See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267855132

Design of Direct Driven Permanent Magnet Generators for Wind Turbines

Article · January 1996

CITATIONS READS

237 2,311

1 author:

Anders Grauers
Chalmers University of Technology
25 PUBLICATIONS   700 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Decision support for electric bus systems View project

All content following this page was uploaded by Anders Grauers on 25 February 2015.

The user has requested enhancement of the downloaded file.

 School of Electrical and Computer Engineering
CHALMERS UNIVERSITY OF TECHNOLOGY
Göteborg, Sweden.

Technical Report No. 292

Design of Direct-driven
Permanent-magnet Generators
for Wind Turbines
by

Anders Grauers

Submitted to the School of Electrical and Computer Engineering,

Chalmers University of Technology, in partial fulfillment of the

requirements for the degree of Doctor of Philosophy.

Department of Electric Power Engineering

Göteborg, October 1996.

2

CHALMERS UNIVERSITY OF TECHNOLOGY

Department of Electric Power Engineering
S - 412 96 GÖTEBORG, SWEDEN

ISBN 91-7197-373-7
ISSN 0346-718X
Chalmers Bibliotek, Reproservice
Göteborg, 1996
 3

Abstract
This thesis presents an investigation of how a direct-driven wind turbine
generator should be designed and how small and efficient such a
generator will be. Advantages and disadvantages of various types of
direct-driven wind turbine generators are discussed, and a radial-flux
permanent-magnet generator connected to a forced-commutated rectifier
is chosen for a detailed theoretical investigation. Further, a design
method is developed for the electromagnetic part of the chosen generator
type. The generator is optimized with a simplified cost function which,
besides including the cost of the active generator parts and the cost of the
structure, also includes the cost of the average losses. Therefore, a method
to calculate the average losses is derived. The design method is used to
investigate the optimization of a 500 kW generator, and the size, efficiency
and active weight of optimized generators from 30 kW to 3 MW are
presented. A result of the investigation is that the outer diameters of the
direct-driven generators are only slightly larger than the width of
conventional wind energy converter nacelles. A comparison of average
efficiency shows that direct-driven generators, including the losses in the
frequency converters, are more efficient than conventional wind energy
converter drive trains. Compared with other direct-driven generators, the
proposed generator type is small, mainly because of the forced-
commutated rectifier and because the generator is not required to produce
a pull-out torque higher than the rated torque.

Preface
The work presented in this thesis was carried out at the Department of
Electrical Power Engineering at Chalmers University of Technology. The
work was financed by NUTEK, partly under the Swedish wind energy
research program, partly as an EU Joule II project. The financial support
is gratefully acknowledged.
I would like to thank Dr Ola Carlson for interesting discussions about the
application of the low-speed generator in wind energy converters. Deborah
Fronko and Margot Bolinder both made an important contribution by
revizing my English. Finally, I wish to express my gratitude to Professor
Jorma Luomi for valuable help with the generator design and for
discussions on how to present the results from my project.
4

Table of Contents
Abstract 3
Preface 3
Table of Contents 4
List of Symbols 6
1 Introduction 11
1.1 Why Use Direct-driven Wind-turbine Generators 11
1.2 Differences Compared with Conventional Generators 12
1.3 Proposed Generator Types 12
1.3.1 Sector Induction Generator 12
1.3.2 Electrically Excited Synchronous Generator 13
1.3.3 Switched Reluctance Generator 14
1.3.4 Permanent-magnet Radial-flux Synchronous Generator 14
1.3.5 Axial-flux Generators 16
1.3.6 Transversal-flux Variable-speed Generator 17
1.4 Discussion of Earlier Research 19
1.5 Goal and Outline of the Thesis 19
2 Generator Specification and Cost Function 21
2.1 Specification 21
2.2 Generator Cost Function 23
2.2.1 Cost of Active Parts 24
2.2.2 Cost of Structure 24
2.2.3 Cost of Average Losses 24
2.2.4 Total Cost Function 26
3 Calculation Method for the Average Losses 27
3.1 Average Losses 27
3.2 Average Efficiency and Average Power 29
3.3 Determining Average Loss Factors 30
4 Generator Types 37
4.1 Electrical Excitation or Permanent Magnets 37
4.2 Direct Grid Connection or Frequency Converter 39
4.3 Surface Magnets or Flux Concentration 40
4.4 Slot Winding or Air Gap Winding 41
4.5 Radial-, Axial- and Transversal-flux Machines 42
4.6 Forced-commutated Rectifier or Diode Rectifier 44
4.6.1 Generator Model 45
4.6.2 Diode Rectifier 45
4.6.3 Forced-commutated Rectifier 46
4.6.4 Rectifier Comparison 48
4.7 Chosen Generator Type 51
4.7.1 Basic Generator Concept 51
4.7.2 Details of the Chosen Generator 51
4.7.3 Materials 52
 5

5 Design Method for a Permanent-magnet Generator 55

5.1 Design Variables 55
5.2 Design Equations 58
5.2.1 General Definitions 58
5.2.2 Magnetic Circuit 60
5.2.3 Stator Inductance and Resistance 61
5.2.4 Material Volume and Weight 63
5.2.5 Losses 64
5.2.6 Voltage, Power and Efficiency 67
5.2.7 Thermal Model and Temperature Rise 68
5.2.8 Irreversible Demagnetization 69
5.3 Calculation Procedure 71
5.4 Test of the Design Method 72
5.4.1 Comparison with Finite Element Calculations 72
5.4.2 Test of Thermal Model 73
6 Generator Optimization 77
6.1 Optimum 500 kW Generators 77
6.1.1 Optimized Reference Generator 77
6.1.2 Optimized Generators for 50 Hz and 200 % Peak Power 80
6.1.3 Optimization Using the Losses at Rated Load 82
6.2 Sensitivity to Variable Changes 84
6.3 Sensitivity to Cost Function Changes 86
6.3.1 Cost of Losses 86
6.3.2 Cost of Iron and Copper 87
6.3.3 Cost of Permanent Magnets 88
6.3.4 Cost of the Structure 89
6.4 Optimum Generator Diameter 90
6.5 Typical 500 kW Permanent-magnet Generator 92
7 Design and Comparison 95
7.1 Generators from 30 kW to 3 MW 95
7.1.1 Generator Data 95
7.1.2 Optimum Variables and Parameter Values 97
7.1.3 Power Limits For the Direct-driven Generators 100
7.2 Comparisons 102
7.2.1 Comparison with Conventional Generators and Gears 102
7.2.2 Comparison with Other Direct-driven Generators 104
8 Conclusions 107
8.1 Different Generator Types 107
8.2 Generator Design and Optimization 108
8.3 Designed Generators and Comparison with Other Generators 108
8.4 Further Work 109
References 111
Appendix A Magnetizing Inductance 115
Appendix B Thermal Model of the Generator 119
Appendix C Average Efficiencies 131
6

List of Symbols

Symbol Unit Description

a – Exponent for the structure cost
A – Parameter for wind speed probability density function
bCu m Conductor width
bd m Tooth width
bm m Magnet width
bs m Slot width
bs1 m Slot opening
^
B d0 T Maximum flux density in the teeth at no-load
Bmin T Min. flux density allowed in the permanent-magnets
^
B s T Peak air gap flux density generated by the stator at
rated current
^
Byr T Maximum flux density in the rotor yoke
^
B ys T Maximum flux density in the stator yoke
^
Bδ0 T Peak air gap flux density at no-load
Bδ(1) T Fundamental air gap flux density (RMS value)
c – Parameter for wind speed probability density function
cC u ECU/kg Specific cost of the copper
cd ECU/kW Specific cost of average losses
c el ECU/kWh Specific cost of electric energy
c Fe ECU/kg Specific cost of the active iron
cm ECU/kg Specific cost of the permanent magnets
cstr ECU Cost of a reference structure
Cact ECU Cost of the active parts of the generator
Cd ECU Cost of the losses of the generator
Cstr ECU Cost of the generator structure
C tot ECU Total cost function for generator optimization
d m Air gap diameter
dref m Diameter of the reference structure
dse m Outer diameter of the stator
E V Internal line-to-line emf
Ep V Internal phase emf
EpN V Internal phase emf at rated speed
 7

f Hz Frequency
fN Hz Rated frequency
gad – Factor for the additional losses
gC u – Factor for the copper losses
g Ft – Factor for the eddy current losses
g Hy – Factor for the hysteresis losses
gt – Factor for the turbine power
gµ – Factor for the windage and friction losses
hC u m Conductor height
hi m Insulation height (in the slots)
hm m Magnet height
hs m Stator slot height
hs1 m Tooth tip height 1
hs2 m Tooth tip height 2
hs3 m Slot height, excluding tooth tips
h yr m Rotor yoke height
hys m Stator yoke height
Hc A/m Coercivity of the magnets
i – Real interest rate
Ia A Armature current
IaN A Rated armature current
I(1) A Fundamental armature current
Js A/m 2 Current density in the armature winding
k Cu — Copper fill factor of the stranded wire
kdad — Average loss factor for additional losses
kdCu — Average loss factor for copper losses
kdFt — Average loss factor for eddy current losses
kdHy — Average loss factor for hysteresis losses
kdµ — Average loss factor for windage and friction losses
kFes — Fill factor for the stator iron
kFtd — Empirical eddy current loss factor for the teeth
kFtys — Empirical eddy current loss factor for the stator yoke
kHyd — Empirical hysteresis loss factor for the teeth
kHyys — Empirical hysteresis loss factor for the stator yoke
kN — Factor for the present value of future costs
kt — Average factor for the turbine power (capacity factor)
8

kw(1) — Winding factor for the fundamental voltage

l m Active length of the generator
lb m End winding length
le m Equivalent core length
lref m Length of the reference structure
l tot m Total length of the stator and end windings
lu m Useful length of the stator core
La H Total armature inductance per phase
Lb H End winding leakage inductance
Lm H Magnetizing inductance per phase
Lsl H Slot leakage inductance
L tl H Tooth tip leakage inductance
Lσ H Total leakage inductance
m — Number of phases
mCu kg Copper weight
mFe kg Iron weight of the active parts of the generator
m Fed kg Teeth weight
m Feyr kg Rotor yoke weight
m Feys kg Stator yoke weight
mm kg Weight of the magnets
m tot kg Total active weight of the generator
n rpm Generator speed
nN rpm Rated speed of the generator
N WEC — Number of years of the assumed wind energy
converter life time
Ny — Number of hours per year
p — Number of pole pairs
pFt W/kg Eddy current loss density at 50 Hz and 1.5 T
pFtm W/m 2 Eddy current loss density of the magnet surface
pHy W/kg Hysteresis loss density at 50 Hz and 1.5 T
Pa W Armature output power
Pad W Additional losses
PadN W Additional losses at rated load
PaN W Rated electrical power
PAv W Average turbine power
P Cu W Copper losses
 9

P CuAv W Copper losses at average ambient tempearture

PCuMax W Copper losses at maximum ambient tempearture
PCuN W Copper losses at rated load (only in Chapter 3)
Pd W Total generator losses
PdAv W Average losses
PFt W Total eddy current losses
PFtd W Eddy current losses in the teeth
PFtm W Eddy current losses in the magnets
PFtN W Total eddy current losses at rated load
PFtys W Eddy current losses in the stator yoke
PHy W Total hysteresis losses
PHyd W Hysteresis losses in the teeth
PHyN W Total hysteresis losses at rated load
PHyys W Hysteresis losses in the stator yoke
PlossAv W Total average losses
PlossMax W Total losses at maximum ambient temperature
PN W Rated mechanical input power of the generator
Pt W Active power from the turbine
Pµ W Friction and windage losses
PµN W Friction and windage losses at rated load
Q — Number of slots
q — Number of slots per pole and phase
Ra Ω Armature resistance
TMax Nm Pull-out torque
TN Nm Rated torque of the generator
Ua V Armature voltage
Uap V Armature phase voltage
UapN V Rated armature phase voltage
v m/s Wind speed
vin m/s Cut-in wind speed
vN m/s Wind speed at which the rated power is reached
vnN m/s Wind speed at which the rated speed is reached
v out m/s Cut-out wind speed
^
v A Mmf drop of the teeth
d
^
v A Mmf drop of the magnets
m
^
v A Mmf drop of the rotor yoke
yr
10

^
v ys A Mmf drop of the stator yoke
^
v A Mmf drop of the air gap
δ
VC u m3 Copper volume
V Fed m3 Stator teeth volume
V Feyr m3 Rotor yoke volume
V Feys m3 Stator yoke volume
VM m3 Magnet volume
w s/m Weibull distributed probabillity density of wind speeds
W m Winding pitch
xa m Per unit armature reactance
Xa m Armature reactance

ϕ Rad Terminal phase angle

ϕN Rad Terminal phase angle at rated load
δ m Mechanical air gap
δef m Effective air gap
θC u K Temperature of the winding
θ CuAv K Tem p. of the winding at average am bient tem p.
θCuN K Maximum temperature of the winding
θm K Temperature of the magnets
τ m Slot pitch
τp m Pole pitch
µ0 — Permeabillity of air
µm — Relative permeabillity of the magnet material
ηav — Average efficiency
ηN — Efficiency at rated load
Ψ Vs Flux linkage of the stator winding
ΨN Vs Flux linkage at rated load
1 Introduction

1.1 Why Use Direct-driven Wind-turbine Generators

Today almost all wind energy converters of a rated power of a few
kilowatts or more use standard generators for speeds between 750 and
1800 rpm. The turbine speed is much lower than the generator speed;
typically between 20 and 60 rpm. Therefore, in a conventional wind energy
converter a gear is used between the turbine and the generator. An
alternative is to use a generator for very low speed. The generator can
then be directly connected to the turbine shaft. Direct-driven generators
are already in use in at least two large commercial wind energy
converters of 230 and 500 kW. The drive trains of a conventional 500 kW
wind energy converter and a wind energy converter with a direct-driven
generator are shown in Figure 1.1.
There are two main reasons for using direct-driven generators in wind
energy converters. The cost of the produced electricity and the noise of the
wind energy converter can both be reduced. Reducing the noise can be
important when applying for permission to erect wind energy converters
close to dwelling places. The reasons why direct-driven generators can
decrease the cost of the produced electricity are as follows:
• they can decrease the cost of the drive train;
• they can decrease the losses of the energy conversion;
• they can improve the availability of the wind energy
converter.
Nevertheless, to achieve all of these advantages simultaneously, an
efficient generator type, which is well optimized and rationally
manufactured, is required.

Gear Generator
1:47 1500 rpm

Generator
32 rpm

Figure 1.1 500 kW drive trains of one conventional wind energy converter
(left) and one with a direct-driven generator (right).
1.2 Differences Compared with Conventional Generators
Theoretically, direct-driven wind turbine generators do not differ much
from other generator types. They can be designed and built in the same
way as other generators. The most important difference between
conventional and direct-driven wind turbine generators is that the low
speed of the direct-driven generator makes a very high rated torque
necessary. This is an important difference, since the size and the losses of
a low-speed generator depend on the rated torque rather than on the rated
power. A direct-driven generator for a 500 kW, 30 rpm wind turbine has
the same rated torque as a 50 MW, 3000 rpm steam-turbine generator.
Because of the high rated torque, direct-driven generators are usually
heavier and less efficient than conventional generators. To increase the
efficiency and reduce the weight of the active parts, direct-driven
generators are usually designed with a large diameter. To decrease the
weight of the rotor and stator yokes and to keep the end winding losses
small, direct-driven generators are also usually designed with a small
pole pitch.

1.3 Proposed Generator Types

Many different generators have been proposed as direct-driven wind-
turbine generators. This section describes some of them and presents
some results from the research on these generators. The generators
described below are either direct grid-connected generators or variable-
speed generators connected to the grid via a frequency converter.
The generators can be divided into electrically excited generators and
permanent-magnet-excited generators. The electrically excited generators
presented are:
• sector induction generator;
• electrically excited synchronous generator;
• switched reluctance generator.
The permanent-magnet generators presented are:
• radial-flux synchronous generator;
• axial-flux synchronous generator with toroidal stator or
double-sided stator ;
• transversal-flux generator.

1.3.1 Sector Induction Generator

Gribnau and Kursten (1991) and Deleroi (1992) have presented a direct-
driven sector induction generator for direct grid connection, shown in
Figure 1.2. The generator is an axial-flux generator with a stator only on a
 Short-circuited rotor winding

Double-sided stator segment

Wind turbine tower
a) b)

laminated steel rotor structure

windings turbine blades

Figure 1.2 Axial-flux sector induction generator. a) Complete generator,

axial view, and b) Active part, tangential view.

segment of the circumference. It has a large diameter in order to achieve

a high air-gap speed, but since the stator segment is fixed to the tower of
the wind energy converter, the structure is simple. Since it is an induction
generator with a high slip, the damping of the generator is no problem for
the design, even though it is direct grid-connected. The generator is
developed by a Dutch company and only very little data is available on it. A
500 kW version with a diameter of about 9 m and a speed of about 40 rpm is
presented (Gribnau and Kursten, 1991). The efficiency is estimated to be
about 80 to 85 % with a rated slip of about 10 to 15 %. A 150 kW prototype
has been built and tested. The efficiency of the prototype is much lower
than the goal for the 500 kW version, only 65 % at a rated slip of about 20 %.

1.3.2 Electrically Excited Synchronous Generator

The two commercial direct-driven wind energy converters both use
electrically excited synchronous generators with frequency converters.
The generators are of the same principal design as hydropower
generators. Very little information is available on these generators. One of
them is a 500 kW generator and has a rated speed of 40 rpm (Anon. 1994a).
The air gap diameter is about 4 m and the generator frequency is lower
than 50 Hz. This generator is in 1996 used in more than 600 wind energy
converters.
1.3.3 Switched Reluctance Generator
de Haan et al. (1994) present a switched reluctance generator which
produces 20 kW at 120 rpm. The generator has been optimized using an
analytical design method. Finite element calculations are carried out for
the chosen design, but the finite element calculations predict a much
higher torque than the analytical model. The active part of the generator
is shown in Figure 1.3.

b) a)

Stator

Rotor

b) a)

a) b)

laminated steel windings

Figure 1.3 A switched reluctance generator. a) Tangential view and b)

Axial view.

1.3.4 Permanent-magnet Radial-flux Synchronous Generator

Spooner and Williamson (1992a,1992b,1992c) have discussed the feasibility
of direct grid-connected, direct-driven wind-turbine generators. The
proposed generators are permanent-magnet synchronous generators, and
when connected directly to the grid, they operate at a constant speed.
Rotors with surface-mounted high-energy magnets have been
investigated, as well as rotors with ferrite magnets and flux
concentration. The two versions of the generator are shown in Figures 1.4
and 1.5. The stator is of a conventional design, but with a very small pole
pitch, approximately 40 mm, in order to generate 50 Hz frequency without
having a large generator diameter. The winding is a three-phase,
fractional-slot winding with less than 1 slot per pole and phase. Spooner
and Williamson show that direct-driven, direct grid-connected generators
can be designed with a small diameter if permanent-magnet excitation is
used. Both the rotor types have been found to be feasible. The main
difference between them is that the generator with ferrite-magnet, flux-
concentrating rotor is heavier and more complicated but leads to a shorter
generator with higher efficiency at rated load. A radial-flux, permanent-
magnet generator with a modular stator design, was also presented by
Spooner et al. (1994).
 b) a)

Stator

Rotor

b) a)
a) b)

laminated steel magnets (magn. direction)

windings solid steel

Figure 1.4 The radial-flux generator with surface-mounted magnets.

a) Tangential view and b) Axial view.

b) a)

Stator

Rotor

b) a)
a) b)

laminated steel magnets (magn. direction)

windings non-magnetic material

Figure 1.5 The radial-flux generator with flux concentration and ferrite
magnets. a) Tangential view and b) Axial view.

Two problems of the proposed design are mentioned in the papers. First,
the sub-harmonic flux waves from the fractional-slot winding are
considered a problem, because they may lead to additional losses. Second,
parallel paths in the winding should be avoided. The reason for this is that
the poles may generate unequal voltages leading to circulating currents
between the parallel coils.
Two small generator prototypes have been made. They showed that a
fractional-slot winding with only 0.75 slots per pole and phase can
generate an almost sinusoidal voltage from a very non-sinusoidal flux
wave form, and that the flux-concentration method can be used to achieve
high flux densities from low-energy magnets.
A mechanical damping system for the direct grid-connected generator is
discussed by Westlake et al. (1996). In conventional synchronous
generators, the damping is provided by damper windings in the rotor. The
direct-driven direct grid-connected wind turbine generator must have a
very small pole pitch if the diameter is not very large. The small pole pitch
makes the damper windings insufficient. Instead, a mechanical damping
of the stator, by means of a spring and a damper, can be used. The
mechanical damping system is shown to be sufficient, but it may be
difficult and expensive to construct for large generators.
Lampola et al. (1995a) present a 500 kW radial-flux permanent-magnet
generator. The generator is of the same design as the one in Figure 1.4 but
it is not designed for direct grid-connection. The generator has been
calculated in detail using the finite element method and time stepping.
The torque ripple, cogging torque and rotor losses are kept minor by using
1.5 slots per pole and phase. The permanent-magnet generator is
compared with a direct-driven induction generator by Lampola (1995b).
The induction generator is found to be larger, heavier and less efficient
than the permanent-magnet generator. In another paper (Lampola et al.
1996b) the influence of the rectifier on the generator rated power and
efficiency is investigated. It is shown that the rated power and efficiency
are lower if the generator is connected to a diode rectifier than if it is
supplied with sinusoidal voltages.

1.3.5 Axial-flux Generators

Honorati et al. (1991), Di Napoli et al. (1991) and Carrichi et al. (1992) have
proposed a permanent-magnet axial-flux synchronous generator. The
generator design is shown in Figure 1.6. It is a generator with a toroidal
stator, air gap winding and two rotor discs.
Two prototypes of approximately 1 kW have been built. A 1 MW generator
design is presented, but it has a rated speed of 100 rpm, too high for a
1 MW wind energy converter. Because of the high speed, the generator
has high efficiency and low weight.
Alatalo and Svensson (1993) have proposed an axial-flux permanent-
magnet synchronous generator with a double-sided stator and air gap
windings. The generator design is shown in Figure 1.7. A 5 kW prototype
is presented by Alatalo (1991). The generator type has low iron weight
because there is no rotor yoke but the magnet weight is high since an air
gap winding is used.
 a)

Rotor 1

Stator

b) b)
Rotor 2
a)
a) b)

laminated steel magnets (magn. direction)

windings solid steel

Figure 1.6 An axial-flux generator with a toroidal stator winding and

surface-mounted magnets on two rotor discs. a) Tangential
view b) Radial view.

a)

Stator 1

Rotor
b) b)
Stator 2
a)

a) b)

laminated steel magnets (magn. direction)

windings non-magnetic material

Figure 1.7 An axial-flux generator with double-sided stator and air gap
windings. a) Tangential view b) Radial view.

1.3.6 Transversal-flux Variable-speed Generator

Weh et al. (1988) have proposed a direct-driven transversal-flux generator
for wind turbines. The generator is a two-phase generator and is
magnetized by permanent magnets with flux concentration. Since the
generator is a two-phase machine, it cannot be direct grid-connected.
Instead, it is connected to two single-phase rectifiers feeding one three-
phase inverter. This generator type is designed for a frequency in the
range of 100 to 200 Hz to get a high force-to-weight ratio. The transversal-
flux generator has a very high force per weight ratio but one disadvantage
is its complex structure. If the generator is connected to a diode rectifier,
the force density is lower than what would be possible to achieve with a
sinusoidal supply voltage because of high inductance. If connected to a
forced-commutated rectifier, the transversal-flux generator is capable of
producing higher force densities than conventional generator designs.
One phase of the transversal-flux generator is shown in Figure 1.8. Each
phase has a double-sided stator with two cylindrical windings around the
generator circumference.
A number of transversal-flux machine-prototypes have been built, one
example of which is a wind-turbine generator. It is a 5.8 kW generator for
195 rpm (Weh et al. 1988). A 55 kW generator design for 78 rpm is also
presented and compared with a conventional drive train consisting of a
gear and a four-pole generator (Weh et al. 1988). It is shown that the
weight of a 55 kW drive train is approximately halved by using the direct-
driven transversal-flux generator instead of a gear and an induction
generator.

b) a)

Upper part
of the stator {
Rotor
{
Lower part
of the stator {
b) a)
a) b)

laminated steel magnets (magn. direction)

windings non-magnetic material

Figure 1.8 One phase of the two-phase transversal-flux generator with

double-sided stator. The flux paths are shown. a) Tangential
view and b) Axial view.
1.4 Discussion of Earlier Research
It is clear, from the papers mentioned above, that many generator types
can be used as direct-driven wind-turbine generators. It is, however,
difficult to compare the generator types based on data given in the papers
since the generators are designed for different specifications, using
different methods and since all data is not presented.
An investigation of direct-driven generators in general, including a
comparison with conventional generators and gears, was made by
Bindner et al. (1995). In the report, the switched reluctance generator, the
induction generator, the electrically excited synchronous generator and
the permanent-magnet synchronous generator were discussed briefly.
Switched reluctance generators and induction generators have a low
power factor leading to a large stator, and electrically-excited
synchronous generators are larger and less efficient than permanent-
magnet synchronous generators. Consequently, the permanent-magnet
synchronous generator was found to be the best suited for a direct-driven
wind-turbine generator.
Generators of up to 1500 kW were designed and the size, the weight and
the efficiency were discussed by Søndergaard and Bindner (1995). In
comparison with conventional generators and gears, direct-driven
generators were found to have a much larger diameter, about the same
efficiency, about the same total weight and, at present, a slightly higher
price. The authors expect that there will be an upper power limit of about
500 to 1000 kW for direct-driven generators. The rated power is limited
mainly because the outer diameter becomes too large as the rated power
increases.
Since the size of the generator is very important for transportation and
manufacturing, it is important to further investigate how small the
diameter can be made. Also generator efficiency has to be investigated
more thoroughly to evaluate the cost of the losses. Since the losses change
with load, the cost of the losses cannot be calculated from the losses at
rated load. Instead, the average losses have to be evaluated based on a
typical wind speed distribution. The average losses of the generator
system have not been calculated in any of the papers mentioned above.

1.5 Goal and Outline of the Thesis

The goal of this thesis is to investigate how to design a permanent-magnet
generator in order to be suitable for application as a direct-driven wind-
turbine generator, and what can be expected of size and performance of
such a generator. The optimum generator diameter and length, and the
average efficiency are investigated in particular. The goal is also to reduce
the size of the generator in comparison with the generators presented in
Section 1.3. The investigation is theoretical and limited to the
electromagnetic part of a permanent-magnet generator, but the influence
of the generator structure on the electromagnetic optimization is also
included.
A specification and a cost function for the design and optimization of
direct-driven generators are defined. To calculate the cost of the losses, a
method to estimate the average losses of a wind turbine generator is
developed. Advantages and disadvantages of various generator types are
discussed, and a radial-flux permanent-magnet generator with a
frequency converter is chosen for further investigation.
A detailed design method is developed for the active parts of the chosen
generator type. The design method is based on well-known analytical
methods and a lumped-parameter thermal model. This design method is
used together with numerical optimization to find a design suitable for
use as a direct-driven wind turbine generator.
The influence of different parameters on the generator design is
investigated in detail for a 500 kW generator. Generators from 30 kW to
3 MW are then designed and their data are discussed. Some of these
generators are compared with conventional generators and gears and the
proposed generator type is also compared with direct-driven generators
proposed by other authors.
2 Generator Specification and Cost Function
This chapter presents a specification for direct-driven wind-turbine
generators of different rated powers. A method to estimate the total cost of
generators is also presented. The total cost includes the cost of the active
parts, the cost of the average losses and the cost of the generator
structure.

2.1 Specification
Only the quantities which significantly affect the generator performance
are included in the specification. A complete specification for the final
design of a generator will include many more detailed requirements. The
rated power used here is the mechanical power from the turbine, not the
electrical power to the grid. The reason for using mechanical power is
that generators designed for the same turbine should be compared with
each other, not with generators of the same output power at rated load.
The difference in generator efficiency is included as a cost of the losses.
The size of a generator depends to a very large extent on the required rated
torque. Consequently, the rated torque is one of the most important parts
of the specification. The rated torque differs for different wind turbines of
the same rated power, because of different turbine speeds. Data from 25
wind energy converters from Bindner et al. (1995) and Anon. (1994b) were
used to find the typical values of the rated torque for different sizes of wind
energy converters. A curve fit to the data was made to find an analytical
expression of the rated torque. The rated torque can be approximated as
P 1.23
TN = 71.1 Nm  N  (2.1)
 1 kW
The rated torque determines the rated speed, which can be expressed as
P –0.23
nN = 134 rpm  N  (2.2)
 1 kW
The empirical function (2.1) for the rated torque can be compared with the
one used by Veltman et al. (1994) and Søndergaard & Bindner (1995)
P 1.5
TN = k  N  (2.3)
 1 kW
where k is approximately 13 Nm for a 500 kW turbine. Equation (2.3)
predicts a faster increase in rated torque as the turbine power increases
than Equation (2.1) does. The difference depends on the assumption that k
is a constant. In order for k to be a constant, the tip speed of the turbine
and the rated power per swept area has to be independent of the rated
power. However, the tip speed increases slightly and the rated power per
swept area also increases as the turbine power increases. In Figure 2.1
the rated torques for the 25 wind energy converters are plotted in a log-log
diagram together with the torque according to Equations (2.1) and (2.3). It
is obvious that Equation (2.3) can only be used to predict the rated torque
for generators with a rated power similar to the rated power for which k
has been determined.
Since the winding temperature is a limiting factor for the rated current,
the maximum allowed winding temperature is also vital to generator
performance. The generators designed in this thesis are made for class F
winding insulation. According to Det Norske Veritas wind energy
converter standard (Anon. 1992, Section 8, p. 11) the allowed temperature
rise for class F insulation is 90˚C and the thermal calculations are to be
made for an ambient temperature of 40˚C. Consequently, the maximum
temperature for the winding is 130˚C.
There are often technical demands that vary between different wind
energy converters. In some wind energy converters, the generator has to
be used as a start motor for the turbine, usually for stall-controlled
turbines. The required peak torque differs between turbines with different
control principles. Stall-controlled turbines need a high peak torque to
limit the turbine speed during wind gusts, while pitch-controlled turbines
do not have to use the generator to limit the speed and, therefore, do not

10000
Rated torque (kNm)

1000

100

10

1
10 100 1000 10000
Rated mechanical power (kW)
Rated torques of 25 wind energy converters
Rated torque according to Equation (2.1)
Rated torque according to Equation (2.3)

Figure 2.1 The rated torque as a function of mechanical power.

need over-torque capacity. If the generator is used for electrical
emergency braking, a high peak torque is also needed. The generators in
this thesis are assumed to be used in a pitch-controlled wind energy
converter and are not required to produce a torque higher than the rated
one. The specification used in this thesis is summarized in Table 2.1.

Table 2.1 The specification used in this thesis for a direct-driven

generator of a rated power of PN.

P 1.23
Rated torque TN = 71.1 Nm  N 
 1 kW

PN  –0.23
Rated speed nN = 134 rpm  1 kW
 

Winding temperature θCuN ≤ 130˚C

Peak torque TMax ≥ TN

2.2 Generator Cost Function

The generator system for a wind energy converter can be chosen by
comparing the total cost of the different alternatives. The total cost
includes more than the cost of purchasing or manufacturing the
generator. It includes both direct costs and indirect costs. Some of the
costs are:
• material costs (direct);
• manufacturing costs (direct);
• cost of losses (indirect);
• cost of maintenance (indirect);
• cost of availability (indirect).
The material and manufacturing cost of the active part of the generator is
estimated from the weights of the active materials. The cost of the
supporting structure is determined from the diameter and length of the
generator structure and the cost of the losses is calculated from the
average losses of the system. Because the maintenance required for a
generator and a gear is very limited, the cost of maintenance is neglected.
Moreover, the cost of the availability is neglected since the availability is
assumed to be very close to 100 % for all generator systems.
Apart from the total cost of the generator system, a comparison can
include other aspects that cannot easily be economically evaluated. For
instance, the noise of the generator can be crucial to the acceptance of the
wind energy converter. Such aspects have not been included here.
2.2.1 Cost of Active Parts
The cost of the active parts of the generator is based on the assumption
that the cost, of both the material and the manufacturing, can be
expressed as a specific cost per weight of the different materials. The cost
of the active parts is, thus, expressed as

Cact = c Cu m Cu + c Fe m Fe + cm m m (2.4)
where m Cu , m Fe and m m are the weight of the copper, the active iron and
the permanent magnets, respectively. The used values of the specific costs
of the different materials cCu , c Fe and cm are given in Section 2.2.4.

2.2.2 Cost of Structure

The cost of the structure has not been analyzed thoroughly. Only an
approximate model is used. Without going into mechanical details, it is
clear that the amount of material used and the difficulty in
manufacturing the structure increase as the diameter and length
increase. Thus, the structural cost is a function of the stator outer
diameter dse and stator length including end windings ltot. In this thesis,
the cost of the structure is approximated as

1   dse  a  ltot  a 
 
2   dref +  lref  
Cstr = cstr (2.5)

where the constant c str is the cost of a reference structure with the
diameter dref and the length lref. The exponent a describes how fast the
cost increases with increasing diameter and length. The cost of a
structure of 2 m diameter and 1 m length is assumed to be 20 000 ECU
(i.e., dref = 2 m, lref = 1 m and cstr=20000). For small generators, the cost of
a structure of 1 m diameter and 0.5 m length is estimated to be 2500 ECU.
The exponent a is then 3.
Of course, this model is only approximate. The real cost function will be
much more complicated and include terms which depend on both
diameter and length as well as terms which are functions of other
variables than the outer dimensions. The real cost function will also be
discontinuous, for instance at the diameter above which the generator can
no longer be transported in one piece, but instead has to be mounted at the
wind energy converter site. Nevertheless, it will be shown in Section 6.3
that the exact shape of the cost function for the structure is not very
important for the optimization of the generator diameter and length.

2.2.3 Cost of Average Losses

The losses decrease the energy production of the wind energy converter
and reduce the income from the sold electric energy. Being proportional to
the reduction of the energy production, the cost of losses is proportional to
the average losses, not proportional to the losses at rated load. The cost of
losses can be expressed as average losses P dAv times the specific cost per
kilowatt of average losses cd, i.e.

Cd = PdAv cd (2.6)

The specific cost per kilowatt of average losses is the present value of all
the annual costs of one kilowatt losses during the economical lifetime of a
wind energy converter. The value of this specific cost of average losses
depends on several variables which are difficult to estimate, such as the
future price of electricity, the real interest rate, and the lifetime of the
wind energy converter. However, these problems are the same for any
type of long-term investment calculation. The specific cost of average
losses can be calculated as

c d = cel Ny kN (2.7)

where cel is the specific cost of electric energy (ECU/kWh), N y the number
of hours per year and kN is the factor for the present value of N WEC years
of losses. With the real interest rate i the factor for the present value is
NWEC
(1 + i) – 1
kN = NWEC (2.8)
i (1 + i)

In Table 2.2, examples of the specific cost of average losses are shown,
with various real-interest rates and electricity prices. The number of
years N WEC is assumed to be 20. Electricity produced in modern wind
turbines on good sites costs about 0.04 to 0.06 ECU/kWh. The real interest
rate is usually about 2-4 % for infrastructure investments, but private
companies often use higher rates. It can be seen that the variation in the
cost of losses is large, from 4000 to 8600 ECU. For the optimization, a
specific cost of average losses of 6000 ECU/kW is used.

Table 2.2 Examples of the specific cost of average losses cd for a wind
energy converter lifetime of 20 years.

Electricity price Real interest rate, i

c el 2% 4% 6%

0.04 ECU/kWh 5700 ECU/kW 4800 ECU/kW 4000 ECU/kW

0.06 ECU/kWh 8600 ECU/kW 7100 ECU/kW 6000 ECU/kW

2.2.4 Total Cost Function
The total cost function used in this thesis includes the cost of the active
parts, the cost of the structure and the cost of average losses, i.e.

Ctot = Cact + Cstr + Cd (2.9)

This cost function is intended for approximate optimization of the
generator and should not be used to estimate the manufacturing cost of a
generator. The parameters for the different parts of the cost function are
presented in Table 2.3.

Table 2.3 The cost function parameters and their nominal values.

Cost parameter Nominal value

Cost of copper c Cu 6 ECU/kg

Cost of iron c Fe 4 ECU/kg

Cost of NdFeB magnets cm 100 ECU/kg

Cost of reference structure c str 20 000 ECU

Reference diameter dref 2m

Reference length lref 1m

Structure exponent a 3

Cost of average losses cd 6000 ECU/kW

3 Calculation Method for the Average Losses
In this Chapter, a method to calculate the average losses and average
efficiency is derived. The average losses are calculated from the different
types of losses by multiplying each type of loss at rated load with an
average loss factor for that type of loss. The average loss factors need be
calculated only once for a generator type. Consequently, the average losses
are easy to include in the cost function for the optimization of the
generator. The calculations are made for a permanent-magnet generator,
but they can be made in a similar way also for other types of generators.

3.1 Average Losses

To find the average losses PdAv the probability density function w(v) is used
as a weighting function for the losses P d(v) at different wind speeds. The
average losses can be calculated as
v out
PdAv = ∫ Pd(v) w(v) dv (3.1)
vin
where v is the wind speed and v in and v out the cut-in and cut-out wind
speed of the turbine. The cut-in wind speed used in the calculation of the
average loss factors is 3 m/s and the cut-out wind speed is 24 m/s.
The probability density of wind speeds is approximated with high
accuracy by a Weibull distribution
c

c v c
w = v  
()v
– A

A 
e (3.2)

where c is a shape parameter which varies a little, but is close to 2 for

normal sites. The parameter A is determined by the average wind speed
of the site. By definition the integral of the wind speed probability density
function over wind speeds from zero to infinity is exactly one. Three
typical wind speed probability density distributions are shown in
Figure 3.1.
The average losses of the generator depend on which site the wind energy
converter is placed on. Therefore, the average losses must be calculated
for a site with wind conditions similar to the ones at the site on which the
wind energy converter will be used. In Section 3.3 the average losses are
calculated for three different sites.
To use Equation (3.1) to calculate the average losses, the losses must be
expressed as a function of wind speed. The main types of losses for a
permanent-magnet generator are: stator copper losses P Cu which are a
 0.14

Probability density (s/m)

Average wind speed 5.5 m/s
0.12 Average wind speed 6.8 m/s
0.1 Average wind speed 8.0 m/s
0.08
0.06
0.04
0.02 Cut-in Cut-out
0
5 10 15 20 25
Wind speed (m/s)

Figure 3.1 Weibull probability density distributions for sites with

different average wind speeds.

function of armature current; stator core hysteresis and eddy current

losses, P Hy and P Ft , which are functions of flux linkage and frequency;
and friction and windage losses P µ which are a function of the rotational
speed of the generator. The additional losses P ad are mainly a function of
the armature current. By defining the armature current, flux linkage,
frequency and rotational speed as functions of wind speed, the different
losses can be expressed as functions of wind speed. The total losses can
then be expressed as

Pd(v) = PCu(v) + PHy(v) + PFt(v) + Pµ(v) + Pad(v) (3.3)

One way to calculate the average losses is to use Pd(v) from Equation (3.3)
directly in (3.1). But when the average losses are to be included in the
optimization of a generator design, this method is inefficient. The average
losses must be calculated by an integral for each new set of generator
variables that is used in the optimization procedure. To simplify the
generator optimization, the average losses can instead be calculated from
the different types of losses at rated load using average loss factors:

PdAv = PCuN kdCu + PHyN kdHy + PFtN kdFt + PµN kdµ + PadN kdad (3.4)

where k dCu , k dHy, k dFt , k dµ and k dad are the average loss factors for the
different types of losses. PCuN, P HyN, P FtN, P µN and P adN are the losses at
rated load.
The average loss factors are independent of how high the losses are at
rated load. If the losses at different wind speeds are expressed as the
losses at rated load multiplied by functions representing the dependence
on the wind speed, the average loss factors can be derived from Equation
(3.1). The different losses are then expressed as

PCu(v) = PCuN gCu(v) (3.5)

PHy(v) = PHyN gHy(v) (3.6)
PFt(v) = PFtN gFt(v) (3.7)
Pµ(v) = PµN gµ(v) (3.8)
Pad(v) = PadN gad(v) (3.9)
By substituting the total losses in (3.1) according to Equations (3.3) and
(3.5)–(3.9), the average power losses can be calculated as a few integrals
representing the different types of average losses. Since the losses at rated
load are constant, irrespective of the wind speed, the average losses can be
expressed
v out
PdAv = PCuN ∫ w(v) gCu(v) dv +
vin
v out v out
+ PHyN ∫ w(v) gHy(v) dv + PFtN ∫ w(v) gFt(v) dv +
vin vin
v out v out
+ PµN ∫ w(v) gµ(v) dv + PadN ∫ w(v) gad(v) dv (3.10)
vin vin

By making a comparison with Equation (3.4) the average loss factors can
be identified. For instance, the average loss factor of the copper losses is
v out
kdCu = ∫ w(v) gCu(v) dv (3.11)
vin
It remains to define the functions gCu(v), gHy(v), gFt(v), gµ(v) and gad(v) in
order to calculate values of the average loss factors. These functions will
be different for different wind energy converters since they depend on how
the turbine and the generator are controlled. In Section 3.3 the average
loss factors are derived for one permanent-magnet generator type.

3.2 Average Efficiency and Average Power

The average efficiency of a generator expresses the percentage of the
mechanical input energy which is converted into electrical energy. The
average efficiency can be calculated from the average input power PAv and
the average losses
P
ηAv = 1 – PdAv (3.12)
Av
The average input power can be calculated in the same way as the
average losses

PAv = kt PN (3.13)
where PN is the rated mechanical power of the turbine and kt the average
factor for the turbine power. The average factor can be calculated by
expressing the turbine power as the power at rated load times a function
gt representing the wind speed dependence, i.e.,

Pt(v) = PN gt(v) (3.14)

The average factor for the turbine power is then

v out
kt = ∫ w(v) gt(v) dv (3.15)
vin
The value of the average factor for the turbine power is almost the same as
the value of the often used capacity factor for the wind energy converter.
The capacity factor expresses the relation between average power to the
grid divided by rated power to the grid.

3.3 Determining Average Loss Factors

In this section, the average loss factors will be calculated for the different
loss components. To calculate these factors, the g-functions, representing
the dependence of the different loss components on the wind speed, are
derived. Finally, the average loss factors are calculated and discussed.
The g-functions can be derived from a loss model for electrical machines.
An approximate model, for how the losses change with armature current
Ia, flux linkage Ψ , frequency f and speed n, is used. The model has been
verified for a wide range of generator speeds, voltages and currents by
measurements on a four-pole synchronous generator (Grauers, 1994,
p. 70-86). The g-function of the copper losses can be expressed as
Ia(v)2
gCu(v) = (3.16)
IaN2
where IaN is the armature current at rated load. Additional losses can be
approximated as proportional to the armature current squared, just like
the copper losses (Adkins & Harley, 1975). Therefore, there is no need for a
special g-function for the additional losses since they can be included in
the copper losses.
The g-function of the eddy current losses in the iron core can be expressed
approximately as
f Ψ 2
gFt(v) =   (3.17)
 fN ΨN
where fN and Ψ N are the frequency at rated load and the flux linkage at
rated load, respectively. The g-function of the hysteresis loss function is
approximated as

f Ψ 2
gHy(v) =   (3.18)
fN  Ψ 
N

The friction and windage losses are bearing friction losses, approximately
proportional to the rotational speed, and fan and windage losses,
approximately proportional to the cube of the rotational speed. The
g-function, therefore, can be expressed as
n n 3
gµ(v) = Cµ1 + Cµ2  n  (3.19)
nN N

where Cµ1 + Cµ2 = 1 and nN is the speed at rated load. The parameter Cµ1
represents friction that is proportional to the speed and C µ2 represents
losses that are proportional to the cube of the speed. Here it is assumed
that C µ 1 = C µ 2 = 0.5. For a 50 kVA, four-pole, electrically excited
synchronous generator C µ1 is 0.28 and C µ2 is 0.72 (Grauers, 1994, p. 74).
The lower the generator speed is, the larger Cµ1 is and the lower Cµ2 is.

The g-functions above have to be defined with the wind speed as a

parameter. Consequently, the generator speed, frequency, current and
flux linkage have to be expressed as functions of the wind speed. In a
variable speed wind energy converter, the speed of the turbine is usually
controlled to maximize turbine efficiency. The turbine speed is increased
linearly with the wind speed until the rated speed n N is reached. The
rated speed is reached at a wind speed vnN, which is lower than the rated
wind speed. (vnN = 10 m/s in this thesis.) Thus, the speed can be defined as

  v  n
 vnN N
if v ≤ vnN
n(v) =  (3.20)
 nN if v > vnN

The frequency varies in the same way as the turbine speed, i.e.,

  v  f
 vnN N
if v ≤ vnN
f(v) =  (3.21)
 fN if v > vnN

The core losses of the generator are determined by the flux linkage of the
generator armature. The rectifier is assumed to control the armature
voltage to keep the flux linkage constant

Ψ (v) = ΨN (3.22)
The current as a function of wind speed can be calculated from the
equation for the generator output power

Pa(v) = 3 Uap(v) I a(v) cos(ϕ) (3.23)

where U ap is the armature phase voltage and cos(ϕ ) the terminal power
factor. The terminal voltage is kept at the same level by the rectifier
control as the internal emf Ep and the emf is proportional to the frequency
f
Uap(v) = EpN   (3.24)
 fN
where E pN is the internal emf at rated speed. The reactance X a of the
generator (= 2 π f (v) La) makes the power factor current-dependent

√
0.5 Ia(v) 2 π f (v) La 2
cos(ϕ) = 1 –   (3.25)
 Uap(v) 
In calculating the current as a function of wind speed, generator
efficiency is approximated as being constant. The electric power is
consequently proportional to the turbine power

Pa(v) ~ Pt(v) (3.26)

The losses in the generator are not neglected, it is only assumed that the
armature power can be scaled in proportion to the turbine power. The
armature power, then, can be expressed as
P (v)
Pa(v) = t
Pt(vN) PaN (3.27)

where P aN is the armature power at rated load and v N is the rated wind
speed, here assumed to be 13 m/s. The power from the turbine Pt(v) is
1
Pt(v) = CP(v, n) 2 ρa At v3 (3.28)

where CP is the power coefficient of the turbine, ρa is the density of the air
and At is the area swept by the turbine. Equations (3.27) and (3.28) can be
used to express the armature power as
CP(v, n(v))  v  3
Pa(v) =
CP(vN, nN)  vN PaN (3.29)

From (3.23) and (3.25) the current as a function of electrical power and
voltage can be found

√
{U √
Uap(v)4 – 1/9 [Pa(v) 2 π f(v) La]2 }
2
2 ap(v) –
I a (v) = (3.30)
[ 2 π f(v) La ]2
The way in which armature current changes with wind speed depends on
the generator design, since the armature inductance L a is not constant.
Therefore, the average loss factor for the copper losses will vary during
the optimization of a generator. If the variation of the average loss factor
is large, the loss factor has to be calculated for each new set of generator
variables used in the optimization. It will later be shown that the variation
in the average loss factor is small enough to be neglected.
The g-functions for the different types of losses can be calculated by means
of the above functions for the current, rotational speed and flux linkage as
functions of wind speed. In Figure 3.2 the g-functions for different types of
losses are plotted. It can be seen that all types of losses decrease from their
rated values when the wind speed decreases. The copper losses decrease
rapidly as the wind speed decreases because the current is almost
proportional to the power. The hysteresis losses are reduced because of the
reduced speed but the reduction is much smaller than for the copper
losses, since the flux of the generator remains constant. The eddy current
losses are reduced more than the hysteresis losses because they decrease
as the square of the frequency. The reduction of the friction and windage
losses is similar to that of the eddy current losses.
The average loss factors are calculated for three different sites with the
wind speed probability density approximated as a Weibull distribution.
The first site is a high wind speed site, the second site is a typical wind
energy converter site, and the third site is a low wind speed site. The
average factors for the turbine power are 0.35, 0.25 and 0.15 on the three
sites. The average loss factors for the different sites and the different types

1.2 Rated speed

Rated power
1
g-functions

0.8
0.6
0.4
0.2
0
5 10 15 20 25
Wind speed (m/s)

Eddy current losses Hysteresis losses

Copper losses Friction losses

Figure 3.2 The g-functions for different types of losses from the cut-in to
the cut-out wind speed.
of losses are shown in Table 3.1. An armature reactance x a of 1 p.u. has
been used.
It is clear that copper losses cause rather low average losses. When the
generator is optimized for a wind energy converter, it will be more
important to keep the core losses low than to keep the copper losses low.
The average loss factors must be calculated for each new type of generator
and drive train because they can differ between different generator
designs and also the control of the generator affects the loss constants.
The average loss factor for the copper losses depends on the armature
reactance. In Figure 3.3 the value of the average loss factor for the copper

Table 3.1 The average loss factors for different types of losses at
different sites.

Average wind speed 5.5 m/s 6.8 m/s 8.0 m/s

Parameter A (c=2) 6.23 7.66 9.06

Average factor for the 0.15 0.25 0.35

turbine power, kt

kdCu 0.07 0.14 0.24

kdHy 0.50 0.61 0.69

kdFt 0.35 0.47 0.57

kdµ 0.38 0.50 0.60

0.15
for copper losses
Average factor

0.1

0.05

0
0.2 0.4 0.6 0.8 1 1.2 1.4
Armature reactance (p.u.)

Figure 3.3 The value of the average factor for the copper losses kdCu as a
function of generator reactance at the medium wind speed
site.
losses is shown as a function of armature reactance for the site with an
average wind speed of 6.8 m/s. The average loss factor decreases with
increasing reactance. For generators with high force density the
reactance is usually between 0.7 and 1.2 p.u. Consequently, it is
reasonable to use a reactance of value 1 p.u. when the average loss factor
for the copper losses is calculated.
In electrically excited generators, the flux linkage can be reduced at low
power to maximize the efficiency at each load, as described by Grauers
(1994, p. 90-91). If this control strategy is used, the average loss factor for
the copper losses will be higher than for this permanent-magnet
generator. The average loss factor for the core losses will instead be lower.
4 Generator Types
In this chapter a generator type suitable for direct-driven wind turbine
generators is chosen. First, some different generator types are discussed
briefly, and then the choice of rectifier is discussed more in detail. Finally,
the chosen generator type is presented.
It is very difficult to compare generator types completely; to find which is
the best for a given application. The aim of this chapter is solely to find a
generator type well suited to be a direct-driven wind turbine generator.

4.1 Electrical Excitation or Permanent Magnets

Synchronous generators can be either electrically excited or excited by
permanent magnets. The question of which type of excitation is best is
determined mainly by comparing the cost of the permanent magnets with
the total cost of the rotor pole, the field winding and the field winding
losses.
The cost and the losses of an electrically excited generator depend on the
pole pitch. A simplified way of showing this is to look into the required
field current at no-load. The mmf required of the field pole is determined
by the required air gap flux density and the magnetic air gap. The mmf
required for the iron has been neglected. In the electrically excited
generator, the magnetic air gap is the distance between the pole shoe and
the stator teeth. As the pole pitch is reduced, the no-load mmf required of
each pole is constant. Consequently, the total field mmf has to be constant,
although there is less room for the field winding as the pole pitch
decreases. To allow constant no-load field mmf, the field pole will have to
be higher as the pole pitch is decreased. In Figure 4.1, generator poles
with electrical excitation are shown for three different pole pitches. The
air gap flux density is 0.7 T and the air gap is 2 mm, which requires a no-
load field mmf of 1100 ampere-turns. The the no-load field winding
current density is assumed to be 2 A/mm2 and the fill factor of the field
winding is 0.5. In the figure it can be seen that it becomes difficult to
decrease the pole pitch below 100 mm because there is not much room left
for the field winding. Since the number of poles increases with decreasing
pole pitch, the field winding losses will increase as the pole pitch
decreases.
Permanent magnets are expensive but they eliminate the excitation losses
and allow smaller pole pitches to be used than electrical excitation does.
The pole pitch of a generator with permanent magnets can be very small.
It is only limited by the leakage flux between the magnets. Just as for the
electrically excited generator, the no-load mmf required of the magnet
 Stator yoke

Pole shoe

1100 A

1100 A

1100 A

1100 A
1100 A Pole body 1100 A

Rotor yoke

120 mm 80 mm 50 mm

Figure 4.1 Electrical excitation for three pole pitches. No-load peak flux
density in the air gap 0.7 T, air gap 2 mm.

does not depend on the pole pitch. The mmf produced by a magnet is the
magnet height times the coercitivity of the permanent-magnet material.
Therefore, the magnet height can be constant as the pole pitch decreases.
In Figure 4.2 permanent magnet excitation is shown for three different
pole pitches. In comparison with Figure 4.1 it is clear that permanent
magnets are a better alternative than electrical excitation if the pole pitch
has to be small.
This comparison is made in a simplified way, not including the armature
reaction of stator currents. Still, it shows the main advantage of
permanent magnets over electrical excitation when using a small pole
pitch. For electrically excited generators the magnetic air gap is small

Stator yoke

Permanent magnet
Rotor yoke

120 mm 80 mm 50 mm

Figure 4.2 Permanent magnet excitation for three pole pitches, 120, 80
and 50 mm. No-load peak flux density in the air gap 0.7 T, air
gap 2 mm.
and, as a consequence, the armature reaction will be important if the pole
pitch is large. For rotors with surface-mounted permanent magnets, the
magnetic air gap is much larger since the permeability of the permanent
magnets is almost equal to that of air. Consequently, the armature
reaction is much smaller in a permanent magnet generator with surface
magnets than in electrically excited generators.
Even though the permanent magnets are very expensive, the losses of the
field winding make permanent-magnet excitation better than electrical
excitation for small pole pitches. Jöckel (1996) has shown that even
expensive NdFeB magnets (≈150 ECU/kg) lead to a lower total cost than
electrical magnetization does. Besides reducing losses, the permanent
magnets lead to a lighter design.
Induction generators are electrically excited, but in contrast to the
electrically excited synchronous generator the magnetizing current flows
in the stator winding. Although the design differs from the design of
electrically excited synchronous generators, induction generators also
suffer from the same negative effects as the pole pitch is reduced. The
magnetizing mmf is constant, but as the pole pitch is reduced a larger
part of the stator current will be needed to magnetize the air gap. This
effect causes the power factor to decrease as the pole pitch decreases. The
practical limit for the minimum pole pitch of induction machines with an
air gap of 2 mm is in the order of 100 mm. An other reason why induction
generators cannot be made with a small pole pitch even if the air gap can
be made small is that the stator winding should be made with at least two
slots per pole and phase to keep the space harmonics of the air gap flux
wave low. Two slots per pole and phase requires at least a 100 mm pole
pitch in a three phase generator.

4.2 Direct Grid Connection or Frequency Converter

Grid connected generators are required to have a frequency of 50 or 60 Hz
and to damp oscillations between the grid and the rotor. Direct-driven
wind turbine generators can only generate 50 Hz with a very large
number of poles, for example 188 poles for a 500 kW generator with a rated
speed of 32 rpm.
With a required pole pitch of more than 100 mm, diameter for electrically
excited generators will be very large, more than 6 m for the 500 kW
generator. Therefore, electrically excited, direct-driven wind turbine
generators should not be designed for a 50 Hz frequency and,
consequently, should not be direct grid connected.
The pole pitch can be made 50 mm, or even less, by using permanent
magnets. The generator can then provide a 50 Hz frequency with a
reasonable diameter, 3 m for a 500 kW generator with a rated speed of
32 rpm. Nevertheless, damper windings cannot be made efficient with
such a small pole pitch and, therefore, the permanent magnet generator
cannot be direct grid connected. There is one solution to this problem;
mechanical damping of a moving stator, discussed by Westlake et al.
(1996). That solution requires a complicated mechanical structure and is
not further analysed in this thesis.
It is clear from the discussion above that it would be very difficult to design
a direct-driven wind turbine generator that is to be direct connected to the
grid if the diameter must remain small. Connecting a frequency
converter between the generator and the grid will solve these problems. To
use a frequency converter is, however, not only a way to avoid problems: A
frequency converter also improves the wind energy converter
performance. It makes it possible to operate the wind turbine at variable
speed, which increases the energy production and reduces the noise at
low wind speeds. The frequency converter can also reduce mechanical
loads and it allows the generator to be optimized with less restrictions.

4.3 Surface Magnets or Flux Concentration

In permanent magnet generators the magnetization can either be
achieved by magnets directly on the rotor surface or by magnets inside the
rotor. One rotor design with surface-mounted magnets and one rotor
design with flux concentration are shown in Figure 4.3.
Magnets on the rotor surface have to have a remanent flux density higher
than the required air gap flux density. Subsequently, it is necessary to use
expensive magnets, like Samarium-Cobalt (SmCo) or Neodymium-Iron-
Boron (NdFeB). SmCo has a remanent flux density of about 1 T and NdFeB
about 1.2 T. The magnet material is utilized best when the flux density in
the air gap is half the remanent flux density. If the air gap flux density
has to be close to the remanent flux density, the amount of permanent

Permanent magnets,
magnetization direction

Figure 4.3 Surface mounted NdFeB magnets (left) and ferrite magnets
with flux concentration (right).
magnets required will be large. Nevertheless, surface magnets lead to a
very simple rotor design with a low weight.
Flux concentration can be used to utilize cheap low-energy magnets and
still obtain a high air gap flux density. The magnets are then placed
inside the rotor and the flux is guided in magnetic circuits which are
narrower at the air gap than at the magnets. A common low-energy
magnet material is ferrite which has a remanent flux density of about
0.4 T.
A more complicated rotor is required for flux concentration than for
surface magnets and it would also normally be heavier, while the cost for
magnets can be much lower than for surface magnets. Today, NdFeB
magnets cost about 30 times more than ferrite magnets and their
maximum magnetic energy product is about 10 times higher than it is for
ferrite.

4.4 Slot Winding or Air Gap Winding

Normally, the armature windings in electrical machines are placed in
slots in the iron core, but in certain machines air gap windings have been
used. A conventional slot winding and an air gap winding are shown in
Figure 4.4.
The air gap winding is interesting for several reasons: more copper can be
placed in the space between the stator yoke and the rotor; cogging torque
caused by the teeth is avoided; and the reactance of the stator winding is
reduced. Nevertheless, the use of air gap windings also results in some
disadvantages: the magnetic air gap is large, requiring a large field
current or much permanent magnet material to be used; the forces act on
the stator winding instead of on the stator iron; the cooling surface
between the windings and the stator core is smaller if an air gap winding
is used; and the conductors of the air gap winding will be penetrated by
the air gap flux, which will cause eddy current losses in the windings.
The last disadvantage is not important if the windings are made of thin

Stator yoke Stator yoke

Rotor yoke Rotor yoke

Figure 4.4 A conventional slot winding (left) and an air gap winding
(right).
wire. In windings made of copper bars with a large cross section, the eddy
current losses caused by the main flux may singnificantly increase the
winding temperature.
Several authors propose air gap windings instead of conventional slot
windings. Air gap windings have been proposed for large turbogenerators
by Davies (1971), for several small machines and also for direct-driven,
wind turbine generators by Honorati et al. (1991) and by Alatalo (1991).
Except for turbogenerators, which have a very large pole pitch, air gap
windings are used in combination with permanent magnet excitation.
Turbogenerators have a large air gap, in the order of 100 mm, even with a
conventional slot winding. Consequently, the air gap winding does not
necessarily increase the necessary magnetizing mmf. The windings are
also usually directly water-cooled. Still, there can be problems with air
gap windings in turbogenerators: The eddy current losses in the winding,
for example, have not been discussed by Davies (1971); since
turbogenerators generally have thick copper bars, these losses can be very
high.
In small electrical machines, the air gap winding can be very useful
because it simplifies the manufacturing of the machine. In these small
machines the windings can be easily glued to the stator, since the forces
are very small. The cooling of the winding is also effective because of thin
windings and the windings are made of thin wire, leading to low eddy
current losses.
In direct-driven wind-turbine generators, the air gap winding seems to
have some disadvantages. The winding is normally indirectly cooled in
generators of this size and, therefore, the decreased cooling surface
between the air gap winding and the stator yoke is a disadvantage. The air
gap is normally only a few millimetres in a slotted machine but in a
machine with air gap windings, the air gap will become several times
larger, leading to a large amount of permanent magnets. The eddy
current losses in the winding can also be a problem, but one that can be
avoided if the winding can be made of stranded wire.

4.5 Radial-, Axial- and Transversal-flux Machines

There are several generator types that are possible to use as direct-driven
wind turbine generators. In this section, the radial-flux, axial-flux and
transversal-flux generators are discussed. Since the radial-flux machine
is the most conventional of the alternatives to be compared, it is used as a
reference in the comparisons. The design of the different machine types
can be seen in Figures 1.4–1.8 in Section 1.3.
Axial-flux machines can, in many respects, be designed in a way similar
to radial-flux machines. One important restriction for axial-flux
machines is that the amount of windings in the air gap is limited by the
available space at the inner radius. The air gap at larger radius cannot be
fully utilized because of this and the utilization of the iron core and
magnets is slightly less efficient in axial-flux machines than in radial-
flux machines. In radial-flux machines, the length of the stator and the
air gap diameter can be chosen independently. If necessary, the radial-
flux machine can be made with a small diameter by using a long stator.
To reduce the diameter of the axial-flux machine, while keeping the rated
torque constant, the difference between inner and outer radius has to be
increased. The maximum torque of an axial-flux machine is, however,
achieved when the inner radius is about 0.6 times the outer radius
(di Napoli 1991). A smaller inner radius will only decrease the rated
torque. Consequently, the diameter of the axial-flux machine cannot be
reduced as much as that of the radial-flux machine. One way of avoiding
a large diameter is to stack a number of axial-flux machines with a small
diameter on the same shaft. Thus, the rated power can be increased
without increasing the diameter. This will, however, lead to an expensive
generator.
To allow a small air gap, the rotor and stator structures have to withstand
the high magnetic force in the air gap. It is easier to make a rotor stiff in
the radial direction than in the axial direction, especially in generators
with a large diameter. Therefore, it is easier to make radial-flux
generators with a small air gap. However, the thermal expansion of the
rotor and stator will in a radial-flux generator influence the air gap, while
in an axial-flux machine it does not affect the air gap. Axial-flux
machines are also difficult to manufacture with a slotted stator, because
the slot pitch varies on stator laminations for different radii.
The axial-flux machine can be made with a double-sided stator more
easily than a radial-flux machine. A double-sided stator is shown in
Figure 1.7. This stator eliminates the need for a rotor yoke as a return
path for the flux. Subsequently, the active weight of the generator can be
reduced. Nevertheless, it is only a rotor yoke made of cheap solid iron that
is eliminated. A more complex non-magnetic rotor structure has to be
used instead to hold the magnets. The double-sided stator also allows the
winding to be divided into two, half as thick parts. In a radial-flux
machine an equivalent electromagnetic design can be achieved by
doubling the stator length instead of using two stator halves. Such a
solution will lead to a lower amount of end windings than the double-sided
stator. If the machine length is not restricted, the axial-flux machine with
a double-sided stator will not be better than a radial-flux machine with a
long stator, from an electromagnetic point of wiev.
A special type of axial-flux generator is the toroidal stator machine
(Shown in Figure 1.6). Besides the above mentioned advantages and
disadvantages of axial-flux machines, the toroidal stator winding leads to
simple end windings, but it becomes more difficult to fix the stator to the
generator structure. One further disadvantage is that the windings,
which have high losses in direct-driven wind turbine generators, are in
the middle part of the machine where they are difficult to cool without
direct air- or water-cooling.
The transversal-flux machine is rather different from the other machine
types, and it is difficult to make any simple comparisons between it and
radial-flux machines. The major difference between radial- or axial-flux
machines and the transversal-flux machine is that the transversal-flux
concept allows an increase in the space for the windings without
decreasing the available space for the main flux; this allows for very low
copper losses. The transversal-flux machine can also be made with a very
small pole pitch compared with the other types. These differences make
the transversal-flux machine capable of producing a higher force density
in the air gap than the other machine types. Unfortunately, the
electromagnetic structure is more complicated than for conventional
generator types, which may make it more expensive to manufacture. The
transversal-flux generator is probably better than the radial-flux
machines from an electromagnetic point of view, but a comparison of
these generator types must include a detailed mechanical investigation
and has, therefore, not been included in this thesis.

4.6 Forced-commutated Rectifier or Diode Rectifier

To make a direct-driven generator small, it is important that the force
density in the air gap is high. A high force density requires a high current
loading, which leads to a high armature reactance. Because of the
reactance, the type of rectifier has a large influence on the phase angle of
the armature current and on the achievable force density.
There are two major types of rectifiers that can be used for variable-speed
generators: machine-commutated rectifiers and forced-commutated
rectifiers. In this section, a machine-commutated diode rectifier is
compared with a forced-commutated sine-wave rectifier. The diode
rectifier is simpler, cheaper and more efficient than a forced-commutated
rectifier. The diode rectifier, however, cannot control the current phase
angle and, if the reactance of the generator is high, the generator will not
be well utilized. The forced-commutated rectifier can supply the generator
with reactive power and, therefore, the phase angle between the current
and the internal emf can be kept small, allowing a high force density,
even if the reactance is high.
A comparison of rectifiers will be made regarding the power at rated
current and peak power. In some wind energy converters, the generator
must be able to produce a peak power higher than the rated power, often
about 150 % of the rated power. The peak power is only required for short
times, in the order of some seconds. During that short time, the heating of
the winding is not assumed to be a problem and, therefore, the current
can be allowed to be much higher than the rated current.

4.6.1 Generator Model

The generator is assumed to be a three-phase permanent-magnet
generator with sinusoidal no-load voltages. Its equivalent circuit is shown
in Figure 4.5. In a permanent-magnet generator of the type discussed in
Chapters 5–7, the synchronous-, transient- and sub-transient reactances
are almost equal. In this thesis, the value of the synchronous reactance Xa
is used in all calculations. The per unit base is the no-load emf of the
generator and the rated current. The per unit resistance of the armature
winding will be a few percent, and has not been included in this
comparison of rectifiers.

Xa Ia Id
∼ —
E Ua
∼
—
∼
Ud
∼
∼
∼
Generator model Rectifier DC-link

Figure 4.5 The equivalent circuit of the generator used for a comparison
of rectifiers.

4.6.2 Diode Rectifier

The commutation voltage drop makes the output voltage U d of a diode
rectifier depend on the dc current Id. The currents in two phases overlap
during the time it takes to commutate the current from one phase to
another. The overlap can be measured as an electrical angle and, when
the overlap angle is less than 60 degrees, the voltage can be calculated as
3 3
Ud =
π √ 2 E – π Xa Id (4.1)

where E is the internal line-to-line emf of the generator and X a is the

armature reactance (Thorborg 1988, p. 117). If the overlap angle exceeds
60 degrees, the diode bridge will be completely short-circuited during a
part of the period. The dc voltage starts to decrease more rapidly with
increasing current and the dc voltage eventually becomes zero. The output
power as a function of the current can be calculated from the output
voltage. Since the voltage always decreases with increasing current, the
output power will have a maximum value which cannot be exceeded, no
matter how high the current is. The output dc voltage and the active
power of a diode rectifier fed by a generator, with a reactance of 0.5 p.u.
and zero resistance, are shown in Figure 4.6. The voltage and active
power as functions of current will always have the same shape,
regardless of the value of the armature reactance. It is only the scales of
the current axis and the power axis which will change if the reactance is
changed.

1
Active power (p.u.)
Dc voltage (p.u.)

0.8
0.6
0.4
Dc voltage
0.2 Active power
0
0 0.5 1 1.5 2 2.5 3
Dc current (p.u.)

Figure 4.6 The dc voltage and active power of a diode rectifier fed by a
generator with an internal emf of 1 p.u. and a reactance of
0.5 p.u.

4.6.3 Forced-commutated Rectifier

If a forced-commutated rectifier is used, the terminal voltage of the
generator can be controlled by supplying the generator with reactive
power. The rated armature current IaN and rated phase voltage U apN are
usually the limit for the generator power, and depending on how the
phase angle of the armature current is controlled, different values of the
active power can be achieved at rated current. To maximize the active
power at rated current, the product of the voltage, current and power
factor has to be maximized.
Three different ways to control the generator terminal voltage, and the
corresponding kVA ratings required of the generator armature and the
rectifier, are shown in Figure 4.7. Maximizing the terminal power factor
(control a) does not maximize the power, because the terminal voltage
must be lower than the internal emf. Since the generator armature and
the rectifier both has to be rated for the no-load emf, which is higher than
the terminal voltage at rated load, the required kVA ratings are high.
Instead, the terminal voltage can be kept at the same level as the no-load
emf (control b), by supplying reactive power from the rectifier. This control
method maximizes the active power from a generator and rectifier with
equal kVA rating and will be used in the following discussion.
If the generator is designed with an emf lower than the rated voltage, the
power can be increased by keeping the phase angle between the internal
emf and the armature current zero (control c). The advantage of this
generator design is that less permanent magnet material is used than if
control b is used and core losses at no-load are lower. The stator flux is
kept high only at high load by supplying reactive power to the armature
winding. One drawback is that the rated active power of the generator is
not maximized. The same generator would be capable of producing
slightly higher power, if the internal emf were increased to the rated
voltage and rectifier control b were used. Control b does not only allow a
lower generator armature kVA rating than control c, it also requires a
lower kVA rating of the frequency converter.

a) b) c)

Ep = UapN Ep = UapN Ep < UapN

Ia = IaN Ia = IaN Ia = IaN
Uap < UapN Uap = UapN Uap = UapN

Ep Ep
Ep
jXa Ia jXa Ia Ia jX a Ia

Ia

Ia Uap Uap Uap

Rating (kVA) if xa = 0.8 p.u. a) b) c)

Frequency converter 1.28 PN 1.09 PN 1.28 PN
Generator armature 1.28 PN 1.09 PN 1.28 PN

Figure 4.7 Phasor diagrams of a generator loaded by a forced-

commutated rectifier at rated current and the neccesary
frequency converter and generator armature kVA ratings. Ep
is the induced no-load phase voltage, Uap the armature phase
voltage and Ia the armature current.
a) cos(ϕ)=1 at the terminals and Ep equal to the rated voltage.
b) Uap and Ep equal to the rated voltage.
c) Uap equal to and Ep lower than the rated voltage.
The active power as a function of current, for a generator with a reactance
of 0.5 p.u. connected to a forced-commutated rectifier using control b, is
plotted in Figure 4.8. The output power increases linearly with increasing
current as the current starts to increase from zero. The reason for this is
that the forced-commutated rectifier can keep the terminal voltage
constant and that the power factor is almost 1. As the current increases
further, the power factor decreases because the rectifier has to supply
reactive power. Therefore, the increase in active power is no longer linear.
Eventually, the active power reaches a maximum value, at a generator
load angle of 90 degrees. If the current increases further, it will only lead
to a decrease in active power.

2
Active power (p.u.)

1.5

0.5

0
0 1 2 3 4
Armature current (p.u.)

Figure 4.8 The active power of a generator with a reactance of 0.5 p.u.
connected to a forced-commutated rectifier. The rectifier keeps
the generator terminal voltage constant.

4.6.4 Rectifier Comparison

The diode rectifier and the forced-commutated sine-wave rectifier were
compared regarding the output power at rated armature current. The
active powers of a generator connected to the two types of rectifiers are
shown in Figure 4.9, as functions of armature current. The value of the
generator reactance does not change the shape of the curves, only the
scaling of the current and power axes. As can be seen in the figure, the
difference in rated power between a diode rectifier and a forced-
commutated rectifier depends on how high the rated current of the
generator is.
If the rated current is I 1 , which corresponds to a reactance of 0.15 p.u.,
the rated power of the forced-commutated rectifier is 12 % higher than if a
diode rectifier is used. A sub-transient reactance of 0.15 p.u. is usual for a
conventional, four-pole, synchronous generator, but it is a low value for a
direct-driven wind turbine generator.
 Active power
Forced-comm. rectifier
Diode rectifier

II1
1 II2
2 I3 I4 II5
5
I3I4
Armature current

Figure 4.9 The active powers of a generator, connected to a diode rectifier

or a forced-commutated rectifier, as functions of armature
current.

In some wind energy converters, the generator has to be able to produce a

peak power of 150 % of the rated power. In this case, the rated current
cannot be higher than I2 if a diode rectifier is used. The rated current I2
corresponds to a reactance of 0.31 p.u. At the same rated current, the
generator can produce 22 % higher power if it is instead connected to a
forced-commutated rectifier. The copper losses are almost the same as if a
diode rectifier is used and, therefore, the forced-commutated rectifier
increases generator efficiency.
If the rated current is I 3 , corresponding to a reactance of 0.66 p.u., the
generator reaches its peak power if it is connected to a diode rectifier. If
connected to a forced-commutated rectifier instead of the diode rectifier,
the same generator can be capable of producing 65 % higher power with
almost the same copper losses.
Connected to a forced-commutated rectifier, the generator can have a
rated current of I4 and still be capable of producing a peak power of 150 %.
The rated current I4 corresponds to a per unit reactance of 0.71 p.u. If a
diode rectifer is used, the maximum reactance for a generator with 150 %
peak power is only 0.31 p.u. Consequently, a forced-commutated rectifier
allows more than twice as high reactance as a diode rectifier does for a
generator system required to produce a peak power of 150 % of the rated
power.
If cooling and efficiency considerations do not limit the current and the
peak power is not required to be higher than the rated power, the
generator is utilized best if the rated current is I 5 and a forced-
commutated rectifier is used. In this case, the reactance is 1.41 p.u. and
the generator can produce a power which is 165 % higher than
(2.65 times) its peak power connected to a diode rectifier. The copper losses
of the generator, however, are 360 % higher (4.6 times) at the current I5
than they are at the current I3, which means much lower efficiency even
though the rated power is 165 % higher. This power is the absolute peak
power the generator can produce if the armature voltage is limited to the
no-load voltage.
Peak power and power at rated current of the generator are presented as
functions of the per unit reactance in Figure 4.10. It can be seen that the
peak power of the generator connected to a forced-commutated rectifier is
always 2.65 times the value obtained for a generator connected to a diode
rectifier. The difference in power at rated current, if a forced-commutated
rectifier is used instead of a diode rectifier, is less than 20 % for reactances
below 0.3 p.u., but it increases quickly as the reactance increases. The
diode rectifier should probably not be used for a generator with a reactance
higher than 0.5 p.u., while generator reactance of up to 1.4 p.u. can be
allowed if a forced-commutated rectifier is used.
The choice of rectifier depends on the total cost for the generator plus
rectifier and the total average efficiency. The diode rectifier is cheap and
very efficient, but it may require a much larger generator and it also
decreases the efficiency of the generator. If it is important that the
generator is small, a high reactance of the generator is difficult to avoid.
Of the two compared rectifier types, the forced-commutated rectifier is
probably the best for a direct-driven wind turbine generator.

1.4 Forced-comm. rectifier:

Peak power
1.2
Power at rated current
Active power (p.u.)

0.8

0.6

0.4 Diode rectifier:

Peak power
0.2
Power at rated current
0
0.5 1 1.5 2
Reactance (p.u.)

Figure 4.10 Peak power and power at rated current as a function of

armature reactance of the generator.
4.7 Chosen Generator Type
In this section, the generator type chosen and the reason for using it are
presented. The comparisons made earlier in this chapter indicate that the
radial-flux permanent-magnet generator has several advantages and that
it is well suited to be used as a direct-driven wind turbine generator.

4.7.1 Basic Generator Concept

Permanent-magnet excitation is chosen because it gives an efficient
generator in which the pole pitch can be made small, leading to a light
core and low end-winding losses. The magnets are of NdFeB material and
are placed on the rotor surface. This placement gives a simple and light
rotor design, however these magnets are more expensive to use than
ferrite magnets.
Radial-flux design is used because it allows a simple generator structure,
good utilization of the active materials and it can easily be made with a
slotted stator. Radial-flux design also allows a small diameter since the
stator can be long. The air gap can more easily be made small in a radial-
flux machine than in an axial-flux machine, which leads to a low amount
of permanent-magnet material.
A stator with a slot winding is used because the cost of the permanent
magnets would otherwise be very high and a slot winding makes it easier
to cool the winding indirectly, through the stator core.
A frequency converter is used since the generator diameter otherwise has
to be very large and also because it allows variable turbine speed and lower
grid interference. The rectifier is a forced-commutated sine-wave rectifier
to allow for high inductance and a small generator.

4.7.2 Details of the Chosen Generator

The winding has three phases and only one slot per pole and phase to
allow for a small pole pitch, without having slots which are very narrow.
Narrow slots lead to a low copper fill factor. A two-layer winding is used to
make the end windings simple, and since it is an integer slot winding, all
slots will contain two coils belonging to the same phase. The slots are
constructed semiclosed to limit losses in the rotor due to slot harmonics
and to limit the cogging torque.
The windings and magnets must be cooled efficiently. Almost all losses
are dissipated in the stator, since the only rotor losses are the losses in the
magnets. The generator is to be totally enclosed (class IP54 described in
IEC Publication 529). Ambient air is not allowed to enter the generator, in
order to reduce the risk of water condensation on the windings and salt
and dust in the generator.
Forced cooling directly on the outer surface of the stator core will be used.
The stator yoke is fixed to a circular beam which guides the cooling air.
Air is forced round the circumference of the stator by means of external
fans. A principal drawing of the cooling system is shown in Figure 4.11.
To achieve a low temperature rise between the stator yoke and the cooling
air, the air velocity is 15 m/s and extra cooling fins are made at the outer
surface of the stator core. The circular stator beam is divided into two or
more separate cooling circuits, in order to make the cooling more
efficient. Internally, the rotor and the end windings are cooled by the
circulation of the inner air. The rotor is equipped with a radial fan which
forces the air to pass the end windings and then out to be cooled at the end
shields.

Cooling duct Radial fan

Stator core

Figure 4.11 Left: Axial view of the cooling of the stator iron core. An
external fan forces air through the cooling duct at the stator
core back.

Right: Tangential view of the cooling of the rotor and end

windings. The rotor is equipped with a radial fan circulating
the internal air.

4.7.3 Materials
The winding is made of stranded wire that has a copper fill factor of 0.8,
excluding the coil insulation. Each coil has an insulation that is 0.5 mm
thick. The slot is made 1 mm wider than the insulated coil. That extra
space is filled with resin during the impregnation, leading to a total
insulation thickness of 1 mm. The value of the total fill factor of the slot,
excluding the slot wedge, is 0.58 for a 40 mm high and 10 mm wide slot
with a two-layer winding.
The magnet material is NdFeB with a remanent magnet flux density of
1.22 T at room temperature. The remanent flux density at operating
temperature, i.e. below 120˚C, will be 1.1 T or higher. NdFeB is chosen
instead of SmCo because of its lower price and higher remanent flux
density.
The stator core is made of 0.5 mm thick, low-loss, electrical steel for large
generators. The losses at 50 Hz and 1.0 T are 1.20 W/kg and the fill factor
of the stator core is 0.97.
5 Design Method for a Permanent-magnet
Generator
In this chapter, an analytical design method is derived for the proposed
three-phase radial-flux generator with a forced-commutated rectifier. The
design variables are discussed in Section 5.1 and the design equations and
the thermal model in Section 5.2. In Section 5.3, the calculation method is
described. Finally, the analytical design method is compared with finite
element calculations and the sensitivity of the thermal model is discussed
in Section 5.4.
Many of the calculations are based on simplified models. The models have
been estimated to be sufficient for a preliminary generator design. If the
design method should be used for other purposes, needing higher
accuracy, the models can be changed to more detailed ones.
Some steps in the complete design of a generator have not been included.
Parts which can be considered as final adjustments and checks, are left
for the detailed design that has to be carried out before a generator is
manufactured. The steps remaining for the final design are: adjusting
the rated voltage to the desired level; rounding off the number of pole pairs
to the nearest integer value; and checking that the permanent magnets do
not risk irreversible demagnetization. The demagnetization calculations
can be included in the design method, but it has been found during this
investigation that the demagnetization is not a problem in this type of
generator, as long as the pole pitch is kept small.

5.1 Design Variables

It is important that the variables chosen for the design calculations are
independent. By starting with the geometrical variables and the current
density in the windings as basic variables, it is easy to assure that they are
independent. The proposed generator is described by 16 basic variables,
the 15 in Figure 5.1 and the stator length. The generator is completely
defined by these 16 basic variables and the material data. All the basic
variables have to have values before the generator data can be calculated.
Nine of the basic variables are assumed to be constant or are defined as
functions of other variables. The stator and rotor yoke heights are
determined by the allowed flux densities in the yokes. The tooth tip height,
the slot wedge height, the slot opening and the coil insulation thickness
are constant. The air gap is kept at its minimum value, limited by
mechanical considerations to 0.1 % of the air gap diameter. The slot pitch
is determined by the required number of slots per pole and phase and the
magnet width is kept as a constant fraction of the pole pitch.
 τ bd
hys
hi

hs

hs2 Js

bs1 δ
hs1 hm

hyr

bm
d
τp

Figure 5.1 Basic variables of the proposed generator.

Five of the basic variables are used directly as design variables: the air
gap diameter, the stator length, the slot height, the pole pitch and the
current density. The two remaining variables, the magnet height and the
tooth width, are defined as functions of the peak, no-load, air gap flux
density and the peak, no-load, teeth flux density, respectively. The reason
for using flux densities as design variables rather than geometrical
variables is that the optimum flux densities can be assumed to vary rather
little as the generator size changes.
The variables are summarized in Table 5.1. Seven variables are used as
design variables. These variables can be used to calculate generator
designs of different rated power and with different temperatures of the
windings. If a generator with a specified rated power and specified
winding temperature is designed, two of the design variables can no
longer be chosen freely. The stator length is adjusted to get the right rated
power of the generator and the current density is adjusted to get the right
winding temperature. The other five design variables can still be varied
freely.
The rated voltage and current are not included in the preliminary design.
It is assumed that the top and bottom conductor in the slots are connected
parallelly. Thus, the phase current is equal to the total current in a slot.
By adjusting the number of slots and the number of branches connected
parallelly and in series, the voltage level can be adjusted.
The pole pitch is used as a variable instead of the number of pole pairs.
Since there are no restrictions on the pole pitch, the number of pole pairs
is usually not an integer in the design calculations. For a generator
having a large number of pole pairs, this will simplify the numerical
optimization of the generator without introducing any significant errors.

Table 5.1 The variables used in the design method.

Design variables Corresponding basic

variables

d Air gap diameter d Air gap diameter

l Stator length l Stator length

hs Slot height hs Slot height

τp Pole pitch τp Pole pitch

Js Current density Js Current density

^
Bδ 0 Peak air gap flux density hm Magnet height
^
Bd0 Peak teeth flux density bd Tooth width

Constants and fixed relations Corresponding basic

variables
^
Bys = 1.2 T Peak stator yoke flux density hys Stator yoke height
^
Byr = 1.2 T Peak rotor yoke flux density h yr Rotor yoke height

hs1 = 1 mm Tooth tip height hs1 Tooth tip height

hs2 = 4 mm Slot wedge height hs2 Slot wedge height

bs1 = 3 mm Slot opening bs1 Slot opening

hi = 1 mm Insulation thickness hi Insulation thickness

δ = 0.001 d Mechanical air gap δ Mechanical air gap

q=1 No. of slots per pole & phase τ Slot pitch

bm = 0.7 τp Magnet width bm Magnet width

5.2 Design Equations

5.2.1 General Definitions

The diameter and pole pitch determine the number of pole pairs
π d
p= (5.1)
2 τp
The number of slots per pole and phase q is set to one to allow for a small
pole pitch without getting a low slot fill factor because of narrow slots. The
number of phases m is three. The total number of slots of the stator is

Q = 2p m q (5.2)
The slot pitch is
τp
τ = m q (5.3)

The slot and the two-layer winding are shown in Figure 5.2. The slot is
described by its depth hs and its width bs. The slot width can be calculated
from the slot pitch and tooth width bd as

bs = τ – bd (5.4)
The slot opening bs1 is assumed to be 3 mm, the tooth tip height hs1 1 mm,
and the slot wedge height hs2 4 mm. The winding height is

hs3 = hs – hs1 – hs2 (5.5)

The conductor height h Cu and width b Cu are determined by the winding

bs
bCu
hi hi
hi
Stator core

h Cu Winding
hs3
hs Insulation
hs2 h Cu hi
2 hi
Slot wedge

hs1 bs1

Figure 5.2 The slot and the two-layer winding.

height, slot width and the coil insulation thickness hi, i.e.,
h – 4 hi
hCu = s3 2 (5.6)

bCu = bs – 2 hi (5.7)
respectively.
For a three-phase machine for which the magnet cost per torque should
be kept low, the ratio of magnet width to pole pitch should be between 0.6
and 0.9 (Lampola et al. 1996a). In the proposed generator the magnet
width is kept at 0.7 times the pole pitch, i.e.,

bm = 0.7 τp (5.8)
The winding is a full-pitch winding and, therefore, the winding pitch is

W = τp (5.9)
The end winding length is assumed to be

lb = 2 W (5.10)
The equivalent core length is approximated by

le = l + 2 δ (5.11)
The useful iron length is

lu = kFes l (5.12)
where kFes is the stator iron fill factor. The frequency at rated speed is

f = p nN (5.13)
where p is the number of pole pairs. The air gap should be small to
minimize the amount of permanent magnets needed. The mechanical
stiffness and the thermal expansion of the generator limits the minimum
air gap which can be used. In this thesis the relation

δ = 0.001 d (5.14)
is used. Because the slot opening is narrow compared with the air gap,
the Carter factor will be 1. The outer diameter of the stator and the
approximate total length of the stator, including the end windings, are

dse = d + 2 hs + 2 hys (5.15)

ltot = l + 3 W (5.16)
respectively. (Error found after the defence of the thesis: The total winding length is
overestimated, in equation 5.16 the total length is more likely to be approx. l tot = l + 2 W .
This error will lead to a small overestimation of the copper weight and copper losses)
5.2.2 Magnetic Circuit
The stator yoke thickness, the rotor yoke thickness, and the stator tooth
width are
^ b l
B
hys = δ0 m e (5.17)
^ l
2 B ys u
^ b l
B
hyr = δ0 ^ m e (5.18)
2 B l
yr
^ τ l
B δ0 e
bd = ^ (5.19)
Bd0 lu
^ the
^ is the no-load peak air gap flux density and B
respectively, where B δ0 d0
no-load peak teeth flux density.
The mmf:s of the iron core can be calculated from the magnetization
curves for the stator and rotor core materials, i.e., Hs(B) and Hr(B). In the
stator yoke the mmf needed for the magnetic flux between two poles can be
approximated as
π (hs + 0.5 hys) 
v ys = c  τp +
^ ^ )
H (B (5.20)
 p  s ys
where c takes into account the variation of the field strength in the yoke.
For the stator, c is assumed to be 0.5.
The mmf needed for the teeth can be expressed approximately as
^ ^ ) (h + 0.5 h ) + H (B
v d = Hs(B ^
d0 s3 s2 s δ) (0.5 hs2 + hs1 ) (5.21)

The mmf of the rotor yoke is

 π (δ + hm + 0.5 hyr)  ^
^
v yr = c  τp –  Hr(Byr) (5.22)
 p 
where c is assumed to be 0.5 for the rotor. Because h m is included in this
expression, the required magnet height has to be calculated by iteration.
The required magnet height can be calculated analytically by replacing hm
with 2 δ in equation (5.22). This causes an error in the order of 1 % in v^yr.
The mmf drops of the magnet and the air gap are
^
Bδ0
^
v m = hm (5.23)
µm µ0
^
B
v δ = δef δ 0
^ (5.24)
µ0
respectively. µ m is the relative permeability of the permanent magnet
material.
The sum of the mmf:s around the magnetic circuit of two poles is zero
^ – ^
2 Hc hm – v v yr – 2 ^
vd – 2 ^
vδ – 2 ^
vm = 0 (5.25)
ys

where H c is the coercitivity of the permanent-magnet material. The

needed magnet height can now be calculated from Equations (5.23) and
(5.25)
^ys + 0.5 v^yr + v^d + v^δ
0.5 v
hm = (5.26)
^
Bδ0
Hc –
µm µ0
The flux density wave of a permanent magnet generator with uniform air
gap and surface magnets is ideally square-shaped. Due to fringing effects,
the real shape is smoother. Since the winding is a full-pitch winding, the
harmonics of the flux wave result in rather large voltage harmonics. The
voltage harmonics can contribute to the active power produced by the
generator if the armature current is non sinusoidal. Usually the
contribution from the harmonics is very small, thus, only the
fundamental components of the flux density wave and the voltage are
considered in the following calculations.
The shape of the flux density wave in the air gap depends mainly on the
magnet width, the pole pitch, the air gap and the magnet height. The
shape of the flux wave with a smooth stator surface was calculated by the
finite element method for several combinations of air gap, magnet height
and pole pitch, including combinations more extreme than what will
occur in the direct-driven generators in this thesis. In the calculations,
the magnet width was kept at 0.7 times the pole pitch. The fundamental
component was calculated from the air gap flux wave. The analysis
showed that the approximate relation

^  0.81 – 0.30 (hm + δ) 

Bδ(1) = Bδ0   (5.27)
 τp 
for

τp ≥ 4 ( hm + δ ) (5.28)
can model the RMS value of the fundamental flux density B δ(1) within
about 2 % of the values calculated with the finite element method.

5.2.3 Stator Inductance and Resistance

In order to calculate the output power of the generator, the inductance
and resistance of the armature winding must be known. In the
calculation of the tooth tip leakage inductance and magnetizing
inductance, the permanent magnets are assumed to have the same
permeability as air.
The slot leakage inductance, tooth tip leakage inductance and end
winding leakage inductance, for a winding without parallel branches,
can be expressed as

Lsl = 2 p q µ0 le λsl (5.29)

Ltl = 2 p q µ0 le λtl (5.30)

Lb = 2 p q µ0 lb λb (5.31)
where λsl, λtl and λb are the specific permeance of the slot leakage, tooth-
tip leakage and end winding leakage, respectively. For the proposed
generator, with equal current in the upper and lower conductor in the
slots, the average specific permeance of the slot leakage for the two coil
sides in the slot can be expressed as (Richter 1951, p. 269-271)
2h 3h h h b
λsl = 3 bCu + 2 b i + bs1 + b –s2b ln b s  (5.32)
s s s1 s s1  s1

The specific permeance of the tooth tip leakage can be calculated by an

approximate expression (Richter 1953, p. 90)
δ + hm
λtl = (5.33)
bs1 + 0.8 (δ + hm)
The specific permeance of the end winding leakage has been determined
experimentally for different winding types in (Richter 1953, p. 91–92). For
this two-layer winding the specific permeance is approximately

λb = 0.25 (5.34)
The magnetizing inductance is used to calculate the required reactive
power. Therefore, the total inductance is important, including the flux
harmonics caused by the stator winding. The single-phase magnetizing
inductance, for the winding with one slot per pole and phase, can be
expressed as

Lm = p µ0 le λm (5.35)

where the specific permeance of the single-phase magnetizing inductance

is
τp
λm = (5.36)
2 (δef + hm)
The expressions (5.35) and (5.36) for the single-phase magnetizing
inductance are derived in Appendix A.
The magnetizing inductance used in an equivalent Y-phase circuit is not
the single-phase inductance. Because of the mutual inductance between
the three phase windings, the apparent inductance of the equivalent Y-
phase is higher. The equivalent Y-phase magnetizing inductance is 4/3
times the value of equation (5.35). The factor 4/3 includes the total flux
generated by the stator winding and it is derived in Appendix A. For
sinusoidally distributed windings this factor is, instead, 3/2.

There is no mutual inductance between the slot leakage and tooth tip
leakage inductances of the different phases, and the magnetic coupling
between the end winding inductances is included in the empirical
permeance coefficient λ b . The equivalent Y-phase inductance of the
armature, therefore, is
4
La =
3 Lm + Lsl + Ltl + Lb (5.37)

and the total leakage inductance is

Lσ = Lsl + Ltl + Lb (5.38)

Since the resistance is temperature-dependent, its value depends on the

generator load and the ambient temperature. The skin effect can be
neglected since the winding is made of stranded wire. A typical value of
the stator per phase resistance at rated load and average ambient
temperature is
2 p q (l + lb)
Ra = ρCu(θCuAv) (5.39)
kCu 2 hCu bCu

where ρCu(θ) is the resistivity of copper at the temperature θ and θCuAv is

the winding temperature at rated load and average ambient temperature.
The annual average temperature in the southern part of Sweden is about
8˚C. Since the maximum ambient temperature is 40˚C and the maximum
winding temperature according to the specification 130˚C, the
temperature of the winding at rated load and average ambient
temperature is

θCuAv = 98˚C (5.40)

5.2.4 Material Volume and Weight

In this section the material consumption for the active parts of the
generator is calculated. The symbol ρ here is the density of the materials,
not the resistivity.
The volumes of the different materials are calculated and multiplied by
the specific weight of the material to find the material weight for the
windings m Cu, stator yoke m Feys, stator teeth m Fed, rotor yoke m Feyr and
magnets m m :

VCu = 2 (l + lb) Q hCu bCu kCu (5.41)

m Cu = ρCu V Cu (5.42)

VFeys = lu π (d + 2 hs + hys ) hys (5.43)

mFeys = ρFe VFeys (5.44)

 (τ – bs1) + bd 
VFed = lu Q  bd hs3 + h + (τ – b ) h  (5.45)
 2 s2 s1 s1 

mFed = ρFe VFed (5.46)

VFeyr = l π (d – 2 δ – 2 hm – hyr ) hyr (5.47)

mFeyr = ρFe VFeyr (5.48)

Vm = 2 p l bm hm (5.49)

mm = ρm Vm (5.50)
where ρ Cu , ρ Fe and ρ m are the density of copper, iron and permanent-
magnets, respectively. The total weight of the active parts of the generator
is

mtot = mCu + mFeys + mFed + mFeyr + mm (5.51)

5.2.5 Losses
The copper losses at a winding temperature of θCu can be calculated from
the resistivity of the copper ρ Cu (θ Cu ), the rms current density J s and the
copper volume V Cu:

PCu(θCu) = ρCu(θCu) Js2 VCu (5.52)

The copper losses will be higher if the ambient temperature is high and
lower if it is low. The thermal design of the generator must be carried out
at the maximum ambient temperature with the maximum value of the
copper losses, while the losses for the average efficiency should instead be
the typical values of the copper losses. Consequently, two values of the
copper losses are calculated: the maximum copper losses P CuMax at an
ambient temperature of 40˚C; and the typical copper losses P CuAv at an
ambient temperature of 8˚C,

PCuMax = PCu(θCuN ) (5.53)

PCuAv = PCu(θCuAv ) (5.54)

The core losses have to be calculated for each part of the iron core. They
are divided into hysteresis losses and eddy current losses. The core losses
can be calculated from the iron manufacturers loss data only if the losses
are multiplied by empirical loss factors for the hysteresis and eddy
current losses, k Hy and k Ft (>1). These empirical factors depend on the
difference between the test conditions and the conditions in a real
machine and they have different values for the yoke and the teeth. The
iron used for the stator yoke, 0.5 mm thick Surahammar CK-30, has the
approximate specific hysteresis and eddy current losses

pHy = 2.04 W/kg (5.55)

pFt = 0.76 W/kg (5.56)

respectively, at 50 Hz and 1.5 T. The core losses in the stator yoke are
^ 2
f  B
PHyys = kHyys mFeys pHy    ys  (5.57)
 50 Hz  1.5 T
^ 2
2  B
 f   ys 
PFtys = kFtys mFeys pFt (5.58)
 50 Hz  1.5 T
The empirical loss factors are approximately

kHyys = 2 kFtys = 1.8 (5.59)

for the yoke (Richter 1951, p. 213). The teeth losses are
^ 2
f B
PHyd = kHyd mFed pHy    d0  (5.60)
 50 Hz  1.5 T
^ 2
2  B
 f   d0 
PFtd = kFtd mFed pFt (5.61)
 50 Hz  1.5 T
The empirical loss factor for the hysteresis losses in the teeth is
approximately (Richter 1951, p. 213)

kHyd = 1.2 (5.62)

Because the flux wave is square-shaped, the eddy current losses in the
teeth are higher than for sinusoidal flux waves. Therefore, the eddy
current loss factor of 1.5, given by Richter (1951, p. 213), is increased to

kFtd = 2.5 (5.63)

It is assumed that the flux density in the iron, and, thereby, the core
losses, at rated load are the same as at no-load. For a direct-driven wind
turbine generator, connected to a forced-commutated rectifier that keeps
the armature voltage constant, the core losses were found to be rather
independent of load by Lampola et al. (1996b).
The core losses in the rotor yoke are small because the main flux is
constant in the rotor. The major rotor losses are the eddy current losses in
the magnets due to flux harmonics. The flux harmonics originate both
from the no-load flux harmonics caused by the stator slots and the flux
harmonics caused by the armature currents. The magnet losses are
reduced by using small magnets to build a complete magnet pole. The
magnet losses can be estimated by time stepping FEM calculations and
the losses are assumed to have a constant loss density at the magnet
surface. The magnet losses, therefore, can be expressed as

PFtm = pFtm 2 p bm l (5.64)

This estimation of the magnet losses is very rough. However, since the
magnet losses are important mainly for the magnet temperature, the
error will not be very important. In Section 5.4.2, it is shown that the
magnet temperature is only to a minor extent determined by the magnet
losses. In a paper by Lampola et al. (1996b) the specific rotor losses of the
generator are about 100 W/m2 at rated load. For the proposed generator,
which is similar to the one investigated by Lampola et al. (1996b), the
specific losses are assumed to be

pFtm = 300 W/m2 (5.65)

because of higher current loading.
Additional losses (stray load losses) in synchronous generators consist,
according to Chalmers (1965), of losses due to slot leakage flux, losses due
to end leakage flux, short-circuit iron losses due to the armature mmf and
rotor pole face losses. In the proposed generator the winding is made of
stranded wire, which eliminates eddy current losses due to slot leakage.
The rotor surface losses P Ftm are dealt with separately and, thus, are not
included in the additional losses in these calculations. The rest of the
additional losses are mainly core losses. They are at rated load assumed
to be about 20 % of the core losses at no load, i.e.,

Pad = 0.2 (PHyys + PFtys + PHyd + PFtd ) (5.66)

The additional losses are assumed to be proportional to the square of the
armature current. In the thermal calculations, the additional losses are
assumed to be located in the stator tooth tips.
Friction and windage losses are caused by the losses in the generator
bearings, the windage losses of the rotor and the radial fan on the rotor
used to circulate the internal air. The bearing braking torque is not
determined so much by the generator design as by the loading from the
turbine. It is assumed to be 0.5 % of the rated torque and to be independent
of the speed. The losses of the radial fans can be calculated approximately
and they are found to be very small, less than 100 W for a 500 kW
generator. Because of the low air gap speed, the windage losses are
neglected. The total friction and windage losses at rated speed are
assumed to be

Pµ = 0.005 PN (5.67)

The power of the external cooling fans, blowing air through the cooling
ducts at the outer surface of the stator yoke, are not included in this
calculation. The total losses at rated load can now be calculated as

PlossMax = PCuMax + PFtys + PHyys + PFtd + PHyd + PFtm + Pad + Pµ (5.68)

The average losses can be calculated by using the average loss factors
derived in Chapter 3:

PlossAv = kdCu (PCuAv + Pad) + kdFt (PFtys + PFtd + PFtm) +

+ kdHy (PHyys + PHyd) + kdµ Pµ (5.69)

5.2.6 Voltage, Power and Efficiency

The induced fundamental no-load armature phase voltage is

Ep = 2 p q kw(1) Bδ (1) le vδ (5.70)

where vδ is the velocity of the flux wave in the air gap, given by

vδ = π d nN (5.71)

and k w(1) is the winding factor for the fundamental flux density wave.
Because the winding is a full-pitch winding, the winding factor k w(1) is 1.
The generator phase voltage is at rated speed controlled by the forced-
commutated rectifier to be equal to the no-load phase emf

UapN = Ep (5.72)
The rated rms phase current is

IaN = Js 2 hCu bCu kCu (5.73)

where k Cu is the fill factor of copper inside the stranded coils, excluding
the coil insulation, equal to 0.8. The phase current from the rectifier will
include harmonics and, therefore, the fundamental component is lower
than the total rms current. The rms value of the fundamental component
is assumed to be 98 % of the total rms value of the armature current, i.e.,

I(1) = 0.98 IaN (5.74)

From the rated voltage, rated current and the current phase angle, the
rated electrical output power can be calculated as
 PaN = 3 UapN I(1) cos(ϕN) (5.75)
where, according to (3.25)

√
I 2πf L 2
cos(ϕN) = 1 –  (1)2 U N a  (5.76)
 apN 
The rated mechanical shaft power can now be calculated from the
electrical output power and the losses, i.e.,

PN = PaN + PlossMax (5.77)

The efficiency at rated load is
P
ηN = PaN (5.78)
N

By using the average factor for the turbine power, derived in Chapter 3,
the average input power can be calculated as

PAv = PN kt (5.79)
The average efficiency is
PlossAv
ηAv = 1 – P (5.80)
Av

The armature reactance is in this thesis expressed in per unit. The per
unit base values used are

Ubase = UapN (5.81)

Ibase = IaN (5.82)

5.2.7 Thermal Model and Temperature Rise

The aim of the thermal calculations is to find the maximum temperature
of the stator winding and the magnets. The generator is represented by a
lumped-parameter thermal network model.
The thermal model represents the generator by the circuit shown in
Figure 5.3. The temperature differences in the circumferential direction
of the generator are neglected. The generator cooling is symmetrical in
the axial direction and, therefore, the two end windings of a coil are
modelled as one. The thermal network model is derived in Appendix B.
The losses in the thermal model are copper losses in the stator winding,
core losses in the stator teeth and yoke, eddy current losses in the magnets
and additional losses. Friction and windage losses are assumed not to
increase the temperature rise of the winding and magnets. Thus, they are
neglected in this thermal model. The copper losses are divided into losses
in the end windings, losses in the bottom-layer coil sides in the slots, and
 Stator cooling air Ambient air

End shield
Stator yoke

Endwinding

Internal air
Lower coil side
Tooth

Upper coil side

Permanent-
magnet

Rotor yoke

Figure 5.3 The thermal model based on one slot pitch, one rotor pole, one
coil, internal air and end shield.

losses in the top-layer coil sides. The magnet losses are assumed to be
distributed homogenously in the magnets, while additional losses are
assumed to be located in the tooth tip. The temperature rise of the cooling
air along the cooling ducts at the outer surface of the stator yoke is
included in the thermal model as an equivalent thermal resistance.
Because the major part of the losses is cooled at the outer surface of the
stator yoke, the maximum winding temperature is the end winding
temperature.

5.2.8 Irreversible Demagnetization

Irreversible demagnetization can be a problem in permanent-magnet
generators. The magnets will be demagnetized if the flux density in them
is lower than a minimum flux density of B min . The high-temperature
NdFeB material used in these calculations can withstand at least 0 T at
the assumed maximum operating temperature of 120˚C. The minimum
flux density may in many cases also be a negative value, usually at low
temperatures.
Of course, the magnets must not be demagnetized by the rated current
and it is also important that they can withstand the most probable faults,
for instance, short circuits. During normal operation the armature
winding generates a peak flux density in the air gap and magnet that
depends on the peak value of the stator mmf v^ s and the magnetic air gap.
It is given by
^
vs
^
Bs = µ0 (5.83)
δ + hm/µm
For the proposed generator type with three phases and one slot per pole
and phase, the peak mmf generated by the armature is the peak value of
the current in one slot

v s = ^I aN
^ (5.84)
To avoid demagnetization of the magnets, the stator must not generate a
^ than
higher peak flux density B s
^ – B
^ < B
Bs δ0 min (5.85)

Equation (5.85), however, does not take the risk of local demagnetization
into account. The flux density may locally be lower than B min because of,
for instance, flux leakage at the magnet ends. The minimum flux density
should be checked in the final design by a finite element calculation.
The risk of demagnetization at a short circuit at the generator terminals
is discussed by Slemon (1992). For a generator with sinusoidally
distributed windings there is no risk of demagnetization if the ratio
between leakage inductance and magnetizing inductance is
Lσ 8 sin(0.5 αM)
Lm > – 1 (5.86)
 Bmin (δ + hm/µm) 
π 1 – 
 Br hm/µm 
where α M is the magnet angle. For a generator with large resistance, the
factor 8 can be changed to 4. The proposed type of generator does not have
a sinusoidally distributed winding. Equation (5.86) is still used as an
indication of the risk of demagnetization during short circuit. Since the
minimum flux density here is zero and the magnet angle is 126˚, the
expression becomes
Lσ
Lm > 1.27 (5.87)

Because of the small pole pitch and the large air gap, demagnetization is
generally not a problem in the proposed type of generator. This was
verified by checking that the peak flux density generated by the stator
winding is below the value in Equation (5.85) and that the leakage
inductance to magnetizing inductance ratio is larger than described by
Equation (5.87).
If there is a risk of irreversible demagnetization, the risk can be
eliminated by increasing the air gap and at the same time increasing the
magnet height in such a way that the air gap flux density is unchanged.
Then, the stator will generate lower flux density at rated current and the
magnetizing inductance will be reduced. The reduced magnetizing
inductance will also increase the possible peak power. Of course, the price
for this change is increased cost for the magnets.

5.3 Calculation Procedure

The design method is implemented in the computer program Matematica
(Version 2.1 for Machintosh), which facilitates easy use of symbolic
mathematics. The generator design starts by executing all the definitions
in Section 5.2 and Appendix B. Once that is done, the program can
calculate any of the defined quantities if only the design variables are
given values. In this thesis seven variables are used. If, for example, the
average efficiency of a generator is to be calculated, the seven design
variables are given their values and the definition for the average
efficiency is executed giving the value of the average efficiency of that
particular generator. The active weight of the same generator can now be
calculated simply by executing that definition, without changing the
variable values. A new generator can be calculated by changing one or
more of the variables and executing the definitions again.
The rated power and winding temperature of a generator does not
automatically fulfil the specification, with arbitrary values of the seven
design variables. The proper winding temperature and rated power is
achieved by adjusting two variables iteratively. In this thesis, the stator
length and winding current density are used to reach a feasible generator
design. Before the iteration, these variables have to have starting values.
The variables are adjusted until the error of the winding temperature is
smaller than 0.1 K and the error of the rated power is smaller than 0.1 W.
During the optimization of the generator, the starting values for the
length and current density are the final values of the generator designed
in the previous step in the optimization.
In the way that the design calculations are carried out, only the
definitions of the quantities which are sought are executed. There is no
program which has to be run through completely for each new generator.
During the optimization procedure only the cost function is executed. The
other definitions will only be used once the performance and data of the
optimized generator are calculated.
Generator optimization is made using a numerical Matematica function
which finds a minimum of a multi-variable function. The generator is
optimized regarding air gap diameter, pole pitch, air gap flux density,
teeth flux density, slot height, and the temperature of the windings. The
search is unconstrained and if any optimum variables are outside the
allowed limits, the optimization is repeated with a fixed value of that
variable. In practice, it is only the winding temperature that sometimes
has an optimum value not allowed by the specification.

5.4 Test of the Design Method

5.4.1 Comparison with Finite Element Calculations

To verify and test the design method, a generator with an air gap
diameter of 2.5 m, a stator length of 0.5 m and a rated speed of 30 rpm was
designed. The generator is a rather typical 550 kW generator, but it was
not optimized. The variables used in the calculations are shown in
Table 5.2.
To verify some of the analytical calculation methods, the performance of
the test generator was also calculated by the finite element method. The
finite element calculations were both steady state calculations and time-
stepping calculations with a rotation of the rotor. A program called
"FCSMEKB" which has originally been developed at Helsinki University
of Technology was used. The program has been modified for permanent
magnet machines by Antero Arkkio and Jorma Luomi at Chalmers
University of Technology. A comparison of the analytical design method
and the finite element calculations is shown in Table 5.3.
The voltage and the rated torque calculated with the analytical model
correspond well to the results of the finite element calculations. The

Table 5.2 The variable values of the test generator

Variable Value

Air gap diameter d 2.5 m

Stator length l 0.5 m
Pole pitch τp 60.4 mm
Slot height hs 55 mm
Current density Js 4.27 A/mm2
^
Peak air gap flux density Bδ 0 0.72 T
^
Peak tooth flux density Bd0 1.63 T
Table 5.3 Comparison of the analytical design calculations and FEM
calculations for the test generator.

Analytical method FEM

Open circuit 246 V 240 V

fundamental voltage
Torque at rated current 175 kNm 177 kNm
Reactances x a = 0.97 p.u. x d = 0.88 p.u.
(x q = x d = x a assumed)
(in d- and q-axis) x q = 0.82 p.u.
Stator core losses 2982 W 2314 W
Magnet losses 824 W 311 W
Rotor yoke losses included in magnet losses 93 W
No-load torque ripple not calculated ± 0.8 kNm

analytically calculated inductance is 10 % higher than predicted by the

finite element calculations, but the results cannot be expected to be much
better since there are simplifications in the inductance calculation
methods. The analytical method is expected to overestimate the
inductance since the mmf of the stator core was not included in the
inductance calculations. The core losses are lower for the finite element
calculations than for the analytical method. However, the analytical
model includes empirical corrections for non-ideal manufacturing which
the finite element calculations do not. The magnet losses of the analytical
model are based on a rough estimation, and the finite element calculation
shows that they are at least not underestimated for this generator. Torque
ripple has not been included in the design calculations because the semi-
closed slots will keep the torque ripple low. The finite element calculations
verify this assumption.

5.4.2 Test of Thermal Model

Since the generator has not been built yet, the thermal model cannot be
verified by measurements. One way of showing how the errors of the
model can influence the design of the generator is to vary the uncertain
parameters of the thermal model to see what effect they have on the
winding temperature and magnet temperature. First the temperatures of
the test generator were calculated using the normal values of the
parameters in the thermal model. The calculations were made for the test
generator described in Section 5.4.1. and the results are shown in
Table 5.4.
The influence of the uncertain thermal parameters on the overall
generator design was tested by changing the parameter values to twice
and to half the nominal values. For each new parameter value, the
temperature of the windings and the magnets were calculated. The
calculations with a parameter value causing a temperature increase
were selected (either twice or half of the nominal parameter value). The
resulting temperature increases in the windings and the magnets can be
found in Figure 5.4. The parameters that were changed are: heat transfer
coefficients at the outer surface of the stator yoke, in the air gap, at the end
shields, at the end windings and at the inner surface of the rotor yoke; the
thermal conductivity of the slot insulation; and the losses in the magnets.
It is not the thermal conductivity of the slot insulation that is the
uncertain parameter, but rather the thickness of the insulation. However,
the thickness cannot be changed in the calculations because that will
increase the conductor cross-section and, therefore, increase the rated
power and the copper losses of the generator. Changing the thermal
conductivity instead will have the same effect on the heat transfer as
changing the insulation thickness, although the generator performance
will not be changed.
Figure 5.4 shows that the winding temperature is very sensitive to the
heat transfer coefficient at the outer surface of the stator yoke. The only
other parameter with an important influence on the winding temperature
is the thermal conductivity of the slot insulation. The reason for this is
that almost all the losses in the windings are cooled through the slot
insulation and the stator yoke.
The magnet temperature is also very sensitive to the thermal resistance
between the stator yoke and the cooling air. The reason is that good
thermal contact between the stator and the rotor makes the magnet

Table 5.4 The calculated temperatures at rated load in the test

generator.
Ambient temperature 40 ˚C
Stator cooling air 50 ˚C
Stator yoke 98.6˚C and 99.3 ˚C
Lower teeth 107 ˚C
Upper teeth 113 ˚C
Lower coil side 115 ˚C
Upper coil side 121 ˚C
End winding 122 ˚C
Magnets 96 ˚C
Internal air 77 ˚C
temperature almost as high as the stator teeth temperature. An increase
in stator teeth temperature will cause an almost as large increase in
magnet temperature. Also the thermal resistance at the end shields and
the inner surface of the rotor yoke, as well as the losses in the magnets,
influence the magnet temperature. It is important to note that the magnet
losses do not influence the magnet temperature very much. The
temperature of the magnets will be a little lower than the temperature of
the stator teeth even if the magnet losses differ greatly from what has been
estimated. Since the winding is not allowed to be warmer than 130˚C and
the teeth are about 10 ˚C colder, it must always be possible to keep the
magnet temperature below 120 ˚C, which is the limit in the specification.
The winding and magnet temperatures can become very high if the stator
cooling is not as efficient as assumed. However, the assumed heat
transfer coefficient at the outer surface of the stator yoke is not higher
than what is achieved in a normal induction machine (Kylander 1993,
p. 66).
Increase in temperature (°C)

40

30

20

10

0
A B C D E F G

Winding temperature Magnet temperature

Figure 5.4 The increase in the temperatures of the windings and

magnets as different thermal parameters are doubled and
halved. The changed parameters are:
A) Heat transfer coefficient at the stator yoke;
B) Heat transfer coefficient in the air gap;
C) Heat transfer coefficient at the end shields;
D) Heat transfer coefficient at the end windings;
E) Heat transfer coefficient at the rotor yoke;
F) Thermal conductivity of the slot insulation and ;
G) Losses in the magnets.
6 Generator Optimization
In this chapter, the generator design method is used together with a
numerical optimization method to investigate the optimization of a
generator of the proposed type. A 500 kW generator has been choosen
since that is a normal size for today's commercial wind energy
converters.
In Section 6.1, five 500 kW generators are optimized. The first generator is
optimized using the nominal cost function, and that generator will be
used as a reference design throughout this chapter. Three other
generators are also optimized with the nominal cost function, but they
were required to produce either 50 Hz or 200 % peak power, or both 50 Hz
and 200 % peak power. The fifth generator is optimized using a cost
function with the losses at rated load instead of average losses.
The influence of different variables on the generator design is investigated
in Section 6.2 and the influence of the cost function in Section 6.3. The
optimum diameter of the generator is especially interesting since the
problems of manufacturing and transportation can be expected to depend
on the generator diameter to a great extent. In Section 6.4, the optimum
diameter is discussed in detail. In Section 6.5 the typical size, average
efficiency and active weight of a 500 kW generator of the proposed type are
summarized.

6.1 Optimum 500 kW Generators

6.1.1 Optimized Reference Generator

In this section, a 500 kW generator is optimized using the nominal values
of the cost function parameters. The cost of the average losses is
6000 ECU/kW and the cost of iron, copper and permanent magnets are 4, 6
and 100 ECU/kg, respectively. The reference diameter of the structure is
2 m, the reference length 1 m, the cost of the reference structure is
20 000 ECU and the exponent for the structure cost is 3. A medium wind
speed site is assumed, i.e., the average loss factors for a site with an
average wind speed of 6.8 m/s are used. The rated speed of the 500 kW
generator is 32 rpm and the rated torque is 148 kNm. Data for the
optimized reference generator are presented in Table 6.1.
The optimum diameter of the generator is 2.15 m and the stator length is
0.55 m. Both the flux density in the air gap and the teeth have normal
optimum values and the optimum winding temperature is lower than
what is allowed in the specification. A higher winding temperature would
lead to an uneconomically low efficiency. The frequency of the optimized
Table 6.1 Data of the optimized 500 kW reference generator.

Ratings: Geometry:

Rated torque TN 148 kNm Stator length l 0.55 m

Rated speed nN 32 rpm Tooth width bd 11.1 mm

Optimized variables: Slot width bs 11.7 mm

Air gap diameter d 2.15 m Air gap δ 2.15 mm

Slot height hs 64.0 mm Magnet height hm 6.3 mm

Pole pitch τp 68.3 mm Stator yoke height hys 15.9 mm

^
Air gap flux den. Bδ 0 0.77 T Rotor yoke height h yr 15.4 mm
^
Teeth flux dens. Bd0 1.64 T End wind. length lb 137 mm

Winding temp. ΘCuN 107 ˚C Conductor width bCu 9.7 mm

Electrical data: Conductor height hC u 27.5 mm

Frequency f 26.5 Hz Weights:

Reactance xa 0.95 p.u. Magnet weight mM 124 kg

Current density Js 3.60 A/mm2 Copper weight mCu 779 kg

Current in a slot IaN 1540 A Stator yoke w. m Fes 467 kg

Losses, efficiency: Teeth weight m Fed 888 kg

Copper losses PCuN 22700 W Rotor yoke weight m Fer 431 kg

Stator yoke losses PFeys 760 W Total active w. m tot 2690 kg

Stator teeth losses PFed 1950 W Miscellaneous:

Magnet losses PFtm 780 W Magnet temp. Θm 90 ˚C

Additional losses Pad 540 W Induct. ratio 1) – 1.7

^
Frict. & windage Pµ 2500 W Stator flux dens. Bs 0.33 T

Average losses PdAv 6430 W Cost of losses Cd 38500 ECU

Average eff. ηAv 94.9 % Cost of active part Cact 24200 ECU

Full-load eff. ηN 94.2 % Cost of structure Cstr 16000 ECU

1) Leakage inductance divided by magnetizing inductance.

generator is only 26.5 Hz and the reactance is 0.95 p.u., which is a high
reactance value compared with grid-connected generators. Such a high
reactance is possible only because the generator is not required to produce
a high peak torque.
The weights of the stator and rotor yokes are low. From a mechanical
point of view, the stator yoke may have to be made thicker than 16 mm to
be stiff enough. The magnet weight is 124 kg, which means that the
magnets will cost about 12 400 ECU. This cost is about the same as the cost
for 2 kW of average losses.
The magnet temperature and the risk of demagnetization have not been
included in the optimization and, thus, have to be checked. As predicted,
the magnet temperature is lower than the limit, and the peak flux density
^
Bs in the air gap caused by the stator winding at rated current is only 0.4
times the flux density generated by the permanent magnets. Therefore,
the magnets will not be demagnetized during normal operation. The ratio
of the leakage inductance to the magnetizing inductance is 1.7, which is
enough to avoid demagnetization during a sudden short circuit according
to equation (5.87).
Both the losses at rated load and the average losses of the generator are
illustrated in Figure 6.1. The copper losses totally dominate at rated load,
but since the average copper losses for the chosen wind energy converter
site are only 14 % of the ones at rated load, the copper losses are still only
half of the average losses. The other loss components are, on the average,
approximately half of their rated load values. Since the total average
losses are 22 % of the total losses at rated load and the average power
produced by the turbine is 25 % of the rated power, the average efficiency
is higher than the efficiency at rated load. The average efficiency is
94.9 %, while the efficiency at rated load is 94.2 %. For conventional, grid-
connected, four- or six-pole wind energy converter generators, the average
efficiency is lower than the efficiency at rated load.

Rated load

Average

0 5 10 15 20 25 30
Losses (kW)
Friction and windage losses Hysteresis losses

Eddy current losses Copper losses

Figure 6.1 The losses of the optimized 500 kW generator at rated load and
average losses.
6.1.2 Optimized Generators for 50 Hz and 200 % Peak Power

The reference generator is designed for a variable-speed, pitch-controlled

wind turbine and requires neither any specified frequency, nor a peak
power higher than the rated power. In this section, a generator for direct
grid-connection is designed (for a pitch- or stall-controlled wind turbine),
as well as a generator for stall-controlled, variable-speed wind turbines. A
generator which is direct connected to the grid has to produce a frequency
of 50 Hz and, for stability reasons, it must have a peak power of about
200 % of the rated power. A stall-controlled wind energy converter with
variable speed also requires about 200 % peak power, to control the turbine
speed, but does not require a frequency of 50 Hz.

A generator design with a frequency of 50 Hz is achieved by keeping the

number of pole pairs constant, which means that the pole pitch is no
longer a free variable. A peak power of 200 % is assured by limiting the
allowed armature reactance to be no more than 0.52 p.u. This maximum
value of the armature reactance acts as a second limit for the current
density, beside the maximum winding temperature.

In Table 6.2, the different generator designs are compared with the
reference generator. To make the comparison more complete, a 50 Hz
generator without any required peak power is also included in the
comparison. The generators are compared regarding average efficiency,
active weight and size. The generator size is, here, expressed as the
volume of the rotor.

The comparison shows that a demand for 200 % peak power leads to about
a 25 % larger rotor volume than for the reference generator, while the

Table 6.2 Generators designed with or without requirement of 200 % of

the rated power and requirement of 50 Hz.

Frequency Peak power to Air gap Stator Average Active Rotor

rated power diameter length efficiency weight volume
(Hz) (%) (m) (m) (%) (kg) (m3)

26.5 120 2.15 0.55 94.9 2690 2.00

29.7 200 2.33 0.60 94.5 2401 2.56

50 124 2.39 0.56 93.6(1 1928 2.51

50 200 2.51 0.60 93.4(1 1795 2.97

(1 The average efficiency is calculated for variable speed operation. At constant
speed the average efficiency will be lower.
average efficiency is 0.4 % lower and the active weight 10 % lower. The
lower active weight depends mainly on the increased air gap diameter.
Requiring 50 Hz frequency of a generator also leads to about a 25 % larger
rotor volume than of the reference generator. To keep the diameter low,
the pole pitch has become only approximately 40 mm and, therefore, the
slots are narrow. Narrow slots lead to a low copper fill factor and higher
resistance. As a consequence, the average efficiency is 1.3 percent-units
lower for the 50 Hz generator than for the reference generator. The active
weight is almost 30 % lower than for the reference generator. About two
thirds of the weight reduction is because of the reduced pole pitch, the
other third because of increased air gap diameter.

A generator for direct grid-connection, both 50 Hz and 200 % peak power,

will have about 50 % larger rotor volume than the reference generator.
The active weight is about 30 % lower than for the reference generator
partly because of the increased air gap diameter, partly because of
reduced yoke thickness owing to a smaller pole pitch. The average
efficiency is 1.5 percent-units lower than for the reference generator.
More data of the direct grid-connected generator is presented in Table 6.3.
No damper windings were included in this design, so the direct grid-
connected generator has to be provided with mechanical damping of the
type presented by Westlake et al. (1996). Damping by means of damper
windings is possible only with a much larger generator diameter.

Table 6.3 Data of an optimized 500 kW generator for direct grid-

connection. The frequency is 50 Hz and the peak power 200 %
of the rated power.

Optimized variables: Electrical data:

Air gap diameter d 2.51 m Reactance xa 0.52 p.u.

Slot height hs 34.0 mm Geometry:

Pole pitch τp 42.2 mm Stator length l 0.60 m

^
Air gap flux den. Bδ0 0.74 T Tooth width bd 6.5 mm
^
Teeth flux dens. Bd0 1.68 T Slot width bs 7.6 mm

Winding temp. ΘCuN 108 ˚C Miscellaneous:

Efficiency: Cost of losses Cd 49300 ECU

Average eff.(1 ηAv 93.4 % Cost of active part Cact 22700 ECU

Full-load eff. ηN 92.6 % Cost of structure Cstr 25700 ECU

(1 The average efficiency is calculated for variable speed operation. At constant
speed the average efficiency will be lower.
6.1.3 Optimization Using the Losses at Rated Load

Because the average factor for the copper losses is only 0.14 while the
average factors for the core losses are about 0.50–0.60, a generator
optimized with the average losses in the cost function can be expected to
have high copper losses at rated load compared with the core losses. If the
losses at rated load are used in the cost function, instead of the average
losses, the optimum generator will have lower copper losses and higher
core losses. The effect of using average losses in the cost function, instead
of the losses at rated load, was tested by changing the cost function for the
optimization. To make sure that the generator is comparable with the
reference generator, the diameter was fixed at 2.15 m and the cost of the
losses at rated load was adjusted until the average efficiency was 94.9 %,
as it is for the reference generator. Data of a 500 kW generator, optimized
using the losses at rated load in the cost function, are shown in Table 6.4.
The differences between this generator and the reference generator are
rather small. The current density is 12 % lower, the slot 7 % higher and
the stator 11 % longer than for the reference generator. The efficiency at
rated load is 0.5 percent units higher. The cost of the active parts of the
generator has increased slightly along with the cost of the structure.

Table 6.4 Data of a 500 kW generator optimized with the losses at rated
load in the cost function.

Optimized variables: Geometry:

Slot height hs 68.4 mm Air gap diameter d 2.15 m
Pole pitch τp 57.9 mm Stator length l 0.61 m
Air gap flux den. ^ 0.75 T Tooth width bd 9.2 mm
Bδ
Teeth flux dens. ^ 1.64 T Slot width bs 10.1 mm
Bd
Winding temp. ΘCuN 92 ˚C Losses, efficiency:
Electrical data: Copper losses PCuN 18700 W
Reactance xa 0.94 p.u. Stator yoke losses PFeys 842 W
Current density Js 3.17A/mm2 Stator teeth losses PFed 2760 W
Miscellaneous: Magnet losses PFtm 861 W
Cost of av. losses Cd 38500 ECU Additional losses Pad 720 W
Cost of active part Cact 25300 ECU Frict. & windage Pµ 2500 W
Cost of structure Cstr 20200 ECU Average eff. ηAv 94.9 %
Full-load eff. ηN 94.7 %
The losses at rated load and the average losses for this generator and the
reference generator are illustrated in Figure 6.2. The difference between
the generator loss distributions is not very large. This low difference
indicates that the very high copper losses of the reference generator are
not only a consequence of the cost function, but also a natural loss
distribution in this type of low-speed generator. Moreover, the efficiency
curves for the two generators were calculated, using the g-functions from
Chapter 3. The efficiency curves, plotted in Figure 6.3, are found to be
rather similar, but with the peak efficiency at different powers.

A: Rated load

A: Average

B: Rated load

B: Average

0 5 10 15 20 25 30
Losses (kW)
Friction and windage losses Hysteresis losses

Eddy current losses Copper losses

Figure 6.2 The losses of two generators with equal average efficiency:
Generator A: Reference generator;
Generator B: Optimized using the cost of the losses at rated
load.

96
95
Efficiency (%)

94
93
92 Optimized using the losses at rated load
91 Optimized using the average losses
90
100 200 300 400 500
Input power (kW)

Figure 6.3 Efficiency curves of two generators. One is optimized with the
losses at rated load in the cost function and one with the
average losses. Both generators have equal average efficiency.
The generator optimized regarding average losses will be smaller and
have a slightly lower total cost than if the losses at rated load are used.
However, one reason to design a generator with high efficiency at rated
load, even if the average efficiency is not improved, would be because the
cooling system might be made simpler and cheaper.

6.2 Sensitivity to Variable Changes

In this section, the influence of six design variables on the stator length,
the active weight, the losses and the cost function is investigated. The
influence of a variable is tested by changing its value to 0.9 or 1.1 times the
optimum value, while the rest of the variables are kept at their optimum
values. A new generator is calculated for each variable that is changed.
The generator is not optimized; only the stator length and current density
are adjusted to get the right rated power and right winding temperature.
The results are presented in Table 6.5.
The main result from this investigation is that the total cost of the
generator is rather insensitive to changes in any of the design variables.
This insensitivity indicates a flat minimum in the cost function. The flux
densities in the air gap and the teeth have the highest influence on the
total cost, but the total cost increases only by approximately 3 % as the flux
densities are changed 10 %.
The pole pitch has almost no influence on the total cost of the generator.
The active weight and the cost of the active parts increase as the pole pitch
increases. However, the core losses decrease at the same time.
A 10 % change in the air gap diameter increases the total cost of the
generator by approximately 1.5 %. As the diameter decreases, the stator
length along with the cost of the active part increase. The copper losses
increase with decreasing diameter while the core losses remain almost
constant.
The slot height has very little effect on the total cost of the generator. A
decrease in the slot height requires an increase in stator length because
the current in a slot has to be decreased in order to maintain the original
winding temperature. The copper losses increase as the slot height
decreases. At the same time, however, the core losses decrease since the
major part of the core losses are losses in the teeth. The total influence on
the average losses is minor.
The air gap flux density has the largest influence on the total cost of the
generator. If the flux density is reduced, the stator length has to be
increased and both copper and core losses increase. The amount of
permanent magnets is decreased as the flux density is decreased, leading
to a lower cost for the active part of the generator. The influence of the
teeth flux density on the generator is similar to the influence of the air gap
flux density, although the total cost is changed slightly less.
The winding temperature has a very low influence on the total cost. The
influence on the copper losses is large, but while the copper losses
increase much with increasing winding temperature the core losses
decrease much, reducing the influence on the average losses. Decreased
winding temperature requires a longer stator and, therefore, the cost of
the active part of the generator increases.

Table 6.5 The influence of the design variables on the generator. The
design variables are changed one by one while the rest have
their optimum values. The changes are expressed in percent.

Changed Stator Active Copper Core Average Cost of Total

variables length weight losses losses losses active cost
part

τp = 0.9 τpOpt +0 –5 +1 +9 +3 –2 +0

τp = 1.1 τpOpt +0 +5 –1 –6 –2 +3 +0

d = 0.9 dOpt +23 +9 +8 +0 +4 +5 +2

d = 1.1 dOpt –17 –7 –6 +0 –3 –4 +1

hs = 0.9 hsOpt +4 –3 +3 –2 +1 –1 +0

hs = 1.1 hsOpt –3 +4 –2 +2 –0 +2 +0

^ ^ +12 +6 +11 +4 +6 –4 +3
Bδ = 0.9 BδOpt

^ ^ –5 –0 –6 +2 –2 +15 +3
Bδ = 1.1 BδOpt

^ ^ +6 +5 +6 –1 +3 –1 +2
Bd = 0.9 BdOpt

^ ^ –4 –2 –5 +2 –2 +14 +3
Bd = 1.1 BdOpt

ΘCuN = +7 +6 –14 +7 –5 +7 +1
0.9 ΘCuNOpt

ΘCuN = –5 –5 +13 –5 +5 –5 +0
1.1 ΘCuNOpt

Reference 0.55 m 2690 22700 3490 6420 W 24200 78700

generator kg W W ECU ECU
6.3 Sensitivity to Cost Function Changes
The cost function used in this thesis is only a rough estimate of the real
costs. Therefore, it is important to investigate how the cost function
influences the generator design. In this section, the cost of losses, cost of
iron and copper, cost of permanent magnets and the cost of the structure
are changed to see how they influence the optimum generator.

6.3.1 Cost of Losses

The effect that the specific cost of losses has on the optimized 500 kW
generator is investigated by changing the cost of average losses between 0,
10, 50, 100, 200 and 1000 % of the nominal value. The extreme values 0, 10
and 1000 % are, of course, not realistic. Nevertheless, they have been used
to show more clearly which parameters are important for the average
efficiency. Data of the generator optimized with different specific cost of
losses are shown in Table 6.6. With increasing specific cost of losses the
diameter increases slightly, the active weight increases considerably and,
of course, the average efficiency increases. The average losses with no
cost of the losses are 8 percent-units and 3.5 percent-units at the highest
cost of the losses. The increased efficiency is mainly reached by
increasing the slot height and decreasing the current density. Thus, a
higher efficiency causes a higher active weight. The efficiency increase is,
to some extent, also reached by increasing the diameter. The air gap flux
density, however, is not increased as the cost of losses is increased.
The average losses of three of the optimized generators are shown in
Figure 6.4. It can be seen that the decrease in average losses is mainly
caused by a decrease in copper losses. The eddy current losses are

Table 6.6 Generators optimized using different loss costs.

Cost of Average Air gap Active Slot Current Air gap flux W i n d i n g
losses efficiency diameter weight height density density temperature
(%) (m) (kg) (mm) (A/mm 2 ) (T) (˚C)

0% 92.0 2.05 1814 46 5.6 0.67 130

10 % 92.6 2.04 1907 48 5.3 0.70 130

50 % 93.7 2.07 2187 55 4.7 0.75 130

100 % 94.9 2.15 2689 64 3.6 0.77 107

200 % 95.6 2.32 3274 71 2.7 0.76 89

1000 % 96.5 2.73 5369 93 1.5 0.75 71

 1000 %

Cost of losses
100 %

10 %

0 2 4 6 8 10
Average losses (kW)

Friction and windage losses Hysteresis losses

Eddy current losses Copper losses

Figure 6.4 The average losses of three generators optimized using

different specific costs of the losses (10, 100 and 1000 % of the
nominal cost of the average losses).

approximately halved but the hysteresis losses remain almost constant.

The windage and friction losses are assumed constant and cannot be
changed by optimization.
One way of decreasing the core losses, which is not included in the
optimization in this thesis, is to use iron with lower losses and perhaps
also thinner laminations.

6.3.2 Cost of Iron and Copper

The influence of the cost of the active iron and copper on the optimum
generator is tested by changing the cost of these two materials to 0 %,
10 %, 50 %, 200 % and 1000 % of their nominal costs. The results are
shown in Table 6.7.
Decreased specific cost of iron and copper leads to a more efficient
generator with a smaller diameter but a higher active weight. The
average losses change between 4.5 and 7.0 percent-units, the diameter
changes from 2.07 to 2.59 m and the weight from 3887 to 1443 kg as the
costs of iron and copper change from 0 to 1000 % of their nominal values.
The costs of iron and copper mainly influence the active weight. The
weight increases as copper and iron costs decrease because of the higher
slots and also because of a slightly increased pole pitch. The pole pitch
increases because the slot width increases. The copper losses are reduced
as the costs of iron and copper decrease since the slots are made higher
and the current density decreases.
Table 6.7 Generators optimized with different specific copper and iron
costs.

Cost of Average Air gap Active Magnet Slot Stator Winding

iron and efficiency diameter weight weight height length temperature
copper
(%) (m) (kg) (kg) (mm) (m) (˚C)

0% 95.5 2.07 3887 134 88 0.60 90

10 % 95.4 2.08 3677 132 84 0.59 92

50 % 95.1 2.11 3106 127 72 0.57 100

100 % 94.9 2.15 2689 124 64 0.55 107

200 % 94.4 2.21 2223 120 55 0.52 118

1000 % 93.0 2.59 1443 118 36 0.47 130

6.3.3 Cost of Permanent Magnets

The specific cost of the permanent magnets is a rather uncertain
parameter. The permanent magnets are expensive, but the price has been
falling for a number of years. It is, therefore, interesting to see what effect
magnet price has on the optimum generator. The specific cost of the
magnets has been changed from 0 to 1000 % of the nominal value,
100 ECU/kg. Some data of the different optimized generators are shown in
Table 6.8.
It can be seen that the amount of magnets increases considerably as the
specific cost of the magnets is reduced. The differences in the air gap and

Table 6.8 Generators optimized with different specific magnet costs.

Cost of Average Air gap Active Air gap Magnet Teeth Magnet
magnets efficiency diameter weight flux height flux weight
density density
(%) (m) (kg) (T) (mm) (T) (kg)

0% 95.3 2.00 3142 0.89 16.2 1.83 285

10 % 95.2 2.03 2988 0.87 12.3 1.78 219

50 % 95.0 2.10 2802 0.81 7.9 1.70 148

100 % 94.9 2.15 2689 0.77 6.3 1.64 124

200 % 94.6 2.23 2549 0.72 5.1 1.59 104

1000 % 93.1 2.49 1977 0.60 3.3 1.49 76

teeth flux densities are, however, not very large. Even with a very low
specific cost of the magnets, the optimum air gap flux density is only
about 0.1 T higher than for the generator optimized with the nominal
value of the specific cost of the magnets. The diameter decreases with
decreasing magnet cost and the active weight increases, but the average
efficiency changes only slightly. The conclusion of these results is that the
average efficiency, the diameter and the active weight are not very
dependent on the specific cost of the magnets. The specific cost of the
magnets will mainly influence the material cost of the generator.
Since the generator is only slightly improved by increasing flux density,
flux concentration will not lead to any significantly different generator
performance than surface magnets do. Ferrite magnets and flux
concentration, however, may lead to a cheaper generator.

6.3.4 Cost of the Structure

The approximate cost of the structure has several parameters. In this
section, the reference structure cost cstr, the structure exponent a and the
reference length lref are changed in order to see which effect they have on
the optimum generator. The results of the changed cost of the structure
are presented in Table 6.9. The parameters have been changed one at a
time, while the other two have their nominal values. The nominal values
of the parameters are cstr = 20 kECU, a = 3 and lref = 1 m.
Even major changes in the cost of the structure have little influence on the
generator efficiency and active weight. The size of the generator is not
changed drastically, unless cstr is reduced to 0 or 2 kECU or a is decreased
to 1. For all other tested values of the cost function parameters, the
diameter varies between 1.97 to 2.50 m and the average efficiency between
94.4 and 95.2 %. The weight of the active parts varies between 2048 and
3129 kg.
If the structure is not included (cstr=0), the optimization is only carried out
for the electromagnetic part of the generator. The optimum diameter then
becomes much larger than if the cost of the structure is included. Note,
however, that the average efficiency of the very large generator is not
much different from that of the reference generator. The cost of the active
part does not decrease much compared with the reference generator. A
large diameter, therefore, is not motivated.
The conclusion from this investigation is that the cost of the structure
must be included in the optimization, because it is the main limiting
factor for the diameter. Exactly how the cost of the structure varies with
diameter does not have to be known in order to find an approximate
optimum diameter.
Table 6.9 Generators optimized with different structure costs.

Changed cost Average Air gap Active Stator Current Reac- Winding
parameter efficiency diameter weight length density tance temp.
(%) (m) (kg) (m) (A/mm2) (p.u.) (˚C)

cstr=0 kECU 95.7 6.33 1902 0.12 2.8 0.80 87

cstr=2 kECU 95.6 3.26 2366 0.32 3.1 0.89 93

cstr=10 kECU 95.2 2.39 2664 0.50 3.4 0.93 99

cstr=20 kECU 94.9 2.15 2689 0.55 3.6 0.95 107

cstr=40 kECU 94.4 1.97 2611 0.58 4.0 0.98 121

cstr=200kECU 93.9 1.78 2690 0.62 4.1 1.00 130

a=1 95.5 3.39 2181 0.28 3.2 0.90 97

a=2 95.1 2.46 2534 0.45 3.4 0.94 102

a=3 94.9 2.15 2689 0.55 3.6 0.95 107

a=4 94.7 2.01 2749 0.60 3.7 0.96 111

lref = 0.5 m 94.4 2.50 2048 0.36 4.1 1.00 130

lref = 1 m 94.9 2.15 2689 0.55 3.6 0.95 107

lref = 2 m 94.9 1.98 3129 0.73 3.4 0.92 96

6.4 Optimum Generator Diameter

The diameter of a direct-driven generator is of great importance to a wind
energy converter manufacturer. The optimum diameter is a compromize
between the electromagnetic part and the structure of the generator. Since
the cost of the structure is highly uncertain, it is interesting to investigate
the influence of the diameter on the optimization.
The main variables that link the optimization of the electromagnetic part
to the structure are the diameter and the length. In Figure 6.5 the length
of the stator is plotted as a function of air gap diameter for optimized
generators. Because the cost of the structure increases both with
increasing length and with increasing diameter, the cost of the structure
will have a minimum value. Both large and small diameters will lead to a
high structure cost.
 1.4
1.2

Stator length (m)

1
0.8
0.6
0.4
0.2
0
1 2 3 4 5 6 7
Air gap diameter (m)

Figure 6.5 The optimum stator length as a function of diameter.

The cost of the active part, the cost of the losses and the cost of the
structure were calculated as functions of air gap diameter. The generator
was optimized for each fixed value of the diameter. The results are shown
in Figure 6.6.
The losses, and consequently the cost of the losses, decreases rapidly as
the diameter increases from 1 to 2 m. As was shown in Section 6.2, the
core losses are rather diameter-independent. The reason the losses
decrease with increasing air gap diameter is that the active parts of the
windings are reduced. At small diameters, when the stator is rather

50000
Cost of structure
40000
Cost (ECU)

30000 Cost of losses

20000

10000 Cost of active parts

0
1 2 3 4 5 6 7
Air gap diameter (m)

Figure 6.6 The three parts of the cost function as functions of air gap
diameter. The generator is optimized for each value of the air
gap diameter.
long, this phenomenon reduces the losses as the diameter increases. At
large diameters, the stator is short and the end windings become more
important. While the active part of the winding decrease with increasing
diameter the amount of end windings increases. If the air gap diameter is
between 3 and 5 m, the total losses are almost independent of the
diameter. For diameters larger than 5 m the losses increase slightly with
increasing diameter.
The cost of the active part of the generator is more diameter-dependent
than the cost of the losses. If the generator diameter is decreased, the
amount of active material is increased. Since the generator cost function
will keep the optimum stator short, the increased cost depends mostly on
increasing slot height and increasing flux density. Increasing flux
density leads to high magnet weight. Under a diameter of 2 m, the cost of
the active material increases rapidly and, therefore, the diameter cannot
be much less than 1.5 m for this 500 kW generator.
At large diameters, over 2 m, the stator is short and the cost of the
structure is determined mainly by the diameter. Consequently, the cost of
the structure increase along with the diameter. For small diameters,
under 1.5 m, the stator is long and, thus, the cost of the structure is
determined mainly by the stator length. Therefore, the cost of the
structure increase as the diameter is decreased below 1.5 m. A large
diameter has to be avoided because of transportation and manufacturing
problems and a small diameter has to be avoided because a long generator
structure is difficult to manufacture.
If only the cost of the active parts and the cost of the losses are considered,
the optimum diameter would be more than 6 m. If the cost of the structure
is included in the optimization, the optimum diameter is decreased
drastically, to about 2 m. The reason is easily seen in Figure 6.6; the cost
of the losses and the cost of active materials are rather diameter-
independent for diameters above 2.5 m.
The optimum generator diameter is not very sensitive to changes in the
shape of the cost function for the structure. The reason is a significant
increase in the cost of the active material and the cost of the losses as the
diameter decreases below 1.75 m, in combination with a very slow
decrease in those costs as the diameter increases beyond 2.5 m. As a
consequence, the optimum diameter is likely to be within 1.75–2.5 m.

6.5 Typical 500 kW Permanent-magnet Generator

Ranges for the efficiency, active weight and air gap diameter can be found
from the data in Section 6.3. The maximum and minimum values of the
average efficiency, active weight and air gap diameter are presented in
Table 6.10. Only the values for generators optimized with specific costs cd,
cFe, cCu, lref, cstr equal to 50 or 200 % of the nominal cost and a equal to 2
and 3 were used. The generators optimized with more extreme values of
the cost function parameters were excluded.
The 500 kW permanent-magnet generator will have an average efficiency
of about 94.6 %, an active weight of about 2600 kg and a diameter of about
2.2 m. The efficiency can increase slightly if the specific cost of the losses
is high or if the specific costs of the iron and copper are low. The
increased efficiency will lead to an increase of the active weight. An
increase in the diameter only leads to a small increase in the efficiency
and a small decrease in the active weight; a decrease in the diameter will
increase the active weight and decrease the efficiency.

Table 6.10 Data of a typical 500 kW direct-driven, permanent-magnet

generator of the proposed type.

Average efficiency Active weight Air gap diameter

94.6 ± 1 % 2600 ± 700 kg 2.2 ± 0.3 m

7 Design and Comparison
In this chapter the performance and data of the proposed generator type is
presented and discussed. Direct-driven generators are designed and
optimized for rated powers from 30 kW to 3 MW. A comparison is made of
the average efficiency and the size of direct-driven generators of the
proposed type and conventional drive trains, consisting of a gear and a
grid-connected induction generator. Some other direct-driven generators
are also compared with the proposed generator type.

7.1 Generators from 30 kW to 3 MW

Direct-driven generators are interesting for all sizes of wind energy
converters. Wind energy converters from about 30 kW and up to 3 MW are
of the same basic design and can be assumed to follow Equations (2.1) and
(2.2) as pertains to rated speed and rated torque as functions of rated
power. Therefore, generators within that power range were investigated.
The calculated generator sizes are 30, 50, 100, 170, 300, 500, 1000, 1700 and
3000 kW. All the generators were optimized using the nominal values of
the cost function parameters presented in Table 2.3. The efficiency, size
and active weight of the direct-driven generators are presented in Section
7.1.1. Optimum values of the design variables and other results from the
electromagnetic design are discussed in Section 7.1.2. In Section 7.1.3 the
differences between generators of different rated power are discussed.

7.1.1 Generator Data

Both the average efficiency and the efficiency at rated load of the optimized
generators are shown in Figure 7.1. The efficiency of the direct-driven
generators increases with the rated power, as it also does for conventional
generators. An important reason for the increase in efficiency is that the
rotor surface velocity increases; a higher rotor surface velocity means that
a higher active power can be produced per square meter of air gap
surface, with a given force density. The average efficiency increases from
92.5 % for a 30 kW generator to 95.8 % for a 3 MW generator, and the rotor
surface velocity increases from 2.58 to 4.56 m/s. The efficiency at rated
load is about 1 percent-unit lower than the average efficiency.
The size of the direct-driven generator is important. Generators will be
more difficult to manufacture the larger they are, but an even more
important problem might be the transportation to the site. The optimum
air gap diameter and stator length of the generator are shown in
Figure 7.2. An optimized 30 kW generator has an air gap diameter of
0.8 m and a 3 MW generator 4.1 m. The corresponding stator lengths are
 96

95

Efficiency (%)
94

93

92 Average efficiency

91 Efficiency at rated load

90
30 100 300 1000 3000
Rated power (kW)

Figure 7.1 The efficiencies of the direct-driven generators

4
Diameter and length (m)

Air gap diameter

3
Stator length

0
30 100 300 1000 3000
Rated power (kW)

Figure 7.2 The air gap diameter and stator length of the direct-driven
generators.

0.2 m for the 30 kW generator and 1.1 m for the 3 MW generator. The
length-to-diameter ratio increases as the rated power increases, but even
the 3 MW generator has a length-to-diameter ratio of less than one third.
The weight of the generator is usually not a problem for the wind energy
converter during operation. Forces on the wind energy converter tower,
for instance, are determined almost exclusively by the forces from the
turbine, not by the generator weight. Nevertheless, the weight can be
important for the erection of the wind energy converter. A heavy generator
demands a larger crane or that the machinery be lifted in several parts.
Since the mechanical part of the generator has not been designed, only the
active weight is calculated. The weight of the generator structure is
expected to be much higher than the active weight. The active weight
increases slightly less than linearly with the rated power, see Figure 7.3.
The figures are 210 kg for a 30 kW generator, 1700 kg for a 300 kW
generator and 14000 kg for a 3 MW generator.
Since it was not included in the design method, the risk of
demagnetization of the magnets was also checked. The larger generators
have a lower minimum flux density in the magnets than the smaller
ones, but none of the generators risks irreversible demagnetization of the
permanent magnets.

15000
Active weight (kg)

10000

5000

0
500 1000 1500 2000 2500 3000
Rated power (kW)

Figure 7.3 The active weight of the generator as a function of rated

power.

7.1.2 Optimum Variables and Parameter Values

All the generators, from 30 kW to 3 MW, are of the same type and
optimized with the same cost function. It is interesting, therefore, to see in
what respect the generators of different rated powers are similar and in
what respect they are different.
The optimum air gap flux density is about 0.76 T and the optimum teeth
flux density about 1.65 T for all generator sizes. The variations are only
about ±1 %.
The force density in the air gap of the generators is presented in
Figure 7.4. The larger generators have a higher force density than the
smaller ones. Since the flux density is constant, the higher force density is
achieved by a higher current loading.
The optimum winding temperature also increases as the rated power of
the generator increases. In Figure 7.5, the diagram shows that the
temperature limit is reached as the rated power is 3 MW or larger. This
means that the smaller machines can be made with simpler cooling
systems than the larger ones. The magnet temperature is also plotted in
the diagram. The magnet temperature is well below 120˚C for all of the
generators.
The per-unit reactance of the optimized generators increases with
increasing rated power. Figure 7.6 shows that the value of the reactance
increases from approximately 0.75 p.u. for a 30 kW generator to 1.05 p.u.
for a 3 MW generator.

50
Force density (kN/m^2)

40

30

20

10

0
30 100 300 1000 3000
Rated power (kW)

Figure 7.4 The force density in the air gap of the optimized generators

130
Temperature (degree C)

120 End windings

110 Magnets

100

90

80

70
30 100 300 1000 3000
Rated power (kW)

Figure 7.5 The maximum winding temperature and magnet

temperature of the optimized generators.
The slot height and pole pitch of the optimized generators both increase
with increasing rated power, as is shown Figure 7.7. It is the increasing
current loading which causes the slot height to increase. The slot pitch
and pole pitch are not optimized independently. Because the number of
slots per pole is constant, the slot pitch is proportional to the pole pitch.
Therefore, the optimum pole pitch and the slot pitch could have other
values if they were optimized separately.
The results presented in this section are rather typical of the design of
electrical machines in general. It is a well known phenomenon that
larger generators have a higher force density than smaller ones. The
higher force density is the reason for the increase in slot height, current

1.1
Reactance (p.u.)

0.9

0.8

30 100 300 1000 3000

Rated power (kW)

Figure 7.6 The reactance of the optimized generators.

0.1
Pole pitch and slot height (m)

Pole pitch
0.08
Slot height

0.06

0.04

0.02

0
30 100 300 1000 3000
Rated power (kW)

Figure 7.7 The slot height and pole pitch of the optimized generators.
loading, reactance and winding temperature as the rated power
increases. The increased force density, however, is not a necessary
consequence of increased generator size; it is caused by the optimization.

7.1.3 Power Limits For the Direct-driven Generators

This section examines whether or not there exists a practical upper or
lower limit for the rated power of direct-driven generators. The reasons
for such a limit may be the total cost, the weight or the size of the
generator.
The three parts of the cost function only give a rough estimate of the cost of
the generator. In Figure 7.8, the three parts of the cost function are shown
to illustrate how the cost changes with rated power. All three parts of the
cost function increase almost linearly with the rated power. This indicates
that the specific cost, ECU per kW, of a direct-driven generator is about the
same for all generator sizes.
The weights of the active materials are plotted as functions of the rated
power in Figures 7.9 and 7.10. The iron weight and the copper weight
increase slightly less than linearly and the magnet weight increases
slightly more than linearly. Since the magnets are only a small part of the
active weight, the total active weight still increases less than linearly with
increasing rated power. Thus the specific active weight, kg per kW, will be
slightly lower for a large generator than for a small one.
The estimated outer diameter of the generators is plotted as a function of
the rated power in Figure 7.11. The diameter increases with increasing
power, but the increase is rather small, because of the increase in force

200
Losses
150 Active parts
Cost (kECU)

Structure
100

50

0
0 1000 2000 3000
Rated power (kW)

Figure 7.8 The cost of active parts, cost of losses and cost of structure
for the generators as a function of rated power.
 Copper and iron weight (kg)
10000
Iron weight

Copper weight

5000

0
0 1000 2000 3000
Rated power (kW)

Figure 7.9 The weight of the iron core and winding as a function of
rated power.

1000
Magnet weight (kg)

500

0
0 1000 2000 3000
Rated power (kW)

Figure 7.10 The weight of the permanent-magnets as a function of rated

power.

density and length-to-diameter ratio with increasing rated power. If there

is a definite maximum diameter, then this type of generator can be made
with a smaller outer diameter than what is shown here. However, in that
case the active weight will increase, the average efficiency decrease
slightly and the length will increase.
The specific cost and specific active weight will not increase as the rated
power increases. From these two aspects the upper power limit for direct-
driven generators for wind energy converters seems to be more than
3 MW. Consequently, it is mainly the size of the generator that can act as a
 5

Appr. outer diameter (m)

4

0
30 100 300 1000 3000
Rated power (kW)

Figure 7.11 The approximate outer diameter of the direct-driven

generators.

practical limit for large direct-driven generators. Compared with other

parts of the wind energy converter, however, like the tower, the nacelle
and the turbine blades, the diameter of the proposed generator type does
not necessarily have to be a large problem.

7.2 Comparisons
Direct-driven generators have to be more cost effective than conventional
systems with a gear and a four- or six-pole generator in order to be
interesting. The main question is whether or not a direct-driven generator
can bring down the cost of the produced energy. This comparison is not
made regarding the total cost, however, since the cost function is too
approximate for that purpose. Instead, the generator systems are
compared regarding weight, efficiency and size.

7.2.1 Comparison with Conventional Generators and Gears

The proposed direct-driven generator type was compared with drive trains
in conventional wind energy converters of three different rated powers.
Four-pole generators were assumed rather than six-pole, since four-pole
generators are smaller, have lower weight and higher efficiency. The
comparison of the generator systems is shown in Table 7.1.
The weights of the gears and induction generators are based on data for
commercial gears and standard generators. For the direct-driven
generators the weights are only the active weight. The total weight of the
gears and induction generators are 2.1 to 3.8 times the active weight of the
Table 7.1 A comparison of the proposed direct-driven generator type
with conventional drive trains. Efficiency values refer to the
efficiency from the turbine shaft to the grid.

Rated power 30 kW 500 kW 3000 kW

Weight of induction generator and 440 kg 7700 kg 53000 kg

gear

Active weight of direct-driven 210 kg 2700 kg 14000 kg

generator

Full-load efficiency of induction 89.7 % 93.7 % 94.3 %

generator and gear

Full-load efficiency of direct-driven 87.4 % 90.3 % 91.4 %

generator and frequency converter

Average efficiency of induction 81.7 % 88.4 % 90.0 %

generator and gear

Average efficiency of direct-driven 88.3 % 90.7 % 91.6 %

generator and frequency converter

Diameter and length of generator 0.6 m 1.5 m 2.5 m

and gear combination 1.3 m 3m 6m

Outer diameter and length of the 1.2 m 2.7 m 5m

direct-driven generator 0.5 m 1.2 m 2m

direct-driven generators. It seems realistic to expect that the total weight

of the two alternative systems will be of the same order.
Both the efficiency at rated load and the average efficiency were compared.
The efficiency of the direct-driven generators includes the frequency
converter, which was assumed to be constructed with insulated-gate
bipolar transistors. Its efficiency at rated load is assumed to be 95.9 % and
the average efficiency is 95.6 %. (The average efficiency of the conventional
systems and the frequency converter for the direct-driven generators are
calculated in Appendix C).
At rated load the conventional systems are more efficient than the direct-
driven generators, but they are less efficient on the average. The
difference in average efficiency is 1.6 and 2.3 percent-units for the 3 MW
and 500 kW systems, respectively. For the 30 kW system the difference is
larger, 6.6 percent-units.
The size of the gear and generator is defined by their maximum diameter,
at the gear box, and the total length, including a coupling on the high-
speed shaft. The outer dimensions of the direct-driven generators are
estimated; they will of course depend on how the structure is constructed.
Note that the compared systems are not completely equal. The
conventional drive trains have a fixed turbine speed, while the direct-
driven generator with a frequency converter allows variable turbine speed.
Variable-speed turbines can produce about 5 to 10 % more energy, which
will make the variable-speed, direct-driven generator more worthwhile
than the grid-connected induction generator. The increased energy
production by means of the variable turbine speed is not included in the
efficiency comparison above. The efficiency discussed is only the efficiency
from the turbine shaft to the grid.
High-slip generators have more recently often been used in large wind
energy converters, in order to limit turbine power peaks and damp power
oscillations. The rotor copper losses of the high-slip generators are 1–2 %
higher at rated load than for the standard induction generators in
Table 7.1. If two-speed generators are used the efficiencies of the induction
generators will also be lower than what is shown here, about 0.5–1
percent-unit lower at rated load. In comparison with these types of
generators, the use of direct-driven generators will lead to an even larger
increase in average efficiency.

7.2.2 Comparison with Other Direct-driven Generators

In the previous section the proposed direct-driven generator type was
compared with conventional drive trains; in this section it is compared
with other direct-driven generators. The comparison is made with three
theoretical design studies and one commercial generator. Unfortunately,
the weight and efficiency of the commercial generator is not available. The
values of the diameter, length, efficiency and active weight of the
generators are shown in Table 7.2. The generators of the proposed type
have all been optimized with the nominal cost function defined in
Chapter 2.
Often, only electromagnetic losses are included in the theoretically
calculated efficiency of generators, because the friction losses are difficult
to determine without knowlegde of the mechanical design. In the
efficiency of the proposed generator 0.5 % friction losses are included. As
pertains to the other direct-driven generators it is not known which losses
are or are not included in the efficiency figures.
The proposed generator is of about the same size and slightly less efficient
at rated load when compared with a transversal-flux generator by Weh
et al. (1988). The transversal-flux generator has a lower active weight but
it needs more permanent magnets than the proposed generator does.
However, the presented transversal-flux generator is designed for diode
rectifier connection which reduces its rated power. If the transversal-flux
generator is designed for a forced-commutated rectifier, it will be smaller
than the proposed generator.
The air gap diameter of the proposed generator is 8 % smaller and the
stator is 38 % shorter than those of the similar generator, with surface

Table 7.2 Comparison of the proposed direct-driven generator with

other direct-driven generators.

Air gap Stator Efficiency at Active

Generator, diameter length rated load weight
rated shaft power (Outer (Outer (Average (Magnet
diameter) length) efficiency) weight)

Transversal-flux 1.00 m – 94 % 176 kg

generator (1.2 m) (0.35 m) – (14 kg)
60 kW at 78rpm
Proposed generator 0.91 m 0.20 m 93.4 % 318 kg
1) 1)
60 kW at 78 rpm (1.15 m) (0.40 m) (93.9 %) (9.4 kg)

Direct grid-connected 2.4 m 0.91 m 94.2 % 4100 kg

Radial-flux generator – – – –
532 kW at 28.3 rpm
Proposed generator 2.28 m 0.58 m 94.0 % 3088 kg
532 kW at 28.3 rpm – – (94.8 %) (145 kg)
2) 3)
Axial-flux generator 3.46 m 0.17 m 97 % 3000 kg
2) 4) 1)
1040 kW at 100 rpm 2.94 m 0.52 m – (>>100kg)
Proposed generator 1.90 m 0.46 m 96.9 % 2456 kg
1040 kW at 100 rpm – – (95.6 %) (103 kg)
1) 1)
Electrically excited 4m 0.2 m – –
Radial-flux generator
1) – – – (0 kg)
≈540 kW at 40 rpm
Proposed generator 2.06 m 0.52 m 94.8 % 2487 kg
540 kW at 40 rpm – – (95.3 %) (111 kg)
1) Values estimated by the author
2) Outer and inner diameter of the stator
3) Axial length of the active part of the generator.
4) Active stator length = two times the difference between inner and outer stator radii.
mounted NdFeB magnets, designed for 50 Hz direct grid-connection
(Spooner and Williamson, 1992c). The efficiency at rated load is almost
equal, and the proposed generator has 31 % lower active weight than the
direct grid-connected generator. One important reason for this large
difference in generator size is that the proposed generator has a reactance
of 0.96 p.u. while the direct grid-connected generator has a reactance of
only 0.31 p.u. The low reactance of the direct grid-connected generator is
necessary to produce a high peak power.
Honorati et al. (1991) presents a 1 MW axial-flux generator designed for
wind energy converters. The outer stator diameter of the axial flux
generator is 82 % larger than the air gap diameter of the proposed
generator, and the active stator length (two times the outer minus inner
stator radius) is 13 % longer than the stator of the proposed generator.
Since the stator is oriented in the radial direction in the axial-flux
generator, however, the axial length of the active generator part is 63 %
shorter than the stator length of the proposed generator. Both the axial-
flux generator and the proposed generator have high efficiency. Such high
efficiency is not realistic for a normal wind energy converter generator;
the high efficiency depends on the very high rated speed of the generators.
The active weight of the axial-flux generator is 22 % higher than that of
the proposed generator. Since the axial-flux generator has an air gap
winding, the amount of permanent magnets needed can be expected to be
very high in comparison with the 103 kg needed for the proposed generator
type.
A commercially used electrically excited generator was included in the
comparison, even if the data of that generator are only approximate. The
generator is used in the Enercon E-40 wind energy converter (Anon.,
1994a). The comparison shows that the electrically excited generator has a
94 % larger diameter and a 62 % shorter stator than the proposed
generator. The rotor volume of the electrically excited generator is about
50 % larger than that of the proposed generator type.
The comparisons in this section show that the proposed radial-flux
permanent-magnet generator with a forced-commutated rectifier can
have a small size compared with the other direct-driven wind turbine
generators. It is much smaller than the electrically excited generator, the
axial-flux generator and the direct grid-connected radial flux generator.
It is of about the same size as the transversal flux generator with a diode
rectifier. The reason for the small size of the proposed generator is mainly
that a high peak power is not required. The efficiency at rated load is
similar for the proposed generator type and the alternatives, but the
average efficiency cannot be compared since it has not been given for the
other generators.
8 Conclusions
A radial-flux permanent-magnet generator is a good option for direct-
driven wind turbine generators. Connected to the grid via a frequency
converter, it can be made small and efficient. The generator type can be
used in wind energy converters, from at least 30 kW up to 3 MW, and it is
more efficient than conventional four-pole induction generators with
gears. By allowing the rated torque to be close to the pull-out torque, the
proposed generator type can be made smaller than many other direct-
driven generators in the literature.

8.1 Different Generator Types

Several generator topologies can be used for direct-driven wind turbine
generators, for instance, radial-flux, axial-flux and transversal-flux
topologies. Radial-flux generators are slightly more efficient and need
slightly less active material than axial-flux generators do. Moreover,
radial-flux topology allows a smaller outer diameter than axial-flux
topology does. The transversal-flux generator is very small, efficient and
light. There is one drawback, however; its structure is more complicated
than the structure of a radial-flux generator. The final choice of the
generator type cannot be made based exclusively on the analysis of the
active part of the generators. The choice depends, to a large extent, on
manufacturing aspects.
Electrically excited generators are difficult to use as direct-driven wind
turbine generators, because a large air gap is required and that leads to
large losses due to the magnetizing current. By making the pole pitch
large the losses due to the magnetizing current can be reduced, but that
results in a very heavy generator. Permanent-magnet generators are
more efficient than electrically excited generators and, if the pole pitch is
made small, they are also lighter.
Generators with low-energy magnets and flux-concentration or high-
energy magnets mounted on the rotor surface will have similar efficiency
and size. Low-energy magnets for flux-concentration are cheaper than
surface-mounted, high-energy magnets, but a rotor for flux-concentration
is more complicated and heavier.
Direct grid-connected generators are more difficult to design than
generators connected to the grid via a frequency converter. A direct grid-
connected 500 kW generator, of the proposed type, is about 50 % larger and
1.5 % less efficient than what a generator connected to a frequency
converter can be. A 500 kW variable-speed generator required to produce a
peak power of 200 % of the rated power will be about 25 % larger and 0.4 %
less efficient than if no peak power capacity is required.
8.2 Generator Design and Optimization
A direct-driven radial-flux permanent-magnet generator with magnets
mounted on the rotor surface was chosen for a theoretical investigation.
This generator type has several favourable properties when compared
with other types. The generator is connected to a forced-commutated
rectifier in order to allow for a small diameter. An analytical design
method was developed for this generator type. The cost function used for
the generator optimization includes the cost of the active part of the
generator, the cost of the average losses and the cost of the generator
structure. A method to calculate the average losses of a wind turbine
generator was also developed. The average losses were calculated by
multiplying the losses at rated load with average loss factors for the
different types of losses. This way of calculating the average losses is very
easy to use in the optimization of a generator.

The cost function used for the optimization is approximate, but it is shown
that the efficiency, active weight and size of an optimized generator do not
change much even if the cost function is changed a great deal. If the
efficiency is increased, the active weight of the generator increases. The
diameter of the generator has a practical lower limit because the stator
length, the losses and the active weight increase rapidly as the diameter
is decreased below a certain diameter. An increase in diameter, above the
practical lower limit, increases the efficiency and reduces the active
weight only slightly. Therefore, the optimum diameter can be expected to
be close to the practical lower limit.

8.3 Designed Generators and Comparison with Other

Generators
Generators from 30 kW to 3 MW were designed. They are all feasible as
wind turbine generators. The active weight per kW and total cost per kW
are about the same for all the generator sizes. The maximum power of
direct-driven generators may be limited by the diameter. The outer
diameter of the proposed generator type, however, is not much larger
than the width of the nacelle in a conventional wind energy converter.

The average efficiency of the proposed direct-driven generator, including

the frequency converter losses, is higher than that of a gear and a direct
grid-connected induction generator. Furthermore, the frequency
converter allows variable turbine speed. The variable speed will increase
energy production in comparison with the direct grid-connected generator
system. Compared with some other direct-driven generators, the proposed
generator type is small.
8.4 Further Work
The generator type proposed in this thesis is promising for use in future
wind energy converters. There are aspects that should be investigated
further, however. The mechanical structure should be investigated in
detail, both for a final optimization of the generator and in order to
determine the generator price. A prototype should be built to demonstrate
that a generator with a high inductance can produce a high force density
in the air gap when connected to a forced-commutated rectifier. Some
parts of the design method can also be improved to increase accuracy.
Inductance calculations can be improved by including the saturation, and
the loss model can be improved. In addition, the optimum slot pitch and
pole pitch should be investigated. In this thesis the number of slots per
pole and phase was constant.
References
Adkins B, Harley R.G. 1975. The general theory of alternating current
machines, London, Great Britain: Chapman and Hall, 279 p.
Alatalo M, 1991. "Konstruktionsstudie av en permanentmagnetiserad
axialflödesmaskin." Göteborg, Sweden: Chalmers University of
Technology, Department of Electrical Machines and Power Electronics,
Technical Report No. 108L. 107 p.
Alatalo M, Svensson T, 1993. "Variable speed direct-driven PM-generator
with a PWM controlled current source inverter." European Community
Wind Energy Conference (ECWEC'93), Lübeck-Travemünde, Germany,
8–12 March 1993, Proceedings, p. 639–642.
Anon., 1992, Guidelines for Certification of Wind Turbine Power Plants.
Norway: Det Norske Veritas.
Anon., 1994a, Product description leaflet from the Enercon company,
Germany, 4 p.
Anon., 1994b, European Wind Turbine Catalogue. Copenhagen,
Denmark: Energy Centre Denmark, 63 p.
Bindner H, Søndergaard L, Damgaard E. 1995. "Undersøgelse af
generatorkoncepter til vindmøller - opsummeringsrapport" (in Danish).
Roskilde, Denmark: Forskningscenter Risø, Risø-R-857(DA), October 1995,
15 p.
Carrichi F, Crescimbini F, Honorati O, Santini E. 1992. "Performance
evaluation of an axial-flux PM generator". International Conference on
Electrical Machines (ICEM'92), Manchester, U.K., 15-17 September 1992,
Proceedings, vol. 2, p. 761-765.
Chalmers B.J. 1965. Electromagnetic problems of A.C. machines.
London, Great Britain: Chapman and Hall Ltd, 102 p.
Davies E.J. 1971. "Airgap windings for large turbogenerators",
Proceedings IEE, vol. 118, No 3/4 1971, p. 529-535.
de Haan S, van Engelen T, Frumau C, Veltman A, Wildenbeest E, Childs
S, Torrey D. 1994. "Development of a gearless drive with variable
reluctance generator for variable speed wind turbines". European Wind
Energy Conference (EWEC'94), Thessaloniki, Greece, 10–14 October 1994,
Proceedings vol. III, p. 130–133.
Deleroi W. 1992. "Linear induction motor in generator use for windmills".
Symposium on Power electronics Electrical drives Advanced electrical
motors (SPEEDAM), Positano, Italy, 19-21 May 1992, Proceedings,
p. 71−76.
Di Napoli A, Carrichi F, Crescimbini F, Noia G. 1991. "Design criteria of
a low-speed axial-flux PM synchronous machine". I n t e r n a t i o n a l
Conference on the evolution and modern aspects of synchronous
machines (SM 100), Zurich, Switzerland, 27-29 August 1991, Proceedings
part 3, p. 1119-1123a.

Grauers A. 1994. "Synchronous generator and frequency converter in

wind turbine applications: system design and efficiency", Göteborg,
Sweden: Chalmers University of Technology, Department of Electrical
Power Engineering, Technical report No. 175 L, 125 p.

Gribnau W.H.J.K., Kursten J.P.J. 1991. "Electrical flexibillity versus

mechanical complexity through direct conversion." European Wind
Energy Conference (EWEC'91), Amsterdam, The Netherlands, 14–18
October 1991, Proceedings Part I, p. 809-813.

Honorati O, Caricchi F, Crescimbini F, Noia G. 1991. "Gear-less wind

energy conversion system using an axial-flux PM synchronous
machine". European Wind Energy Conference (EWEC'91), Amsterdam,
The Netherlands, 14-18 October 1991, Proceedings Part I, p. 814-818.

Jöckel S, 1996, "Gearless Wind Energy Converters with Permanent

Magnet Generators–An Option for the Future?". European Union Wind
Energy Conference (EUWEC'96), Göteborg, Sweden, 20–24 May 1996,
Proceedings, (in press).

Kylander G. 1993. "Modellering och mätning av temperatur och förluster

i mindre asynkronmaskiner" (In Swedish). Göteborg, Sweden: Chalmers
University of Technology, Department of Electrical Machines and Power
Electronics, Technical Report No. 147L, 107 p.

Lampola P, Perho J, Saari J. 1995a. "Electromagnetic and thermal design

of a low-speed permanent magnet wind generator". Stockholm Power
Tech Conference, Stockholm, Sweden, 18-22 June 1995, Proceedings
vol. "Electrical machines and drives", p. 211-216.

Lampola P. 1995b. "Dirtectly driven generators for wind power

applications". EWEA special topic conference: The economics of wind
energy, Finland, 5-7 September 1995, Proceedings p. E4-1 to E4-6.

Lampola P, Perho J, Väänänen J. 1996a. "Analysis of a low-speed

permanent-magnet wind generator". European Union Wind Energy
Conference (EUWEC'96), Göteborg, Sweden, 20–24 May 1996, Proceedings,
(in press).

Lampola P, Perho J, Väänänen J. 1996b. "Analysis of a low-speed

permanent-magnet wind generator connected to a frequency converter".
International Conference on Electrical Machines (ICEM'96), Vigo, Spain,
10-12 September 1996, Proceedings, vol II, p. 393-398.
Richter R, 1951. Elektrische Machinen I. 2nd ed. Basel, Germany: Verlag
Birkhäuser. 630 p.

Richter R. 1953. Elektrische Machinen II. 2nd ed. Basel, Germany:

Verlag Birkhäuser. 707 p.

Slemon G.R, Xian L. 1992. "Modeling and design optimization of

permanent magnet motors" Electric Machines and Power Systems,
vol. 20, 1992, p. 71-92.

Spooner E. 1992a "Direct-coupled, direct-connected, permanent-magnet

generators". Wind energy conversion 1992, Nottingham, U.K., 25-27
March 1992, Proceedings, p. 149-154.

Spooner E, Williamson A.C. 1992b. "The feasiabillity of direct-coupled

permanent-magnet generators for wind power applications". Symposium
on Power electronics Electrical drives Advanced electrical motors
(SPEEDAM), Positano, Italy, 19-21 May 1992, Proceedings, p. 105-112.

Spooner E, Williamson A.C. 1992c. "Permanent-magnet generators for

wind power applications". International Conference on Electrical
Machines (ICEM'92), Manchester, U.K., 15-17 September 1992,
Proceedings, p. 1048–1052.

Spooner E, Williamson A.C, Thompson L. 1994. "Direct-drive, grid-

connected, modular permanent-magnet generators". Wind Energy
Conversion 1994 (BWEA'94), Sterling, U.K., 15–17 June 1994, Proceedings,
p. 339–343.

Søndergaard L, Bindner H, 1995, "Undersøgelse af generatorkoncepter til

vindmøller - Direkt drevet generator til gearløs vindmølle" (in Danish).
Roskilde, Denmark: Forskningscenter Risø, Risø-R-801(DA), 52 p.

Thorborg K. 1988. Power Electronics. United Kingdom: Prentice Hall,

504 p.

Veltman A, Thijssen G, de Haan S. 1994. "Direct drive generators for

wind turbines". European Wind Energy Conference (EWEC'94)
Thessaloniki, Greece, 10-14 October 1994, Proceedings vol. I, p. 541-546.

Weh H, Hoffman H, Landrath J,Mosebach H, Poschadel J. 1988.

"Directly-driven permanent-magnet excited synchronous generator for
variable speed operation". European Wind Energy Conference (EWEC'88),
Herning, Denmark, 6-10 June 1988, Proceedings p. 566-572.

Westlake A.J.G, Bumby J.R, Spooner E, 1996. "Damping the power-angle

oscillations of a permanent-magnet synchronous generator with
particular reference to wind turbine applications". IEE Proc.-Electrical
Power Applications , Vol. 143, No. 3, p. 269-280.
Appendix A Magnetizing Inductance
In this appendix the magnetizing inductance of the armature winding is
derived. The total air gap flux generated by the armature currents is
included in the inductance calculations, while only the fundamental flux
density wave from the magnets is included in the torque calculations. The
reason is that no torque can be produced by flux-density harmonics since
the armature currents are assumed to be sinusoidal, but the flux-density
harmonics do increase the required reactive power consumed by the
generator.

The assumptions for the inductance calculations are that:

• the number of slots per pole and phase is one;
• each slot contains two conductors which are parallel
connected (Consequently, the total current in one slot is
equal to the phase current);
• the winding pitch is equal to the pole pitch;
• the flux in the air gap is assumed to cross the air gap
perpendicular to the rotor and stator surfaces;
• the permeabilities of the stator and rotor iron are assumed
to be infinite and the relative permeability of the magnets is
assumed to be 1.

In Figure A.1 the flux caused by a current in the R-phase is illustrated,

the two other phase currents are assumed to be zero. The self-inductance
of phase R can be expressed as
ΨRR
LRR = I (A.1)
R

where Ψ RR is the total flux linkage of phase R caused by the current IR .

The total flux linkage is the pole flux Φ RR times the number of pole pairs,
i.e.,

ΨRR = p ΦRR (A.2)

The air gap reluctance of a pole pitch is

δ ef + hm
ℜ= (A.3)
µ0 le τp

and since the magnetizing mmf of one pole is IR and the flux passes the
air gap twice, the pole flux (excluding the leakage flux) becomes
IR µ0 le τp
ΦRR = (A.4)
2 (δ ef + hm)
 R-phase pole flux S-phase pole flux
caused by IR (ΦRR) caused by IR (–1/3 ΦRR)

+R –T +S –R +T –S

δ +hm
τp

Figure A.1 The flux of the different phases caused by a current in the
R-phase.

Now the self-inductance, for p pole pairs, can be expressed as

τp
LRR = p µ0 le (A.5)
2 (δ ef + hm)

which makes it possible to identify the permeance coefficient λ m of

Equation (5.35). The mutual inductances between the phases will all be
equal, and can be calculated in a way similar to the self-inductance, i.e.,
ΨSR
M= I (A.6)
R

where Ψ SR is the flux linkage of the S-phase caused by the current in the
R-phase. Figure A.1 shows the pole flux linking to phase S to be minus
one third of the pole flux linking to phase R. Consequently, the mutual
inductance of the phases will be

M = –1/3 LRR (A.7)

The equivalent Y-phase inductance must include both the self-inductance
of one phase and the mutual inductance to the two other phases. The total
flux linkage in phase R caused by all the windings is

ΨR = LRR IR + M IS + M IT (A.8)
Since the neutral point of the Y-connected windings is not connected, the
sum of the three phase currents is always zero. Therefore,

IS + IT = –IR (A.9)
The flux linkage of phase R can with Equations (A.8) and (A.9) be written

ΨR = (LRR – M ) IR (A.10)
Using Equation (A.7), the equivalent Y-phase inductance can be
expressed as

Leq = 4/3 LRR (A.11)

Appendix B Thermal Model of the Generator
A thermal model of the generator is used to calculate the temperature of
the hottest part of the stator winding, which is the end windings, and the
magnet temperature. Only steady-state temperatures are calculated.
To make it easy to calculate the values of the thermal resistances in the
complete model, the thermal resistances are first derived for thermal
models of one slot pitch of the stator, one coil, one pole pitch of the rotor
and the end shields. The values of these thermal resistances can be
calculated by using definitions for one-, two- or three-dimensional heat
flow in rectangular elements. The thermal model of the generator is
derived by combining the detailed models for Q slots, Q coils, 2 p rotor
poles, and two end shields. Finally, the complete thermal model of the
generator is slightly simplified. All nodes which are not needed for the
calculation of the winding and magnet temperatures are eliminated and
series and parallel connected thermal resistances are replaced with the
total thermal resistance. The result is a thermal network model with
twelve nodes and eighteen thermal resistances.

Basic Theory
Three-dimensional heat flow can be approximately modelled by a lumped-
parameter thermal circuit (Perez and Kassakian, 1979). The generator is
divided into rectangular elements, like the one in Figure B.1, and they are
represented by the simplified network models in Figure B.2. The total
thermal resistances through the body in the x, y and z-directions are
lx ly lz
Rx = Ry = Rz = (B.1)
ly lz λ lx lz λ lx ly λ
where lx, ly and lz are the lengths of the body in x, y and z directions and λ

iy

ly TAv Pd λ
iz
ix
lz
lx

Figure B.1 A rectangular body with homogeneously distributed internal

losses P d , the heat conductivity λ and the average
temperature TAv .
 Ty2

1/2 Ry
TAv Pd T z2
-1/6 Ry
Rz
1/2

-1/6 Rx
z
R
6/
-1
T x1 1/2 Rx 1/2 Rx T x2

1/2 Ry
Rz
1/2
T z1 Ty1

Figure B.2 Lumped-parameter thermal models for three-dimensional

heat flow through the rectangular body.

is the thermal conductivity of the material. The average temperature of

the body is TAv and the power losses Pd are assumed to be homogeneously
distributed within the body. If there is no heat flow in one direction, the
corresponding thermal resistances are excluded (i.e., infinite thermal
resistance).

The generator outer surface of the stator core is cooled by air forced
through circumferential cooling channels. The temperature increase of
the cooling air is included in the model, as an equivalent thermal
resistance. The temperature increase of the cooling air depends on the
heat flow P c, the volumetric flow q vc, the density ρ c and the specific heat
capacitivity kthc of the cooling air. The temperature rise of the cooling air
is
Pc
∆θc = (B.2)
qvc ρc kthc

This extra temperature rise will occur in the parts of the stator close to an
outlet of the stator cooling channels. In the model, the cooling air at the
stator yoke is assumed to have this temperature rise. The model then
represents the warmest part of the stator. The magnet temperature is,
therefore, overestimated by about 0.5 ∆ θ c . The error of the magnet
temperature is in the order of 5˚C.
The temperature rise of the cooling air is included in the thermal model
by introducing the equivalent thermal resistance of the cooling duct
1
Req = (B.3)
qvc ρc kthc
which represents the heating of the cooling air. The total volumetric
cooling air flow depends on the number of cooling circuits and the flow in
each circuit. It is assumed that the cooling channel length τAir should be
2 m. The number of cooling circuits has not been restricted to an integer
number. Instead, it is defined as
π dse
NAir = (B.4)
τAir
The volumetric flow in each cooling channel is determined by the stator
length l, the height of the cooling channel hAir and the cooling air velocity
vAir as

qvAir = vAir l hAir (B.5)

The total volumetric cooling air flow is

qvc = NAir qvAir (B.6)

The parameters for the stator cooling have the following values

ρc = 1.1 kg/m3 (constant)

kthc = 1010 J/(kg K) (constant)
hAir = 100 mm (design variable)
vAir = 15 m/s (design variable)
τAir = 2 m (design variable)

The Detailed Model

The detailed models of a slot pitch, a rotor pole, a coil and the two end
shields are presented in Figures B.3 and B.4. Note that the thermal
models of the two figures are connected. All parts of the generator are
modelled by a one-, two- or three-dimensional version of the thermal
model in Figure B.2. The definitions of the thermal resistances in these
figures are found in Table B.1 and the values of the thermal constants
used are found in Table B.2. Some clarifications of the model and the
thermal resistances follow below.
The unconnected thermal conductors marked with I up to VII are
connected to the other unconnected thermal conductor with the same
number. Pa to Pf are losses in different parts of the generator.
 IV
R0

R1 R2

R3 R4
I I
R5 R5 R6 R6
R7 R8

R9 Pa Pb R10
R3 R4

R11

One stator slot pitch

R12 R13
II II
R14 R14 R15 R16 R16 R15
R17 R18 V

R19 Pc Pd R20
R12 R13

R11

R11

R12 R13
III III
R14 R14 R15 R16 R16 R15
R17 R18 VI

R12
R19 Pc Pd R20
R13

R11

R21

R22

R23

R24
One rotor pole

Pe
R24 R25

R26

R27

R28

VII

Figure B.3 The detailed model of the rotor and stator and the model of the
radial and circumferential heat flow in the coil in the slots
 Two end shields
IV
R29 R29

R29 R29

V
R30

R31 R31 R32

R32 Pf Pf
R33 R34 R33 R34

One coil
R35 R36 R37 R36 R35 R35 R36 R37 R36 R35

R38 R39 R39 R38 R38 R39 R39 R38

VI
R32 R30 R32

R31 R31

VII

Figure B.4 The detailed model of the axial heat flow in the coil in the slots
and the complete model of the end windings and the end
shields.

The heat flow in the iron core is assumed to be two-dimensional. No heat

flow in the axial direction is assumed because the thermal conductivity is
about 30 times lower in that direction than along the laminations.
The heat flow in the coils is three-dimensional. The model of the slot pitch
is connected to the model of a coil through the average temperatures
(points V and VI) of the upper and lower coil sides.
The heat transfer coefficient α 1 at the outer surface of the stator yoke is
estimated from data by Luke (1923). With a cooling air velocity of about
15 m/s, the value is assumed to be 60 W/(K m2 ). The stator yoke is
assumed to have cooling fins that increase the cooling surface to three
times its own outer surface.
Table B.1 Definitions of thermal resistances in the detailed model.

1 hs1 + hs2
R0 = R21 =
q vc ρ c k thc lu (0.5 τ + 0.5 bd) λFe
1 1
R1 = R22 =
3 l bd α1 l (τ – b s1) α 2
1 1 hm1
R2 = R23 = +
3 l bs α1 l bm α2 l bm λGRP
0.5 hys 0.5 hm
R3 = R24 =
lu bd λFe l bm λm
0.5 hys 1
R4 = R25 = – R24
lu bs λFe 3
0.5 bd hm0
R5 = R26 =
lu hys λFe l bm λglue
0.5 bs hyr
R6 = R27 =
lu hys λFe l τp λ Fe
1 1
R7 = – R5 R28 =
3 l τp α 5
1 1
R8 = – R6 R29 =
3 π (0.5 d + hs + hys)2 α3
1 1
R9 = – R3 R30 = – R31
3 3
1 0.5 l
R10 = – R4 R31 =
3 hCu bCu kCu λCu
hi 0.5 lb
R11 = R32 =
l bCu λi hCu bCu kCu λCu
0.5 (hCu + 2 hi) 1
R12 = R33 = – R32
lu bd λFe 3
0.5 hCu 1
R13 = R34 = – R36
l b Cu λ coil 3
0.5 bd 1
R14 = R35 = 2
lu (hCu + 2 hi) λFe lb bCu α4
hi 0.5 hCu
R15 = R36 =
l h Cu λ i lb bCu λCoil
0.5 bCu 1
R16 = R37 = – R39
l h Cu λ coil 3
1 1
R17 = – R14 R38 = 2
3 lb hCu α4
1 0.5 bCu
R18 = – R16 R39 =
3 lb hCu λCoil
1
R19 = – R12
3
1
R20 = – R13
3
The heat transfer coefficient at the tooth tip α2 is assumed to be lower than
at the stator yoke back, because the rotor surface velocity is less than
15 m/s. The air flow in the air gap is assumed to be turbulent because of
the rough rotor surface and, therefore, only the thermal resistances of the
convective heat transfer at the tooth tip and at magnet surfaces are
included in the model. The heat transfer coefficients at the end shields α3,
the end windings α 4 and the inner surface of the rotor yoke α 5 are all
assumed to be equal.

The rotor pole model is simple and is mainly included to show that the
temperature rise of the magnets should not be a problem. It includes
losses in the magnets and cooling through the magnet and rotor yoke to
the internal air of the generator. In addition, the thermal resistances of

Table B.2 The thermal constants used.

Heat transfer coefficient at the stator yoke back α1 = 60 W/(K m2)

Heat transfer coefficient in the air gap α2 = 40 W/(K m2)

Heat transfer coefficient at the end shields α3 = 25 W/(K m2)

Heat transfer coefficient at the end windings α4 = 25 W/(K m2)

Heat transfer coefficient at the rotor yoke back α5 = 25 W/(K m2)

Thermal conductivity of iron λFe = 38 W/(K m)

Thermal conductivity through the coil λCoil = 1.8 W/(K m)

Thermal conductivity of copper, along the coil λCu = 400 W/(K m)

Thermal conductivity of insulation λi = 0.2 W/(K m)

Thermal conductivity of NdFeB magnets λm = 9 W/(K m)

Thermal conductivity of the magnet glue λglue = 0.7 W/(K m)

Thermal conductivity of the GRP magnet λGRP = 0.2 W/(K m)

protection

Thickness of magnet glue hm0 = 0.1 mm

Thickness of magnet reinforcement hm1 = 0.5 mm

Coil insulation thickness hi = 1 mm

the magnet glue and the glass fibre reinforcement over the magnets are
included. The thickness of the magnet glue is h m0 and of the glass fibre
reinforcement h m1 .

The internal air in the generator is assumed to have a homogeneous

temperature except in the air gap. The cooling is also assumed to be
equally efficient at both end shields.

The value of the thermal resistance R21 is derived assuming that the tooth
tip is rectangular instead of trapezoidal. It is assumed that the end
windings overlap so that only half of their outer surface is used for
cooling. Therefore, R35 and R38 include the factor 2.

The Simplified Model

From the detailed models for the different parts described above, a
simplified thermal model of the complete generator can be derived. The
simplified model is derived by connecting Q parallel models for a stator
slot pitch, Q parallel models of a coil, 2 p parallel models of a rotor pole
and the model for the internal air and the two end shields.

The thermal model is simplified by using symmetry to reduce the number

of thermal resistances in the yoke, teeth, coil sides, end windings and end
shields. The network is simplified as much as possible while keeping only
the nodes that are necessary to model the temperature of the end
windings and magnets accurately. Figures B.5 and B.6 illustrate how the
simplified model is derived from the detailed models. The thermal
resistances of the simplified model are defined in terms of the thermal
resistances of the basic models and their definitions are shown in
Table B.3.

The symmetry between the two end windings and end shields is used to
reduce the model. Since the core losses are smaller than the copper losses
and the temperature rise inside the stator iron is not so large, the core
losses are moved in the thermal network to allow simplifications. The
losses P1, P2, P3 and P9 are all moved to the other side of R9, R10, R19 and
R19, respectively. By doing so, the number of nodes and thermal
resistances can be decreased but the temperature rise of the iron core will
be slightly overestimated. The same simplification is used for the magnet
losses, thus, R25 is neglected. Since the aim of the model is only to show
that the temperature in the magnets can be low enough, it is justified to
overestimate the temperature rise of the magnets slightly. The thermal
resistances ending at the slot wedge are excluded.
 X
R50

R51 R52
R53

P1 P2

R54 R55

R56
XI

Q stator slots
P3 P4 R57

R58
R59

R56

XII
P9 P10

R60
2 p rotor poles

P11

R61

XIII

Figure B.5 Simplified thermal network for the stator and rotor.
 Two end shields
X

R62

XI
R63

P7

Q coils
R64 R65

R63

XII

XIII

Figure B.6 Simplified thermal network for the end windings and end
shields.
Table B.3 Thermal resistances of the simplified model.

R50 = R0 2 R13 + 2 R11 + R20

R59 =
Q
R1 + R3
R51 =
Q R12 + R21 + R22
R60 = +
Q
R2 + R4
R52 =
Q R23 + R24
+
2p
R7 + R8 + R9 + R10 + 0.5(R5 + R6)
R53 =
Q R24 + R26 + R27 + R28
R61 =
2p
R3 + R12
R54 =
Q R62 = R29

R4 + R11 + R13 R30 + 0.5(R31 + R32)

R55 = R63 =
Q Q

R19 + R17 + R18 R64 = R64a//R64b (in parallel)

R56 = +
Q
R34 + 0.5(R36 + R35)
R64a = 0.5
R14 + R15 + R16 Q
+ 0.5
Q
R37 + 0.5(R39 + R38)
R64b = 0.5
R20 Q
R57 =
Q 0.5 R33
R65 =
2 R12 Q
R58 =
Q

Summary of the Thermal Model

The final thermal model is shown in Figure B.7 and the losses used in it
are given in Table B.4. The thermal model has twelve nodes, plus the
ambient temperature, and eighteen thermal resistances. The
temperature rise problem is formulated as a matrix equation. The vector
of temperature rises is evaluated by multiplying the loss vector by the
inverse of the thermal conductance matrix.

References:
Luke G. E. 1923. "The Cooling of Electric Machines". Transactions of the
AIEE 42, p. 636–652.
Perez I.J., Kassakian J.G., 1979. "A Stationary Thermal Model for
Smooth Air-gap Rotating Electric Machines". Electric Machines and
Electromechanics 3, 1979, p. 258-303.
 0 Ambient air
R50
R51 R52

1 2
R53 R62
R54 R55

3 R63
5
R56 4 R57 6 8
R58 R59 R65 7 R64
9
R56 10 R63
R60
R61
11

Figure B.7 The simplified thermal model.

Table B.4 The nodes and the losses of the simplified thermal model.

0: Stator cooling air (average temperature) P0 = 0

1: Temperature in the yoke above a tooth bd

P1 =
τ (PHyys + PFtys)
2: Temperature in the yoke above a slot bs
P2 =
τ (PHyys + PFtys)
3: Temperature in a tooth at the bottom coil side P3 = 0.5 (PHyd + PFtd)

4 : Temperature in a bottom coil in a slot l

P4 = 0.5 l + l PCu
b

5: Fictitious model temperature P5 = 0

6: Fictitious model temperature P6 = 0

7: Temperature in an end winding l

P7 = l +bl PCu
b

8: Temperature of the internal air P8 = 0

9: Temperature in a tooth at the top coil side P9 = P3 + Pad

10: Temperature in a top coil side in a slot P10 = P4

11: Temperature in the magnets P11 = PFtm

Appendix C Average Efficiencies
In this appendix, the average efficiencies of a direct grid-connected
induction generator and its gear are derived. In addition, the average
efficiency of the frequency converters for the proposed direct-driven
generators are derived.
The calculations are made in the same way as the average efficiency is
calculated for the direct-driven generators, with average loss factors
calculated with the wind speed probability density function. A medium
wind speed site is assumed, with an average wind speed of 6.8 m/s and a
capacity factor for the turbine power of 0.25.

Induction Generator Efficiency

The losses of a constant-speed induction generator can be divided into
three types with different average loss factors:
• No-load losses which are independent of the load.
(Includes stator core losses, friction and windage losses);
• Stator copper losses which are proportional to the square of
the stator current;
• Rotor copper losses which are proportional to the square of
the rotor current.
The additional losses are not neglected; they are included in the copper
losses.
To calculate the average loss factors, the rotor and stator currents have to
be expressed as functions of the wind speed. For this purpose, the
following approximations are made. The rotor current is assumed to be
proportional to the produced torque, and, thus, proportional to the active
power. The stator current has been divided into one reactive part, which is
assumed to be constant, and one active part which is assumed to be
proportional to the active power. The reactive current, magnetizing
current, is about 30 % of the rated current. With the function for the active
power given as a function of wind speed, Equation (3.29), the average loss
factors can be calculated. Note that the generator speed is constant for this
generator system.
For the medium wind speed site, the average loss factor for the no-load
losses is 0.77, for the stator copper losses 0.24, for the rotor copper losses
0.15 and the average factor for the active power is 0.25. The reason that the
average loss factor for the no-load losses is less than 1 is that the wind
energy converter is only operating 77 % of the year.
With these average loss factors and loss data for different machine sizes
(ABB, 1991), the average efficiencies can be calculated. The losses and the
average efficiencies are shown in Table C.1. Note that all the generators
are standard induction generators. Recently, high-slip generators have
been used in large wind energy converters, to limit turbine power peaks
and damp power oscillations. The high slip generators have rotor copper
losses of 2–3 % at rated load. Two-speed generators will also have lower
efficiency than assumed here, in the order of 0.5–1 % lower at rated load.

Table C.1 The different losses of the induction generators and their
efficiencies.
Rated No-load Stator Rotor Efficiency Average Average
power losses copper copper at rated losses efficiency
losses losses load
(kW) (p.u.) (p.u.) (p.u.) (%) (p.u.) (%)

30 0.035 0.020 0.020 92.5 0.0348 86.1

500 0.020 0.009 0.007 96.4 0.0186 92.6

3000 0.018 0.010 0.007 96.5 0.0173 93.1

Gear Efficiency
The gear losses can be divided into (Shipley, 1991):
• No-load losses which are independent of load (Include
friction and windage losses);
• Gear mesh losses, which are a constant percentage of the
active power.
Since the gear mesh losses are a constant percentage of the power, their
average factor will be equal to the average factor for the active power. The
average factor for the gear mesh losses is 0.25 and for the no-load losses it
is 0.77. In Table C.2, the losses and the average efficiencies of the different
gear sizes are shown.

Table C.2 The different gear losses and the gear efficiencies.
Rated No-load Gear mesh Efficiency at Average Average
power losses losses rated load losses efficiency
(kW) (p.u.) (p.u.) (%) (p.u.) (%)

30 0.010 0.020 97.0 0.0127 94.9

500 0.008 0.020 97.2 0.0112 95.5

3000 0.005 0.018 97.7 0.0084 96.7

 Frequency Converter Efficiency
The losses in the frequency converter are calculated in a slightly
simplified way, which overestimates them. The simplifications made are
the following: the anti-parallel diodes of the converter valves are assumed
to have the same voltage drop as the transistors; and the resistive losses,
being rather small, are included in the voltage-drop losses. The frequency
converters for the different rated powers are all assumed to have the same
efficiency.
The frequency converter losses are, here, divided into:
• No-load losses which are constant (Include power to the
cooling fans and the control circuits);
• Voltage drop losses which are proportional to the current
(Include the semiconductor voltage drop and the switching
losses).
For the rectifier, the current will be equal to the generator current and the
average loss factor for the voltage drop losses is 0.26. For the inverter
losses the average factor is different. The inverter is assumed to be
controlled to keep the power factor at 1 and the grid voltage is constant, 1
p.u. This means that the current will be proportional to the active power
and, consequently, the average loss factor for the voltage drop losses will
be equal to the average factor for the power, 0.25.
The average factor for the no-load losses, of both the inverter and the
rectifier, is 0.86. This average factor for the no-load losses differs from
that of the direct grid-connected system because a variable speed turbine
can operate in lower wind speeds and, therefore, it operates a larger part
of the year.
The voltage drop losses of the inverter and rectifier were assumed to be
0.02 p.u. for each unit and the total no-load losses 0.001 p.u. The efficiency
at rated load is, thus, 95.9 % and the average efficiency is 95.6 %.

References

ABB, 1991. Product Catalogue from ABB Motors, Sweden, 77 p.

Shipley E.E., 1991. "Loaded Gears in Action", In Dudley D.W. ed.,

Dudley's Gear Handbook, 2nd edition, New York, Townsend, 1991,
p. 12.1–12.39.

View publication stats