairs, typically from two
to three, which permits to rotate at a lower speed. A second alternative is
changing the slip in the generator by controlling the energy dissipation
in the generator’s rotor resistance. This is possible with a variable rotor
resistance, which yields a range of speed variations up to 10% above
synchronous speed. A more recent development enables an energy ex-
change between the rotor and the grid through a frequency converter.
This DFIG is consequently only partly in direct grid connection, which
extends the normal speed range usually with 60%.
2. A second common design concept is a synchronous generator in an indi-
rect grid connection. These generators can be made with a high number
of pole pairs, yielding a low synchronous speed. In “direct drive wind
turbines” this speed is as low as the desired speed of the wind turbine’s
rotor and, consequently, no gearbox is required. Since a high number
of pole pairs implies a large and heavy generator, a compromise for this
concept is found in the combination of a smaller synchronous generator
and a gearbox, with a smaller reduction ratio than for the asynchronous
generator in a direct grid connection.
The power control of a wind turbine aims at optimising the power output be-
low rated wind speed (Vrat ) and at keeping the power output constant above
this wind speed. Three different concepts are in use for the power control. In
the passive and the active stall control, the power is limited above Vrat as a
result of a decreasing lift coefficient during stall. In the former concept, this
occurs automatically at a certain wind speed, but the power output can fluctu-
ate considerably. In the latter concept, the blades are pitched into stall and a
constant power output can be controlled actively. Here, the power can also be
optimised below Vrat by pitching the blades to higher α values. This optimi-
sation is also applied in the pitch control concept. Pitch control actively steers
the blades’ pitch angles and differs from the active stall control at wind speeds
above Vrat . Here, the blades are turned towards the wind yielding a smaller α
and, consequently, a smaller lift coefficient. This latter concept is combined
with a variable speed generator.
2.5 Conclusions 43

During the last decades, the interest for using renewable energy sources for
electricity generation increased, often promoted by political support mecha-
nisms. One of its results is a boom in the wind turbine industry since ten years.
The global installed wind power capacity reached 59.3 GW at the end of 2005.
Europe comprises almost 70% of this capacity and Germany is the global mar-
ket leader with a share of 31%. The rapid growth is expected to continue in the
coming years and new technological improvements should further reduce the
cost of wind turbines.
44
3

State-of-the-art in the design of


a wind turbine drive train

3.1 Introduction
This chapter describes the current state-of-the-art in simulating the design
loads for the drive train in a wind turbine. Since the drive train is an integrated
part of the whole wind turbine, this process is part of the prediction of loads
for the complete wind turbine. Therefore, section 3.2 starts with a description
of the general design specifications for a wind turbine. The consideration of all
specifications, implies the combination of all relevant external conditions with
various possible operation modes leading to a large set of different load cases.
This is a complex procedure, which is computerised in specialised wind turbine
codes for the simulation of load time series. Section 3.3 describes the principle
behind these codes and gives an overview of existing software packages and
their validation. In addition, it discusses the structural model representation in
these traditional wind turbine codes, including only one degree of freedom to
represent the drive train, and its consequences for the traditional drive train de-
sign. Subsequently, section 3.4 describes the limitations of the existing design
codes and the need for additional simulation models, which is the main objec-
tive in the present research. The MBS formulation is introduced as the most
appropriate modelling technique for a more detailed drive train model. The
section concludes with an overview of recent publications on similar research
activities.

3.2 Design specifications of a wind turbine


In principle, one single wind turbine could be designed for one specific site.
However, it is clear that this would not lead to a cost-efficient production of
wind turbines. Therefore, wind turbine manufacturers offer in general a lim-
45
46 3. State-of-the-art in the design of a wind turbine drive train

ited range of products suitable for a broad range of sites. The design specifi-
cations of each of these wind turbines are a summary of the requirements at
different sites. However, the generalisation of the designs has also limitations.
In practice, a commercially available wind turbine is certified for installation
at a specific IEC wind class according to the IEC61400-1 standard [99] (cfr.
table 2.1). A site assessment prior to a wind turbine project development de-
termines the wind class of the site and allows to choose an appropriate wind
turbine. Only a few parameters of the chosen wind turbine can be changed
without implying a completely new design. These are site-dependent param-
eters and typically alternatives for the hub height or the rotor diameter. As an
example, the Vestas V52 850 kW wind turbine is available in a wide range of
tower heights from 40 to 74 m [212].

The development of a new wind turbine is preceded by a thorough market sur-


vey which assesses the future success of a wind turbine with a certain capacity
for a specific wind class, taking into account the technological state-of-the-art
and several market trends. This way an imaginary and ideal wind turbine is
invented which forms the target for a new design. The first step in the design
process is listing the required design specifications, considering the feasibility
and economical benefits of eventual design parameters. The summary of the
design specifications is further used in the design of all necessary components.

Based on the target IEC wind class, the most determining factors in the design
specifications are already known. These are the wind speed and the turbulence
intensity. Additional wind related specifications are the wind direction and the
occurrence of transient wind phenomena, often in combination with extreme
wind conditions. Other external conditions considered for a wind turbine de-
sign are:
• the environmental temperature, because of e.g. its influence on the oil
lubrication
• the air density ρ, because of e.g. its influence on the loads and the power
curve
• the air humidity and atmospheric corrosion, e.g. for offshore wind tur-
bines exposed to a saline environment
• the precipitation (rain, snow and hail)
• ice formation on rotor blades
• earthquakes
• lightning strokes
3.2 Design specifications of a wind turbine 47

Guidelines for the consideration of the majority of these external conditions


are also described in the IEC61400-1 standard. Furthermore, additional re-
quirements with respect to the power transfer to the grid are included.

Regarding the wind turbine operation, normal operation and fault situations
are usually distinguished. This leads to another important part of the design
specifications for a new wind turbine, namely the description of the different
load cases that can occur. These cases are typically a combination of external
conditions and a specific operational or fault situation. Not only their def-
inition is included, but also an estimation of their frequency of occurrence.
In general, a wind turbine is designed for a 20 years lifetime which equals
175.200 hours. Combinations of external conditions and turbine states can
be made endlessly, but an overview of relevant load cases is summarised in
the IEC61400-1 standard. Moreover, other standards are also used in industry
which specify sometimes more load cases. Such alternative standards are the
Germanischer Lloyd’s Regulation for the Certification of Wind Energy Conver-
sion Systems, commonly referred to as the GL rules, and the national standard
NEN 6096 in The Netherlands and DS 472 in Denmark [37].

Based on the design specifications, a wind turbine manufacturer can start de-
signing the new wind turbine. Similar to any other design, this is an iterative
process with a lot of trade-offs aiming at a product that yields electricity as
cheaply and efficiently as possible. In general, most of the design decisions
rely on existing experience and proven concepts. However, since the wind tur-
bine market is still expanding, new wind turbines are usually bigger than their
predecessors and, therefore, the application of existing techniques should al-
ways be done with care.

The design process starts typically with the calculation of the loads acting on
the wind turbine. The following list shows all sources of loading:
• Aerodynamic loads: Section 2.3.1.1 introduces the BEM method to
calculate aerodynamic loads for a given wind speed. Using the wind
speed characteristics in the design specifications as input, a probability
distribution for these loads can be calculated.
• Gravitational loads: Gravitational loading leads to constant forces point-
ing downwards and corresponding bending moments.
• Inertial loads: The inertial loads include all acceleration, centrifugal
and gyroscopic effects.
48 3. State-of-the-art in the design of a wind turbine drive train

• Operational loads: The most important operational load is the gen-


erator torque. Other operational loads are induced by certain control
actions, such as e.g. blade pitching, starting up, (emergency) braking or
yawing.

Of course the different types of loading - if present - can all work simulta-
neously. Therefore, it is necessary to combine them for all load cases in the
design specifications in an appropriate way. In a first approach, all load sources
can be calculated independently. A combination of the relevant load compo-
nents yields maximum values for the loads, further called extreme loads, on the
wind turbine components. This initial calculation is a static approach which
gives a first estimation for the extreme loads, however, it has also some major
drawbacks.

1. An independent calculation of external loads from different sources com-


plicates the consideration of interdependence. For example:

• the inertial loads corresponding to the acceleration of the drive


train depend on the generator torque, which is an operational load
• the control system of the blade pitch angle and the rotor speed
setting influences the aerodynamic loads

2. It is impossible to analyse the dynamic interaction between the individ-


ual wind turbine components in a static and independent calculation of
different load sources. The dynamic response of a structure describes the
amplification and damping of a given excitation spectrum and depends
on the combination of the mass, stiffness and damping characteristics of
all its components. Dynamic effects cause load variations and can yield
higher extreme loads than a static calculation. Their influence is often
included by applying empirical load safety factors for lack of dedicated
simulation tools. However, the industry feels more and more the need
for accurate predictions and tries to make progress in this direction.

3. The last but certainly not the least drawback of the static determination
of extreme loads is the lack of information about load variations. After
all, a wind turbine is subjected to a severe fatigue loading regime during
its 20 year life. The fatigue loads originate from different sources. One
of them is the stochastic wind speed variation described by the turbu-
lence in the wind. In addition, all other kinds of wind changes can lead
to load variations. Another source of fatigue loads is the cyclic loading
caused by the rotation of the rotor and all other drive train components.
For example, the rotor of a 2 MW wind turbine will rotate in total more
than 108 times, with each revolution causing a complete gravity stress
3.3 Traditional wind turbine design codes 49

reversal for each blade. Finally, certain load variations originate from
vibrations or other dynamic effects in the wind turbine. For the indi-
vidual component designs, the combination of all fatigue loads is often
more demanding than the requirements for extreme loading. Therefore,
an accurate prediction of all load cycles is of extreme importance for a
reliable design.

It is clear that a static determination of extreme loads is by far not sufficient


for the design of all components in a wind turbine. In addition, it is even
more clear that an accurate prediction of the load levels and cycles, including
the interaction between different load sources and possible dynamic effects, is
no straightforward job. This is even further complicated by the fact that this
procedure needs to be repeated for all defined load cases. It is therefore not sur-
prising that, since the earliest days in modern wind turbine design, the industry
has been developing software codes to computerise the prediction of wind tur-
bine loads and to replace more simplified and empirical load calculations. A
number of specialised wind turbine codes have become standard design tools
in the wind turbine industry and perform quite well in the prediction of rotor
and tower loads. However, they still impose considerable limitations with re-
spect to the design of a robust drive train in a wind turbine. In order to better
understand these limitations, section 3.3 describes the existing codes and their
capabilities in more detail. In addition, section 3.4 describes their main short-
comings and summarises the actual needs as objectives in the present research.

3.3 Traditional wind turbine design codes

Present wind turbine drive trains are designed for simulated loads, which result
from specialised wind turbine software codes. These codes have to combine
all relevant external conditions with all possible operational and fault condi-
tions, while calculating all kind of loads on the complete wind turbine with
sufficient detail. However, the drive train model in the existing codes is over-
simplified and, as a consequence, they do not give the necessary insight in the
drive train dynamics. With the purpose of understanding the shortcomings of
the existing codes, this section elaborates on their implementation and their
features. Section 3.3.1 starts with a description of the general principle behind
these codes. Subsequently, section 3.3.2 gives an overview of existing wind
turbine codes and discusses their validation. Finally, section 3.3.3 describes
the typical structural model representation in the existing wind turbine codes
with only one DOF to represent the torsion in the drive train.
50 3. State-of-the-art in the design of a wind turbine drive train

3.3.1 Principle

In general, the output of the traditional specialised wind turbine codes is a


prediction of the mechanical loads formatted in time series, i.e. as a function
of time and, thus, including load variations. These time series can be easily
processed to summarising statistics such as for example mean, maximum or
minimum values, which are already sufficient to perform certain initial design
calculations. The code behind the load predictions is a “model” description of
the complete wind turbine including all relevant external conditions at its site.
The detail and accuracy of the predictions depend on the detail and accuracy
of this description, which is a combination of theories from diverse domains.
Therefore, these codes typically originate from the cooperation of various ded-
icated specialists or research groups.

The splitting up of the modelling work into sub-domains can also be found in
the typical implementation of the codes. Figure 3.1 describes this subdivision
in different “modules”, with a distinction between external and internal mod-
ules. The wind, the electricity grid and the sea waves (in case of an offshore
turbine) are external modules, which are only considered as inputs for the sim-
ulation. This is debatable with respect to the electricity grid, since there will
be a mutual interaction with the wind turbine [197]. However, an independent
grid is usually assumed for mechanical load calculations.

Figure 3.1: Schematic representation of the implementation of existing wind


turbine design codes (dark grey: external modules, light grey: internal mod-
ules).
3.3 Traditional wind turbine design codes 51

Each individual internal module describes the behaviour of a specific wind tur-
bine component and can mutually interact with other internal modules. This
way, the combination of these modules describes the behaviour of the complete
wind turbine. The complete model description requires furthermore not only
the application of correct theories in each module, but also sufficient informa-
tion about the required parameters. Examples of data required in the internal
modules are:

• the geometric properties of the wind turbine and its components

• the blade profile characteristics

• generator characteristics

• control algorithms

• mass, stiffness and damping values for all relevant components

Most of this information is wind turbine dependent. However, some of these


parameters vary with the operational state of the wind turbine and, therefore,
all load cases in the design specifications are simulated individually. The re-
quired case dependent information is adapted accordingly for each new sim-
ulation and, finally, each time series corresponds with one load case. Case
dependent data includes furthermore the inputs for the external modules, such
as the wind and wave characteristics and the electric grid behaviour.

Not only loads, but also various other signals need to be calculated in the differ-
ent modules. All the values that are monitored during simulation and exported
as a time series, are often referred to as “sensors” by analogy with the devices
that can measure them during operation. Examples of typical sensors, other
than load sensors, are:

• the rotor position and speed

• the generator position and speed

• the power output

• tower vibrations and deflections

• blade deformations

• control signals such as pitch and yaw settings


52 3. State-of-the-art in the design of a wind turbine drive train

All different modules and the way they interact with each other compose the
model description of the complete wind turbine system. This description con-
tains dedicated theories and relations which are translated into a set of equa-
tions. These numerical expressions can be easily computerised and their solu-
tion is typically calculated by a time stepping solver that yields time series for
all desired sensors. The set of equations consists of four different parts.
1. One set of equations corresponds to the external modules in figure 3.1,
which determine all parameters in the internal modules.
(a) The wind module translates the given wind characteristics into
wind speed time series. This implies the calculation of a wind
field over the rotor at each time step, taking into account the wind
statistics, the wind direction, the wind shear and all other relevant
parameters. Popular models for the calculation of a wind field over
a rotor disc are the Veers model developed at the Sandia National
Laboratories [209] and the Mann model developed at the Risø Na-
tional Laboratory [143].
(b) In case of an offshore wind turbine, the sea waves module trans-
lates the given wave spectrum into wave time series at the support-
ing structure below sea level.
(c) The electricity grid module translates the grid behaviour into rele-
vant parameters at the generator side. This occurs also in the for-
mat of time series allowing variations in the grid.
2. A second set of equations describes the loads acting in the internal mod-
ules, which includes the translation of information from the external
modules into loads.
(a) The BEM theory, implemented in the rotor module, leads to a wind
speed distribution along the span of the rotor blades and conse-
quently to the loads on the rotor.
(b) The generator module determines the torque and speed of the gen-
erator for the given parameters.
(c) The tower module includes models for the calculation of wave
loads based on the wave time series.
(d) Other equations describe the gravitational loads and certain opera-
tional loads.
3. A third set of equations describes the wind turbine’s control system in
the code. During time simulation, the control system module checks
specific sensor values before proceeding to the next time step and under-
takes appropriate measures, such as adjusting the required parameters or
3.3 Traditional wind turbine design codes 53

changing from operational to fault situation. This way, it simulates the


operation of the controller in “real-time”.

4. The fourth set of equations describes the mechanical behaviour of the


complete system. These equations are further called the structural model,
which is discussed below.

The structural model

A structural model describes all motions and deformations in a system. Three


modelling approaches are in use to implement such a model.

1. The first approach is the multibody system (MBS) formulation and is


typically used for the simulation of (large) motion. All structural com-
ponents in a system are represented by individual discrete bodies with
maximally six DOFs, corresponding to three translational and three ro-
tational motions. Each body has mass and inertia properties, but cannot
deform. The joints between the bodies represent the flexibility and the
damping in the system. This approach yields a set of ordinary differen-
tial equations (ODEs) in the traditional wind turbine design codes.

2. The second approach is the finite element (FE) modelling technique.


FE models are typically used for the “internal” analysis of individual
components. Such a component is not represented as a discrete body, but
as a deformable mesh of numerous nodes and elements. Mass and inertia
properties are linked to the nodes, whereas flexibility and damping are
included in the elements. Each node has maximally six DOFs and a
complete FE model can have a large number of DOFs, in the order of
magnitude of 10,000 up to 1,000,000, resulting in long calculation times.
This approach yields a set of partial differential equations.

3. The third approach combines the MBS formulation with the FE formula-
tion. Each component in a system has six rigid-body DOFs representing
its overall motion and, in addition, an extra set of DOFs to represent
its internal deformation. This latter set is derived from an FE model of
the component using the component mode synthesis (CMS) technique.
This approach is also known as a modal formulation or a flexible MBS,
which is discussed elaborately in section 4.6. It can be considered, on
the one hand, as an extension of the MBS formulation with the purpose
of simulating more details or, on the other hand, as a reduction of the FE
formulation in order to reduce the computational time.
54 3. State-of-the-art in the design of a wind turbine drive train

Finally, the structural model describes the physical relation between all DOFs
(and its derivatives) in a system, by expressing an equilibrium of all acting
forces1 . This expression yields a set of equations of motion, which is generally
formulated as:
[M] · {ẍ} + [C] · {ẋ} + [K] · {x} = { f } (3.1)

[M] the mass matrix


[C] the damping matrix
[K] the stiffness matrix
{f} set of force vectors acting on the system
{x} vectors containing the DOFs of all nodes in the system

Section 3.3.2 describes which of the three structural model formulations is


applied in various existing wind turbine design codes. In addition, section 3.3.3
describes the actual structural model in these codes extensively.

3.3.2 Existing software codes and their validation


Section 3.3.1 describes the methodology in traditional wind turbine codes to
calculate load time series on a wind turbine, including the loads on the drive
train. This section gives an overview of various existing codes, which all orig-
inate from specific research groups in the wind energy business. A lot of effort
in the development of these codes has been spent on modelling the interaction
between the aerodynamic loads and the wind turbine’s structure. This inter-
action is often referred to as “aeroelasticity” and, therefore, the corresponding
time simulations are also called aeroelastic simulations. [174] and [178] de-
scribe the evolution and specific issues of modelling the aeroelasticity of wind
turbines. [159] discusses the results of the EU-JOULE project “Verification
of European Wind Turbine Design Codes (VEWTDC)”. This report compares
the simulations of eight codes: PHATAS, Alcyone, HAWC, GH Bladed, Flex4,
Flexlast, Vidyn and Alcyone (free wake). [149] describes furthermore six ex-
tra codes: ADAMS/WT, DUWECS, FAST-AD, GAST, Twister and YawDyn.
The design code DHAT completes the list of fifteen codes, which is considered
here as a comprehensive collection of codes, which are currently applied for
research purposes as well as in industrial designs. Small modifications of the
different codes are frequently required to yield extra insight in specific load
cases or new designs. These requirements lead often to ad hoc code modifi-
cations which are, especially in industry, kept as in-house developments and
rarely published. This makes it difficult to assess all capabilities of the individ-
1 Various methods exist to describe the force equilibrium, but they are not further elaborated

in the present work. One popular example is Lagrange’s equation, which describes a relation
between the kinetic and potential energy in a system.
3.3 Traditional wind turbine design codes 55

ual codes based on a literature survey; however, the basic principles of these
codes remain, despite such modifications.

1. ADAMS/WT (Automatic Dynamic Analysis of Mechanical Systems /


Wind Turbine) [67]. The ADAMS software is a general-purpose multi-
body package developed by Mechanical Dynamics, Inc. (MDI). ADAMS
/WT is an add-on module developed under contract to the National Re-
newable Energy Laboratory (NREL) in Colorado (US). It includes fur-
thermore two subroutines developed at the University of Utah [82]: Aero-
Dyn for the computation of aerodynamic loads and the package YawDyn
described below. The structural modelling in ADAMS/WT is based on
an MBS formulation.
2. Alcyone is developed by the Center for Renewable Energy Sources (CRES)
in Greece in cooperation with the National Technical University of Athens
(NTUA) [159]. The structural modelling in this code is based on an FE
formulation.
3. Alcyone (free wake) originates also from the NTUA, but differs from
Alcyone and, furthermore, almost all other design codes in the type of
aerodynamic model which is applied. It uses the free wake panel method
in contrast with the BEM method [159]. Its other modules are similar to
Alcyone, such as the FE formulation for the structural modelling.
4. GH Bladed is a development of Garrad Hassan and Partners Limited
(GH), Bristol, UK [31, 74]. It is used by GH, as an in-house tool in their
wind turbine design and analysis projects, as well as by many of the
leading manufacturers, research institutes, universities and certification
organisations in the entire wind industry. The software has been verified
and accepted by the organisation Germanischer Lloyd to perform certi-
fication load calculations for wind turbine designs. The structural model
description in GH Bladed is a modal formulation with the option to in-
clude some additional DOFs to represent the drive train, the generator
and other subsystems. Moreover, this design code acts also as a basis for
detailed analyses in several research projects as described among others
by Bossanyi [26–30] and Witcher [222].
5. DHAT (Dynamic analysis of Horizontal Axis Turbines) originates from
Germanischer Lloyd (GL) WindEnergie GmbH, which is an important
certifying body for wind turbines. GL uses DHAT as an in-house tool
and the structural model description in this code is based on a modal
formulation. The drive train is represented as a two-mass spring-damper
system. Buhl and Manjock [35] compared the load simulations in ADAMS
/WT and FAST-AD with simulations in GL’s DHAT.
56 3. State-of-the-art in the design of a wind turbine drive train

6. DUWECS (Delft University Wind Energy Converter Simulation Pro-


gram) originates from the Delft University of Technology and is main-
tained by its Institute for Wind Energy [24]. It is based on a modal
formulation and considers one DOF for the drive train.

7. FAST-AD (Fatigue, aerodynamics, structures, and turbulence) [219].


This design code originates from the FAST code which is developed
at the Oregon State University under the authority of NREL. It uses a
modal formulation for the structural model description. FAST-AD in-
cludes the AeroDyn package from the University of Utah to calculate
aerodynamic loads.

8. Flex4 is developed by Stig Øye at the Fluid Mechanics Department of


the Technical University of Denmark (DTU) [158]. Its application is
widespread, especially in the Danish industry, and it is accepted as a
certified design code. Its implementation is based on a modal formula-
tion including one DOF for the drive train. Flex5 is an extended version
with for instance extra DOFs at the tower foundation to describe the
behaviour of an offshore turbine.

9. Flexlast (Flexible Load Analyzing Simulation Tool) is developed by


Stork Product Engineering, Amsterdam, The Netherlands [214]. It is
mainly used by the Dutch industry for design and certification calcula-
tions and it is described as an MBS tool.

10. GAST (General Aerodynamic and Structural prediction Tool for wind
turbines) [183] originates from the NTUA, similar to the Alcyone codes.
It is developed at the Fluids Section of the Department of Mechanical
Engineering and combines the MBS and FE formulation for structural
modelling.

11. HAWC (Horizontal Axis Wind turbine Code) is developed at the Wind
Energy Department of the Risø National Laboratory in Denmark [168–
170]. This code is used in-house at Risø for research and customer-
oriented design and analysis calculations, but is also commercially avail-
able to the industry. Furthermore, it is accepted as a simulation tool for
certified design loads. The structural model in HAWC is a modal formu-
lation and Nim [155] describes the coupling and reduction of the equa-
tions of motion. For in-house research projects, HAWC often acts as a
basis for more elaborate models [122–124].

12. PHATAS-IV (Program for Horizontal Axis wind Turbine Analysis and
Simulation) is the 4th version of PHATAS, a design code developed at
the Energy Research Center of the Netherlands (ECN) [134]. Structural
3.3 Traditional wind turbine design codes 57

model descriptions in PHATAS are a combination of FE and modal for-


mulations.

13. Twister is a design tool that originates from the consulting office Stentec
B.V., Heeg, The Netherlands [204]. It was known as FKA before 1997
and it is used for wind turbine design and certification purposes.

14. Vidyn is developed since 1983 by Teknikgruppen AB, Täby, Sweden [72].
It can perform static and dynamic analyses of wind turbines based on a
modal model description. [73] describes how a code-generating system
is applied to derive the equations of motion in Vidyn.

15. YawDyn (Yaw Dynamics computer program) [83, 84] is described as


a limited design code to predict preliminary loads in an initial design
phase. It is developed at the University of Utah supported by NREL and
is included in the ADAMS/WT code [82] as described above.

Comparing the simulations of different wind turbine design codes is possible


through benchmark calculations on a specific wind turbine. The correlation
of these simulations with experimental data from this wind turbine allows fur-
thermore to assess the accuracy of all codes. However, performing an extensive
benchmark study is a difficult task and requires the effort of various specialists,
each of them capable of modelling the same wind turbine in a different design
software. Consequently, only few publications describe the results of such
studies. One of them is the valuable VEWTDC report [159] with an extensive
list of conclusions. The general conclusions are summarised as follows:

• The comparison of calculations and measurements was in several cases


hampered by misunderstandings or errors in the specified input. This
emphasises the need for correct communication and quality assurance
during both the design calculations and the measurements. Furthermore,
in all design codes it is difficult to tune the existing uncertainties in the
specified input to better match the measured time series. This is mainly
due to the large number of DOFs in the model descriptions. Proper
insight in the model description can often help to overcome these prob-
lems.

• Power curve calculations agree well with measured data for wind speeds
(far) below rated wind speed Vrat . Near this wind speed, the predic-
tions for stall controlled wind turbines can differ by more than 15%.
Pitch controlled wind turbines have a controlled constant power output
above Vrat and, consequently, negligible differences between predicted
and measured power values.
58 3. State-of-the-art in the design of a wind turbine drive train

• The mutual deviation of the load calculations in the eight design codes
was smallest for the mean blade loads with a limit of ±10%. The simu-
lated fatigue loads on the blades differed up to ±15% and the difference
between the calculated loads on the other components is usually in the
order of ±20% to ±30%.

• A similar trend is visible in the assessment of the deviations between


calculations and measurements. Again, the smallest differences are no-
ticed for the mean blade loads and are in the order of 5% to 10%. The
fatigue loads were compared as 1 Hz equivalent loads and differences in
the order of 5% to 20% are found for the blades. Equivalent loads on the
remaining components (shaft, tower and nacelle) differed often between
10% and 40%. Large deviations of more than 50% were noticed for
the tower top roll moment2 and, furthermore, for several extreme loads
during special load cases.

All reported load time series have a length of 10 minutes which is the typical
standard used in all wind turbine design calculations and measurements. Al-
though it is not specified in the report, it is assumed that the sample frequency
of these time series also equals the industrial standard of 20 Hz. This value lim-
its the bandwidth of the dynamic response to 10 Hz for both the simulations
and measurements. This limitation is often accepted as sufficient to describe
the loads on all wind turbine components and has the advantages of limiting
the calculation time and, moreover, the amount of data that needs to be stored
and often exchanged. However, a reduction of the bandwidth to only 10 Hz
puts considerable limitations on the analysis of dynamic drive train loads as
further discussed in section 3.4.

[36] is a second publication which describes a comparison of different de-


sign codes, namely ADAMS/WT, GH Bladed, FAST-AD and YawDyn. It fo-
cusses mainly on the differences in predicted structural response at the rotor
blades. The analysis of these deviations lead to important code improvements
and correspondingly a better correlation between the simulations. However, no
comparison with experimental results is included. This is a frequent problem
in the wind turbine industry. Since new designs are continuously increasing,
the application of the design codes requires always a new experimental vali-
dation. This is a very time-consuming activity and, therefore, always lagging
(far) behind the design process. As a consequence, the modern multi-megawatt
wind turbine designs are mainly based on validations of wind turbines in the
2 Note that the roll moment on the tower top corresponds closely to the torque in the drive

train. This means that a repeatedly bad correlation between simulations and experiments might
indicate an inaccuracy in the drive train model.
3.3 Traditional wind turbine design codes 59

class below 1 MW. This demands for sufficient reliability in the design, es-
pecially when new concepts are introduced, such as variable speed generators
or pitch control. Reliability assessment leads typically to the application of
specific safety factors. Partial safety factors increase the simulated load levels
and their magnitude depends on the accuracy and availability of experimental
validations, as described for instance in the IEC61400-1 standard [99].

Besides experimental validation of the design codes, further model extensions


can also help to improve the quality of load predictions. Therefore, acquiring
more accurate load simulations is the main driver behind all new software de-
velopments in this field. During the last two decades, considerable effort has
been put in the development of the wind turbine design codes, especially with
respect to aerodynamics. However, despite the substantially achieved progress
in this field, the structural model formulation in these state-of-the-art design
codes is still capable of considerable improvement. Section 3.3.3 gives more
insight in this structural model and its consequences for the drive train design.

3.3.3 Structural model in the traditional codes


In order to comprehend the shortcomings in the traditional wind turbine de-
sign codes with respect to the simulation of drive train loads, it is important
to know the limitations of the structural model in these codes. However, no
exact model description for all existing codes is found in the available public
literature. Nevertheless, based on the author’s knowledge of and experience
with such codes, as well as based on published results from normal modes cal-
culations in such codes (cfr. section 3.3.3.2), the concept of all these codes is
found to be more or less similar:

The structural model in a traditional wind turbine design code has about
16 to 24 DOFs, which can be further classified into:

1. one or two pairs of tower bending modes and one torsional tower
mode (3 - 5 DOFs)

2. two or three pairs of bending modes and no torsional modes for


each blade (12 - 18 DOFs for a three bladed rotor)

3. one additional DOF to represent the rotation of the generator and,


consequently, the torsion in the drive train (1 DOF)

This statement and the applied terminology is further clarified in the remain-
der of this section. Firstly, section 3.3.3.1 describes the different DOFs in the
structural model individually. Section 3.3.3.2 continues with a presentation
60 3. State-of-the-art in the design of a wind turbine drive train

of the eigenmodes and eigenfrequencies calculated for such a model. Finally,


section 3.3.3.3 summarises the consequences for a traditional drive train de-
sign.

3.3.3.1 Clarification of the different DOFs


Figure 3.2 shows the composition of a structural model in a traditional wind
turbine design code. The rotor and the tower are known as the most flexible
parts in a wind turbine and, therefore, their DOFs are usually represented by
their first eigenmodes using a modal formulation.

flexible rotor torque path


( 12 - 18 DOFs )

KDT
hub generator ( 1 DOF )
generator torque
?
tower top

flexible tower
load path ( 3 - 5 DOFs )
(exclusive torque)

Figure 3.2: Schematic representation of the DOFs in a structural model of a


three-bladed wind turbine in a traditional design code.

Tower
The tower is a long, slender and symmetric structure with a huge mass on
top and, consequently, its two first bending modes determine typically the first
modes of the entire wind turbine. These modes differ only slightly in fre-
quency and the corresponding mode shapes are bending deformations of the
wind turbine: one in a direction parallel to the rotor plane and another one
perpendicular to this. Figure 3.3 shows these mode shapes, where the former
is called “transverse bending” and the latter “longitudinal bending”. One ad-
ditional DOF represents usually the first tower torsion mode.
3.3 Traditional wind turbine design codes 61

(a) 1st tower transverse bending mode (b) 1st tower longitudinal bending mode

Figure 3.3: The 1st tower bending modes of a wind turbine (undeformed and
deformed models are shown).

Rotor

A three-bladed rotor model consists of three identical blade models. Each


blade model is represented by two or three pairs of bending modes. One pair
of bending modes is a combination of a flapwise mode and an edgewise mode.
This description of the mode shapes refers to the bending direction with re-
spect to the blade profile. Figure 3.4(a) shows the mode shape at the lowest
frequency which is called the flapwise mode, corresponding to the “flapping”
of a blade. Here, the bending direction is quasi perpendicular to the chord,
since a wind turbine blade is most flexible in this direction. The stiffest direc-
tion, on the other hand, is always quasi parallel to the chord line and yields a
bending mode at a higher frequency. Figure 3.4(b) shows the corresponding
mode shape, which is called the edgewise mode. The same terminology is
applied for the 2nd, 3rd and higher bending modes. The blade torsion modes
are usually omitted because the corresponding eigenfrequencies are typically
several times higher than the dynamics of interest [182].

At normal operation the blades are pitched near 0◦ (cfr. section 2.3.4) which
means that the blade chord lies quasi parallel to the rotor plane and, thus,
the edgewise direction approximately in the rotor plane. Consequently, the
flapwise direction is perpendicular to the rotor plane. Therefore, the terms
“in-plane” and “out-plane” are sometimes used as synonyms for edgewise and
flapwise respectively. However, since variable pitch turbines turn their blades
about 90◦ while braking, these former terms are not coherent and misleading.
62 3. State-of-the-art in the design of a wind turbine drive train

(a) 1st flapwise mode: top view

(b) 1st edgewise mode: side view

Figure 3.4: The 1st bending modes of a wind turbine blade clamped at the root
(undeformed and deformed models are shown).

Each blade mounted on the rotor hub introduces the flapwise and edgewise
bending modes in the model as described above. Since all blades mounted on
the hub are equal, these bending modes appear as a multiple of the number of
blades. For a typical rotor with three blades, this consequently yields three 1st
rotor flapwise modes, three 1st rotor edgewise modes, etc. The eigenfrequen-
cies corresponding to these three modes may differ mutually because of their
“coupling” to the tower. Two of these rotor modes are an asymmetric coupling
of blade modes and one rotor mode is a symmetric coupling. Figure 3.5 shows
this phenomenon for the three first flapwise and edgewise rotor modes for a
wind turbine with one blade in horizontal position.

Drive train

The rotation of the generator in the structural model is represented by one DOF.
The difference between this motion and the rotation of the rotor is the torsion
in the drive train. This twist is represented in figure 3.2 by a torsional spring
with a stiffness value KDT 3 and is determining for the torque transferred to the
generator. This simplified drive train model does not give any insight in the
drive train loads at the level of the bearings and the gears.

3 The determination of the drive train stiffness K


DT requires the calculation of one torsional
stiffness value representative for all flexibilities in the drive train. This is a complex calculation,
which can be simplified by using a multibody system technique, as presented in section 5.5
3.3 Traditional wind turbine design codes 63

(a) 1st rotor flapwise mode (b) 1st rotor flapwise mode (c) 1st rotor flapwise mode
A (asymmetric) B (asymmetric) C (symmetric)

(d) 1st rotor edgewise mode (e) 1st rotor edgewise mode (f) 1st rotor edgewise mode
A (asymmetric) B (asymmetric) C (symmetric)

Figure 3.5: The 1st rotor modes of a wind turbine: side view for the flap-
wise modes and front view for the edgewise modes (undeformed and deformed
models are shown).

The torsion of the drive train couples with the symmetric rotor edgewise mode
in a wind turbine (figure 3.5(f)), since this latter mode has a global torque
component on the drive train. Therefore, the “1st drive train mode” in a wind
turbine corresponds to the “1st rotor torsion mode” or the “1st shaft torsion
mode”. This is further demonstrated in appendix A, which presents a detailed
study of the drive train modes in the traditional wind turbine design codes.
This study includes the following main items.

• A normal modes analysis of a finite element model, which is equivalent


to the structural model of a wind turbine in the traditional design codes,
indicates the origin of the drive train modes.
64 3. State-of-the-art in the design of a wind turbine drive train

• The study indicates the importance of considering the rotor flexibility in


an appropriate way, when calculating the drive train modes. It is proven
that a reduction of the rotor to a single rigid mass, which is an industrial
practice, is inadequate.

• The discussion of various sensitivity analyses indicates the influence of


the following parameters on the drive train modes: the rotor inertia and
flexibility, the blade pitch angles, the drive train stiffness, the genera-
tor inertia, the gearbox support stiffness, the tower flexibility and the
generator behaviour.

3.3.3.2 Results of normal modes calculations


Some publications discuss the output of a normal modes calculation for a wind
turbine in a traditional design code. This section summarises these results with
the purpose of:

1. further clarifying the structural model description

2. yielding insight in the wind turbine’s dynamic behaviour

The terminology for the mode shapes applied in different publications is not al-
ways consistent. Therefore, appendix B contains a verbatim description of the
reproduced references and indicates some of these inconsistencies. Table 3.1
summarises the results with a generalised consistent terminology, based on the
definition of rotor and tower modes as described above. This table includes
results for the following wind turbines:

Power Description
A 500 kW a wind turbine with 19 m blades (Bonus) [171]
B 600 kW a stall regulated wind turbine [124]
C 600 kW a stall regulated wind turbine with a rotor diameter of
44 m (Bonus) [86]
D 1.8 MW a wind turbine with a rotor diameter of 66 m [182]
D1: blades in normal position, i.e. 0◦ pitched
D2: blades 90◦ pitched
E 2 MW a fixed pitch, stall regulated, constant speed wind
turbine with a rotor diameter of 76 m [122]
F 2.75 MW a pitch regulated, variable speed 2.75 MW wind
turbine [86] (in normal production at 5 m/s wind)
G > 2 MW an example of a wind turbine in the HAWCModal demo [181]
3.3 Traditional wind turbine design codes 65

The presented results are all valid for wind turbines at standstill, unless stated
otherwise. This standstill situation implies that the generator end of the drive
train is fixed. The interpretation of the presented results yields the following
insights:

• A modern wind turbine has numerous eigenmodes in the low frequency


range up to 5 Hz. Therefore, the maximum frequency in the traditional
load simulations is usually not higher than 10 Hz.

• It should be noted that the order of the calculated modes is not always
the same for the different wind turbines: e.g. the 1st rotor torsion mode
is the first mode for wind turbine C, while it is the third mode in all other
cases.

• For wind turbines A & C, the results include the frequencies of the first
bending modes for a single blade clamped at the root. The link between
these modes and the global wind turbine modes is further discussed in
appendix A.

• Although table 3.1 contains results for a very limited number of wind
turbines, some trends are visible for increasing power capacity. The
tower bending modes tend to decrease in frequency, indicating the appli-
cation of more flexible tower designs in larger wind turbines. Likewise,
the rotor modes tend to decrease which is an indication for the use of
more flexible blade designs.

• The presented calculations are valid for a stationary wind turbine, except
for wind turbine F. [85] describes how the influence of the rotor rotation
can be included in the calculations by implementing the so-called multi-
blade coordinate transformation [104]. The rotational effect yields a
shift in the eigenfrequencies for the two asymmetric rotor modes. One
mode shifts down in frequency and is called “backward whirling mode”
and the other one shifts up and is called “forward whirling mode”. The
shift in frequency equals 1P, which is the rotational speed of the rotor
(in Hz). The frequencies of the symmetric rotor modes are usually not
affected by this whirling behaviour. For the wind turbine F, this means
that the flapwise rotor mode B at 1.05 Hz is the symmetric one and that
1P equals 0.25 Hz. This corresponds to a rotor speed of 15 RPM, which
is plausible for a wind turbine of 2.75 MW at 5 m/s wind speed.
66

Eigenfrequency (Hz)
Eigen-
Description A B C D1 D2 E F (*) G
mode
500 kW 600 kW 600 kW 1.8 MW 2 MW 2.75 MW > 2 MW
1 1st tower transverse 0.75 0.76 0.73 0.418 0.417 0.39 0.45 0.31
2 1st tower longitudinal 0.80 0.80 0.78 0.419 0.420 0.40 0.45 0.33
3 1st rotor torsion 0.9 0.92 0.57 0.805 0.704 0.62 0.65 0.41
4 1st rotor flap A 1.39 1.41 1.24 0.979 (**) 1.002 0.87 0.70 0.59
5 1st rotor flap B 1.56 1.52 1.48 1.000 1.064 0.94 1.05 0.63
6 1st rotor flap C 1.85 1.81 1.76 1.067 1.769 1.07 1.20 0.75
7 1st rotor edge B 2.91 3.08 2.85 1.857 1.032 1.72 1.65 1.18
8 1st rotor edge C 2.93 3.23 2.95 - - 1.78 2.1 1.23
9 2nd rotor flap A 3.54 2.85 3.43 - - 2.05 2.1 1.60

from different sources in the literature.


10 2nd rotor flap B 4.22 3.54 3.74 - - 2.20 - 1.70
11 2nd rotor flap C - 5.10 - - - 2.58 - 2.10
12 3rd rotor flap A - 4.21 - - - 3.87 - -
13 3rd rotor flap B - - - - - 3.97 - -
14 2nd rotor edge A - - - - - 4.08 - -
15 3rd rotor flap C - - - - 4.43 - -

Blade clamped at the root


1 1st flapwise 1.8 - 1.7 - - - - -
2 1st edgewise 2.9 - 2.94 - - - - -

(*): these results are calculated for normal production at 5 m/s wind
(**): a supposed typing error in the description of the mode shape is corrected

turbines (at standstill) in traditional wind turbine design codes, reproduced


Table 3.1: Typical results of a normal modes calculation for modern wind
3. State-of-the-art in the design of a wind turbine drive train
3.3 Traditional wind turbine design codes 67

3.3.3.3 Consequences for a traditional drive train design


As a result of the simplification of the drive train model in the traditional design
codes, the simulated load time series contain the drive train loads as a force
vector at the hub, containing three forces and three moments, and a torque
value at the generator. This means that for the design of the drive train, the
simulated output needs to be further processed to loads on the individual com-
ponents, such as gears, bearings and shafts. This quasi-static approach imposes
considerable limitations for the drive train design. These limitations are further
discussed in section 3.4.

Nevertheless, today’s certified design calculations still occur with the tradi-
tional design codes, including the limited drive train model. Therefore, the
remainder of this section describes the consequences for the traditional drive
train design process. This process implies, on the one hand, the determination
of extreme load levels for strength calculations and, on the other hand, a re-
duction to relevant fatigue information. This latter reduction can be done by
applying various cycle counting techniques, such as rain-flow counting, range
counting or peak counting on the load variations [182].

Together with the main excitations, the drive train mode dominates the low-
frequency content of the torque variations in the drive train and is consequently
determining in the fatigue calculations. Such calculations typically define the
accumulated damage of a component, based on a material characteristic such
as a S − N Wöhler curve, where S stands for a stress range and N for the
number of corresponding stress cycles leading to failure. The Smith and Haigh
diagrams are similar characteristics, but include furthermore the influence of
the mean stress on the fatigue limits. The translation of the specified stress
histogram for a component during its lifetime to an accumulated damage value
can be done according to the Palmgren-Miner’s rule [196] or the Niemann-
Winter method [154]. The former rule defines the cumulative damage as:

∆n(Si )
D=∑ (3.2)
i N(Si )

∆n(Si ) the number of load variations with a stress level Si


N(Si ) the number of cycles at a stress level Si leading to failure

According to this definition, fatigue failure occurs when the cumulative dam-
age D exceeds a threshold of 1.0. This means that the accuracy of the predicted
drive train mode considerably affects the fatigue calculations for the drive train
components. The amplitude of the corresponding vibration determines the
stress range and the eigenfrequency mainly determines the number of torque
68 3. State-of-the-art in the design of a wind turbine drive train

variations during the wind turbine’s lifetime. This eigenfrequency lies in the
frequency range below 2 Hz, which is covered by the traditional load calcu-
lations. Appendix A describes the importance of using correct values for all
model parameters in order to get an accurate prediction of this frequency and
its corresponding mode shape.

The limitation of the frequency range to 10 Hz in traditional load simulations,


seemed acceptable from the point of view that the internal drive train dynamics
are in a frequency range well above the overall wind turbine dynamics. After
all, the first eigenmodes of the wind turbine dominate these latter dynamics and
they typically lie in the frequency range [ 0 - 10 Hz ], as demonstrated above.
An additional explanation for the traditional limited scope with respect to dy-
namics are the main excitations of a wind turbine. These originate from the
stochastic turbulence in the wind, from the rotation of the rotor and from the
blades passing the tower. A spectral density of the wind speed indicates how
the energy in the wind turbulence is distributed between various frequencies.
Frequently used power spectral densities are the Kaimal spectrum, the Harris
spectrum or the spectrum limitations from the IEC61400-1 standard [99, 182].
They all indicate negligible excitation sources in the wind above 10 Hz. Like-
wise, the excitations at the rotational frequency and at the blade-passing fre-
quencies are mainly below 10 Hz. For a modern three-bladed wind turbine,
these are dominant at 1P and the first multiples of 3P, where P stands for the
rotational speed of the rotor (in Hz). Nevertheless, the presented assumptions
do not cover all drive train issues, as elaborated in the next section.

3.4 Drive train design loads


3.4.1 Limitations of the traditional wind turbine design codes
The simplification of the drive train model in a traditional wind turbine de-
sign code imposes considerable limitations on the reliability of the drive train
design. This section elaborates on these limitations.

3.4.1.1 Dynamic loads


The structural model in the traditional wind turbine design codes includes only
one DOF for the representation of the drive train. This simplification implies
that the design of all drive train components is quasi-static: no vibrations of
the internal drive train components and, as a consequence, no dynamic loads,
are included in the drive train design loads. This approach is only valid when
the drive train’s internal components contribute by any means only negligi-
bly to the dynamics of interest. This approach seemed acceptable from the
3.4 Drive train design loads 69

point of view that the internal drive train dynamics are expected in a frequency
range well above the overall wind turbine dynamics. However, this argument
has never been experimentally demonstrated and extra caution should be ob-
served, since drive train components in new wind turbines are increasing and,
consequently, eigenfrequencies in the drive train will decrease. In addition,
this argument does not cover the complete range of phenomena that can occur
in the drive train. After all, not only external low-frequency excitation of the
drive train is possible, but also internal excitation at higher frequencies exists,
such as a.o. from the following sources.

• The excitation at the gear mesh frequency is a result of the once-per-


tooth variation of the transmission error in a gear mesh, as described
in section 4.3.2. For a parallel gear system, the gear mesh frequency
equals:
fgm = z · ngear , (3.3)
with z the number of teeth of the gear and ngear its rotational speed (in
Hz). For a planetary gear system with a fixed ring wheel, as described
in appendix C, the gear mesh frequency equals:

fgm = zrw · ncarrier , (3.4)

with zrw the number of teeth of the ring wheel and ncarrier the rotational
speed of the planet carrier (in Hz).

• The behaviour of the generator could possibly excite the drive train at
higher frequencies, as a result of:

– passing the notches in the generator stator (the notch passing fre-
quency equals normally the number of poles times the rotational
speed of the rotor in the generator)
– controlling the frequency of a generator indirectly coupled to the
grid, with possible higher harmonics in the electric voltage

• During vibration measurements on a gearbox, it will often be possible to


identify so-called bearing frequencies. Each of these excitations orig-
inates from a particular bearing behaviour. They are widely used for
condition monitoring of bearings; however, they will rarely be consid-
ered as important excitations for load calculations in a drive train.

Such excitations might introduce energy in the range of the internal eigenfre-
quencies and, consequently, lead to harmful resonant behaviour. In order to
avoid resonance in the drive train and predict the dynamic loads on all compo-
nents, new simulation methods are required.
70 3. State-of-the-art in the design of a wind turbine drive train

3.4.1.2 Detailed component loads


The wind turbine industry has been struggling with a series of component fail-
ures in the drive train during recent years. Although the occurrence of such
damage in a commercial product is of course rather kept quiet than made pub-
lic, some exceptional publications describe particular cases. Rasmussen et
al [179] discuss a specific problem with spherical roller bearings in a wind tur-
bine gearbox. Another publication is a recent article by De Vries [51], which
lists a series of trouble spots in wind turbine gearboxes and relates the start
of about 80 % of all trouble cases to faulty bearings. It suggests furthermore
various possible causes for the gearbox problems.
1. Wind gusts at an overall low wind speed can cause load reversals in the
drive train, which may lead to undesired axial load cycles on the high
speed bearings in a helical gear stage.
2. Rapid speed variations at low load levels are problematic for large spher-
ical roller bearings.
3. An insufficient rigidity of the bed plate (cfr. figure 2.11) can result in the
misalignment of gearbox and generator shafts and, consequently, lead
to unwanted vibrations through e.g. a cardan coupling between these
shafts.
4. Lack of a well-dimensioned oil filtration system can lead to accelerated
gearbox failures.
This indicates particular issues in the drive train design on the level of individ-
ual bearings and gears. More insight from more advanced simulation models
is required in order to guarantee a robust design of these components. The
traditional structural model is insufficient to determine accurate local loads on
or local stresses in these components, which are required in detailed fatigue
calculations and in the assessment of allowable stresses.

In addition, the article by De Vries [51] attributes the occurrence of gearbox


problems partly to the continuous market pressure to increase the size and
capacity of wind turbines. This trend led to a gap in know-how, especially
regarding the understanding of the design loads. The main concern is the lack
of insight in the drive train behaviour during various transient phenomena. For
instance, particular control or fault transients at the generator can occur during
specific control actions, short-circuits or grid disturbances. These transients
can lead to unexpected torque reversals or torque peaks, which could then be
harmful to certain drive train components, including the bearings. They might
also cause higher acceleration levels than expected, which is considered as un-
desirable for specific bearing types (cfr. above: large spherical roller bearings),
3.4 Drive train design loads 71

although this effect is still not clearly identified. A new simulation method is
required to identify possible harmful transient phenomena.

3.4.1.3 Safety factors

In order to deal with the uncertainties in the traditional design process, specific
load application factors are recommended in industry for the fatigue calcula-
tion of various drive train components. The ANSI/AGMA/AWEA 6006-A03
standard: “Design and Specification of Gearboxes for Wind Turbines” [6] rec-
ommends a minimum dynamic factor Kv = 1.05 for the design of gears. Other
standards, such as DIN 3990 [54], ISO 6336 [102] and the DNV Classification
Notes 41.2 [58] advise a similar approach for gears. The DIN ISO 281 stan-
dard [55] describes the design calculations for the bearings and corresponding
recommendations for the application factors. Likewise, the DIN 743 [56] stan-
dard is a reference for the fatigue calculations of shafts.

All these application factors are safety factors on the side of the load simula-
tions. In general, safety factors account for the uncertainties in a design and
correspond to a certain reliability. A better knowledge of the reliability yields
always a more cost-effective design, be it through a reduction of the number
of damages or, on the other hand, through a reduction of superfluous design
conservatism and corresponding costs. The former case corresponds to an in-
sufficient safety factor and the latter one to redundancy in the safety factor. The
assessment of the safety factors requires improved simulations of the loads on
the internal drive train components. This implies again the need for additional
numerical simulation methods.

3.4.2 New approach: the MBS formulation


The previous section clarified the limitations in the traditional wind turbine
design codes and the corresponding need for more advanced simulation meth-
ods. An additional motive for this need is the restriction on gaining sufficient
experimental experience. First of all, experiments are only possible on existing
wind turbines and cannot give additional information in the design phase of a
new wind turbine. Furthermore, some experiments are hardly possible, such
as the following examples.

• The experimental validation of a load case with a long recurrence period


(e.g. 1 years) is practically impossible. This complicates for instance
the identification of transient phenomena, which are assumed to occur
rarely, such as extreme wind gusts or generator short-circuits.
72 3. State-of-the-art in the design of a wind turbine drive train

• The measurement of local stresses on various gearbox components, such


as the bearings, is extremely difficult and requires dedicated and expen-
sive sensors, which complicates these experiments.

A summary of all motives yields the main objective in this dissertation:

“The development of a more detailed drive train model to guarantee the


structural integrity of all drive train components with a higher reliabil-
ity.”

Therefore, the focus is on the analysis of drive train loads, rather than on noise
calculations or vibration monitoring. The new method should finally permit
to:

1. predict the dynamic loads on all drive train components: this includes
the calculation of all internal eigenfrequencies in order to avoid reso-
nance in the drive train as well as an accurate simulation of the response
for given excitations.

2. identify harmful transient phenomena: gaining insight in e.g. load rever-


sals or rapid accelerations in particular load cases is important in eval-
uating the suitability of certain drive train components, such as specific
bearing solutions.

3. determine the level and variation of local loads and stresses: an accurate
prediction of the stress ranges and corresponding number of cycles for a
particular drive train component yield more reliable fatigue calculations.

4. assess the redundancy or insufficiency of the applied safety factors: a


logical consequence of the items above, will be an adjustment of the
safety factors.

The translation of these demands into a comprehensive analysis model is fur-


ther complicated by the fact that the development requires already insight in
the internal dynamics and related issues, while this knowledge is often only
available after dedicated simulations or a correct interpretation of experimen-
tal data. It is clear that this development is part of an iterative learning process,
where each step tries to bring the model closer to reality, but can also identify
new relevant phenomena. The present study is an initial step in this process. It
focusses only on the effect of applying various methods and can be considered
as a predecessor for the development of a validated industrial tool.
3.4 Drive train design loads 73

A first option for the implementation of a more detailed drive train model is
adapting changes to the existing design codes. However, this approach has the
following drawbacks:
• Traditional wind turbine codes have typically a limited modularity: their
implementation is often dedicated to specific wind turbine types and dif-
ficult to adapt for new designs and concepts.

• Including extra DOFs is usually not straightforward, which complicates


desired model extensions for a more detailed analysis of drive train
loads.

• The purchase and use of a wind turbine design code by industrial or


research companies, only involved in the design of drive train compo-
nents, may furthermore be considered as an overkill, especially when
other standard tools for structural modelling are already available.
A second alternative is the development of a new model in a general-purpose
FE or MBS software package. These software tools offer generally a large va-
riety of features for structural modelling and, additionally, the description of a
model is often less complicated and easily extendable. These tools lack, how-
ever, the specific wind turbine modules implemented in the traditional wind
turbine design codes. This conflict leads to two trends in the evolution of wind
turbine modelling. On the one hand, the wind turbine design codes grow to-
wards general-purpose tools with more features and a better extendability. On
the other hand, dedicated developments occur in standard FE or MBS soft-
ware to better describe the wind turbine behaviour. A combined effort by ex-
changing (partial) model descriptions can further accelerate the improvement
of wind turbine models.

In a comparison of the two options for developing a new drive train model, the
advantages of the large modularity, the better extendability and the variety of
available features in the latter approach are judged more suitable in an iterative
development process. Therefore, this dissertation introduces model extensions
in a general-purpose software package. The MBS formulation with the option
of including flexible elements is the most appropriate modelling technique to
meet the identified needs. The latter option is required to predict local stresses
in drive train components and to include their flexibility accurately. It implies
that the particular MBS software package should have built-in features to in-
clude reduced FE models. The software “DADS” fulfils all requirements and
is selected as the best alternative to implement the new models.

DADS: (Dynamic Analysis and Design System) originates from the company
CADSI (Computer-Aided Design Software, Inc.) which was founded
74 3. State-of-the-art in the design of a wind turbine drive train

in 1983 in Coralville, Iowa (US). In 1999, DADS became LMS DADS


when CADSI was bought by LMS International from Leuven (Belgium).
Since 2001, LMS DADS is an integrated part of the virtual prototype
simulation software LMS Virtual.Lab and is known as LMS Virtual.Lab
Motion [207]. The software LMS DADS Revision 9.6 [138] is used in
the present work and further called DADS. It is a general-purpose multi-
body package and has the required built-in option to include reduced FE
models. In addition, it has features to communicate with MATLAB and
to implement user-defined subroutines. Section 3.4.3 compares the main
specifications of DADS with other software packages.

Chapter 4 describes the actual application of the MBS formulation for the mod-
elling of a drive train4 . It presents three levels of complexity in the multibody
models, going from one DOF of freedom to six DOFs per body and, finally, to
flexible bodies by including a reduced FE model.

3.4.3 Overview of similar research activities


The specific modelling of drive trains in wind turbines is rather new and un-
known. The limited number of publications on this topic and their recent dates
of publication affirm the novelty of this issue. This section summarises the
known research activities in this domain, which comprise only developments
starting from general-purpose software packages. No evidence is found in the
literature for the implementation of new drive train models in existing wind
turbine design codes. Nevertheless, this evolution is assumed to occur too, al-
beit rather as in-house developments of wind turbine manufacturers and, there-
fore, less communicated to the general public. The summary below compares
the applied modelling techniques and corresponding software with the MBS
methodology in the software DADS. Table 3.2 shows a comparison of various
features for the different software packages.

1. Krull [115, 116] describes the prediction of natural frequencies and the
load simulation during two transient phenomena in a drive train of a
wind turbine by means of the software package DRESP [79] (Dreh-
schwingungs-Simulation-Programm). This is a German noncommercial
software developed at the Institut für Maschinenelemente und Maschi-
nengestaltung (IME) of the RWTH Aachen University under the author-
ity of the Forschungsvereinigung Antriebstechnik e.V. (FVA). It is avail-
able to all members of the FVA. DRESP is an MBS software dedicated
4 Although the methodologies and modelling techniques in the present work could be applied

to direct-drive wind turbines as well as to wind turbines including a gearbox, the focus is limited
here to the latter type. Models of such wind turbines are considered to be comprehensive with
respect to the defined interests.
3.4 Drive train design loads 75

to the analysis of torsional vibrations in drive trains. Therefore, only one


DOF per body is considered in the model description.

DYLA III (Dynamische Lagerkräfte) [100] is another noncommercial


MBS software tool from the FVA developed at IME. It can predict dy-
namic bearing loads in various types of gear stages and predict natural
frequencies in the drive train. DYLA III is very similar to DRESP, but
its structural model includes six DOFs per body. However, the imple-
mentation of this option is not yet completely finished for the modelling
of planetary gear stages. No evidence is found for any research activities
in DYLA III related to wind turbine drive trains.

A dynamic simulation tool of the FVA dedicated to planetary gear stages


is the SIMPLEX software developed in the FVA project No. 51/I-IV
at the Lehrstuhl für Maschinenelemente, Getriebe und Kraftfahrzeuge
(LMGK) of the Ruhr-Universität Bochum (Germany). A SIMPLEX
model includes six DOFs for each gear in a planetary system and can
be used to calculate a.o. the system’s natural frequencies and its dy-
namic response, including the elastic deformations of the teeth and the
bearings. Blümm [23] illustrates the use of SIMPLEX for the calcula-
tion of dynamic load factors required in the design of a planetary gear
stage. He considers furthermore a system of two individual planetary
stages, which are coupled purely torsionally. Polifke [173] emphasises
in addition the importance of modelling the coupling between two stages
in more detail and describes its implementation in SIMPLEX. Although
this software could offer various valuable features to model planetary
gear stages, which is required in the modelling of a drive train in a wind
turbine, it is limited in its capabilities to model a complete drive train
integrated in a wind turbine. The author is not aware of any research ac-
tivities in SIMPLEX related to the simulation of a drive train in a wind
turbine.

2. Heege [90–92] describes a dynamic analysis of a wind turbine based on


a MBS formulation with flexible bodies in the Samcef-Mecano software
package. The focus in the study is on the prediction of gearbox loads and
the applied methodology is similar to the present work. Furthermore,
Samcef-Mecano is a general-purpose multibody package like DADS.
It is developed by the Samtech company which was founded in 1986
from the Aerospace Laboratory of the University of Liège in Belgium.
Resulting from Heege’s recent developments in this software related to
wind turbines, Samtech launched in 2004 “Samcef for Wind Turbines”.
76 3. State-of-the-art in the design of a wind turbine drive train

3. Schlecht et al [185–187] from the IMM department at the TU Dres-


den are also developing numerical models for wind turbines with more
details for the drive train. They use the software package SIMPACK,
which is again a general-purpose MBS software with an optional inclu-
sion of flexible bodies. SIMPACK is developed by INTEC GmbH in
Germany [101].

4. Gold et al [80] from the IME department at the RWTH Aachen use both
DRESP and SIMPACK models to simulate the drive train loads in a
wind turbine. They use the DRESP model in a pre-simulation to predict
the external loads at the generator side. These loads are imported in the
SIMPACK model of the complete wind turbine, which includes flexible
bodies.

5. Larsen, Thomsen and Rasmussen [124] from Risø describe a dynamic


analysis of a single spur planetary gear stage in a gearbox for a 600 kW
wind turbine. Their model includes 20 D0Fs and is not coupled to any
other gear stage, nor to the rest of the wind turbine. The mathematical
description of gear contacts and bearings is based on the work of Lin
and Parker [129], which is also used as a starting point for modelling
planetary gears in the present work. The software used for the imple-
mentation of this mathematical description is not mentioned. The study
includes response calculations in the time domain using external loads
calculated in Risø’s wind turbine design code HAWC (cfr. section 3.3).
However, the gearbox model and HAWC model are independent and no
further evolution towards an integration is known by the author.

6. The add-on module ADAMS/WT is developed in the general-purpose


multibody package ADAMS and was already included in the list of spe-
cialised wind turbine design codes (cfr. section 3.3). Although this mod-
ule could eventually be used as a basis for the development of more
detailed drive train models in ADAMS, no evidence has been found for
such evolutions, nor any recent information about the status of ADAMS/WT.
Therefore, ADAMS is considered here as a general-purpose software
only. ADAMS is currently sold and further developed as MSC.ADAMSr
by the MSC.Software Corporation, California (US) [150]. MSC.ADAMSr
is capable of including reduced FE models in the MBS formulation.

The overview describes the use of various general-purpose software packages


to simulate wind turbines with the purpose of gaining more insight in the drive
train behaviour. Although these packages have numerous built-in features,
the essence in all research activities above is still the model creation. This
3.4 Drive train design loads 77

implies often the implementation of user-defined subroutines to fulfil all de-


mands. This requires sufficient experience and a thorough insight in differ-
ent methodologies. The present work gives an overview of various possible
methodologies and elaborates on their implementation as well as on the corre-
sponding consequences during simulation.
78

Software Research Publications 1 DOF 6 DOFs flexible user-


institute related to per per bodies defined
(*) wind turbines body body subroutines
ADAMS x x x x

loads in a wind turbine.


DADS KULeuven cfr. p.239 x x x x
TU Bielefeld [14]

DRESP IME (RWTH Aachen) [80, 115, 116] x

DYLA III x x

SIMPLEX x x

Samcef SAMTECH (**) [90–92] x x x x

SIMPACK IMM (TU Dresden) [185–187] x x x x


IME (RWTH Aachen) [80]

(*): Research institute using this software for the modelling of wind turbines
(**): SAMTECH is the company which is developing the Samcef software, rather than a research institute

Table 3.2: Comparison of software packages for the simulation of drive train
3. State-of-the-art in the design of a wind turbine drive train
3.5 Conclusions 79

3.5 Conclusions
The design specifications of a wind turbine are defined according to dedicated
standards, such as the IEC61400-1 standard, the GL rules, the NEN 6096 stan-
dard or the DS 472 standard. These specifications include various external
conditions, with the wind speed and turbulence as the most important param-
eters. The combination of these conditions with all possible normal and fault
situations for the wind turbine yields a large set of different load cases, which
can occur during the wind turbine’s twenty year lifetime. The determination of
correct levels for the aerodynamic, gravitational, inertial and operational loads
and the respective load variations during all these load cases, requires dedi-
cated software codes.

The state-of-the-art in these software packages are specialised wind turbine de-
sign codes. They consist typically of different modules, classified as external
and internal modules. The wind, the electricity grid and the sea waves are ex-
amples of the former type, since they can be considered as pure model inputs,
independent of the wind turbine behaviour. The internal modules describe the
loads acting on the wind turbine, the wind turbine’s control system and its
mechanical behaviour. This latter part is included in the so-called structural
model. The MBS formulation, the FE formulation and the modal formulation
are three modelling approaches to implement this model. Most of the exist-
ing wind turbine design codes apply the latter approach and the solution of
the corresponding equations of motion yields, among others, load time series
with a typical bandwidth of 10 Hz. The present work contains an overview
of the main specifications of the following codes: ADAMS/WT, Alcyone, Al-
cyone (free wake), GH Bladed, DUWECS, FAST-AD, Flex4, Flexlast, GAST,
HAWC, PHATAS-IV, Twister, Vidyn and Yawdyn.

The experimental validation of these codes is complicated, since new designs


are continuously increasing in size and often require adapted simulation tools.
Therefore, it is typically lagging far behind the design process. Nevertheless,
the search for a good correlation between experimental and numerical results
drives the development of model extensions and improvements. Consider-
able effort has been put already into the aerodynamics, however, the struc-
tural model description is still rather limited. This description is found to be
very similar for all design codes and includes sixteen to twenty-four DOFs
to describe the complete wind turbine. These DOFs are classified into three
to five DOFs for the tower, twelve to eighteen DOFs for the rotor and one
DOF to represent the torsion in the drive train. Appendix A presents a study
of the calculation of drive train modes based on this model description. The
consideration of only one DOF for the drive train in a wind turbine imposes
considerable limitations on the reliability of its design.
80 3. State-of-the-art in the design of a wind turbine drive train

1. As a consequence of the model limitations, no dynamic loads on the in-


ternal drive train components are included in the traditional design loads.
However, internal excitation and corresponding load amplifications are
possible in the drive train, such as from the gear mesh frequencies or
from the generator. Extra caution should be observed, since the internal
eigenfrequencies are decreasing for new and larger wind turbines.

2. A recent article of De Vries [51] relates a series of gearbox failures in


wind turbines to the lack of insight in local loads and stresses in a drive
train and an insufficient understanding of the design loads, which are
consequences of the limited structural model.

3. The structural model does not allow to assess the redundancy or insuffi-
ciency of the applied safety factors.

These limitations imply a need for additional numerical simulation methods to


guarantee the structural integrity of all drive train components with a higher
reliability. The flexible MBS formulation is the most appropriate modelling
technique to meet the identified needs and the software DADS is selected
as the best alternative to implement the new models. Other general-purpose
multibody packages used by certain research institutes for modelling a drive
train in a wind turbine are DRESP, Samcef-Mecano and SIMPACK.
4

Detailed modelling of the drive


train in a wind turbine

4.1 Introduction
The main objective in this dissertation is the development of a more detailed
drive train model to guarantee the structural integrity of all drive train com-
ponents in a wind turbine with a higher reliability. Chapter 3 indicates the
limitations of the existing wind turbine design codes and presents the applica-
tion of the MBS formulation in the software DADS as the best alternative to
implement more detailed drive train models. This section describes the appli-
cation of this methodology in more detail.

Firstly, section 4.2 further elaborates the problem formulation by describing


the load transfer in a drive train of a wind turbine. This includes a description
of the general layout of the drive train and a detailed discussion of the loads
in a gearbox. Subsequently, section 4.3 elaborates on various drive train mod-
elling techniques. It surveys relevant modelling issues found in the literature
on gear dynamics and summarises the state-of-the-art in mathematical expres-
sions to describe the load transfer, the flexibility and the damping in the tooth
contact as well as in the bearings and other drive train components. The next
sections describe a combination of these expressions with the implementation
of further model extensions in three different modelling approaches.

These sections describe a step by step extension of the drive train model. The
approach in section 4.4 considers only one DOF per drive train component,
while the implementation in section 4.5 comprises all six DOFs. Section 4.6
describes the addition of extra DOFs to represent the flexibility and the dy-
namic behaviour of individual components, which is based on the component

81
82 4. Detailed modelling of the drive train in a wind turbine

mode synthesis (CMS) technique. This gradual increase in DOFs aims at as-
sessing their individual contribution to the internal dynamics and their added
value in the present analyses and for the design process. Chapters 5 and 6
discuss the results of dedicated analyses, based on the different modelling ap-
proaches, and attempt to yield insight in the possibilities and limitations of the
implemented solutions.

4.2 Load transfer in the drive train


The scope in the present work is on wind turbines including a gearbox. Sec-
tion 4.2.1 describes a general drive train layout for such wind turbines. In
addition, section 4.2.2 discusses the load transfer in a gearbox.

4.2.1 General drive train layout

Rotor hub Coupling


Brake disk 
Main bearing Gearbox
E  Generator
E 
E 
Main shaft 

Bed plate

Figure 4.1: Layout of a wind turbine drive train including a gearbox, repro-
duced from [25].

Section 2.3.3.2 introduces different drive train concepts for a wind turbine with
a gearbox. Figure 4.1 shows the layout of the three-point-suspension type
with a more detailed focus on the different drive train components. The main
shaft connects the rotor hub with the gearbox. The main bearing and a second
bearing inside the gearbox carry this main shaft and transfer all loads from
the hub to the bed plate, except for the torque. The main shaft introduces this
torque at low speed into the gearbox, where it is transformed to a lower torque
at a higher speed. Equation (4.1) describes this transformation as:
4.2 Load transfer in the drive train 83

Tin · ωin = Tout · ωout · ηgb , (4.1)

where ηgb is the gearbox efficiency. Considering the gearbox as a black box, it
is clear that its support should carry the resulting reaction torque which equals
the difference of Tin and Tout :

Tres = Tin − Tout


ωin 1
= Tin · (1 − · )
ωout ηgb
1
= Tin · (1 − ), (4.2)
igear · ηgb

where igear is the gearbox ratio, which is negative when the generator rotates in
the opposite direction of the rotor. Modern wind turbine applications with an
induction generator have typically an overall gear ratio from forty up to more
than one hundred. As a result, the reaction torque equals approximately the
input torque. The gearbox support transfers this torque to the bed plate, which
carries furthermore the generator. This implies supporting the weight and the
reaction torque of the generator. The connection between the generator and
the high speed output shaft of the gearbox occurs with a coupling, which is
usually flexible in bending and axial direction, but rather stiff in torsional di-
rection. Between this coupling and the gearbox, the mechanical brake disk is
usually mounted.

Both spur gears and helical gears are used in wind turbine gearboxes as well
as parallel gear stages and planetary gear stages. Figure 4.2(a) shows a pin-
ion and a wheel in a helical parallel gear stage and figure 4.2(b) describes the
concept of a helical planetary gear stage with three planets. A planetary gear
stage consists of a planet carrier, a ring wheel, a sun and a number of planets.
This number is a design variable which is most often chosen equal to three.
Planetary gears are successful in wind turbine applications, because they yield
a high torque density in comparison with parallel stages. This means that they
transfer more torque for the same amount of material required in the design. A
corresponding weight reduction is favourable in wind turbines. Furthermore,
the bearing loads are reduced and planetary gears are more compact. Depend-
ing on the desired overall gear ratio, a wind turbine gearbox is typically a
combination of one low speed planetary stage with two parallel stages or two
planetary stages with only one high speed parallel stage. In general, the plan-
etary stage is designed for gear ratios up to seven and a parallel stage usually
up to five.
84 4. Detailed modelling of the drive train in a wind turbine

ring wheel

planet
@@
I
@

sun





planet carrier

(a) Helical parallel gear stage (b) Helical planetary gear stage with three planets

Figure 4.2: Common gear stages in a wind turbine gearbox.

The Willis formulas, published by Robert Willis in 1870 [218], describe the
kinematics of a planetary gear stage and indicate that such a system has two
kinematic DOFs. Therefore, it needs a constraint in order to yield a fixed gear
ratio. Depending on whether the ring wheel, the sun or the planet carrier is
fixed, a different ratio is achieved. The most popular concept in wind turbine
gearboxes is using the planet carrier as input and the sun as output, while the
ring wheel is fixed. In this case, the sun rotates at a speed ωsun which equals:
zrw
ωsun = ω pc · (1 + ) (4.3)
zsun

ωsun rotational speed of the sun


ω pc rotational speed of the planet carrier
zrw number of teeth of the ring wheel
zsun number of teeth of the sun

The input and output shaft of this planetary stage are concentric and rotate in
the same direction. Appendix C further elaborates the kinematic relations be-
tween the different components for a planetary stage with a fixed ring wheel
and, in addition, it discusses an example.

The input torque acting on the planet carrier, in the specific system as described
above, goes through the planet bearings towards the planets. All planets are in
contact with the ring wheel as well as with the sun and the load transfer results
4.2 Load transfer in the drive train 85

in an output torque at the sun (Tsun ) and a reaction torque at the ring wheel
(Tres ). Based on equations (4.1), (4.2) and (4.3), these torques equal (in the
friction less case):
zrw
Tout = Tsun = Tin /igear = Tin /(1 + ) (4.4)
zsun
1 zrw
Tres = Trw = Tin · (1 − ) = Tin · (4.5)
igear zsun + zrw

Figure 4.3 shows a cross section of a typical gearbox for a wind turbine in the
1 MW class with one planetary stage and two parallel gear stages. The plan-
etary stage has spur gears and its ring wheel is fixed in the gearbox housing.
The second gear stage is a helical parallel stage. Its wheel is driven by the sun
of the first stage and its pinion drives the wheel of the third stage, which is
also a helical parallel stage. The pinion of this latter stage is connected to the
generator. Since both parallel gear stages cause a change in the direction of
rotation, the generator rotates in the same direction as the rotor. According to
equation (4.2), the reaction torque of the gearbox on the bed plate is therefore
slightly smaller than the input torque. This torque acts in the same direction as
the rotation of the rotor, which is clockwise when looking at the wind turbine
in the direction of the wind. The torque on the generator support acts also in
this direction.

4.2.2 Gearbox loads


The internal components of the gearbox are the gears, the components carry-
ing the gears and the bearings supporting these components. The gear teeth
transfer the torque between the different gears. They have an involute profile
and figure 4.4 shows an example of an external spur gear pair. It is a standard
gear pair, i.e. without addendum modifications. The input torque Tin drives the
pinion which makes contact with the wheel in two contact points: CP1 and
CP2. When a pair of teeth is in contact, the teeth roll on each other and the
contact point moves along the line of contact (LOC). This line is perpendicular
to the tooth flanks in contact and tangent to the base circles (BC). The common
tangent (TAN) of the pitch circles (PC) is the line on which the gears seem to
roll. The angle between this line and the line of contact is the pressure angle


αn . The direction of the contact force between two meshing gears ( Fbn ) lies
furthermore along the line of contact. Consequently, this force has a tangential
component along the common tangent and a radial component along the cen-
treline. The tangential component causes a torque Tout on the wheel. For the
given standard spur gear pair, the following equations describe the magnitude
of the tooth contact force and its components for the friction less case accord-
ing to [151]:
86 4. Detailed modelling of the drive train in a wind turbine

pinion 3rd
stage H H HH

gear 3rd
stage H
HH

shrink disk
@ HH
HH
pinion 2nd st.

A
A HH
 A H
 A H
 gear 2nd stage
 A
 A
planet carrier @ A
sun
@
@
B planet
B
B
torque arm H ring wheel
H H

Figure 4.3: Gearbox for a wind turbine in the 1 MW class, with one spur
planetary gear stage (1st stage) and two helical parallel gear stages (2nd and
3rd stage).

T1,2 Ft
Ft = Fbn =
d1,2 /2 Fr = Ft · tan αn cos αn
(4.6) (4.7) (4.8)

T1 the reaction torque on the pinion,


T2 the output torque at the wheel,
Fbn the tooth contact force on the respective gear directed along
the line of contact and, consequently, tangent to the base circle
(b) and normal to the involute profile (n),
Ft the tangential component of the tooth contact force,
Fr the radial component of the tooth contact force,
4.2 Load transfer in the drive train 87

wheel

ωout Tout BC2


CP2 


t
 αn HH


PC2
t

TAN 
CP1
A PC1
A 
LOC
ωin Tin

BC1
pinion H
H
centreline

Figure 4.4: Involute profile of a standard external spur gear pair.

d1 and d2 are the respective pitch diameters and define the gear ratio as:

igear = d1 /d2 = ω2 /ω1 = T1 /T2 (4.9)

For a given gear module m, the number of teeth equals:

z = d/m, (4.10)

and, since meshing gears have an equal module, the gear ratio in equation (4.9)
can be written as:
igear = z1 /z2 (4.11)

For gears with certain addendum modifications, the line on which the gears
seem to roll is no longer the common tangent of the pitch circles. Here, the
circles which seem to roll on each other are called the operating pitch circles.
They differ from the former pitch circles, which are usually described as ref-
erence pitch circles since they define the size of the tooth manufacturing tools.
As a result, the operating pitch diameters d10 and d20 substitute the reference
pitch diameters in equations (4.6)-(4.9) for non-standard spur gears and, fur-
thermore, the actual pressure angle α0n is used instead of the reference value
αn . This latter value equals 20◦ for most industrial gears.
88 4. Detailed modelling of the drive train in a wind turbine


− Fa
Fbn

Ft Fr

Fbn Fa
βb

Fbt
(a) Force vector on the tooth flank (b) Plane of action

Ftn

αn Ftn Fa

Fbn
Fr β

Ft
(c) Cross section normal to the involute profile (d) Top view

Ft Fa
αt αa

Fr Fr
Fbt Fba

(e) Transversal cross section (f) Longitudinal cross section

Figure 4.5: Contact forces on a helical tooth.

For a helical gear pair, the plane normal to the involute profile no longer co-
incides with the transversal plane of the gears. It is rotated by an angle β for
one gear wheel and, consequently, by -β for the other wheel in order to mesh


appropriately. Figure 4.5 shows how the tooth contact force ( Fbn ) acts on a
single tooth of a helical gear. This force vector can be resolved into different
components as shown in the cross sections of different planes.

• Figure 4.5(b): the plane of action is tangent to the base circles of both
gears; the helix angle in this plane is therefore called βb . The axial
component of the force vector is consequently:

Fa = Fbn · sin βb (4.12)


4.2 Load transfer in the drive train 89



The projection of Fbn in the transversal plane equals:

Fbt = Fbn · cos βb (4.13)




• Figure 4.5(c): Fbn acts in a plane normal to the involute profile; its radial
component is consequently:

Fr = Fbn · sin αn (4.14)




with αn the pressure angle in this plane. The component of Fbn in this
plane perpendicular to Fr equals:

Ftn = Fbn · cos αn (4.15)

• Figure 4.5(d): the helix angle β is defined in this plane. Based on equa-
tion (4.15), the tangential force on the gears equals:

Ft = Ftn · cos β = Fbn · cos αn · cos β (4.16)

This component causes the torque on both gears and complies accord-
ingly with equation (4.6).

• Figure 4.5(e): based on equations (4.14) and (4.16), the transversal pres-
sure angle can be written as:

tan αt = tan αn /cos β (4.17)

• Figure 4.5(f): the axial pressure angle αa is defined in the longitudinal


plane, but is further rarely used.

• The relation between βb and β can be derived from the equations above
as:

sin βb = sin β · cos αn (4.18)


tan βb = tan β · cos αt (4.19)
cos αn sin αn
cos βb = cos β · = (4.20)
cos αt sin αt

The application of certain addendum modifications exists also for helical gears
and implies the use of the actual pressure angles (α0n , αt0 , α0a ), the actual helix
angles (β0 , β0b ) and the operating pitch and base diameters (d 0 and db0 ) in the
force calculations, with:
db0 = d 0 · cos αt0 (4.21)
90 4. Detailed modelling of the drive train in a wind turbine

Using these actual values, makes equations (4.12)-(4.20) generally applicable


for all gear pairs. The generalised formulation of equations (4.6)-(4.8) is:

T
Ft = Fr = Ft · tan αt0
d 0 /2
(4.22) (4.23)

Ft
Fa = Ft · tan β0 Fbn =
cos α0n · cos β0
(4.24) (4.25)

All individual parts of a gearbox, which are loaded as a result of the torque
transfer, will deform under the respective load components, such as axial, tor-
sional, shear and bending loads. The relation between the load and the re-
sulting deformation corresponds to the flexibility of the part. The following
overview introduces different categories for these flexibilities.
1. Tooth flexibility: all teeth in contact of a gear pair under load exhibit
bending deformation, which can be represented as a tooth stiffness be-
tween the gears (gear mesh stiffness). Since the tooth contact force lies
along the line of contact, the corresponding stiffness is defined in this
direction. This stiffness value is further called κgear and is defined as
the average normal distributed tooth force in the normal plane causing
the deformation of one or more engaging tooth pairs, over a distance
of 1 µm, normal to the involute profile in the normal plane; this defor-
mation results from the bending of the teeth in contact between the two
gear wheels of which one is fixed and the other is loaded. δ denotes this


deformation and, consequently, the magnitude of Fbn equals:

Fbn = δ · κgear (4.26)

Figure 4.6 clarifies this equation. Note that δ corresponds to a relative


rotation of both gears on top of their absolute rotation, which is defined
by the kinematic gear ratio.

2. Bearing flexibility: a bearing constrains specific displacements and/or


rotations of a gearbox component, depending on the type of bearing. The
constrained motion implies a corresponding load component and, conse-
quently, a resulting deformation of the bearing. Therefore, the flexibility
of a bearing is typically represented by individual stiffness values for the
different load components.
4.3 Drive train modelling techniques 91

line of contact
T
δ

κgear

base circle pinion


base circle wheel

The torque T drives the pinion, while the wheel is kept fixed. The
resulting tooth contact force lies along the line of contact, which is
tangent to the base circles. This force causes a bending deformation of
the engaging tooth pairs, which is represented as a deformation δ along
the line of contact. The ratio of the force and the deformation equals
the gear mesh stiffness κgear .
Figure 4.6: Definition of the gear mesh stiffness κgear in the normal plane of
two gears in contact.

3. Component flexibility: this represents the relation between loads and cor-
responding deformations of all parts of the gearbox, other than the gears
and the bearings. This includes among others the shafts, planet carriers,
keys, splines and furthermore the gearbox housing. The representation
of the corresponding stiffness values in a drive train model differs for an
MBS, an FE or a flexible MBS formulation, as described hereafter.

The next section describes the modelling of a drive train in more detail and
gives an overview of the existing literature on gear dynamics.

4.3 Drive train modelling techniques


4.3.1 Introduction
Further insight in the internal dynamics of a drive train requires a proper struc-
tural model to describe the load transfer and flexibilities, as defined in the
previous section, in combination with all inertia properties and damping val-
ues of the drive train components. Section 3.3.1 introduced three modelling
approaches for a structural model, which are the FE formulation, the MBS
formulation and the flexible MBS formulation.
92 4. Detailed modelling of the drive train in a wind turbine

1. The FE formulation is typically applied for discretisations of flexible


structures into a large number of DOFs. In general, the number of el-
ements varies between the orders of magnitude from hundred up to a
million. As a result, these models yield detailed information about the
internal stresses and strains, but they are relatively slow with respect to
time domain calculations for large systems. Therefore, in case of a drive
train, the use of FE models is generally limited to the analysis of an
individual drive train component.

2. A structural model of a complete drive train is usually based on an MBS


formulation. This approach implies a splitting up of the drive train into
discrete rigid bodies, which yields typically a relatively small number of
DOFs and consequently the fastest calculation times. The interest in an
MBS analysis is the overall motion of the drive train components, rather
than their deformation. This motion corresponds to the DOFs of a body.
Several books related to mechanical vibrations and more specifically to
rotating machinery, such as [52, 59, 75, 119, 180], describe that the anal-
ysis of dynamic drive train loads starts with the investigation of the drive
train torque. Consequently, at least the torsional DOFs of all bodies in
a drive train are required. Multibody models in which the DOFs per
body are limited to this torsional DOF only, are further called (purely)
torsional multibody models. Section 4.4 elaborates on their implemen-
tation. In addition, section 4.5 discusses the implementation of more
detailed models with six DOFs per body.

3. The MBS approach assumes that each body does not deform. This as-
sumption is only accurate when the behaviour of the individual compo-
nents does not interact (dynamically) with the analysed phenomenon.
For instance, when a drive train component has an eigenmode at a fre-
quency of 4 Hz, it can usually not be considered as rigid in the cal-
culation of a frequency response function up to 10 Hz. Moreover, the
assumption does not allow any insight in the internal stresses of a com-
ponent. These problems can only be tackled with a flexible MBS for-
mulation. Here, an FE model of the component is reduced to its modal
representation, which includes usually its static deformation properties
and its dynamic response. This modal representation is then combined
with the MBS formulation of the drive train. Section 4.6 discusses the
implementation of flexible multibody models.

A discussion of drive train modelling techniques can be found in all publica-


tions related to research on “drive train dynamics”. This concept covers in
general all issues related to vibrations and/or noise in the drive train. This is a
very broad perspective additional to the calculation of the static load transfer
4.3 Drive train modelling techniques 93

starting from the specifications of the external forces. Each individual issue
requires typically dedicated analyses and modelling tools, which leads to the
introduction of various fields of application and corresponding domains of re-
search. As a result, a large number of publications is available covering a wide
range of drive train issues. A non-comprehensive subdivision of this range
gives further insight in relevant topics.

1. Drive train loads:

• prediction of absolute load values and variations for a specified


external load spectrum;
• investigation and assessment of internal excitations, e.g. from mesh-
ing gears;
• analyses to avoid resonance and its corresponding load amplifica-
tion;
• research of nonlinear phenomena and their influence on the loads,
such as for instance friction and backlash;

2. Noise radiation:

• tracing of excitation sources;


• analysis of the transfer path from the vibration source to the noise
radiator;
• calculation of actual noise radiation;

3. Condition monitoring:

• definition of parameters that can describe the machine condition;


• translation of monitoring requirements into a suitable data-acquisition
(DAQ) system;
• processing from the measured data into indicators for the machine
condition and assessment of specific indicator values;

4. Control:

• design and implementation of control actuators;


• development of control algorithms, e.g. to reduce loads and noise,
to optimise the efficiency, to guard the machine state, . . .
94 4. Detailed modelling of the drive train in a wind turbine

A particular interest with respect to the drive train may be related to a number
of topics from this artificial subdivision. For instance, dedicated DAQ tech-
niques for condition monitoring purposes can also be useful for load validation
or in control systems. Likewise, the load variation prediction could yield the
required data for noise radiation calculations. Furthermore, the consideration
of nonlinear effects is also required for control issues.

This brief overview sketches the elaborateness of the research domain related
to drive train dynamics. The remainder of this section does not aim at giving
a comprehensive overview of this domain, but limits its focus to the most rele-
vant part for the present work, which is the research on gear dynamics. Various
publications related to this subject yield insight in dedicated modelling tools.

4.3.2 Review of the literature on gear dynamics


Özgüven and Houser [160] have described a comprehensive review of mathe-
matical models used in gear dynamics, published before 1987. Another review
is from Parey and Tandon [162]. They refer to the former review, who focussed
mainly on gears without defects. Parey and Tandon, on the other hand, shift
the focus to spur gear dynamic models which can include gear defects.

Özgüven and Houser [160] refer in their review to the large variation in objec-
tives in dynamic modelling of gear systems. They summarise these goals as an
analysis dedicated to one (or some) of the following interests: bending and/or
contact stresses in teeth; loads on supporting machine components (especially
on bearings); reliability and durability; natural frequencies, vibratory motion
and stability regions of the system; radiated noise; transmission efficiency; pit-
ting and scoring; whirling of rotors. In their review, Özgüven and Houser re-
mark that classifying the various mathematical models is not straightforward.
Whichever categorisation is used, there will always be some models that might
correspond to more than one group. Nevertheless, they consider the following
classification appropriate in their literature survey.

1. Simple dynamic factor models which aim at a simplified determination


of dynamic factors that can be used in gear root stress formulae. Parey
and Tandon defined the dynamic factor as the ratio of the maximum
dynamic tooth root stress to the maximum static tooth root stress. This
type of models are included in the DIN and AGMA gear standards [5,54]
and contain typically empirical or semi-empirical information.

2. Models with tooth compliance include only the tooth stiffness. Flexibil-
ities of all other drive train components are neglected.
4.3 Drive train modelling techniques 95

3. Models for gear dynamics include additionally also the flexibilities of


the other drive train components, often as a reduction of the tooth stiff-
ness along the line of contact.

4. Models for geared rotor dynamics consider furthermore the transverse


vibrations of the shafts supporting the gears, by including two perpen-
dicular translational DOFs per shaft. This allows prediction and investi-
gation of whirling of the shafts.

5. Models for torsional vibrations are all models dedicated to torsional vi-
bration problems in which the flexibility of the gear teeth is neglected.

Özgüven and Houser state furthermore that, despite the large variation in mod-
elling approaches, a good correlation between numerical and experimental re-
sults is reported for models from all groups. This indicates that even the rather
simple models are valuable, especially because they require typically less cal-
culation time and less implementation work. In addition, it is important to
note that a good agreement with experimental observations requires in the first
place a proper satisfaction of the basic assumptions in the models, which can
be very dedicated to specific applications.

Özgüven and Houser describe moreover different kinds of analyses found in


the literature.

1. Normal modes analyses are made when the main objective is to find the
system natural frequencies and mode shapes. Knowledge of these char-
acteristics allows to tune specific excitation frequencies in the system to
avoid drive train resonances. It gives furthermore valuable insight for
response analyses and for the assessment of natural frequency sensitivi-
ties.

2. A dynamic response calculation aims at predicting the drive train loads


for a defined excitation and implies usually a transient or a harmonic vi-
bration analysis. Various phenomena are often modelled as excitations,
such as gear errors, a tooth stiffness variation and even nonlinear effects
caused by loss of tooth contact or by friction between meshing teeth.

Only the models with tooth compliance (2), the models for gear dynamics
(3) and the models for geared rotor dynamics (4) are further considered as
relevant for the present work. These models include mathematical expressions
to describe the load transfer, the flexibility and the damping in the tooth contact
as well as in the bearings and the drive train components. The remainder of
this section discusses these items individually.
96 4. Detailed modelling of the drive train in a wind turbine

4.3.2.1 Tooth contact

The representation of the tooth contact forces is further called the gear mesh
model. Most drive train models are multibody systems and apply a spring
or a spring-damper representation for the gear mesh model, in accordance
with equation (4.26). Some exceptions, such as Howard and Wang [215],
Litvin [135] and Barone [11], use an FE model for the teeth in contact, which
requires a very dense mesh near the contact zone. This puts much more de-
mands on the computational time and modelling effort in such analyses. There-
fore, they are normally only used for numerical validation of equivalent, but
more simplified models or for dedicated analyses with respect to tooth stresses
and strains. Parker et al [165] describe the successful implementation of an FE
approach with an analytical description of the contact zone, without the need
for a fine mesh. This yields still the desired contact behaviour, but reduces the
computational demands. An important factor in this contact behaviour is the
so-called transmission error. An accurate representation of this error in a gear
model is important for the analysis of internal excitations in a gear system.
The next paragraph describes the transmission error extensively. A subsequent
paragraph discusses the classification of different gear mesh models.

A. Transmission error

The transmission error is defined as the difference between the actual and the
ideal position of the driver gear, where the latter position refers to a situation
determined by the gear ratio and a perfectly conjugate mesh action without
errors or deflections. This error is usually expressed as a linear displacement
along the line of contact, similar to the tooth deflection δ as introduced in
equation (4.26). Based on this equation, it is clear that a transmission error
can result in a load (variation). Another way of expressing the transmission
error is as a deviation of the relative angular position of two gear shafts. This
is popular in studies with purely torsional models, since these include only ro-
tational DOFs of the gears and torque loads (cfr. section 4.4). Furthermore,
distinction is made in the literature between a static and a dynamic transmis-
sion error, where the former one is only valid during very slow rotation. The
dynamic transmission error represents all deviations on top of the static error
during operation at higher speeds.

Various factors contribute to the transmission errors, such as tooth geometry


errors, elastic deformations and imperfect gear mounting. The second factor
includes the effects of imperfect load distribution at the gear mesh and fluc-
tuating mesh loads caused by external dynamic influences or a varying gear
mesh stiffness κgear . Numerous articles describe the individual effect of such
4.3 Drive train modelling techniques 97

factors on the transmission errors and, consequently, on the response spectra of


the gears. After all, these errors are considered as the prime source of vibration
in the system and, as a result, the main cause of gear noise [191]. Figure 4.7
shows a typical transmission error trace during one gear revolution. In gen-
eral, this consists of a fairly regular once-per-tooth pattern superimposed on
a once-per-revolution waveform. The former excitation is usually dominating
and its frequency is known as the gear mesh frequency. The amplitude of this
excitation depends on the magnitude of the transmission error, which can be
measured experimentally as described in [192, 193].
transmission error

0 gear revolution 1

Figure 4.7: Typical transmission error trace during one gear revolution, repro-
duced from [194].

A considerable contribution to the once-per-tooth variation of the transmis-


sion error comes from the variation of the stiffness value κgear during rotation.
This stiffness determines the deflection of the gear mesh, which is part of the
transmission error and, as a result, a variation results in a variation of the trans-
mission error. The once-per-tooth pattern in this variation may originate from
the number of teeth in contact that changes during a single mesh period. This
varies for a spur gear typically from a single tooth pair in contact to a double
tooth pair in contact, where the latter situation yields a higher gear mesh stiff-
ness. Wang and Howard analysed this mesh stiffness variation in detail [215]
and Kuang and Lin derived an analytical formulation for this fluctuation [118].
The mesh stiffness in general depends on many parameters such as the ap-
plied load, the material properties, the number of teeth in contact, the gear face
width and possible profile modifications [114]. In practice, only two values are
often used to represent the gear mesh stiffness variation of a spur gear. This
is considered as a sufficient approximation [110], when analysing the excita-
tion caused by this variation. Figure 4.8 shows this representation of the mesh
stiffness variation between a single tooth contact and a double tooth contact
for such a gear pair. This variation is in practice small compared to the average
stiffness [110] and is assumed to be negligible in an analysis without consid-
eration of the internal excitation.
98 4. Detailed modelling of the drive train in a wind turbine

κgear (t)
double tooth contact
max(κgear ) @
@

mean(κgear )

min(κgear )
single tooth contact

Figure 4.8: Approximation of the gear mesh stiffness variation for a spur gear
pair.

The excitation caused by the mesh stiffness variation is a parametric excita-


tion and, correspondingly, the situation where the gear mesh frequency equals
a natural frequency can lead to parametric instabilities. Such instabilities and
parametric excitation more generally can occur in various dynamic systems,
which has been studied extensively and is reviewed in publications from Ibrahim
et al [97, 98] and Nayfeh and Mook [153]. The particular occurrence in gear
systems can lead to large amplitude vibrations, as demonstrated by experi-
ments of Benton and Seireg [18, 19] and Kahraman and Blankenship [109].
Therefore, it is investigated thoroughly for different applications using divers
modelling approaches, and described accordingly by Amabili and Rivola [4],
Benton and Seireg [20], Kahraman and Blankenship et al [22, 108, 112, 113],
Parker et al [167] and Randall and Du et al [60, 61]. Numerical studies on
parametric instabilities for multi-mesh gear systems require more complicated
models. Kahraman et al [2,3,106,107] and Parker and Lin et al [133,164–166]
describe such analyses and indicate that experimental validation of these com-
plex models is difficult.

The consideration of the varying mesh stiffness in a gear model may occur by
including a fluctuating κgear (t) in equation (4.26), which gives:

Fbn (t) = δ · κgear (t) (4.27)

Since κgear (t) determines only a part of the transmission error, it can also be
included in a function which represents the overall transmission error. Such a
function can be denoted e(t) and is typically added to the gear deformation δ
4.3 Drive train modelling techniques 99

to represent the time-varying effect of various errors along the line of contact:

Fbn (t) = [ δ + e(t) ] · κgear (4.28)

The function e(t) can include furthermore for instance the effect of profile er-
rors, backlash or gear defects and is often considered as a forcing vector at
the gear mesh causing vibrations, which can lead to noise [12, 38, 39, 144].
The analyses of gear defects and their influence on the noise or vibration spec-
trum is highly relevant for condition monitoring of gearboxes. Various stud-
ies from among others Bartelmus [12], El Badaoui et al [65], Fernandez et
al [71], Howard et al [95, 103], Kuang and Lin [117], Li et al [127], Velex et
al [210, 211] describe the application of dedicated models for these analyses.
Next to the simulation of gear defects, a related domain of research exists with
the focus on the experimental investigation for condition monitoring purposes.
This includes mainly the development of particular hardware, the implemen-
tation of specific signal processing algorithms and the definition of assessment
criteria for the evaluation of gearbox conditions. Randall [176, 177] gives a
general overview of the state of the art in monitoring rotating machinery and
he and Antoni [7, 175] discuss furthermore the particular analysis of gear and
bearing faults. Various dedicated signal processing techniques are described by
Baydar [15], Cpadessus et al [40], El Badaoui [64, 66], Sung [205], Wei [216],
Lin et al [128], Meltzer et al [145–147] and Yuan et al [223, 224].

B. Classification of gear mesh models

Both the FE approach and the MBS approach are applied for gear mesh mod-
els. However, generally the latter approach is chosen, since it yields sufficient
insight with much faster calculation times. Within this approach, Kahraman
and Singh [113] distinguished between four different mathematical gear mod-
els:
1. linear time invariant (LTI) models
2. linear time-varying (LTV) models which include a varying mesh stiff-
ness and no backlash
3. LTV models which include backlash and a constant average mesh stiff-
ness
4. LTV models which include both backlash and mesh stiffness variations
It is clear that the LTV models focus on the effect of a varying transmission
error in the gear system and are therefore dedicated to response analyses with
respect to internal excitations.
100 4. Detailed modelling of the drive train in a wind turbine

The number of DOFs in a gear mesh model yields a second classification.


Parey and Tandon [162] describe a model with a spring-damper representation
which considers only the rotations of the gears around their symmetry axis.
This means that only one DOF of each gear is coupled to one DOF of the
other gear or, in other words, that the gear mesh model neglects the dynamic
coupling between transverse and torsional vibrations (i.e. 1 DOF/body). Al-
though this is a popular approach, it has been observed experimentally that the
neglected dynamic coupling can have a considerable influence on the system
behaviour.

Therefore Kahraman et al [111] stated, following the conclusions of Bagci and


Rajavenkateswaran [9], that a coupled torsional and flexural modal analysis is
the best procedure to find natural frequencies and corresponding mode shapes.
They introduced a gear mesh model that couples not only the rotation, but also
the translation along the contact line of two gears (i.e. 2 DOFs/body).

Afterwards, Kahraman [105] extended this two-DOF coupling by including


two more DOFs, namely the axial translation and one tilting rotation of the
gears (i.e. 4 DOFs/body). This allows assessing the effect of helical teeth on
the gear system behaviour.

The extension towards a coupling of all six DOFs of both gears is presented
in the work of Kahraman [106] on multi-mesh helical gear trains (i.e. 6 DOFs
/body).

A special case of multi-mesh gear trains are planetary gear systems [107]. Lin
and Parker [129] presented an analytical model to investigate the natural fre-
quencies and corresponding mode shapes of such a system. Their gear mesh
model includes only in-plane DOFs of the gears, since their analysis is limited
to spur planetary gear stages(i.e. 3 DOFs/body). For an example system, they
demonstrated furthermore that the calculated numerical results are characteris-
tics of general planetary gears. They distinguished different types of vibration
modes and classified them as rotational, translational and planet modes. This
classification seems to be valid more generally for planetary gears with un-
equally spaced planets, satisfying a specific condition [131]. Moreover, the
analytical approach of Lin and Parker allows to investigate the sensitivity of
natural frequencies and mode shapes in a more pragmatic way [130,132]. [166]
and [133] describe their analyses of parametric instabilities in multi-mesh gear
systems and in [164], Parker demonstrates the effectiveness of planet phasing
to suppress planetary gear vibrations and, consequently, reduce possible gear
noise problems [163].
4.3 Drive train modelling techniques 101

Section 4.5 describes the implementation of a gear mesh model with six DOFs
per body, similar to the work of Kahraman. This model is furthermore suited
for the analysis of helical planetary gear systems, which is an extension of Lin
and Parker’s model for a spur planetary gear stage.

4.3.2.2 Bearings

The most popular bearing type in drive trains is a roller bearing. These bear-
ings have an inner and and outer ring with a series of rollers between them.
Figure 4.9 shows four different roller bearings, which are popular in a wind
turbine drive train. These bearings normally have a certain clearance. This
means that no rollers are in contact during no load conditions, which corre-
sponds to a zero stiffness value. Under load conditions, a number of rollers are
in contact with the inner and outer ring. These rollers transfer the loads and
represent a certain flexibility and damping. Both FE models and multibody
models exist to describe their physical behaviour.

The detailed FE approach is a rarely chosen option, because of its high com-
putational demands. It is only applied for detailed stress calculations in indi-
vidual rollers or in the inner and outer ring. Such calculations require a very
dense mesh in combination with the use of sufficient contact elements. The
non-linearity in the contact analysis can further complicate the calculations
and increase the calculation time.

Therefore, the application of a simplified spring or spring-damper model is


more popular. Lin and Parker [129], Vedmar and Andersson [208] and Vinayak
et al [213] describe this approach and Sopanen and Mikkola [200,201] discuss
it in more detail. The spring-damper model represents both the deformation
and damping characteristics of a bearing. The stiffness properties depend a.o.
on the number of rollers in contact, which can vary with the applied load. The
relation between the bearing deflection and the applied load, is consequently
a nonlinear behaviour as shown in figure 4.10 for a deep groove ball bearing
with a diameter of 80 mm. Palmgren [161] and Brandlein et al [34] describe
the bearing properties in more detail.

Although the bearing stiffness varies for different loads, it is often linearised
as a constant stiffness to keep the resulting model linear. This occurs for a
given load set defined in a specific operating point and limits the validity of
the corresponding analyses results to a working range around this point. Espe-
cially during load reversals, this model is not longer adequate because of the
clearance in a bearing.
BALL

BALL
102 4. Detailed modelling of the drive train in a wind turbine

BEARINGS
There are five basic types of anti-friction bearings: tapered, needle, ball, spherical
and cylindrical. Each is named for the type of rolling element it employs.
SPHERICAL
ROLLERS
SPHERICAL
(a) Spherical roller bearing
ROLLERS

BEARINGS
There are five basic types of anti-friction bearings: tapered, needle, ball, spherical
and cylindrical. Each is named for the type of rolling element it employs.
TAPERED
CYLINDRICAL
ROLLERS
(b) Cylindrical roller bearing
ROLLERS
CYLINDRICAL
ROLLERS

NEEDLE
ROLLERS
TAPERED
(c) Taper roller bearing
ROLLERS

NEEDLE
ROLLERS
BALL
(d) Deep-groove ball bearing

Figure 4.9: Different types of roller bearings used in the drive train design of
a wind turbine.

4.3.2.3 Component flexibilities

Various alternatives exist also for the representation


BALL of the component flexi-
bilities in combination with their inertias in aSPHERICAL
mathematical model. The most
ROLLERS
straightforward approach is considering a component as a rigid body and ne-
glect its flexibility. However, this is only valid when the contribution of this
component to the dynamics of interest is negligible. If not, the flexibility has
to be taken into account as a discrete stiffness in an MBS formulation or as a
distributed property in an FE model or a flexible multibody model.

SPHERICAL
CYLINDRICAL
ROLLERS
ROLLERS
4.3 Drive train modelling techniques 103

Deflection [µm]

Radial load [kN]

(a) Radial deflection


Stiffness [kN/mm]

Radial load [kN]

(b) Radial bearing stiffness

Figure 4.10: Radial stiffness of a deep groove ball bearing with a diameter of
80 mm, based on rules of thumb published in [34].

This dissertation describes the implementation based on an MBS formulation


and on a flexible MBS approach. This latter approach includes furthermore
directly the components’ inertia. Ho [94], Spanos and Tsuha [202], Cyril et
al [49], El Absy and Shabana [62, 63], De Fonseca [50], Schwab [188] and
Braccesi et al [32, 33] describe this technique. However, publications, which
describe the application of this technique for the modelling of drive trains and
gear systems more particularly, are still rare. One exception is an article of
Lundvall et al [139], who applied the flexible MBS methodology for the mod-
elling of the frictional contact in a spur gear. They represented the gears as
reduced FE models, but the shafts and the bearings as linear discrete springs.
104 4. Detailed modelling of the drive train in a wind turbine

This dissertation goes further by representing all shafts and gears in a gearbox
as reduced FE models.

As described in section 3.3, it is already common practice in the traditional


wind turbine design codes to consider the flexibility of the blades and the
tower through a modal formulation. Zhao et al [225], Lee et al [125] and
Ahlström [1] focus on this implementation, independent of any existing design
code. Baumjohann et al [14] apply this approach using a highly accurate model
for the wind turbine blades, aiming at a prediction of stresses and strains inside
these components. However, as described in section 3.4.3, the implementation
of a flexible multibody model to investigate the drive train in a wind turbine
in more detail is rarely found in the literature. This dissertation presents the
combination of an MBS approach with six DOFs per body in the gear mesh
and bearing models with the use of flexible bodies.

4.3.2.4 Damping in the drive train


The actual representation in a model of the physical phenomena which intro-
duce damping in a drive train is not straightforward. Especially the determina-
tion of absolute values to define the damping behaviour is complicated. These
values are represented by the matrix [C] in the general formulation of the equa-
tions of motion (3.1). However, the use of this absolute damping matrix is only
rarely found in the literature. Generally, various relative damping values are
used to describe particular energy losses in the system, such as e.g. due to ma-
terial damping or friction in the gears and the bearings. A widespread method
to include relative damping is as a modal damping factor ζ, which corresponds
to the damping of the system’s response at a particular eigenfrequency.

Gerber [76] introduces an empirical equation for the modal damping factor,
which represents the damping between two gears in contact:

ζ = 2.2 · 10−4 · (as − 23)0.55 · (η + 39)0.27 · (vt − 5)0.53 (4.29)

as distance between axes of two gears [mm]


η viscosity of the lubrication fluid [mPa · s]
vt peripheral speed at the pitch circle [ ms ]

This expression uses physical parameters to calculate the damping and is de-
rived from a large series of experiments. The application of this empirical for-
mulation requires furthermore the knowledge of the reduced mass of the gear
system. Its validity is limited to the ranges 15 ms < vt < 50 ms and 50 mm <
as < 250mm.
4.3 Drive train modelling techniques 105

Equation (4.26) describes the contact force between two meshing gears as the
product of the deformation along the line of contact δ and a given gear mesh
stiffness κgear . An equivalent formulation for the damping force between two
meshing gears requires the knowledge of an absolute gear mesh damping coef-
ficient cgear . The product of cgear and δ̇ yields consequently the desired damp-
ing force:
F˙bn = δ̇ · cgear (4.30)
The reference [3] from Al Shyyab and Kahraman is one of the rare publica-
tions found in the literature with a value for cgear . For a mean mesh stiffness
κgear = 5 · 108 Nm , the corresponding mesh damping equals cgear = 2721 Ns m.
When κgear is only half of the defined value, likewise the mesh damping cgear
is taken only half.

The present work does not further discuss specific issues related to damp-
ing in the drive train, nor does it include any quantified damping values for
the developed models. Since damping in the gears, in the bearings and in
the components’ material is relatively small, its influence on the calculation
of eigenmodes and eigenfrequencies is negligible. For the calculation of fre-
quency response functions (FRFs), damping is considered only qualitatively.
The damping in a wind turbine introduced by the generator characteristic and
by the aerodynamic interaction between the rotor and the wind is not negligi-
ble. However, the determination of their influence is outside the scope of this
work.

4.3.3 Summary and conclusions


This dissertation presents a study of applying the MBS methodology for the
analysis of a drive train in a wind turbine, to guarantee the structural integrity
of all drive train components with a higher reliability (cfr. section 3.4.2). This
limits the focus on the analysis of drive train loads, rather than on noise calcu-
lations or vibration monitoring.

The transmission error in a gear system can be considered as an excitation,


which determines the radiated noise level and the load level during parametric
instabilities. These instabilities occur when a gear mesh frequency equals an
internal eigenfrequency. This resonance behaviour should be avoided, which
requires a prediction of the natural frequencies in the drive train and the cor-
responding mode shapes. In this prediction, the influence of the transmission
error is negligible. Therefore, this variation of the gear mesh deformation is
not implemented in the gear mesh models in the present work. Its influence is
only analysed by applying the transmission error as an excitation in a response
calculation. As a result, the gear mesh models are LTI models, with a constant
106 4. Detailed modelling of the drive train in a wind turbine

gear mesh stiffness. In addition, all bearing stiffness values are linearised and
kept constant.

Three modelling techniques with increased levels of complexity are analysed


subsequently in the present study. Every addition to the model leads to spe-
cific additional information about the internal dynamics of the drive train, but
makes the modelling and simulation more complex. Therefore, depending on
the aim of the analysis, a designer has to decide how much detail is required
in his models. The presentation of the step by step increase in complexity en-
ables the drive train designer and in particular the gearbox designer, to get an
overview of the advantages and the limitations of the different levels of mod-
elling. Each level can be used as a separate tool in specific design phases to
estimate the significance of dynamic loads.

Section 4.4 describes the simplest level of modelling, where exactly one DOF
per drive train component is used to simulate only torsional vibrations in the
drive train. Their implementation is based on the state-of-the-art in modelling
torsional models. Flexibility is assumed to be concentrated in shafts and gear
teeth. Bearings are considered to be rigid in radial and axial directions. These
models are called purely torsional multibody models. The analysis of such
models gives insight in their usefulness and their limitations.

Since a coupling of torsional and flexural motion in a model yields a better pre-
diction of natural frequencies and corresponding mode shapes in a gear system,
section 4.5 presents more elaborate models. Here, all individual drive train
components have six DOFs and these models are further called rigid multi-
body models. Linear springs represent the gear and bearing flexibilities, as
presented by Kahraman for a helical parallel gear system [105]. These tech-
niques are further combined with the work of Lin and Parker on planetary spur
gears [129]. The performed synthesis makes it possible to analyse additionally
a single-stage helical planetary gear set. Finally, this approach yields further-
more three-dimensional, generic models to simulate the dynamics of complete
gearboxes integrated in a wind turbine drive train.

The most detailed level of modelling is the flexible MBS formulation. Sec-
tion 4.6 discusses this further extension in which the drive train components
are represented by reduced FE models instead of rigid bodies. This approach
is necessary to gain insight in local stresses in drive train components. It yields
furthermore a more accurate and realistic representation of the inertia and flex-
ibility properties of the components. The combination of reduced FE models
with the gear mesh and bearing models derived for the rigid multibody mod-
els, is a further extension of the state-of-the-art modelling techniques in gear
dynamics.
4.4 Purely torsional multibody models 107

4.4 Purely torsional multibody models


A first approach in modelling the internal dynamics of a drive train is only
focusing on torsional vibrations. In a torsional multibody model, all bodies
have exactly one DOF, namely the rotation around their axis of symmetry. The
five other DOFs are fixed; thus, they can be left out of the equations of mo-
tion and the coupling of two bodies involves only two DOFs (θ1 ,θ2 ). Only
the torsional inertia is needed as input for the rigid bodies; furthermore, their
torsional stiffness and the gear mesh stiffnesses are the only flexibilities repre-
sented in a direct way. Taking into account the influence of other flexibilities is
only possible indirectly as a reduction of the former stiffness values. However,
the calculation of such reduction factors is not straightforward and complicates
the modelling. Torsional models can be used for the dynamic analysis of the
torque in the drive train; the other force components can only be derived by fur-
ther processing the torque. The torsional flexibility of a shaft (Kshaft ) between
two bodies is included in the equations of torque as shown in equation (4.31).
Material damping is neglected in this model.

T1 = −T2 = Kshaft · (θ2 − θ1 ) (4.31)




The tooth contact force Fbn between two wheels acts in the plane of action
along the contact line (normal to the tooth surface at the point of contact),
as described in section 4.2. Its representation in a torsional model is a linear
spring which couples the two DOFs of the wheels and includes the transmis-
sion ratio between them, as described in figure 4.11. The bending deformation
of the teeth as a function of these two DOFs equals:

δ = (rb1
0
θ1 − rb2
0
θ2 ) · cos β0b , (4.32)
0 and r 0 the base circle radii of the respective wheels. The transverse
with rb1 b2


component of Fbn is written as Fbt = Fbn · cos β0b , based on equation (4.13).
This force vector works on both gears and causes a higher torque on the larger
gear. The direction of this force is such that the resulting torque on the driving
wheel is always opposite to the input torque. Furthermore, based on equa-
tion (4.26), its magnitude equals1 :

Fbt = δ · κgear · cos β0b (4.33)


= 0
(rb1 θ1 − rb2
0
θ2 ) · κgear · cos2 β0b

1 Note that when the tooth stiffness value would be defined in the transversal plane, such

as in DIN 3990 [54], the influence of the helix angle would be included in this value. Here,
β0b is explicitly included in the equations of the deformation and κgear can be kept constant for
varying helix angles.
108 4. Detailed modelling of the drive train in a wind turbine

+ θ2
- 0
rb2



Fbt T2
κgear
θ1 angle of rotation of pinion Td +
θ2 angle of rotation of gear wheel
0 0
rb1
rb1 base circle radius of pinion
0
rb2 base circle radius of gear wheel θ1
T1
J1 pinion inertia
J2 gear wheel inertia ?
igear 0 /r 0 )
transmission ratio (rb2 b1
Td positive driving torque κgear · (rb1
0 cos β0 )2
b
applied to the pinion

− OC
Fbt tooth contact force J1 J2
in the transversal plane igear
κgear gear mesh stiffness
T1 reaction torque on pinion
T2 reaction torque on gear wheel

T1 = −Td = −Fbt · rb1 0


= −κgear · cos β0b (rb1
0
θ1 − rb2
0
θ2 ) ·rb1
0
cos β0b
| {z }
(b) (c) (a)
z }| { z }| {
= − κgear · (rb1
0
cos β0b )2 · (θ1 − igear · θ2 ) (4.34)

1
T2 = −T1 · igear = − κgear · (rb2
0
cos β0b )2 · (θ2 − · θ1 ) (4.35)
| {z } i gear
| {z }
(d) (e)

(a) deformation along the line of contact (> 0)


(b),(c) torsional stiffness, torsional deformation - referred to the pinion
(d),(e) torsional stiffness, torsional deformation - referred to the gear wheel

Figure 4.11: A torsional model for the tooth contact force between a driving
pinion and a driven gear wheel. Td is a positive driving torque applied to
the pinion causing a negative reaction torque T1 on the pinion and a positive
reaction torque T2 on the gear wheel.
4.5 Rigid multibody models with discrete flexible elements 109

In the gear contact model, the time-varying components due to the varying
transmission error are not considered. Furthermore, no damping or friction
forces are included. From a physical understanding, it is clear that the present
spring will only work under compression. To ensure that this limitation will
not be exceeded during simulation, the following extra assumption is made
here. No contact loss between the gears will occur, something that could hap-
pen for a system with backlash when the dynamic mesh force becomes larger
than the static force transmitted. This assumption is considered to be valid for
heavily to moderately loaded gears [105] not running near resonance.

θ1 and θ2 in figure 4.11 are defined as the rotations of the pinion and the gear
wheel in their respective reference frame. For a parallel gear stage, these ref-
erence frames remain fixed to the gearbox housing. However, the same for-
mulation is valid when the reference frame of a wheel follows the rotation of
a component, which implies a kinematic coupling between the wheel and this
component. This makes it also applicable for a planetary gear stage where the
reference frame of a planet follows the rotation of the planet carrier. Thus,
by keeping the gear contact formulation independent from the definition of
the reference frame, it can be used as a generic module for all possible gear se-
tups. This independency is possible in the multibody software package DADS,
since coordinate systems can be created and referenced freely. Note that the
base circle radius should be taken negative, when using the formulation for a
wheel with internal teeth. Chapter 5 describes the application of the present
formulations for a parallel and a planetary gear stage.

4.5 Rigid multibody models with discrete flexible ele-


ments
The extension of a torsional model to a rigid multibody model adds the possi-
bility to investigate the influence of bearing flexibilities on the internal dynam-
ics of the drive train, without the complicated calculation of the stiffness reduc-
tion factors during the model implementation as described in section 4.4. Fur-
thermore, the analysis can also yield insight in dynamic bearing loads, which
are coupled with the displacements of the bodies in their bearings. All drive
train components are still treated as rigid bodies, but now have a full set of
six DOFs instead of only one. This implies that the linkages in the multibody
model, representing the bearing and tooth flexibilities, now need to couple
twelve DOFs.

Linear springs are used here to model the bearing and gear mesh stiffnesses.
An individual formulation of these models in section 4.5.1 and 4.5.2 yields
110 4. Detailed modelling of the drive train in a wind turbine

two three-dimensional plug-in components, ready to use in a generic mod-


elling approach for the drive train. This means both models are suitable for a
rather simple parallel gear stage as well as for a more complex helical plane-
tary stage with any number and any positioning of the planets. Furthermore,
the application of these models is not limited to single gear stages, but also
usable for the analysis of complete gearboxes.

4.5.1 Modelling of the bearing flexibility

The six DOFs of the rigid bodies need appropriate constraints in the bearing
model. This model is represented by a linear spring and implemented as a
6 × 6 stiffness matrix (Kb ) defined in the XYZ coordinate system as shown
in figure 4.12. Damping is neglected and all bearings are assumed to have
an axisymmetric behaviour without coupling between the individual DOFs.
Therefore, all off-diagonal terms are zero and both the radial and tilt stiffnesses
are equal. Practically, the bearing component in a multibody model connects
the XYZ coordinate system, fixed to a certain body, with the X’Y’Z’ system
fixed to this body’s reference frame. This reference frame can be for instance
the fixed housing, however, it can also be the planet carrier, e.g. for the planet
bearings. Since the present models are LTI, the bearing stiffness values need
to be determined for a specific operating point and have a limited validity with
respect to the load range.

4.5.2 Modelling of the tooth contact forces

Similar to the purely torsional equivalent of the gear mesh model, the tooth


contact force Fbn is represented here by a linear spring, but now this spring
involves a coupling between twelve DOFs instead of only two. Figure 4.13
shows the representation of this force vector for the general case of two helical
gears in contact. Td is the magnitude of the driving torque, which acts on
gear1 . The operating pitch circle diameters are d10 and d20 and, in addition, the
corresponding base circle radii:

0
rb1 = d10 /2 · cos αt0
0
rb2 = d20 /2 · cos αt0

The assumptions postulated for the gear mesh model in section 4.4 are still
valid. For the sake of completeness, they are repeated in the list of assumptions
that are made here:
4.5 Rigid multibody models with discrete flexible elements 111



Fb = Kb · →

q (4.36)

with
 
kradial 0 ··· ··· ··· 0
 .. ..

 kradial . .

 .. ..

=  kaxial . .
Kb ,
 
 .. ..
.


 ktilt 
.
 ktilt 0 
0



q = [x y z ρX ρY θ]T ,

where x, y, z, ρX , ρY and θ are the projections in the XYZ system of the


translations and rotations from the position and orientation of the gear in its
reference frame X’Y’Z’.

Figure 4.12: Schematic representation of a bearing model: a linear spring con-


nects the XYZ coordinate system fixed to the body, with the X’Y’Z’ system
fixed to this body’s reference frame. The spring is modelled by the stiffness


matrix Kb and Fb is the force working on the gear in the XYZ system.
112 4. Detailed modelling of the drive train in a wind turbine

Y’1 Gear1 Gear1 is the driving wheel (Td < 0)


0 : base circle radius
rb1

Gear2 is the driven wheel


0 : base circle radius
rb2
Z’1 β0b κgear
Td X’1 κgear is the gear mesh stiffness
as defined in section 4.2
Gear2
Y’2

Z’2

X’2

Figure 4.13: Representation of the tooth contact force vector as a linear spring
for two helical gears in contact.

1. The gear mesh model is an LTI model. A varying transmission error and
corresponding excitation is not considered and, therefore, no phasing
relationships between gear meshes are included, nor a variable stiffness
caused by a fluctuation in the number of tooth pairs in contact. The va-
lidity of these assumptions for the present linear analyses can be justified
as in section 4.4.
2. Sliding of teeth in contact and corresponding friction forces are ne-
glected as well as any other possible damping in the system.
3. Occurrence of tooth separation is considered to be non-existent and, con-
sequently, the modelling of gear backlash is not included. This implies
that the spring is always under compression.
4. Coriolis accelerations of gears that are rotating and simultaneously trans-
lating (e.g. planets on their carrier) are neglected and all gyroscopic ef-
fects as described by Lin and Parker [129] are excluded. These assump-
tions are valid for wind turbine applications, since planetary gear stages
in wind turbines are only rarely used as high speed stages.

A further elaboration of the tooth contact force vector representation in fig-


ure 4.13 towards an actual formulation of forces and moments on both gears,
is based on the approach shown in figure 4.14.
4.5 Rigid multibody models with discrete flexible elements 113

Y1 ’ Y2 ’ X2
u1
u2
Y1 Y2
0
rb1 αt0
ψ2 ’< 0
ψ1 ’< 0 X1 ’ αt0 X2 ’
0
rb2

X1

β0b
X1 ’ X2 ’

Z1 ’ Z2 ’

Figure 4.14: Modelling of the tooth contact forces for two helical gears in
contact (β0b > 0).

• Coordinate systems X1 ’Y1 ’Z1 ’ and X2 ’Y2 ’Z2 ’ are oriented with X’ along
the centreline pointing from gear1 to gear2 ; Z’ is lying along the axis of
rotation. These coordinate systems remain fixed to the reference frames
of the respective wheels (as introduced in section 4.4 and figure 4.12).

• X1 Y1 Z1 and X2 Y2 Z2 are fixed to the respective gears and in their start-


ing position, they coincide with the corresponding X’Y’Z’.

• αt0 is the actual pressure angle of the gear mesh in the transversal plane.
It is defined as the angle measured from the centreline towards the nor-
mal on the contact line in the corresponding X’Y’Z’, as introduced in
figure 4.5. The sign of this angle changes when the driving direction of
the system changes.

• ψ01 and ψ02 are the angles measured respectively from X1 ’ to X1 and from
X2 ’ to X2 along the corresponding Z’: ψ1 = αt0 − ψ01 and ψ2 = αt0 − ψ02 .

• β0b is the helix angle which is positive when the teeth of gear1 are turned
“left” from a reference position where β0b = 0; β0b > 0 in figures 4.13
and 4.14.
114 4. Detailed modelling of the drive train in a wind turbine

The compression of the linear spring (δ) represents the bending deformation
of the teeth. This deformation is a function of the vectors →

q1 and →

q2 , as in-
troduced in figure 4.12. Since the spring works always under compression, δ
should be positive.

δ = (x1 · sin ψ1 − x2 · sin ψ2 − y1 · cos ψ1 + y2 · cos ψ2 − u1 − u2 ) (4.37)


· sign(αt0 ) · cos β0b
−(z1 − z2 + wX1 · sin ψ1 + wX2 · sin ψ2 − wY 1 · cos ψ1 − wY 2 · cos ψ2 )
· sign(αt0 ) · sin β0b

with

[x1 y1 z1 wX1 wY 1 u1 ]T = diag(1, 1, 1, r0b1 , r0b1 , r0b1 ) · →



q1
[x2 y2 z2 wX2 wY 2 u2 ]T 0 0
= diag(1, 1, 1, r , r , r ) · q0 →

b2 b2 b2 2

The stiffness value of the linear spring is κgear , which is the ratio of the contact
force on a tooth over the resulting displacement of the contact point, as defined


in equation (4.26). The spring force represents the tooth contact force Fbn and
causes forces and moments on the gears, which can be projected in the XYZ
coordinate systems and, thus, written as:

FX1 = −δ · κgear · sin ψ1 · cos β0b · sign(αt0 )


FY 1 = δ · κgear · cos ψ1 · cos β0b · sign(αt0 )
FZ1 = δ · κgear · sin β0b · sign(αt0 )
TX1 = δ · κgear · rb1
0
· sin ψ1 · sin β0b · sign(αt0 )
TY 1 = −δ · κgear · rb1
0
· cos ψ1 · sin β0b · sign(αt0 )
TZ1 = δ · κgear · rb1
0
· cos β0b · sign(αt0 )

FX2 = δ · κgear · sin ψ2 · cos β0b · sign(αt0 )


FY 2 = −δ · κgear · cos ψ2 · cos β0b · sign(αt0 )
FZ2 = −δ · κgear · sin β0b · sign(αt0 )
TX2 = δ · κgear · rb2
0
· sin ψ2 · sin β0b · sign(αt0 )
TY 2 = −δ · κgear · rb2
0
· cos ψ2 · sin β0b · sign(αt0 )
TZ2 = δ · κgear · rb2
0
· cos β0b · sign(αt0 )
By writing the force components and the compression δ in the XYZ systems,
the gear contact formulation is a generic module. Its implementation in DADS
is a user-defined subroutine, which can be used to couple any two gears in
all possible gear setups, keeping in mind that a wheel with internal teeth is
4.5 Rigid multibody models with discrete flexible elements 115

described with a negative radius. The formulation of the gear forces in matrix
form yields: " →
− #    → − 
F1 k11 k12 q1
− = κgear ·
→ · → − (4.38)
F2 k21 k22 q2

− →

where F1 and F2 are written as:


F1 = [FX1 FY 1 FZ1 TX1 TY 1 TZ1 ]T


F2 = [FX2 FY 2 FZ2 TX2 TY 2 TZ2 ]T

The sub-matrices k11 , k12 , k21 and k22 all have a similar structure. Therefore,
only k11 is given here. Appendix D describes the details for all sub-matrices.

 
k11, A k11, B
k11 = (4.39)
k11,C k11, D
with

−c2 β s2 ψ1 c2 β0b cψ1 sψ1 cβ0b sβ0b sψ1


 

k11, A =(∗)  c2 β0 cψ1 sψ1 −c2 β0 c2 ψ1 −cβ0 sβ0 cψ1 


b b b b
cβ0b sβ0b sψ1 −cβ0b sβ0b cψ1 −s2 β0b

cβ0b sβ0b s2 ψ1 −cβ0b sβ0b cψ1 sψ1 c2 β0b sψ1


 

k11, B =  −cβ0b sβ0b cψ1 sψ1 cβ0b sβ0b c2 ψ1 −c2 β0b cψ1 
−s βb sψ1
2 0 s βb cψ1
2 0 −cβ0b sβ0b

0 cβ0 sβ0 s2 ψ
1 −rb1 s βb sψ1
0 cβ0 sβ0 cψ sψ 0 2 0
 
rb1 b b 1 −rb1 b b 1
k11,C 0 cβ0 sβ0 cψ sψ
=  −rb1 0 cβ0 sβ0 c2 ψ
rb1 0 s2 β0 cψ
rb1
b b 1 1 b b 1 b 1 
rb1 c βb sψ1
0 2 0 −rb1 c βb cψ1
0 2 0 0 0 0
−rb1 cβb sβb

0 s2 β0 s2 ψ 0 s2 β0 cψ sψ 0 0 0
 
−rb1 b 1 rb1 b 1 1 −rb1 cβb sβb sψ1
k11, D 0 s2 β0 cψ sψ
=  rb1 −rb10 s2 β0 c2 ψ 0 cβ0 sβ0 cψ
rb1
b 1 1 b 1 b b 1 
0 0 0 0 0
−rb1 cβb sβb sψ1 rb1 cβb sβb cψ1 0 −rb1 c βb
0 2 0

(*) cβ0b = cos β0b ; sβ0b = sin β0b ; cψ = cos ψ; sψ = sin ψ.

When β0b 6= 0, it is possible to have no zero components in the ki j -matrices. At


that moment, all DOFs of both gears are coupled to each other. Chapter 5 de-
scribes an example of the application of the present formulations for a parallel
and a planetary gear stage. In addition, chapter 6 discusses an example of a
complete drive train in a wind turbine.
116 4. Detailed modelling of the drive train in a wind turbine

4.6 Flexible multibody models


Typically, multibody models consist of rigid bodies which are linked by joints
and stiffness elements. The stiffness values can include an equivalent discrete
stiffness for the flexibility of the individual components. However, the reduc-
tion to an equivalent stiffness and the discretisation method, complicate the
modelling, especially for more complex systems. As a result, geometrically
complex bodies are in practice often assumed to be rigid and no flexibility is
further taken into account. This assumption implies that there is no relative
displacement between the particles within such a body. However, this sim-
plification may differ considerably from reality. Considering the component’s
flexibility as a property of the body leads to a more realistic understanding of
the models. Firstly, it may give further insight in the role of this flexibility
in the overall dynamic behaviour. Estimating this influence for non standard,
geometrically complex and rather flexible parts used today in wind turbines
is barely possible with the traditional multibody formulation. Moreover, this
approach is the only alternative, which yields the desired insight in the local
deformations and corresponding stresses in drive train components.

Therefore, the extension towards a flexible multibody formulation is presented


here. This technique is no straightforward adaptation of the traditional method,
since it implies the inclusion of additional generalised DOFs to represent the
deformations of an individual body. These DOFs are introduced as component
modes, which are calculated in an FE analysis and subsequently synthesised
into a global set of modes for one body. The component mode synthesis (CMS)
technique is the general term for the reduction and coupling of FE models of
individual substructures. The first section below introduces the fundamentals
of this technique and the next section discusses its application in combination
with the traditional rigid multibody approach.

4.6.1 Component mode synthesis


The CMS technique originates from the need to reduce the number of DOFs in
the FE models of individual substructures, which form in combination an en-
tire structure. Typically, the individual FE models can have a large number of
DOFs, in the order of magnitude of 10,000 up to 1,000,000. The correspond-
ing demanding requirements for a dynamic analysis with respect to computer
resources and calculation time indicate the main driver behind the model re-
ductions, namely the necessity to accelerate and optimise the solution process.
Another reason for using CMS is the fact that the modelling and the analysis of
an entire structure can be split up partially into separate tasks, each focussing
on a different substructure. When different people or organisations design dis-
4.6 Flexible multibody models 117

tinct parts of an entire structure, it may be difficult to assemble an FE model


of the entire structure in a timely manner. The CMS simplifies this procedure
of concurrent engineering, since the resulting model of the entire structure is a
coupling of the individual substructure models, which are reduced to a set of
component modes tuned for the desired analysis. The set of component modes
is a smaller set of DOFs, e.g. in the order of magnitude of 1 up to 100.

For the reasons above, methods have been developed to subdivide entire struc-
tures into components or substructures and combine models of these individ-
ual parts in an approximate mathematical model of the full structural system2 .
Common names of these methods are methods of CMS or methods of sub-
structuring. Hurty [96] and Craig [46] performed during the late 1960’s pio-
neering work in the development of these model reduction techniques and a
comprehensive overview of various methods is given by Craig [43–45, 48] and
Spanos and Tsuha [202]. The remainder of this section overviews briefly the
concept of CMS based on these publications.

The objective of CMS is to calculate the dynamic behaviour of an entire struc-


ture, starting from the knowledge of the dynamic behaviour of its substruc-
tures. Numerical models of reduced sizes are used to describe the substruc-
tures. The general principle of sub-structuring consists of approximating the
dynamic behaviour of a substructure by a linear combination of qs vectors that
form a Ritz basis. This implies a model reduction from Ns physical DOFs, rep-
resented by the set ys , to a set of qs generalised coordinates cs . This reduction
is performed and described by the coordinate transformation:
{ys } Ns ×1 = [Ts ] Ns ×qs · {cs } qs ×1 , (4.40)
where Ts is a matrix of preselected component modes (the subscript s refers
to the sth sub-structure). Note that some of the generalised coordinates may
correspond to physical coordinates. Others correspond to non-physical modal
coordinates. The various sub-structuring methods differ in the choice of the
component modes (1), i.e. Ts , and the technique for assembling the substruc-
tures (2), which are the two basic steps in a CMS. The respective differences
in methods and techniques are further elaborated below.

4.6.1.1 Selection of component mode sets


The determination of an appropriate set of component modes for a particular
substructure starts with dividing its physical coordinates ys into a set of N j
interface coordinates y j 3 and a set of Ni interior coordinates yi . The interface
2 Another popular name for substructures in CMS is super-elements.
3 The character j denotes juncture, which is synonymous with interface.
118 4. Detailed modelling of the drive train in a wind turbine

coordinates are those coordinates where substructures are joined together and
the corresponding nodes are the interface nodes, in contrast with the interior
nodes. The coordinates of the displacement vector ys are accordingly:
 
yj
{ys } = (4.41)
yi

This permits furthermore to write the equations of motion for a component in


a partitioned form, as demonstrated here for an undamped substructure:

Ms · {y¨s } + Ks · {ys } = { fs } (4.42)


         
M j j M ji y¨j K j j K ji yj fj
· + · = (4.43)
Mi j Mii ÿi Ki j Kii yi fi
Using equation (4.40) in the equation of motion (4.42) yields:

Ms · Ts · {c¨s } + Ks · Ts · {cs } = { fs } (4.44)

For normalisation purposes, as explained below, equation (4.44) is multiplied


by the transpose of Ts :

TsT · Ms · Ts · {c¨s } + TsT · Ks · Ts · {cs } = TsT · { fs }


M̂s · {c¨s } + K̂s · {cs } = TsT · { fs } (4.45)

where M̂s = TsT · Ms · Ts is the reduced mass matrix and K̂s = TsT · Ks · Ts the
reduced stiffness matrix.

The derivation of the transformation matrix Ts is based on the partitioning into


interface and interior nodes and may lead to a combination of various sets of
component modes. In general, normal modes are used to represent the natural
vibration of the substructure and static modes are used to account for localised
loading and deformation caused by coupling the component at its interface
nodes. These latter coupling effects are already considered in the reduction of
a component by applying specific boundary conditions to the interface DOFs,
such as:

• free-interface, fixed-interface, hybrid-interface or loaded-interface con-


ditions for the calculation of normal modes

• unit displacements at interface DOFs for the calculation of static con-


straint modes

• unit loads at interface DOFs for the calculation of static attachment


modes
4.6 Flexible multibody models 119

These boundary conditions and the corresponding different types of normal


and static modes are further introduced briefly. The appropriateness of the
various types in the reduction depends on the type of application and analysis.
The final choice of the flexible component modes for a particular substructure
is a critical step in the model reduction and requires modelling experience and
engineering judgement. Only those modes, which are relevant for use in the
DADS flexible multibody models, are further discussed in section 4.6.1.2.

1. Normal modes.
The normal modes are the mode shapes of the substructure, which are
obtained using the eigenvalue problem of the form:
(Ks − ω2 Ms )Φ = 0 (4.46)
including a specific boundary condition applied to the interface DOFs.
Three types of boundary condition at the interface are possible, which
leads to fixed-interface normal modes, free-interface normal modes or
hybrid-interface normal modes depending on whether all, none or part
of the interface coordinates are restrained. In addition, some synthesis
methods [17] employ loaded-interface normal modes. This implies a
modification of the mass and stiffness properties of the interfaces by
adding stiffness and mass coefficients to the matrices Ks and Ms .
The set of normal modes is generally normalised with respect to Ms :
ΦTn Ms Φn = Inn ,

ΦTn Ks Φn = Λnn ≡ diag(ω2n ) (4.47)


where Φn is the matrix whose columns are the component normal modes
and Λnn contains the squared eigenvalues. This complete set of normal
modes is usually truncated to a smaller set of modes, which is denoted as
Φk . This truncation determines partly the accuracy of the model reduc-
tion and depends on many factors. For inclusion of the reduced model in
an MBS, a rule of thumb is suggested [202] that normal modes with fre-
quencies above two times the system frequency of interest be truncated.
2. Constraint modes.
A constraint mode corresponds to the static response of the substruc-
ture when statically imposing a unit displacement on one interface DOF
while holding the remaining interface DOFs fixed. The number of con-
straint modes is therefore equal to the number of interface coordinates.
One constraint mode (for the jth interface DOF) is defined by the equa-
tion:     
K j j K ji ujj fj
= (4.48)
K ji Kii ψc, j 0
120 4. Detailed modelling of the drive train in a wind turbine

where the vector u j j is a vector containing zeros except at the jth DOF
where it is equal to one, representing a unit displacement imposed on
this coordinate of the interface. The vector ψc, j represents the static dis-
placements at the interior DOFs for this case and the vector f j denotes
the reaction forces due to the known unit displacements. Solving equa-
tion (4.48) yields an expression for both vectors:

ψc, j = (−Kii −1 Ki j ) u j j (4.49)


−1
f j = (K j j − K ji Kii Ki j ) u j j (4.50)

The complete set of constraint modes is included in the matrix Ψc :


     
u11 uN j N j
Ψc = ··· (4.51)
ψc,1 ψc,N j
Note that solving the equations (4.49) and (4.50) requires that the par-
tition of the stiffness matrix Kii be nonsingular, which implies that the
group of interface coordinates y j is sufficient to prevent any rigid-body
motion. If this is not the case, a special consideration of rigid-body
modes is required, as described below.
3. Attachment modes.
An attachment mode is defined as the static deflection of the substructure
obtained by applying a unit force or moment to one interface coordinate,
while all other interface DOFs are free of external loads and kept fixed.
The number of attachment modes is again equal to the number of inter-
face coordinates and one attachment mode (for the jth interface DOF) is
defined by the equation:
   
K j j K ji ujj
ψa, j = (4.52)
K ji Kii 0
where the vector u j j represents the same unit vector as defined above.
The complete set of attachment modes is included in the matrix Ψa :

Ψa = ψa,1 · · · ψa,N j
 
(4.53)

Note again that the partition of the stiffness matrix Kii must not be sin-
gular in order to solve equation (4.52), which physically means that the
substructure must not possess any rigid-body freedom. Again, if this is
not the case, a special consideration of rigid-body modes is required.

Note furthermore that an attachment mode obtained by inverting equa-


tion (4.52) may include contributions from the normal modes of the sub-
structure. Since it would be undesirable to include their contribution
4.6 Flexible multibody models 121

twice in the columns of the reduction matrix Ts , it is important to re-


move this from the attachment mode. After this removal, the resulting
mode is a so-called residual attachment mode. These modes are intro-
duced in the CMS techniques of MacNeal [140] and Rubin [184] who
propose to compute them as:

r j = ψa, j − (Φk Λ−1


nn Φk ) u j j
T
(4.54)

where Φk is the modal matrix, which forms the truncated basis already
included in the columns of the reduction matrix Ts . Φk does not include
rigid-body motions.

4. Rigid-body modes.
Substructures that are unconstrained and have consequently rigid-body
freedom, require consideration of rigid-body modes. These modes may
be obtained in the process of calculating the component normal modes,
but they can also be seen as a special case of constraint modes, corre-
sponding to a set of physical coordinates used to restrain the substructure
against rigid-body motion. The rigid-body modes represent the gross
motion of a substructure. Since this motion is already included as a DOF
in the multibody system, inclusion of these modes is redundant. For this
reason, they are omitted from the global set of component modes for
each individual body.

As described above, the calculation of attachment modes for substruc-


tures with rigid-body freedom requires an alternative approach, which
leads to so-called inertia relief attachment modes. These are obtained
by applying to a body an equilibrated load system, which consists of the
originally specified unit forces and moments on all interface DOFs equi-
librated by the rigid-body d’Alembert (inertial) forces. The definition of
the applied load system results in no rigid-body motion and explains
the concept of “inertia relief”. These modes were also used by Mac-
Neal [140] and Rubin [184] and, in addition, by Craig and Chang [47].
Removing the contribution of these modes which is already included in
the truncated modal matrix Φk , leads in a similar way as described above
to residual inertia relief attachment modes.

Finally, the selected set of qs modes included in the matrix Ts , determines a set
of qs generalised coordinates cs according to equation (4.40). Each individual
generalised coordinate will be included as an additional DOF for the substruc-
ture in the DADS model. Many combinations of static and normal modes may
be chosen for inclusion in DADS, as long as they are linearly independent.
122 4. Detailed modelling of the drive train in a wind turbine

Here, the chosen alternative is a combination of static constraint modes and


fixed-interface normal modes.

1. Although the residual attachment modes are a solution for the possi-
ble interdependency of normal modes and attachment modes, the latter
modes are not considered in this work. The limitation to the use of static
constraint modes assures all modes to be independent.

2. The application of fixed-interface boundary conditions in the calculation


of normal modes yields a set without rigid-body modes, as required in
the flexible MBS formulation. In addition, these conditions correspond
best to the actual boundary conditions of drive train components, which
are supported by bearings and can make contact in the gear mesh. Con-
sequently, the mode shapes in a set of fixed-interface normal modes will
correspond best to the actual deformation of a drive train component,
resulting in a more accurate description of the overall deformation.

This combination of static constraint and fixed-interface normal modes is re-


ferred to as a set of “Craig-Bampton” modes [46]. The sub-structuring method
of Craig-Bampton is the technique used most intensively in the industry, owing
to its well-established accuracy of the approximation and its ease of numerical
implementation. The next sections describe how this type of mode set is in-
cluded in the flexible multibody models in DADS. MSC/NASTRAN is further
used for the calculation of these modes and all FE models include no damping.

4.6.1.2 Substructure coupling

This section describes how the set of Craig-Bampton modes is further pro-
cessed towards inclusion in the equations of motion in DADS. The global set
consists of qs modes or, more specifically, a truncated set of qk fixed-interface
normal modes Φk and a set of qc static constraint modes Ψc . As a result, equa-
tion (4.40) can be written as:

{ys } Ns ×1 = [Ts ] Ns ×qs · {cs } qs ×1


 
ck
{ys } = [ Φk Ψc ] ·
cc
{ys } = Φk · ck + Ψc · cc (4.55)

where ck is the vector of qk generalised coordinates corresponding to the nor-


mal modes and cc the vector of qc generalised coordinates corresponding to
the static constraint modes. The latter vector equals furthermore the vector of
interface coordinates, cc ≡ y j , which implies that qc = N j . This equality and
4.6 Flexible multibody models 123

the normal modes being fixed-interface, permits to write equation (4.55) in the
partitioned form:
     
yj 0 I qc ×qc ck
= · (4.56)
yi Φki Ψci cc

where Φki and Ψci are sub-matrices of Φk and Ψc respectively, corresponding


to the Ni interior DOFs. From equation (4.51), Ψci is given by:

Ψci = ψc,1 · · · ψc,N j


 
(4.57)

Based on equation (4.49), this can be written as:

Ψci = − Kii −1 Ki j (4.58)

Finally, using equation (4.56) in combination with the formulation for the re-
duced mass matrix in equation (4.45), yields:

0T ΦTki
     
M j j M ji 0 I qc ×qc
M̂s = · ·
I qc ×qc ΨTci Mi j Mii Φki Ψci
 
M̂kk M̂kc
=
M̂ck M̂cc

Assuming a normalisation of the normal modes according to equation (4.47)


leads to:

M̂kk = I qk ×qk
T
M̂kc = M̂ck = ΦTki (Mii Ψci + Mi j )
M̂cc = ΨTci (Mii Ψci + Mi j ) + M ji Ψci + M j j (4.59)

In a similar way, the reduced stiffness matrix can be written as:

0T ΦTki
     
K j j K ji 0 I qc ×qc
K̂s = · ·
I qc ×qc ΨTci Ki j Kii Φki Ψci
 
K̂kk K̂kc
=
K̂ck K̂cc

Carrying out matrix multiplications and replacing Ψci according to equation (4.58)
gives:

K̂kk = Λkk
T
K̂kc = K̂ck = 0 (4.60)
−1
K̂cc = K j j − K ji Kii Ki j (4.61)
124 4. Detailed modelling of the drive train in a wind turbine

where K̂cc corresponds to the “reduced stiffness matrix” often used in static
substructure analyses. This matrix represents the stiffness relation between a
limited set of coordinates in an FE model and is a reduction of the complete
stiffness matrix. The reduction in static analyses is also performed to decrease
the problem size and optimise the calculations.

Equation (4.60) indicates that the particular choice of a Craig-Bampton set as


Ritz basis “decouples” the stiffness matrix, meaning that there is no stiffness
coupling term between the coordinates ck and cc , which simplifies the calcula-
tion of the equation of motion (4.45). On the other hand, in the reduced mass
matrix there is inertia coupling between the normal mode coordinates and the
constraint mode coordinates.

The equations of motion of the reduced system are the additional equations
considered for a flexible body included in DADS. The DADS analysis code
solves for each of the generalised coordinates cs . The final solution can then
be transformed back into physical coordinates according to equation (4.40).
Section 4.6.2 describes in more detail the particularities for combining the re-
duced substructure with a multibody model in DADS.

4.6.2 Inclusion of a flexible body in DADS


For all bodies in a rigid multibody model, a reduced FE model can be included.
Equations of motion in DADS for the resulting flexible multibody systems,
contain coupling terms which yield a linear elastic response superimposed on
nonlinear rigid-body motion [138]. This means that the so-called flexible body
can experience elastic deformation in addition to a large overall rigid-body
motion.
1. The description of the rigid-body motion is based on the MBS formula-
tion presented in section 4.5 with six DOFs per body. The connection
between the flexible body and other (flexible) bodies in the model is only
possible at the interface nodes. These connections are implemented in a
similar way as for the rigid multibody systems, with joints or spring ele-
ments. This means that the presented formulations for the bearing flexi-
bility (cfr. section 4.5.1) and the tooth contact forces (cfr. section 4.5.2)
are still applicable here.
2. The generalised coordinates, based on a Craig-Bampton CMS reduction,
represent the elastic deformation of the flexible body in DADS. The con-
sideration of these extra DOFs has two objectives.
(a) The set of component modes describes the effect of a body’s static
flexibility and of its dynamic behaviour on the complete system in
4.6 Flexible multibody models 125

a more realistic way than in a rigid multibody model. This yields


more accurate equations of motion and, consequently, more accu-
rate load simulations.
(b) Knowledge of the elastic deformation of a body is indispensable
in the calculation of local stresses, which determine the fatigue
calculations in a drive train design.

Although both objectives are covered by the flexible MBS formulation,


the choice of component modes may differ slightly, depending on the
focus in the analysis. The former objective requires an accurate repre-
sentation of the body’s overall behaviour, while the latter objective may
require detailed insight at specific locations in the body with high stress
levels. As described above, the final choice of component modes re-
quires engineering judgement. The next discussion describes this proce-
dure and its consequences in order to get an accurate and efficient DADS
model.

The inertia and flexibility properties of a flexible body in DADS are repre-
sented by a set of Craig-Bampton modes and a nodal mass matrix.

1. The static flexibility is represented by the set of static constraint modes.


Including all static constraint modes, makes this technique general-purpose
and prevents any guesswork in the mode selection. However, it has the
disadvantage that often some irrelevant modes are considered, which
needlessly slows down the analysis. The following measures can opti-
mise the numerical calculation procedure.

(a) It is usually not necessary to include a static constraint mode, cor-


responding to applying a unit deformation at a specific interface
coordinate, when there is no constraint or force acting on this DOF
in the final DADS model. In this case, the nodal coordinate can
be removed from the set of interface coordinates, or the irrelevant
static mode can be removed from the global set of modes. In this
latter case, caution should be observed that the combination with
the fixed-interface normal modes is still a complete basis. There-
fore, static modes are only excluded from the sets in the flexible
multibody models in this work, when no normal modes are con-
sidered.
(b) The reduced mass matrix M̂cc and stiffness matrix K̂cc of a compo-
nent, corresponding to its static constraint mode coordinates cc (cfr.
equation (4.61)), often have off-diagonal terms. This indicates a
coupling between different coordinates, which may slow down the
126 4. Detailed modelling of the drive train in a wind turbine

analysis. Therefore, the reduced mass matrix M̂s and stiffness ma-
trix K̂s are usually orthogonalised in DADS. This procedure yields
a new diagonal stiffness matrix, including the eigenvalues of the
system, and the identity matrix for the mass matrix. It has fur-
thermore the added benefit of filtering out any rigid-body modes
that could originate from a static constraint mode set, as a result of
applying unit deformations in the same direction to several inter-
face nodes. Section 4.6.1.1 describes that these modes should be
omitted in a flexible multibody model.

2. The set of fixed-interface normal modes represents the dynamic be-


haviour of a flexible body. A truncation of this set reduces the number
of DOFs and simplifies the equations of motion, which reduces the cal-
culation time. The truncation depends typically on the frequency range
of interest in the analysis. In the present work, a rule of thumb for this
truncation is applied. This implies that all normal modes up to twice the
frequency of interest are included.

3. The nodal mass matrix of the flexible body’s FE model is also included
in the DADS analysis. This matrix represents the body’s inertial prop-
erties, which implies that these are no longer input parameters, but are
automatically calculated from the reduced model. Since the FE models
are usually derived directly from CAD models with a very realistic rep-
resentation of geometry, the mass distribution will be close to reality as
long as a sufficient amount of nodes is used in the FE model. This means
that the FE mesh should not only represent the stiffness and dynamic be-
haviour of the substructure accurately, but also its mass lumping. Gen-
erally, when the mesh fulfills the former requirement, it will also meet
the latter one.

Section 6.3 demonstrates the application of the flexible MBS formulation in a


static and a dynamic analysis of two individual gear stages in a wind turbine
gearbox. The objective of using flexible bodies in these analyses is having an
accurate representation of the body’s behaviour rather than focussing on local
stresses.

The calculation of local stresses from the simulated output of flexible multi-
body models requires an additional step in the post-processing of the results.
After all, this output is limited to numerical values for the generalised coordi-
nates cs . According to equation (4.40), these values can be transformed into
numerical values for all original nodal coordinates in the FE model. Since the
general coordinates describe only an approximation of the body’s deforma-
tion, this yields only an approximation of the original nodal coordinates. The
4.7 Conclusions 127

accuracy of this transformation is consequently determined by the choice of


component modes. This implies a trade-off between more modes and, thus, a
higher accuracy or less modes and faster calculation times. The transforma-
tion of general coordinates into nodal coordinates is possible in the FE soft-
ware. In this software, the nodal coordinates can subsequently be processed
into stresses and strains in the component, which can then be used for strength
and fatigue analyses. Baumjohann [14] demonstrates this technique for the
calculation of stresses in wind turbine blades based on DADS models. The
present work does not elaborate on this technique.

4.7 Conclusions
The drive train model in traditional wind turbine design codes comprises gen-
erally only one DOF to describe its behaviour, which imposes considerable
limitations on the simulation of drive train loads. This chapter presents differ-
ent modelling techniques to analyse these loads in more detail. The focus is
limited to wind turbines including a gearbox. These gearboxes include usually
both parallel gear stages and planetary gear stages as well as spur gears and
helical gears. A proper description of the forces acting between two meshing
gears is presented. The definition of these forces as a result of the tooth de-
formation along the line of contact, leads to the introduction of the gear mesh
stiffness. Other flexibilities in the drive train are defined for the deformation
of the bearings and for the components. An accurate dynamic model requires
a correct description of these flexibilities as well as of all components’ inertias
and of joints between the different components. Three structural modelling
approaches are available.

1. A structural model based on an FE formulation has typically a large


number of DOFs for each individual drive train component. This implies
high computational demands and, as a result, this approach is rarely used
in dynamic models of gear systems.

2. The MBS formulation, on the other hand, is the most popular approach
in the literature on gear dynamics. All publications about the application
of this technique describe models with at least one DOF per body, which
are purely torsional models, to investigate the torque in the drive train.
More accurate simulations are obtained when all six rigid-body DOFs
of a body are included. In most publications of such multibody mod-
els for gear systems, the gear mesh is represented as a spring-damper
element. Kahraman and Singh [113] distinguish here further between
LTI and LTV models. The latter models focus mainly on the effect of a
128 4. Detailed modelling of the drive train in a wind turbine

varying transmission error in the gear system, which can be considered


as the main source of excitation in a gear mesh. The prediction of this
excitation is not within the scope of this work and, therefore, the gear
mesh element is simplified to a spring with constant stiffness. Regard-
ing the representation of the bearings in dynamic gear models, the most
popular approach is also a spring-damper element. The element in this
dissertation represents a linear approximation of the bearing flexibility,
which is valid in a limited load range.

3. Most publications about dynamic gear models are limited to the MBS
approach and do not discuss the application of the flexible MBS for-
mulation. This dissertation presents an important step forward, since
it combines the modelling techniques for gear systems with the use of
flexible bodies.

Three modelling approaches with increased level of complexity are investi-


gated subsequently in this dissertation. Their methodology is summarised be-
low and table 4.1 compares the different techniques. Chapters 5 and 6 discuss
the application of these techniques and give an overview of the advantages and
the limitations of the different levels of modelling, which helps in deriving
a consistent modelling approach for the drive train in a wind turbine. Ap-
pendix E gives an overview of the numerical calculation procedures in DADS
and includes a comparison of the computational time for the three multibody
models.

1. The first method is using a purely torsional multibody model, in which


each drive train component is represented by one rotational DOF. In
these models, only the flexibility concentrated in shafts and gear teeth
is included in a direct way. Taking into account the effect of other flexi-
bilities is only possible as a reduction of the equivalent stiffness values,
which often complicates the modelling. Furthermore, due to the limita-
tion in DOFs, only the torque in the drive train is simulated directly.

2. The second technique is using a rigid multibody model with discrete flex-
ible elements, in which each body has six DOFs. This adds the possibil-
ity of investigating the influence of the bearing flexibilities on the torque
dynamics, without the complicated calculation of the stiffness reduc-
tion factors. Furthermore, the analysis can also yield insight in dynamic
bearing loads, which are coupled with the displacements of the bodies
in their bearings. The bearing model is an LTI 6 × 6 stiffness matrix and
the model of the gear mesh between two gears is an LTI spring element,
which couples twelve DOFs. Kahraman [105] used these models in the
4.7 Conclusions 129

analysis of a helical parallel gear system and Lin and Parker applied
these techniques for the study of spur planetary gears. The synthesis of
their work in this dissertation yields two three-dimensional plug-in com-
ponents to represent a bearing and a gear mesh respectively, which can
be used in more complex models of helical planetary gear systems and,
furthermore, of complete gearboxes integrated in a wind turbine.

3. The third method is using flexible multibody models. These models in-
clude additional DOFs per body to represent their deformation, such
that the flexible body can experience elastic deformation in addition to a
large overall rigid-body motion. This approach has two objectives.

(a) A more realistic description of a body’s static flexibility and of its


dynamic behaviour yields a more accurate simulation of the drive
train loads.
(b) The simulation of internal deformation can be transferred into local
stresses, which are required in fatigue calculations.

The CMS technique is a reduction method to determine and include


these extra DOFs. It can be split up in two steps:

(a) the reduction of an FE model of the drive train component into a


global set of component modes, which can be used as a Ritz basis
to approximate its static and dynamic behaviour
(b) the coupling of the reduced component with the rest of the structure
at predetermined interface nodes

Different types of component modes exist, such as normal modes, static


constraint modes, static attachment modes and rigid-body modes. Nor-
mal modes describe the dynamic behaviour of a flexible body, whereas
its static flexibility is represented by the set of static modes. Various
sets of modes may be combined in the global set, but the inclusion of
rigid-body modes is redundant in a flexible multibody model. For the
modelling of drive trains in DADS, the set of Craig-Bampton modes is
chosen as the best alternative. This set includes fixed-interface normal
modes and static constraint modes. A nodal mass matrix represents the
inertia properties of a flexible body.

Various measures can optimise the calculation time for flexible multi-
body models, such as a truncation of the component modes set. As a
rule of thumb, all normal modes up to twice the frequency of interest
are included. All modes at a higher frequency are omitted. There is no
130

Approach DOFs realistic representation of insight in variations of


per gear bearing component torque six load local
body mesh flexibility flexibility components stresses
Purely torsional
multibody 1 + + (**)
models

Rigid multibody

in the analyses of various gear systems.


models with discrete 6 + + + +
flexible elements

Flexible
multibody 6 + (*) + + + + + +
matrix can further improve the calculation time.

models

(*): an additional set of component modes represents the elastic deformation of a body
(**): only torque is simulated, but other load components can be derived in post-processing

work. Chapters 5 and 6 demonstrate the differences between these techniques


Table 4.1: Comparison of three modelling approaches applied in the present
dent. In addition, an orthogonalisation of the reduced stiffness and mass
rule of thumb for truncating the set of static modes. This is case depen-
4. Detailed modelling of the drive train in a wind turbine
5

Analysis of parallel and


planetary gear stages

5.1 Introduction

The generic multibody formulation described in chapter 4 is a general mod-


elling technique applicable to various kinds of drive trains. This chapter demon-
strates the use of purely torsional multibody models and of rigid multibody
models with discrete flexible elements for the modelling of two gear systems,
that originate from the literature. A comparison of the natural frequencies
calculated in both examples with the results from the literature, aims at a nu-
merical verification of the developed generic methodologies.

1. The first example is a helical parallel gear pair introduced by Kahra-


man [105]. Section 5.2 describes first a torsional model of this system,
including only one rotational DOF per component. The analysis pro-
ceeds with a rigid multibody model, including six DOFs per component,
and an investigation of the sensitivity from the system’s natural frequen-
cies to variations of the helix angle and the bearing stiffness values.

2. Section 5.3 describes subsequently the analysis of a planetary spur gear


system introduced by Lin and Parker [129]. Again, the discussion starts
with a torsional model and continues with a rigid multibody model. Fre-
quency response functions (FRFs) are calculated for this latter model, to
indicate the importance of different modes in e.g. the torque response of
the system. Furthermore, the analysis focusses on the sensitivity of the
results to bearing stiffness variations.

131
132 5. Analysis of parallel and planetary gear stages

In addition, section 5.4 presents the application of the multibody formulation


for a two-stage gearbox, which consists of the two gear stages, namely the
parallel gear pair coupled to the planetary gear system. The analysis is lim-
ited to the calculation of natural frequencies and corresponding mode shapes
and compares the results with those calculated for the individual gear stages.
Finally, section 5.5 describes how a multibody model can also be used to de-
termine the torsional gearbox stiffness, which is included in the value for the
drive train stiffness in the traditional wind turbine design codes.

All details about the numerical calculation of the normal modes, the FRFs and
the torsional stiffness in DADS are included in appendix E.

5.2 Parallel helical gear pair

Kahraman investigated the effect of the helix angle on the dynamics of a


generic gear system [105]. Figure 5.1 shows this system, which is used as
a demonstration and verification model in the present analyses. It consists
of a helical gear pair with a gear ratio of one. Both gears are mounted on
two rigid shafts1 supported by flexible rolling element bearings assembled in
a rigid housing. The clearances and stiffness changes of these bearings are
neglected and, furthermore, damping in the whole system is omitted, because
the analyses are limited to normal mode calculations. In addition, the input
and output shafts are free at their boundaries or, in other words, the torsional
effect of inertias at both sides is not included. Both shafts and gears have equal
masses m, polar moments of inertia J (around their rotation axis) and mass
moments of inertia I (around the bending axes of the shafts).

Section 5.2.1 describes the analysis of the presented system as a torsional


model with only one DOF per gear. Subsequently, section 5.2.2 discusses the
implementation of this system as a rigid multibody model, including six DOFs
for each gear. Both sections include an investigation of the effect of the helix
angle on the calculated results. A sensitivity analysis of the bearing stiffnesses
for the latter approach shows the influence of these support flexibilities.

1 Kahraman described both the shafts and the bearings as flexible, but included only one
equivalent stiffness value to represent both. Here, the same stiffness value is implemented, but
it is considered as a representation of rigid shafts in flexible bearings. This does not affect the
calculated results, but simplifies the discussion of the contribution of the individual flexibilities
in the sensitivity analyses.
5.2 Parallel helical gear pair 133

model input
igear 1
κgear (N/m) 2 · 108
α0n (◦ ) 20
rb0 (mm) 50
m (kg) 2.0
Td J (kg·m2 ) 2.9 · 10−3
I (kg·m2 ) 1.45 · 10−3
kbrad (N/m) 3.5 · 108
kbax (N/m) 1.0 · 108
kbtilt (Nm/rad) 277.5 · 103

Figure 5.1: Helical gear system. The two bearings supporting each of the rigid
shafts can be considered together as one equivalent radial (kbrad ), one axial
(kbax ) and one tilt stiffness value (kbtilt ). The helix angle β0b is varied through
the analysis and both the input and the output shaft are free at their boundaries.

5.2.1 Torsional model

The helical gear system is first modelled as a purely torsional model with the
gear mesh as only flexibility. This means that the bearings are considered to
be rigid in radial and axial directions (kb = ∞) and that the model has only
two DOFs as presented in figure 5.2. These DOFs are on the one hand the
coupled rotation of the gears in their bearings and on the other hand the defor-
mation of the teeth. A normal modes analysis shows that only the latter DOF
yields a non-zero eigenfrequency. This frequency corresponds to the analytical
solution of the following equation:

v !
u
1 u 1 1
ffree−free = · tκgear · (rb0 cos β0b )2 · +
2π J J · i2gear

κgear · (rb0 cos β0b )2


r
1
= · 2· (5.1)
2π J
134 5. Analysis of parallel and planetary gear stages

κgear · (rb0 cos β0b )2

eigenfrequencies (β0b = 0)
OC
J J
(1) 0 Hz
igear
(2) 2955 Hz

Figure 5.2: A torsional model for the helical gear system. Both the input and
the output shaft have a free boundary.

Equation (5.1) indicates that the eigenfrequency of this system varies linearly
with the cosine of β0b . Since the variation is limited to 10% for helix angles
below 25◦ , the influence of β0b is rather small. The effect of fixing
√ one of the
gears at its boundary reduces the eigenfrequency with a factor of 2:

κgear · (rb0 cos β0b )2


r
1
ffree−fix = · · (5.2)
2π J
For a helix angle of 0◦ , the free-fix eigenfrequency equals accordingly 2090 Hz.

5.2.2 Rigid multibody model


The previous torsional model is extended here to a model with six DOFs per
shaft. For each shaft, one stiffness matrix represents the bearing flexibilities.
This corresponds to the modelling approach of Kahraman. However, he used
only eight DOFs to represent the gear system instead of 12. He did not include
the displacement of the gears in the direction normal to the plane of action,
which does not change the model’s behaviour from the one presented here,
since the friction forces are assumed to be negligible and, therefore, any dis-
placement in this direction is uncoupled from the other gear in the presented
formulation. Furthermore, he neglected the rotations around the X-axis in fig-
ure 4.14, since they are usually small. Since this assumption is also valid for
the presented modelling approach, this can only cause a minor change in the
results.

Table 5.1 shows a comparison between the results from a normal modes calcu-
lation in DADS and the eigenfrequencies calculated by Kahraman for β0b = 0◦ and
β0b = 20◦ . Because the DADS model has four DOFs more, four extra eigenfre-
quencies are found, namely ω 6−7 and ω 10−11 . The former pair corresponds to
the displacement of the gears normal to the plane of action and the latter to the
rotations around the X-axis. These mode pairs cannot be used for verification
by means of Kahraman’s results. The other calculated eigenfrequencies match
5.2 Parallel helical gear pair 135

almost perfectly, which proves the validity of the model implementation in the
frictionless case. Furthermore, a qualitative visual comparison of the calcu-
lated mode shapes confirms this conclusion.

Eigen- β0b =0◦ β0b =20◦


frequency Peeters Kahraman Peeters Kahraman
1 0 0 0 0
2 1125 1125 1058 1064
3 1125 1125 1125 1125
4 1566 1566 1519 1519
5 2105 ng 2105 ng
6 2105 - 2105 -
7 2105 - 2105 -
8 2202 2201 2173 2173
9 2202 2201 2202 2201
10 2202 - 2202 -
11 2202 - 2202 -
12 3972 3972 4150 4149

Table 5.1: Comparison of the eigenfrequencies (Hz) calculated in DADS with


those presented by Kahraman [105] for a model of a helical gear system. The
5th eigenfrequency was not given (ng) in Kahraman’s results.

Figure 5.3 shows the mode shapes for ω 2 , ω 4 , ω 8 and ω 12 . The 4th mode
shape (for β0b = 0◦ ) corresponds best to the eigenmode, which was calculated
with the purely torsional model and which has the biggest impact on the torque.
Note the drop in frequency (2955 Hz → 1566 Hz) for this mode as a result of
including the bearing stiffness values in the rigid multibody model, which were
lacking in the torsional model. Users of torsional models are aware of this lim-
itation in their models and, therefore, often use gear mesh stiffness reduction
factors to include the effect of the bearing flexibilities. This is no longer nec-
essary in the rigid multibody model, which is clearly a way of representing the
gear system that is closer to physical reality. This yields in addition directly
more accurate predictions for the torque dynamics. Furthermore, several new
modes appear in the same frequency range, which indicates that the results are
not longer limited to the torque DOF only. This demonstrates further the added
value of the rigid multibody approach in comparison with a torsional model.
136 5. Analysis of parallel and planetary gear stages

β0b = 0◦ ω 2 = 1125 Hz β0b = 0◦ ω 4 = 1566 Hz

β0b = 20◦ ω 2 = 1058 Hz β0b = 20◦ ω 4 = 1519 Hz

(a) Mode 2 (b) Mode 4

β0b = 0◦ ω 8 = 2202 Hz β0b = 0◦ ω 12 = 3972 Hz

β0b = 20◦ ω 8 = 2173 Hz β0b = 20◦ ω 12 = 4150 Hz

(c) Mode 8 (d) Mode 12

Figure 5.3: Natural mode shapes for the helical gear system (wireframe: unde-
formed; solid: deformed).
5.2 Parallel helical gear pair 137

5.2.2.1 Influence of the bearing flexibilities

Figure 5.4 illustrates the property that increasing all bearing stiffness values
(kbax , kbrad , kbtilt ) for the helical gear system to infinity, yields the equivalent
of the torsional model. For this purpose, the real bearing stiffness values are
multiplied with a factor, which is taken equal for kbax , kbrad and kbtilt , since the
focus is not on the individual sensitivity of these values. All eigenfrequencies
increase towards infinity, except for ω4 (1566 Hz); this frequency approaches
asymptotically the torsional eigenfrequency (2955 Hz), which corresponds to
the conclusions above.
Natural frequency [Hz]

multiplication factor

Figure 5.4: The effect of multiplying kbax ,kbrad and kbtilt with an equal factor on
the natural frequencies of the helical parallel gear system (β0b = 0).

5.2.2.2 Influence of the helix angle

Table 5.1 shows the eigenfrequencies for β0b = 0◦ and β0b = 20◦ . The helix
angle only influences the eigenfrequencies ω 2 , ω 4 , ω 8 and ω 12 and the cor-
responding mode shapes, as shown in figures 5.5 and 5.3 respectively. For all
β0b values, the largest relative change in eigenfrequency occurs for ω 2 and is
limited to the cosine of β0b . This means that a simplification of a parallel he-
lical gear system to a spur gear pair yields an error of 6 % on the calculated
eigenfrequencies, when the helix angle is below 20◦ .
138 5. Analysis of parallel and planetary gear stages

ω 12
Natural frequency [Hz]

ω8

ω4

ω2

β0b [◦ ]

Figure 5.5: The effect of the helix angle on the natural frequencies of the
helical parallel gear system.

5.3 Planetary gear stages

The generic formulation of the present methodology can also be used for plan-
etary gear systems. There is no limitation on the number of planets, nor on
their positioning around the sun. The sun can be constrained by a bearing
model or can be modelled as floating, depending on the application. Further-
more, the choice of which component is constrained as non-rotating is an input
parameter for the model. The planetary gear systems introduced by Lin and
Parker [129] are used here as demonstration and verification models. They
developed an analytical model of a planetary gear system and applied it on a
single planetary stage with spur gears. The number of planets varies in their
analysis from three to five as shown in figure 5.6.

Since Lin and Parker focussed on spur gears, they limited their analysis to
planar vibrations and included only the 3 planar DOFs. Therefore, the mass
moments of inertia for tilting motion are not required, nor other than radial
bearing stiffness values. Furthermore, all planets are identical as well as all
sun-planet and planet-ring mesh stiffness values and corresponding pressure
angles. The stiffness values of the radial supports of the planet carrier, the
sun, the planets and the ring wheel are all equal (krad ). The planet carrier and
5.3 Planetary gear stages 139

the sun are free at their boundaries and the ring wheel is constrained as non-
rotating in the gearbox housing. The torsional deformation of the ring wheel
and the housing is represented by a torsional spring element between the ring
wheel and the ground.

Sun Planet Carrier Ring


Mass (kg) 0.4 0.66 5.43 2.35
J (·10−3 kgm2 ) 0.58 1.53 49.2 56.7
rb0 (mm) 38.7 50.2 96.9 -137.5
Mesh stiffness κgear = 5 · 108 N/m
Bearing Stiffness krad = 108 N/m
Torsional Stiffness kring−housing = 19 · 106 Nm/rad
Nominal Pressure Angle αt0 = 24.6◦
Helix Angle β = 0◦

(a) Model input parameters.

(b) The three planets system. (c) The four planets system. (d) The five planets system.

Figure 5.6: Planetary gear systems introduced by Lin and Parker [129]. In all
systems the planet carrier and the sun are free at their boundaries.

Firstly, section 5.3.1 presents the analyses of the planetary gear systems as
torsional models with only one DOF per gear. Subsequently, section 5.3.2 dis-
cusses the extension towards a rigid multibody model with six DOFs for each
gear and indicates the advantages of this approach for a planetary gear system.
Here, a sensitivity analysis of the bearing stiffness values shows furthermore
the influence of the radial support flexibilities.

5.3.1 Torsional model


Similar to the discussion of the parallel gear system, this section starts with a
torsional model of the planetary gear system with the gear mesh and the tor-
sional ring wheel constraint as only flexibilities. Figure 5.7 shows a schematic
overview of this approach for the three planets system. Table 5.2 presents the
140 5. Analysis of parallel and planetary gear stages

planet-ring mesh stiffness + gear ratio


kinematic constraint
flexibility torsional ring wheel support 2
sun-planet mesh stiffness + gear ratio

1 planet carrier
2,3,4 planets 1 3 6
5 ring wheel
6 sun

5 4

Figure 5.7: Representation of the torsional model for the three planets system.
The ring wheel is constrained as non-rotating and connected with the ground
through a torsional spring element. The planet-ring mesh elements include
a stiffness value and a gear ratio, representing the contact between the plan-
ets and the ring wheel. Likewise, the sun-planet mesh elements represent the
contact between the planets and the sun. The kinematic constraint elements
between the planets and the planet carrier describe the motion of the planets
relative to the planet carrier.

corresponding natural frequencies of the planetary gear systems with N plan-


ets (N=3,4,5). The mode at 0 Hz corresponds to the kinematic rotation of the
system. The calculation yields furthermore four non-zero eigenfrequencies.
One of them is quasi identical for the three systems (6.4 kHz) and has multi-
plicity N-1, resulting from the symmetry in the planetary stage. However, the
frequency response functions (FRFs) for the four planets system in figure 5.8
indicate that its influence on the torque fluctuations in the ring wheel, the plan-
ets and the sun is not visible for an excitation at the planet carrier. Therefore,
these modes do further not matter in a torsional analysis.

On the other hand, the FRFs indicate amplified torque levels at the three other
natural frequencies. These functions are used for qualitative evaluation only,
since the amplitude levels depend on the amount of damping in the system,
which is not within the scope of this analysis. Appendix E describes the nu-
merical calculation of the FRFs. The excitation signal used in the present
example has a power spectrum as indicated in figure 5.8(a). An excitation of
0.2 seconds is calculated with an iterative solver and a maximum time step
of 1/(60 kHz). The FRF calculations for the different torque time series are
performed using 10 averages, which yields a frequency resolution of 50 Hz.
5.3 Planetary gear stages 141

Natural N
frequency 3 4 5
(1) 0 0 0
(2) 2217 2138 2059
(3) 6159 6451 (×3) 6444(×4)
(4) 6444 (×2) 6688 7105
(5) 11205 12577 13810

Table 5.2: Natural frequencies calculated for the torsional model of the plane-
tary gear stages with N planets.
PSD [dB/Hz]

PSD [dB/Hz]

[Hz] [Hz]
(a) Power spectrum of the torque excitation (b) FRF from the planet carrier torque to the
signal at the planet carrier. torque at the ring wheel.
PSD [dB/Hz]

PSD [dB/Hz]

[Hz] [Hz]
(c) FRF from the planet carrier torque to the (d) FRF from the planet carrier torque to the
torque at one of the planets. torque at the sun.

Figure 5.8: Response calculation for a sinusoidal excitation of the torsional


model of the four planets system at the planet carrier.
142 5. Analysis of parallel and planetary gear stages

5.3.2 Rigid multibody model


The rigid multibody model includes all flexibilities presented in figure 5.6.
Since the helix angle is zero and, consequently, no forces are acting out-of-
plane, the analysis is limited to planar vibrations in the gear system, as in
the work of Lin and Parker [129]. This limitation in our model is made by
constraining all components with bearings that are infinitely stiff for the dis-
placement and rotations out-of-plane (axial and tilt stiffness). Table 5.3 divides
the natural frequencies of the three planetary systems according to the classifi-
cation by Lin and Parker. They distinguished three categories of planar modes
for planetary systems with N ≥ 3, satisfying equation (5.3) for the angular po-
sitioning ψN around the sun [131]. Figure 5.9 shows a planetary gear stage
with three planets and clarifies how the angle ψN is defined.

∑ sin ψN = ∑ cos ψN = 0 (5.3)

planet 2

ψ2 = 120◦

sun ψ1 = 0 ◦

ψ3 = 240◦

planet 1
planet 3

Figure 5.9: Definition of the angular positioning of the planets in a planetary


gear system with N=3.
5.3 Planetary gear stages 143

Mode N
shape 3 4 5



m=1 R1 0 0 0


R2 1425 1519 1538


R3 2032 2079 2082


R4 2644 2630 2602


R5 7500 7805 8086


R6 11744 13052 14237


m=2 T 1a,b 770 759 745


T 2a,b 1101 1092 1073


T 3a,b 1989 1947 1921


T 4a,b 2238 2328 2421


T 5a,b 7060 7249 7427


T 6a,b 9582 10392 11136


m = N-3 P1 1959 1959


P2 6450 6444


P3 6497 6497

Table 5.3: Classification of the natural frequencies for the planetary gear sys-
tems with N planets, as introduced by Lin and Parker [129].

Figure 5.10 shows an example of a mode shape for each of the three cate-
gories. Figure 5.10(a) shows the undeformed model. The deformation in the
mode shapes can be seen as the difference between this figure and the fig-
ures 5.10(b), 5.10(c) and 5.10(d) respectively. This deformation is a combi-
nation of the deformation in the bearings (planets, sun, planet carrier and ring
wheel), the deformation in the gear meshes (sun-planet, planet-ring) and the
torsional deformation of the ring wheel and the housing.

1. Six rotational modes always have multiplicity m = 1 for various num-




bers of planets N. The mode shapes ( R 1−6 ) have pure rotation of the
carrier, ring and sun and all planets have the same motion in phase.

2. Six translational modes always have multiplicity m = 2 for different N.




Here, the six mode shape pairs ( T 1a,b−6a,b ) have pure translation of the
carrier, ring and sun.
144 5. Analysis of parallel and planetary gear stages

3. Three planet modes exist for N > 3 and have multiplicity m = N − 3.


The carrier, ring and sun have no rotation, nor translation in the corre-


sponding mode shapes ( P 1−3 ).



(a) undeformed model (b) rotational mode ( R 2 : 1519 Hz):
the motion of all planets relative to
their planet bearings is identical. They
move furthermore in phase. The mo-
tions of the planet carrier, the ring
wheel and the sun are pure rotations.


− →

(c) translational mode ( T 1a : 759 (d) planet mode ( P 1 : 1959 Hz): the
Hz): the motions of the planet carrier, planets move, while the planet carrier,
the ring wheel and the sun are pure the ring wheel and the sun stand com-
planar translations. pletely still.

Figure 5.10: Classification of the mode shapes for a planetary gear system
(N=4).
5.3 Planetary gear stages 145

This classification implies that such planetary gear systems can only have fif-
teen different natural frequencies, since additional planets do not yield addi-
tional frequencies, but change only the multiplicity of the planet modes. The
natural frequencies of these modes are furthermore identical for the four and
five planets systems. The classification of the DADS results is based on anima-
tions of the mode shapes and the corresponding natural frequencies correlate
well with those calculated by Lin and Parker, which further proves the validity
of the model implementation for planetary stages.

Figure 5.11 shows the FRFs for a torque excitation at the planet carrier and the
responses in the torque at the ring wheel, the planets and the sun2 . Qualitative
evaluation of these functions indicates amplified torque levels at the frequen-
cies of the rotational mode shapes only. This means that the other types of
modes cannot be excited by an external torque excitation. However, these
modes may eventually be excited by other excitation sources, such as e.g. a
radial force excitation originating internally from the gear mesh vibrations.

The effect of the rotational modes on the torque response in a planetary sys-
tem is further clarified by a comparison between the results of the torsional
models and the rigid multibody approach. In the latter models, extra DOFs
are taken into account and, furthermore, realistic radial bearing flexibilities for
the supports of the sun, the planets, the carrier and the ring wheel. Comparing
table 5.2 with table 5.3 indicates first of all that additional modes are identified
in a rigid multibody model, as a result of the extra DOFs. Moreover, the con-
sideration of the bearing stiffness values has a major impact on the results.

This is further demonstrated in figure 5.12 for the three planets system. For
a multiplication factor of one, the curves give the natural frequencies of the
rigid multibody model. The higher factors correspond to a gradual increase of
the radial bearing stiffness values. This increase is a theoretical approach to
describe what happens when the flexibility in the bearings is omitted, which
clearly differs from physical reality3 . The frequencies corresponding to the
first four rotational modes and the first translational double mode approach
asymptotically the results from the torsional model when the stiffness values
approach infinity.

2 The excitation signal and the simulation parameters used in the numerical calculation of
the FRFs are identical with the torsional model of the four planets system.
3 A stiffness variation up to a multiplication factor of three may be considered as physically

feasible. The higher multiplication factors are only added to show the trend towards a purely
torsional model with the omission of the flexibility in the bearings.
146 5. Analysis of parallel and planetary gear stages

PSD [dB/Hz]

PSD [dB/Hz]
[Hz] [Hz]
(a) Power spectrum of the torque excitation (b) FRF from the planet carrier torque to the
signal at the planet carrier. torque at the ring wheel.
PSD [dB/Hz]

PSD [dB/Hz]

[Hz] [Hz]
(c) FRF from the planet carrier torque to the (d) FRF from the planet carrier torque to the
torque at one of the planets. torque at the sun.

Figure 5.11: Response calculation for a sinusoidal excitation of the rigid multi-
body model of the four planets system at the planet carrier. Only the frequen-
cies corresponding to the rotational modes are visible in the response spectra.

This phenomenon is numerically shown in table 5.4 where the natural frequen-
cies calculated with a rigid multibody model are compared with the results for
a torsional model. The FRFs for the torsional models in figure 5.8 indicated the
negligible effect of the double mode in a torque response, which corresponds
to the conclusion that only the rotational modes have an effect on the torque
dynamics.

This example shows again the added value of the rigid multibody approach
with respect to the torsional models, which is the more realistic consideration
of the bearing stiffness values and the identification of additional modes.
5.4 Gearbox with a parallel and a planetary gear stage 147

Natural frequency [kHz]



R4


T 1a,b



R3


R2



R1

Multiplication factor

Figure 5.12: Influence of increasing the radial bearing stiffness values on the
natural frequencies of the planetary gear system with three planets.

5.4 Gearbox with a parallel and a planetary gear stage

In various wind turbine gearboxes, both parallel and planetary gear stages ex-
ist. This section investigates the coupling of such stages, based on the systems
introduced in the previous sections. Figure 5.13 shows a coupling of the plan-
etary gear system (N = 4), as presented in figure 5.6, with the helical parallel
gear system (β0b = 20◦ ), as presented in figure 5.1. This coupling is imple-
mented as a rigid link between the torsional DOF of the sun and the torsional
DOF of the parallel input shaft; all other DOFs are not considered. This im-
plies that only torque is transmitted via the coupling. Furthermore, both the
planet carrier and the parallel output shaft are free at their boundaries.

Table 5.5 shows the natural frequencies of the presented combined system,
calculated with a rigid multibody model. Comparison of these results with the
results calculated for the individual stages, yields an interesting conclusion.
Several modes appear in the combined system as in the individual system, at
quasi the same frequency.
148 5. Analysis of parallel and planetary gear stages

Mode shape Rigid multibody model Torsional model




1 0 ( R 1) 0
→−
2 770 ( T 1 ) H * 2217


− HH 
3 1101 ( T 2 )  3 6159

− HH
4 1425 ( R 2 )   j
H 6444 (×2)

− 
5 1989 ( T 3 )  3 11205




6 2032 ( R 3 )  

− 
7 2238 ( T 4 ) 

− 
8 2644 ( R 4 ) 


9 7060 ( T 5 )


10 7500 ( R 5 )


11 9582 ( T 6 )


12 11744 ( R 6 )

Table 5.4: Numerical comparison of the natural frequencies (Hz) calculated


with a rigid multibody model and with a torsional model for the planetary stage

− →

with three planets. ( R ) are the rotational modes (m = 1) and ( T ) the transla-
tional modes (m = 2) found in the former model. The arrows indicate how the
frequencies shift from the rigid multibody model results to the torsional model
results, when the radial bearing stiffness values are increased towards infinity.

B
B
B
B
B
B
BN
Rigid coupling of torsional DOFs

Figure 5.13: Coupling of a planetary spur gear system and a helical parallel
gear stage with a torsionally rigid link.
5.5 Torsional stiffness of a gearbox 149

1. For the planetary stage, these are five of the six rotational modes, all
translational and all planet modes, which are categorised accordingly.

2. For the parallel stage, seven of the eleven non-zero natural frequencies
remain constant, when coupled to the planetary stage. Note that both
the planetary and the parallel gear system have a natural frequency at
1519 Hz, which makes it hard to assess whether mode no. 7 at 1519 Hz
should be classified as a rotational mode of the planetary gear system
or as a mode in the parallel stage. Here, the corresponding mode shape
yields more insight and identifies mode no. 7 as a mode in the planetary
stage.

3. Other modes change in frequency and are classified as global modes.


Moreover, since the rigid coupling of two DOFs removes one DOF, the
analysis yields one natural frequency less, namely one rigid-body mo-
tion at 0 Hz.

The actual coupling of two gear stages in a gearbox is case dependent and will
influence the natural frequencies of the overall gearbox. However, the present
example demonstrates that certain internal eigenmodes can be insensitive to
external boundaries, in which case an analysis of the individual gear stage
yields already valuable insight.

5.5 Torsional stiffness of a gearbox

Section 3.3.3.1 describes how the drive train in the traditional wind turbine
design codes is represented by one torsional spring element. This element rep-
resents the flexibilities of all drive train components, which contribute to the
torsional deformation in the drive train. The gearbox is one of these drive train
components. Its flexibility can be calculated in a static FE analysis. However,
building an FE model for a multistage gearbox with a large number of com-
ponents, including an accurate representation of all gear contacts and bear-
ing supports is labour-intensive. Therefore, the torsional gearbox stiffness is
usually calculated as a combination of many stiffness values for the different
gearbox components, such as the gears, the bearings and the shafts. Since
these values are all included in the presented multibody models, as well as a
correct description of the load path, these models can also yield a value for
the torsional gearbox stiffness. This section demonstrates this method for the
gearbox above.
150 5. Analysis of parallel and planetary gear stages

No. Global mode Planetary stage Parallel stage


Rotational Translational Planet
mode mode mode
(m = 1) (m = 2) (m = 1)
1 0
2 759
3 973
4 1092
5 1125
6 1220
7 1519
8 1947
9 1959
10 2072
11 2105 (×3)
12 2173
13 2202 (×3)
14 2328
15 2589
16 3950
17 6450
18 6497
19 7249
20 7804
20 8028
22 10392

Table 5.5: Natural frequencies calculated with a rigid multibody model of the
four planets gear system coupled torsionally to the helical parallel gear stage.
These frequencies are put in the category of an individual stage, when they are
equal to the individually calculated results; otherwise, they are global modes.

Figure 5.14 shows a schematic overview of the gear setup. The planetary stage
is driven by the torque Td and the housings of both gear stages are fixed. The
input shaft of the parallel stage follows the rotation of the sun output shaft of
the planetary stage. The driving torque is balanced by fixing the rotation of the
output shaft of the parallel stage4 . When Td = 500 Nm, the tooth contact forces
on this shaft cause a radial deformation of the supporting bearings of 6.27 µm.
Since the planet carrier rotates 0.35 mrad, the overall torsional stiffness of the
gear setup equals 1.42 MNm/rad, related to the low speed side.
4 Appendix E describes the numerical procedure for performing a static analysis in DADS.
5.5 Torsional stiffness of a gearbox 151

iplanetary = 4.55 iparallel = 1

Td

Figure 5.14: Schematic representation of the two-stage gearbox introduced in


figure 5.13.

When specific flexibilities or parameters are varied, this stiffness changes as


described here:

No. Description Torsional


stiffness
(MNm/rad)
1 normal setup 1.42
2 rigid connection of the radial planet bearing supports 2.62
3 rigid connection of the torsional ring wheel support 1.49
4 rigid connection of the radial bearing supports of the 1.54
output shaft
5 doubling the gear mesh stiffness in the parallel 1.54
stage
6 changing the helix angle in the parallel stage 1.69
from β0b = 20◦ to β0b = 0◦
7 doubling the sun-planet and planet-ring gear mesh 1.46
stiffness in the planetary stage

These arbitrary examples indicate the individual contribution of different flex-


ibilities and parameters to the torsional stiffness value. For instance, the radial
stiffness of the planet bearings has a considerable effect on the overall stiff-
ness, while the influence of the sun-planet and planet-ring gear mesh is quasi
negligible.

When a gearbox is modelled as a flexible multibody model, the results of a


static analysis include the elastic deformation of all flexible bodies. As de-
scribed in section 4.6, these results can be further processed to stress distri-
152 5. Analysis of parallel and planetary gear stages

butions in these components. This technique offers a valuable alternative for


non-linear FE contact analyses of gear systems, which imply high computa-
tional demands.

5.6 Conclusions
The analyses of two gear systems from the literature aim at validating the
implementation of two modelling techniques described in chapter 4 and at
demonstrating the use of these techniques. The former purpose is based on
a comparison of natural frequencies and corresponding mode shapes found in
the literature, with the results from calculations in DADS for the two systems.
The latter demonstration is based on the state-of-the-art in modelling gear dy-
namics. Chapter 6 presents the application of the modelling techniques for
helical planetary gear stages and flexible multibody models, which go further
than the state-of-the-art.

The first system is a parallel helical gear pair presented by Kahraman [105].
Both the frequencies and mode shapes calculated in DADS for a rigid multi-
body model of this system match with Kahraman’s results, which proves the
validity of the implementation for a parallel gear stage. Furthermore, a com-
parison of these DADS results with the results for a purely torsional multibody
model, indicates the enhanced capabilities of the more elaborate rigid multi-
body approach which considers six DOFs per body.

1. This approach includes bearing flexibilities, which have an important


impact on the torsional eigenfrequency calculated in the purely torsional
model. Moreover, it permits to calculate additional relevant modes in
the system, which may also be excited by e.g. external torque variations.

2. Since all bearings are included as separate spring elements, the rigid
multibody approach permits to examine their individual influence in a
direct way. A sensitivity analysis for Kahraman’s helical gear pair indi-
cates that the results from the rigid multibody approach shift towards the
single torsional mode for increasing bearing stiffness values. A second
sensitivity analysis indicates the minor influence of the helix angle on
the eigenfrequencies.

The second example considers a collection of three planetary spur gear sys-
tems presented by Lin and Parker [129], with three, four and five planets re-
spectively. According to Lin and Parker, only fifteen different eigenmodes of
these systems can be identified, which can subsequently be classified as rota-
tional, translational and planet modes. This classification is clearly recognised
5.6 Conclusions 153

in the results calculated in DADS with a rigid multibody model of these sys-
tems. In addition, a good correlation between the corresponding eigenfrequen-
cies proves further the validity of the implementation for planetary gear stages.
Based on a frequency response calculation, it is demonstrated that only the ro-
tational modes in a planetary spur gear system can be excited by an external
torque variation. The frequencies of these rotational modes shift furthermore
for increasing bearing stiffness values towards the frequencies, which can be
calculated with a purely torsional model. However, this latter approach does
not permit a direct consideration of the bearing flexibilities.

When the models of the parallel gear system and the planetary gear system
with four planets are coupled with a torsional rigid link, they compose a sim-
ple two-stage gearbox. The analysis of the eigenmodes and corresponding
frequencies of this gearbox indicates that part of them are identical to the re-
sults calculated for the individual stages. These eigenmodes are the so-called
local modes, in contrast with the global modes of the system. This means that,
for the former modes, the analysis of an individual stage yields already valu-
able insight. Finally, when a multibody model of a gearbox is available, it can
offer a valuable alternative to a large FE model in a static analysis, which aims
at determining the torsional stiffness of a gearbox or local stresses in gearbox
components.
154
6

Analysis of the drive train in a


modern wind turbine

6.1 Introduction
Chapter 5 describes the use of purely torsional models and of rigid multibody
models for the analysis of a helical parallel gear stage and of a spur planetary
gear system, both found in the literature. This chapter demonstrates the appli-
cation of the three presented modelling approaches for a drive train in a wind
turbine. Especially the investigation of a helical planetary gear stage, the use
of flexible multibody models and the analysis of a multistage gearbox model,
are important improvements of the state-of-the-art. In addition, the gearbox
model is integrated in a model of the complete wind turbine and an overview
of various types of analyses is illustrated.

Section 6.2 starts with a description of the drive train, which is a generic exam-
ple. This example is representative for a drive train in a modern wind turbine
and it consists of a gearbox with one parallel and two planetary gear stages.
Both the high speed planetary stage and the parallel stage are helical gear sys-
tems. Section 6.3 presents dedicated models for the three gear stages individ-
ually.
1. For the parallel stage, this includes a torsional model and a rigid multi-
body model, similar to the approach in chapter 5, but furthermore an
extension towards a flexible multibody model. In this latter model, both
shafts are reduced FE models, which include the inertia as well as the
stiffness properties of the bodies. A static analysis for the respective
models of the parallel stage demonstrates the particular influence of the
gear mesh, the bearings and the components on the overall torsional
stiffness of this stage. In addition, a dynamic analysis based on the

155
156 6. Analysis of the drive train in a modern wind turbine

calculation of eigenmodes and corresponding frequencies indicates the


possibilities and limitations of the different modelling approaches and
yields insight in the dynamics of the parallel gear stage.
2. The individual analysis of the helical planetary gear stage is performed
in a similar fashion as described for the parallel stage. However, the
dynamic analysis is limited here to the approach with the rigid multibody
model only. This analysis leads to the introduction of a new category of
mode shapes.
3. The section ends with a discussion of the low speed planetary stage,
which has spur gears.
Subsequently, section 6.4 presents an extension of the individual models to-
wards a model of the complete gearbox. The dynamic analysis of this model
focusses on how the gearbox boundaries influence the eigenfrequencies of this
system. Finally, section 6.5 investigates the integration of this gearbox model
in a model of the complete wind turbine. This is done in two steps. Firstly,
only the additional components in the drive train at the generator side of the
gearbox are added. The second step makes the model complete by taking fur-
thermore the rotor and the tower into account.

This latter comprehensive model is a significant extension of the structural


model in the traditional wind turbine design codes, as described in chapter 3.
The traditional model is considered to be applicable for all analyses in the
frequency range [ 0 - 10 Hz]. The extended model presented in this chap-
ter includes much more detail for the drive train and enables consequently to
gain more insight in the drive train behaviour, including loads at frequencies
above 10 Hz. The additional benefits are illustrated in section 6.5. Firstly, an
overview of the eigenmodes and eigenfrequencies of the system is given. Sec-
ondly, a frequency response analysis in the range [ 10 - 1500 Hz] is described
to demonstrate how certain excitations can lead to amplified load levels in
the drive train. Thirdly, the simulation of a transient load case is discussed.
Finally, the features of the new simulation approach in this research are sum-
marised according to the research objectives of this dissertation.

6.2 Drive train layout


Figure 6.1 shows the wind turbine for the present application. It is a generic
example of which the results are representative for a modern wind turbine. The
drive train concept is similar to the concept introduced in figure 2.17(c), which
has one main bearing integrated in the gearbox carrying the wind turbine rotor.
The generator is a doubly fed induction generator (DFIG) and the gearbox
6.2 Drive train layout 157

design is a combination of two planetary stages with one high speed parallel
stage. The wind turbine rotor is connected to the planet carrier of the first
planetary stage. This stage has spur gears and its ring wheel is fixed in the
gearbox housing. This housing is assumed to be rigid as well as its connection
to the bed plate. This latter frame supports also the generator and rests on
the yaw bearing, which connects the complete nacelle with the tower. The
second gear stage in the gearbox is a helical planetary stage. Its planet carrier
is driven by the sun of the first stage and its ring wheel is also fixed in the
gearbox housing. The sun of this stage drives the gear of the third stage, which
is a parallel stage with helical gears. The pinion of this stage rotates at the
speed of the generator. A brake disk is mounted on this output shaft and a
flexible coupling connects it with the input shaft of the generator. Since only

rotor hub
C
C
C
C high speed planetary stage
C
C high speed parallel stage
C
C
C
C flexible coupling
C 


 
 

 D
low speed planetary stage D
D 
 D 
 
D 



 D generator
bed plate DD
 brake disk
yaw bearing

Figure 6.1: Simplified representation of the three bladed wind turbine with a
zoom on its drive train.
the parallel gear stage causes a change in the direction of rotation, the high
speed pinion and the generator rotate in the opposite direction of the rotor.
According to equation (4.2), the reaction torque of the gearbox on the bed
plate is therefore slightly larger than the input torque. This torque acts in the
same direction as the rotation of the rotor, which is clockwise when looking
at the wind turbine in the direction of the wind. The torque on the generator
support acts in the other direction.
158 6. Analysis of the drive train in a modern wind turbine

6.3 Individual gear stages


As described in chapter 5, the analysis of individual gear stages can yield al-
ready valuable information concerning the dynamic behaviour of a complete
gearbox or drive train, which includes these stages. This section follows this
strategy and investigates the statics and dynamics of the individual stages.
Since this implies a splitting up of the drive train, each stage requires specific
artificial boundary conditions. The sun of the first stage is supported together
with the planet carrier of the second stage; splitting up the drive train requires
extra boundaries to replace this support. Similarly, the sun of the second stage
needs an additional support replacing the common support with the gear of the
third stage. The following individual analyses consider furthermore no exter-
nal inertias at input or output and, therefore, the boundaries are called “free”.
The discussion starts with the parallel gear stage and deals subsequently with
the high speed and the low speed planetary gear stages.

6.3.1 Parallel stage


6.3.1.1 Modelling
The high speed stage of the gearbox is a parallel helical gear system. For this
stage, three successive models are built, with each time an increased level of
detail and consideration of more flexibilities in the system:

1. a torsional model with the gear mesh as only deformable part;

2. a rigid multibody model, including additionally the spring elements for


the bearing deformations;

3. a flexible multibody model, considering in addition the components’ flex-


ibilities.

In all three models, both the gear and the pinion are free to rotate in their bear-
ings. Figure 6.2 shows these models. The torsional model in figure 6.2(a) has
only two DOFs and the torsional stiffness related to the low speed side of this
stage equals Kgear , according to figure 4.11. This stiffness represents only the
deformation of the teeth. Figure 6.2(b) indicates the location of the bearings,
which are modelled as discrete spring elements in this rigid multibody model.
Here, both the gear and the pinion have six DOFs. The flexible multibody
model in figure 6.2(c) is the most realistic representation of the helical gear
pair, since it includes for each body additional DOFs to describe its respective
deformation.
6.3 Individual gear stages 159

Kgear = κgear · (rb,gear


0 · cos β0b )2
OCC
C
Jgear Jpinion
igear

(a) Torsional model with one DOF per body, including the gear mesh as only
deformable part.

pinion bearings
gear bearings 
PP
H 
HPP 
HH 
H
HH
H
H
H
Z
Y
6
Q
sX
Q
(b) Rigid multibody model with six DOFs per body, including additional
bearing spring elements.

Z
Y
6
Q
sX
Q
(c) Flexible multibody model with additional DOFs per body to represent the
respective deformation.

Figure 6.2: Three models of the helical parallel gear stage.


160 6. Analysis of the drive train in a modern wind turbine

The derivation of the additional DOFs in the flexible multibody model is done
according to the CMS technique as described in section 4.6.1. The first step
in the CMS procedure is the calculation of an appropriate set of component
modes for an FE model of each body. Figure 6.3 shows both FE models, which
consist of solid elements. Here, five-sided elements with six nodes are used at
the inner part of the shafts and six-sided elements with eight nodes for the
rest of the structure1 . These implementations are based on a meshing of CAD
models, which implies furthermore that the FE models represent automatically
the correct mass and inertial properties. Since the lumped mass matrices of
both FE models are included in the DADS analyses, this guarantees an accu-
rate consideration of the inertial properties in the flexible bodies. Note that the
number of nodal coordinates in the FE model determines the size of these ma-
trices. In addition, it defines, in combination with the set of component modes,
the level of detail in the post-processing of the simulations, with respect to the
local deformations. The DADS software includes a useful tool to reduce the
number of nodal coordinates for each flexible body, which decreases the level
of detail in the simulations, but can accelerate the calculation process.

Both shafts in the parallel stage interface with each other at the gear mesh and
with the housing via two bearings. This leads to a set of three interface nodes
per body. The model representation of the gear mesh and the bearings, implies
a spring coupling at the respective interface nodes, as described in section 4.5.
Figure 6.3(a) describes the location of these nodes in the pinion model. They
lie at the symmetry axis of the pinion and are rigidly connected with all nodes
at the outer diameter of the shaft at that location. This rigid connection implies
that the corresponding cross sections of the pinion do not deform, which intro-
duces a stiffening effect that is further neglected. Figure 6.3(b) describes the
same approach for the gear model. Since this is a hollow shaft, the interface
nodes are additional points at the symmetry axis, not part of any solid element.
Again they are rigidly connected to all nodes at the outer diameter, as demon-
strated for one bearing interface. Figure 6.4 shows a schematic representation
of the flexible multibody model of the helical parallel gear stage, which indi-
cates the discrete flexible elements to represent the bearings and the gear mesh
and the reduced FE models to represent the pinion and the gear.

The load transfer at the gear mesh interface differs from the load transfer at
the bearing interface. The gear mesh introduces loads in the six coordinates of
the interface node, whereas the bearings form no boundaries for the rotation in
any direction. This results in the definition of the set of static modes:

1 The specific name of the five-sided and six-sided elements in the FE software MSC/Nastran

is CPENTA and CHEXA respectively.


6.3 Individual gear stages 161

( interface nodes at the bearings


(((
((((
( (((( 


A 
A
A
A
A
A
interface node at the gear mesh

(a) Pinion model (2937 nodes and 2720 elements).

Q
Q
Q
Q
Q
Q
interface node at one bearing

(b) Gear model (1139 nodes and 816 elements).

Figure 6.3: FE models of the two components in the helical parallel gear stage.
which are reduced to a set of component modes according to the CMS tech-
nique.

• At the interface node of the gear mesh, six static modes are required,
which are calculated as constraint modes by applying unit displacements
at all six coordinates of this node.
162 6. Analysis of the drive train in a modern wind turbine

1 2 No. Description
1, 2 bearing support pinion

3, 4 bearing support gear

5 6
5 gear mesh model

7 reduced FE model
6
of the pinion

3 4 reduced FE model
7
of the gear

Figure 6.4: Schematic representation of the flexible multibody model of the


helical parallel gear stage.

• For both interface nodes at the bearings, only three static modes are
required, which are calculated as constraint modes by applying unit dis-
placements only at the three translational coordinates of these nodes2 .

The global set of component modes includes consequently a total of twelve


static modes per component. The static modes determine the static deforma-
tion of the components for specific loads at the interface nodes. The CMS tech-
nique allows furthermore to include a set of normal modes for each component,
to represent their dynamic behaviour. The Craig-Bampton reduction tech-
nique, which is applied here, implies the calculation of all fixed-interface nor-
mal modes. Including normal modes is only required for components which
influence the dynamics in the frequency range of interest. This work focusses
on a range up to 1.5 kHz, which is more than sufficient, since 1 kHz can gen-
erally be considered as a maximum for the gear mesh excitation frequencies.
Following the rule of thumb from section 4.6.2, only those normal modes with
a frequency lower than twice this frequency need to be included in the set of
component modes.

Based on this rule, five normal modes are included in the global set of the
pinion. Since all coordinates at the interface nodes belong to the set of interface
DOFs, the inclusion of normal modes requires that six static constraint modes
are included per interface node. Therefore, eighteen static modes are included
2 Both bearings of as well the pinion as the gear are considered as radial and axial supports.

In case one of them is only a radial support, the static mode in axial direction at this interface
becomes redundant.
6.3 Individual gear stages 163

in the reduction of the pinion, instead of only twelve. After orthogonalisation


of this global set, the normal modes correspond to two pairs of bending modes
and one torsional mode for free-free boundary conditions. Figure 6.5 shows
these modes and their frequencies, which are all below 3 kHz. Since the gear
has no normal modes in this frequency range, no fixed-interface normal modes
need to be included for this component. The set of twelve static modes is then
sufficient. The following summary gives an overview of the component modes
included for the pinion and the gear.

1. pinion:

• six static constraint modes at the interface node, where the tooth
contact force acts
• six static constraint modes at both interface nodes, which represent
the points of support at the bearings (twelve modes)
• five fixed-interface normal modes
• TOTAL: 23 component modes

2. gear:

• six static constraint modes at the interface node, where the tooth
contact force acts
• three static constraint modes at both interface nodes, which repre-
sent the points of support at the bearings (six modes)
• no fixed-interface normal modes
• TOTAL: 12 component modes

6.3.1.2 Analysis
Similar to the discussion in section 5.2, a comparison of the results for the var-
ious models yields insight in both the capabilities of the different modelling
approaches and the individual contribution of the properties that are taken into
account.

Static analysis

The individual contribution of the various model parameters is demonstrated


in table 6.1, which gives the torsional stiffness of the gear stage for the differ-
ent models; additionally, the effect of a varying helix angle is included. The
variation in one column indicates the effect of respectively adding the bearing
flexibilities to the torsional model and adding the component flexibilities to the
164 6. Analysis of the drive train in a modern wind turbine

(a) 1st bending mode (×2) at 1.2 kHz. (b) 1st torsional mode at 2.0 kHz.

(c) 2nd bending mode (×2) at 2.9 kHz.

Figure 6.5: Representation of the normal modes after orthogonalisation of the


global component mode set of the pinion, which is considered for the calcu-
lation of the additional DOFs for this body in the flexible multibody model of
the helical parallel gear stage.

β0b (◦ )
0 5 10 15 20
Torsional model 394 391 382 367 347
Rigid multibody model 195 189 175 154 131
Flexible multibody model 152 148 138 123 106

Table 6.1: The equivalent static torsional stiffness (MNm/rad) of the high
speed parallel gear stage for the different modelling approaches in figure 6.2
and a varying helix angle. The stiffness value is related to the side of the gear.

rigid multibody model. Based on these results, it is possible to calculate these


individual contributions as equivalent stiffness values, which can be combined
in series to get the overall torsional stiffness. For example:
• the value 394 MNm/rad is the equivalent stiffness of the gear mesh when
β0b = 0◦ ;
• for this β0b , the value 195 MNm/rad is the equivalent stiffness of the
gear mesh and the bearings, which leads to an equivalent stiffness of
1 1 −1
386 MNm/rad for the bearings only (i.e. ( 195 − 394 ) );
6.3 Individual gear stages 165

• a similar derivation leads to an equivalent stiffness of 689 MNm/rad for


1 1 −1
the components only (i.e. ( 152 − 195 ) ).

These results are summarised in table 6.2, which indicates that the equiva-
lent gear mesh stiffness and bearing stiffness are of the same magnitude. The
components are considerably stiffer in this case. A more convenient way of
comparing the relative magnitudes of the stiffness values is by a representation
of their contribution to the overall flexibility as a ratio in percent, which is in-
cluded in the second part of the table. The table includes also the effect of a
varying helix angle. This variation causes a decrease of the overall torsional
stiffness as indicated already in table 6.1. However, table 6.2 demonstrates
furthermore that this is not only a consequence of the expected effect on the
gear mesh stiffness (cfr. figure 6.2(a)). In fact, the equivalent stiffness values
of both the bearings and the components decrease relatively more than that of
the gear mesh and, vice versa, their relative importance for the overall flex-
ibility increases correspondingly. Especially the contribution of the bearings
increases, going from 39 % to 50 % for an increase of the helix angle of 20◦ .
This is mainly due to the asymmetric bearing stiffness in this gear stage: the
axial stiffness is lower than the radial stiffness and, therefore, its overall effect
on the torsional flexibility increases when the axial loads become larger as a
result of the increase in helix angle.

Dynamic analysis

Table 6.3 compares the eigenfrequencies calculated for the three different mod-
els with a helix angle β0b = 20◦ . The eigenfrequency 1426 Hz calculated with
the torsional model drops to 696 Hz in the rigid multibody model, which is a
clear result of the decrease in stiffness as shown in table 6.1. This indicates
the impact of the bearing flexibilities on the torque dynamics. Furthermore,
other additional modes are found and are given with the main component of
their corresponding mode shape. By adding the components’ flexibilities in
the flexible multibody model, the eigenfrequencies decrease further. However,
the impact of these flexibilities on the overall stiffness, and thus also on the fre-
quencies, is smaller. Only the bending flexibility of the rather long and slender
high speed pinion, causes a considerable decrease of the eigenfrequencies cor-
responding to its y-z-rotation modes (ω4 , ω5 ). Figure 6.6 shows the fourth
mode shape of the gear system with a clear deformation of the pinion. This
analysis shows how the flexible multibody technique makes it possible to eval-
uate also the dynamic effect of the components’ flexibilities, without reducing
them to discrete stiffnesses. Furthermore, the eigenmodes of the components
themselves are also taken into account, which is impossible in a rigid multi-
body model.
166 6. Analysis of the drive train in a modern wind turbine

β0b (◦ )
0 5 10 15 20
gear mesh stiffness 394 391 382 367 347
bearing stiffness 386 366 323 265 210
components’ stiffness 689 682 653 611 555

β0b (◦ )
0 5 10 15 20
gear mesh stiffness 39 % 38 % 36 % 33 % 31 %
bearing stiffness 39 % 40 % 43 % 46 % 50 %
components’ stiffness 22 % 22 % 21 % 20 % 19 %

Table 6.2: above: Equivalent torsional stiffness values (MNm/rad) for the in-
dividual contribution of the gear mesh, the bearings and the components to the
overall torsional stiffness of the high speed parallel gear stage, including the
effect of a varying helix angle. All stiffness values are related to the side of the
gear.
below: The relative importance of the individual contribution to the overall
flexibility.

Z
6 Y
Q
sX
Q

Figure 6.6: The fourth mode shape of the helical parallel gear stage, calculated
with the flexible multibody model (solid: undeformed; wireframe: deformed).
6.3 Individual gear stages 167

No. Torsional Rigid Flexible Corresponding mode shape


model mbm mbm
(Hz) (Hz) (Hz)

1 0 0 0 Rigid-body mode (a)


2 - 409 400 x-Translation pinion (b)
3 - 490 473 x-Translation gear (c)
4 1426 696 510 y-z-Rotation pinion (d)
5 751 510 y-z-Rotation pinion (e)
6 755 609 y-z-Rotation gear (f)
7 764 648 y-z-Rotation gear (g)
8 772 682 y-z-Translation gear (h)
9 784 715 y-z-Translation gear (i)
10 1045 874 y-z-Translation pinion (j)
11 1285 1074 y-z-Translation pinion (k)
12 2011 1642 x-Rotation pinion & gear (l)

Table 6.3: Comparison of the eigenfrequencies of the helical gear pair


(β0b = 20◦ ) calculated with a torsional model, with a rigid multibody model
(mbm) and with a flexible multibody model.

6.3.2 High speed planetary stage


6.3.2.1 Modelling

The high speed planetary gear stage consists of three identical planets with he-
lical teeth. The planet carrier in this stage is a cage type as shown in figure 6.7,
which is supported by one bearing. The second point of support is the sun of
the first stage, which is replaced here by an artificial bearing. Furthermore, the
sun is floating in radial direction near the planets; its support at the gear of the
third stage is also replaced by an artificial bearing. Both the planet carrier and
the sun are free to rotate. The ring wheel is fixed; this fixation is not modelled
as a connection through a torsional spring with its housing (as in the example
of section 5.3), but, instead, its DOFs are removed from the equations of mo-
tion. This section discusses the implementation of three different models for
this planetary stage, similar to the discussion above of the high speed parallel
stage.

The first is a purely torsional model, which includes only one DOF per com-
ponent and the different gear meshes as only deformable parts in the system.
168 6. Analysis of the drive train in a modern wind turbine

(a) View on the low speed side. (b) View on the high speed side.

Figure 6.7: The high speed planetary stage has a cage type planet carrier and
helical gears.

The second is a rigid multibody model and includes in addition extra DOFs per
component and furthermore the bearing flexibilities. Finally, a flexible multi-
body model is an extension of this latter model with additional DOFs for the
planet carrier and the sun to represent their internal deformation. These ex-
tra DOFs are derived from the respective FE models of these components, as
shown in figure 6.8.

The sun model in figure 6.8(a) consists of five-sided and six-sided solid ele-
ments and has only two interface nodes. Its global set of component modes
includes six static modes per interface. Figure 6.9 shows the three normal
modes taken into account for this component. On the other hand, the planet
carrier model in figure 6.8(b) consists of four-sided solid elements3 and has
five interface nodes: three at the planet bearings and two at the planet carrier
supports. The global set of component modes includes six static modes per in-
terface, but no normal modes. Section 4.4 describes how the planet carrier acts
as the reference frame for the planets, which is necessary in the gear contact
formulation. This is included in the reduced FE model of the planet carrier as a
coordinate system fixed to an artificial node on the symmetry axis of the planet
carrier, which moves with the rigid-body motion of this component. The fol-
lowing summary gives an overview of the component modes included for the
sun and the planet carrier and figure 6.10 shows a schematic representation of
the flexible multibody model of the high speed planetary gear stage.
3 The specific name of the four-sided elements in the FE software MSC/Nastran is CTETRA.
6.3 Individual gear stages 169

interface node at the planets










 

interface node at the gear of the third stage

(a) Model of the sun (2194 nodes and 1664 elements).

interface nodes at
  the planet bearings

  

    
 
 
 
 


 
 
   
  
  

interface nodes at
the points of support

(b) Model of the planet carrier (2585 nodes and 10635 elements).

Figure 6.8: The FE models of the sun and the planet carrier of the high speed
planetary stage, which are reduced to a set of component modes using the CMS
technique.
170 6. Analysis of the drive train in a modern wind turbine

1. sun:

• six static constraint modes at the interface node, where the tooth
contact forces acts (three sun-planet gear meshes)
• six static constraint modes at the interface node near the gear of the
third stage
• three fixed-interface normal modes
• TOTAL: 15 component modes

2. planet carrier:

• six static constraint modes at each of the three interface nodes for
the planet bearings (eighteen modes)
• six static constraint modes at both interface nodes, which represent
the points of support at the planet carrier bearings (twelve modes)
• no fixed-interface normal modes
• TOTAL: 30 component modes

(a) 1st bending mode (×2) at 1.4 kHz. (b) 1st torsional mode at 1.9 kHz.

Figure 6.9: Representation of the normal modes after orthogonalisation of the


global component mode set of the sun in the high speed planetary gear stage.
6.3 Individual gear stages 171

No. Description
1, 2 bearing support planet carrier
3 bearing support sun
reduced FE model
4
of planet carrier
5 gear mesh model sun-planet
6 reduced FE model of sun
7 bearing support planet
8 rigid planet body
9 gear mesh model planet-ring
rigid ring wheel body
10 fixed to the gearbox
housing

Figure 6.10: Schematic representation of the flexible multibody model of the


high speed planetary gear stage. For the sake of clarity, only one of the three
planets is included here.

6.3.2.2 Analysis

A static analysis yields an equivalent torsional stiffness of the complete second


gear stage of 346 MNm/rad, related to the low speed side at the planet carrier.
In contrast with the conclusions for the parallel gear stage, the components
contribute here most to the overall flexibility, namely by 70 %. This is a com-
bined result of the torsional flexibility of the planet carrier and the sun. Similar
to the results for the parallel gear stage, the gear mesh and the bearings have
the same impact on the overall flexibility, namely each 15 %.

The individual dynamic analysis of the second gear stage is further limited to
the rigid multibody model only. This simplifies the categorisation of the mode
shapes, which is based on their visualisation, since any modal deformation of
172 6. Analysis of the drive train in a modern wind turbine

a flexible component is excluded this way and cannot complicate the visual
interpretation. This allows furthermore a better definition of a new category of
mode shapes. However, this implies a neglect of the significant flexibilities of
the planet carrier and the sun. These could, however, be included as a torsional
stiffness at the input, respectively output side, which is done in the model of
the complete gearbox in section 6.4.

Rotational mode (m = 1)

− →
− →
− →
− →
− →

R1 R2 R3 R4 R5 R6
0 703 1035 1563 2243 -
Translational mode (m = 2)

− →
− →
− →
− →
− →

T1 T2 T3 T4 T5 T6
131 822 1110 1532 2269 4645
Out-of-plane mode

− →
− →
− →
− →
− →
− →

O1 O2 O3 O4 O5 O6 O7
(m = 2) (m = 2) (m = 2)
90 545 581 3048 3054 3062 3127

Table 6.4: Eigenfrequencies (Hz) of the rigid multibody model of the high
speed helical planetary gear stage. The results are divided in three categories,
based on their corresponding mode shapes.

Table 6.4 shows the eigenfrequencies for the rigid multibody model of the
high speed planetary stage. The categorisation presented in section 5.3 for
a planetary system is used, but an extra category of out-of-plane modes is
introduced, since the out-of-plane motion is not fixed here. The relevance
of the out-of-plane modes is indicated by the fact that they lie in the same
frequency range as the other modes, which could interfere with the range of
e.g. the gear mesh excitations. Furthermore, these excitations include out-of-
plane forces because of the helical teeth, which enables energy input in the
out-of-plane modes. Figure 6.11 shows a mode shape corresponding to an
out-of-plane mode. Moreover, since all DOFs of the ring wheel are removed
from the equations of motion, only five rotational modes are found in this
planetary system instead of six, as in the benchmark example of section 5.3.
Although the frequency range of interest is limited to 1.5 kHz, all rotational
and translational modes are included (even those outside this range), for the
sake of completeness. In contrast, three out-of-plane modes above 5 kHz are
omitted.
6.3 Individual gear stages 173

Figure 6.11: Example of an out-of-plane mode (545 Hz), calculated for the
rigid multibody model of the high speed planetary stage. Note that only one
side of the cage type planet carrier is shown. The arrows indicate the respective
out-of-plane displacement of two planets.

6.3.3 Low speed planetary stage


The first gear stage in the present example is similar to the second stage, but has
spur gears instead of helical gears. As a result, it can be modelled and analysed
analogously to the second stage. Therefore, the discussion on the modelling of
this stage is not further elaborated here; only the results are summarised. The
static analysis is furthermore limited to a comparison of the relative contribu-
tion of the gear mesh and the bearing flexibilities. This analysis demonstrates
again that they are of an equal magnitude. In addition, the torsional deforma-
tion of the planet carrier and the sun are included as a discrete torsional spring
at the input, respectively output side of this stage in the model of the complete
drive train.

Table 6.5 presents the eigenfrequencies for the rigid multibody model of the
first planetary stage. Since this stage has spur gears and, consequently, only
in-plane forces consist, no out-of-plane modes are included in this table. How-
ever, both the planet carrier and the sun can experience out-of-plane loads
through their coupling with other components in the drive train and, therefore,
the corresponding modes are included in the analysis of the complete drive
train. As a result of excluding out-of-plane motions here, only rotational and
translational modes are identified for the first planetary stage.
174 6. Analysis of the drive train in a modern wind turbine

Rotational mode (m = 1)

− →
− →
− →
− →
− →

R1 R2 R3 R4 R5 R6
0 429 693 1024 1482 -
Translational mode (m = 2)

− →
− →
− →
− →
− →

T1 T2 T3 T4 T5 T6
501 747 957 1032 1439 7287

Table 6.5: Eigenfrequencies (Hz) of the rigid multibody model of the low
speed planetary gear stage, which are categorised as rotational and transla-
tional modes.

6.4 Complete gearbox

This section discusses the analysis of the gearbox shown in figure 6.12. It
consists of the three gear systems as analysed individually above. The artificial
boundaries of the individual stages, introduced above to replace the common
supports of two gear stages, are now removed again. The actual supports and
coupling between the stages are therefore implemented by appropriate stiffness
matrices:

• the sun of the first stage and the planet carrier of the second stage are
coupled and rest on a common support bearing;

• the sun of the second stage is floating near the planets and is supported
at its other end in connection with the gear wheel of the third stage.

Firstly, a static analysis demonstrates how the equivalent torsional stiffness is


determined by the different individual stages. Subsequently, a dynamic anal-
ysis, based on a rigid multibody approach, focusses on a comparison of the
results for the complete gearbox with those for the individual stages. This de-
scribes how the boundaries of the individual stages in the gearbox influence
their internal behaviour. In addition, a final analysis determines the effect of
adding inertia to the input and output of the complete gearbox, as a first step
towards integration in a wind turbine drive train.
6.4 Complete gearbox 175

Figure 6.12: The complete gearbox consists of a planetary spur gear stage, a
planetary helical gear stage and a parallel helical gear stage. The bearings are
not included in this figure.

6.4.1 Static analysis

The static analyses in section 6.3.1 and 6.3.2 resulted in an individual torsional
stiffness of the parallel and second planetary gear stage of 106 MNm/rad and
346 MNm/rad respectively, both referred to their low speed side. The torsional
stiffness of the first planetary stage, including the coupling with the second
stage, equals 1.8 GNm/rad referred to the low speed side of the gearbox. Fi-
nally, the equivalent torsional stiffness of the complete gearbox referred to its
input side equals 1.5 GNm/rad, which is a result of putting the stiffness values
of the individual stages - referred to the low speed side of the gearbox - in
series. Referring the stiffness values to the input side implies that, although
the stiffness value for the first stage is considerably larger than the values for
the second and third stage, it still determines the overall flexibility for more
than 80 %. After all, referring the individual stiffness values of the second and
third stage to the low speed side of the first stage, requires a multiplication
with the square of the first stage gear ratio and, in addition, with the square of
the second stage gear ratio for the stiffness value of the third stage only. Since
the stiffness values decrease less than proportional to the square of the gear
ratio, going from the first to the third stage, their contribution to the overall
equivalent torsional stiffness value decreases from the first to the third stage.
176 6. Analysis of the drive train in a modern wind turbine

6.4.2 Dynamic analysis

Section 5.4 describes the analysis of a combination of a planetary and a parallel


gear stage, which are coupled only in torsion. This example demonstrated that
only few eigenfrequencies calculated for the individual stages are influenced
by coupling both stages. This is no longer completely valid for the results of
the complete gearbox analysed in this section, because the coupling between
the different stages is not only in torsion and includes a common support for
separate components. This has a major impact on the behaviour of the individ-
ual stages and leads to a larger amount of global modes, which are defined as
modes that cannot be attributed to a single gear stage. Table 6.6 summarises
all results with, where possible, a categorisation according to the location of
the nodes in the corresponding mode shapes and their type. A comparison of
these eigenfrequencies with those calculated for the individual stages yields
the following conclusions:

• The eigenfrequencies found for the parallel gear stage are hardly influ-
enced by coupling it in the gearbox, except for the two y-z-translation
modes of the gear wheel (h & i) - (cfr. table 6.3).

• The rigid-body rotational modes in both planetary stages disappear as a


logic consequence of coupling the different stages. On the other hand,
the four non rigid-body rotational modes can still be identified in the
model of the complete gearbox. In both stages, the fourth rotational
mode is identified in the individual stage and in the complete gearbox
at the same frequency. For the high speed planetary system, the other
rotational modes lie all lower in frequency in the complete gearbox. For
the low speed planetary system, on the contrary, two of the four modes
increase in frequency, as a result of removing the artificial boundaries
from the individual stage - (cfr. table 6.4 and 6.5).

• Four out of six translational modes in the 1st and 2nd stage are also
identified in the complete gearbox. Compared with the eigenfrequen-
cies calculated for the individual stages, they lie all lower in frequency,
except for two modes in the high speed planetary system - (cfr. table 6.4
and 6.5).

• The out-of-plane modes in the high speed planetary stage remain quasi
constant in frequency in the complete gearbox, which indicates that the
influence of the boundaries on these modes is negligible - (cfr. table 6.4).
6.4 Complete gearbox 177

No. High speed planetary stage


Rotational Translational Out-of-plane
mode mode mode
No. Global mode (m=1) (m=2)
1 0 2 81
5 147 3 140
34,35 1041 11 342
43 1518 17,18 542
44 1527 19 565
50 2390 20 625
51 2401 29 854
52 2560 36 1093
39 1152
45 1558
47 1563
53 3048
54,55 3054
56 3066
57,58 3123
No. Low speed planetary stage
Rotational Translational Out-of-plane
No. Parallel mode mode mode
stage (m=1) (m=2)
12 408 (b) 6 224
13 441 (c) 8,9,10 227
22 701 (d) 14 507
24 750 (e) 16 528
25,26 751 (f,g) 23 742
32 1018 (j) 27 785
40 1285 (k) 30 1005
49 2005 (l) 33 1026
38 1100
41,42 1440
48 1641

Table 6.6: Eigenfrequencies (Hz) of the rigid multibody model of the complete
gearbox with free boundaries, which are classified according to the location of
the nodes in the corresponding mode shapes and their type. The characters
behind the eigenfrequencies for the parallel stage refer to the results in the
individual stage (cfr. table 6.3).
178 6. Analysis of the drive train in a modern wind turbine

The results from table 6.6 are valid for a gearbox with free boundaries, i.e.
not connected to any inertia. However, the position of the gearbox in a wind
turbine is between the rotor and the generator, which are components with a
considerable inertia. When the rotor inertia is included as a fixation of the
gearbox input and the generator as an infinite torsional inertia at the output,
the eigenfrequencies change to the results presented in table 6.7. This artificial
situation is the outer limit of adding inertia to the input and output shafts and
indicates the following:

• The x-translation mode of the pinion in the parallel stage decreases in


frequency from 408 Hz to 264 Hz (b). Furthermore, one of its y-z-
rotational modes (d) and one of its y-z-translation modes (j) decrease
from 701 Hz to 510 Hz and from 1043 Hz to 867 Hz respectively. The
x-rotation mode of the pinion and the gear (l) decreases from 2005 Hz
to 1728 Hz.

• Only two of the internal mode shapes and corresponding frequencies of


the high speed planetary stage are influenced by the outer boundaries of
the gearbox. These are the first and second rotational mode shape.

• The fixation of the input implies a fixation of the low speed planet car-
rier. As a result, most of its internal frequencies decrease and, further-
more, the out-of-plane modes corresponding to the motion of the planet
carrier disappear.

Since the analysed situations are the outer limits with respect to the addition of
inertias at the boundaries, the comparison of the respective results indicates the
outer limits for the final results of the gearbox in the wind turbine. Section 6.5
discusses the integration of the gearbox in the drive train of the wind turbine
more elaborately.
6.4 Complete gearbox 179

No. High speed planetary stage


Rotational Translational Out-of-plane
mode mode mode
No. Global mode (m=1) (m=2)
1 0 2 81
30,31 1030 3 140
37 1509 10 264
38 1527 16 535
44 2390 17 565
45 2396 18 625
46 2560 26 838
32 1092
35 1152
39 1559
41 1564
47 3048
48,49 3054
50 3066
51,52 3123
No. Low speed planetary stage
Rotational Translational Out-of-plane
No. Parallel mode mode mode
stage (m=1) (m=2)
10 264 (b) 5,6,7 227
12 437 (c) 8 244
13 510 (d) 11 364
23 750 (e) 14 525
24,25 751 (f,g) 20 650
27 867 (j) 22 731
36 1285 (k) 28 1001
43 1728 (l) 30 1011
42 1640

Table 6.7: Eigenfrequencies (Hz) of the rigid multibody model of the com-
plete gearbox with a fixed input and an infinite torsional inertia at the output,
which are classified according to the location of the nodes in the correspond-
ing mode shapes and their type. The characters behind the eigenfrequencies
for the parallel stage refer to the results in the individual stage (cfr. table 6.3).
180 6. Analysis of the drive train in a modern wind turbine

6.5 Drive train integrated in the wind turbine


This section describes the analysis of the drive train, when it is integrated in a
model of the complete wind turbine. The first part describes the extension of
the gearbox model with the extra components at the generator side only. Then,
the second part extends this model further with the consideration of the rotor
and the tower as additional components.

6.5.1 Model without rotor and tower


The completion of the drive train model at the generator side requires the ad-
dition of the brake disk, the flexible coupling and the generator itself (cfr. fig-
ure 6.12). The flexible coupling is considered as very flexible in all directions,
except for its torsional deformation and, therefore, it is modelled as a torsional
spring. Its inertia is included in the generator, for which only one rotational
DOF is considered. The generator is furthermore free to rotate. No boundaries
are considered here, which corresponds to the free behaviour of a DFIG, as
described in section 2.3.3.1. The brake disk is modelled as an additional body
clamped on the pinion.

Table 6.8 shows the eigenfrequencies calculated for this model. A comparison
with the results for the gearbox individually (cfr. table 6.6), yields the follow-
ing conclusions:

• An additional global mode (mode no. 2) is identified at 62 Hz and cor-


responds to the torsional deformation of the flexible coupling.

• The local modes at the parallel stage decrease considerably in frequency


because of the additional inertia of the brake disk, which is mounted on
the pinion.

• All other eigenmodes of the gearbox change only slightly in frequency,


except for one rotational mode of the high speed planetary stage, which
drops from 342 Hz to 307 Hz.
6.5 Drive train integrated in the wind turbine 181

No. High speed planetary stage


Rotational Translational Out-of-plane
mode mode mode
No. Global mode (m=1) (m=2)
1 0 3 81
2 62 4 140
6 145 12 307
36,37 1040 20,21 542
44 1506 23 568
45 1527 24 633
51 2390 31 864
52 2396 37 1093
53 2560 40 1151
46 1558
48 1563
54 3048
55,56 3054
57 3066
58,59 3123
No. Low speed planetary stage
Rotational Translational Out-of-plane
No. Parallel mode mode mode
stage (m=1) (m=2)
13 349 (b) 6 224
14 406 (d) 8,9,10 227
15 409 (e) 14 507
16 430 (c) 16 527
22 562 (f) 23 742
27 750 (g) 27 785
30 832 (j) 30 1008
41 1238 (k) 33 1026
50 1731 (l) 38 1100
41,42 1440
48 1641
Table 6.8: Eigenfrequencies (Hz) of the rigid multibody model of the drive
train, including the gearbox, the brake disk, the generator, but no rotor. The
results are classified according to the location of the nodes in the corresponding
mode shapes and their type. The characters behind the eigenfrequencies for the
parallel stage refer to the results in the individual stage (cfr. table 6.3).
182 6. Analysis of the drive train in a modern wind turbine

6.5.2 Model including rotor and tower

Section 3.3.3 describes that the structural model of a wind turbine in the tra-
ditional design codes has about twenty-four DOFs. Five of these DOFs repre-
sent tower modes and eighteen modes represent blade modes. This indicates
the importance of the structural properties of the rotor and the tower for the
calculation of a wind turbine’s response in a frequency range up to 10 Hz,
which is typically the limit in the traditional simulations. Shifting this limit
to 1.5 kHz with the purpose of simulating detailed drive train loads, requires -
in theory - the consideration of all rotor and tower modes up to approximately
3 kHz. However, since this is an impracticable task, the analyses are further
split into a low frequency range [ 0 - 10 Hz] and a high frequency range [ 10 -
1500 Hz]. Both analyses start from the the multibody model of the drive train
as described above in section 6.5.1.

1. [ 0 - 10 Hz]
The drive train model is further elaborated with two additional bodies,
representing the rotor and the tower. Each body has six rigid-body DOFs
and, furthermore, an extra set of DOFs to represent the internal deforma-
tions of these components. The latter DOFs of the so-called flexible bod-
ies are derived from an FE model of the rotor and the tower respectively,
using the CMS technique. The set of component modes is composed
such that it accurately represents the dynamic behaviour of the individ-
ual components up to 10 Hz. This requires a consideration of four pairs
of normal bending modes for each blade and the first ten normal modes
of the tower. This flexible multibody model of the wind turbine is con-
sidered to be similar to the structural model in a traditional design code.
Therefore, its analysis is discussed separately in appendix A.

2. [ 10 - 1500 Hz]
Since both the rotor and the tower have more than ten modes below
this frequency range, it is assumed that they will act as a large inertia
with respect to an excitation at higher frequencies. Therefore, instead
of adding two flexible bodies for the rotor and the tower, only one rigid
body with six DOFs is included to represent the large inertia of the rotor.
The tower is considered as a rigid ground, which supports the gearbox
and the generator. This implies no need for an additional body to rep-
resent the tower. The remainder of this section describes three types of
analysis for this model: a normal modes analysis, a frequency response
analysis and a transient load simulation.
6.5 Drive train integrated in the wind turbine 183

6.5.2.1 Normal modes analysis


Table 6.9 shows the eigenfrequencies calculated for the model of the wind
turbine. A comparison with the results calculated for the gearbox individually,
including a fixation of the rotor side and an infinite torsional inertia at the
generator side (cfr. table 6.7), yields the following conclusions:
• An additional global mode (mode no. 2) is identified at 68 Hz and cor-
responds to the torsional deformation of the flexible coupling.
• The global mode at 5.6 Hz corresponds to a torsional deformation of the
drive train or a so-called drive train mode. However, as demonstrated in
appendix A, it is inaccurate to calculate such modes without the consid-
eration of the rotor flexibility. Therefore, this mode is not considered as
a relevant result.
• The double mode at 32 Hz corresponds to a tilting motion of the rotor
in the main bearing. Again, an omission of the rotor flexibility for the
calculation of this mode is inaccurate, since the flapwise rotor modes
have a determining influence. Consequently, also this double mode is
not considered as a relevant result.
• The eigenmodes identified in the parallel stage correspond exactly to
those presented in table 6.8 for the model without rotor. The decrease
in frequency, compared to the results from table 6.7, is a result of the
additional inertia of the brake disk at the pinion side.
• The eigenmodes in the high speed planetary stage are hardly influenced
by the gearbox boundaries: only one rotational mode increases in fre-
quency from 264 Hz to 302 Hz. This is a result of the lower inertia at
the generator side in the actual model.
• One rotational mode of the low speed planetary stage increases in fre-
quency from 364 Hz to 513 Hz as a result of the lower inertia at the
generator side. Two translational modes decrease in frequency and the
out-of-plane modes in this stage (at 227 Hz and 443 Hz) are further not
considered to be relevant.
Summarising these conclusions, it may be stated that a combination of a fixed-
input gearbox model with a correct model description of the components at the
generator side is sufficient to determine the internal eigenmodes of the drive
train in the wind turbine in the frequency range above 10 Hz. This statement
is valid under the assumption that in this frequency range the rotor can be
considered as a large inertia and the tower as a rigid supporting structure. Fur-
thermore, in this particular drive train, the first relevant internal eigenmode is
identified at 68 Hz. Therefore, it is sufficient to focus in the frequency response
analyses on a frequency range starting from 50 Hz.
184 6. Analysis of the drive train in a modern wind turbine

No. High speed planetary stage


Rotational Translational Out-of-plane
mode mode mode
No. Global mode (m=1) (m=2)
1 0 6 81
2 5.6 7 140
3,4 32 14 302
5 68 23,24 527
38,39 1033 25 568
44 1506 27 630
45 1527 34 864
51 2390 40 1093
52 2396 42 1151
53 2560 46 1558
48 1564
54 3048
55,56 3054
57 3066
58,59 3123
No. Low speed planetary stage
Rotational Translational Out-of-plane
No. Parallel mode mode mode
stage (m=1) (m=2)
17 346 (b) 9 205
18 406 (d) 11,12,13 227
19 409 (e) 15 305
21 430 (c) 20 423
25 562 (f) 22 513
32 750 (g) 29 660
33 832 (j) 31 731
43 1238 (k) 35 1003
50 1731 (l) 37 1011
49 1640
Table 6.9: Eigenfrequencies (Hz) of the rigid multibody model of the drive
train, including the gearbox, the brake disk and the generator. The rotor is
added as a large inertia. The results are classified according to the location
of the nodes in the corresponding mode shapes and their type. The characters
behind the eigenfrequencies for the parallel stage refer to the results in the
individual stage (cfr. table 6.3).
6.5 Drive train integrated in the wind turbine 185

6.5.2.2 Frequency response analysis


This section describes how the results of the normal modes analysis should be
interpreted in order to avoid resonance and, furthermore, it demonstrates how
the drive train loads can be simulated for a sinusoidal load excitation. Both
analyses imply a calculation of the frequency response.

In order to avoid resonance, it is important to know which eigenmodes can


considerably influence the loads in the drive train. This means that in one way
or another there should be a coupling between these modes and a source of
excitation. Various excitation sources might exist, as described in section 3.4.
The coupling with such an excitation can be estimated from a detailed inter-
pretation of the mode shapes, although, this is not straightforward. An easier
method is to calculate a frequency response function (FRF) between the exci-
tation source and a specific load in the drive train and check which eigenmodes
lead to amplified load levels. This is demonstrated for two excitations:

1. the load in the generator,

2. the gear mesh excitation in the high speed planetary stage.

Both excitations are considered here as a variation of an input torque, at the


generator and the high speed sun respectively. The input torque is a multisine
with a minimal frequency of 50 Hz, according to the statement above. The
upper frequency limit is 2.0 kHz, which guarantees a proper excitation at all
frequencies in the range from 50 to 1.5 kHz. Figure 6.13 shows the spec-
trum of the multisine signal. The DADS solver calculates the output loads for
this signal using a time integration procedure as explained in appendix E. A
maximum time step of 0.0001 second is chosen. The numerical integration is
only possible with a certain amount of damping in the drive train, which is in-
cluded by adding a viscous damper element to the DADS model. The damping
defines the amplitudes of the response, especially at the eigenfrequencies. Sec-
tion 4.3.2.4 describes the difficulties in quantifying accurate damping values.
Since this determination is not within the scope of this analysis, the responses
are only considered qualitatively.

Excitation in the generator torque

Figure 6.14 shows the FRFs calculated for the torque excitation at the generator
side. The torque on the high speed pinion and on the two suns are the respec-
tive outputs in these calculations. The maximum frequency in these plots is
only 1.0 kHz, since no relevant amplified torque levels are identified at higher
frequencies. The analysis of these results, leads to the following conclusions:
186 6. Analysis of the drive train in a modern wind turbine

PSD [dB/Hz]

Frequency [Hz]

Figure 6.13: Power spectrum of the torque excitation signal used in the FRF
calculations.

1. The global mode at 68 Hz, corresponding to the deformation of the flex-


ible coupling is clearly excited and identified as a dominant frequency
in the response at all gear stages.
2. The excitation of the first rotational mode in the high speed planetary
stage (at 302 Hz) is also visible in the response at all gear stages.
3. For the high speed pinion, the response above 300 Hz is mainly deter-
mined by its local modes (cfr. figure 6.14(a)). Note that e.g. the axial
translation of the gear with a resonance frequency at 430 Hz can be ex-
cited by a generator torque variation. Furthermore, a qualitative compar-
ison of the load amplification for this particular frequency indicates that
it is as relevant as the other local modes in this stage. In other words, it
is no longer possible to identify for this stage one single so-called “tor-
sional mode” which is dominant with respect to the torque dynamics.
This emphasises the importance of the rigid multibody models next to
the purely torsional models. After all, even when the latter model is
tuned to simulate one of the local modes accurately, it still includes an
insufficient number of DOFs to simulate the torque response completely.
4. For the high speed planetary stage, the local translational mode at 140 Hz
can lead to a considerable torque amplification (cfr. figure 6.14(b)). The
response above 300 Hz is also mainly determined by the local modes in
the parallel stage.
5. Additionally, in the torque response for the low speed planetary stage,
the local translational mode at 205 Hz is identified, which influence is
hardly transferred to the other gear stages (cfr. figure 6.14(c)).
6.5 Drive train integrated in the wind turbine 187

PSD [dB/Hz]

Frequency [Hz]
(a) FRF from the generator torque to the torque at the high speed pinion.
PSD [dB/Hz]

Frequency [Hz]
(b) FRF from the generator torque to the torque at the high speed sun.
PSD [dB/Hz]

Frequency [Hz]
(c) FRF from the generator torque to the torque at the low speed sun.

Figure 6.14: Response calculation for a sinusoidal excitation of the drive train
applied as a generator torque variation.
188 6. Analysis of the drive train in a modern wind turbine

Gear mesh excitation in the high speed planetary stage

Figure 6.15 shows the FRFs calculated for a torque excitation of the sun in the
high speed planetary stage. Again, the torque on the high speed pinion and
on the two suns are the respective outputs in these calculations. Note that the
FRF to the torque at the high speed sun is a direct FRF. For the same reason
as described above, the maximum frequency in these plots is only 1.0 kHz. A
detailed investigation of these results, yields the following insight:

1. The global mode at 68 Hz does still contribute to the torque response in


the different gear stages, albeit considerably less than for the excitation
at the generator side.

2. For the high speed pinion, mainly the local modes at 430 Hz and at
562 Hz are dominant in the torque response (cfr. figure 6.15(a)).

3. The direct FRF to the torque on the sun of the high speed planetary stage,
indicates that mainly the local modes in the parallel stage, at 346 Hz and
430 Hz respectively, can lead to a considerable torque amplification (cfr.
figure 6.15(b)). This indicates the importance of analysing the drive train
as a whole. The analysis of an individual gear stage yields insight in its
local modes, but it does not permit to determine the mutual interaction
between dynamic loads in different stages.

4. The torque spectrum at the low speed sun is dominated by one local
mode in the parallel stage (346 Hz), one translational mode in the low
speed planetary stage (305 Hz) and by six local modes in the range [ 500
- 700 Hz].
6.5 Drive train integrated in the wind turbine 189

PSD [dB/Hz]

Frequency [Hz]
(a) FRF to the torque at the high speed pinion.
PSD [dB/Hz]

Frequency [Hz]
(b) FRF to the torque at the high speed sun.
PSD [dB/Hz]

Frequency [Hz]
(c) FRF to the torque at the low speed sun.

Figure 6.15: Response calculation for a gear mesh excitation in the high speed
planetary stage, which is applied as a sinusoidal torque excitation of the high
speed sun.
190 6. Analysis of the drive train in a modern wind turbine

The excitation signal used in this analysis has a broadband spectrum in order
to get an overall idea about which eigenmodes can lead to amplified torque
levels. However, for a particular speed of the drive train, the gear mesh ex-
citation frequency in the high speed planetary stage is exactly known. This
permits to determine whether it coincides with an important eigenfrequency.
Figure 6.16 demonstrates this procedure. It is a Campbell diagram, which indi-
cates how the gear mesh frequency varies with the rotational speed of the rotor
in the wind turbine. In addition to the actual gear mesh frequency, its first and
second harmonic are also plotted as excitation frequencies. The eigenfrequen-
cies included in this figure are the horizontal lines, which correspond to the
dominating peaks from the direct FRF in figure 6.15(b). These frequencies are
considered as the only important frequencies for this excitation with respect
to possible torque amplifications in this gear stage. The intersection of lines
indicate possible resonances. For a fixed-speed wind turbine the focus can
be limited to a single speed. However, modern variable-speed wind turbines
require the consideration of a certain speed range. For the present example,
two cursors indicate such a speed range for the wind turbine rotor from 10 to
20 RPM. In this range, the following intersections are found:

1. the gear mesh frequency intersects the eigenfrequency at 140 Hz;

2. the first harmonic intersects the eigenfrequency at 302 Hz;

3. the second harmonic intersects four eigenfrequencies above 300 Hz.

The results from this analysis are valuable input for assessing whether or not a
drive train resonance can occur as a result of this excitation. The same analysis
can be performed for other excitations and with the focus on other loads. It is
obvious that avoiding all intersections between excitations and eigenfrequen-
cies in a quite broad speed range is impossible. However, keeping in mind
that the actual gear mesh frequencies are usually more important than their
harmonics, the insights from the calculated FRFs yield already valuable in-
formation for the evaluation of a drive train design in order to avoid severe
resonances. To gain more experience in this evaluation and, more generally, to
gain further confidence in the present analysis techniques and their results, it
is recommended to perform sufficient experimental validation measurements
(cfr. chapter 7).

A more detailed numerical investigation of the intersections in the Campbell


diagram at a particular frequency, requires a prediction of the amplified load
levels. This simulation requires an accurate consideration of the drive train
damping values and an appropriate description of the excitation signal, which
represents the transmission error in the gear mesh (cfr. section 4.3.2.1). Since
6.5 Drive train integrated in the wind turbine 191

this is not within the scope of the present dissertation, the load simulation at
resonance is not further elaborated.
Eigenfrequencies [Hz]

Rotor speed [RPM]

Figure 6.16: Campbell diagram which indicates the position of the gear mesh
frequency and its harmonics (inclined lines) of the second gear stage for a
varying rotor speed, in comparison to the dominant eigenfrequencies (hori-
zontal lines) from the direct torque response function in figure 6.15(b). The
cursors (vertical lines) indicate the speed range during operation.

6.5.2.3 Simulation of a transient load case


This section demonstrates how the load simulation with a detailed drive train
model yields the desired insight in its behaviour and gives much more and
more accurate information than obtained with the simulations in the traditional
wind turbine design codes. For the present demonstration, the simulation of
a transient load case is used. An accurate description of the external forces
during a transient load case requires a complete model of the wind turbine
system, including a model for the generator, for the aerodynamics, . . . (cfr.
section 3.3.1). Since these models are not available, an assumption is made
here for the load excitation. This is sufficient for the present illustration. An
integration of the presented drive train model in a traditional wind turbine de-
sign code can help in further refining the load simulations.
192 6. Analysis of the drive train in a modern wind turbine

The simulated transient load case includes a sudden torque variation at the gen-
erator with a high amplitude. This phenomenon can have various causes, such
as disturbances in the electrical grid as described by Soens et al [198] and Se-
man et al [189, 190]: e.g. frequency disturbances, a voltage dip or swell and a
network short circuit. In the present example, the torque variation occurs dur-
ing a start-up of the wind turbine. This mean that the generator torque would
normally be increasing as shown in figure 6.17(a). Firstly, the simulation is
done for this reference signal, i.e. without the sudden torque peak. The slope
of this signal equals 1 kNm/second and the time series has a length of 1 sec-
ond. Subsequently, the torque variation is added at t = 0.5 s, which is visible in
figure 6.17(b). The shape and the duration of the torque variation may largely
differ for various grid disturbances and is furthermore highly dependent on the
type of generator. Here, a damped sinusoidal variation with a frequency of
20 Hz is considered; it has a duration of two periods (100 ms) and a maximum
amplitude of 9.1 kNm. This example is based on the description in [189] of
the generator torque variation during a network short circuit in a DFIG with an
over-current protection system.

(a) 1st load case: a normal start-up (reference (b) 2nd load case: a grid disturbance during
signal) start-up

Figure 6.17: Generator torque used in the simulations of two transient load
cases.

Figure 6.18 shows a comparison of the simulations calculated for the two load
cases. In this example, the focus is put on:

1. the level of the torque, which acts on the pinion of the high speed stage

2. the rotational acceleration of the pinion in its bearings

3. the axial displacement of the pinion in its bearings


6.5 Drive train integrated in the wind turbine 193

(a) Torque at the pinion of the high speed stage

(b) Rotational acceleration of the pinion (high speed stage) in its bearings

(c) Axial displacement of the pinion (high speed stage) in its bearings
(0µm corresponds to no load)

Figure 6.18: Comparison of the results calculated for two transient load cases
(dashed: normal start-up; solid: grid disturbance).
194 6. Analysis of the drive train in a modern wind turbine

The comparison of the two load cases in figure 6.18 yields the following con-
clusions:

• The sudden torque peak in the generator torque, as a result of the grid
disturbance, causes a torque peak at the pinion. The level of this latter
peak equals 2.5 kNm, which is about 3.5 times lower than the level of
the torque peak in the generator. Further experimental validation of the
numerical models and a proper consideration of the damping in the drive
train is required to assess the accuracy of this absolute level. This level
depends furthermore on the type of coupling used in the drive train, as
explained below.
The sudden impact in the drive train excites moreover the 1st drive train
mode of the wind turbine. As a result, various torque reversals occur
during this start-up. This may lead to backlashing in the bearings (cfr.
below), which should be investigated for the bearing design. Note that a
proper simulation of the 1st drive train mode requires the consideration
of the rotor and the tower flexibility as described in appendix A.

• The grid disturbance and resulting torque variation cause the pinion to
accelerate rapidly in its bearings. The acceleration peak level in this ex-
ample is about 30 times higher than during the normal start-up. This
should also be considered with care in the design of the bearings as well
as the negative acceleration, which follows rapidly after the positive ac-
celeration peak. In the resulting variation of the acceleration, the eigen-
mode at 68 Hz is clearly visible, which corresponds to the deformation
of the flexible coupling.

• Since the investigated model includes linear bearing models as presented


in section 4.5, the axial displacement of the pinion follows the torque
variation. This implies that the displacement peaks when the torque
peak occurs. Subsequently, it becomes negative. Under the assumption
that the drive train is at no load conditions for a displacement value of
0 µm the change in sign for the displacement corresponds to passing
the clearance at the pinion. The simulation of this so-called backlashing
may be further improved by using a non-linear bearing model, including
clearance.

The presented analyses are further elaborated by investigating the influence of


the stiffness of the flexible coupling on the level of the torque and the accel-
eration at the pinion. This demonstrates how the simulation approach in this
dissertation can be used as an effective tool to assess the influence of design
changes. Figure 6.19 shows the simulated torque and acceleration signals for
the original coupling (A), for a coupling with a stiffness value of 10% (B) and
6.5 Drive train integrated in the wind turbine 195

of 1000% (C) of the original. The absolute stiffness values of the three cou-
plings are shown in table 6.10, including the dimensions of a hypothetical steel
shaft which has this stiffness.

Coupling Stiffness Length Diameter


[MNm/rad] [m] [mm]
A 1.4 0.5 97
B 0.14 1 65
C 14 0.3 155

Table 6.10: Stiffness values for the three flexible couplings used in the transient
simulation. The dimensions (length and diameter) of a hypothetical steel shaft
with a corresponding stiffness are included.

A comparison of the simulations for the three couplings yields the following
conclusions.

• The level of the torque peak at the pinion decreases when a coupling
with a lower stiffness (B) is used. In the present example, the peak is
1.6 times lower than for the original coupling (A). On the other hand,
using a coupling with a higher stiffness (C), yields a higher torque peak.
In the present example, the maximum torque level is 5% higher.

• The maximum level of the rotational acceleration of the pinion in its


bearings is for both additional couplings lower than for the original cou-
pling.

6.5.3 Features of the new simulation approach


Section 3.4.2 describes the main objective in this dissertation and enumerates
a list of four research objectives. This section repeats these objectives with
a clear description of how they are achieved, using the presented multibody
modelling approach.

1. OBJECTIVE:
Predict the dynamic loads on all drive train components:

• prediction of the internal eigenfrequencies in order to avoid reso-


nance in the drive train
• an accurate simulation of the response for given excitations
196 6. Analysis of the drive train in a modern wind turbine

(a) Torque at the pinion of the high speed stage

(b) Rotational acceleration of the pinion (high speed stage) in its bearings

Figure 6.19: Comparison of the simulations for three different couplings:


solid: original coupling (A); dashed: lower stiffness (B); dotted: higher stiff-
ness (C).

ACHIEVEMENT:
• The use of three different types of multibody models is demon-
strated for the prediction of eigenfrequencies in a drive train. The
rigid multibody modelling approach is judged to be the most effi-
cient technique to perform this analysis. Section 6.5.2.1 describes
such an analysis for a model of a complete wind turbine. In addi-
tion, section 6.5.2.2 demonstrates the use of a frequency response
calculation in order to avoid resonance in the drive train of this
wind turbine. The Campbell diagram is a useful tool in this analy-
sis.
• Section 6.5.2.2 demonstrates how the loads in the drive train of a
wind turbine are simulated for a known sinusoidal load excitation.
6.5 Drive train integrated in the wind turbine 197

Again, the rigid multibody modelling approach is judged to be the


most efficient technique for this analysis, except for the determi-
nation of local stresses in drive train components (cfr. objective
no. 3).

2. OBJECTIVE:
Identify harmful transient phenomena.
ACHIEVEMENT:

• Section 6.5.2.3 describes how the drive train behaviour during two
transient load cases is analysed. This demonstrates how the new
simulation method yields detailed insight in the torque variations
in the drive train, the acceleration levels of bearings and the mo-
tion of all drive train components. This is valuable information for
the identification of harmful transient phenomena. In addition, the
study shows how a sensitivity analysis is used to assess the influ-
ence of particular design changes.
• In order to improve the present simulations, it is recommended:
(a) to further investigate the influence of (1) including a flexible
rotor and a flexible tower and of (2) considering damping and
non-linearities in the drive train, on the results of a transient
simulation
(b) to apply more realistic external loads on the drive train model,
which is possible when the drive train model is integrated in a
traditional wind turbine design code

3. OBJECTIVE:
Determine the level and variation of local loads and stresses.
ACHIEVEMENT:

• Section 4.6 describes how a combination of the CMS technique


and a multibody formulation yields a flexible multibody model.
Section 6.3 demonstrates the application of this method for (1) the
determination of the torsional stiffness of a parallel and a plane-
tary gear stage and for (2) a normal modes analysis of the former
stage. Additionally, appendix E describes (3) a load simulation for
a flexible multibody model. This example indicates that the com-
putational time is considerably larger than for an identical simu-
lation with a rigid multibody model. This consequence of adding
“flexible DOFs” should be weighed with the advantage of using a
flexible multibody model in a load simulation. Such a simulation
yields additional information about the level and variation of local
198 6. Analysis of the drive train in a modern wind turbine

stresses in drive train components. The calculation of these stresses


is not demonstrated in this dissertation. Nevertheless, an overview
is given of the possibilities of the flexible multibody simulation.
– The present flexible MBS formulation permits to simulate the
stresses in all drive train components, except for the local
stresses at the tooth contact and in the bearings. This latter
limitation is a result of the representation of the tooth contact
and the bearing as a linear spring. The accuracy of the stress
distribution in the drive train components depends a.o. on the
number of component modes which is included in the analy-
sis. The simulated stress time series can furthermore be used
directly in fatigue calculations for the structural components.
– The flexible MBS method permits to include the deformation
of a planet carrier in a planetary stage in a much more real-
istic way than is possible in a rigid multibody model. This
makes it possible to investigate for instance the influence of
this deformation on the load distribution between the planets.
– A simulation with a flexible MBS can be an effective opti-
misation tool during the design process. The calculation of
a stress distribution indicates whether or not a component is
over-dimensioned. In addition, various sensitivity analyses
may yield valuable insight in the influence of specific design
changes.

4. OBJECTIVE:
Assess the redundancy or insufficiency of the applied safety factors.
ACHIEVEMENT:

• Safety factors represent uncertainties in the design process. Since


the new simulation approach yields more confidence in the load
simulations, it permits to evaluate the traditional safety factors.

6.6 Conclusions
The generic nature of the multibody modelling techniques, presented in chap-
ter 4, permits to apply them for the dynamic analysis of a drive train in a mod-
ern wind turbine. However, building an accurate model can be considerably
complex due to the difficulties in determining correct input parameters. Espe-
cially the reduction of all drive train flexibilities into equivalent discrete spring
elements between the bodies may be difficult. An accurate consideration of
all flexibilities is nevertheless of major importance. This is demonstrated for a
6.6 Conclusions 199

drive train in a wind turbine with a gearbox, which consists of a spur planetary
gear stage, a helical planetary gear stage and a helical parallel gear stage. An
individual static analysis for the latter gear stage indicates that the gear mesh
flexibility and the bearing flexibility contribute equally to the overall torsional
flexibility of this stage. The importance of the flexibility of the components
is only half. For the planetary stages, this latter flexibility is determining for
70 % of the overall value. Similar to the parallel stage, the gear mesh and bear-
ing influences are equal. For the overall torsional flexibility of the complete
gearbox, it is demonstrated that the equivalent stiffness of the first planetary
stage is determining for more than 80 %. This is a result inherent to the torque
reduction in a gearbox.

Other difficulties in the modelling process are often the discretisation of the
drive train into various bodies and a correct definition of the boundary con-
ditions of the model. The former issue requires experience and engineering
judgement of the modeller. The effect of the latter issue is demonstrated by
comparing the results of several eigenmode calculations.

1. Firstly, the eigenmodes for the individual gear stages with a free in-
put and a free output are identified. These results are compared with
those calculated for the complete gearbox with similar boundaries (cfr.
table 6.6). This yields two conclusions:

(a) Various eigenmodes of the individual stages can also be identified


in the model of the complete gearbox and they are consequently
called the local modes, in contrast with the global modes.
(b) The frequencies of the local modes identified in the individual
stages are not equal to the frequencies of the same eigenmodes
identified in the complete gearbox. This indicates the effect of the
change in boundary conditions for the individual stages.

2. A comparison of the eigenfrequencies calculated for the model of the


complete gearbox with, on the one hand, a free input and a free output
(cfr. table 6.6) and, on the other hand, fixed boundaries (cfr. table 6.7),
indicates that most of the eigenfrequencies decrease in the latter case.
These two situations are considered as the outer limits in boundary con-
ditions and, therefore, the results indicate the maximum range in which
the respective eigenfrequencies can vary.

The introductory analyses of the individual stages permit to relate the final
eigenmodes in the complete gearbox and drive train to a single gear stage. In
addition, the individual analyses permit to evaluate the appropriateness of the
different modelling techniques for the simulation of drive train loads.
200 6. Analysis of the drive train in a modern wind turbine

1. For the high speed parallel gear stage, a purely torsional model, a rigid
multibody model and a flexible multibody model are implemented and
analysed. The consideration of more than only the torsional DOFs in the
second approach gives more relevant insight in the dynamic behaviour
of the drive train, since extra eigenmodes are found in the same fre-
quency range, which can also be excited by e.g. gear mesh vibrations.
The rigid multibody approach includes the flexibility of the bearings in
a more realistic way. The impact of these flexibilities on the torque dy-
namics cannot be neglected, since the corresponding eigenfrequencies
shift considerably when they are taken into account.

A comparison of the eigenmodes and eigenfrequencies from the rigid


and the flexible multibody model indicates that the components’ flexi-
bilities need to be taken into account in a dynamic model, because of
their influence on the eigenfrequencies. The component mode synthe-
sis applied in the latter model is an appropriate technique to consider
these flexibilities. In addition, it guarantees a proper representation of a
component’s mass and inertias. The application of this technique in the
analysis of a gear system is an important contribution of this work.
Note that the application of this reduction technique requires the com-
position of an FE model and the calculation of a set of static and normal
modes, which are relatively large modelling efforts. When the normal
modes of the individual components lie considerably higher in frequency
than the eigenmodes calculated with the rigid multibody model, it is
considered to be sufficient to neglect these normal modes and represent
the components’ flexibilities by discrete spring elements. Since this is
typically the case for the drive train components in a gearbox, the rigid
multibody approach is considered as the most appropriate and most ef-
ficient modelling technique when calculating normal modes. However,
when insight is required in the internal deformation of drive train com-
ponents, the use of flexible multibody models is the only available alter-
native.
2. The high speed planetary gear stage is also modelled using the three
modelling techniques, but the dynamic analysis is limited to the rigid
multibody model only. This is a helical planetary gear stage and its anal-
ysis is an extension of the state-of-the-art in modelling gear dynamics,
as described in section 4.3.3. This analysis indicates that, in addition to
the rotational and translational modes introduced for the spur planetary
gear stage in section 5.3, an additional category of out-of-plane modes is
found in this gear system. The corresponding frequencies can interfere
with internal excitation frequencies, which indicates their relevance.
6.6 Conclusions 201

Regarding the conclusions with respect to the eigenmodes of the drive train
in the wind turbine, a distinction is made between the frequency ranges [ 0 -
10 Hz] and [ 10 - 1500 Hz]. The analysis in the former range is based on a
model with an accurate description of the rotor, the tower and the complete
drive train and is described in appendix A. This chapter includes a discussion
of the latter analysis. A combination of a fixed-input gearbox model with an
accurate description of the components at the generator side is found to be
sufficient for a normal modes analysis in this frequency range. This is valid
under the assumption that in this frequency range the rotor can be considered
as a large inertia and the tower as a rigid supporting structure. The drive train
model in this analysis has approximately 70 DOFs. A first relevant eigenmode
in this model is identified at 68 Hz. A subsequent frequency response analysis
in the range [ 50 - 1500 Hz] for a torque excitation at the generator side and
at the gear mesh in the high speed planetary gear stage, indicates the impor-
tance of the different eigenmodes for the torque in the drive train during these
excitations. Based on the FRFs, a set of eigenmodes, which can lead to am-
plified torque levels, is identified and, subsequently, compared with possible
excitation frequencies in a Campbell diagram. This is valuable information for
assessing whether or not a drive train resonance can occur.

In addition, the frequency response analyses demonstrate how the loads in the
drive train can be simulated for a sinusoidal load excitation. The simulation of
two transient load cases is the final study in this chapter. A comparison is made
between a normal start-up and a start-up including a grid disturbance. This
disturbance causes a torque variation with a high amplitude in the generator
torque. This torque peak causes a.o.:

• a torque peak on the pinion of the high speed stage

• high rotational acceleration levels for the bearings of the pinion

• an oscillating axial displacement of the pinion in its bearings, which may


include backlashing

A sensitivity analysis indicates how a flexible coupling with a lower stiffness


value yields a lower torque peak and, vice versa, how a stiffer coupling leads
to more severe torque peaks.

Section 6.5.3 concludes this chapter by repeating the objectives in this disser-
tation and by summarising the corresponding achievements.
202
7

Measurement campaign on a
modern wind turbine

The actual version of this chapter includes confidential business information


from Hansen Transmissions International NV and is not published in the present
public version of the dissertation. The public version of this chapter gives only
a brief overview of the measurement campaign.

7.1 Overview of the measurement campaign


The measurement campaign is performed on a modern multi-MW wind tur-
bine and the purpose of the measurements is to identify the operation and the
behaviour of the wind turbine and, more particularly, of the drive train. This
purpose is threefold:

1. Characterisation of the general behaviour under normal operation:


Since the wind speed, which is the most determining parameter for the
operation of a wind turbine, is uncontrollable, the corresponding mea-
surements should be taken over a long time period, such that sufficient
experimental data is available for all wind speeds. The length of this
period is in the order of (at least) several weeks, which excludes the op-
tion of acquiring all data during a single measurement campaign on-site.
This indicates the need for an automatic and stand-alone data-logger in-
stalled in the wind turbine for such measurements. Considerable effort
has been put in acquiring a huge and unique database of time series,
which covers a.o. the normal operation of the wind turbine.

203
204 7. Measurement campaign on a modern wind turbine

2. Investigation of transient load cases:


The occurrence of transient load cases related to specific variations in
the wind, such as for instance an extreme operating gust, cannot be con-
trolled. These load cases are difficult to measure, since they require a
continuous monitoring system or a dedicated trigger to start the mea-
surements. Due to practical limitations, these load cases could not be
included in the measurement campaign.
Other external factors that influence the operation of the wind turbine
are often uncontrollable either, such as e.g. the occurrence of:
• voltage dips in the electricity grid
• a short-circuit in the electricity grid or in the generator
• component failures in the wind turbine
Again, due to practical limitations, the load cases corresponding to these
unpredictable phenomena could not be included in the measurement
campaign. On the other hand, those transient load cases, which could
be forced by the wind turbine’s control system, have been measured. As
a result, the database of time series includes measurements of various
transient load cases, such as starting up, shutting down and stopping in
an emergency case. Forcing the occurrence of these load cases is done
manually in the wind turbine’s control system and, consequently, the
measurement of these load cases was part of a measurement campaign
on-site.
3. Identification of the dynamic behaviour of the wind turbine:
It should be stated that the identification of eigenmodes from a wind tur-
bine in operation is rather complex. After all, its operation is not fully
controllable because of its dependence on the wind speed and the lim-
ited freedom in commercial wind turbine controllers, e.g. to operate at
any desired rotational speed. This latter example is required to calculate
a Campbell diagram, which is a useful tool to identify eigenmodes. The
database of measured time series includes information for all rotational
speeds of the wind turbine. However, outside its operational interval,
the wind turbine was not producing power and, consequently, at low
load conditions. This complicates the identification of eigenfrequen-
cies. Therefore, it is recommended for further experimental validation
of the numerical gearbox models, to to do future measurements also on
a gearbox test rig, which is a more controllable environment than a wind
turbine.
A second useful approach to identify eigenmodes, is analysing the re-
sponse after an emergency stop, which can be considered as an impact
excitation of the wind turbine.
7.1 Overview of the measurement campaign 205

Table 7.1 gives a summarising overview of the signals which are measured
during the campaign and have been analysed for the purposes above.

Sensor fs (*) max. bandwidth


[Hz] [Hz]
vibrations on the gearbox housing adj. [5 - fs /2]
(three triaxial accelerometers)
torque in the drive train adj. [0 - fs /2]
power production 20 [0 - 10]
generator speed 20 [0 - 10]
rotor speed 20 [0 - 10]
nacelle position 20 [0 - 10]
wind speed and wind direction 20 [0 - 10]
pitch speed and pitch postion 20 [0 - 10]
longitudinal and transversal tower acceleration 20 [0 - 10]
blade root moments (edgewise and flapwise) 20 [0 - 10]
moment about x-axis at tower bottom and top 20 [0 - 10]
moment about y-axis at tower bottom and top 20 [0 - 10]
acceleration of the nacelle in x-direction adj. [0 - fs /2]
acceleration of the nacelle in y-direction adj. [0 - fs /2]
acceleration of the nacelle in z-direction adj. [0 - fs /2]
(*): fs is the sampling rate
(**): adj. stands for adjustable

Table 7.1: Overview of the signals measured in the wind turbine


206
8

General conclusions

8.1 Overview and main contributions


Chapter 2 gives an introduction to modern wind turbines, including an overview
of various concepts for the drive train in a wind turbine and a summary of
the current status of wind powered electricity.

An important distinction between different wind turbine types is related to the


choice of the generator type.

1. The most popular drive train concept in modern wind turbines is the
combination of a gearbox with an induction generator, which is (at least
partly) in a direct grid connection. This concept has a market share of
more than 80% in the wind turbine industry. It is applied with three
alternatives for the connection between the rotor hub and the gearbox in
the wind turbine:

(a) a main shaft supported by two bearings outside the gearbox


(b) a main shaft supported by one bearing inside and one outside the
gearbox: a so-called three-point-suspension
(c) a single main bearing integrated in the gearbox

Traditional induction generators rotate near their synchronous speed,


which is typically 1500 RPM for a connection to the European 50 Hz
AC grid. The allowable speed variation is very small and, therefore,
their application is known as fixed-speed. However, since a variation in
speed can lead to a higher aerodynamic efficiency and a noise reduction,
induction generators are usually combined with dedicated developments
to achieve variable speed. A distinction is made between discrete and
continuous variable speed systems:

207
208 8. General conclusions

(a) The synchronous speed of the induction generator changes when


the number of pole pairs changes. This causes a discrete variation
of the generator speed, e.g. from 1500 RPM to 1000 RPM.
(b) Changing the slip in the generator continuously causes a continu-
ous variation of the rotational speed. The slip variation is achieved
by dissipating energy in a variable rotor resistance or by exchang-
ing energy with the grid through a partial frequency converter in
a DFIG. The allowable speed variations are typically about 10%
for the former alternative and 60% for the latter one. The latter
alternative is very popular in modern multi-MW wind turbines.

2. A second popular drive train concept is the combination of a synchronous


generator with a full frequency converter. When the generator rotates at
the speed of the wind turbine’s rotor, no gearbox is required. This con-
cept imposes furthermore less limitations on the variable speed range.
However, its application for wind turbines in the multi-MW class leads
to huge and heavy generators. Therefore, the same concept is also used
in combination with a gearbox, which keeps the generator dimensions
within limits.

Besides a controller in the generator to regulate the rotational speed and the
loads in the drive train, most wind turbines have pitch regulated blades to con-
trol the rotor speed and the rotor loads. In an active stall control the blades
turn “out of the wind” to limit the power output at high wind speeds. In a pitch
controlled wind turbine they turn in the opposite direction in this situation.

The wind turbine industry is booming since ten years and this rapid growth
is expected to continue in the coming years. This is a result of the increasing
interest for using renewable energy sources for electricity generation, which
is often promoted by political support mechanisms. The status of wind pow-
ered electricity at the end of 2005 was a global installed capacity of 59.3 GW.
Almost 70% of this capacity is installed in Europe and Germany is the global
market leader with a share of 31%.
8.1 Overview and main contributions 209

Chapter 3 describes the state-of-the-art in the design of a wind turbine drive


train and introduces a new modelling approach for the simulation of more
detailed drive train loads.

The calculation of design loads for the drive train in a wind turbine is part
of the load simulations for the complete wind turbine. Various wind turbine
design codes exist to carry out these simulations. The structural model descrip-
tion in these codes is found to be very similar for all design codes and includes
sixteen to twenty-four DOFs to represent the complete wind turbine. Only one
DOF corresponds to the torsion in the drive train, which imposes considerable
limitations on the reliability of its design. It implies a quasi-static design of
all drive train components, while dynamic load amplifications can occur as a
result of internal excitation sources. It gives furthermore no insight in local
stress levels in the different components, nor in the load variations during var-
ious transient phenomena. More detailed simulation models are required to
tackle these limitations. The flexible MBS formulation is the most appropri-
ate modelling technique to meet the identified needs and the software package
DADS from LMS International is selected as the best alternative to develop
the new models.

Additionally, appendix A describes a study of the so-called drive train modes


in traditional wind turbine design codes. Based on an FE model and a flexible
multibody model, which are similar to the structural model of a wind turbine
in such a code, the study demonstrates how the flexibility of the rotor is de-
termining to a considerable extent for the eigenfrequencies of these modes.
This contradicts the validity of a rule of thumb, which is sometimes applied in
industry, where all flexibility is assumed to be in the drive train part between
the rotor and the generator. In addition, the analysis shows that the boundary
conditions for the generator, which are usually simplified as “free” or “fixed”,
have an important influence on these modes. The rotor inertia, the blade pitch
settings, the generator inertia, the tower properties and the gearbox support
stiffness have furthermore been identified as additional determining parame-
ters.
210 8. General conclusions

Chapter 4, 5 and 6 present the main contribution of the author. This is the
development of a consistent modelling approach to correctly describe the dy-
namic behaviour of a complex drive train in a wind turbine and the transfer
of torque. The developed methodology covers the low- and mid-frequency
range. Coupling effects between the drive train and the other components of
the wind turbine (tower, rotor and generator) are taken into account.

An accurate dynamic model of the drive train requires a correct description of


the flexibilities corresponding to the deformation of the gear mesh, the bear-
ings and the other components as well as of all components inertias and of the
joints between the different components. The MBS formulation is the most
popular modelling approach in the literature on gear dynamics. This disser-
tation gives further details of the development of three MBS modelling tech-
niques with distinct levels of detail. Since the prediction of internal excitations
in the drive train is not within the scope of this study, the use of LTI models is
considered to be sufficient. These models include spring elements to represent
the bearings and the gears, assuming furthermore that their stiffness as well as
the stiffness of the components is constant and that the influence of damping is
negligible. The general-purpose multibody software package DADS offers a
generic framework to develop the three techniques and is complemented with
an accurate description of all components that are relevant for the dynamic
behaviour in the frequency range that is considered.

1. The purely torsional multibody approach includes one rotational DOF


for each body. Taking all flexibilities in the drive train into account in an
appropriate way complicates the use of this technique. The correspond-
ing analyses have furthermore an output which is limited to the torque
in the drive train.

2. The rigid multibody approach includes six DOFs for each body. The
development of an individual formulation of the bearing and the gear
mesh model yields two three-dimensional plug-in components suitable
for modelling spur gears and helical gears in parallel and planetary gear
systems as well as in complete drive trains. Especially the generic method-
ology behind this formulation is a valuable feature for the modelling of
drive trains.

3. In a flexible multibody model it is possible to include static and normal


modes of individual drive train components as additional DOFs for each
body. These modes represent the static and the dynamic deformation of
a body respectively and they are the result of a CMS calculation for a
FE model of the body. This dissertation demonstrates the application of
the CMS technique according to Craig-Bampton, which requires a set
8.1 Overview and main contributions 211

of fixed-interface and static constraint modes. The combination of this


technique with the gear mesh and bearing models of the rigid multibody
approach, is an important step forward towards the simulation of local
stress levels in gear systems.
The developed techniques are generic methods to describe the dynamic be-
haviour of a drive train in a wind turbine. Analysis of the corresponding mod-
els can yield specific insight in the drive train loads. This is demonstrated by
the application of the different methods.

An application of the first and second modelling technique on a helical parallel


gear stage, introduced by Kahraman [105], and on three spur planetary gear
stages, introduced by Lin and Parker [129], demonstrates that a rigid multibody
model is superior to a purely torsional multibody model because:
1. bearing flexibilities can be taken into account in a correct way
2. a normal modes calculation of this model yields a series of additional
modes, which may also be excited by external torque variations and are
therefore not negligible
A good correlation between the results of a normal modes calculation and the
corresponding results published by Kahraman and Lin and Parker respectively
proves the validity of the model implementation. The latter authors introduced
a classification for the eigenmodes in a spur planetary gear stage. This distin-
guishes between rotational, translational and planet modes. The same distinc-
tion is identified in the present example and it is demonstrated that only the
former modes are excited by external torque variations. Additionally, a normal
modes calculation for a model of the two example gear stages, coupled by a
torsional spring, yields a distinction between local modes and global modes.
The former modes are identical to the results identified in a normal modes
analysis of the individual stages.

Chapter 6 demonstrates the application of the developed techniques for the


analysis of a drive train in a wind turbine. The particular example consists
of a drive train with a single main bearing integrated in the gearbox and a
DFIG. It is a three stage gearbox with one spur planetary gear stage, one helical
planetary gear stage and one helical parallel gear stage. A static analysis of
the individual gear stages and of the complete gearbox yields the following
conclusions.
1. The bearing stiffness and the gear mesh stiffness both contribute by 40%
to the global torsional flexibility of the parallel gear stage. The stiffness
of the components is determining for 20% and the relative contribution
of the bearings increases for increasing helix angle.
212 8. General conclusions

2. In contrast, in the planetary stage the stiffness of the components, espe-


cially the stiffness of the planet carrier and the sun, determine the global
torsional flexibility by 70%. The respective contribution of the bearings
and the gear meshes both equal 15%.

3. The global torsional stiffness of the gearbox equals 1.5 GNm/rad. The
contribution of the first stage’s flexibility to the global flexibility is dom-
inant, which is usually valid for all types of gearboxes. In this particular
gearbox, its share is more than 80%.

The subsequent investigation yields gradually more insight in the dynamic be-
haviour of the drive train. It starts with the analysis of the individual gear
stages, which gives insight in their eigenfrequencies and eigenmodes.

1. The analysis of the helical parallel gear stage with a purely torsional
multibody model and a rigid multibody model demonstrates again the
lack of information based on the former model and the more realistic
inclusion of bearing flexibilities in the latter model. Subsequently, the
process of including additional DOFs as a set of component modes is
demonstrated, yielding a flexible multibody model of this stage. This
model includes implicitly a proper representation of the components’
flexibilities as well as their masses and inertias. The consideration of
these flexibilities has a considerable impact on the eigenfrequencies and,
therefore, they cannot be neglected. However, since the components’
normal modes lie typically substantially high in frequency, an imple-
mentation as a rigid multibody model is sufficient, when a designer is
only interested in identifying possible harmful resonance frequencies in
the gearbox. This implies a correct inclusion of the components’ stiff-
ness values in the discrete spring elements, which represent the bearings
and the gears. It excludes furthermore any insight in the components’
deformations, but it simplifies the modelling work noticeably and, there-
fore, the rigid multibody approach is considered here as the most ap-
propriate and most efficient modelling technique for the calculation of
normal modes.

2. A normal modes calculation for the helical planetary gear stage yields an
additional category of so-called out-of-plane modes, which correspond
to an out-of-plane motion of one or more components. The axial forces
on helical gears are out-of-plane forces and, the fact that the gear mesh
frequencies can interfere with the eigenfrequencies corresponding to the
out-of-plane modes, indicates the relevance of these modes.
8.1 Overview and main contributions 213

A normal modes calculation for the complete gearbox model indicates again
the distinction between modes in the individual stages, the so-called local
modes, and the global modes. The eigenfrequencies of these modes, corre-
sponding to free and fixed boundary conditions respectively, determine the
maximum range in which these frequencies can vary.

The dynamic analysis of the drive train in the wind turbine is split up into the
frequency ranges [ 0 - 10 Hz] and [ 10 - 1500 Hz]. The study in the former
range is described in appendix A, which focusses on the analysis of the drive
train modes. This appendix includes the results of a normal modes calcula-
tion and of various sensitivity analyses. Chapter 6 describes the latter analysis,
where the focus is limited on the local eigenmodes in the drive train. The anal-
ysed model includes a rigid body with a large inertia to represent the rotor and
no additional body for the tower, since this acts as a rigid supporting structure
in this frequency range. The coupling with the generator is included as a tor-
sional spring element between the gearbox output shaft and the discrete mass
of the generator.

This final model of the drive train in the wind turbine has about 70 DOFs
and has its first eigenfrequency at 68 Hz. A frequency response analysis in
the range [ 50 - 1500 Hz] for a torque excitation at the generator side and
at the gear mesh in the high speed planetary gear stage, indicates the impor-
tance of the different eigenmodes for the torque in the drive train during these
excitations. Based on the FRFs, a set of eigenmodes, which can lead to am-
plified torque levels, is identified and, subsequently, compared with possible
excitation frequencies in a Campbell diagram to identify possible drive train
resonances. In addition, the frequency response analyses demonstrate how the
loads in the drive train can be simulated for a sinusoidal load excitation. Fi-
nally, the investigation of two transient load cases demonstrates how a distur-
bance, which causes a torque variation with a high amplitude in the generator
torque, yields a torque peak on the pinion of the high speed stage, high rota-
tional acceleration levels for the bearings of the pinion and an oscillating axial
displacement of the pinion in its bearings. A sensitivity analysis indicates how
a flexible coupling with a lower stiffness value can reduce the amplitude of the
torque peak on the pinion.

The various examples in the present dissertation demonstrate how the applica-
tion of the (flexible) multibody modelling approach yields valuable informa-
tion for the design of the drive train. The presented methodology fulfils the
objectives formulated in chapter 1 and a correct application can consequently
guarantee the structural integrity of the drive train with a higher reliability.
214 8. General conclusions

Chapter 7 describes an extensive measurement campaign on a modern wind


turbine.

Since chapter 7 includes confidential business information, only a brief overview


of the measurement campaign is given in the public version of this dissertation.
The measurements were performed to (1) characterise the general behaviour of
the wind turbine during normal operation, to (2) investigate the behaviour dur-
ing various transient load cases and to (3) identify the dynamic behaviour of
the wind turbine. A huge database of time series is acquired to do the corre-
sponding analyses.

As a general remark it can be stated that the identification of eigenmodes from


a wind turbine in operation is rather complex. After all, its operation is not
fully controllable because of its dependence on the wind speed and the limited
freedom in commercial wind turbine controllers, e.g. to operate at any desired
rotational speed. Therefore, it is recommended to validate the detailed gearbox
models further through experiments on a gearbox test rig, which is a more
controllable environment.

8.2 Recommendations for future research


This dissertation indicates the need for using more advanced models in the
simulation of drive train loads in a wind turbine. Subsequently, it describes the
development of three modelling approaches and demonstrates their use to meet
the objectives in this dissertation. This section presents several suggestions to
further improve and extend the present work.

1. Section 6.5.2.3 demonstrates the simulation of a transient load case for


a particular disturbance in the electricity grid. Since the accuracy of
the predicted load levels in a transient analysis depend highly on the
accuracy of the load excitation, it is recommended to further investigate
the occurrence and the appropriate description of such excitations during
realistic load cases. The most efficient solution for this problem seems to
be an integration of the extended drive train model in one of the existing
wind turbine design codes, since these codes contain usually most of the
desired features to describe the applied loads on the wind turbine.
8.2 Recommendations for future research 215

2. A second recommendation in order to further improve the quality of load


simulations is a further improvement of the model representation.

• The effect of damping and non-linearities, such as backlash and a


variable gear mesh stiffness, can become important for load simu-
lations of transient phenomena, as for instance: starting and stop-
ping of the wind turbine, performing an emergency stop, operation
at low wind speeds and sudden voltage dips or short-circuits in the
electricity grid.
• The use of more accurate values for model parameters can help in
making the models more realistic and, consequently, yield a better
correlation with experimental measurements. Especially the deter-
mination of stiffness values for bearings and gears needs further
investigation. This can be done using detailed FE analyses in com-
bination with experimental validation. Additionally, the influence
of the flexibility of the housing and the gearbox support in the load
simulation should be further investigated. This is possible by using
a flexible MBS including a reduced FE model of these components.

3. A further development of the post-processing of the simulated loads to-


wards stress levels in the drive train components, can yield a valuable
tool for fatigue calculations or for an optimisation in the design process.

4. The recommendations above are related to the simulation of drive train


loads. In addition, the presented techniques can be further extended for
other purposes, such as for noise radiation calculations or to gain insight
in vibration monitoring techniques.
216
Bibliography

[1] A. Ahlström. Aeroelastic simulation of wind turbine dynamics. PhD


dissertation, Royal Institute of Technology, Department of Mechanics,
Stockholm, Sweden, April 2005.
[2] A. Al Shyyab and A. Kahraman. Non-linear dynamic analysis of
a multi-mesh gear train using multi-term harmonic balance method:
period-one motions. Journal of Sound and Vibration, 284(1-2):151–
172, 2005.
[3] A. Al Shyyab and A. Kahraman. Non-linear dynamic analysis of a
multi-mesh gear train using multi-term harmonic balance method: sub-
harmonic motions. Journal of Sound and Vibration, 279(1-2):417–451,
2005.
[4] M. Amabili and A. Rivola. Dynamic analysis of spur gear pairs: Steady-
state response and stability of the SDOF model with time-varying mesh-
ing damping. Mechanical Systems and Signal Processing, 11(3):375–
390, 1997.
[5] ANSI/AGMA. ANSI/AGMA 2001-D04: Fundamental Rating Factors
and Calculation Methods for Involute Spur and Helical Gear Teeth.
2001.
[6] ANSI/AGMA/AWEA. ANSI/AGMA/AWEA 6006-A03 standard: De-
sign and Specification of Gearboxes for Wind Turbines. 2003.
[7] J. Antoni and R. B. Randall. Differential diagnosis of gear and bearing
faults. Journal of Vibration and Acoustics-Transactions of the Asme,
124(2):165–171, 2002.
[8] AWEA. http://www.awea.org/faq/basicen.html. 1998.
[9] C. Bagci and S. K. Rajavenkateswaran. Critical Speed and Modal
Analyses of Rotating Machinery Using Spatial Finite-Line Element

217
218 Bibliography

Method. Proceedings of the Fifth International Modal Analyses Con-


ference, pp.1708-1717, Imperial College, London, England, April 6-9,
1987.

[10] T. L. Baker. A Field Guide to American Windmills. Norman, Oklahoma:


University of Oklahoma Press, 1985.

[11] S. Barone, L. Borgianni, and P. Forte. Evaluation of the effect of mis-


alignment and profile modification in face gear drive by a finite element
meshing simulation. Journal of Mechanical Design, 126(5):916–924,
2004.

[12] W. Bartelmus. Mathematical modelling and computer simulations as an


aid to gearbox diagnostics. Mechanical Systems and Signal Processing,
15(5):855–871, 2001.

[13] R. J. Barthelmie, G. Larsen, S. Pryor, H. Joergensen, H. Bergstrom,


W. Schlez, K. Rados, B. Lange, P. Voelund, S. Neckelmann, S. Mo-
gensen, J. Schepers, T. Hegberg, L. Folkerts, and M. Magnusson. EN-
DOW (Efficient Development of Offshore Wind Farms): Modelling
wake and boundary layer interactions. Wind Energy, 7:225–245, 2004.

[14] F. Baumjohann, M. Hermanski, R. Diekmann, and J. Kroning. 3D-Multi


Body Simulation of Wind Turbines with Flexible Components. DEWI
Magazin, 21:63–66, 2002.

[15] N. Baydar and A. Ball. Detection of gear failures via vibration and
acoustic signals using wavelet transform. Mechanical Systems and Sig-
nal Processing, 17(4):787–804, 2003.

[16] R. Belmans and W. Geysen. Elektrische machines en aandrijvingen.


2 : Wisselstroommachines, vermogenselektronica voor aandrijvingen.
Leuven: Garant, 1994.

[17] W. A. Benfield and R. F. Hruda. Vibration Analysis of Structures by


Component Mode Substitution. AIAA Journal, 9:1255–1261, 1971.

[18] M. Benton and A. Seireg. Simulation of Resonances and Instabil-


ity Conditions in Pinion-Gear Systems. Mechanical Engineering,
100(1):112, 1978.

[19] M. Benton and A. Seireg. A Dynamic Absorber for Gear Systems Op-
erating in Resonance and Instability Regions. Journal of Mechanical
Design-Transactions of the Asme, 103(2):364–371, 1981.
Bibliography 219

[20] M. Benton and A. Seireg. Factors Influencing Instability and Reso-


nances in Geared Systems. Journal of Mechanical Design-Transactions
of the Asme, 103(2):372–378, 1981.

[21] A. Betz. Schraubenpropeller mit geringstem energieverlust. Gottinger


Nachrichten, Germany, 1919.

[22] G. W. Blankenship and A. Kahraman. Steady state forced response of


a mechanical oscillator with combined parametric excitation and clear-
ance type non-linearity. Journal of Sound and Vibration, 185(5):743–
765, 1995.

[23] A. Blümm. Simplex 5. Ruhr-Universität Bochum, Germany,


http://www.lmgknet.ruhr-uni-bochum.de/ger/planet ger.htm, 2002.

[24] P. M. M. Bongers. Duwecs reference guide v2.0, delft university wind


energy converter simulation package. DUT-MEMT 26, Delft University
of Technology, The Netherlands, 1993.

[25] Bonus. Bonus-Info Newsletter. 1998.

[26] E. A. Bossanyi. Prediction of flicker produced by wind turbines. Wind


Energy, 1(1):35–51, 1998.

[27] E. A. Bossanyi. The Design of closed loop controllers for wind turbines.
Wind Energy, 3(3):149–163, 2000.

[28] E. A. Bossanyi. Individual Blade Pitch Control for Load Reduction.


Wind Energy, 6(2):119–128, 2003.

[29] E. A. Bossanyi. Wind Turbine Control for Load Reduction. Wind En-
ergy, 6(3):229–244, 2003.

[30] E. A. Bossanyi. Further load reductions with individual pitch control.


Wind Energy, 8(4):481–485, 2005.

[31] E. A. Bossanyi. Gh bladed theory manual. Document 282/BR/009, Issue


13, Garrad Hassan and Partners Limited, Bristol, UK, 2005.

[32] C. Braccesi and F. Cianetti. Development of selection methodologies


and procedures of the modal set for the generation of flexible body
models for multi-body simulation. Proceedings of the Institution of
Mechanical Engineers.Proceedings.Part K: Journal of multi-body dy-
namics, 218:19–30, 2004.
220 Bibliography

[33] C. Braccesi, L. Landi, and R. Scaletta. New dual meshless flexible


body methodology for multi-body dynamics: simulation of generalized
moving loads. Proceedings of the Institution of Mechanical Engineers
Part K-Journal of Multi-Body Dynamics, 218(1):51–62, 2004.
[34] J. Brandlein, P. Eschmann, and L. Hasbargen. Ball and roller bearings:
theory, design and application. Chichester: Wiley, 1999.
[35] M. L. Buhl and A. Manjock. A Comparison of Wind Turbine Aeroelas-
tic Codes Used for Certification. 44th AIAA Aerospace Sciences meeting
and exhibition, Reno, Nevada, US, January 9-12, 2006.
[36] M. L. Buhl, A. Wright, and K. G. Pierce. Wind Turbine Design Codes:
A Comparison of the Structural Response. Proceedings of the ASME
Wind Energy Symposium, Reno, Nevada, US, January 10-13, 2000.
[37] T. Burton, D. Sharpe, E. A. Bossanyi, and N. Jenkins. Wind Energy
handbook. John Wiley and Sons, Ltd., 2004.
[38] Y. Cai. Simulation on the rotational vibration of helical gears in consid-
eration of the tooth separation phenomenon (a new stiffness function of
a helical involute tooth pair). Journal of Mechanical Design, 117:460–
469, 1995.
[39] Y. Cai and T. Hayashi. The linear approximated equation of vibration
of a pair of spur gears (theory and experiment). Journal of Mechanical
Design, 116:558–563, 1994.
[40] C. Capdessus, M. Sidahmed, and J. L. Lacoume. Cyclostationary pro-
cesses: Application in gear faults early diagnosis. Mechanical Systems
and Signal Processing, 14(3):371–385, 2000.
[41] G. P. Corten. Flow Separation on Wind Turbine Blades. PhD disserta-
tion, University of Utrecht, Utrecht, The Netherlands, January 2001.
[42] G. P. Corten and H. F. Veldkamp. Insects cause double stall. Proceed-
ings of the European Wind Energy Conference (EWEC2001), Copen-
hagen, Denmark, July 1-7, 2001.
[43] R. R. J. Craig. Structural dynamics. John Wiley & Sons, Inc., 1981.
[44] R. R. J. Craig. A review of time-domain and frequency-domain
component-mode synthesis methods. International Journal of Analyti-
cal and Experimental Modal Analysis, 2(2):59–72, 1987.
[45] R. R. J. Craig. Substructure Methods in Vibration. Journal of Mechan-
ical Design, 117:207–213, 1994.
Bibliography 221

[46] R. R. J. Craig and M. C. C. Bampton. Coupling of substructures for


dynamic analyses. AIAA Journal, 6(7):1313–1319, 1968.

[47] R. R. J. Craig and C. J. Chang. Free-interface methods of substructure


coupling for dynamic analysis. AIAA Journal, 14(11):1633–1635, 1976.

[48] R. R. J. Craig, F. Hemez, J. Bennighof, D. Kammer, Y.-T. Chung, and


C. Pickrel. Roy Craig, Engineering Educator And Pioneer Contributor
To Component Mode Synthesis. Proceedings of the 22nd International
Modal Analysis Conference (IMAC), Dearborn, Michigan, US, January
26-29, 2004.

[49] X. Cyril, J. Angeles, and A. Misra. Dynamics of Flexible Multibody


Mechanical Systems. Transactions of the Canadian Society for Me-
chanical Engineering, 15(3):235–256, 1991.

[50] P. De Fonseca. Simulation and optimisation of the dynamic behaviour


of mechatronic systems. PhD dissertation, K.U.Leuven, Department of
Mechanical Engineering, Division PMA, Heverlee, Belgium, 2000.

[51] E. De Vries. Trouble spots: Gearbox failures and design solutions. Re-
newable Energy World, 9(2):37–47, 2006.

[52] J. P. Den Hartog. Mechanical vibrations. Dover Publications, 1956.

[53] W. D’haeseleer. Energie vandaag en morgen : Beschouwingen over


energievoorziening en -gebruik. Leuven : Acco, 2005.

[54] DIN. DIN 3990, Deutsches Institut für Normung e.V.: Calculation of
load capacity of cylindrical gears. 1987.

[55] DIN. DIN 281, Deutsches Institut für Normung e.V.: Rolling bearings:
dynamic load ratings and rating life. 1990.

[56] DIN. DIN 743, Deutsches Institut für Normung e.V.: Shafts and axles,
calculation of load capacity. 2000.

[57] A. Dixit and S. Suryanarayanan. Towards Pitch-Scheduled Drive Train


Damping in Variable-Speed, Horizontal-Axis Large Wind Turbines.
Proceedings of the Conference on Decision and Control 2005, pp. 1295-
1300, Seville, Spain, 2005.

[58] DNV. DNV 41.2: Calculation of Gear Rating for Marine Transmission.
2003.

[59] H. Dresig. Schwingungen mechanischer Antriebssyteme: Modelbil-


dung, Berechnung, Analyse, Synthese. Springer-Verlag, 2001.
222 Bibliography

[60] S. Du. Dynamic modelling and simulation of gear transmission error for
gearbox vibration analysis. PhD dissertation, University of New South
Wales, Sydney, Australia, 1997.

[61] S. Du, R. B. Randall, and D. W. Kelly. Modelling of spur gear mesh


stiffness and static transmission error. Proceedings of the Institution
of Mechanical Engineers Part C-Journal of Mechanical Engineering
Science, 212(4):287–297, 1998.

[62] H. El Absy and A. A. Shabana. Coupling between rigid body and defor-
mation modes. Journal of Sound and Vibration, 198(5):617–637, 1996.

[63] H. El Absy, A. A. Shabana, and A. A. Shabana. Geometric stiffness


and stability of rigid body modes. Journal of Sound and Vibration,
207(4):465–496, 1997.

[64] M. El Badaoui, J. Antoni, F. Guillet, J. Daniere, and P. Velex. Use Of


The Moving Cepstrum Integral To Detect And Localise Tooth Spalls
In Gears. Mechanical Systems and Signal Processing, 15(5):873–885,
2001.

[65] M. El Badaoui, V. Cahouet, F. Guillet, J. Daniere, and P. Velex. Model-


ing and detection of localized tooth defects in geared systems. Journal
of Mechanical Design, 123(3):422–430, 2001.

[66] M. El Badaoui, F. Guillet, and J. Daniere. New applications ofthe real


cepstrum to gear signals, including definition of a robust fault indicator.
Mechanical Systems and Signal Processing, 18:1031–1046, 2004.

[67] A. S. Elliott and A. D. Wright. Adams/wt 2.0 user’s guide, version 2.0.
Mechanical Dynamics Inc. (Ann Arbor, Michigan, US) and National
Renewable Energy Department (Golden, Colorado, US), 1998.

[68] Enercon. Product information. http://www.enercon.de, Germany, 2006.

[69] ESCO. ESCOWIND couplings: DWMO / FWMO cou-


plings for wind turbines with integrated torque limiter.
http://www.escocoupling.com/Wind.pdf, Belgium, 2006.

[70] ESM GmbH. Product Information. http://www.esm-gmbh.de, Germany,


2005.

[71] A. Fernandez, F. Viadero, J. Pascual, P. Garcia, and R. Sancibrian. Vi-


bration Behaviour Modelling for a Low-Speed Gearbox. Proceedings
of the International Conference on Noise and Vibration Engineering
(ISMA 2002), Leuven, Belgium, September 16-18, 2002.
Bibliography 223

[72] H. Ganander. The Vidyn story. Proceedings of the 28th IEA Meeting of
Experts “State of the Art of Aerolelastic Codes for Wind Turbine Cal-
culations”, pp. 77-90, Technical University of Denmark, Lyngby, Den-
mark, April 11-12, 1996.

[73] H. Ganander. The Use of a Code-generating System for the Derivation


of the Equations for Wind Turbine Dynamics. Wind Energy, 6(4):333–
345, 2003.

[74] Garrad Hassan and Partners Ltd. GH Bladed: Product Informa-


tion. http://www.garradhassan.com/products/ghbladed/index.php, UK,
2006.

[75] M. Géradin and D. Rixen. Mechanical Vibrations: Theory and Appli-


cation to Structural Dynamics, 2nd Edition. Wiley, 1997.

[76] H. Gerber. Innere dynamische Zusatzkräfte bei Stirnradgetrieben: Mod-


ellbildung, innere Anregung un Dämpfung. Phd dissertation, TU Mu-
nich, Munich, Germany, 1984.

[77] H. Glauert. Airplane propellors. Aerodynamic Theory (Springer,


Berlin), 4:169–360, 1935.

[78] H. Glauert. The elements of airfoil and airscrew theory (second edition).
Cambridge University Press, 1959.

[79] P. Gold. DRESP 7.0 Documentation. IME-RWTH Aachen, FVA, Ger-


many, 2000.

[80] P. W. Gold, R. Schelenz, J. Fechler, W. Frenschek, and A. Klein.


Simulation des dynamischen Verhaltens von Windkraftanlagen. VDI-
Berichte Nr. 1749, Schwingungen in Antrieben 2003, pp.271-286,
Fulda, Germany, April 2-3, 2003.

[81] GWEC. Global Wind Report. 2006.

[82] A. Hansen and D. Laino. Users Guide to the Wind Turbine Dynamics
Computer Programs YawDyn and Aerodyn for Adams - Version 11.0.
University of Utah, Salt Lake City, US, 1998.

[83] A. C. Hansen. Yaw dynamics of horizontal axis wind turbines: Final


report. NREL/TP-442-4822, 73p., National Renewable Energy Labora-
tory (NREL), US, 1992.

[84] A. C. Hansen. User’s guide to the yaw dynamics computer program:


Yawdyn. report, 32p., 1993.
224 Bibliography

[85] M. H. Hansen. Improved Modal Dynamics of Wind Turbines to Avoid


Stall-induced Vibrations. Wind Energy, 6:179–195, 2003.

[86] M. H. Hansen, P. Fuglsang, K. Thomsen, and T. Knudsen. Two Methods


for estimating aeroelastic damping of operational wind turbine modes
from experiments. Proceedings of the European Wind Energy Confer-
ence (EWEC2004), Wembley, Londen, UK, November 22-25, 2004.

[87] M. O. L. Hansen. Aerodynamics of Wind Turbines. James & James


(Science Publishers), 2000.

[88] R. Harrison, E. Hau, and H. Snel. Large wind turbines. John Wiley and
Sons, Chichester, 2000.

[89] E. Hau. Wind turbines. Springer Verlag, 2000.

[90] A. Heege. Evaluation of wind turbine gearbox fatigue load spectra by


coupled structural and mechanism analysis. Dresdner Maschinenele-
mente Kolloquium (DMK2005), Dresden, Germany, December 1-2,
2005.

[91] A. Heege. Quantification of wind turbine gearbox loads by coupled


structural and mechanisms analysis. Proceedings of the 7th German
Wind Energy Conference (DEWEK2004), Wilhelmshaven, Germany,
Oktober 20-21, 2004.

[92] A. Heege. Computation of Dynamic Loads in Wind Turbine Power


Trains. DEWI Magazin, 23:59–64, 2003.

[93] W. Heylen, S. Lammens, and P. Sas. Modal analysis Theory and Testing.
Leuven : Acco, 1999.

[94] J. Y. L. Ho. Direct Path Method for Flexible Multi-Body Spacecraft


Dynamics. Journal of Spacecraft and Rockets, 14(2):102–110, 1977.

[95] I. Howard, S. X. Jia, and J. D. Wang. The dynamic modelling of a spur


gear in mesh including friction and a crack. Mechanical Systems and
Signal Processing, 15(5):831–853, 2001.

[96] W. C. Hurty. Dynamic analysis of structural systems using component


modes. AIAA Journal, 3(4):678–685, 1965.

[97] R. A. Ibrahim and A. D. Barr. Parametric vibration part-I: Mechanics


of linear problems. The Shock and Vibration Digest, 10:9–24, 1978.
Bibliography 225

[98] R. A. Ibrahim and M. A. El Sayad. Simultaneous parametric and inter-


nal resonances in systems involving strong non-linearities. Journal of
Sound and Vibration, 225(5):857–885, 1999.

[99] IEC. IEC61400-1 (2nd edition): Wind turbine generator systems - Part
1: Safety Requirements. 1999.

[100] IME. DyLa Documentation. IME-RWTH Aachen, FVA,


http://www.ime.rwth-aachen.de/index.php?id=99, Germany, 2005.

[101] INTEC GmbH. Simpack: Product information.


http://www.simpack.com, Germany, 2005.

[102] ISO. ISO6336-2: Calculation of load capacity of spur and helical gears.
Calculation of surface durability (pitting). 1996.

[103] S. X. Jia and I. Howard. Comparison of localised spalling and crack


damage from dynamic modelling of spur gear vibrations. Mechanical
Systems and Signal Processing, 20(2):332–349, 2006.

[104] W. Johnson. Helicopter Theory. Princeton University Press: Princeton,


NJ, 1980.

[105] A. Kahraman. Effect of axial vibrations on the dynamics of a helical


gear pair. Journal of Vibration and Acoustics, 115:33–39, 1993.

[106] A. Kahraman. Dynamic analysis of a multi-mesh helical gear train.


Journal of Mechanical Design, 116:706–712, 1994.

[107] A. Kahraman. Planetary gear train dynamics. Journal of Mechanical


Design, 116:713–720, 1994.

[108] A. Kahraman and G. W. Blankenship. Interactions between commen-


surate parametric and forcing excitations in a system with clearance.
Journal of Sound and Vibration, 194(3):317–336, 1996.

[109] A. Kahraman and G. W. Blankenship. Experiments on nonlinear dy-


namic behavior of an oscillator with clearance and periodically time-
varying parameters. Journal of Applied Mechanics-Transactions of the
Asme, 64(1):217–226, 1997.

[110] A. Kahraman and G. W. Blankenship. Effect of Involute Contact Ratio


on Spur Gear Dynamics. Journal of Mechanical Design, 121:112–118,
1999.
226 Bibliography

[111] A. Kahraman, H. N. Özgüven, D. R. Houser, and J. J. Zakrajsek. Dy-


namic analysis of geared rotors by finite elements. Journal of Mechan-
ical Design, 114:507–514, 1992.

[112] A. Kahraman and R. Singh. Nonlinear Dynamics of A Spur Gear Pair.


Journal of Sound and Vibration, 142(1):49–75, 1990.

[113] A. Kahraman and R. Singh. Interactions Between Time-Varying Mesh


Stiffness and Clearance Nonlinearities in A Geared System. Journal of
Sound and Vibration, 146(1):135–156, 1991.

[114] R. Kasuba and J. W. Evans. An Extended Model for Determining


Dynamic Loads in Spur Gearing. Journal of Mechanical Design-
Transactions of the Asme, 103(2):398–409, 1981.

[115] F. Krull. Triebstrangschwingungen in Windkraftgetrieben. Dresd-


ner Maschinenelemente Kolloquium (DMK2003), Dresden, Germany,
September 23-24, 2005.

[116] F. Krull. Vibrations and dynamic behaviour of gearboxes in drive trains


of wind turbines. Proceedings of the 7th German Wind Energy Confer-
ence (DEWEK2004), Wilhelmshaven, Germany, Oktober 20-21, 2004.

[117] J. H. Kuang and A. D. Lin. The effect of tooth wear on the vibra-
tion spectrum of a spur gear pair. Journal of Vibration and Acoustics-
Transactions of the Asme, 123(3):311–317, 2001.

[118] J. H. Kuang and A. D. Lin. Theoretical aspects of torque responses in


spur gearing due to mesh stiffness variation. Mechanical Systems and
Signal Processing, 17(2):255–271, 2003.

[119] M. Lalanne and G. Ferraris. Rotordynamics: Prediction in engineering.


John Wiley and Sons, 1997.

[120] B. Lange, H.-P. Waldl, and R. J. Barthelmie. Modelling of Offshore


Wind Turbine Wakes with the Wind Farm Program FLaP. Wind Energy,
6(1):87–104, 2003.

[121] G. C. Larsen, M. H. Hansen, A. Baumgart, and I. Carlen. Modal analy-


sis of wind turbine blades. Ris-r-1181, Risø National Laboratory, Den-
mark, 2002.

[122] T. Larsen, M. H. Hansen, and F. Iov. Generator dynamics in aeroelastic


analysis and simulations. Ris-r-1395, Risø National Laboratory, Den-
mark, 2003.
Bibliography 227

[123] T. Larsen, H. A. Madsen, and K. Thomsen. Active load reduction using


individual pitch, based on local blade flow measurements. Wind Energy,
8(1):67–80, 2005.
[124] T. Larsen, K. Thomsen, and F. Rasmussen. Dynamics of a Wind Tur-
bine Planetary Gear Stage. Proceedings of the European Wind Energy
Conference (EWEC2003), Madrid, Spain, June 16-19, 2003.
[125] D. Lee, D. H. Hodges, and M. J. Patil. Multi-flexible-body Dynamic
Analysis of Horizontal Axis Wind Turbines. Wind Energy, 5:281–300,
2002.
[126] J. G. Leishman. Challenges in Modeling the Unsteady Aerodynamics
of Wind Turbines. Proceedings of the 21st ASME Wind Energy Sympo-
sium, Reno, Nevada, US, January 14-17, 2002.
[127] C. J. Li, H. Lee, and S. H. Choi. Estimating Size Of Gear Tooth Root
Crack Using Embedded Modelling. Mechanical Systems and Signal
Processing, 16(5):841–852, 2002.
[128] D. Lin, M. Wiseman, D. Banjevic, and A. K. S. Jardine. An approach
to signal processing and condition-based maintenance for gearboxes
subject to tooth failure. Mechanical Systems and Signal Processing,
18:993–1007, 2004.
[129] J. Lin and R. G. Parker. Analytical Characterization of the Unique Prop-
erties of Planetary Gear Free Vibration. Journal of Vibration and Acous-
tics, 121:316–321, 1999.
[130] J. Lin and R. G. Parker. Sensitivity Of Planetary Gear Natural Frequen-
cies And Vibration Modes To Model Parameters. Journal of Sound and
Vibration, 228(1):109–128, 1999.
[131] J. Lin and R. G. Parker. Structured Vibration Characteristics Of Plan-
etary Gears With Unequally Spaced Planets. Journal of Sound and Vi-
bration, 233(5):921–928, 1999.
[132] J. Lin and R. G. Parker. Natural frequency veering in planetary gears.
Mechanical Structures and Machinery, 29(4):411–429, 2001.
[133] J. Lin and R. G. Parker. Planetary gear parametric instability caused by
mesh stiffness variation. Journal of Sound and Vibration, 249(1):129–
145, 2002.
[134] C. Lindenburg and T. Hegberg. Phatas-iv user’s manual - program for
horizontal axis wind turbine analysis and simulation - version iv. ECN-
C-99-093, Energy Research Centre (ECN), The Netherlands, 2000.
228 Bibliography

[135] F. L. Litvin, J. S. Chen, J. Lu, and R. F. Handschuh. Application of


finite element analysis for determination of load share, real contact ratio,
precision of motion, and stress analysis. Journal of Mechanical Design,
118(4):561–567, 1996.

[136] LM Glasfiber A/S. Company Brochure. Denmark, 2004.

[137] LM Glasfiber A/S. Product information. Denmark, 2005.

[138] LMS. DADS Revision 9.6 Documentation. LMS International, Bel-


gium, 2000.

[139] O. Lundvall, N. Stromberg, and A. Klarbring. A flexible multi-body


approach for frictional contact in spur gears. Journal of Sound and
Vibration, 278(3):479–499, 2004.

[140] R. H. MacNeal. A hybrid method of component mode synthesis. Com-


puters and Structures, 1(4):581–601, 1971.

[141] H. A. Madsen and F. Rasmussen. A near wake model for trailing vortic-
ity compared with the blade element momentum theory. Wind Energy,
7:325–341, 2004.

[142] S. F. Madsen. Interaction between electrical discharges and materials for


wind turbine blades particularly related to lightning protection. PhD
dissertation, Technical University of Denmark (DTU), Órsted, Den-
mark, 2004.

[143] J. Mann. Wind field simulation. Probabilistic Engineering Mechanics,


13(4):269–282, 1998.

[144] S. Meisner and B. Campbell. Development of Gear Rattle Analytical


Simulation Methodology. Proceedings of the SAE Noise and Vibration
Conference, US, 1995.

[145] G. Meltzer and N. P. Dien. Fault diagnosis in gears operating under non-
stationary rotational speed using polar wavelet amplitude maps. Me-
chanical Systems and Signal Processing, 18:985–992, 2004.

[146] G. Meltzer and Y. Y. Ivanov. Fault Detection In Gear Drives With Non-
Stationary Rotational Speed - Part II: The Time-Quefrency Approach.
Mechanical Systems and Signal Processing, 17(2):273–283, 2003.

[147] G. Meltzer and Y. Y. Ivanov. Fault detection in gear drives with non-
stationary rotational speed part I: the time-frequency approach. Me-
chanical Systems and Signal Processing, 17(5):1033–1047, 2003.
Bibliography 229

[148] Middelgrunden wind turbine cooperative. Documentation.


http://www.middelgrunden.dk, Denmark, 2005.

[149] D.-P. Molenaar. Cost-effective design and operation of variable speed


wind turbines. PhD dissertation, Delft University of Technology, Delft,
The Netherlands, 2002.

[150] MSC. MSC.Software: Product Information.


http://www.mscsoftware.com, 2005.

[151] D. Muhs, H. Wittel, and M. Becker. Roloff/Matek Machineonderdelen:


normering, berekening, vormgeving. 1998.

[152] Multibrid. Product Information. http://www.multibrid.com, 2005.

[153] A. H. Nayfeh and D. T. Mook. Nonlinear Oscillations. New York:


Wiley, 1979.

[154] G. Niemann and H. Winter. Maschinenelemente. Springer, Berlin, 1983.

[155] E. Nim. Coupling and reduction of the HAWC equations. Ris-r-1294,


Risø National Laboratory, Denmark, 2001.

[156] Nordex. Technical description NORDEX S70/S77: Product Informa-


tion. http://www.nordex-online.com, 2005.

[157] NVN. NVN 11400/0: Regulations for the Type-Certification of Wind


Turbines: Technical Criteria. 1999.

[158] S. Øye. FLEX 4 - Simulation of Wind Turbine Dynamics. Proceedings


of the 28th IEA Meeting of Experts “State of the Art of Aerolelastic
Codes for Wind Turbine Calculations”, pp. 71-76, Technical University
of Denmark, Lyngby, Denmark, April 11-12, 1996.

[159] S. Øye, J. Schepers, J. Heijdra, D. Foussekis, R. Smith, M. Belessis,


K. Thomsen, T. Larsen, I. Kraan, B. Visser, I. Carlen, H. Ganander,
and L. Drost. Verification of European Wind Turbine Design Codes,
VEWTDC: Final report. ECN-C-01-055, Energy Research Centre of
the Netherlands (ECN), 2003.

[160] H. N. Özgüven and D. R. Houser. Mathematical models used in gear


dynamics: a review. Journal of Sound and Vibration, 121(3):383–411,
1988.

[161] A. Palmgren. Grundlagen der Wälzlagertechnik. Franckh, 3rd Edition,


Göteborg, 1964.
230 Bibliography

[162] A. Parey and N. Tandon. Spur Gear Dynamic Models Including Defects:
A Review. The Shock and Vibration Digest, 35(6):465–478, 2003.

[163] R. G. Parker. Progress and problems in gear vibration and noise. Pro-
ceedings of the 2nd International Workshop on Damping Technologies,
Stellenbosch, South Africa, March, 2003.

[164] R. G. Parker. A physical explanation for the effectiveness of planet


phasing to suppress planetary gear vibration. Journal of Sound and
Vibration, 236(4):561–573, 2000.

[165] R. G. Parker, V. Agashe, and S. M. Vijayakar. Dynamic Response of


a Planetary Gear System Using a Finite Element/Contact Mechanics
Model. Journal of Mechanical Design, 122:305–311, 2000.

[166] R. G. Parker and J. Lin. Mesh Stiffness Variation Instabilities in Two-


Stage Gear Systems. Journal of Vibration and Acoustics, 124(1):68–76,
2002.

[167] R. G. Parker, S. M. Vijayakar, and T. Imajo. Non-linear dynamic re-


sponse of a spur gear pair: modelling and experimental comparisons.
Journal of Sound and Vibration, 237(3):435–455, 2000.

[168] J. T. Petersen. The aeroelastic code HAWC - model and comparisons.


Proceedings of the 28th IEA Meeting of Experts “State of the Art of
Aerolelastic Codes for Wind Turbine Calculations”, pp. 129-135, Tech-
nical University of Denmark, Lyngby, Denmark, April 11-12, 1996.

[169] J. T. Petersen. Kinematically nonlinear finite element model of a hor-


izontal axis wind turbine. part 1: Mathematical model and results. re-
port, Risø National Laboratory, Denmark, 1990.

[170] J. T. Petersen. Kinematically nonlinear finite element model of a hori-


zontal axis wind turbine. part 2: Supplement. inertia matrices and aero-
dynamic model. report, Risø National Laboratory, Denmark, 1990.

[171] J. T. Petersen, H. A. Madsen, A. Bjorck, P. Enevoldsen, S. Óye,


H. Ganander, and D. Winkelaar. Prediction of dynamic loads and in-
duced vibrations in stall. Ris-r-1045, Risø National Laboratory, Den-
mark, 1998.

[172] A. Petersson. Analysis, Modeling and Control of Doubly-Fed Induction


Generators for Wind Turbines. PhD dissertation, Chalmers Univer-
sity of Technology, Department of Energy and Environment, Division of
Electric Power Engineering, Göteborg, Sweden, 2005.
Bibliography 231

[173] G. Polifke. Dynamisches Verhalten von mehrstufigen Plan-


etenradgetrieben. Ruhr-Universität Bochum, Germany,
http://www.lmgknet.ruhr-uni-bochum.de/ger/planet ger.htm, 1997.

[174] D. C. Quarton. The evolution of wind turbine design analysis - a twenty


year progress review. Wind Energy, 1:5–24, 1998.

[175] R. B. Randall. Editorial For Special Edition On Gear And Bearing Di-
agnostics. Mechanical Systems and Signal Processing, 15(5):827–829,
2001.

[176] R. B. Randall. State of the art in monitoring rotating machinery - Part


1. Sound and Vibration, 38(3):14–21, 2004.

[177] R. B. Randall. State of the art in monitoring rotating machinery - Part


2. Sound and Vibration, 38(5):10–17, 2004.

[178] F. Rasmussen, M. Hansen, K. Thomsen, T. Larsen, F. Bertagnolio, J. Jo-


hansen, H. A. Madsen, C. Bak, and A. Hansen. Present Status of Aeroe-
lasticity of Wind Turbines. Wind Energy, 6:213–228, 2003.

[179] F. Rasmussen, K. Thomsen, and T. Larsen. The gearbox problem revis-


ited. AED-rb-17(EN), Risø National Laboratory, Denmark, 2004.

[180] N. F. Rieger and J. F. Crofoot. Vibrations of Rotating Machinery. Vi-


bration Institute, 1977.

[181] Risø. Demo version of HAWCModal. Risø National Laboratory, Wind


Energy Department, Roskilde, Denmark.

[182] Risø and DNV. Guidelines for Design of Wind Turbines. Det Norske
Veritas (DNV) and Wind Energy Department, RisøNational Laboratory,
2002.

[183] V. A. Riziotis and S. Voutsinas. GAST: A General Aerodynamic and


Structural Prediction Tool for Wind Turbines. Proceedings of the Eu-
ropean Wind Energy Conference (EWEC1997), pp. 448-452, Dublin,
Ireland, October 6-9, 1997.

[184] S. Rubin. Improved component-mode representation for structural dy-


namic analysis. AIAA Journal, 13(8):995–1006, 1975.

[185] B. Schlecht and S. Gutt. Multibody-System-Simulation of Drive Trains


of Wind Turbines. Proceedings of the Fifth World Congress on Compu-
tational Mechanics (WCCM V), Vienna, Austria, 2002.
232 Bibliography

[186] B. Schlecht, T. Schulze, and T. Hähnel. Multibody-system-simulation


of drive trains in wind turbines. Proceedings of The International Con-
ference on Noise and Vibration Engineering (ISMA2004), Leuven, Bel-
gium, September 20-22, 2004.
[187] B. Schlecht, T. Schulze, and T. Hähnel. Today’s techniques of the
assessment of dynamic loads in drive trains of wind turbines using
multibody-system-simulation. Proceedings of the 7th German Wind
Energy Conference (DEWEK2004), Wilhelmshaven, Germany, Oktober
20-21, 2004.
[188] A. L. Schwab. Dynamics of Flexible Multibody Systems: small vibra-
tions superimposed on a general rigid body motion. PhD dissertation,
Delft University of Technology, Delft, The Netherlands, 2002.
[189] S. Seman, S. Kanerva, J. Niiranen, and A. Arkkio. Transient Analysis
of Doubly Fed Power Induction Generator Using Coupled Field-Circuit
Model. Proceedings of the 16th International Conference on Electrical
Machines, Krakow, Poland, September 5-8, 2004.
[190] S. Seman, J. Niiranen, S. Kanerva, and A. Arkkio. Analysis of a 1.7
MVA Doubly Fed Wind-Power Induction Generator during Power Sys-
tems Disturbances. Proceedings of the Nordic Workshop on Power and
Industrial Electronics (NORPIE/2004), Trondheim, Norway, June 14-
16, 2004.
[191] J. D. Smith. Gears and their vibration. Marcel Dekker, New York and
MacMillan, London, 1983.
[192] J. D. Smith. Gear Transmission Error Accuracy with Small Rotary En-
coders. Proceedings of the Institution of Mechanical Engineers Part
C-Journal of Mechanical Engineering Science, 201(2):133–135, 1987.
[193] J. D. Smith. Accuracy of small encoders for gear transmission errors.
Proceedings of the Institution of Mechanical Engineers Part C-Journal
of Mechanical Engineering Science, 215(C8):995–998, 2001.
[194] J. D. Smith and D. B. Welbourn. Gearing research in cambridge
1827-2000. University of Cambridge, Department of engineering,
http://www-g.eng.cam.ac.uk/125/achievements/gears/index.htm, UK,
2000.
[195] H. Snel. Review of the present status of rotor aerodynamics. Wind
Energy, 1(S1):46–69, 1999.
[196] R. Snoeys. Dimensioneren tegen vermoeiing. Acco, Leuven, 1988.
Bibliography 233

[197] J. Soens. Impact of wind energy on a future power grid. PhD dis-
sertation, K.U.Leuven, Department of Electrical Engineering (ESAT),
Division ELECTA, Heverlee, Belgium, 2005.

[198] J. Soens, J. Driesen, and R. Belmans. Interaction between Electri-


cal Grid Phenomena and the Wind Turbine’s Behaviour. Proceedings
of The International Conference on Noise and Vibration Engineering
(ISMA2004), pp. 3969-3988, Leuven, Belgium, September 20-22, 2004.

[199] T. Soerensen and M. H. Brask. Lightning protection of wind turbines.


Technical report, DEFU, RisøNational Laboratory, Denmark, 1999.

[200] J. Sopanen and A. Mikkola. Dynamic model of a deep-groove ball bear-


ing including localized and distributed defects. Part 1: theory. Pro-
ceedings of the Institution of Mechanical Engineers.Proceedings.Part
K: Journal of multi-body dynamics, 217:201–211, 2003.

[201] J. Sopanen and A. Mikkola. Dynamic model of a deep-groove ball


bearing including localized and distributed defects. Part 2: implemen-
tation and results. Proceedings of the Institution of Mechanical Engi-
neers.Proceedings.Part K: Journal of multi-body dynamics, 217:213–
223, 2003.

[202] J. T. Spanos and W. S. Tsuha. Selection of Component Modes for Flexi-


ble Multibody Simulation. Journal of Guidance Control and Dynamics,
14(2):278–286, 1991.

[203] D. A. Spera. Wind Turbine Technology: Fundamental concepts of wind


turbine engineering. ASME Press, 1998.

[204] Stentec b.v. Twister: Product information.


http://www.stentec.com/windframe.html, The Netherlands, 2005.

[205] C. K. Sung, H. M. Tai, and C. W. Chen. Locating defects of a gear


system by the technique of wavelet transform. Mechanism and Machine
Theory, 35(8):1169–1182, 2000.

[206] K. Thomsen and H. A. Madsen. A new simulation method for turbines


in wake - Applied to extreme response during operation. Wind Energy,
8:35–47, 2005.

[207] H. Van den Wijngaert. Personal communication, November 17, 2005.

[208] L. Vedmar and A. Andersson. A method to determine dynamic loads


on spur gear teeth and on bearings. Journal of Sound and Vibration,
267:1065–1084, 2003.
234 Bibliography

[209] P. S. Veers. Three-dimensional wind simulation. Technical Report


SAND88-0152, Sandia National Laboratories, US, 1988.

[210] P. Velex and V. Cahouet. Experimental and numerical investigations on


the influence of tooth friction in spur and helical gear dynamics. Journal
of Mechanical Design, 122(4):515–522, 2000.

[211] P. Velex and M. Maatar. A mathematical model for analyzing the in-
fluence of shape deviations and mounting errors on gear dynamic be-
haviour. Journal of Sound and Vibration, 191(5):629–660, 1996.

[212] Vestas. Product Information. http://www.vestas.com, Denmark, 2006.

[213] H. Vinayak, R. Singh, and C. Padmanabhan. Linear Dynamic Anal-


ysis Of Multi-Mesh Transmissions Containing External, Rigid Gears.
Journal of Sound and Vibration, 185(1):1–32, 1995.

[214] B. Visser. The aeroelastic code FLEXLAST. Proceedings of the 28th


IEA Meeting of Experts “State of the Art of Aerolelastic Codes for Wind
Turbine Calculations”, pp. 161-166, Technical University of Denmark,
Lyngby, Denmark, April 11-12, 1996.

[215] J. D. Wang and I. Howard. The torsional stiffness of involute spur gears.
Proceedings of the I MECH E Part C Journal of Mechanical Engineer-
ing Science, 218:131–142, 2004.

[216] J. Wei. On-line surveillance monitoring of gearboxes. Application note


CM3067, SKF Reliability Systems, San Diego, California, US, 2003.

[217] WHTP. Wind & Hydropower Technologies Program: History of Wind


Energy. http://www.eere.energy.gov/windandhydro/wind history.html,
US, 2005.

[218] R. Willis. Principles of Mechanism, 2nd Edition. London: Longmans,


Green and Co, 1870.

[219] R. E. Wilson, S. N. Walker, and P. Heh. Technical and user’s manual


for the FAST-AD advanced dynamics code. OSU/NREL Report 99-
01, Department of Mechanical Engineering, Oregon State University,
Corvallis, US, 1999.

[220] Windpower. http://www.windpower.org. 2005.

[221] WinWinD. Product Information. http://www.winwind.fi, Finland, 2005.

[222] D. Witcher. Seismic analysis of wind turbines in the time domain. Wind
Energy, 8(1):81–91, 2004.
Bibliography 235

[223] X. Yuan and L. Cai. Variable amplitude Fourier series with its applica-
tion in gearbox diagnosis–Part I: Principle and simulation. Mechanical
Systems and Signal Processing, 19(5):1055–1066, 2005.

[224] X. Yuan and L. Cai. Variable amplitude Fourier series with its applica-
tion in gearbox diagnosis–Part II: Experiment and application. Mechan-
ical Systems and Signal Processing, 19(5):1067–1081, 2005.

[225] X. Zhao, P. Maisser, and P. Tenberge. Stability analysis of a vari-


able speed wind turbine with power splitting transmission using multi-
flexible-body methodology. VDI-Berichte Nr. 1606, Schwingungen in
Anlagen und Maschinen, pp. 95-112, Veitshöchheim, Würzburg, Ger-
many, May 16-17, 2001.
236
Curriculum Vitae

Personal data

Joris Louis Magda Maria Erik Peeters


Address: Geerdegemstraat 114
B-2800 Mechelen, Belgium
E-mail: joris.peeters@lid.kviv.be
Place and date of birth: Hasselt, June 25th 1979
Nationality: Belgian

Professional experience

• since June 2006:


employee at Hansen Transmissions International NV, Division
R&D Technology

• September 2001 - May 2006:


Ph.D. student at the Katholieke Universiteit Leuven, Department
of Mechanical Engineering, Division Production engineering, Ma-
chine design and Automation (PMA)

237
238 Curriculum Vitae

Education

• 1996 - 2001: Student Mechanical-Electrotechnical Engineer (op-


tion Mechatronics), Faculty of Applied Sciences, Katholieke Uni-
versiteit Leuven

– June 1997, First year: Great distinction


– June 1998, Second year: Great distinction
– June 1999, Third year: Distinction
– June 2000, Fourth year: Great distinction
– July 2001, Fifth year: Great distinction
Master thesis: Design of a rehabilitation-device for arm-
trunk-coordination for stroke patients

• 1990-1996: Secondary school: Science-Mathematics at the Hu-


maniora Kindsheid Jesu, Hasselt
List of publications

International peer reviewed journal articles


[1] J. Peeters, D. Vandepitte, P. Sas. Analysis of internal drive train dy-
namics in a wind turbine, Wind Energy, 9(1):141-161, 2006.

Full papers in proceedings of international conferences


[1] J. Peeters, D. Vandepitte, P. Sas, S. Lammens. Comparison of analysis
techniques for the dynamic behaviour of an integrated drive train in a
wind turbine, Proceedings of ISMA 2002, International Conference on
Noise and Vibration Engineering, pp. 1397-1406, Leuven (Belgium),
September 16-18, 2002.

[2] J. Peeters, D. Vandepitte, P. Sas, W. Smook. Analysis of the dynamic


behaviour of an integrated drive train in a wind turbine, Proceedings of
the VDI conference Schwingungen in Antrieben 2003, pp. 287-300,
Fulda (Germany), April 2-3, 2002.

[3] J. Peeters, D. Vandepitte, P. Sas. Dynamic analysis of an integrated


drive train in a wind turbine, Proceedings of the European Wind En-
ergy Conference 2003, Madrid (Spain), CD-Rom, June 16-19, 2003.

[4] J. Peeters, D. Vandepitte, P. Sas. Flexible multibody model of a


three-stage planetary gearbox in a wind turbine, Proceedings of ISMA
2004, International Conference on Noise and Vibration Engineering,
pp. 3923-3942, Leuven (Belgium), September 20-22, 2004.

239
240 List of publications

[5] J. Peeters, D. Vandepitte, P. Sas. Multibody simulation of a three-stage


planetary gearbox in a wind turbine, Proceedings of DEWEK 2004,
7th German Wind Energy Conference, CD-Rom, Wilhelmshaven (Ger-
many), October 20-21, 2004.

[6] J. Peeters, D. Vandepitte, P. Sas. Analysis of internal drive train dy-


namics in a wind turbine, Proceedings of the European Wind Energy
Conference 2004, London (UK), CD-Rom, November 22-25, 2004.

Full papers in proceedings of national conferences


[1] J. Peeters, D. Vandepitte, P. Sas. Dynamic analysis of an integrated
drive train in a wind turbine, Proceedings of the 6th National Con-
ference on theoretical and applied Mechanics, Ghent (Belgium), CD-
Rom, May 27-28, 2003.
Appendix A

Drive train modes in traditional


wind turbine design codes

A.1 Introduction
Chapter 3 describes the state-of-the-art in the traditional wind turbine design
codes and introduces a general concept for the structural model, which is used
in these codes. Section 3.3.3.3 describes the importance of the so-called drive
train modes for the simulation of the drive train loads in such codes. This
appendix gives proper insight in the model used to calculate these modes and
investigates the influence of various parameters on the corresponding eigenfre-
quencies. The overall scope of the present analysis is limited to the frequency
range [ 0 - 10 Hz], which includes three drive train modes.

For the analysis of the drive train modes, two models are build, which are
considered to be similar to the structural model in the traditional wind turbine
design codes. In addition, a popular industrial simplification of the structural
model for the calculation of the 1st drive train frequency is discussed. The
appendix is split up in three parts.

1. Section A.2 describes the analysis of an FE model of the wind turbine.


The modelling of the rotor, the tower and the complete wind turbine are
separately introduced, with a description of their eigenmodes and corre-
sponding frequencies. These results correspond to the typical results of
normal modes calculations in traditional design codes, as discussed in
section 3.3.3.

2. Section A.3 discusses an industrial approach for the calculation of the


1st drive train mode.

241
242 A. Drive train modes in traditional wind turbine design codes

3. Section A.4 presents an investigation of the influence of various model


parameters on the eigenfrequencies of the drive train modes. Firstly, the
FE model of the wind turbine is used for the sensitivity analyses of the
rotor properties, the drive train stiffness and the generator inertia (sec-
tion A.4.1). For the sensitivity analyses of the tower properties and the
coupling between the drive train and the tower, a more detailed model
of the drive train is required. Section A.4.2 introduces therefore a flex-
ible multibody model of the wind turbine. Section A.4.3 discusses the
sensitivity analyses based on this latter model.

All model inputs are based on a generic wind turbine, representative for a
modern multi-megawatt wind turbine (>2 MW). The rotor of this wind tur-
bine is carried by a main bearing, which is integrated in the gearbox (cfr. fig-
ure 2.17(c)).

A.2 FE model of the wind turbine


A.2.1 Rotor
The wind turbine rotor consists of three equal blades clamped at the rotor hub.
Each individual blade is modelled with 28 one-dimensional beam elements.

No. Description Eigenfrequency (Hz)


1 1st flapwise 1.15
2 1st edgewise 1.49
3 2nd flapwise 3.09
4 2nd edgewise 4.46
5 3rd flapwise 6.05
6 3rd edgewise 9.49
7 4th flapwise 10.0
8 1st torsion 14.7
9 5th flapwise 15.0
10 4th edgewise 16.8

Table A.1: Results of a normal modes analysis of a single blade clamped at the
root.

Table A.1 shows the results of a normal modes calculation for such a blade
clamped at the root. These results comply with the general rules introduced in
section 3.3.3:

• The flapwise mode in a pair of bending modes is always lowest in fre-


quency.
A.2 FE model of the wind turbine 243

• A blade is stiff in torsional direction and, thus, the corresponding mode


occurs at a relatively high frequency. As a result, the blade torsion is gen-
erally omitted [121], unless the occurrence of flutter is analysed. Since
the scope in this section is limited to a frequency of 10 Hz, the torsional
mode at 14.7 Hz is out of the scope.
Three of these blade models are connected with their roots at a hub, which
is represented by a point mass, and compose in this way the complete rotor.
When the hub is rigidly connected to the ground (fixed), all modes of an indi-
vidual blade are found logically three times, as shown in table A.2.

No. Description Eigenfrequency (Hz)


1 1st rotor flapwise A 1.15
2 1st rotor flapwise B 1.15
3 1st rotor flapwise C 1.15
4 1st rotor edgewise A 1.49
5 1st rotor edgewise B 1.49
6 1st rotor edgewise C 1.49
7 2nd rotor flapwise A 3.09
8 2nd rotor flapwise B 3.09
9 2nd rotor flapwise C 3.09
10 2nd rotor edgewise A 4.46
11 2nd rotor edgewise B 4.46
12 2nd rotor edgewise C 4.46
13 3rd rotor flapwise A 6.05
14 3rd rotor flapwise B 6.05
15 3rd rotor flapwise C 6.05
16 3rd rotor edgewise A 9.49
17 3rd rotor edgewise B 9.49
18 3rd rotor edgewise C 9.49

Table A.2: Results of a normal modes analysis of the rotor with a hub fixed to
the ground.

Finally, the degree of freedom from the hub which represents the drive train
rotation is set free. Consequently, in the results of a normal modes analysis
for this model, a rigid-body mode is present. In addition, for each triple of
rotor edgewise modes, one mode is shifted in frequency. This is the symmetric
edgewise rotor mode, which has a non zero torque component on the hub. The
increase in frequency for the 1st symmetric edgewise mode - from 1.49 Hz to
2.77 Hz - is a consequence of removing the fixed boundary. After all, removing
the fixation of the hub can be seen as removing a huge inertia, which causes an
increase in eigenfrequency.
244 A. Drive train modes in traditional wind turbine design codes

No. Description Eigenfrequency (Hz)


0 rigid-body mode 0
1 1st rotor flapwise A 1.15
2 1st rotor flapwise B 1.15
3 1st rotor flapwise C 1.15
4 1st rotor edgewise A 1.49
5 1st rotor edgewise B 1.49
6 1st rotor torsion (edgewise C) 2.77
7 2nd rotor flapwise A 3.09
8 2nd rotor flapwise B 3.09
9 2nd rotor flapwise C 3.09
10 2nd rotor edgewise A 4.45
11 2nd rotor edgewise B 4.45
12 3rd rotor flapwise A 6.05
13 3rd rotor flapwise B 6.05
14 3rd rotor flapwise C 6.05
15 2nd rotor torsion (edgewise C) 7.05
16 3rd rotor edgewise A 9.49
17 3rd rotor edgewise B 9.49
18 4th rotor flapwise A 10.0
19 4th rotor flapwise B 10.0
20 4th rotor flapwise C 10.0
21 3rd rotor torsion (edgewise C) 12.9

Table A.3: Results of a normal modes analysis of the rotor with a torsionally
free hub.

A.2.2 Tower
The tower is a tubular steel structure and is modelled with 29 one-dimensional
beam elements. Table A.4(a) presents the results of a normal modes calcula-
tion for this structure clamped at the ground. Secondly, table A.4(b) shows the
results for an extension of this model with an extra discrete mass on top, which
represents the inertia of all nacelle components, rotor inclusive.

Since the tower is a symmetric structure, there is no distinction between a


transversal or a longitudinal bending mode for the model without a top mass
element in table A.4(a). For the model with a top mass element, this distinc-
tion can be identified, as indicated in table A.4(b). The transversal bending
mode is typically lowest in frequency. Adding the extra mass element on top
has furthermore a considerable impact on the torsional tower modes: the 1st
torsion mode shifts from 16.1 Hz to 1.98 Hz.
A.2 FE model of the wind turbine 245

Eigen- Eigen-
No. Description frequency No. Description frequency
(Hz) (Hz)
1 1st twr bending 0.87 1 1st twr bending t 0.34
2 1st twr bending 0.87 2 1st twr bending l 0.34
3 2nd twr bending 3.67 3 1st twr torsion 1.98
4 2nd twr bending 3.67 4 2nd twr bending t 2.05
5 3rd twr bending 9.05 5 2nd twr bending l 2.25
6 3rd twr bending 9.05 6 3rd twr bending t 4.31
7 1st twr torsion 16.1 7 3rd twr bending l 4.82
8 4th twr bending 16.8 8 4th twr bending t 9.21
9 4th twr bending 16.8 9 4th twr bending l 9.30
10 2nd twr torsion 21.1 10 2nd twr torsion 11.1

(a) Tower without top mass. (b) Tower with a discrete element on top,
which represents the mass and inertia of
the rotor, the hub and all nacelle compo-
nents.
Table A.4: Results of a normal modes analysis of the tower (twr) clamped at
the ground. (t: transversal; l: longitudinal)

A.2.3 Wind turbine


The FE model of the complete wind turbine is implemented by coupling the
model of the rotor with the model of the tower. Figure A.1 gives a schematic
representation of this model:
• The mass and inertia of the nacelle and its components are included as
a discrete mass element at the tower top node. This is the same element
as was used in the model of table A.4(b), however, the mass and inertia
of the rotor are no longer included here.
• The hub is included as a discrete mass in the rotor model and its rota-
tional DOF around the drive train axis is connected through a torsional
spring with a rigid generator. All other DOFs of the hub are rigidly
connected with the tower top.
• The torsional drive train stiffness (KDT ) is referred to the rotor side and
equals 5 GNm/rad. This value is estimated as an upper limit of the drive
train stiffness and corresponds to a stiff drive train design, as a result of
the absence of a main shaft in this wind turbine concept. A sensitivity
analysis investigates the influence of decreasing the stiffness value.
• The generator inertia (Igenerator ) equals 147 kgm2 and is converted to a
value, which is referred to the rotor side, by multiplying it with the gear-
246 A. Drive train modes in traditional wind turbine design codes

flexible rotor torque path

KDT
hub generator

tower top

flexible tower
load path
(exclusive torque)

Model specifications
rotor inertia (Irotor ) 4.38E6 kgm2
drive train stiffness (KDT ) 5 GNm/rad (referred to the rotor side)
gear ratio 100
generator inertia (Igenerator ) 1.47E6 kgm2 (referred to the rotor side)

Figure A.1: Schematic representation of the FE model of the wind turbine


(model A).

box ratio squared. The gearbox ratio equals 100. Initially, the generator
has pinned boundary conditions, which means that it can rotate freely
around the drive train axis (model A). This implies that no (reaction)
torque acts upon the tower top node. This is not realistic, but acts as
a good starting point for further analysis of the torque coupling to the
tower. A second model includes a rigid connection between the gener-
ator rotation and the tower top node, which implies a direct torque path
from the drive train to the tower (model B). This representation is al-
ready closer to reality and, since all presented analyses are at standstill,
this corresponds to the situation for the wind turbines discussed in sec-
tion 3.3.3. A more elaborate discussion of the influence of the coupling
between the drive train and the tower is given in section A.4.3.
A.2 FE model of the wind turbine 247

• The influence of the small cone and tilt angles is neglected in this model.

Table A.5 shows the results of a normal modes calculation for the wind turbine
model with one blade in horizontal position and all blades pitched for normal
operation. The table includes results for model A with a pinned generator, as
well as for model B with a generator fixed to the tower top. The calculated
eigenfrequencies and corresponding eigenmodes match very well with the or-
der and type of results found in the traditional wind turbine design codes (cfr.
table 3.1). A comparison of the results for model A, with those calculated
for the individual rotor and tower models as well as with model B yields the
following conclusions:

• The first bending modes of the tower can be identified approximately


with a model of the tower only, including a mass element on top (cfr.
table A.4).

• The three first flapwise rotor modes of the wind turbine lie close to the
triple of modes calculated for the model of the rotor individually (cfr.
table A.3).

• The two first asymmetric edgewise rotor modes of the wind turbine
hardly shift in frequency, compared with the model of the rotor indi-
vidually (cfr. table A.3).

• The frequencies of the rotor torsion modes from model A lie below the
values found for the model of the pinned individual rotor (cfr. table A.3).
Moreover, the corresponding frequencies calculated for model B de-
crease further in frequency and lie below the values found for the fixed
individual rotor (cfr. table A.2). Except for the 20th and the 22nd mode,
only these modes change in frequency when the generator is fixed. This
ratifies their name of drive train modes and limits the scope in the next
analyses to these modes.
248 A. Drive train modes in traditional wind turbine design codes

Eigen-
No. Description frequency
(Hz)
A B
0 rigid-body mode (drive train rotation) 0 -
1 1st tower longitudinal 0.33 0.33
2 1st tower transversal 0.34 0.33
3 1st asymmetric rotor flap/yaw (A) 1.03 1.03
4 1st asymmetric rotor flap/tilt (B) 1.11 1.11
5 1st symmetric rotor flap (C) 1.17 1.17
6 1st asymmetric rotor edge (B) 1.49 1.49
7 1st asymmetric rotor edge (C) 1.50 1.50
8 1st rotor torsion (A) 2.23 1.27
9 2nd asymmetric rotor flap/tilt + 2nd tower bending 2.45 2.45
10 2nd asymmetric rotor flap/yaw + 2nd tower bending 2.53 2.45
11 2nd asymmetric rotor flap/yaw + 1st tower torsion 2.70 2.67
12 2nd symmetric rotor flap 3.10 3.10
13 2nd asymmetric rotor flap/tilt 3.18 3.18
14 2nd asymmetric rotor edge (B) 4.44 4.44
15 2nd asymmetric rotor edge (C) 4.46 4.46
16 3rd asymmetric rotor flap/yaw + tower torsion 4.88 4.88
17 2nd rotor torsion 5.05 4.00
18 3rd asymmetric rotor flap/tilt 5.63 5.63
19 3rd symmetric rotor flap 6.10 6.10
20 3rd asymmetric rotor flap/yaw + 3rd tower bending 6.87 5.80
21 3rd asymmetric rotor flap/tilt + 3rd tower bending 6.90 6.90
22 3rd asymmetric rotor flap/yaw + 2nd tower torsion 7.67 7.54
23 3rd asymmetric rotor edge (B) 9.28 9.28
24 3rd rotor torsion 9.47 9.01
25 3rd asymmetric rotor edge (C) 9.55 9.55
26 4th asymmetric rotor flap/yaw 9.74 9.74
27 4th symmetric rotor flap 10.1 10.1
28 4th asymmetric rotor flap/tilt 10.3 10.1

Table A.5: Results of a normal modes analysis of the wind turbine model with
one blade in horizontal position and all blades pitched for normal operation.
The generator is pinned in model A and fixed to the tower top in model B.
A.3 An industrial approach to calculate the 1st drive train mode 249

A.3 An industrial approach to calculate the 1st drive


train mode
Although the importance of the 1st drive train mode is never underestimated in
the literature, common related discussions are often based on a model descrip-
tion for the drive train which is oversimplified. [37] describes this drive train
model as a connection of the following elements in series:

• a rigid body with a rotational inertia representing the wind turbine rotor,

• a torsional spring representing the drive train stiffness,

• a rigid body with a rotational inertia representing the generator rotor,

• a torsional damper representing the generator torque as a function of the


slip (cfr. section 2.3.3.1),

• a body with an infinite rotational inertia which is the mechanical equiv-


alent of the electrical grid with a constant frequency.

Figure A.2 sketches this modelling approach which corresponds very much to
a popular intuitive representation of the drive train as a rigid rotor and a rigid
generator connected by the drive train stiffness. The influence of the generator
torque characteristic is often generalised as “fixed” or “free”. The former term
corresponds to the operation of a traditional asynchronous generator where a
steep torque-slip curve allows only a negligible rotation of the generator rela-
tive to the mechanical equivalent of the electric grid. The latter term describes
the characteristic corresponding to the typical behaviour of a variable speed
generator which implies no considerable limitation on the relative rotation of
the generator. This generalisation is described in [203]. It makes the drive train
model equal to a “single-mass system” in the former case and a “two-mass sys-
tem” in the latter case. The corresponding drive train mode is often called a
free-fixed mode or a free-free mode respectively. Equations (A.1) and (A.2)
describe the formulas for the calculation of the corresponding eigenfrequen-
cies. [122] describes also the use of equation (A.2) for the approximation of
the free-free eigenfrequency.
r
1 KDT
ffree−fixed = · (A.1)
2π Irotor

v !
u
1 u 1 1
ffree−free = · tKDT · + (A.2)
2π Irotor Igenerator · i2gear
250 A. Drive train modes in traditional wind turbine design codes

Irotor the rotor inertia


KDT torsional drive train stiffness
Igenerator the generator inertia
Cgenerator the torsional damping characteristic of the generator torque

Figure A.2: A popular model for the drive train in a wind turbine [37]. The
torsional drive train stiffness and the generator inertia should be referred to the
rotor side.

The wind turbine model A in section A.2.3 has a free generator. Based on the
given specifications, the free-free eigenfrequency of the drive train mode is:
s  
1 1 1
ffree−free = · 5E9 · + Hz
2π 4.38E6 147 · 1002
= 10.7 Hz

This is by far not the “1st rotor torsion frequency” of 2.23 Hz calculated with
the FE model of the wind turbine (cfr. table A.5), although the “only” principal
difference between the models is the omission of the rotor flexibility in the
equivalent two-mass system. The big difference indicates that this flexibility
cannot be neglected and, consequently, the equivalent two-mass model is not a
valid representation for the drive train. The two-mass system can only be valid
when the drive train stiffness only is by far the most determining flexibility
in the structure or, in other words, when the eigenfrequencies of the rotor and
tower individually lie much higher than the free-free eigenfrequency. This is
not the case for a wind turbine, where the overall behaviour is determined by
a combined effect of the following parameters:
1. the (distributed) rotor inertia and flexibility (including the pitch angle
and rotor position)
2. the drive train stiffness
3. the generator inertia
4. the coupling between the drive train and the tower top (e.g. the gearbox
support and the generator characteristic)
5. the (distributed) tower inertia and flexibility
A.3 An industrial approach to calculate the 1st drive train mode 251

An accurate drive train model should yield a drive train eigenfrequency which
matches well with the eigenfrequency measured in the torque signal. When
this latter frequency is known, it is possible to “tune”1 the torsional drive train
stiffness KDT in equation (A.2) to find the correct eigenfrequency. Figure A.3
shows how this frequency changes with KDT . However, the equivalent two-
mass model does not represent the physical reality, which can lead to wrong
interpretations or conclusions, e.g. when applying design changes in order to
avoid resonances as described in [37]. The sensitivity analyses in the next
section describe the influence of the different model parameters on the drive
train modes and confirm the insufficiency of the equivalent two-mass model as
a “rule of thumb”.
Natural frequency [Hz]

Torsional drive train stiffness KDT [Nm/rad]

Figure A.3: Influence of the torsional drive train stiffness KDT on the free-free
eigenfrequency calculated in equation (A.2). The cursor indicates the initial
frequency calculated for the specifications in figure A.1.

1 “Tuning” means here changing the torsional drive train stiffness value in the two-mass

model to get a good correlation between the measured and simulated eigenfrequency. Since
this stiffness value is generally hard to assess in the design phase, it is often considered as an
appropriate parameter to tune.
252 A. Drive train modes in traditional wind turbine design codes

A.4 Sensitivity analyses


This section presents an investigation of the influence of the five parameters
listed in the previous section, on the eigenfrequencies of the drive train modes.
Model A in figure A.1 is used for the sensitivity analyses of the rotor prop-
erties, the drive train stiffness and the generator inertia. The drive train fre-
quencies of this model are initially calculated at 2.23 Hz, 5.05 Hz and 9.47 Hz
respectively (cfr. table A.5 - model A).

Section A.4.2 presents a flexible multibody model of the wind turbine, which
includes a more detailed model of the drive train. This is required for the sensi-
tivity analyses of the tower properties and the coupling between the drive train
and the tower, which are discussed in section A.4.3.

Each of the following analyses is limited to the investigation of one single


parameter’s influence only and aims at giving insight in their contribution to
the overall behaviour.

A.4.1 Sensitivity analyses for the FE model


A.4.1.1 Rotor inertia and flexibility
Although the FE model of the rotor is considered as accurately representing
reality (i.e. no model updating is needed), a sensitivity analysis is done for its
stiffness and inertia as a theoretical exercise to describe their influence. This
analysis occurs by a variation in the FE model of the E-modulus and the density
of the blade material respectively. Figure A.4 shows how the frequencies of the
first two drive train modes change when the E-modulus of the blades is multi-
plied with a stiffness factor. The first frequency asymptotically approaches to
10.7 Hz for infinitely stiff blades, which corresponds logically to the frequency
calculated for the two-mass system in section A.3, since the rotor acts here as a
rigid body. The second frequency goes to infinity for rigid blades2 . Both drive
train modes decrease towards 0 Hz for stiffness values lower than one.

Figure A.5 shows how the 1st and 2nd drive train frequencies change when the
density of the blades is multiplied with a factor. Both frequencies approach to
zero for high density and to infinity for low density values.

2 The Modal Assurance Criterion (MAC) [93] is used to track the drive train modes in all

sensitivity analyses, since it is not possible to track them by only looking at the calculated
eigenfrequencies.
A.4 Sensitivity analyses 253

Natural frequency [Hz]

Natural frequency [Hz]


Stiffness factor Stiffness factor
(a) 1st rotor torsion mode. (b) 2nd rotor torsion mode.

Figure A.4: Sensitivity of the 1st and 2nd drive train frequencies to the blade
flexibility. The stiffness factor is the factor multiplied with the original E-
modulus of the blade material and the cursor indicates the frequency for the
reference stiffness.
Natural frequency [Hz]

Natural frequency [Hz]

Density factor Density factor


(a) 1st rotor torsion mode. (b) 2nd rotor torsion mode.

Figure A.5: Sensitivity of the 1st and 2nd drive train frequencies to the blade
material density value. The density factor is the factor multiplied with the
original blade density and the cursor indicates the frequency for the reference
density.

In addition, the influence of the rotor position and the pitch angle variation
on the drive train modes is also investigated. The first analysis demonstrates
that the shift in eigenfrequencies for different rotor positions is negligible. The
scope in the second analysis is limited to the first 9 modes in table A.5, which
includes only one drive train mode. Generally, the variation of the pitch angle
254 A. Drive train modes in traditional wind turbine design codes

during normal operation is rather limited; e.g. for a pitch controlled wind tur-
bine the angle varies typically from a few degrees below zero up to maximum
20◦ . However, when the turbine stops, the pitch angle can increase to 90◦ .
Since each blade is symmetric, it is sufficient to analyse a variation of 0◦ up
to 90◦ . After all, the eigenfrequencies for small negative pitch angles are the
same as for small positive values. In this analysis, the three blades are pitched
simultaneously.

The eigenfrequencies found for a pitch angle of 0◦ correspond to those from


table A.5. The rigid-body mode and the first tower bending modes of this list
are hardly influenced by a pitch angle variation. On the contrary, the frequen-
cies of the next rotor modes change drastically as demonstrated in figure A.6.
The change in pitch angle causes a gradual shift from flapwise to edgewise
rotor frequencies and vice versa, since flapwise becomes gradually in-plane
and edgewise out of the rotor plane. It is furthermore clear from the plot that
the frequency lines intersect, which indicates that the order of mode shapes
switches. The interpretation of the mode shapes as described in table A.5 is
only valid for a pitch angle of 0◦ , however, it is possible to track the switching
of several modes by using a MAC matrix.

Figure A.7 shows this MAC matrix. It indicates the correlation between the
mode shapes calculated at a pitch angle of 0◦ and 90◦ respectively3 . Based on
these results, it can be concluded that the asymmetric edgewise rotor modes at
a pitch angle of 0◦ (6th and 7th mode) correspond to the asymmetric flapwise
rotor modes at a pitch angle of 90◦ . Vice versa, the 4th and 5th flapwise rotor
modes at an angle of 0◦ correspond to edgewise rotor modes at an angle of 90◦ .
This can be understood from the fact that the three corresponding flapwise fre-
quencies increase, since the stiffer edgewise direction becomes now out of the
rotor plane and the more flexible flapwise direction in-plane. The 1st drive
train frequency decreases, since it is now determined by the more flexible flap-
wise blade stiffness. There is also a quite high correlation between this mode
and the rigid-body mode of the drive train, which is explained by the similarity
between the drive train rotation (rigid-body mode) and the drive train torsion
(drive train mode).

3 The MAC value lies in the range [0 1] and a high value means a good correlation.
A.4 Sensitivity analyses 255

Natural frequency [Hz]

Pitch angle [◦ ]

Figure A.6: Shift in eigenfrequencies for a variation in pitch angle. The num-
ber of the modes corresponds to the results from table A.5.

θ = 90◦ θ = 0◦

Figure A.7: MAC matrix which represents the correlation between the mode
shapes calculated for a pitch angle of 0◦ and 90◦ .
256 A. Drive train modes in traditional wind turbine design codes

A.4.1.2 Drive train stiffness

The torsional drive train stiffness implemented in the FE model represents a


torsional equivalent for the flexibility of all components between the rotor hub
and the generator. The determination of such a stiffness value is discussed
more elaborately in section 6.4. Here, an initial estimation of 5.0 GNm/rad is
used as a reference value. Figure A.8 shows how the frequencies of the drive
train modes change with a variation of this drive train stiffness. Remarkable
is the fact that only the frequencies of these modes change with varying drive
train stiffness, which further ratifies the terminology: “drive train modes”. Fig-
ure A.8(a) shows how the frequency of the first drive train mode changes. This
is a completely different behaviour than shown in figure A.3 for the simplified
two-mass model in figure A.2, which confirms the insufficiency of the simpli-
fied model.
Natural frequency [Hz]

Natural frequency [Hz]

Drive train stiffness [Nm/rad] Drive train stiffness [Nm/rad]


(a) 1st drive train mode (b) 2nd drive train mode
Natural frequency [Hz]

Natural frequency [Hz]

Drive train stiffness [Nm/rad] Drive train stiffness [Nm/rad]


(c) 3rd drive train mode (d) first 3 drive train modes

Figure A.8: Sensitivity of the drive train frequencies to the torsional drive train
stiffness. The cursors indicate the eigenfrequencies for the reference value
KDT = 5.0 GNm/rad.
A.4 Sensitivity analyses 257

From figure A.8(d) it can be seen that the points of inflection of the different
frequency curves move to higher stiffness values for the higher modes. This
means that the influence of the drive train stiffness is bigger for the higher drive
train modes. Moreover, it can be seen from this plot that when the stiffness
value decreases, the frequencies approach the results given in table A.3 for
the individual rotor mounted torsionally free. The first drive train mode shape
becomes the rigid-body mode, the second mode shape becomes the first and
the third one becomes the second. This is confirmed by the calculation of the
MAC values which indicates these mode switches, however, these switches
are not indicated in the respective figures. This means that the influence of the
drive train stiffness on the drive train modes disappears for very low stiffness
values: the rotor acts here as if it is not coupled with the generator.

A.4.1.3 Generator inertia


The generator inertia of 147 kgm2 is used as the reference value. Figure A.9
shows how the three drive train frequencies change with a varying generator
inertia. Again, only these frequencies change in this sensitivity analysis. For
high inertia values, the frequencies approach the results from table A.2 for an
individual rotor with a fixed hub. Due to the influence of the drive train stiff-
ness, the asymptotic values lie a bit below those frequencies. Analogously,
the results for small inertia values approach the results from table A.3 for an
individual rotor which is torsionally free.

From a comparison of figure A.8(a) and figure A.9(a), it can be concluded that
a small variation of the generator inertia has a much bigger impact on the first
drive train frequency of this wind turbine, than a small variation of the drive
train stiffness. This means that, if a designer wants to change this frequency,
an adaptation of the generator inertia is much more effective. This conclusion
cannot be drawn based on the simplified model of figure A.2. On the contrary,
the use of this model could lead to false conclusions, since an adaptation of the
drive train stiffness seems also effective as demonstrated in the figure A.3.
258 A. Drive train modes in traditional wind turbine design codes

Natural frequency [Hz]

Natural frequency [Hz]


Generator inertia [kgm2 ] Generator inertia [kgm2 ]
(a) 1st drive train mode (b) 2nd drive train mode
Natural frequency [Hz]

Natural frequency [Hz]

Generator inertia [kgm2 ] Generator inertia [kgm2 ]


(c) 3rd drive train mode (d) first 3 drive train modes

Figure A.9: Sensitivity of the drive train frequencies to the generator in-
ertia. The cursor indicates the eigenfrequencies for the reference value
Igenerator = 147 kgm2 .

A.4.2 Flexible multibody model of the wind turbine


In order to analyse the influence of the tower and the coupling between the
drive train and the tower on the drive train dynamics in a proper fashion, it is
necessary to include the physical principle of a gearbox in the FE model of
the wind turbine. This principle is shown in figure A.10 for the analysed wind
turbine and is discussed more generally in chapter 4. The schematic illustra-
tion indicates the various load flows and focusses on the torque flow. The drive
train includes a gearbox with three gear stages. Each stage is represented by
a torque split, with on the one hand a path towards the gearbox housing and
on the other hand a path towards the generator. The former torque path is in
connection with the tower top via the gearbox support, which is further called
A.4 Sensitivity analyses 259

the torque arm. The flexibility of this support is represented by the torque arm
stiffness. Since the gearbox ratio equals 100, the resulting torque flow towards
the generator is only 1% of the input torque. The generator controller defines
the torque on the generator rotor and the resulting torque on the generator sta-
tor, which is mounted on the tower top.

Figure A.10: Schematic overview of the load flow in the wind turbine.

The drive train in the FE model of the wind turbine, which is introduced in
section A.2 and used in the sensitivity analyses of the previous section, is a
torque path represented by an equivalent spring between the flexible rotor and
the generator. Furthermore, all inertia values from the gearbox components
are assumed negligible and no torque path is coupled with the tower top. It is
clear from figure A.10 that this differs from physical reality. However, the im-
plementation of the torque split is not straightforward in a standard FE model.

Therefore, a flexible multibody model of the complete wind turbine is built,


including a detailed representation of the gearbox. This model is presented in
section 6.5.2. It includes four pairs of normal bending modes for each blade
and the first ten normal modes of the tower. The generator in the resulting
model is initially modelled as free, which means that the generator controller
in figure A.10 does not apply any torque. The eigenfrequencies of the drive
train modes in the resulting model are at 2.03 Hz, 4.42 Hz and 8.71 Hz respec-
260 A. Drive train modes in traditional wind turbine design codes

tively. These are the reference values in the sensitivity analyses discussed in
the next section.

Note that the difference between the drive train frequencies calculated in the
FE model with a free generator (cfr. model A - table A.5) and in the flex-
ible multibody model is a result of the difference in drive train stiffness in
both models. In the former model a reference value of 5 GNm/rad is used,
whereas the gearbox stiffness in the latter model equals 1.5 GNm/rad (cfr.
section 6.4.1). This does not influence the trends presented in the following
sensitivity analyses.

A.4.3 Sensitivity analyses for the flexible multibody model


A.4.3.1 Tower flexibility
Similar to the sensitivity analysis of the rotor properties, the influence of the
tower properties on the drive train modes is investigated here, but the analysed
properties are limited to the flexibility only. Figure A.11 shows how the eigen-
frequencies of the drive train modes change with a variation of the bending
stiffness of the tower. This variation is a result of multiplying the E-modulus
with a stiffness factor. The three drive train modes clearly decrease in fre-
quency for decreasing bending stiffness, which indicates the influence of the
tower flexibility on the drive train dynamics. For stiffer tower designs, the first
drive train mode increases only slightly in frequency, which indicates that it is
near its asymptote or, in other words, the tower’s flexibility can be neglected
in the calculation of the first drive train frequency for this design. On the other
hand, the asymptotes for the second and third drive train mode lie higher in
frequency.

Note that for a similar analysis based on the FE model described in figure A.1,
there is no influence on the frequencies of the drive train modes, since there
is no coupling between the torque in the drive train and the tower. This em-
phasises the importance of using a correct drive train representation, even in
simplified equivalent models.
A.4 Sensitivity analyses 261

Natural frequency [Hz]

Natural frequency [Hz]


Stiffness factor Stiffness factor
(a) 1st drive train mode (b) 2nd drive train mode
Natural frequency [Hz]

Natural frequency [Hz]

Stiffness factor Stiffness factor


(c) 3rd drive train mode (d) first 3 drive train modes

Figure A.11: Sensitivity of the drive train frequencies to the bending stiffness
of the tower. The stiffness factor is the factor multiplied with the E-modulus
of the tower material. The cursor indicates the frequencies for the reference
stiffness.

A.4.3.2 The coupling between drive train and tower


Figure A.10 shows two torque paths from the drive train to the tower in the
presented wind turbine. The first is through the torque arm and the second
via the generator support. The torque in this latter path is determined by the
generator controller. The influence of both torque paths on the drive train
dynamics is investigated individually.

Torque arm
The influence of the torque arm is analysed by a variation of the corresponding
stiffness value in the flexible multibody model. This stiffness represents all
flexibilities between the gearbox housing and the tower top. For the analysed
design, a reference value of 10.0 GNm/rad is taken, which corresponds to a
very stiff connection. Torque arm stiffness values can be a factor one hundred
lower for wind turbine gearboxes mounted on rubber bushings, as is typically
262 A. Drive train modes in traditional wind turbine design codes

done in the drive train concepts described in figures 2.17(a) and 2.17(b).
Natural frequency [Hz]

Natural frequency [Hz]


Torque arm stiffness [Nm/rad] Torque arm stiffness [Nm/rad]
(a) 1st drive train mode (b) 2nd drive train mode
Natural frequency [Hz]

Natural frequency [Hz]

Torque arm stiffness [Nm/rad] Torque arm stiffness [Nm/rad]


(c) 3rd drive train mode (d) first 3 drive train modes

Figure A.12: Sensitivity of the drive train frequencies to the torque arm stiff-
ness. The cursor indicates the frequencies for the reference stiffness value of
10.0 GNm/rad.

Figure A.12 shows how the drive train frequencies change with a varying
torque arm stiffness. The changes in frequency of the first and second drive
train mode are similar to what was calculated for a variation in drive train stiff-
ness as shown in figure A.8. This means that the influence of their stiffness
works in a similar fashion or, in other words, that they can be combined as
springs in series, keeping in mind that they both need to be referred to the ro-
tor side.

Note that the variation of the third drive train frequency correlates less with the
results for a variation of the drive train stiffness in figure A.8. This is a result of
the influence of various tower bending modes in the flexible multibody model
used here, which is not existing in the FE model.
A.4 Sensitivity analyses 263

Generator characteristic
The generator type and controller define the characteristic behaviour of the
generator. This defines the instantaneous torque on the generator rotor and sta-
tor and, thus, the coupling between the drive train and the tower. This section
does not aim to investigate the exact influence of the generator on the wind
turbine, since this requires the implementation of elaborate and generator type
dependent models as described among others by Soens [197]. However, it
only aims at gaining insight in the influence of a simplified model for an asyn-
chronous generator, as used by Larsen [122] in a wind turbine model.

This model describes the behaviour of an asynchronous generator as a so-


called damper as presented in figure 2.15. The electric torque in such a gen-
erator can be described by the Kloss equation as a function of the generator
speed variation around its synchronous speed [16]. A linearisation of the cor-
responding Kloss curve yields a characteristic value for the damping action in
[Nm/(rad/sec)]. Based on realistic data of a wind turbine generator, it is con-
cluded that this damping can be in the order of 10-50 kNm/(rad/sec). However,
Larsen describes that the simplified representation of the generator behaviour
implies an overestimation of the damping at higher frequencies. Therefore,
the reference value of 50 kNm/(rad/sec) is considered as the maximum that
can possibly occur.

The implementation of the damper in the flexible multibody model occurs by


adding a viscous damper between the generator rotor and the generator sup-
port. This implies that the Kloss curve is shifted from the synchronous gen-
erator speed to standstill, which is considered as a valid approximation, since
no other speed dependent factors are included. Since the eigenmode calcula-
tion for the flexible multibody model is based on a perturbation method (cfr.
appendix E), the influence of the damper is included in this analysis.

Figure A.13 shows how the first two drive train modes change in frequency for
a variation in damping value4 . At the reference value of 50 kNm/(rad/sec), the
results are quasi identical to what was calculated for the FE model of the wind
turbine with the generator fixed on top of the tower (cfr. model B in table A.5).
This corresponds to the term “free-fix”, which is used in industry for a drive
train with an asynchronous generator.

4 The third drive train mode is not longer considered, since it does not add value to this

sensitivity analysis because of its coupling with various tower bending modes as described for
the analysis of the torque arm.
264 A. Drive train modes in traditional wind turbine design codes

Frequency [Hz]

Frequency [Hz]
Damping value [Nms] Damping value [Nms]
(a) 1st drive train mode (b) 2nd drive train mode

Figure A.13: Sensitivity of the first two drive train frequencies to the damping
value of the generator, which is considered as a characteristic for its behaviour.
The cursor indicates the frequency for the reference value of 50 kNm/(rad/sec).

Investigation of the eigenfrequencies for lower damping values, indicates in


addition that the generator behaviour has a considerable influence: for damp-
ing values approaching zero, the generator acts completely free, which is con-
firmed by the good correlation with the results calculated for the FE model
with the pinned generator on top of the tower (cfr. model A in table A.5). This
corresponds to the term “free-free”, which is used in industry for a drive train
with a variable speed generator.

Although it seems that this sensitivity analysis gives reasonable physical re-
sults, the interpretation should still be done with care. Especially at the sensi-
tive part between the “free” and “fixed” behaviour, the validity of the results is
hard to assess, since only little experience with such high damping factors and
their numerical consequences is available. No further validation of the present
trend is given in this work, than a reference to the correspondence with what
is intuitively expected and generally accepted in industry.

A.5 Conclusions
An accurate prediction and interpretation of the drive train modes in a wind
turbine is important in the simulation of drive train loads. This appendix de-
scribes a thorough analysis of these modes using two models, which are similar
to the structural model description in a traditional wind turbine design code.
The first model is an FE model and the second model is a flexible multibody
A.5 Conclusions 265

model. The subject in this appendix is a generic wind turbine.

A step by step implementation of the FE model demonstrates the origin of the


drive train modes. This study starts with an analysis of an individual blade
clamped at the root which has sequential pairs of bending modes; each pair
consists of a flapwise mode and an edgewise mode, where the former is always
lowest in frequency. The stiffness of a blade in torsional direction is relatively
high and, as a result, the torsional modes are high in frequency and generally
not considered. Mounting three identical blades on a fixed rotor hub leads to
triples of the modes and eigenfrequencies found for the individual blade. Two
of these triples are always an asymmetric coupling of the blade bends and one
is symmetric. When the rotor is pinned at the hub, the symmetric edgewise ro-
tor modes increase in frequency, as a result of removing their boundary. These
modes become further the drive train modes.

The tower is a structure which determines the first eigenmodes of a wind tur-
bine. These are two bending modes and their frequencies can be defined ap-
proximately using a model of the tower only, including a discrete mass element
on top to represent the mass and inertia of the rotor and all other nacelle com-
ponents. The FE model of the wind turbine is a combination of the rotor and
the tower model, including a torsional spring and a torsional inertia to repre-
sent the drive train stiffness and the generator inertia respectively. The 1st drive
train mode in this model lies at 2.23 Hz for a pinned generator and at 1.27 Hz
when the generator is fixed on the tower top. The former result corresponds
to what is often called the “free-free” eigenfrequency and the latter one to the
so-called “free-fix” eigenfrequency.

[37] describes a popular simplified model to calculate the 1st drive train fre-
quency with only one spring element, to represent only the torsional stiffness
of the drive train in the wind turbine. However, based on the corresponding
equation (cfr. equation (A.2)) for this model, the free-free eigenfrequency of
the generic wind turbine equals 10.7 Hz. The large difference between this
value and the frequency 2.23 Hz, which is calculated using the FE model, in-
dicates an unacceptable omission of the important contribution of the rotor
flexibility in the simplified model. Nevertheless, since it is straightforward to
adapt the torsional drive train stiffness to make this frequency and the mea-
sured drive train frequency match, the simplified equation could apparently
yield an accurate result. However, the frequency variation for a variation of
the torsional drive train stiffness in the simplified model (cfr. figure A.3) dif-
fers completely from the result calculated for the FE model of the wind turbine
(cfr. figure A.8(a)). The 1st drive train frequency in this latter model (2.23 Hz)
lies at a maximum asymptote, when considering only variations of the drive
266 A. Drive train modes in traditional wind turbine design codes

train stiffness value. This indicates clearly one limitation of the simplified
model. The perceptibility of this limitation is enhanced here by the fact that
the torsional drive train stiffness in the generic wind turbine (5.0 GNm/rad) is
high. However, this does not undermine the general validity of the importance
of the rotor flexibility for the calculation of the 1st drive train frequency.

The influence of various other parameters on the drive train frequencies is anal-
ysed in the last part of this appendix. The FE model is used to investigate the
effect of a variation of the rotor flexibility, which confirms the importance of a
proper consideration of the rotor flexibility for the calculation of the 1st drive
train frequency. Additionally, the FE model is used in three other sensitivity
analyses. Changing the pitch angle (1) from 0◦ to 90◦ causes a decrease from
2.23 Hz to 1.66 Hz for the 1st drive train frequency. The influence of the rotor
position (2) was found to be negligible. A sensitivity analysis of the generator
inertia (3) demonstrates furthermore its considerable impact on the 1st drive
train frequency. This indicates, consequently, that an adaptation of this param-
eter can be an effective way of changing this frequency.

For the investigation of the influence of the tower properties and the coupling
between the drive train and the tower, it is necessary to have a realistic repre-
sentation of the torque path in the wind turbine. Therefore, a flexible multibody
model of the wind turbine is built, which includes a detailed representation of
the gearbox. This model is used in three sensitivity analyses.

1. The influence of the torsional stiffness of the gearbox support on the


drive train frequencies is similar to the effect of a variation of the drive
train stiffness. This latter variation is analysed using the FE model and,
since the results for a variation in support stiffness do not differ, they are
not included in the text.

2. A second study demonstrates how a decrease of the tower bending stiff-


ness yields lower drive train frequencies and vice versa.

3. The last sensitivity analysis tries to indicate the influence of the gener-
ator characteristic on the drive train frequencies. This characteristic is
drastically simplified and represented by the action of a viscous damper.
For high damping values, corresponding to the behaviour of an asyn-
chronous generator, the generator boundaries act as if it is “fixed”. For
low damping values, on the contrary, the generator acts as “free”.
Appendix B

Eigenmodes and
eigenfrequencies of modern
wind turbines

This appendix lists the eigenmodes and eigenfrequencies of various modern


wind turbines, which are found in literature and have been calculated using one
of the traditional wind turbine design codes. The overview contains a verbatim
description of the reproduced references, which indicates that the terminology
for the mode shapes applied in different publications is not always consistent.
Chapter 3 includes a summary of these results with a consistent terminology.

• Table B.1: a 500 kW wind turbine with 19 m blades (Bonus) [171]

• Table B.2: a stall regulated 600 kW wind turbine [124]

• Table B.3: a stall regulated 600 kW wind turbine with a rotor diameter
of 44 m (Bonus) [86]

• Table B.4: a 1800 kW wind turbine with a rotor diameter of 66 m [182]

• Table B.5: a fixed pitch, stall regulated, constant speed 2 MW wind


turbine with a rotor diameter of 76 m [122]

• Table B.6: a pitch regulated, variable speed 2.75 MW wind turbine [86]

• Table B.7: an example of a wind turbine (> 2 MW) in the HAWCModal


demo [181].

267
268 B. Eigenmodes and eigenfrequencies of modern wind turbines

Eigen- Description Eigenfrequency


mode (Hz)
1 1st tower transverse 0.75
2 1st tower longitudinal 0.80
3 1st rotor torsion 0.9
4 1st rotor flap A 1.39
5 1st rotor flap B 1.56
6 1st rotor flap C 1.85
7 1st rotor edge B 2.91
8 1st rotor edge C 2.93
9 2nd rotor flap A 3.54
10 2nd rotor flap B 4.22
Blade clamped at the root
1 1st flapwise 1.8
2 1st edgewise 2.9

Table B.1: Results of a normal modes calculation for a 500 kW wind turbine
with 19 m blades (Bonus) [171].

Eigen- Description Eigenfrequency


mode (Hz)
1 1st tower transverse 0.76
2 1st tower longitudinal 0.80
3 1st shaft torsion 0.92
4 1st asymmetric rotor flap/yaw 1.41
5 1st asymmetric rotor flap/tilt 1.52
6 1st symmetric rotor flap 1.81
7 2nd asymmetric rotor flap/yaw 2.85
8 1st rotor edge 3.08
9 1st rotor edge 3.23
10 2nd asymmetric rotor flap/tilt 3.54
11 3rd asymmetric rotor flap/tilt 4.21
12 2nd symmetric rotor flap 5.10

Table B.2: Results of a normal modes calculation for a stall regulated 600 kW
wind turbine [124].
269

Eigen- Description Eigenfrequency


mode (Hz)
1 1st rotor torsion 0.57
2 1st tower longitudinal 0.73
3 1st tower transverse 0.78
4 1st rotor flap A 1.24
5 1st rotor flap B 1.48
6 1st rotor flap C 1.76
7 1st rotor edge B 2.85
8 1st rotor edge C 2.95
9 2nd rotor flap A 3.43
10 2nd rotor flap B 3.74
Blade clamped at the root
1 1st flapwise 1.7
2 1st edgewise 2.94

Table B.3: Results of a normal modes calculation for a stall regulated 600 kW
wind turbine with a rotor diameter of 44 m (Bonus) [86].

Eigen- Description Eigenfrequency (Hz)


mode Blades Blades
0◦ pitched 90◦ pitched
1 1st tower transverse 0.418 0.417
2 1st tower longitudinal 0.419 0.420
3 1st rotor torsion 0.805 0.704
4 1st asymmetric rotor A (*) 0.979 1.002
5 1st asymmetric rotor B (yaw) 1.000 1.064
6 1st symmetric rotor (flap) 1.067 1.769
7 1st edgewise mode 1.857 1.032
8 2nd edgewise mode 1.045

Table B.4: Results of a normal modes calculation for a 1800 kW wind tur-
bine with a rotor diameter of 66 m [182]. (*) a supposed typing error in the
reference is corrected.
270 B. Eigenmodes and eigenfrequencies of modern wind turbines

Eigen- Description Eigenfrequency


mode (Hz)
1 1st tower transverse 0.39
2 1st tower longitudinal 0.40
3 1st rotor torsion 0.62
4 1st asymmetric rotor flap/yaw (A) 0.87
5 1st asymmetric rotor flap/tilt (B) 0.94
6 1st symmetric rotor flap (C) 1.07
7 1st rotor edge B 1.72
8 1st rotor edge C 1.78
9 2nd asymmetric rotor flap/yaw (A) 2.05
10 2nd asymmetric rotor flap/tilt (B) 2.20
11 2nd symmetric rotor flap (C) 2.58
12 3rd asymmetric rotor flap/tilt (A) 3.87
13 3rd asymmetric rotor flap/yaw (B) 3.97
+ 1st tower yaw
14 2nd rotor edge 4.08
15 3rd asymmetric rotor flap/yaw (C) 4.43
+ 2nd tower bending

Table B.5: Results of a normal modes calculation for a fixed pitch, stall regu-
lated, constant speed 2 MW wind turbine with a rotor diameter of 76 m [122].

Eigen- Description Eigenfrequency


mode (Hz)
1 1st tower transverse 0.45
2 1st tower longitudinal 0.45
3 1st shaft torsion 0.65
4 1st backward whirl rotor flap 0.7
5 1st symmetric flap 1.05
6 1st forward whirl rotor flap 1.2
7 1st backward whirl edge 1.65
8 1st forward whirl edge 2.1
9 2nd backward whirl rotor flap 2.1
these results are calculated for normal production at 5 m/s wind

Table B.6: Results of a normal modes calculation for a pitch regulated, variable
speed 2.75 MW wind turbine [86].
271

Eigen- Description Eigenfrequency


mode (Hz)
1 1st tower transverse 0.31
2 1st tower longitudinal 0.33
3 1st rotor torsion 0.41
4 1st rotor flap A 0.59
5 1st rotor flap B 0.63
6 1st rotor flap C 0.75
7 1st rotor edge B 1.18
8 1st rotor edge C 1.23
9 2nd rotor flap A 1.60
10 2nd rotor flap B 1.70
11 2nd rotor flap C 2.10

Table B.7: Results of a normal modes calculation for an example of a wind


turbine (> 2 MW) in the HAWCModal demo [181].
272
Appendix C

Gear ratio of a planetary gear


with a fixed ring wheel

This appendix describes the relation between the rotational speed of the com-
ponents in a planetary gear stage with a fixed ring wheel.

The kinematics of a planetary gear stage are determined by the Willis formu-
las [218]. These formulas indicate that such a system has two kinematic DOFs
and needs, therefore, a constraint in order to yield a fixed gear ratio. The spe-
cific case where the ring wheel (rw) is fixed, the planet carrier (pc) is the input
and the sun is the output of the gear stage, is the most popular concept in wind
turbine gearboxes. For such a system, the following relations are valid, where
ω stands for the rotational speed and z for the respective number of teeth.
• The total gear ratio is the ratio of the output to the input speed and equals:
ωsun zrw
= 1+ (C.1)
ω pc zsun
Note that the planet carrier and sun rotate in the same direction.
• The ratio between the speed of the planets and the planet carrier equals:
ω planet z planet − zrw
= (C.2)
ω pc z planet
This is a negative ratio: the planets rotate consequently in the opposite
direction of the planet carrier.
• The ratio between the speed of the sun and the speed of the planets is
then derived as:
ωsun zrw z planet − zrw
= (1 + )/( ) (C.3)
ω planet zsun z planet

273
274 C. Gear ratio of a planetary gear with a fixed ring wheel

This is again a negative ratio indicating the rotation reversal from the
planets to the sun.

Figure C.1 describes an example of a planetary gear stage with a fixed ring
wheel. Based on the equations above, table C.1 describes the rotational speed
of the different components for an input speed of the planet carrier equal to
1 RPM. Note that the ratio between the speed of the sun and the planets equals
-3.87, which corresponds to equation (C.3).

Number of teeth
ring wheel 89
planets 36
sun 19

Figure C.1: Planetary gear stage with three planets.

Rotational speed (RPM)


planet carrier (input) 1
ring wheel 0
planets -1.47
sun (output) 5.68

Table C.1: Rotational speed of the different components in the planetary stage
described in figure C.1, for an input speed of 1 RPM.
Appendix D

Tooth contact forces for a


helical gear pair

This appendix describes the formulation of the tooth contact forces acting be-
tween the gears of a helical gear pair. This approach is used in the rigid
multibody models with discrete flexible elements, which are introduced in
section 4.5, and in the flexible multibody models, which are introduced in
section 4.6.

Figures 4.13 and 4.14 in chapter 4 describe the modelling approach for the
tooth contact forces between two helical gears in contact. These forces can be
written in matrix form as:
" →
− #   →
F1

k11 k12 −
q1

− = κgear ·
→ · →− (D.1)
F2 k21 k22 q2


− → − −
where F1 , F2 , →
q1 and →

q2 are written as:


F1 = [FX1 FY 1 FZ1 TX1 TY 1 TZ1 ]T


F2 = [FX2 FY 2 FZ2 TX2 TY 2 TZ2 ]T


q1 = [x1 y1 z1 ρX1 ρY 1 θ1 ]T ,


q = [x y z ρ ρ θ ]T ,
2 2 2 2 X2 Y2 2

And the sub-matrices k11 , k12 , k21 and k22 are given on the next page, with:
cβ0b = cos β0b ; sβ0b = sin β0b ; cψ = cos ψ; sψ = sin ψ.

275
276

 
−c2 β s2 ψ1 c2 β0b cψ1 sψ1 cβ0b sβ0b sψ1 cβ0b sβ0b s2 ψ1 −cβ0b sβ0b cψ1 sψ1 c2 β0b sψ1
 2
c β0b cψ1 sψ1 −c2 β0b c2 ψ1 −cβ0b sβ0b cψ1 −cβ0b sβ0b cψ1 sψ1 cβ0b sβ0b c2 ψ1 −c2 β0b cψ1 
 
 cβ0b sβ0b sψ1 −cβ0b sβ0b cψ1 −s2 β0b −s2 β0b sψ1 s2 β0b cψ1 −cβ0b sβ0b 
k11 =  0 0 0 0 s2 β0 s2 ψ 0 0

1
 rb1 cβ0b sβ0b s2 ψ1 −rb1 cβ0b sβ0b cψ1 sψ1 −rb1 s2 β0b sψ1 −rb1 
 b rb1 s2 β0b cψ1 sψ1 −rb1 cβ0b sβ0b sψ1 
0 cβ0 sβ0 cψ sψ 0 cβ0 sβ0 c2 ψ 0 s2 β0 cψ 0 0 s2 β0 c2 ψ 0 cβ0 sβ0 cψ
1 1 1 1 1 1
 −rb1 
b b rb1 b b rb1 b rb1 s2 β0b cψ1 sψ1 −rb1 b rb1 b b
0 c2 β0 sψ 0 c2 β0 cψ 0 cβ0 sβ0 0 cβ0 sβ0 sψ 0 0 c2 β0
rb1 b 1 −rb1 b 1 −rb1 b b −rb1 b b 1 rb1 cβ0b sβ0b cψ1 −rb1 b
 
c2 β0b sψ1 sψ2 −c2 β0b cψ2 sψ1 −cβ0b sβ0b sψ1 cβ0b sβ0b sψ1 sψ2 −cβ0b sβ0b cψ2 sψ1 c2 β0b sψ1
 −c2 β0b cψ1 sψ2 c2 β0b cψ1 cψ2 cβ0b sβ0b cψ1 −cβ0b sβ0b cψ1 sψ2 cβ0b sβ0b cψ1 cψ2 −c2 β0b cψ1 
 
 −cβ0b sβ0b sψ2 cβ0b sβ0b cψ2 s2 β0b −s2 β0b sψ2 s2 β0b cψ2 −cβ0b sβ0b 
k12 = 
 0 0 0 0 0 0


 −rb1 cβ0b sβ0b sψ1 sψ2 rb1 cβ0b sβ0b cψ2 sψ1 rb1 s2 β0b sψ1 −rb1 s2 β0b sψ1 sψ2 rb1 s2 β0b cψ2 sψ1 −rb1 cβ0b sβ0b sψ1 
0 cβ0 sβ0 cψ sψ 0 cβ0 sβ0 cψ cψ 0 s2 β0 cψ 0 s2 β0 cψ sψ 0 s2 β0 cψ cψ 0 cβ0 sβ0 cψ
1 2 1 2 1 1 2 1 2 1
 rb1 
b b −rb1 b b −rb1 b rb1 b −rb1 b rb1 b b
0 c2 β0 sψ 0 c2 β0 cψ 0 0 cβ0 sβ0 sψ 0 0 c2 β0
−rb1 b 2 rb1 b 2 rb1 cβ0b sβ0b −rb1 b b 2 rb1 cβ0b sβ0b cψ2 −rb1 b
 
c2 β0b sψ1 sψ2 −c2 β0b cψ1 sψ2 −cβ0b sβ0b sψ2 −cβ0b sβ0b sψ1 sψ2 cβ0b sβ0b cψ1 sψ2 −c2 β0b sψ2
 −c2 β0b cψ2 sψ1 c2 β0b cψ1 cψ2 cβ0b sβ0b cψ2 cβ0b sβ0b cψ2 sψ1 −cβ0b sβ0b cψ1 cψ2 c2 β0b cψ2 
 
 −cβ0b sβ0b sψ1 cβ0b sβ0b cψ1 s2 β0b −s2 β0b sψ1 s2 β0b cψ1 cβ0b sβ0b 
k21 = 
 0 0 0 0 0 0


 rb2 cβ0b sβ0b sψ1 sψ2 −rb2 cβ0b sβ0b cψ1 sψ2 −rb2 s2 β0b sψ2 −rb2 s2 β0b sψ1 sψ2 rb2 s2 β0b cψ1 sψ2 −rb2 cβ0b sβ0b sψ2 
0 cβ0 sβ0 cψ sψ 0 cβ0 sβ0 cψ cψ 0 s2 β0 cψ 0 s2 β0 cψ sψ 0 s2 β0 cψ cψ 0 cβ0 sβ0 cψ
2 1 1 2 2 2 1 1 2 2
 −rb2 
b b rb2 b b rb2 b rb2 b −rb2 b rb2 b b
0 c2 β0 sψ 0 c2 β0 cψ 0 cβ0 sβ0 0 cβ0 sβ0 sψ 0 cβ0 sβ0 cψ 0 c2 β0
rb2 b 1 −rb2 b 1 −rb2 b b −rb2 b b 1 rb2 b b 1 −rb2 b
 
−c2 β0b s2 ψ2 c2 β0b cψ2 sψ2 cβ0b sβ0b sψ2 −cβ0b sβ0b s2 ψ2 cβ0b sβ0b cψ2 sψ2 −c2 β0b sψ2
 c2 β0b cψ2 sψ2 −c2 β0b c2 ψ2 −cβ0b sβ0b cψ2 cβ0b sβ0b cψ2 sψ2 −cβ0b sβ0b c2 ψ2 c2 β0b cψ2 
 
 cβ0b sβ0b sψ2 −cβ0b sβ0b cψ2 −s2 β0b s2 β0b sψ2 −s2 β0b cψ2 cβ0b sβ0b 
k22 =  0 cβ0 sβ0 s2 ψ 0 0 0 s2 β0 s2 ψ 0 0

2 2
 −rb2 
 b b rb2 cβ0b sβ0b cψ2 sψ2 rb2 s2 β0b sψ2 −rb2 b rb2 s2 β0b cψ2 sψ2 −rb2 cβ0b sβ0b sψ2 
0 0 cβ0 sβ0 c2 ψ 0 s2 β0 cψ 0 0 s2 β0 c2 ψ 0 cβ0 sβ0 cψ
2 2 2 2
 rb2 cβ0b sβ0b cψ2 sψ2 −rb2 
b b −rb2 b rb2 s2 β0b cψ2 sψ2 −rb2 b rb2 b b
0 c2 β0 sψ 0 c2 β0 cψ 0 0 cβ0 sβ0 sψ 0 0 c2 β0
2 2 2
D. Tooth contact forces for a helical gear pair

−rb2 b rb2 b rb2 cβ0b sβ0b −rb2 b b rb2 cβ0b sβ0b cψ2 −rb2 b
Appendix E

Numerical calculations in
DADS

This appendix gives an overview of the procedures to perform the numerical


calculations in DADS, which are described in the present dissertation:

1. the normal modes calculations

2. the time integration procedure for the simulations of motion and of loads

3. the static analysis to calculate the torsional stiffness of a gearbox

E.1 Normal modes calculation


The normal modes calculation in DADS, which is used in the present research,
is part of a linearisation module in the software package. This module is called
DADS/Linear [138] and it permits to reduce nonlinear second order differential
algebraic equations (DAEs), which describe a multibody mechanical system,
into linear first order ordinary differential equations (ODEs). The linear first
order form of the equations of motion is consequently written as:

{δẋ} = [A] · {δx} (E.1)

{x} a selection of independent generalised coordinates (*)


δx = x − x0
δẋ = ẋ − x˙0 x0 and x˙0 : initial conditions
[A] the coefficient matrix (**)

(*): The selection of the independent generalised coordinates occurs automat-


ically in DADS.

277
278 E. Numerical calculations in DADS

(**): The derivation and calculation of the coefficient matrix is described


in [138]. [A] is finally written as a function of the derivatives of x and ẋ.
Its calculation is based on the perturbation theory and implies a numer-
ical evaluation at the operating point. For non-linear systems, the linear
equivalent is only valid in the neighbourhood of the operating point.
The eigenvalues of the numerically evaluated [A] are the eigenvalues of
the mechanical system at the operating point. These results include the
eigenfrequencies and mode shapes, which are used in the present disser-
tation.

The computational time of all normal modes calculations in the present dis-
sertation increases with the number of DOFs in the models. However, since
all calculations were finished within 5 seconds, no further investigation of the
computational time is performed.

E.2 Simulation of motion and loads


The numerical procedure for the time integration of the equations of motion
in DADS is used for the calculation of frequency response functions (FRFs)
in chapters 5 and 6. In each simulation a specific force excitation is applied at
one of the bodies in the model. This excitation is an input time series, which
is described in the respective examples.

Various integration methods are available in DADS. For all simulations in


this dissertation, the “backward differentiation formula” (BDF) has been used,
since (1) it is described as inherently stable for stiff systems, where the eigen-
values differ by large margins [138], and since (2) it yielded the fastest cal-
culation time for identical simulations using other methods. According to the
DADS manual [138], the BDF is an implicit integrator which uses a backward
differentiation formula method. This implies iteratively solving a large system
of DAEs to converge to a solution at each time step. The method uses an iter-
ation matrix, which is calculated by a finite difference scheme.

In all simulations in this dissertation, the default toleration values for conver-
gence and the estimated errors have been applied. The maximum time inte-
gration step is chosen equal to the desired time step in the calculated time
series. This latter value is chosen based on the desired bandwidth in the dif-
ferent simulations. Finally, it is assumed that an optimisation of all simulation
parameters may considerably reduce the required computational time in all
calculations. This has not been further investigated in the present research.
E.2 Simulation of motion and loads 279

E.2.1 Comparison of computational time for different types of multi-


body models
This section describes a comparison of the computational time, which is re-
quired for a load simulation using the three MBS formulations presented in
this dissertation. This is an important parameter in assessing the feasibility of
the different models for the design of a drive train. The comparison is made
for three models of the high speed parallel helical gear stage described in sec-
tion 6.3.1.

1. The first model is a purely torsional model (cfr. figure 6.2(a)) including
one DOF per body; this model has consequently two DOFs.

2. The second model is a rigid multibody model (cfr. figure 6.2(b)) includ-
ing six DOFs per body; this model has consequently twelve DOFs.

3. The third model is a flexible multibody model (cfr. figure 6.2(c)) includ-
ing six rigid-body DOFs per body. Additionally, a set of twenty-three
component modes is included for the pinion and a set of twelve compo-
nent modes for the gear. This model has in total forty-seven DOFs.

For all three models two identical simulations are performed: a time series
is calculated using identical simulation parameters and an identical excitation
signal. This excitation is a torque fluctuation applied on the pinion during
1 second and 6 seconds respectively. The same excitation signal is used for the
FRF calculation in section 6.5.2. Figure 6.13 shows its power spectrum. The
maximum time integration step for the present analyses equals 0.0001 second.
Table E.1 summarises the absolute CPU seconds for the six simulations and
gives a relative comparison for the three models, represented as one CPU sec-
ond per DOF in the model and per second of the simulated time series. This
table yields the following conclusions.

Model DOFs CPU seconds CPU seconds


per DOF and per
second time series
A B A B
Torsional model 2 3.5 18.0 1.8 1.5
Rigid multibody model 12 9.2 58.5 0.8 0.8
Flexible multibody model 47 161 1065 3.4 3.8

Table E.1: Comparison of the computational time for three different MBS for-
mulations (A: load simulation for a time series of 1 second; B: load simulation
for a time series of 6 seconds).
280 E. Numerical calculations in DADS

• The comparison of the computational time for the short time series (1 sec-
ond) and the long time series (6 seconds) yields a logic result: the num-
ber of CPU seconds increases directly proportional to the length of the
time series and, as a result, the values in the two last columns are quasi
equal for each model respectively.

• The comparison for the different models indicates that the computational
time increases with increasing number of DOFs. However, it is not clear
why the calculation for the torsional model takes relatively more time
per DOF than the calculation for the rigid multibody model. It is as-
sumed that the value 0.8 CPU seconds per DOF and per second time
series for the latter model is a more accurate estimation of the compu-
tational time. This value is four times larger for the flexible multibody
model. This indicates that the cost of calculating a flexible multibody
model is considerably larger; however, it can yield important additional
insight in the stress levels of the components, as described in the exam-
ple of section 6.5.2.

E.3 Static analysis


The numerical procedure for a static analysis in DADS is used for the cal-
culation of the torsional gearbox stiffness in chapters 5 and 6. The proce-
dure is elaborately described in the DADS manual [138]. The applied method
searches the equilibrium position of the multibody system at which the sum
of the applied and the reaction forces is zero. This implies solving a set of
force-balancing equations. Two solver methods are available in a static anal-
ysis in DADS. In the presented examples, the so-called “QRFULL” method is
used, since it is described to be better for ill-conditioned problems because it
computes a least squares static solution.
Simulatie van de dynamische
belasting in de aandrijflijn van
een windturbine

1 Inleiding
1.1 Situatieschets en probleembeschrijving
Tijdens het afgelopen decennium is het gebruik van windenergie voor het op-
wekken van elektriciteit sterk toegenomen. Figuur 1.1 toont de evolutie van
de wereldwijd geı̈nstalleerde capaciteit aan elektrisch vermogen op basis van
windenergie sinds 1995. Aan het einde van 2005 bedroeg deze capaciteit
59 GW, waarvan er 20% werd geı̈nstalleerd gedurende datzelfde jaar. De
windturbine-industrie groeit zeer snel en investeert veel in onderzoek en ont-
wikkeling om de werking van de huidige windturbines nog verder te verbeteren
en hun capaciteit nog verder op te drijven. De grootste moderne windturbines
hebben reeds een capaciteit van 5 MW en hebben wieken met een lengte van
60 m en een toren met een hoogte tot 100 m. Deze indrukwekkende ma-
chines werken bovendien vaak in complexe omstandigheden, zodat het garan-
deren van hun structurele integriteit gedurende een levensduur van 20 jaar een
enorme uitdaging is.

Tijdens het ontwerp van een windturbine wordt er gebruik gemaakt van spe-
ciale simulatiecodes1 voor het voorspellen van de belastingsniveaus en belas-
tingswisselingen op de verschillende componenten in de machine. Het struc-
tureel model van de windturbine in deze traditionele software is meestal vol-
doende gedetailleerd om een nauwkeurige belasting te voorspellen op de ro-
tor en de toren. Dit model bevat echter slechts één enkele vrijheidsgraad die
1 “Codes”verwijst in dit proefschrift naar software die tijdens het ontwerpproces van een
windturbine gebruikt wordt voor de simulatie van de belasting op de verschillende onderdelen.

I
II Nederlandse samenvatting

[MW]

Figuur 1.1: De evolutie van de wereldwijd geı̈nstalleerde capaciteit aan elek-


trisch vermogen op basis van windenergie van 1995 tot 2005 [81].

het gedrag van de volledige aandrijflijn beschrijft. Dit is onvoldoende om de


dynamische belasting op de verschillende componenten in de aandrijflijn te
voorspellen. Bijgevolg gebeurt het huidige ontwerp van deze componenten
volgens quasi-statische methodes, waarin specifieke veiligheidsfactoren wor-
den gebruikt om de dynamische belasting in rekening te brengen. Dit leidt
echter tot beperkingen in de betrouwbaarheid van het ontwerpproces voor de
aandrijflijn.

Een dynamische simulatie van de aandrijflijn is vereist om de ontwerpbelas-


ting beter te begrijpen, om de interne dynamica juist te voorspellen en om
lokale spanningsniveaus in de componenten te voorspellen. Uiteindelijk zal de
nieuwe simulatiemethode bijdragen tot meer rendabele windturbines door het
vermijden van schadegevallen of door te besparen op een onnodig overged-
imensioneerd ontwerp. Dit proefschrift beschrijft de ontwikkeling van een
consistente modelleringstechniek om het dynamisch gedrag van een complexe
aandrijflijn in een windturbine op een correcte manier te simuleren. Het be-
schrijft tevens de resultaten van een unieke meetcampagne op een moderne
“multi-MW” windturbine.
Nederlandse samenvatting III

1.2 Doelstelling van het onderzoek


Het onderzoek in dit proefschrift vertrekt van de state-of-the-art in het ont-
werp van de aandrijflijn van een windturbine. Hoofdstuk 2 geeft hiervan een
overzicht en beschrijft de beperkingen in de traditionele simulatiesoftware. Dit
toont de nood aan een nieuwe simulatiemethode. Bij de ontwikkeling hiervan
stelt zich de centrale vraag: “Welke methode kan leiden tot een robuuster en
meer rendabel ontwerp van de aandrijflijn?”

Dit proefschrift legt de focus op de aandrijflijn van windturbines met een tand-
wielkast. Figuur 1.2 toont een voorbeeld van zo een aandrijflijn. De wieken
van de windturbine zijn bevestigd aan de naaf en zetten de energie in de wind
om naar mechanische energie in de aandrijflijn. De naaf drijft de hoofdas aan.
Deze as wordt ondersteund door een hoofdlager en door een tweede lager in de
tandwielkast. De tandwielkast zorgt voor een verhoging van het toerental van
de hoofdas tot het werkingstoerental van de generator. De uitgaande as van de
tandwielkast is verbonden met de generator d.m.v. een flexibele koppeling.
Naaf Flexibele koppeling

Hoofdlager Tandwielkast  Generator


Hoofdas 

Figuur 1.2: Opbouw van de aandrijflijn in een windturbine met een tand-
wielkast [25].

1.3 Overzicht van het proefschrift


Hoofdstuk 2 beschrijft het ontwerpproces van de aandrijflijn in een windtur-
bine en de state-of-the-art voor het simuleren van de belastingen met de focus
op het gebruikte structureel model. Dit toont de beperkingen aan van het model
voor het ontwerp van de aandrijflijn. Het hoofdstuk introduceert uiteindelijk
de formulering op basis van flexibele meerlichamen-systemen (MLS) als de
beste methode voor de ontwikkeling van een meer gedetailleerd model van de
aandrijflijn.
IV Nederlandse samenvatting

Hoofdstuk 3 beschrijft een generieke methodologie voor het modelleren van


de aandrijflijn gebaseerd op drie MLS-technieken. De eerste methode is be-
perkt tot de analyse van torsietrillingen. De tweede techniek laat toe van de
lagers en de vertanding in de aandrijflijn op een realistischere manier voor te
stellen. De generieke implementatie van deze techniek kan gebruikt worden
voor de modellering van schuine en rechte vertanding in zowel parallelle als
planetaire tandwieltrappen. De derde methode is de flexibele MLS-techniek,
die naast de globale beweging van de componenten in de aandrijflijn ook
inzicht biedt in hun elastische vervorming.

Hoofdstuk 4 illustreert de toepassing van de eerste en de tweede methode voor


het analyseren van een planetaire tandwieltrap.

Hoofdstuk 5 beschrijft het gebruik van de drie simulatietechnieken voor de


analyse van de aandrijflijn in een windturbine. De tandwielkast in deze aan-
drijflijn bestaat uit drie trappen: een planetaire trap met rechte vertanding, een
planetaire trap met schuine vertanding en een parallelle trap met schuine ver-
tanding. Eerst worden de tweede en de derde trap afzonderlijk geanalyseerd.
Vervolgens bespreekt dit hoofdstuk de analyse van de aandrijflijn met inbe-
grip van de koppeling met de toren, de rotor en de generator. De analyses zijn
gericht op het laagfrequente en middenfrequente bereik en tonen aan hoe re-
sonantie in de aandrijflijn kan geı̈dentificeerd worden. Verder wordt gedemon-
streerd hoe de belasting in de aandrijflijn berekend wordt voor, enerzijds, een
sinusoı̈dale excitatie en anderzijds een piekwaarde in het koppel van de gene-
rator.

Hoofdstuk 6 geeft een overzicht van de meetcampagne op een multi-MW


windturbine.

Hoofdstuk 7 vat de belangrijkste besluiten van dit proefschrift samen.

2 Ontwerp van de aandrijflijn in een windturbine


De ontwerpspecificaties van een windturbine worden gedefinieerd volgens spe-
cifieke normen en richtlijnen zoals de internationale IEC61400-1 norm, de
richtlijnen van Germanischer Lloyd en de nationale normen NEN-6096 en
DS-472 uit respectievelijk Nederland en Denemarken [37]. Deze specificaties
beschrijven o.a. de (weers)omstandigheden waaraan een windturbine wordt
blootgesteld, met daarin de beschrijving van de windsnelheid als de belang-
rijkste parameter. De combinatie van deze omstandigheden met alle mogelijke
werkingscondities van een windturbine leidt tot een grote verzameling belas-
tingsgevallen.
Nederlandse samenvatting V

Voor de voorspelling van de belasting op de windturbine in al deze gevallen


zijn speciale simulatiecodes vereist. Deze codes beschrijven de windturbine
als een systeem en bestaan typisch uit verschillende externe en interne mo-
dules. Figuur 2.1 toont de opbouw van dit systeem. De externe modules be-
schrijven de input voor het systeem en zijn onafhankelijk van het gedrag van
de windturbine. Ze beschrijven de wind, het elektriciteitsnetwerk en de zee-
golven (voor offshore windturbines). De interne modules beschrijven de be-
lasting op de windturbine, het controlesysteem en het mechanisch gedrag van
de windturbine. Dit laatste wordt voorgesteld door het structureel model, dat
kan beschreven worden op basis van een MLS-formulering, een formulering
met eindige-elementen (EE) of een flexibele MLS-formulering. De meeste be-
staande simulatiecodes zijn gebaseerd op deze laatste methode.

Figuur 2.1: Voorstelling van de windturbine als een systeem in de traditionele


simulatiecodes (donkergrijs: externe modules, lichtgrijs: interne modules).

De opbouw van het structureel model van de windturbine is gelijkaardig voor


de meeste bestaande simulatiecodes. Figuur 2.2 toont dit model dat zestien tot
vierentwintig vrijheidsgraden bevat:

1. één of twee paar buigmodes van de toren en één torsiemode (3 - 5 vrij-


heidsgraden)

2. twee of drie paar buigmodes voor iedere wiek (12 - 18 vrijheidsgraden


voor een rotor met drie wieken)

3. één vrijheidsgraad voor de torsionele vervorming van de aandrijflijn


(1 vrijheidsgraad)
VI Nederlandse samenvatting

flexibele rotor koppel


( 12 - 18
vrijheidsgraden )
KDT
naaf generator ( 1 vrijheidsgraad )
generatorkoppel
?
top van de toren

flexibele toren
belasting ( 3 - 5 vrijheidsgraden )
exclusief koppel

Figuur 2.2: Schematische voorstelling van het structureel model van een wind-
turbine met drie wieken in een traditionele simulatiecode.

Het structureel model bevat slechts één vrijheidsgraad voor de aandrijflijn en


dit leidt tot beperkingen in de betrouwbaarheid van het ontwerp.

1. De bestaande simulatiecodes voorspellen geen dynamische belasting op


de afzonderlijke onderdelen van de aandrijflijn. De bandbreedte van de
belastingssimulaties is daarenboven typisch beperkt tot 10 Hz. Deze
benadering leek aanvaardbaar, omdat de interne dynamica van de aan-
drijflijn als verwaarloosbaar werd beschouwd t.o.v. de dynamica van de
windturbine. Dit werd echter nooit experimenteel aangetoond en extra
aandacht is hier vereist, omdat de interne eigenfrequenties van de aan-
drijflijn in nieuwe en grotere windturbines dalen. Bovendien kunnen
interne excitaties aan hogere frequenties, zoals de ingrijpfrequenties van
de tandwielen of excitaties in de generator, ook leiden tot verhoogde
belastingsniveaus a.g.v. resonantie.

2. Een recente publicatie van De Vries [51] wijdt een hele reeks van scha-
degevallen in tandwielkasten van windturbines aan (1) een gebrek aan
inzicht in de lokale belasting en in de spanningsniveaus in de onderdelen
van de aandrijflijn en aan (2) een onvoldoende begrip van de ontwerp-
belasting. Beide argumenten zijn een gevolg van de beperkingen in het
structureel model.
Nederlandse samenvatting VII

3. Door het gebrek aan inzicht en de beperkte nauwkeurigheid van de be-


lastingssimulatie is het onmogelijk om de overtolligheid of de ontoerei-
kendheid van de veiligheidsfactoren in te schatten.

De beschreven beperkingen leiden tot de nood aan een meer gedetailleerde


simulatiemethode. Deze methode moet het mogelijk maken om:

1. de dynamische belasting op alle onderdelen van de aandrijflijn nauwkeu-


rig te simuleren en alle interne eigenmodes te berekenen om resonantie
te kunnen vermijden

2. schadelijke transiënte fenomenen in de aandrijflijn te identificeren

3. het niveau en de wisseling van de lokale spanningen in de onderdelen


te voorspellen om de betrouwbaarheid van vermoeiingsberekeningen te
verhogen

4. de onzekerheid, die is vervat in de veiligheidsfactoren, te reduceren

De flexibele MLS-formulering is de meest geschikte modelleringstechniek om


deze doelstellingen te verwezenlijken. De software LMS DADS Revision 9.6
(DADS) [138] van LMS International in Leuven (België) werd gekozen als
beste alternatief om de nieuwe modellen in te ontwikkelen2 .

3 Modelleren van de aandrijflijn in een windturbine


Een correct dynamisch structureel model van de aandrijflijn in een windturbine
vereist een nauwkeurige beschrijving van de massa en inertie van de verschil-
lende onderdelen, van de flexibiliteiten en de demping in de aandrijflijn en van
het verloop van alle krachten. Het meest complexe onderdeel van de aandrijf-
lijn is de tandwielkast. Figuur 3.1 toont een doorsnede van een tandwielkast
in een windturbine. Voor het modelleren van een tandwielkast wordt er verder
onderscheid gemaakt tussen de vertanding, de lagers en de andere componen-
ten. Drie verschillende modelleringstechnieken kunnen gebruikt worden voor
dynamische analyses van tandwielkasten.

1. Een structureel EE-model heeft typisch een groot aantal vrijheidsgraden


voor ieder onderdeel van de aandrijflijn. Dit leidt tot lange rekentijden
tijdens simulatie, waardoor deze techniek slechts zelden wordt gebruikt
in dynamische analyses van tandwielkasten. Dit proefschrift beschrijft
geen modellen op basis van de EE-formulering.
2 Sinds
2001 maakt LMS DADS deel uit van de simulatiesoftware LMS Virtual.Lab en heet
het LMS Virtual.Lab Motion [207].
VIII Nederlandse samenvatting

rondsel 3de
trap HH
HH

wiel 3de
trap H
HH

krimpschijf
@ HH
HH
rondsel 2de trap

A
A HH
 A H
 A H
 wiel 2de trap
 A
 A
planetendrager @ A
zon
@
@
B planeet
B
B
ophanging H ringwiel
H H

Figuur 3.1: Tandwielkast in een windturbine uit de 1 MW klasse, met één


planetaire trap met rechte vertanding (1ste trap) en twee parallelle trappen met
schuine vertanding (2de en 3de trap).

2. De meeste publicaties over dynamische analyses van tandwielkasten be-


schrijven modellen op basis van de MLS-formulering. Deze modellen
bevatten minstens één vrijheidsgraad per lichaam om het koppel in de
aandrijflijn te simuleren. Paragraaf 3.1 beschrijft dergelijke puur tor-
sionele MLS. Simulaties op basis van modellen met zes vrijheidsgraden
per lichaam leiden tot nauwkeurigere resultaten. Paragraaf 3.2 beschrijft
een generieke ontwikkeling van deze techniek, die analyses van com-
plexe tandwielkasten, inclusief schuinvertande planetaire trappen, mo-
gelijk maakt. Dit draagt bij aan de huidige state-of-the-art.

In de meeste MLS-modellen uit de literatuur wordt een veer-demper ele-


ment gebruikt voor de beschrijving van het krachtenverloop in de ver-
tanding. Kahraman en Singh [113] maken hier nog verder onderscheid
Nederlandse samenvatting IX

tussen lineair-tijdsinvariante (LTI) modellen en lineair-tijdsvariante (LTV)


modellen, waarbij LTV modellen hoofdzakelijk worden toegepast voor
het voorspellen van de excitatie in de vertanding. Dit wordt verder niet
beschouwd in dit proefschrift en de vertanding wordt vereenvoudigd
voorgesteld door een veer met een constante stijfheid. Dezelfde vereen-
voudiging wordt ook doorgevoerd voor de lagers. Deze benadering is
hier slechts geldig binnen een beperkt bereik voor de belasting op de
lager.

3. Paragraaf 3.3 beschrijft de dynamische analyse van een tandwielkast op


basis van een flexibele MLS-formulering. Hierin worden technieken uit
de MLS-formulering voor tandwielkasten gecombineerd met flexibele
lichamen. Dit is een belangrijke bijdrage aan de state-of-the-art.

3.1 Torsionele meerlichamen-systemen


In een puur torsioneel meerlichamen-systeem telt ieder lichaam één rotationele
vrijheidsgraad. In een dergelijk model kan enkel de flexibiliteit in de assen
en de vertanding op een directe manier voorgesteld worden. De invloed van
andere flexibiliteiten kan meegenomen worden als een reductiefactor voor de
torsionele stijfheid van deze componenten. Dit bemoeilijkt de opbouw van het
model. Bovendien wordt enkel het koppel in de aandrijflijn met deze tech-
niek op een directe manier gesimuleerd en niet de krachten of buigmomenten.
Figuur 3.2 toont hoe de tandcontactkracht in een parallelle tandwieltrap wordt
voorgesteld als een koppel in een torsioneel model.
X Nederlandse samenvatting

+ θ2
- 0
rb2



Fbt T2
κgear
θ1 rotatie rondsel Td +
θ2 rotatie wiel
0 0
rb1
rb1 basiscirkel rondsel
0
rb2 basiscirkel wiel θ1
T1
J1 inertie rondsel
J2 inertie wiel ?
igear overbrengingsverhouding (rb2 0 /r 0 )
b1
positief aandrijfkoppel op rondsel κgear · (rb1 cos βb )
0 0 2
Td


Fbt tandcontactkracht COC
in het transversaal vlak J1 J2
κgear tandveerstijfheid igear
T1 reactiekoppel op rondsel
T2 reactiekoppel op wiel

T1 = −Td = −Fbt · rb1 0


= −κgear · cos β0b (rb1
0
θ1 − rb2
0
θ2 ) ·rb1
0
cos β0b
| {z }
(b) (c) (a)
z }| { z }| {
= − κgear · (rb1
0
cos β0b )2 · (θ1 − igear · θ2 ) (3.1)

1
T2 = −T1 · igear = − κgear · (rb2
0
cos β0b )2 · (θ2 − · θ1 ) (3.2)
| {z } igear
| {z }
(d) (e)

(a) vervorming volgens de ingrijplijn (> 0)


(b),(c) torsiestijfheid, torsievervorming - gerelateerd naar het rondsel
(d),(e) torsiestijfheid, torsievervorming - gerelateerd naar het wiel

Figuur 3.2: Torsioneel model van de tandcontactkracht tussen een drijvend


rondsel en een aangedreven wiel. Td is een positief aandrijfkoppel op het rond-
sel. Dit veroorzaakt een negatief reactiekoppel T1 op het rondsel en een positief
reactiekoppel T2 op het wiel.
Nederlandse samenvatting XI

3.2 Meerlichamen-systemen met discrete flexibele elementen


In een MLS met discrete flexibele elementen telt ieder lichaam zes vrijheids-
graden. Dit laat toe om de invloed van de flexibiliteit van de lagers op de
dynamische koppelbelasting op een directe manier te analyseren. Bovendien
geeft dit ook inzicht in de dynamische belasting van de lagers, die gekoppeld
is met de verplaatsing van de lichamen in hun lagers. Figuur 3.3 toont het
model voor een lager: een LTI 6 × 6 stijfheidsmatrix beschrijft de flexibiliteit
in de lager. Ook de koppeling tussen twee tandwielen wordt gemodelleerd
als een LTI veer-element dat twaalf vrijheidsgraden koppelt, zoals voorgesteld
in figuur 3.4. Deze benaderingen zijn gebaseerd op de modellen van Kahra-
man [105], in zijn analyse van een parallelle tandwieltrap met schuine vertan-
ding, en op de modellen van Lin en Parker [129], in hun studie van planetaire
tandwieltrappen met rechte vertanding. De synthetische aanpak in dit proef-
schrift beschrijft twee driedimensionale modellen voor respectievelijk een la-
ger en de vertanding, die toelaten om complexere systemen van planetaire
tandwieltrappen met schuine vertanding te simuleren, alsook volledige tand-
wielkasten geı̈ntegreerd in een windturbine.



Fb = Kb · →

q

Kb = diag(kradiaal , kradiaal , kaxiaal , ktilt , ktilt , 0)



q = [x y z ρX ρY θ]T

x, y, z, ρX , ρY en θ zijn de projecties in het XYZ-assenstelsel van de translaties


en rotaties van de positie en de oriëntatie van het tandwiel in zijn referentie-
assenstelsel X’Y’Z’.

Figuur 3.3: Schematische voorstelling van het model voor een lager: een lin-
eair veer-element verbindt het XYZ-assenstelsel (vast aan het lichaam) met
een X’Y’Z’-assenstelsel (vast aan de referentie van het lichaam). De veer-


karakteristiek wordt beschreven door de stijfheidsmatrix Kb en Fb is de kracht
op het tandwiel in het XYZ-assenstelsel.
XII Nederlandse samenvatting

Y’1 Wiel1 Wiel1 drijft aan (Td < 0)


0 : straal basiscirkel
rb1

Wiel2 wordt aangedreven


0 : straal basiscirkel
rb2
Z’1 β0b κgear
Td X’1 κgear is de tandveerstijfheid
Wiel2
Y’2

Z’2

X’2

" →
− #
F1

k11 k12
  →

q1

− = κgear ·
→ · →
− (3.3)
F2 k21 k22 q2

Figuur 3.4: Voorstelling van de krachtvector tussen twee schuinvertande wie-


len als een lineair veer-element. Appendix D beschrijft de stijfheidsmatrix in
vergelijking (3.3).

3.3 Flexibele meerlichamen-systemen


Flexibele MLS zijn MLS, waarin de lichamen extra vrijheidsgraden hebben om
hun elastische vervorming te beschrijven. Voor een dergelijk flexibel lichaam
kan tijdens simulatie naast zijn beweging ook zijn vervorming voorspeld wor-
den. De toepassing van deze techniek heeft twee doelstellingen:
1. Een realistischere beschrijving van de flexibiliteit van een lichaam kan
de nauwkeurigheid van de belastingssimulaties verbeteren.

2. Op basis van de gesimuleerde vervorming van een lichaam kunnen zijn


interne spanningen berekend worden. Deze zijn vereist in vermoeiings-
berekeningen.
De beschrijving van de vervorming van een lichaam gebeurt op basis van een
aantal zogenaamde component modes. De methode om deze modes te bereke-
nen en vervolgens te integreren in het MLS is de CMS techniek3 . Dit is een
3 CMS verwijst naar het Engelse component mode synthesis.
Nederlandse samenvatting XIII

reductietechniek, die een EE-model van het lichaam herleidt tot een set van
component modes. De koppeling van het gereduceerde model met de rest van
het systeem gebeurt op vooraf bepaalde knopen van het EE-model.

Er bestaan verschillende types van component modes, met daarin een onder-
scheid tussen dynamische modes en statische modes voor het beschrijven van
respectievelijk het dynamische en het statische gedrag van een lichaam. Voor
het modelleren van een aandrijflijn in DADS wordt de keuze voor de set van
component modes volgens Craig-Bampton [43–45,48] als het beste alternatief
beschouwd.

4 Analyse van een planetaire tandwieltrap


Dit hoofdstuk beschrijft de analyse van drie planetaire tandwieltrappen met
rechte vertanding, die worden beschreven door Lin en Parker in [129]. Figuur
4.1 toont deze systemen bestaande uit respectievelijk drie, vier en vijf plane-
ten. Het ringwiel staat steeds stil en zijn verbinding met de behuizing van de
tandwielkast wordt voorgesteld door een torsioneel veer-element met een stijf-
heid gelijk aan kring−huis . De planetendrager en de zon vormen respectievelijk
de input- en de ouputzijde van het systeem en hebben vrije randvoorwaarden.
Verder is de stijfheid in de radiale steunpunten van de planetendrager, de zon,
de planeten en het ringwiel gelijk (krad ).

In eerste instantie wordt een puur torsioneel MLS van de planetaire tandwiel-
trappen geanalyseerd. In deze modellen wordt de radiale stijfheid (krad ) ver-
waarloosd. Enkel de tandveerstijfheid en de torsionele stijfheid van het ring-
wiel worden beschouwd. Tabel 4.1 toont de eigenfrequenties voor de drie tor-
sionele MLS. De modes bij 0 Hz komen overeen met de kinematische rotatie
van het systeem. Verder zijn er telkens vier andere eigenfrequenties, waarvan
er één identiek is voor de drie systemen (6.4 kHz). Deze heeft een multipliciteit
N − 1 en is een gevolg van de symmetrie in een planetair systeem. Op basis
van een frequentie-respons analyse wordt echter aangetoond dat deze mode
geen invloed heeft op koppelvariaties in het systeem.

Vervolgens wordt voor ieder systeem een MLS, zoals voorgesteld in para-
graaf 3.2, geanalyseerd. Aangezien iedere planetaire trap rechte vertanding
heeft en er bijgevolg geen krachten zijn die uit het vlak werken, wordt de ana-
lyse beperkt tot trillingen in het vlak. Dit is analoog aan de aanpak van Lin en
Parker [129]. Volgens hen heeft ieder systeem maximaal vijftien verschillende
eigenmodes, die kunnen opgedeeld worden als rotationele, translationele en
planeet modes.
XIV Nederlandse samenvatting

Zon Planeet Drager Ringwiel


Massa (kg) 0.4 0.66 5.43 2.35
J (·10−3 kgm2 ) 0.58 1.53 49.2 56.7
rb0 (mm) 38.7 50.2 96.9 -137.5
Tandveerstijfheid κgear = 5 · 108 N/m
Lagerstijfheid krad = 108 N/m
Torsiestijfheid kring−huis = 19 · 106 Nm/rad
Nominale drukhoek αt0 = 24.6◦
Helixhoek β = 0◦

(a) Model-parameters.

(b) Systeem met drie plane- (c) Systeem met vier plane- (d) Systeem met vijf plane-
ten. ten. ten.

Figuur 4.1: Planetaire tandwieltrappen beschreven door Lin en Parker in [129].

Eigen- N
frequentie 3 4 5
(1) 0 0 0
(2) 2217 2138 2059
(3) 6159 6451 (×3) 6444(×4)
(4) 6444 (×2) 6688 7105
(5) 11205 12577 13810

Tabel 4.1: Eigenfrequenties berekend op basis van een torsioneel model van
de planetaire tandwieltrappen met N planeten.

Figuur 4.2 toont een voorbeeld van een modevorm uit iedere categorie voor het
systeem met vier planeten. Figuur 4.2(a) toont eerst een onvervormd model.
De vervorming in de modevormen kan vervolgens geı̈nterpreteerd worden als
het verschil tussen deze figuur en respectievelijk de figuren 4.2(b), 4.2(c) en
4.2(d). De vervorming is steeds een combinatie van de vervorming in de lagers
(planeten, zon, planetendrager en ringwiel), van de vervorming in de tandcon-
tacten (zon-planeet, planeet-ringwiel) en van de torsionele vervorming van het
ringwiel en de behuizing.
Nederlandse samenvatting XV



(a) onvervormd model (b) rotationele mode ( R 2 : 1519 Hz):
de relatieve beweging van alle plane-
ten t.o.v. de planeetlagers is identiek
en gebeurt in fase. De beweging van
de planetendrager, van het ringwiel en
van de zon is een pure rotatie.


− →

(c) translationele mode ( T 1a : 759 (d) planeet mode ( P 1 : 1959 Hz):
Hz): de beweging van de plane- de planeten bewegen terwijl de plane-
tendrager, van het ringwiel en van de tendrager, het ringwiel en de zon stil-
zon zijn pure translaties in het vlak. staan.

Figuur 4.2: Classificatie van de modevormen voor een planetaire tandwieltrap


met 4 planeten.

1. Er zijn zes rotationele modes met multipliciteit m = 1. De modevormen




( R 1−6 ) beschrijven een pure rotatie van de planetendrager, van het ring-
wiel en van de zon. De beweging van alle planeten is identiek en hun
beweging gebeurt in fase.

2. Er zijn zes translationele modes met multipliciteit m = 2 voor verschil-




lende N. De zes koppels van modevormen ( T 1a,b−6a,b ) beschrijven een
pure translatie van de planetendrager, van het ringwiel en van de zon.
XVI Nederlandse samenvatting

Mode- N
vorm 3 4 5



m=1 R1 0 0 0


R2 1425 1519 1538


R3 2032 2079 2082


R4 2644 2630 2602


R5 7500 7805 8086


R6 11744 13052 14237


m=2 T 1a,b 770 759 745


T 2a,b 1101 1092 1073


T 3a,b 1989 1947 1921


T 4a,b 2238 2328 2421


T 5a,b 7060 7249 7427


T 6a,b 9582 10392 11136


m = N-3 P1 1959 1959


P2 6450 6444


P3 6497 6497

Tabel 4.2: Classificatie van de eigenfrequenties voor de planetaire tandwiel-


trappen met N planeten, zoals beschreven door Lin en Parker in [129].

3. Er zijn drie planeet modes voor N > 3 met een multipliciteit m=N-3.
De planetendrager, het ringwiel en de zon staan volledig stil in de bijbe-


horende modevormen ( P 1−3 ).

Tabel 4.2 toont deze classificatie voor de resultaten die werden berekend in
DADS. Er is een goede overeenkomst tussen de berekende eigenfrequenties en
de resultaten van Lin en Parker. Dit toont de juistheid aan van de modelimple-
mentatie. Verder toont een frequentie-respons analyse dat enkel de rotationele
modes kunnen geëxciteerd worden door het aanleggen van koppelvariaties aan
de planetendrager (input) of de zon (output).

Een vergelijking van de eigenfrequenties in tabel 4.2 met de resultaten bere-


kend op basis van het torsionele MLS (tabel 4.1), leidt tot volgende besluiten:

1. Het model met meerdere vrijheidsgraden per lichaam beschouwt de ver-


schillende lagers als afzonderlijke veer-elementen. Dit laat toe van hun
Nederlandse samenvatting XVII

individuele invloed op een directe en realistischere manier te analyseren.


De vergelijking van de eigenfrequenties toont aan dat de flexibiliteit van
de lagers een belangrijke invloed heeft op de torsionele eigenfrequenties
in een planetaire tandwieltrap.

2. De vergelijking van de analyses toont ook aan dat de berekening op basis


van het model met meerdere vrijheidsgraden extra eigenmodes oplevert.
Aangezien deze in hetzelfde frequentiebereik liggen als de torsionele
modes, is het belangrijk om ook deze te kunnen identificeren. Dit is
enkel mogelijk op basis van de modelbeschrijving in paragraaf 3.2.

5 Analyse van de aandrijflijn in een moderne windtur-


bine
Dit hoofdstuk bespreekt de analyse van de aandrijflijn in een moderne wind-
turbine op basis van de generieke modelleringstechnieken uit hoofdstuk 3.
Figuur 5.1 toont de windturbine voor deze analyse. De aandrijflijn bevat geen
hoofdas zoals in figuur 1.2, maar wel een hoofdlager geı̈ntegreerd in de tand-
wielkast, die de naaf en de rotor van de windturbine draagt. De tandwielkast
bestaat uit twee planetaire tandwieltrappen en één parallelle tandwieltrap. De
rotor van de windturbine drijft de planetendrager van de 1ste planetaire trap
aan. Deze trap heeft rechte vertanding en het ringwiel staat stil t.o.v. de behui-
zing van de tandwielkast. De behuizing en de ophanging van de tandwielkast
in het frame worden star verondersteld. Het frame ondersteunt ook de gene-
rator en steunt op zijn beurt op een kruilager, dat een rotatie van de volledige
gondel t.o.v. de toren mogelijk maakt.

De 2de planetaire tandwieltrap heeft schuine vertanding en zijn planetendrager


wordt aangedreven door de zon van de 1ste trap. Ook hier staat het ringwiel
stil t.o.v. de tandwielkast. De zon van deze trap drijft het wiel van de pa-
rallelle tandwieltrap aan. Deze laatste trap heeft schuine vertanding en het
rondsel roteert aan de snelheid van de generator. De schijfrem is bevestigd op
de rondselas, die via een flexibele koppeling verbonden is met de generator.
Aangezien enkel de parallelle trap een omkering van de draaizin veroorzaakt,
draait de generator tegengesteld aan de rotor.

Paragraaf 5.1 bespreekt de analyse van de parallelle trap in de tandwielkast


als toepassing van de flexibele MLS-formulering. Vervolgens beschrijft para-
graaf 5.2 de analyse van de 2de tandwieltrap als voorbeeld van een planetaire
trap met schuine vertanding. Uiteindelijk bespreekt paragraaf 5.3 de analyse
van de aandrijflijn met inbegrip van de koppeling met de toren, de rotor en de
generator.
XVIII Nederlandse samenvatting

naaf
C
C
2de planetaire trap
C
C
C
C parallelle trap
C
C
C
C flexibele koppeling
C 


 
 

 D
1ste planetaire trap D

 D
D 
  D 

generator
 D
 
frame
D
D
schijfrem

kruilager

Figuur 5.1: Vereenvoudigde voorstelling van de windturbine en de aandrijflijn.

5.1 Een flexibel meerlichamen-model van de parallelle trap


Voor de analyse van de parallelle trap worden de drie modelleringstechnieken
uit hoofdstuk 3 toegepast. Figuur 5.2 toont de drie opeenvolgende modellen,
waarin het aantal vrijheidsgraden steeds toeneemt en er steeds meer flexi-
biliteiten in rekening worden gebracht.

1. Het puur torsioneel model in figuur 5.2(a) heeft twee vrijheidsgraden en


het tandcontact is het enige vervormbare element in dit systeem.

2. Het MLS in figuur 5.2(b) bevat zes vrijheidsgraden voor het wiel en
voor het rondsel. Dit model bevat bovendien veer-elementen voor het
beschrijven van de flexibiliteit in de lagers.

3. Het flexibel MLS in figuur 5.2(c) is de meest realistische voorstelling


van de parallelle tandwieltrap. Dit model bevat bijkomend 23 compo-
nent modes voor het rondsel en 12 component modes voor het wiel om
hun respectievelijke vervormingen te beschrijven.
Nederlandse samenvatting XIX

κgear · (rb,gear
0 · cos β0b )2
OCC
C
Jwiel Jrondsel
igear

(a) Torsioneel model met één vrijheidsgraad per lichaam: het tandcontact is
het enige vervormbare element.

lagers van het rondsel



lagers van het wiel 
PP
H
HP HP 
H 
HH
H
HH
H
Z
Y
6
Q
sX
Q
(b) MLS met zes vrijheidsgraden per lichaam en bijkomend veer-elementen
voor het beschrijven van de lagers.

Z
6 Y

Q
sX
Q
(c) Flexibel MLS met respectievelijk 23 en 12 extra vrijheidsgraden voor het
rondsel en het wiel om hun respectievelijke vervorming te beschrijven.

Figuur 5.2: Drie modellen voor de parallelle tandwieltrap.


XX Nederlandse samenvatting

Nr. Torsioneel Star Flexibel Modevorm


model MLS MLS
(Hz) (Hz) (Hz)

1 0 0 0 Rigid-body mode (a)


2 - 409 400 x-Translatie rondsel (b)
3 - 490 473 x-Translatie wiel (c)
4 1426 696 510 y-z-Rotatie rondsel (d)
5 751 510 y-z-Rotatie rondsel (e)
6 755 609 y-z-Rotatie wiel (f)
7 764 648 y-z-Rotatie wiel (g)
8 772 682 y-z-Translatie wiel (h)
9 784 715 y-z-Translatie wiel (i)
10 1045 874 y-z-Translatie rondsel (j)
11 1285 1074 y-z-Translatie rondsel (k)
12 2011 1642 x-Rotatie rondsel & wiel (l)

Tabel 5.1: Vergelijking van de eigenfrequenties van de parallelle tandwieltrap


berekend met een torsioneel model, met een MLS met zes vrijheidsgraden per
lichaam (star MLS) en met een flexibel MLS.

Tabel 5.1 bevat de resultaten van de berekening van de eigenmodes voor de


drie modellen en leidt tot volgende conclusies:

1. De vergelijking van de eigenfrequenties voor het torsionele model en


voor het MLS met zes vrijheidsgraden per lichaam, toont aan dat de
tweede benadering een betere beschrijving geeft voor het dynamisch
gedrag. Er worden extra eigenmodes geı̈dentificeerd in hetzelfde fre-
quentiebereik, die ook kunnen geëxciteerd worden door b.v. de ingrijp-
frequenties. Bovendien bevat de tweede benadering een realistischere
voorstelling van de flexibiliteiten in de lagers. De vergelijking toont aan
dat deze niet mogen verwaarloosd worden, aangezien ze een grote in-
vloed hebben op de eigenfrequenties van het systeem.

2. De vergelijking van de eigenfrequenties voor het MLS met zes vrijheids-


graden per lichaam en het flexibele MLS toont aan dat ook de flexi-
biliteiten van beide assen een niet te verwaarlozen invloed hebben op
de eigenfrequenties. De CMS-techniek is een geschikte techniek om
deze flexibiliteiten te beschrijven en de toepassing ervan voor de dy-
namische analyse van tandwielkasten is een belangrijke bijdrage van dit
proefschrift.
Nederlandse samenvatting XXI

De toepassing van de CMS-reductietechniek vereist het opstellen van


een EE-model en het berekenen van een set statische en dynamische
modes. Aangezien de dynamische modes van de afzonderlijke onderde-
len in een tandwielkast typisch hoger liggen in frequentie dan de eigen-
modes berekend met een MLS met zes vrijheidsgraden per lichaam, kun-
nen de dynamische modes meestal verwaarloosd worden. Het volstaat
dan om de flexibiliteiten van de onderdelen voor te stellen door discrete
veer-elementen in een MLS. Dit bespaart het werk voor het opstellen
van een EE-model en maakt van de MLS-formulering met zes vrijheids-
graden per lichaam de meest geschikte en meest efficiënte modellerings-
techniek voor het berekenen van eigenmodes. Wanneer er echter inzicht
vereist is in de lokale vervorming van afzonderlijke onderdelen is de
flexibele MLS-formulering het enige bruikbare alternatief.

5.2 “Out-of-plane” modes van een planetaire tandwieltrap


Hoofdstuk 4 beschrijft hoe de eigenmodes voor een planetaire tandwieltrap
met drie planeten en met rechte vertanding kunnen opgesplitst worden in twee
categorieën, nl. rotationele modes en translationele modes. Deze paragraaf
introduceert een derde categorie voor planetaire systemen met schuine vertan-
ding, nl. out-of-plane modes4 .

De analyse gebeurt m.b.v. een MLS met zes vrijheidsgraden per lichaam voor
de 2de planetaire tandwieltrap in de aandrijflijn van de windturbine. Tabel 5.2
toont de resultaten van de berekening van de eigenmodes. De relevantie van de
out-of-plane modes wordt aangetoond door hun frequentiebereik. Dit overlapt
met het bereik van de andere modes en tevens met het bereik van de ingrijpfre-
quenties. Bovendien impliceren deze laatste ook excitaties uit het vlak o.w.v.
de schuine vertanding, zodat de out-of-plane modes kunnen aangestoten wor-
den. Figuur 5.3 toont een voorbeeld van de modevorm voor een out-of-plane
mode.

5.3 Model van de volledige windturbine


Voor de dynamische analyse van de volledige windturbine wordt er een on-
derscheid gemaakt tussen het frequentiebereik onder en boven 10 Hz. In het
eerste bereik is het belangrijk om de flexibiliteit van de rotor en de toren op
een juiste manier te beschrijven, zoals besproken in hoofdstuk 2 voor de tradi-
tionele simulatiecodes. Deze paragraaf bespreekt een analyse in het frequen-
tiebereik boven 10 Hz. In dit bereik wordt er verondersteld dat de rotor zich
gedraagt als een star lichaam met een grote inertie en dat de toren reageert als
een starre ondersteuning van de gondel van de windturbine.
4 Determ “out-of-plane” betekent “uit het vlak” en verwijst naar de overeenkomstige be-
weging van de onderdelen van de planetaire tandwieltrap.
XXII Nederlandse samenvatting

Rotationele mode (m = 1)

− →
− →
− →
− →
− →

R1 R2 R3 R4 R5 R6
0 703 1035 1563 2243 -
Translationele mode (m = 2)

− →
− →
− →
− →
− →

T1 T2 T3 T4 T5 T6
131 822 1110 1532 2269 4645
Out-of-plane mode

− →
− →
− →
− →
− →
− →

O1 O2 O3 O4 O5 O6 O7
(m = 2) (m = 2) (m = 2)
90 545 581 3048 3054 3062 3127

Tabel 5.2: Eigenfrequenties (Hz) berekend met een MLS met zes vrijheidsgra-
den per lichaam voor de planetaire trap met schuine vertanding. De resultaten
zijn onderverdeeld in drie categorieën op basis van de overeenkomstige mod-
evormen.

Figuur 5.3: Voorbeeld van een out-of-plane mode (545 Hz), berekend op basis
van een MLS met zes vrijheidsgraden per lichaam voor de 2de planetaire tand-
wieltrap in de windturbine. De pijlen duiden de respectievelijke verplaatsing
uit het vlak aan van twee planeten.
Nederlandse samenvatting XXIII

Paragrafen 5.3.1, 5.3.2 en 5.3.3 bespreken achtereenvolgens de identificatie


van de eigenmodes, een frequentie-respons analyse en de simulatie van een
transiënt belastingsgeval voor de windturbine. Het volledige model bestaat
uit een model van de tandwielkast op basis van een MLS met zes vrijheids-
graden per lichaam, een generator met één rotationele vrijheidsgraad en een
starre rotor met zes vrijheidsgraden. Het model telt 70 vrijheidsgraden, wat
een aanzienlijke vermeerdering is t.o.v. het structurele model in de traditionele
simulatiecodes.

5.3.1 Identificatie van de eigenmodes


Tabel 5.3 toont de eigenmodes die werden berekend voor het model van de
windturbine. Deze resultaten worden als volgt samengevat.

• Er wordt een onderscheid gemaakt tussen globale modes en lokale mo-


des. De modevorm van lokale modes beschrijft enkel een vervorming in
een bepaalde tandwieltrap. Uit vergelijking met de berekende eigen-
modes voor deze tandwieltrap blijkt bovendien dat een lokale mode
reeds bij benadering kan voorspeld worden op basis van een model van
de individuele tandwieltrap. Globale modes beschrijven een vervorm-
ing van de volledige aandrijflijn en vereisen bijgevolg een model van het
volledige systeem.

• De globale mode met een frequentie van 5.6 Hz beschrijft een torsionele
vervorming van de volledige aandrijflijn. Een correcte voorspelling van
deze mode vereist een juiste beschrijving van de flexibiliteit van de rotor
en wordt daarom verder niet beschouwd in de analyses in dit hoofdstuk.
Dit geldt ook voor de dubbele mode bij 32 Hz, die overeenkomt met een
doorbuiging van de rotor in het hoofdlager.

• De eerste relevante eigenmode van het systeem is bijgevolg de globale


mode met een frequentie van 68 Hz (mode nr. 2). Deze komt overeen
met de torsionele vervorming van de flexibele koppeling.

5.3.2 Frequentie-respons analyse


Deze paragraaf demonstreert hoe de resultaten uit de vorige paragraaf moeten
worden geı̈nterpreteerd om resonantie in de aandrijflijn te vermijden en hoe
de belasting kan worden gesimuleerd voor een sinusoı̈dale excitatie. Beide
analyses zijn gebaseerd op de berekening van een frequentie-respons-functie
(FRF).
XXIV Nederlandse samenvatting

Nr. 2de planetaire trap


Rotationele Translationele Out-of-plane
Globale mode mode mode
Nr. mode (m=1) (m=2)
1 0 6 81
2 5.6 7 140
3,4 32 14 302
5 68 23,24 527
38,39 1033 25 568
44 1506 27 630
45 1527 34 864
51 2390 40 1093
52 2396 42 1151
53 2560 46 1558
48 1564
54 3048
55,56 3054
57 3066
58,59 3123
Nr. 1ste planetaire trap
Rotationele Translationele Out-of-plane
Nr. Parallelle mode mode mode
trap (m=1) (m=2)
17 346 9 205
18 406 11,12,13 227
19 409 15 305
21 430 20 423
25 562 22 513
32 750 29 660
33 832 31 731
43 1238 35 1003
50 1731 37 1011
49 1640

Tabel 5.3: Eigenfrequenties (Hz) van het MLS van de volledige windturbine,
waarin de rotor en de toren werden beschouwd als starre lichamen. De resul-
taten zijn geordend volgens de positie van de knopen in hun overeenkomstige
modevormen en, voor de planetaire tandwieltrappen, volgens het type mode.
Nederlandse samenvatting XXV

Om schadelijke resonantie te kunnen vermijden, moet er nagegaan worden


welke eigenmodes de belasting in de aandrijflijn aanzienlijk kunnen beı̈nvloed-
en. Dit zijn de eigenmodes die kunnen aangestoten worden door een bepaalde
excitatiebron en aanleiding geven tot een verhoogd belastingsniveau. Dit kan
onderzocht worden door het berekenen van een FRF tussen de excitatiebron en
een bepaalde belasting in de aandrijflijn.

Dit principe wordt gedemonstreerd voor een excitatie in het generatorkoppel.


De excitatie heeft een minimale frequentie van 50 Hz, omdat de eerste rele-
vante eigenmode van het systeem geı̈dentificeerd werd op 68 Hz. Figuur 5.4
toont de FRF tussen deze excitatie en de belasting op de zon in de 2de plane-
taire trap.
PSD [dB/Hz]

Frequentie [Hz]

Figuur 5.4: Frequentie-respons-functie die de kwalitatieve relatie weergeeft


tussen een sinusoı̈dale excitatie in het generatorkoppel en het resulterend kop-
pel op de zon van de 2de planetaire tandwieltrap.

Een vergelijking van de eigenmodes uit tabel 5.3 en de FRF in figuur 5.4 geeft
volgende inzichten:

1. De globale mode bij 68 Hz wordt duidelijk geëxciteerd en is de domi-


nante frequentie in het resulterend koppel op de zon.

2. De lokale translationele mode van de 2de planetaire tandwieltrap (140 Hz)


kan ook aanleiding geven tot een verhoging van het koppel op de zon.

3. De eerste rotationele mode van de 2de planetaire trap (302 Hz) wordt
eveneens aangestoten door een variatie van het koppel in de generator en
kan bijgevolg leiden tot verhoogde belastingsniveaus in de aandrijflijn.

4. De respons boven 302 Hz wordt voornamelijk bepaald door lokale mo-


des in de parallelle tandwieltrap.
XXVI Nederlandse samenvatting

Dezelfde inzichten kunnen afgeleid worden voor andere excitatiebronnen in de


aandrijflijn. Wanneer de excitatiefrequenties variëren met het toerental van de
windturbine (b.v. de ingrijpfrequenties), kan een voorstelling in een Campbell-
diagram aantonen of er resonantie kan optreden bij bepaalde toerentallen.

5.3.3 Simulatie van een transiënt belastingsgeval

Deze paragraaf illustreert hoe de simulatie van een transiënt belastingsge-


val met een gedetailleerd model voor de aandrijflijn meer inzicht geeft en
leidt tot nauwkeurigere resultaten dan met de traditionele simulatiecodes. Een
transiënte analyse vereist een nauwkeurige beschrijving van de externe krachten
op de windturbine. Deze beschrijving is enkel mogelijk op basis van gede-
tailleerde modellen voor alle modules in figuur 2.1. Deze zijn hier niet beschik-
baar en daarom wordt er een veronderstelling gemaakt voor de externe belas-
ting op de windturbine.

(a) normale opstartprocedure (b) een plotse koppelpiek tijdens het opstarten

Figuur 5.5: Verloop van het generatorkoppel tijdens het transiënte belastings-
geval.

Het gesimuleerde belastingsgeval betreft een plotse aanzienlijke verhoging van


het koppel in de generator. Een dergelijk fenomeen kan optreden a.g.v. ver-
schillende oorzaken, zoals storingen in het elektriciteitsnetwerk beschreven
door o.a. Soens e.a. [198] en Seman e.a. [189, 190]. In dit voorbeeld treedt de
koppelpiek op tijdens het opstarten van de windturbine. Figuur 5.5(a) toont
het generatorkoppel tijdens een normale opstartprocedure. Figuur 5.5(b) be-
schrijft hetzelfde verloop met daarop gesuperponeerd een plotse koppelpiek
bij t = 0.5 s. De beschrijving van deze piek is gebaseerd op een voorbeeld van
een kortsluiting in het elektriciteitsnetwerk beschreven in [189].
Nederlandse samenvatting XXVII

Figuur 5.6 toont een vergelijking tussen de simulaties voor de normale opstart-
procedure en de opstart met een plotse koppelpiek. Figuur 5.6(a) toont het
verloop van het resulterend koppel op het rondsel in de parallelle tandwieltrap.
Figuur 5.6(b) geeft weer hoe dit rondsel versnelt in zijn lagers. De vergelijking
leidt tot de volgende conclusies:

• De plotse koppelpiek in de generator a.g.v. een storing in het elektriciteit-


snetwerk leidt tot een koppelpiek op het rondsel. Het niveau van de
resulterende koppelpiek op het rondsel ligt in dit voorbeeld ongeveer
3.5 keer lager dan het niveau van de piek in de generator. Bovendien
exciteert de plotse impact een laagfrequente mode in de aandrijflijn, die
leidt tot het omkeren van het koppel. Hierdoor zal de aandrijflijn her-
haaldelijk door zijn speling gaan, hetgeen belangrijk kan zijn voor het
ontwerp van de lagers.

• De storing in het elektriciteitsnetwerk leidt tot grote versnellingsniveaus


voor het rondsel in zijn lagers. De piekwaarde van deze versnelling
ligt in dit voorbeeld ongeveer 30 keer hoger dan tijdens een normale
opstartprocedure. Ook dit kan een impact hebben voor het ontwerp van
de lagers.

6 Meetcampagne op een moderne windturbine


De beschrijving van de meetcampagne bevat confidentiële informatie van Han-
sen Transmissions International NV en werd niet gepubliceerd in de publieke
versie van dit proefschrift. De publieke versie van dit hoofdstuk geeft enkel een
kort overzicht van de meetcampagne.

6.1 Overzicht van de meetcampagne


De meetcampagne werd uitgevoerd op een moderne multi-MW windturbine en
het doel van de metingen is de werking en het gedrag van de windturbine en de
aandrijflijn te identificeren. Een grote verzameling meetdata werd aangelegd
en de experimentele analyses werden opgesplitst in drie deeltaken:

1. karakteriseren van het gedrag tijdens normale werking

2. onderzoek van transiënte belastingsgevallen

3. identificatie van het dynamisch gedrag van de windturbine


XXVIII Nederlandse samenvatting

(a) Koppel op het rondsel van de parallelle tandwieltrap

(b) Rotationele versnelling van het rondsel in zijn lagers

Figuur 5.6: Vergelijking van de resultaten voor de transiënte simulaties


(streeplijn: normale opstartprocedure; volle lijn: opstart inclusief koppelpiek).

7 Algemene conclusies
De state-of-the-art in de simulatie van de belasting op de aandrijflijn in een
windturbine is het gebruik van simulatiecodes, die specifiek zijn ontwikkeld
voor het ontwerpen van windturbines. Het structureel model in zo een simu-
latiecode bevat slechts zestien tot vierentwintig vrijheidsgraden voor het be-
schrijven van de volledige windturbine. Slechts één enkele vrijheidsgraad stelt
de torsie in de aandrijflijn voor en dit leidt tot beperkingen in de betrouw-
baarheid van het ontwerp. Dit impliceert immers een quasi-statisch ontwerp
van alle onderdelen van de aandrijflijn, terwijl er ook dynamische belastings-
verhogingen kunnen optreden a.g.v. interne excitaties. Bovendien geeft dit
geen inzicht in lokale spanningsniveaus, noch in de belastingswisselingen tij-
dens bepaalde transiënte belastingsgevallen. Meer gedetailleerde simulatie-
modellen zijn hier vereist en dit proefschrift beschrijft de ontwikkeling van een
consistente modelleringstechniek op basis van de (flexibele) MLS-formulering.
Nederlandse samenvatting XXIX

Een correct dynamisch structureel model van de aandrijflijn in een windturbine


vereist een nauwkeurige beschrijving van de massa en inertie van de verschil-
lende onderdelen, van het verloop van alle krachten en van de flexibiliteiten en
de demping in de lagers, in de vertanding en in de afzonderlijke componenten.
Voor de beschrijving van de lagers en de vertanding in dit proefschrift volstaat
het gebruik van lineaire veer-elementen zonder demping. Voor de opbouw van
het structureel model worden drie afzonderlijke technieken voorgesteld met
een toenemende mate van detaillering:

1. De torsionele MLS tellen één rotationele vrijheidsgraad per lichaam. In


deze modellen kunnen de flexibiliteiten in de aandrijflijn moeilijk op een
realistische manier in rekening worden gebracht. Bovendien is de output
van de simulaties hier beperkt tot het koppel in de aandrijflijn.

2. De MLS met discrete flexibele elementen tellen zes vrijheidsgraden per


lichaam. De ontwikkeling in dit proefschrift van een individuele beschri-
jving voor een lager en voor de tandcontactkracht geeft twee driedimen-
sionale modellen, die kunnen gebruikt worden in analyses van parallelle
en planetaire tandwieltrappen, zowel met rechte als met schuine ver-
tanding, maar ook in de analyse van volledige aandrijflijnen. Vooral
de generieke methodologie achter deze formulering is waardevol in het
modelleren van dergelijke systemen.

3. De flexibele MLS-formulering laat toe van statische en dynamische mo-


des van een component in de aandrijflijn als extra vrijheidsgraden te be-
schouwen in het model. De modes stellen respectievelijk de statische en
de dynamische vervorming van een lichaam voor en worden berekend op
basis van de CMS-reductietechniek voor een EE-model van de compo-
nent. De combinatie van deze techniek met de modellen voor de lagers
en de vertanding is een belangrijke ontwikkeling voor de simulatie van
lokale spanningsniveaus in tandwielkasten.

Dit proefschrift illustreert de voordelen van de tweede modelleringstechniek


t.o.v. de eerste aan de hand van een analyse van drie planetaire tandwieltrap-
pen, die werden beschreven door Lin en Parker in [129]:

1. de flexibiliteit van de lagers kan op een correcte manier beschreven wor-


den

2. de berekening van de eigenmodes op basis van het tweede model iden-


tificeert een aantal bijkomende modes, die ook geëxciteerd kunnen wor-
den door bepaalde belastingsvariaties en daarom niet verwaarloosd mo-
gen worden
XXX Nederlandse samenvatting

De analyses bespreken ook de classificatie van de eigenmodes in rotationele,


translationele en planeet modes.

Uiteindelijk beschrijft dit proefschrift de toepassing van de ontwikkelde tech-


nieken voor de analyse van de aandrijflijn in een windturbine. Het voorbeeld
betreft een tandwielkast met een geı̈ntegreerd hoofdlager, die bestaat uit een
planetaire trap met rechte vertanding, een planetaire trap met schuine ver-
tanding en een parallelle trap met schuine vertanding. De analyse van deze
laatste tandwieltrap m.b.v. drie verschillende modellen toont aan dat een MLS-
formulering met zes vrijheidsgraden per lichaam de meest geschikte en meest
efficiënte modelleringstechniek is voor het berekenen van eigenmodes. De
berekening van de eigenmodes voor de planetaire tandwieltrap met schuine
vertanding leidt tot de introductie van een nieuw type modes, nl. out-of-plane
modes.

Het finale model voor de analyse van de volledige windturbine in het frequen-
tiebereik boven 10 Hz, telt 70 vrijheidsgraden. De rotor en de toren worden in
dit model star verondersteld. Het volledige systeem heeft een eerste eigenmode
op 68 Hz. De berekening van een FRF tussen een koppelvariatie in de gene-
rator en het resulterend koppel op de zon van de 2de planetaire trap, illustreert
hoe resonantie kan geı̈dentificeerd worden en hoe de belasting kan gesimuleerd
worden voor een sinusoı̈dale excitatie. De simulatie van een transiënt belas-
tingsgeval wordt gedemonstreerd voor een plotse verhoging van het koppel in
de generator tijdens het opstarten van de windturbine. De impact veroorzaakt
een koppelpiek op het rondsel en hoge versnellingen van het rondsel in zijn
lagers.

De beschreven voorbeelden tonen aan hoe de doelstellingen uit hoofdstuk 2


gerealiseerd worden.

You might also like