Design, Modelling and Control of Electrical Machines - With Applications to Iron-

powder Machines and Acoustic Noise

Martinez, David

2005

Link to publication

Citation for published version (APA):

Martinez, D. (2005). Design, Modelling and Control of Electrical Machines - With Applications to Iron-powder
Machines and Acoustic Noise. Department of Industrial Electrical Engineering and Automation, Lund Institute of
Technology.

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors
and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the
legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study
or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove
access to the work immediately and investigate your claim.

LUNDUNI
VERSI
TY

PO Box117
22100L und
+4646-2220000
Design, Modelling and Control
of Electrical Machines

With Applications to Iron-powder

Machines and Acoustic Noise

David Martı́nez Muñoz

Doctoral Dissertation in Industrial Electrical Engineering

Department of Industrial Electrical Engineering
and Automation
Department of
Industrial Electrical Engineering and Automation
Lund Institute of Technology
Lund University
P.O. Box 118
SE-221 00 LUND
SWEDEN

http://www.iea.lth.se

ISBN 91-88934-35-7
CODEN:LUTEDX/(TEIE-1043)/1-338/(2004)

c David Martı́nez Muñoz, 2004

Printed in Sweden by Media-Tryck
Lund University
Lund 2004

ii
 To my parents:

Manuel and Josefa

iii
iv
Abstract

This thesis consists of two parts, the first dealing with the design of iron-powder
synchronous machines, and the second with the analysis and prediction of the
acoustic noise in electrical machines.
In Part I, a 1.6 kW electrically magnetized claw-pole machine with mag-
netically conducting end-plates has been analyzed and a prototype tested. The
machine is built from soft magnetic composite material (SMC), also known as
iron-powder. The magnetic isotropy of SMC gives enormous flexibility in elec-
trical machine design, enabling new topologies exploiting three dimensional
flux paths. This is the main advantage compared to conventional machines
using laminations, where the flux is constrained into two dimensions. The
novelty of the machine presented lies in that the slip-rings in the rotor are no
longer needed, since the field coils are removed from the rotor and placed in
magnetically conducting end-plates attached to both sides of the stator. This
also improves the cooling capability of the copper losses from the field wind-
ing, allowing an increased electric loading. The rotor is of the claw-pole type,
and the end-plates close the magnetic circuit between the stator and the rotor.
The machine has been optimized using a magnetic equivalent circuit model
allowing rotation, where non-linearities have been included using an iterative
approach based on the linearisation of the BH curve. The traditional leakage
paths in claw-pole machines are modified because of the magnetically conduct-
ing end-plates, and alternatives are proposed to reduce them. The machine has
also been compared to two alternative topologies with electrical magnetization
and another with permanent magnets. The comparison has been carried out for
a similar temperature rise in the windings, and thermal models have been de-
veloped for every machine to determine their maximum electric loading. The
rotational and alternating components of the iron losses are calculated using
the finite element method (FEM). The results from the measurements indi-

v
cate that the average torque is 14% lower than predicted. This is probably due
to a leakage path between the end-plates through the shaft, which carries the
homopolar flux, and that was not considered in the predictions. The concept
of series magnetization has also been tested. This consists of feeding the field
winding directly from the rectified three phase armature currents at the neutral
point, therefore eliminating the need for the d.c. power source.
Electrical machines are finding application in new environments where lower
noise levels are demanded. The main focus of Part II is the measurement and
prediction of the noise emissions from induction motors using the vector con-
trol technique as well as the analysis of some structural changes to reduce these
emissions. A digital drive system has been developed for a 2.2 kW induction
motor, and its dynamic capabilities demonstrated for a wide range of the fre-
quency spectrum. This tool has been used for the experimental evaluation of
the noise emissions when the flux and/or the torque are modulated with high
frequency noise signals. The results showed that the noise emissions were higher
when the flux was modulated compared to the torque, although the differences
were considerably reduced when the machine was loaded. It was also observed
that the noise emissions were decreased importantly at load. Sound pressure
and sound intensity measurements have been conducted with the rotor station-
ary and rotating at low speed, showing that the most proper way to quantify
the noise emissions from electric machinery is to measure the sound power.
A method for the prediction of the noise emissions has been proposed, based
on the interactive use of commercial packages for mechanical, electromagnetic
and acoustic analyses based in the finite and boundary element methods. The
results show that the accuracy of the noise prediction depends on the proper
calculation of the modes of vibration in the structural analysis, as well as a suit-
able selection of the material damping. Skewing also needs to be modelled in
order to account for the rotor harmonics. A study has been conducted to assess
the effectiveness of introducing peripheral air gap layers around the stator core
to reduce the noise emissions. It was observed that the acoustic behaviour was
not improved, since the reduction of the stiffness in the outer part of the core
actually increased its sensitivity to the vibrations.

vi
Preface

This thesis constitutes the work I have carried out in my pursuit of a PhD in
Industrial Electrical Engineering at the Department of Industrial Electrical En-
gineering and Automation (IEA), Lund University, Sweden. During the course
of my PhD studies I also had the opportunity to spend almost one year as a
guest research fellow at the Acoustics and Vibration Unit (AVU), University
College, The University of New South Wales, at the Australian Defence Force
Academy, in Canberra, Australia.
The thesis addresses two topics in electrical machines. The first is concerned
with the design of iron-powder machines. The second is about modelling acous-
tic noise in electrical machines. The thesis is divided into two parts accordingly,
both written as a monograph. For the sake of clarity, each part has been orga-
nized in a series of chapters followed by the bibliography and the appendices
related to that part.
During the years of PhD studies, part of the material in the thesis has been
published at other occasions. The research on iron-powder machines has been
published in the following articles:

• Martı́nez-Muñoz, D. and Alaküla, M. (2003). “Comparison between a novel

claw-pole electrically magnetized synchronous machine without slip-rings and
a permanent magnet machine”. IEEE International Electrical Machines and
Drives Conference, IEMDC’03 Conf. Proc., Madison, WI, USA, June 2003,
pp. 1351-1356.

• Martı́nez-Muñoz, D. and Alaküla, M. (2004a). “Alternatives for leakage re-

duction in a novel claw-pole electrically magnetized synchronous machine”.
IEE International Conference on Power Electronics, Machines and Drives, PEMD
2004 Conf. Proc., Edinburgh, UK, April 2004, pp. 386-391.

vii
• Martı́nez-Muñoz, D. and Alaküla, M. (2004b). “A MEC network method
based on the BH curve linearisation: study of a claw-pole machine”. Inter-
national Conference on Electrical Machines, ICEM’04 Conf. Proc., Cracow,
Poland, Sept. 2004, 6 pp.
• Martı́nez-Muñoz, D., Reinap, A. and Alaküla, M. (2004a). “Comparison
between three iron-powder topologies of electrically magnetized synchronous
machines”. International Conference on Electrical Machines, ICEM’04 Conf.
Proc., Cracow, Poland, Sept. 2004, 6 pp.
• Martı́nez-Muñoz, D., Reinap, A. and Alaküla, M. (2004b). “Comparison
between four topologies of synchronous machines using SMC”, invited pa-
per, The UK Magnetics Society, Seminar on SMC and their use in electrical
machines, Newcastle, UK, Nov. 2004, 7 pp.

The thesis author has also co-authored the following publication in the
same field. The contribution made by the thesis author in this paper is minor:

• Reinap, A., Martı́nez-Muñoz, D. and Alaküla, M. (2004). “Iron loss calcula-

tion in a claw-pole structure”. IEEE Nordic Workshop on Power and Industrial
Electronics, NORPIE’04 Workshop Proc., Trondheim, Norway, June 2004, 5
pp.

The research on acoustic noise in electrical machines is reported in the fol-

lowing articles:

• Martı́nez-Muñoz, D., Pulle, D. W. J., Alaküla, M. and Weibull, H. (1999).

“A drive system for acoustical analysis”. IEE International Conference on Elec-
trical Machines and Drives, EMD’99 Conf. Proc., Canterbury, UK, Sept.
1999, pp. 147-150.
• Martı́nez-Muñoz, D. and Lai, J. C. S. (2003). “Acoustic noise prediction in a
vector controlled induction machine”. IEEE International Electrical Machines
and Drives Conference, IEMDC’03 Conf. Proc., Madison, WI, USA, June
2003, pp. 104-110.
• Martı́nez-Muñoz, D., Lai, J. C. S. and Pulle, D. W. J. (2003). “Acoustic noise
radiated from vector controlled induction motor drives”. European Conference
on Power Electronics and Applications, EPE’03 Conf. Proc., Toulouse, France,
Sept. 2003, 10 pp.

viii
• Martı́nez-Muñoz, D., Lai, J. C. S. and Alaküla, M. (2003). “A study of the
acoustical properties of stators with an outer-surface layer of air gaps”. Eu-
ropean Conference on Power Electronics and Applications, EPE’03 Conf. Proc.,
Toulouse, France, Sept. 2003, 10 pp.

Acknowledgements
Along the years of PhD studies I have had the privilege to work with different
supervisors. I would like to express my sincere gratitude to Prof. Mats Alaküla,
who welcomed me at the department already as an exchange student and gave
me the opportunity to continue with the PhD studies. Mats is very encouraging
and really enjoys exploring unconventional ideas. I am very grateful to Prof.
Joseph C. S. Lai (AVU) who was my advisor for the research on acoustic noise
and invited me to work at the AVU. Joseph is very friendly and helpful and
without his support the acoustic experiments and simulations would not have
been possible. I am also thankful to Prof. Duco W. J. Pulle for his advice in the
first stages of my PhD studies.
I would like to especially thank my colleague and friend Avo Reinap, who
has been an extraordinary working companion. Avo is always willing to discuss
any topic and to fill the white board in our office.
A number of people earn my gratitude for their support during the ex-
perimental work. Dr Thomas Rundqvist (ProEngCo Permedyn) and Ludvig
Wadsten (Motor & Bilelektra) have been very helpful with the construction of
the prototype machine in Part I. Andrew Roberts, Glenn Torr and Dr Andrew
Dombek at the AVU helped with practical aspects related to the acoustic noise
experiments in Part II. At the AVU I also would like to thank Antti Pappiniemi
for enjoyable discussions.
The structural models of the test motor included in Part II were created
by Dr Chong Wang during his PhD studies at the AVU, and their availability
is acknowledged. Dr Alex Michaelides at the VectorFields support in Oxford
has been a good help in the initial stages of the electromagnetic finite element
modelling. I would also like to thank my former colleague Tech. Lic. Svante
Andersson for interesting discussions.
I would like to express my gratitude to the staff at IEA. Even if all the people
at the department are helpful and friendly there are those that I owe a special
thank. Getachew Darge has provided very valuable help with practical issues for

ix
the experiments. Dr Morten Hemmingsson takes care of the LaTeX package in
UNIX, which I use for most of the writing. Dr Gunnar Lindstedt helps to keep
my PC healthy and also allowed me to “occupy” the computers in the lab with
my simulations. I am also grateful to Prof. Gustaf Olsson, who I believe played
an important role in my admission as a PhD student at the department.
Apart from my supervisors, a number of people earn my gratitude for proof
reading parts of the thesis. Part I has been reviewed by Prof. Sture Eriksson and
Tech. Lic. Svante Andersson. Part II has been reviewed by Prof. Lars Gertmar,
Dr Kelvin Maliti, and Tech. Lic. Anders Daneryd in a seminar. Their comments
have certainly contributed to improve the quality of the thesis.
I would also like to thank Dr Ola Stenlåås, Dr Rolf Egnell and Dr Olof Er-
landsson from the joined project for Part I. Together with the members of the
steering committee, Prof. Sture Eriksson (EME/KTH), Göran Masus (Finnve-
den Powertrain), Dr Göran Johansson (Volvo), and Tech. Lic. Joachim Lind-
ström (Volvo) they are acknowledged for fruitful comments and discussions.
The projects in Part I and Part II have been financially supported by STEM
(The Swedish Energy Agency), and NUTEK (The Swedish Business Develop-
ment Agency), respectively. This support is gratefully acknowledged.
Finally, I would like to thank my parents Manuel and Josefa, and my two
brothers Manuel and José Andrés, in Spain, for all their love, trust and sup-
port over the years. I would like to pay tribute to my parents, who always have
supported my projects and have encouraged me to travel. Their wisdom, under-
standing and care has enabled me and my brothers to attain graduate education.
This thesis is dedicated to them.

Lund, November 20, 2004,

David Martı́nez Muñoz

Contents

I Design of Iron-powder Synchronous Machines 1

1 Introduction 3
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . 3
1.2 Basic design principle . . . . . . . . . . . . . . . . . . . . . 7
1.3 A novel HEV application . . . . . . . . . . . . . . . . . . . 9
1.4 Objectives and contributions . . . . . . . . . . . . . . . . . 13
1.5 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Magnetic-Equivalent-Circuit model 17
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 MEC model of the EMSM . . . . . . . . . . . . . . . . . . 18
2.3 Mesh rotation . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Non-linear iterative method . . . . . . . . . . . . . . . . . . 23
2.5 FEM and MEC results . . . . . . . . . . . . . . . . . . . . . 27
2.6 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3 Finite Element model 35

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 FEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 FE model of the EMSM . . . . . . . . . . . . . . . . . . . . 37
3.4 Iron losses . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Iron losses in the EMSM . . . . . . . . . . . . . . . . . . . . 42
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 47

xi
4 Thermal model 49
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Electrical calculations . . . . . . . . . . . . . . . . . . . . . 50
4.3 Thermal network . . . . . . . . . . . . . . . . . . . . . . . 54
4.4 Heat transfer . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5 Alternatives for leakage reduction 67

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Leakage paths in the rotor . . . . . . . . . . . . . . . . . . . 67
5.3 Topologies for leakage reduction . . . . . . . . . . . . . . . . 68
5.4 Flux distribution results . . . . . . . . . . . . . . . . . . . . 69
5.5 Torque response . . . . . . . . . . . . . . . . . . . . . . . . 75
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6 Comparison with three alternative machines 81

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.2 EMSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.3 Outer rotor EMSM . . . . . . . . . . . . . . . . . . . . . . 84
6.4 Conventional EMSM . . . . . . . . . . . . . . . . . . . . . 93
6.5 Surface mounted PMSM . . . . . . . . . . . . . . . . . . . . 102
6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 110

7 Prototype and measurements 113

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.2 Construction of the prototype . . . . . . . . . . . . . . . . . 113
7.3 Adaptation of the thermal model . . . . . . . . . . . . . . . 121
7.4 Experimental set-up and instrumentation . . . . . . . . . . . 126
7.5 Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.6 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.7 Series Magnetization . . . . . . . . . . . . . . . . . . . . . . 140
7.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 144

8 Conclusions 147
8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . 149

xii
Bibliography 151

A Equations for the MEC network 157

B Parametrization of the machine 161

C Equations for the thermal network 165

D Dimensions of the alternative machines 167

II Acoustic Noise in Electrical Machines 171

9 Introduction 173
9.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.2 Objectives and contributions . . . . . . . . . . . . . . . . . 180
9.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

10 Control system 183

10.1 Introduction to Vector Control . . . . . . . . . . . . . . . . 183
10.2 Noise emissions in vector controlled drives . . . . . . . . . . . 186
10.3 Stator flux control . . . . . . . . . . . . . . . . . . . . . . . 189
10.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . 192
10.5 Musical drive test . . . . . . . . . . . . . . . . . . . . . . . 195
10.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 198

11 Acoustic measurements 201

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 201
11.2 Sound intensity measurement technique . . . . . . . . . . . . 201
11.3 Experimental set-up and instrumentation . . . . . . . . . . . 204
11.4 No load tests results and analysis . . . . . . . . . . . . . . . . 211
11.5 Load tests results and analysis . . . . . . . . . . . . . . . . . 218
11.6 Sound intensity measurements . . . . . . . . . . . . . . . . . 222
11.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 227

12 Analytical force computation 229

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 229
12.2 One phase winding distribution . . . . . . . . . . . . . . . . 229

xiii
 12.3 Three phase winding distribution . . . . . . . . . . . . . . . 236
12.4 Modelling the slots . . . . . . . . . . . . . . . . . . . . . . . 236
12.5 FEM validation . . . . . . . . . . . . . . . . . . . . . . . . 240
12.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 243

13 Acoustic noise prediction 245

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 245
13.2 Structural modal analysis . . . . . . . . . . . . . . . . . . . . 247
13.3 Electromagnetic analysis . . . . . . . . . . . . . . . . . . . . 255
13.4 Acoustic analysis . . . . . . . . . . . . . . . . . . . . . . . . 261
13.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 266

14 Acoustic properties of stators with a layer of air gaps 271

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 271
14.2 Basic model . . . . . . . . . . . . . . . . . . . . . . . . . . 272
14.3 Zig-zag gap model. . . . . . . . . . . . . . . . . . . . . . . . 275
14.4 Four gap model . . . . . . . . . . . . . . . . . . . . . . . . 281
14.5 Ideal gap model . . . . . . . . . . . . . . . . . . . . . . . . 286
14.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 289

15 Conclusions 293
15.1 Summary of results . . . . . . . . . . . . . . . . . . . . . . . 293
15.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . 295

Bibliography 297

E Current controllers 303

E.1 Controller for the direct current . . . . . . . . . . . . . . . . 303
E.2 Controller for the quadrature current . . . . . . . . . . . . . 304

F Acoustic terminology 309

G No load sound intensity measurements 313

H Load sound intensity measurements 319

xiv
 Part I

Design of Iron-powder
Synchronous Machines

1
Chapter 1

Introduction

1.1 Background and Motivation

There is a broad consensus among leading climate scientists that emissions from
human activities are enhancing the natural greenhouse effect, that this is already
having a discernible effect on climate, and that the ultimate effect will be much
larger. This fact was a conclusion from The Royal Commission on Environmen-
tal Pollution’s 22nd Report (2000), where it was stated that in the 20th century
the global mean temperature of the Earth’s surface rose by about 0.6o C, and it
was predicted that it would rise at a rate of up to 3o C by the end of the 21th
century. A direct consequence of the global warming is the raise in the sea level,
which was estimated to increase the number of people in the world affected by
coastal flooding from today’s 13 million per year to 94 million each year by the
2080’s, unless there were to be large-scale migration away from threatened areas.
Four fifths of the people affected would be in south and south-east Asia.
Carbon dioxide is an important greenhouse gas, and the most important in
terms of human impact on the atmosphere. Its concentration in the atmosphere
can vary naturally, and Figure 1.1 shows the variations in the carbon cycle that
have occurred over the last 400.000 years (upper curve) and how these related
to changes in the temperature (lower curve). During relatively warm periods
(similar to the climate over the last 10.000 years) the concentration of carbon
dioxide in the atmosphere was, as in the pre-industrial period, about 270-280
parts per million by volume (ppmv); during the coldest parts of glacial periods it
was significantly lower, about 180-190 ppmv. However, over the last 250 years,
as industrialization has taken place, the concentration has risen to about 370

3
4 Chapter 1. Introduction

ppmv. It is now increasing by 0.4% a year on average. Two-thirds of the current

enhancement in the greenhouse effect is estimated to be due to this increased
concentration of carbon dioxide. Nearly four-fifths of the extra carbon dioxide
entering the atmosphere since 1750 is estimated to have come from burning
fossil fuels.

year AD
1,000 1,500 2,000
380
360
340
320
300
280
260
240 global carbon dioxide
concentration
temperature ºC

2 220 ppmv
0
-2
local

-4
-6
-8

400,000 350,000 300,000 250,000 200,000 150,000 100,000 50,000

years before the present

Figure 1.1: Carbon dioxide concentration and temperature: evidence from ice
cores. (The Royal Commission on Environmental Pollution’s 22nd
Report, 2000)

The world energy supply is dominated by fossil fuels, which account for
about 80% of the total. Oil is the most important energy source, meeting 37%
of the demand, followed by coal at 22% and natural gas at 21%. Only in Swe-
den, with a population of 9 million people, the annual oil consumption (petrol
and diesel fuel) in the transport sector is equivalent to around 80 TWh, which
corresponds to more than 19% of the total final energy use in the country (The
Swedish Energy Agency, 2003). The transport sector in Sweden is responsible
for 40% of the total carbon dioxide emissions. In the rest of the world, as in
Sweden, the trend is increasing both in a short and long term perspective, and
it is therefore important to use fuels more effectively in vehicles to reduce their
consumption.
1.1. Background and Motivation 5

Fuel cell and hybrid vehicles

Technology advances both through improvements in existing technology and in
the form of completely new technologies. Of the latter, looking further ahead
than ten years, the automotive industry is pinning considerable hopes on fuel
cell vehicles (FCV). Fuel cells convert energy stored in chemical form in the fuel
directly into electricity and heat. The working principle is comparable to that
of a battery, with the difference that the fuel cell produces electricity and heat
as long as it is supplied with fuel. The electrical efficiency of fuel cells can be
as high as 60-70%. The waste product is the burnt fuel. Normally the fuel is
hydrogen, and mixed with the oxygen in the atmosphere produces water. The
operating temperature is too low to produce NOx. Therefore, apart from effi-
cient, fuel cells have the advantage of being non-polluting and silent. The draw-
backs of FCV are twofold: first, there is presently no refueling infrastructure;
second, there are substantial differences in cost, complexity and safety concerns
between a gas station and a liquefied hydrogen refueling station. Technological
improvements are needed to address these challenges, which leads to yet another
drawback what regards the cost of the technology, both in the vehicle itself and
in the fueling infrastructure. An alternative to FCV are ethanol-fueled vehicles
or even flexible fuel vehicles (FFV), which use different fuels simultaneously,
i.e. ethanol and petrol. Although these vehicles use more or less conventional
combustion engine technology, they share the problem with fuel cells in that
there is no refueling infrastructure for ethanol nowadays.
The alternative that is likely to achieve a commercial breakthrough in the
next ten years is the hybrid electric vehicle (HEV). Hybrid vehicles have two
combined systems, i.e. both an electric motor and a combustion engine. A
definition of the concept of hybrid vehicle from one of the biggest automotive
manufacturers says: “A Hybrid vehicle is a conventionally fueled and operated
vehicle that has been equipped with a power train capable of implementing at
least the first three of the following four hybrid functions:

• Engine shut-down when power demand is zero or negative.

• Engine down-size for improved thermal efficiency.

• Regenerative braking for recovery and re-use of braking energy.

• Engine-off propulsion at low power (when the engine is inefficient).”

6 Chapter 1. Introduction

The most important feature of a hybrid vehicle is the possibility of choosing

a working point for the internal combustion engine (ICE) independent, or al-
most independent, of the drivers choice of driving power. The balance between
the installed power of the ICE and the electrical machines determines the ef-
fectiveness in reduced fuel consumption and emissions, keeping similar driving
characteristics as a traditional ICE vehicle of the same type.
There are different types of hybrid vehicles, namely mild hybrids, power hy-
brids and energy hybrids. Mild hybrids have only 5-15 kW installed electric
power, mainly intended for energy generation for electrically driven loads on-
board, and for driving at low speeds. Power hybrids have 20-40 kW installed
electric power and can therefore contribute significantly to acceleration at any
speed, but they still have too low battery capacity (a few km pure electric driv-
ing). Finally, energy hybrids have 70-100 kW installed electric power and have
therefore energy for long distance driving. The battery is still the weakest point
in hybrid vehicles, with low energy density and high price, and therefore the
trend to minimize its size. This is the reason for the car industry to focus on
mild and power hybrids with relatively small batteries.
The trend in the automotive industry is to use permanent magnet syn-
chronous machines (PMSM) for traction in hybrid vehicles (Kazuaki et al.,
2000), due to their high torque to weight ratio. However, the price and the
availability of magnet material is an issue that needs to be considered, especially
if these machines have to satisfy mass production demands from the car in-
dustry in the future. Regarding thermal performance, operation of permanent
magnet machines in the field weakening region requires better cooling since
the electrical loading in the stator winding is increased. There are also con-
cerns about using an additional current to counteract the magnetic field at high
speeds. If this current disappears, due to a control failure or power electronics
malfunction, the sudden increase in the back e.m.f. could destroy the power
electronics and cause a breakdown in the machine. This risk of over voltage is
prevented in an electrically magnetized synchronous machine (EMSM), where
the field is provided by a separate coil and the current is in fact decreased in or-
der to weaken the field. A lower current is also advantageous from the cooling
point of view. However, there are drawbacks using electrical magnetization in
a conventional synchronous machine, such as the need of slip-rings to feed the
current into the rotor, and the extra power electronics to control this current.
This part of the thesis focuses on the study of an alternative design to eliminate
the slip-rings.
1.2. Basic design principle 7

1.2 Basic design principle

The structure of the electrically magnetized synchronous machine analyzed

more in detail in this part of the thesis is shown in Figure 1.2(a). The nov-
elty of the machine lies in that the slip-rings are removed by placing the field
coils in magnetically conducting end-plates attached to both sides of the stator.
The iron plates close the magnetic circuit between the stator and the claw-pole
rotor. The field coils are wound around the salient part of these plates, re-
maining therefore stationary. The stator coils are wound around a single tooth,
forming a three phase concentrated winding. This has advantages from the
manufacturing point of view since the coils can be pressed in order to increase
the filling factor (Jack et al., 2000). The teeth have rounded edges in order to
decrease the length of the end-windings and improve the heat transfer from the
coils through a better fit. A simplified plot of the machine with the flux flow
is shown in Figure 1.2(b). The magnetizing flux flows from one of the rotor
claw-poles towards the stator and then back to the same claw-pole through the
corresponding iron plate, as indicated in the figure. The flux crosses two air-
gaps in a complete loop, the radial between the rotor claw-pole and the stator
teeth, and the axial between the iron plate and the rotor claw-pole. When the
a.c. coils are excited, a flux flows between the teeth in the tangential direction,
and perpendicular to the magnetization path in Figure 1.2(b). Therefore, the
current loading imposes a three-dimensional flux flow, and the machine is an
optimal application for iron-powder technology.

Soft Magnetic Composites

Soft magnetic composite (SMC) materials consist of iron powder particles where
the surface of every particle is insulated using a continuous oxide layer (Skarrie,
2001). The particles are compacted, together with a lubricant and possibly a
binder, at high pressure into a bulk material. The lubricant eases compaction
and the ejection of the components after compaction, while the binder increases
the strength of the material. The lubricant and the binder also provide the in-
sulation between particles. During compaction, internal stresses are generated
in the material, which can be relieved with a heat treatment process (curing).
This treatment also increases the strength of the material. Conventional uniax-
ial pressing at ambient temperatures with lubricant additives is readily available
for mass production.
8 Chapter 1. Introduction

Armature Stator Field

winding winding
X O
X X

Rotor
Z

Flux path Flux path

south pole north pole
O O
O X

(a) 3-D view. (b) Axial plot.

Figure 1.2: Structure of the EMSM.

Iron powder materials can be considered as magnetically, thermally and me-

chanically isotropic in their behaviour. Due to the small insulated powder par-
ticles the properties are uniform in the bulk material. The magnetic isotropy
gives enormous flexibility in electrical machine design, enabling new topologies
exploiting 3-D flux paths rather than being constrained to the 2-D flux flow in
laminated machines. The main drawbacks of SMC compared to laminations
are the smaller permeability (maximum 800 vs. >3000) and the higher iron
losses at practical frequencies. This is illustrated in Figure 1.3, which shows a
plot with the BH curve for a typical laminated material and for the iron pow-
der used in this thesis, SomaloyT M 500 + 0.6% LB1. There are also practical
limits on shapes that can be pressed, e.g. 6:1 aspect ratio (Jack, 2003). Simply
replacing an existing design using laminations by SMC is bound to be worse in
performance and probably more expensive. In general, SMC materials are more
conveniently used in machines with separate excitation, since the demands on
the permeability of the magnetic material is lower compared to machines with-
out separate field provision. Apart from building complex 3-D structures, such
as claw-poles, SMC can also be used to improve the tooth shapes, decreasing
the amount of copper by decreasing the length of the end-windings. The bet-
1.3. A novel HEV application 9

ter fitting between the coil and the teeth allows increased copper filling factor
and heat transfer to the core, which can be further increased by pressing the
coils. The smoother tooth surface also makes possible to reduce the slot wall
insulation. All this allows to change the shape of the machine to improve the
overall product, e.g. smaller and easier to make. The material can also be tai-
lored for a specific application by changing the composition of the powder and
adapting the fabrication process. Looking at environmental aspects the material
waste can be minimized because of the pressing. The brittleness of the material
(strength around 50-100 MPa TRS) simplifies the separation of the core from
the winding when recycling.

2 .5

1 .5
B [T ]

0 .5
D K 7 0
S o m a lo y 5 0 0
0
0 1 e4 2 e4 3 e4 4 e4
H [A /m ]

Figure 1.3: Comparison between the BH curve in laminated material (DK70)

and SMC (Somaloy 500 + 0.6% LB1).

1.3 A novel HEV application

Today’s ICE descends from the end of the 19th century. Since then, its devel-
opment has been characterized by stepwise improvements and increasing fine-
tuning of the basic principle. A subsystem that during this long evolution has
been kept more or less unchanged is the crankshaft movement. The reason for
this is that this way of controlling the movement of the pistons has been sim-
ply the best. For direct driven vehicles, the transmission of the forward and
backwards movement of the pistons to a rotating shaft has been a simple and
natural mechanism. When the ICE is used as a primary energy converter, i.e.
10 Chapter 1. Introduction

fuel into electricity in a hybrid system, it is no longer obvious that the classical
crankshaft movement and flywheel are the best alternatives. There might be
other waveforms that give better engine performance (Blarigan et. al, 1998).
The basic thought behind the novel HEV application was to analyze what
improvements could be done in the efficiency and emissions in todays vehicles
if the piston movement could be controlled freely by electrical means. The
efficiency of the combustion process and the emissions are affected by the pres-
sure and the temperature profiles in the combustion chamber, which in turn
are determined by the trajectory that the piston describes. A free piston move-
ment can be realized in practice for example by using the electro-mechanical
system presented in Figure 1.4(a). The piston is connected to two plates, one of
them mounted on the shaft of the electrical machine and the other represents
the crankshaft itself. The piston can move freely along the cylinder by rotating
both shafts independently in any direction at any speed. An example is shown
in Figure 1.4(b), where the stroke is plotted as a function of the crankshaft angle
(CAD). The wave referred to as the ‘equal speed’ curve corresponds to a rota-
tion of both shafts in the same direction and with the same speed. This gives a
waveform similar to the one obtained from a conventional combustion engine.
The wave referred to as the ‘double speed’ curve is obtained by rotating both
shafts in the same direction but one with double the speed as the other. It is
thus illustrated how the degrees of freedom in the ICE can be increased with
this configuration.

Alvar motor
In order to explore the idea of the free piston movement, it is cost-effective to
try the idea using a known mechanical solution, rather than constructing a com-
pletely new system. The Alvar motor shown in Figure 1.5 was found very appro-
priate. This motor is a Swedish invention of Alvar Gustavsson, from Skärblacka,
Östergötland. The idea behind this combustion engine was to achieve a vari-
able compression ratio, which is a way of increasing the efficiency in the Otto
motor at partial load. The main difference compared to a conventional ICE lies
in the cylinder head, where there is room for an extra crankcase, crankshaft and
cylinders. These cylinders are connected to the combustion chamber, and the
space required is obtained compromising in the size and location of the valves.
Typically, the top cylinder has half the bore and stroke as the base engine. The
compression volume in the motor can be changed by adjusting the synchroniza-
1.3. A novel HEV application 11

(a) Electro-mechanical system (b) Comparison of piston movements

Figure 1.4: Alternative piston movement.

tion between the two crankshafts, which in turn affects the compression ratio.
The unique feature in the Alvar motor is that the extra crankshaft (secondary
shaft) rotates with half the speed than the primary shaft, which allows to control
the emissions and provides a higher expansion and compression ratio. Previous
studies have demonstrated the potential of the Alvar motor as a machine with
variable compression ratio (Stewart, 1997), (Erlandsson, 1998), (Wong et. al,
1998), (Erlandsson et. al, 1998).
The secondary shaft of the Alvar motor can be coupled to an electrical
machine, which can be controlled to achieve the desired piston movement, as
it was done with the system in Figure 1.4(a). The Alvar motor available was
based in a Volvo B5254 FS motor. Four of the five cylinders in the engine block
were deactivated, using only one for the experiments. The displaced volume is
around 0.5 liters per cylinder, and the compression could be changed from 6:1
to 15:1. The study of the optimized combustion cycle and the experiments were
carried out by Stenlåås (2004). Simulations were performed in order to find the
torque requirements for the electrical machine to adapt the piston cycle, and the
results are shown in Figure 1.6. This figure shows the torque as a function of a
12 Chapter 1. Introduction

Secondary shaft
Secondary piston

Primary piston

Transmission
Primary shaft
(Crankshaft)

Figure 1.5: Alvar motor plot.

normalized mechanical angle for one revolution of the electrical machine, and
a speed of 500 r.p.m. It can be observed that the peak torque is quite high, 108
Nm, while the continuous torque calculated using (1.1) is much lower, 24.5
Nm. These calculations assumed that the inertia of the electrical machine was
0.6·10−3 kg·m2 . It is important to have a low inertia in the secondary shaft and
in the electric motor to allow quick changes of the volume of the combustion
chamber during the short period when the combustion takes place.


 t
 cycle
 1
Trms = T 2 (t)dt (1.1)
tcycle
0

It was realized that the requirements for the Alvar motor application could
not be fulfilled with the electrical machine presented in section 1.2. The prin-
ciple of having magnetically conducting end-plates implies that the machine
should be short, in order to minimize the magnetization path and thus the leak-
age. In a short machine, the rotor diameter needs to be high in order to increase
the torque output, which also increases its inertia. It was therefore decided that
a commercial servo motor should be used instead to implement the modified
1.4. Objectives and contributions 13

1 2 0
p eak
c o n tin u o u s
m e c h a n ic a l to rq u e [N m ] 1 0 0

8 0

6 0

4 0

2 0

− 2 0
0 9 0 1 8 0 2 7 0 3 6 0
m e c h a n ic a l a n g le [d e g ]

Figure 1.6: Torque requirements for the Alvar application.

combustion cycle using the Alvar motor principle. The study and development
of the electrical machine presented in this thesis was carried out independently,
and the machine was down-scaled so that it could be easily tested with the avail-
able equipment and laboratory facilities.
The commercial machine used for the Alvar application had an inertia of
2.2·10−3 kg·m2 , and the total inertia including the secondary shaft and other
parts became 4.2·10−3 kg·m2 , almost double the value used in the simulations.
This higher inertia limited the dynamic characteristics, i.e. how fast the com-
pression or volume ratio can be changed during combustion. Nevertheless,
experimental and simulation results showed that the possibility of affecting the
emissions by adjusting the piston movement are limited, without compromising
the efficiency, and not cost-effective. The reader is referred to (Stenlåås, 2004)
for a detailed and extensive description of the theory, simulations, experiments
and conclusions of the study.

1.4 Objectives and contributions

The main objective of this part of the thesis is to explore the possibilities of using
SMC to modify the conventional design of electrically magnetized synchronous
machines by removing the coils from the rotor, and therefore also the slip-rings.
14 Chapter 1. Introduction

A prototype has been built based on this idea and tested.

A magnetic equivalent circuit (MEC) model has been developed for the
optimization of the prototype design, and it allows backwards and forward ro-
tation of the rotor mesh. Non-linearities have been included in the model using
an iterative approach based on the linearisation of the BH curve, which proves
to be very fast and accurate.
The traditional leakage paths in claw-pole machines are modified in the
prototype design because of the magnetically conducting end-plates. A sensitiv-
ity study has been carried out in FEM to assess how different permanent magnet
materials can be used to minimize the leakage from the rotor claw-poles. The
study considers machines with 12, 16, 20 and 24 poles.
The prototype design has been compared by simulation to three alterna-
tive machines, an outer rotor EMSM, a conventional EMSM, and a surface
mounted PMSM. The comparison has been carried out for the same tempera-
ture rise in the windings, and thermal models were developed for each machine
to determine the maximum current loading. The performance of the machines
was evaluated using the 3-D computational finite element method (FEM). This
method has also been used to calculate the iron losses in the machines using a
detailed formulation that includes both alternating and rotational field compo-
nents.
Finally, an idea consisting of feeding the field winding from the rectified
three phase armature currents at the neutral point (series magnetization) has
been tested with the prototype machine. In this way the d.c. power source for
the field winding can be removed.

1.5 Outline
The material in Part I is divided in eight chapters. In the first Chapter the
structure of the machine has been described, as well as the application where
the machine was initially meant to be used. The rest of the chapters are focused
on different aspects related to the design, modelling and testing of the electrical
machine.
In Chapter 2, the MEC model of the machine is described, together with
the method to include non-linearities in the model. This model is used to
optimize the machine and to assess the shape of the torque response. The results
are validated with FEM calculations.
In Chapter 3, the FEM model is presented. The calculation of the iron
1.5. Outline 15

losses in FEM considering alternating and rotational field components is also

described. The losses are calculated at load and no load in the stator and in the
rotor as a function of the harmonic number.
In Chapter 4, the thermal model is presented, together with the electrical
calculations. Heat transfer theory is used to calculate the heat transfer coeffi-
cient in the water-cooled machine, as a function of the properties of the flow
and the geometry.
In Chapter 5, alternatives to reduce the high leakage in the machine are
studied. A sensitivity analysis is carried out for machines with different number
of poles.
In Chapter 6, three alternative machine topologies are studied. Basically,
a similar analysis to that described in Chapter 3 and Chapter 4 is now carried
out with the new topologies, and their performance is compared to the original
machine by simulation.
In Chapter 7, the construction of the prototype is described. Measurements
are carried out in order to assess the characteristics of the machine and validate
the electromagnetic and thermal models. The concept of series magnetization
is also tested.
Finally, Chapter 8 presents a summary of the most important results and
conclusions of this part of the thesis, together with some suggestions for future
work in this area.
16 Chapter 1. Introduction
Chapter 2

Magnetic-Equivalent-Circuit
model

2.1 Introduction

One of the numerical methods employed for electrical machine analysis is the
magnetic equivalent circuit (MEC) approach. This method is well described by
Ostović (1989). Basically, the complex magnetic circuit of the machine is trans-
formed into a simple, resistive electrical network, which is solved using electric
circuit theory. The fewer number of elements compared to the finite element
method (FEM) decreases considerably the computation time, although the ac-
curacy of the solution is also compromised. For highly saturated devices, the
number of elements in the MEC model may need to be increased (Moallem
and Dawson, 1998). The advances in computing technology during the last
decade have contributed to extend the use of FEM for the analysis of electro-
magnetic devices, specially what regards two-dimensional problems. However,
the CPU time needed to solve three-dimensional problems is still prohibitive,
which jeopardizes the use of FEM for 3D optimization (Ostović et al., 1999).
In this chapter, a 3D MEC model has been developed for the electrically
magnetized machine presented in Chapter 1. This model is used for the op-
timization of the machine using a non-linear iterative method based on the
linearisation of the BH curve. This method allows very fast convergence. The
MEC model has been extended to accomplish forward and backward rotation,
which is used for the assessment of the torque response. Results compared to
FEM simulations demonstrate the feasibility of the MEC model for predicting

17
18 Chapter 2. Magnetic-Equivalent-Circuit model

the performance of the machine.

Due to the initial specifications on the electrical machine for the Alvar mo-
tor application, the study and the optimization was carried out for a machine
with an outer diameter of 400 mm and an axial length of 152 mm. However,
the final design that is analyzed in the next chapters was down-scaled to half the
size in the axial and radial directions, maintaining exactly the same proportions.

2.2 MEC model of the EMSM

The MEC model of the machine is shown in Figure 2.1. The machine has 12
poles and 18 teeth, and the MEC model represents one pole pair at the position
of maximum torque along the negative pole period. Iron and air reluctances
are coloured in grey and white respectively. Each reluctance is identified with a
number which represents the main mesh that the reluctance belongs to. There
are three main meshes, associated with the three stator teeth. The reluctances
linking the main meshes are identified with their respective numbers. The fol-
lowing iron reluctances were considered: the tip of the stator tooth (Rsc), the
body of the stator tooth (Rps), the yoke in the axial direction (Ryx) and in
the tangential direction (Ryp), the end-plates at the level of the stator teeth
(Rss) and at the level of the rotor (Rsra, Rsrb), the rotor plates attached to the
claw-poles (Rcra1), and the claw-poles (Rcra2). There are four airgap reluc-
tances (Rgr), and ‘Rgr1’ and ‘Rgr4’ are connected together since the first stator
tooth is located exactly between the two claw-poles. They also have double the
value than the rest of the airgap reluctances. The leakage reluctances are iden-
tified with the symbol ‘σ’. The following leakage reluctances were considered:
between the stator tooth tips (Rcsσ ), between the stator tooth tips and the end-
plates (Rsfσ ), between the sides of the claw-poles (Rcrσ ), between the tips of the
claw-poles and the end-plates (Rsrσ ), and the reluctance for the zig-zag leakage
through the claw-poles (Rlea). The equations used to calculate the reluctances
are included in Appendix A. The m.m.f. of the armature coils wound around
the stator teeth is referred to as ‘Fph’. They contribute to drive the flux through
the four meshes adjacent to one stator tooth, and therefore a source with one
quarter of the total m.m.f. in the corresponding coil was placed in each one
of these meshes. The m.m.f. of the field coils wound around the end-plates is
referred to as ‘Fex’. They contribute to drive the magnetizing flux in the axial
direction and they are located between the axial core and end-plate reluctances.
 Stator tooth 3 Stator tooth 2 Stator tooth 1
        
Fph3/4 Fph3/4 Ryp23 Fph2/4 Fph2/4 Ryp12 Fph1/4 Fph1/4 Ryp13
Fex Fex Fex
Ryx3 Ryx2 Ryx1
Fph3/4Fph3/4 Fph2/4Fph2/4 Fph1/4Fph1/4
FexRyx3 FexRyx2 FexRyx1
2.2. MEC model of the EMSM

Rss3 Rss2 Rss1

Rps3 Rps2 Rps1
Rss3 Rss2 Rss1
Rsc3 Rsc2 Rsc1
Rcs23σ Rcs12σ Rcs13σ
Rsf3σ Rsf2σ Rsf1σ
Rsf3σ Rsr3a Rsf2σ Rsr2a Rsf1σ Rsr1a
Rgr3 Rgr2 Rgr1 Rgr4
Rsr3a Rsr2a Rsr1a
Rsr3b Rsr4b
Rge3 Rsr2b Rsr1b Rge4
Rcr3a1
Rsr3σRcr3a2 Rcr14σ Rcr4a1
Rcr23σ Rcr2a2Rsr2σ Rlea12 Rcr1a2Rsr1σ Rsr4σRcr4a2 Rlea13
Rcr2a1
Rge2 Rcr1a1
Rge1
Rsr3b Rsr2b Rsr1b Rsr4b
        

Figure 2.1: MEC model of the EMSM.

Rotor claw pole 1 Rotor claw pole 2 Rotor claw pole 1
19
20 Chapter 2. Magnetic-Equivalent-Circuit model

2.3 Mesh rotation

The dynamic characteristics of the machine can be studied by rotating the rotor
mesh with respect to the stator. When the rotor is rotated, the radial and axial
airgap reluctances, together with the axial leakage paths between the claw-poles
and the end-plates, are modified at each step. The position of maximum torque
along the negative pole represented in Figure 2.1 will be referred to as position
zero. From this position, it was desired that the rotor could rotate 90 electrical
degrees forward and backward in steps of 15 electrical degrees. Forward rotation
will be linked to counter-clockwise rotation, i.e. from right to left in Figure 2.1,
and the opposite is true for backward rotation. The arrangement used for for-
ward rotation is shown in Figure 2.2, and for backward rotation in Figure 2.3.
The rotor mesh below the first stator tooth has been modified grouping together
‘Rsr1b’ and ‘Rsr4b’ from Figure 2.1 into ‘Rsr1b’ in Figures 2.2 and 2.3.
In general, the principle for rotating the rotor mesh is that each one of the
radial and axial air reluctances linking the stator and the rotor are by-passed
by another reluctance from the same origin in the stator side but ending at the
point in the rotor mesh which approaches with the rotation. The original re-
luctances linking the stator and the rotor at position zero will be referred to as
direct reluctances and their value as nominal reluctance value. The new reluc-
tances added to accomplish rotation will be referred to as by-pass reluctances.
The by-pass reluctances for the radial airgap are ‘Rgr21’, ‘Rgr32’, and ‘Rgr13’
for forward rotation, and the same names but swapping the numbers for back-
ward rotation. The rest of the new reluctances above the rotor mesh are by-pass
reluctances for the axial airgaps, including leakage. These reluctances have the
same name as the direct reluctances that they by-pass, followed by a number
which indicates the part of the stator mesh (first number) that is linked to the
part of the rotor mesh (second number) that is approaching.
For convenience, it is important to note that each stator tooth will see ex-
actly the same rotor part as the adjacent tooth after eight steps in any direction.
This ratio is easily obtained from the step angle and considering that the stator
is divided in three parts, each one associated with one tooth, spanning 120 elec-
trical degrees. Therefore, the direct and by-pass reluctances will be increased
and decreased respectively by one eighth of the nominal reluctance value after
each rotation step. At position zero, the direct reluctances will have the nomi-
nal value and the by-pass reluctances will be infinite. From the first step to the
seventh, the direct reluctances will be increased from 8/7 to 8/1 the nominal
 2.3. Mesh rotation

Rsr32σ Rge24
Rge13
CC AA AA
Rgr32 Rgr21
Rgr13 Rgr3 Rgr2 Rgr1 Rgr4
CC
BB Rge21
Rge32
Rsr13σ BB

Rsr3b
Rge3 Rsr2b Rsr1b Rge4
Rcr3a1
Rsr3σRcr3a2 Rcr4a1
Rcr23σ Rcr2a2Rsr2σ Rlea12 Rcr1a2Rsr1σ Rcr14σ Rsr4σRcr4a2 Rlea13
Rcr2a1
Rge2 Rcr1a1
Rge1
Rsr3b Rsr2b Rsr1b

Figure 2.2: Arrangement for forward rotation.

21
 22

Rge23
Rge34
AA
AA
Rgr31 Rgr23 Rgr12 Rsr12σ
Rgr3 Rgr2 Rgr1 Rgr4
Rge31 CC
CC Rsr23σ Rge12 BB
BB
Rsr3b
Rge3 Rsr2b Rsr1b Rge4
Rcr3a1
Rsr3σRcr3a2 Rcr14σ Rcr4a1
Rcr23σ Rcr2a2Rsr2σ Rlea12 Rcr1a2Rsr1σ Rsr4σRcr4a2 Rlea13
Rcr2a1
Rge2 Rcr1a1
Rge1
Rsr3b Rsr2b Rsr1b

Figure 2.3: Arrangement for backward rotation.

Chapter 2. Magnetic-Equivalent-Circuit model
2.4. Non-linear iterative method 23

value, and the by-pass reluctances will be decreased from 8/1 to 8/7 the nomi-
nal value. After eight steps, the direct reluctances will be infinite and the by-pass
reluctances equal to the nominal values.

2.4 Non-linear iterative method

The flux density in the iron reluctances is a non-linear function of the field
strength. The BH curve of the composite material (or any iron material) can
be divided in a series of linear intervals, as shown in Figure 2.4(a). Each one
of these intervals is linearly extrapolated until they intersect with the B-axis at
a certain level, depending on their slope. The intersection level and the slope
will be referred to as B0f e and µf e respectively. For simplicity, let us assume a
magnetic circuit with a yoke and an airgap, excited with a magnet and a coil, as
shown in Figure 2.4(b). The iron, the magnet and the airgap in this structure
are considered to form a single reluctance element each. The flux density in
the iron at any interval can be expressed as in (2.1). The flux density in the
permanent magnet is given by (2.2). Assuming a linear characteristic for the
magnet, B0pm represents the remanence of the material. Otherwise, the BH
curve of the magnet could be linearised in a similar way as it is done for the
iron. The flux density in the air is given by (2.3). These flux densities can also
be expressed as a function of the flux through their respective area, as in (2.4).

Bf e = B0f e + µ0 · µf e · Hf e (2.1)
Bpm = B0pm + µ0 · µpm · Hpm (2.2)
Bδ = µδ · Hδ (2.3)

φf e φpm φδ
Bf e = Bpm = Bδ = (2.4)
Af e Apm Aδ

Applying Ampere’s law to the circuit yields (2.5). Including Hf e from (2.1),
Hpm from (2.2), and Hδ from (2.3) into (2.5), and using the equivalence from
(2.4) yields (2.6). Assuming that there is no leakage flux, then the flux flowing
through the iron, the permanent magnet and the airgap is the same, denoted by
φ. No fringing flux is assumed in the airgap. Equation (2.6) can be re-written
as in (2.7). Leaving φ alone yields (2.8), where Rf e , Rpm and Rδ are the iron,
magnet and air reluctances respectively, given in (2.9). Equation (2.8) shows
that both the iron and the permanent magnet have an ‘internal m.m.f.’ source.
24 Chapter 2. Magnetic-Equivalent-Circuit model

2
Af e , Lf e , µf e

1 .5
Apm , Lpm , µpm
[T ]

1
Aδ , Lδ , µδ
B

0 .5
N ·i

0
0 1e4 2e4 3e4 4e4 5e4 6e4
H [A /m ]

(a) Linearised BH curve (b) Yoke with coil and magnet

Figure 2.4: Linearised BH curve (Somaloy500) and yoke structure.

For the non-linear iron, this m.m.f. is a function of the values of B0f e and µf e
in the interval where its flux density is located.

N · i = Hf e · lf e + Hpm · lpm + Hδ · lδ (2.5)

φ
Af e − B0f e φ
Apm − B0pm φ
Aδ
N ·i = · lf e + · lpm + · lδ (2.6)
µ0 · µf e µ0 · µpm µ0
φ lf e lpm lδ B0f e
N ·i = + + − · lf e
µ0 Af e · µf e Apm · µpm Aδ µ0 · µf e
B0pm
− · lpm (2.7)
µ0 · µpm
B0f e B0pm
N ·i+ µ0 ·µf e · lf e + µ0 ·µpm · lpm
φ = (2.8)
Rf e + Rpm + Rδ

lf e lpm lδ
Rf e = Rpm = Rδ = (2.9)
µ0 · µf e · Af e µ0 · µpm · Apm µ0 · Aδ

This formulation can be easily extrapolated to the MEC model of the ma-
chine. At position zero, the MEC model contains 43 iron reluctances and 25
2.4. Non-linear iterative method 25

air reluctances, and a number of 27 fluxes need to be calculated. When the ro-
tor mesh is rotated, the MEC model contains 9 additional air reluctances, and
36 fluxes need to be calculated. The flux density in one iron reluctance k at
any interval can be expressed as in (2.10). Applying Ampere’s law to the mesh
containing that reluctance yields (2.11), where ‘w’, ‘n’ and ‘r’ are the number
of winding m.m.f.’s, iron and air reluctances included in that mesh respectively.
Following the same procedure explained above, the flux in a mesh ‘x’ can be
expressed as in (2.12).

Bf e (k) = B0f e (k) + µ0 · µf e · Hf e (k) (2.10)

w n r
N (h) · i(h) = Hf e (k) · lf e (k) + Hδ (p) · lδ (p) (2.11)
h=1 k=1 p=1

w n B0f e (k)
h=1 N (h) · i(h) + k=1 µ0 µf e lf e (k)
φ(x) = n r (2.12)
k=1 Rf e (k) + p=1 Rδ (p)

Equation (2.12) is solved iteratively for all the meshes using M ATLAB, and
the procedure is described in the flowchart shown in Figure 2.5. In the def-
inition stage, the BH curve is linearised, and the limits for the intervals are
calculated together with their slopes and the extrapolation at Hf e = 0. In the
initialization stage, all the reluctances are set to be within the first interval (lin-
ear solution), and thereafter the iteration process is started. The reluctance (R)
and the m.m.f. (F) for each element are calculated using the updated value of
µf e and B0f e after each iteration. The total m.m.f. matrix (Ftotal ) is obtained
by adding the internal m.m.f. of each element within a mesh to the m.m.f.
value of the coils that are included in that mesh. The flux matrix is calculated
dividing the total m.m.f. matrix by the reluctance matrix using the back-slash
operator in M ATLAB. The flux densities are then calculated at each element and
compared to the limits of the corresponding interval. If they are located within
these limits, the proper Hf e and µf e are calculated using the non-linearised BH
curve (BHmaterial ). If the value is higher than the upper limit or lower than
the lower limit, the values of B0f e and µf e are updated to the corresponding
values of the next or the previous interval respectively, and the program moves
to the next iteration. It was observed that the process converged after less than
20 iterations (N ), taking less than 2 seconds in computer equipped with a 1.2
26 Chapter 2. Magnetic-Equivalent-Circuit model

B e g in

IN D E X
D E F IN E
- I: n r o f r e l u c t a n c e s - N: n r o f i t e r a t i o n s
- Blimit ; Hlimit : l i m i t s l i n e a r i n t e r v a l s
- m: n r o f m e s h e s - j: r e l u c t a n c e u n i t
- P fe : s l o p e s i n t h e i n t e r v a l s
- A: r e l u c t a n c e a r e a - R: r e l u c t a n c e m a t r i x
- Bexpo : e x t r a p o l a t e d B a t H = 0
- L: r e l u c t a n c e l e n g t h - : fl u x

IN IT IA L IZ A T IO N S
B 0fe (1 : I ) = 0
fe (1 : I ) = P fe (1)
k(1 : I ) = 1
fo r a = 1 to N
R(1 : I ) = L(1 : I )=(0
fe(1 : I )
A(1 : I ))
F (1 : I ) = B 0fe (1 : I )
L(1 : I )=(0
fe (1 : I ))
Ftotal = Fcoils (1 : m) + F (1 : I )m

 = RnFtotal
fo r j= 1 to I
Btemp (j ) = (j )=A(j )

B 0fe (j ) = Bexpo(k(j ))
Blim (k(j ))  Htemp (j ) = interpl(BHmaterial ; Btemp (j ))
Btemp (j )  Y es
fe (j ) = (Btemp (j ) Blimit (k(j )))/
Blim (k(j ) + 1) ::: 0
(Htemp (j ) Hlimit (k(j )))
N o

B 0fe (j ) = Bexpo(k(j ) + 1)
Btemp (j ) > Y es
fe (j ) = P fe (k(j ) + 1)
Blim (k(j )+1)
k(j ) = k(j ) + 1
N o
B 0fe (j ) = Bexpo(k(j ) 1)
fe (j ) = P fe (k(j ) 1)
k(j ) = k(j ) 1

Y es
en d a en d j
Y es N o
E n d n ex t j

Figure 2.5: Flowchart for the non-linear iteration process.

2.5. FEM and MEC results 27

GHz Pentium III processor and 1 Gb RAM. The results of the flux densities ob-
tained at the elements of the MEC model (Figure 2.1) are shown in Figure 2.6,
at no load and load. The limits of the linearised intervals are also indicated in
the figures. It can be observed that most of the reluctances are located in the
first (linear) interval.

1 .6 1 .6
R c r− to p R c r− b o tto m
1 .4 R c r− b o tto m
1 .4
R ps2
R ps2
1 .2 R ps3 1 .2 R c r− to p
R ps3
R ss
1 1 R ss
R sra
[T ]

R ps1
[T ]

0 .8 R yx 0 .8 R sra
B

R yx
B

R yp23
0 .6 R sc2 0 .6 R yp12
R sc3 R sc2
0 .4 0 .4 R yp23
R yp12
R yp31 R sc1
0 .2 0 .2 R sc3
R sc1, R ps1 R yp31
0 0
0 0 .5 e 4 1e4 1 .5 e 4 2e4 0 0 .5 e 4 1e4 1 .5 e 4 2e4
H [A /m ] H [A /m ]

(a) No load (b) Load

Figure 2.6: Flux densities at the elements in the MEC model.

2.5 FEM and MEC results

The machine was simulated using finite element (FE) analysis. The details
about the FE model set-up will be described in Chapter 3, but some results
will be included in this section for comparison. The FE model reproduced
the same part of the machine as the MEC model, and it was solved applying
symmetric boundary conditions at the edges. The number of elements was
451248. The rotor mesh could be rotated in steps of 2.5 degrees, comprising
12 steps per pole. One simulation at a fixed rotor position required 1.5 hours
CPU time using the same computer equipment described above for the MEC
model implementation.
28 Chapter 2. Magnetic-Equivalent-Circuit model

Flux distribution

The flux flow through the teeth and the end-plates was evaluated for the FEM
and MEC simulations at position zero, both at no load and load. The results
are summarized in Table 2.1. The sign indicates the direction of the flux, which
was considered positive in the radial direction outwards. In the table, the link-
ing flux is obtained adding the absolute value of the flux flowing through the
three teeth, and the total flux is obtained adding the flux through both end-
plates. The leakage is calculated as the percentage of the total flux that does not
contribute to the linking flux. The results show a very good agreement at no
load. At load there are more discrepancies in the values at the teeth. This may
be due to the assumption that the m.m.f. from the armature coils contributes
equally to the flux flow in the four meshes adjacent to one tooth. Nevertheless,
the results are acceptable especially comparing the total, linking and leakage
fluxes. In both cases the difference in the leakage predicted from the FEM and
MEC simulations is less than 2%.

Torque response

It was desired to assess the shape of the torque response, and this could be
achieved by rotating the mesh. The points 90 electrical degrees before and after
position zero will be referred to as position minus six and position six respectively,
indicating the number of steps that the rotor mesh needs to be rotated until
that position. The change of torque between two positions was calculated using
(2.13), where θ is the angle rotated. In this equation, as a first approximation,
Wmag was calculated from the magnetic energy in the system, when in fact it
should be the co-energy. However, in a first stage the effects of magnetic sat-
uration in the torque production were neglected. This is not too unreasonable
since most of the reluctances in Figure 2.6 are located in the first interval, in-
dicating that the machine is not too saturated. Under these circumstances the
magnetic energy and co-energy were assumed to be similar. The energy after
each step was calculated using (2.14), which is derived from the equivalence be-
tween the equations for the magnetic energy density wmag in (2.15). In these
equations, B represents the flux density in the selected reluctance element, and
V its volume.
2.5. FEM and MEC results 29

Table 2.1: Flux results [wb] from the FEM and MEC simulations
No Load Load
Part FEM MEC FEM MEC
Tooth 1 -4.37e-7 -1.88e-5 -8.97e-4 -6.93e-4
Tooth 2 9.61e-4 10.02e-4 11.81e-4 10.28e-4
Tooth 3 -9.62e-4 -9.37e-4 -1.86e-4 -6.02e-4
Plate1 32.37e-4 31.85e-4 29.02e-4 32.90e-4
Plate2 -32.36e-4 -32.31e-4 -29.51e-4 -30.23e-4
Linking 19.24e-4 19.58e-4 22.64e-4 23.23e-4
Total 64.74e-4 64.16e-4 58.51e-4 63.13e-4
Leakage 70.27% 69.47% 61.30% 63.20%

∆Wmag
∆T = (2.13)
∆θ
Ngap
1 1
Wmag = Bi2 · Vi (2.14)
2 µ0
i=1

1 1 2 Wmag
wmag = B wmag = (2.15)
2 µ0 V
It can be shown that the torque does not depend on the state in the iron
and/or leakage paths, but on the conditions in the airgap only (Ostović, 1989).
Furthermore, the energy stored in the axial airgaps does not contribute to the
torque production in the direction of rotation. Therefore, equation (2.14)
was only evaluated in the direct and by-pass reluctances along the radial air-
gap (Ngap ). It should be noted that the torque expression obtained combining
(2.13) and (2.14) is equivalent to (2.16) (Miller, 1993), where i is the current,
and L the inductance at a particular position.
1 ∆L
∆T = i2 (2.16)
2 ∆θ
A value of zero torque was assigned to positions six and minus six. The
MEC model for backward rotation was used to calculate the torque change
30 Chapter 2. Magnetic-Equivalent-Circuit model

between positions zero and minus six, starting from position minus six. The
model for forward rotation was used between positions zero and six, starting
from position six. The results from the FEM and MEC simulations are shown
in Figure 2.7. In this figure, the response from MEC has been scaled to fit the
amplitude of the FEM response, since it was observed that the method described
above was quite sensitive to the dimensions of the radial airgap reluctances. In
fact, a change in this area affected significantly the amplitude of the torque
response, while the overall shape was little affected. In order to fit the amplitude
of the FEM response, the area of the radial airgap reluctances was increased by
45% with respect to their original value. The airgap reluctances affect directly
the values of the magnetic energy in (2.14) or the inductance in (2.16). An
acceptable error in reluctance approximation can produce too high an error
when it is differentiated, thus giving an inaccurate result for torque. Overall, it
can be observed that the shape of the torque response is reasonably predicted.
It should be pointed also that the area of the radial airgap reluctances in the
MEC model corresponds to that below the tip of the teeth, and their length is
the airgap length. Having a slot opening of 33% the tooth pitch, the fringing
effects in the airgap region may not be negligible, which in turn would increase
the effective reluctance area.

2 0

0
M e c h a n ic a l to rq u e [N m ]

− 2 0

− 4 0

− 6 0

− 8 0

− 1 0 0 F E M
M E C
− 1 2 0
1 8 0 2 4 0 3 0 0 3 6 0
E le c tric a l a n g le [d e g ]

Figure 2.7: Torque response from FEM and scaled from MEC.
2.6. Optimization 31

2.6 Optimization

The results presented in the previous section were calculated for a geometry ob-
tained from an optimization process. The optimization criteria was to achieve
the maximum possible torque for a given volume, and within the specified ther-
mal constrains. Simple cut and try was used first in FEM to set the approximate
dimensions of the different components. Thereafter, an optimization routine
was implemented in M ATLAB using the MEC model in Figure 2.1. Although
the torque could not be calculated directly from this model, the goal was to
maximize the linking flux and the armature current for a given copper filling
factor in the coils (0.75 assuming pressed windings). For optimization, the iron
losses were calculated using Steinmetz formulation (2.17), in W/m3 , where f
is the frequency and B the flux density. The parameters kh and kdyn account
for the hysteresis and dynamic losses. The dynamic losses consist of eddy cur-
rent and anomalous losses. These parameters together with n, nB and nf have
been empirically calculated by Skarrie (2001) for the iron powder used in the
machine (kh = 1160, n = 1.6, kdyn = 57, nB = 1.85, nf = 1.40).

Pf e = kh B̂ n f + kdyn B̂ nB f nf (2.17)

The calculation was carried out at no load and only for the fundamental
frequency component, so the losses in the rotor were neglected. It was assumed
that the flux variation anywhere else in the machine followed a sinusoidal shape
with that frequency. This is clearly not true, but predicting analytically the
real flux change pattern at different parts of the machine is difficult and time-
consuming. Instead, this will be done using FEM in Chapter 3. It was assumed
that the heat could be removed from all the external surfaces of the machine
with a rate of 5000 W/m2 . In fact, this factor depends of course on the tem-
perature gradient between the surface and the coolant, the speed of flow and
the geometry of the cooling surface, among others. A detailed formulation for
the calculation of the heat transfer coefficient from the different surfaces will be
described in Chapter 4. Finally, it was assumed that the iron losses in the teeth
and the back core, together with the a.c. copper losses, were cooled through the
external cylindrical surface. The iron losses in the end-plates together with the
d.c. copper losses were cooled through the surface area of the end-plates.
The volume of the machine was kept constant during the optimization pro-
cess. The rest of the design variables were optimized in groups. The first group
32 Chapter 2. Magnetic-Equivalent-Circuit model

included the stator yoke, the axial length of the stator teeth body, the thickness
of the end-plates at the yoke connection, and the thickness of their salient part,
where the magnetizing coils are resting. Changing this last dimension modified
the magnitude of the magnetizing current, since the available space for the coil
was affected. The axial length of the stator teeth also determined the magnitude
of the load current, since it affected the space available for the accommodation
of the end-windings. An iteration process was set-up for each geometry to find
the current loading in the coils corresponding to the maximum copper and iron
losses that could be dissipated. For a given current, the iron losses were cal-
culated from the flux density at each reluctance element in the MEC model,
which was obtained using the non-linear iterative method presented in section
2.4.
The second group of variables included the thickness of the rotor body and
the shape of the claw-pole. This shape was defined by the thickness and the
angle span of both the base and the tip of the claw-pole. Finally, the thickness
of the tooth tip, the slot opening, the axial and radial airgap lengths, and the
shaft diameter were optimized individually. The length of the radial and axial
airgaps was 0.6 mm and 0.5 mm respectively. The radial airgap diameter was
270 mm. The slot opening was 33% of the tooth pitch to reduce the leakage
between the teeth. This will of course increase the ripple in the machine, and it
was expected that it could be compensated electronically introducing a suitable
harmonic in the currents from the control system. It was observed that some
material could be removed from the surface of the end-plates closer to the shaft.
The weight of the active parts of the machine was 102 kg, distributed in 77.7
kg of iron, and 24.3 kg of copper (14.9 kg for the armature coils and 9.4 kg for
the field coils).

2.7 Conclusions
In this chapter, a method to include non-linearities in MEC models has been
presented. The method is based on the linearisation of the BH curve, and it
has been applied to calculate the leakage and the torque response in a three-
dimensional electrically magnetized synchronous machine. The method pro-
vides very fast and accurate results for the flux distribution in the machine at
load and no load, predicting the leakage with an error lower than 2% in a few
seconds. Modifications to accomplish forward and backward rotation were in-
corporated in the model. A simplified method was used to assess the shape of
2.7. Conclusions 33

the torque response. It was observed that a change in the area of the airgap re-
luctances affected significantly the amplitude of the torque response, while the
shape was reasonably predicted. Finally, the optimization process has been pre-
sented, and the results will be used for the down-scaled prototype, maintaining
exactly the same proportions.
34 Chapter 2. Magnetic-Equivalent-Circuit model
Chapter 3

Finite Element model

3.1 Introduction

MEC models are useful during the optimization process of an electrical ma-
chine, but more advanced methods are needed to assess more accurately the
final characteristics of the design. Today, numerical methods are increasingly
used for the solution of electromagnetic fields. There is a variety of commercial
computer programs based on the finite element method (FEM) used for this
purpose. In this chapter the finite element (FE) model of the machine will be
presented, after a short introduction to FEM. Thereafter, the FE model will be
used to predict the iron losses in the machine. As it will be shown in Chapter 5,
the output torque of the machine is most optimal for a 16 pole design, which
will be the configuration used for the final prototype. Therefore, the machine
studied in this and next chapters is not only down-scaled compared to the de-
sign in Chapter 2 but it also has a higher pole number. The inner and outer
radius are 19 mm and 100 mm, and the total length of the machine is 76 mm.

3.2 FEM

The method of the finite elements is a numerical approach by which the general
differential equations describing a certain physical phenomena in a structure can
be solved in an approximate manner. These equations are assumed to hold over
a certain region of the structure, which can be one-, two- or three-dimensional.
In fact, the region is divided into smaller parts, so-called finite elements, and the
approximation is then carried out over each element. This approximation, usu-

35
36 Chapter 3. Finite Element model

ally a polynomial, is actually some kind of interpolation over the element, and
it is assumed that the variable is known at certain points within the element,
called nodal points. The precise manner in which the variable changes between
its values at the nodal points is expressed by the specific approximation, which
may be linear, quadratic, cubic, etc. The coefficients of the polynomials are cho-
sen in such a way that a variational principle is approximately satisfied. Having
determined the behaviour of all elements, they are then patched together, using
some specific rules, to form the entire region. The collection of all elements is
called a finite element mesh. More information about the basic formulation used
in FEM can be found for example in (Ottosen and Petersson, 1992).
The emergence of FEM took place in the early 1960’s and since then its
use has spread to virtually all fields of engineering. The first applications on
electrical machine problems were presented in the early 1970’s. In the 1980’s,
the research on numerical field computation methods expanded rapidly, and
there are nowadays several commercial program packages available for the com-
putation of two- and three-dimensional magnetic fields and eddy currents. In
general, the electromagnetic field is defined by Maxwell’s equations, which are
a rather simple formulation of the field problem but they are difficult to solve,
especially in electrical machines. This is due to the complicated geometry, the
time-dependency of the magnetic field, and the non-linearities because of the
magnetic saturation of the iron. Furthermore, the equations of the magnetic
field are coupled with the electric circuit equations of the windings and the
motion of the rotor.
The variable that is solved in electromagnetic FEM is the magnetic potential
at the nodes. The solution algorithm is often based on the minimisation of
a mathematical function that is related to the stored potential energy in the
field. From the node potentials, the parameters of the machine are obtained
using different formulations. Surely the most difficult quantity to calculate
is the torque, and better formulations are still being developed. The virtual
work method is normally approximated evaluating the rate of change of the
magnetic coenergy between two positions with respect to the step angle. If the
distance is short, the coenergy values at the two positions differs only slightly.
Therefore, the accuracy of the result obtained by subtraction may be low, if
the errors of the two coenergy values are not similar. A more common method
is based on the Maxwell stress tensor, defined in (3.1). It allows to calculate
the force acting on all parts within a volume (the rotor) by evaluating a surface
integral around the volume (Γj ). The volume should be totally enclosed by the
3.3. FE model of the EMSM 37

surface and the surface itself should span through a nonferromagnetic region
(the airgap). The flux density vector B  is given by (3.2), where Bn is the normal
component, Bt is the tangential component, n is the unit normal vector and t
is the unit vector in the direction of the projection of B on the tangential plane
Γj . The calculations using this method may be inaccurate since the error in the
direction of the flux density vector is amplified in the force calculation by the
multiplication of the flux density components. However, the accuracy can be
improved by using higher-order finite elements. A more detailed explanation of
the application of the FE method to electric machine analysis can be found in
the specialised literature (Hamdi, 1994), (Luomi, 1993), (Zhou, 1993).

1  − 1 B 2 n dΓ =
F = Bn B
µ0 2µ0
Γj
1 1
= (B 2 − Bt2 ) n + Bn Bt t dΓ (3.1)
2µ0 n µ0
Γj
 = Bn n + Bt t
B (3.2)

3.3 FE model of the EMSM

The FE model was implemented in a commercial package, O PERA 3D. The

design procedure was automated in M ATLAB using a set of libraries already
available to generate the O PERA code for the different steps in the definition of
a FE model. To automate the process for a specific design is time-consuming,
but once it is completed a new FE model with changes in the original geometry,
or in the formulation of the model, can be obtained in a matter of seconds.
This is of course very useful for fine-tuning of the design.
The FE model of the claw-pole machine was divided in two meshes, cor-
responding to the part of the machine above and below the airgap. To reduce
the size of the model, symmetry was exploited and only one pole pair was mod-
elled. For each mesh a grid of macroelements (facets) was built from a set of
circumferential lines of constant radius, corresponding to the radius of the dif-
ferent parts of the machine, and a set of radial lines separated by a constant
angle. This angle defined the minimum step of rotation, and it was selected for
a resolution of 24 steps per pole-pair. The sides of the facets were then divided
into smaller parts, which formed the finite elements. Thereafter the grid was
38 Chapter 3. Finite Element model

extruded in a number of steps in the axial direction, forming hexaedral volumes.

This shape of the elements was mandatory using the pre-processor module of
the program. The material properties were assigned to the facets at the extru-
sions, which implied that all the finite elements within the volume between two
extrusions on one side of the facet shared the same material properties. The
model was solved using a magnetostatic analysis, where no eddy currents are
induced in the iron, and a total scalar potential was defined for the iron, and
a reduced scalar potential for all the air. The conductors were built adapting
pre-defined geometries available in the software and incorporated in the mesh.
Positive periodic boundary conditions were defined at the facets located in the
symmetry planes, with the condition that the normal derivative vanishes at the
boundary (Neumann condition). A thin (5 mm) layer of air was defined around
the rest of the external surfaces, and the condition that the flux density has only
tangential component (Dirichlet condition) was imposed on the air external
surface. Finally, the coordinates of the nodes in the grid at each extrusion were
modified in the radial, axial and circumferential directions to fit the desired ge-
ometry of the machine. A plot of the final 3D grid is shown in Figure 3.1. The
finite element mesh of the model will be shown in next sections where the iron
losses are analyzed.

Figure 3.1: Plot of the 3D grid of the FE model.

3.4. Iron losses 39

The mesh continuity has to be ensured over the whole model. Dividing the
grids of the two meshes in the same number of divisions in the circumferential
direction facilitated the coupling between them through the radial airgap. This
airgap was divided into four layer of elements, two located in each mesh. This
is recommended since the torque calculations are jeopardized if they are carried
out through elements linking materials with very different permeability. The
four meshes also allow to compare two alternative paths of integration, through
the middle of the second and third layers. The axial airgaps were also divided
into four layer of elements. Quadratic elements, with an extra node on each
side, were used in the airgaps in order to further increase the accuracy of the
fields in these critical regions, while the potential variation in the rest of the
elements was assumed to be linear. The non-linearities in the iron were defined
with a table containing the values of the BH curve of the material provided
by the manufacturer. The total number of elements in the model was 154224,
requiring 33 minutes to solve one simulation at load for a certain rotor position.
Triplicating the number of elements changed the torque response less than 1%.
The post-processing module of the software had in-built functions to perform
calculations over a specified path. The torque density was calculated applying
the Maxwell stress tensor in Cartesian coordinates using (3.3)-(3.7), where H
represents the magnetic field intensity. The calculations were performed over
a grid of 50x400 points in the axial and circumferential directions through the
middle of the third airgap layer all around the rotor. The total torque was
obtained by integrating T qz along the grid surface. The action point for the
torque was located at the geometrical center of the machine (x0 , y0 , z0 ).

Hn = Hx nx + Hy ny + Hz nz (3.3)
Hm = Bx Hx + By Hy + Bz Hz (3.4)
Hm
Fx = Bx Hn − nx (3.5)
2
Hm
Fy = By Hn − ny (3.6)
2
T qz = (x − x0 )Fy − (y − y0 )Fx (3.7)

3.4 Iron losses

After copper losses, core losses are generally the second largest component of
power loss or inefficiency in electrical machines. They arise from the variation
40 Chapter 3. Finite Element model

of the magnetic flux density through the core. In general, the iron losses can
be divided into three types depending on the physical background of the loss:
hysteresis, eddy current, and anomalous loss. Hysteresis loss is a static loss that
is due to the hysteresis of the material. It is the part of the applied energy
used to move the microscopic domain walls when they are magnetised, and it
is converted into heat instead of being stored in the material. Eddy current and
anomalous losses are dynamic losses, both depending on eddy currents in the
material, but in different scales. The macroscopic eddy current loss, sometimes
referred to as the classical eddy current loss, origins from currents in the material
set up by an applied varying magnetic field, while the microscopic eddy current
loss, often called anomalous or excess loss, is due to eddy currents generated by
domain wall motion. This is caused by the inhomogeneity in the magnetisation
between adjacent domains, or domain groups depending on their size, which
gives rise to eddy currents at the walls surrounding them (Skarrie, 2001).
The lower permeability of SMC compared to laminations implies that the
hysteresis losses are higher at most relevant frequencies (a few hundreds hertz)
(Höganäs AB, 2001). However, the dynamic losses are lower in SMC, since the
material is built up by small particles electrically isolated from each other. If the
insulation was perfect, there would be only eddy currents circling inside the iron
particles but no currents in-between them. This implies that the eddy current
loss in iron powder would be independent of the geometry of the specimen but
dependent on the size and shape of the particles. However, this assumption is
in most cases not true since the insulation between particles is not perfect and
currents between particles exist. Nevertheless, the dynamic losses are in general
lower in SMC compared to laminations.

Iron loss prediction

Iron loss prediction is an important issue in both design and analysis of electri-
cal machines. It is a difficult task, due to the complexity of the machine, flux
distribution and rotational variation of flux. It has long been realized that a
considerable amount of the total iron loss in the stator core of a rotating elec-
trical machine is caused by the rotating magnetic field. However, the lack of
data and proper models for rotational core losses has traditionally led to the
use of the Steinmetz equation for alternating core loss instead. This trend has
changed in the last years, when important advances have been made regarding
the modelling of rotational losses.
3.4. Iron losses 41

Zhu and Ramsden (1998) reported novel formulations of rotational losses

for electrical sheet steels, and they were applied to electrical machines. Useful
values of the empirical coefficients used in the formulations were provided. A
similar study was performed later by Guo et al. (2003) using SMC material. Ma
et al. (2003) studied the effects of the iron loss caused by flux harmonics. In all
these contributions the method was evaluated following a procedure where the
finite element problem was solved first, and then the hysteresis effects were in-
serted at the post-processing level. Bottauscio et al. (2002) made an assessment
about how the results are affected if the hysteresis effects are included directly
inside the solution process. A comparison with the previous procedure, or even
with a direct calculation based on the specific loss data-sheet provided by the
manufacturer, showed very little difference in the iron loss results.
The empirical coefficients in the formulation described by Guo et al. (2003)
are calculated for the same iron powder material used in this thesis, but using a
different lubricant (Somaloy 500 + 0.5% Kenolube). However, this is the most
similar measured data available for the iron powder, and is therefore used for
the iron loss calculations in this chapter. In general, the iron losses in a rotat-
ing electrical machine consist of an alternating and a rotating component, and
can be expressed as in (3.8). For pure alternation and rotation the trajectory
of the flux density loci describes a line and a circle respectively. But gener-
ally, alternating and rotating effects interact yielding an elliptical trajectory, and
Bmajor and Bminor represent the major and minor axis of the ellipse. Their
ratio RB = Bminor /Bmajor determines the contribution of the alternating
and rotating components to the total core losses. When RB is 0 or 1 the losses
are purely alternating or rotational respectively.
Ptot = RB Prot + (1 − RB )2 Palt (3.8)
The specific alternating and rotational components in (3.8) are calculated
according to the procedure presented by Guo et al. (2003). The alternating
losses were separated into hysteresis, eddy current and anomalous losses using
(3.9), where B̂ is the peak value of flux density and f the frequency. Similarly,
the specific circular core loss was also separated into three parts using (3.10),
where Phr is the rotational hysteresis loss, and Cer and Car are the coefficients
for the rotational eddy current and anomalous losses. The rotational hystere-
sis loss per cycle was expressed in terms of four parameters a1 , a2 , a3 and Bs
(3.11). The value of s was calculated from (3.12). The loss coefficients were de-
termined experimentally and they are summarized in Table 3.1. It was pointed
42 Chapter 3. Finite Element model

out that the rotational excess loss is generally a function of the flux density and
eventually reduces to zero when the material is saturated and all domain walls
disappear.
Palt = Cha · f · B̂ h + Cea (f · B̂)2 + Caa (f · B̂)1.5 (3.9)
2 1.5
Prot = Phr + Cer (f · B̂) + Car (f · B̂) (3.10)
 
1 1
 2−s 
Phr = f · a1   s
 − 2  (3.11)
1 2
a2 + s + a23 a2 + 1
+ a23
2−s

B̂ 1
s = 1− 1− (3.12)
Bs a22 + a23

3.5 Iron losses in the EMSM

The model described in the previous section was adapted to calculate the iron
losses in the EMSM, both in the stator and the rotor. The FE model was
solved at no load at 24 positions along one electrical period. This implied that
the losses could be calculated up to the 11th harmonic. A table was created
with the field values calculated in the center of all the elements in the model.
This procedure was jeopardized by the fact that the element numbering was
automatically modified in all the model after each rotor rotation. Therefore,
a table with the coordinates of the elements at the first position was created,
and the field values were evaluated at these coordinates in the rest of the steps.
Although this procedure was effective in the stator, the problem in the rotor is
that the coordinates of the elements change after each step. The only way of
identifying the rotor elements was to rotate back the whole model so that the
rotor mesh coincided with its first position. At the same time the stator will be
displaced and the field in its elements corrupted, and this is the reason why the
stator and rotor losses were calculated separately.
The field tables were processed in M ATLAB. The flux density values were
transformed from Cartesian to polar coordinates, and the FFT of the radial,
circumferential and axial components was calculated. The ratio RB was com-
puted at each element and harmonic using (3.13). The major and minor axis of
the ellipse were obtained from the maximum and minimum between the mod-
ulus of the radial and axial components of the flux density together, and the
tangential component. Inherent with this procedure is the approximation that
3.5. Iron losses in the EMSM 43

Table 3.1: Loss coefficients for Somaloy 500 (Guo et al., 2003)
Coefficient Value Coefficient Value
Cha 0.1402 Cer 2.303 · 10−4
Cea 1.233 · 10−5 a1 6.814
Caa 3.645 · 10−4 a2 1.054
h 1.548 a3 1.445
Car 0 Bs 2.134 T

either the maximum or the minimum radius of the ellipse are located exactly
along the plane defined by the axial and radial components, or in the tangen-
tial direction. This is not necessarily true, but it is reasonable since the pattern
of flux flow in the machine occurs mainly in these two directions. Actually, in
most parts of the machine one component or the other will be negligible. When
the machine rotates, the change in magnetization will give rise to an increased
tangential component, especially in the teeth tips and the back core right above
the teeth, and probably also in the end-plates at the level of the rotor, but in a
smaller scale.
 
min 2 + B2 , B
Brad ax tan
RB =   (3.13)
max 2 + B2 , B
Brad ax tan

Finally, tables were created with the harmonic components of the flux den-
sity in each element, and the index RB . These tables were imported into the FE
post-processor, where (3.8) was implemented for each element and harmonic,
and the losses were calculated performing a volume integral and multiplying by
the density of the material. The total loss at each element was approximated
simply by adding the fundamental and all its harmonic components. Finally,
the losses from the elements corresponding to the same region in the thermal
model (analysed in next chapter) were added together. It should be noted that
Phr in (3.11) becomes negative for values of B̂ > Bs . Although the total flux
density in some local heavily saturated part of the machine passed this limit, it
was observed that this condition was never satisfied for the fundamental or the
harmonic components on their own.
44 Chapter 3. Finite Element model

Results

Simulations were carried out to assess the losses in the machine at load and no
load with a current loading calculated for the thermal limit in the windings,
which will be treated in next chapter. Figure 3.2(a) shows the distribution of
the alternating and rotational losses in the stator of the machine for the funda-
mental flux density at no load. The RB distribution shows that the alternating
losses are concentrated in the body of the teeth, while the losses in the tips are
dominated by the rotational component. In the back core, rotational losses
appear around the regions where the teeth are connected to the core, while
alternating losses are more important in the regions between the teeth. The
rotational losses are clearly dominating in the regions of the end-plates close
to the rotor. However, although RB provides information about the ratio of
change of the components of the flux density vector at each element, the total
losses due to these components depends directly on the level of the flux den-
sity. Figure 3.2(b) shows the flux density distribution in the stator for the same
conditions as Figure 3.2(a). It can be observed that the flux density variation
is concentrated mainly on the teeth body and tips (around 1.4 T), while the
values in the back-core and in the end-plates are much lower. The flux density
in the back-core is maximum right above the teeth (1 T), and 0.35 T in the
surface of the end-plates closest to the rotor. Therefore it can be concluded that
the iron losses in the stator at the fundamental frequency is dominated by the
alternating component in the body of the teeth.
The distribution of RB and the flux density in the rotor for the third har-
monic are shown in Figure 3.2(c) and Figure 3.2(d) respectively. It can be
observed how the rotational components are concentrated in the upper surface
of the claw-poles and to a lower extent in the sides closest to the end-plates.
However, the maximum flux density is mainly located in the upper surface of
the claw-poles. It can be therefore concluded that the losses in the rotor at this
frequency are due to the rotating component in their surface below the radial
airgap.
A summary of the losses in the stator and in the rotor at load and no load
including the harmonics is presented in Figure 3.3(a) and Figure 3.3(b) respec-
tively. The load simulations where performed rotating the current vector in the
stator together with the rotor for one cycle, in order to obtain the maximum
torque at each position. Above each plot the total loss is referred to as ‘Tot’, and
the losses in the stator and the rotor are referred to as ‘St’ and ‘Rt’ respectively.
3.5. Iron losses in the EMSM 45

(a) Stator RB , fundamental (b) Stator B, fundamental

(c) Rotor RB , 3rd harmonic (d) Rotor B, 3rd harmonic

Figure 3.2: RB and flux density distribution for the fundamental in the stator
and the third harmonic in the rotor, no load.
46 Chapter 3. Finite Element model

It can be observed that the highest losses are due to the fundamental compo-
nent in the stator. At load, the losses at this frequency are also increased by
30% compared to the no load case. The rotor losses appear only at harmonics
multiple of 3, which is due to the slotting harmonics. The losses in the rotor at
load due to the third harmonic are considerably bigger than at other frequen-
cies, and together with the stator losses at that harmonic account for 23% of
the total losses at load. In general, comparing the load and no load case, the
total losses are increased at load by 40%, the stator losses by 38% and the rotor
losses by 55%. It is also clear from the plots the rapid decay of the losses with
the frequency. Therefore neglecting the harmonics above the 11th component
has little effect in the total predicted iron losses.

T o t: 1 3 8 , S t: 1 2 0 , R t: 1 8 W T o t: 1 9 3 , S t: 1 6 5 , R t: 2 8 W
1 0 0 1 0 0
s ta to r s ta to r
ro to r ro to r
8 0 8 0
Iro n lo s s e s [W ]
Iro n lo s s e s [W ]

6 0 6 0

4 0 4 0

2 0 2 0

0 0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r H a rm o n ic n u m b e r

(a) No load (b) Load

Figure 3.3: Distribution of the iron losses in the machine as a function of the
harmonic number.

Table 3.2 shows a summary of the alternating and rotating loss components
calculated at different parts of the machine, both at load and no load. The values
are obtained from the sum of all the harmonics for the respective component at
each part. In general, it can be observed that the losses are dominated by the
rotating component, which represents 78% of the total losses at no load and
71% at load. The only place in the machine where the alternating losses are
more important is in the body of the teeth, which represents 75% of the total
alternating losses at no load, and 70% at load. At no load, the contribution
to the total rotating losses is biggest at the end-plates, followed by the back-
3.6. Conclusions 47

Table 3.2: Summary of the predicted iron losses in the machine [W]
No Load Load
Part Alternation Rotation Alternation Rotation
back-core 3.8 27.1 8.2 43.4
teeth body 22.8 5.6 38.5 5.6
teeth tip 2.2 23.1 4.0 29.3
end-plates 0.9 34.7 3.2 32.8
rotor sides 0.1 3.0 0.5 3.5
claw-poles 0.5 13.7 1.2 22.7
Total 30.3 107.2 55.6 137.3

core, the teeth tips and to a lower extent the claw-poles. However, at load the
contribution from the back-core is the most important, increasing about 60%,
although the other parts also experience an increase, except in the case of the
end-plates. Nevertheless, the back-core stands alone for 54% of the increase in
the rotational losses between no load and load. The alternating losses in the
teeth body are increased by almost 70% at load. The other parts also experience
an increase in the alternating losses, but their effect in the final value is more
modest. The teeth body stand alone for 62% of the increase in the alternating
losses between no load and load. The losses at load will be used for the thermal
model presented in next chapter.

3.6 Conclusions
In this chapter it has been shown how 3D FE analysis can be used to calculate
the iron losses in the machine using a method to separate the losses in alter-
nating and rotational components. It has been shown that the rotational losses
represent around 70% of the total losses in the machine, and that these losses
dominate everywhere except in the body of the teeth. Although this method
to calculate the losses may be accurate, it is not practical for optimization. To
the time of approximately 13 hours needed to simulate the 23 rotor positions
for given load conditions, around 6 hours more need to be added for the loss
calculation. However, once the design is quasi-optimized, this method allows to
48 Chapter 3. Finite Element model

calculate precisely where the losses are located, their components, and how they
behave with the frequency. The losses at different parts of the machine can then
be imported into the thermal model to assess how they affect the temperature
distribution in the machine.
Chapter 4

Thermal model

4.1 Introduction

The thermal characteristics of an electrical machine are important since the

temperature rise in the windings usually defines the power that can be obtained
from the machine. The lumped parameter thermal model has been used for a
long time for calculating the temperature rise in electrical machines. Advanced
models for machines of TEFC design, mainly induction motors, have been re-
ported in the literature (Mellor et al., 1991), (Jokinen and Saari, 1997), (Kylan-
der, 1995), (Boglietti et al., 2004). In recent years there has been also a trend to
use computer FEM models to calculate the thermal characteristics of machines,
where the magnetic and thermal simulations are usually coupled (Driesen et al.,
2001), (Vong and Rodger, 2003). In this chapter, a lumped parameter thermal
network model will be presented for calculating the temperature distribution in
the machine at steady state operation. The thermal resistances are directly cal-
culated as a function of the geometry details, which are defined in Appendix B.
In general, the most important source of heat in electrical machines are
the copper losses from the windings. Therefore the electrical calculations are
described first, which will serve to predict these losses. The thermal network
with the path for the losses through different parts of the machine is described
before introducing a more detailed formulation of the heat transfer problem.
The machine is water cooled, and it is assumed that the coolant flows directly
along the external surface of the back-core and the end-plates.

49
50 Chapter 4. Thermal model

4.2 Electrical calculations

The d.c. link voltage Ulink in the converter was selected as 275 V, around 10%
below the maximum of 300 V available in the laboratory equipment. The ma-
chine will be driven using vector control, and it can be shown that the peak
phase voltage can be increased by 15% using third harmonic injection (sym-
metrization) compared to the case with sinusoidal√ modulation (Alaküla, 2002).
The peak phase voltage Û fac is then Ulink / 3. The number of conductors in
the a.c. winding was calculated using (4.1). The value of ωs corresponds to the
fundamental electrical frequency f in rad/sec, and φ̂ is the fundamental peak
flux-linkage of one phase. This value was preliminary taken from the MEC
model at no load. For the prototype, φ̂ was calculated by doing the FFT of
the flux linkage of one phase at the 24 FEM solutions along one electrical cy-
cle. The number of turns in the d.c. winding were calculated using (4.2). This
equation is deduced making equal (4.3) and (4.4), which are two different ex-
pressions to calculate the d.c. coil resistance from an electrical and a geometrical
point of view respectively. In these equations, U fdc is the voltage, Adc the slot
area, kf ill the copper fill factor, Itdc the Ampere turns in the slot, and r̄dc the
average radius of the coil. The voltage was selected as 12 V per field coil (24 V
connected in series).

Û fac
N cac = (4.1)
ωs · φ̂ · p
U fdc · Adc · kf ill
N cdc = (4.2)
ρ · 2π · Itdc · r̄dc
U fdc N cdc · U fdc
Rcoildc = = (4.3)
ifdc Itdc
2π · r̄dc · N cdc
Rcoildc = ρ (4.4)
Adc · kf ill /N cdc

The diameter of the copper conductors was calculated from (4.5). Fs is

the slot fill factor, including the insulation, and normal values for non-pressed
windings range from 0.6 to 0.7. Nc is the number of conductors in one coil,
and Aslot the slot area. Np1 and Np2 are the number of parallel strands with
diameter dis1 and dis2 , where dis = dcu · kis . The factor kis accounts for
the coating, and it was selected in order to increase the diameter of the copper
strand by 7% (Sadarangani, 2000). Equation (4.5) can be simplified into (4.6)
4.2. Electrical calculations 51

considering single diameter conductors in the coils and no parallel strands. In

the a.c. winding there are two coils sharing one slot, so N c was twice the num-
ber of conductors in one coil.
  
Fs = N c 0.25π Np1 · d2is1 + Np2 · d2is2 /Aslot (4.5)

Fs · Aslot
dcu = 2 (4.6)
0.25π · N c · kis

Although the distribution of the three phases in the teeth is straightforward,

it is instructive to determine it following the more general procedure for con-
centrated windings described by Cros and Viarouge (2002). The number of
slots Ns per pole Np and per phase Nph is defined as Spp in (4.7) and has to
be reduced to a fraction of two non divisible integers b and c. For our distri-
bution of three phases accommodated in three slots per pole pair the values are
b=1 and c=2. A repeatable sequence of 0 and 1 specific to the winding can be
derived from this relation. The number of 1 is equal to b and the number of 0
is equal to c − b. The 1 and 0 have to be distributed as evenly as possible in
the sequence, which is then repeated three times. In our case this is straight-
forward and the sequence becomes 101010. To this sequence it is associated
the usual phase sequence AC’BA’CB’. Then the conductors associated with the
number 1 are extracted to form the first layer of the winding. The second layer
of the winding is obtained reproducing and shifting the initial layer by a tooth.
From this simple procedure, the distribution of the windings in the machine is
obtained, as shown in Figure 4.1.

Ns b
Spp = = (4.7)
Np · Nph c

The coating material was bonding epoxy, with a temperature class of 180o
C (class H). For the slot wall insulation, a 0.5 mm thick layer of Kapton mate-
rial was used. The thermal conductivity of the materials is listed in Table 4.1.
The equivalent thermal conductivity in the a.c. and d.c. coils was calculated
using (4.8). This expression was deduced by Arkkio (2002), for a squared slot
and round conductors. The value of κi is the conductivity of the coating, d1
is the diameter of the copper strand, d2 is the diameter of the copper strand
and coating together, and δi is the shortest distance between the surface of two
52 Chapter 4. Thermal model

A A’ B B’ C C’

Figure 4.1: Sequence of the winding distribution.

Table 4.1: Thermal conductivity κ of the materials

Material κ [W/mK]
Somaloy500 13
Copper 360
Bonding epoxy 0.64
Kapton 0.12

copper strands, which was approximated as two times the thickness of the coat-
ing. Jack et al. (2000) stated that the thermal resistance of the pressed windings
(78% copper fill factor) was reduced by 46%. Assuming that this thermal resis-
tance would be similar in our machine with a 75% copper fill factor, then the
value calculated from (4.8) for the d.c. and a.c. windings was increased by 92%,
and κcoil =17.6 W/mK.
d1 δi
κcoil = κi + (4.8)
δi d2

Copper losses
The total copper losses in the d.c. and the a.c. windings are calculated as in
(4.9) and (4.10). The currents in the a.c. coils are Ia = Itac , Ib = −Itac /2,
and Ic = −Itac /2, where Itac is the maximum peak Ampere turns in one
coil. The losses in all the a.c. coils in the machine are calculated by multiplying
with the number of pole pairs p. It should be noted that this way of calculating
the losses for the a.c. winding is equivalent to the more familiar expression
4.2. Electrical calculations 53

√
P = 3 · I 2 · R, where the effective current is I = Itac /( 2 · N cac ) and
R = ρ · N cac · Lac /(Aac /N cac ). The resistance of the d.c. coil (4.11) is
calculated at the average radius r̄dc (4.12). The value of insl in the coil cross
sectional area Adc (4.13) corresponds to the thickness of the wall insulation.
The resistance of the a.c. coil is calculated as in (4.14), where Lac (4.15) is
simply obtained by adding the average length of the two axial sides (4.16) and
the two tangential sides (4.17). In these equations, the coil thickness cwthac is
calculated as in (4.18), at the average radius given in (4.19). The cross sectional
area of one a.c. coil Aac (4.20) was calculated as the minimum between the area
available in the axial (4.21) and the tangential (4.22) directions. The weight
of the conductors was calculated as in (4.23) and (4.24) for the a.c. and d.c.
windings, where δcu is the density of the copper (8900 kg/m3 ). It should be
noted that the resistivity of the copper ρcu is calculated at a temperature of
140o C, using the equivalence ρcu = ρcu0 (1 + αcu (Tcu − 20)), where ρcu0 is
the resistivity at 20o C (1.673 · 10−8 Ω m) and αcu is the temperature coefficient
(3.93 · 10−3 /o C).

P cudc = 2 · It2dc · Rdc (4.9)

P cuac = (Ia2
+ Ib2 + Ic2 )
· Rac · p (4.10)
2π · r̄dc
Rdc = ρcu (4.11)
Adc · kf ill
ros − dy + ris
r̄dc = (4.12)
2
Adc = (ros − dy − ror − 2 · insl) ·
·(zdy + g2 − 2 · insl) (4.13)
Lac
Rac = ρcu (4.14)
Aac · kf ill
Lac = 2 · Laxiac + 2 · Ltanac (4.15)
Laxiac = zts + cwthac (4.16)
Ltanac = wt + cwthac (4.17)
1
cwthac = (2πr̄ac /Ns − wt − 2 · insl) (4.18)
2
1
r̄ac = (ros − dy + ris + dos) (4.19)
2
Aac = min(Aaxiac , Atanac ) (4.20)
54 Chapter 4. Thermal model

1 2π/Ns  
Aaxiac = ( (ros − dy − insl)2 − (ris + dos + insl)2 −
2 2
wt
−2( + insl)((ros − dy − insl) −
2
−(ris + dos + insl))) (4.21)
1
Atanac = ( (ztps − zts) − insl)((ros − dy − insl) −
2
−(ris + dos + insl)) (4.22)
Wac = Aac · 2(Ltanac + Laxiac ) · Ns · δcu (4.23)
Wdc = 2 · Adc · 2πr̄dc · δcu (4.24)

4.3 Thermal network

The losses considered to contribute to the heat production in the machine are
the iron and copper losses. Friction and stray losses are not included at this
point, but they will certainly appear in the measurements in Chapter 7. Two
paths were defined for the dissipation of the losses from the stator, and they are
shown in Figure 4.2. The copper losses of the a.c. winding (‘Pcu1’) and the iron
losses of the stator teeth and the stator yoke are referred to as ‘Pac’. These losses
are dissipated in the radial direction through the outer cylindrical surface of the
yoke. The whole surface area of the end-plates is used to cool the copper losses
of the d.c. winding (‘Pcu2’) and the iron losses of the end-plates, and they are
referred to as ‘Pdc’.
It was assumed that 20% of the copper losses in the a.c. winding was trans-
ferred directly to the back core through the top of the coil and that the other
80% was transferred through the teeth, following the path shown in Figure 4.2.
This is not unreasonable since the contact area between the coil and the back
core is 60% of the contact area between the coil and the teeth, which implies
that in principle 38% of the copper losses would be dissipated directly to the
back core. However, in practice it is more complicated to fit the coil with the
surface of the core than with the teeth, especially with the type of modular wind-
ing used. Therefore the heat transfer to the core was reduced to almost half the
theoretical value. The iron losses calculated for the FE model at load (Chap-
ter 3) were grouped into the macro-elements in the thermal model, namely the
tooth tip, the tooth body, the stator core, the end-plates, the rotor claw-poles
and the rotor sides attached to these claw-poles. Since the rotor is completely
enclosed by the surrounding stator parts, it is reasonable to assume that its losses
4.3. Thermal network 55

will be transferred to the stator through the closest path. Therefore, the losses
from the rotor claw-poles were added to ‘Pac’, whereas the losses from the rotor
sides were added to ‘Pdc’.

Pac

Pcu1

X O
X X
Pcu2 Pcu2
Pdc Pdc

Figure 4.2: Thermal dissipation in the EMSM.

The study of the thermal circuit for the dissipation of the ‘Pac’ losses is
simplified considering a pole pair, where three teeth are allocated, as shown in
Figure 4.3. The copper losses in the armature coils transferred to the teeth are
referred to as ‘Pcua1 ’, ‘Pcub1 ’ and ‘Pcuc1 ’, while those transferred directly to
the back core are referred to as ‘Pcua2 ’, ‘Pcub2 ’ and ‘Pcuc2 ’. In these labels the
superscript ‘2 ’ should be read as a label, not as a square symbol. The thermal
resistance in the coils and the wall insulation around the teeth is referred to as
‘Rcw’. This resistance was not included in the path for the losses directly trans-
ferred to the core. This is because the temperature in the coils was calculated
for the most critical conditions, i.e. following the path through the teeth. Most
of the losses are dissipated through this way, which is also the longest. The iron
losses in the three teeth are referred to as ‘Pfea’, ‘Pfeb’ and ‘Pfec’ respectively, and
the thermal resistances of the teeth are referred to as ‘Rfea’, ‘Rfeb’ and ‘Rfec’.
The iron losses in the core and its thermal resistance are referred to as ‘Pfe2ac ’
and ‘Rfe2’ respectively. The thermal resistance from the surface of the machine
to the coolant is ‘Rv’. There are a series of assumptions that can be made in this
circuit:
56 Chapter 4. Thermal model

Pcua1 Rcwa Rfea

Pcua2 Pfea

Pcub1 Rcwb Rfeb Rfe2 Rv

Pcub2 Pfeb Tcool

Pfe2ac
Tsurface
Pcuc1 Rcwc Rfec

Pcuc2 Pfec

Figure 4.3: Thermal circuit for ‘Pac’ losses in one pole pair.

• the copper losses in the three windings during one a.c. cycle are the same
(Pcua1,2 = Pcub1,2 = Pcuc1,2 ).

• the iron losses in the three teeth during one a.c. cycle are the same (Pfea
= Pfeb = Pfec).

• the thermal resistances in the coils and wall insulation around the three
teeth is the same (Rcwa = Rcwb = Rcwc).

• the thermal resistance in the body of the three teeth is the same (Rfea =
Rfeb = Rfec).

Then the circuit in Figure 4.3 can be simplified into the circuit in Fig-
ure 4.4, where the copper losses are Pcu1ac = Pcua1 + Pcub1 + Pcuc1 , and Pcu2ac
= Pcua2 + Pcub2 + Pcuc2 . The iron losses in the teeth are Pfe1ac = Pfea + Pfeb
+ Pfec, and the thermal resistances are Rcwt = RcwaRcwbRcwc, and Rfe1 =
RfeaRfebRfec, in parallel.
The thermal model for ‘Pdc’ in one end-plate is represented in a similar way,
see Figure 4.5. The copper losses of one d.c. coil are Pcudc /2 and the iron losses
in the plate are Pfedc . The thermal resistance in the coil and wall insulation is
Rcwp and the resistance in the end-plate is Rfep.
4.4. Heat transfer 57

Pcu1ac
Rcwt Rfe1 Rfe2 Rv

Pfe1ac Tcool
Pfe2ac
Tsurface
Pcu2ac

Figure 4.4: Simplified thermal circuit for ‘Pac’ losses in one pole pair.

Pcudc /2 Rcwp Rfep Rv

Tcool
Pfedc
Tsurface

Figure 4.5: Simplified thermal circuit for ‘Pdc’ losses in one end-plate.

4.4 Heat transfer

The two mechanisms governing the heat transmission in electrical machines are
conduction and convection. In conduction, the heat is transmitted from one
part to another without relative movement between these parts. The equation
for conduction takes the form in (4.25), where q is the heat transmitted through
the surface A along a material with thermal conductivity κ, between two parts
separated by ∆x and with a temperature difference ∆T . In convection, a fluid
moves relative to the cooled part, removing the heat, and (4.26) describes this
process. The parameter hconv is the heat transfer coefficient, which depends
on the properties of the fluid, the pattern of movement, and the geometry of
the cooled surface. It therefore needs to be calculated for each case depending
on the conditions.
∆T
qcond = A (4.25)
∆x/κ
qconv = A · hconv · ∆T (4.26)

The cooling system was first adapted to the one available for the Alvar motor
application. The total water flow Q̇ was limited to 1.2 liter/minute with a
58 Chapter 4. Thermal model

temperature of 27o C, between the heat exchanger and the engine. At this stage,
it was assumed that half of the coolant flow is used to cool the armature winding,
and the other half the field winding through separate circuits. The maximum
temperature rise allowed in the coils was 100o C above an ambient temperature
of 40o C. It was assumed that the coolant would flow through a duct of exactly
the same shape as the cooled surface and a thickness of 3 mm. With these
conditions, the heat transfer coefficient from the surfaces of the machine was
calculated from the known formulas for simple geometries given in the basic
heat transfer theory (Wong, 1977), (Holman, 1992).

Heat transfer coefficient

In principle, the heat transfer coefficient for the cylindrical surface used to cool
‘Pac’ could be calculated using the formulation for forced convection in a cir-
cular cylinder with normal flux, as shown in Figure 4.6. In the machine, the
flux flow would hit perpendicularly the surface from the top (Figure 4.6 ro-
tated clockwise 90o ), and then split into two paths parallel to the end-plates.
However, the fact that the fluid is contained in a duct does not allow to repro-
duce the behaviour associated with this formulation in the textbooks, where the
fluid is not constrained into a tight surface around the cylinder. In this case,
as the flow progresses along the front side of the cylinder, the pressure would
decrease and then increase along the back side of the cylinder, resulting in an in-
crease of free-stream velocity on the front side and a decrease on the back side.
This could eventually cause reverse flow in the back if a phenomenon called
boundary-layer separation occurs. If the fluid is contained inside a narrow duct,
the velocity profile could be assumed to be similar in the front and in the back,
and the heat transfer behaviour could then be approximated as that of a fluid
flowing along a plane surface.

Figure 4.6: Cylinder with normal flux

4.4. Heat transfer 59

The heat transfer coefficient from the surface of the cylinder was therefore
approximated using the formulation for forced convection over a plate, which
assumes that the flow is parallel to a plane surface. Due to symmetry, this for-
mulation is applied to only one of the two paths around the cylinder. The
properties of the coolant were evaluated at the so-called film temperature, de-
fined as the arithmetic mean between the surface (Tsf ) and free-stream (T∞ )
temperatures (4.27). The value of T∞ was calculated as the average between the
inlet (27o C) and outlet temperatures (4.28). The increase in temperature in the
fluid at the output can be calculated using (4.29), where Pac are the copper and
iron losses dissipated through this path, q̇ac the coolant volumetric flow (Q̇/4),
ρw the density and cpw the specific heat. These properties were calculated at
the inlet water temperature.
1
Tm = (Tsf + T∞ ) (4.27)
2
1
T∞ = (Tin + Tout ) (4.28)
2
Pac
∆Tw = (4.29)
q̇ac · ρw · cpw
The determination of the film temperature involves an iteration process,
since the temperature in the surface is a priori unknown. A table was created
with the physical properties of water between 273 K and 375 K in steps of 5
degrees. The values for the specific heat, density, viscosity (µ), and thermal
conductivity (κ) were then extrapolated automatically depending on the calcu-
lated surface temperature. The speed of flow is obtained from (4.30), where
zdy is the axial length of the core and wdct is the thickness of the cooling
duct. A so-called Reynolds number (Re) appears in forced convection, and it
gives information about the characteristics of the flow, laminar or turbulent. In
general, this number is calculated as in (4.31), where x is the geometric charac-
teristic of the problem. In our case, this corresponds to the length of the plane
(Lcyl = π · ros), in order to calculate the average Reynolds number in the sur-
face. The Prandlt number (P r) appears in any heat transfer process and relates
the velocity and temperature fields. It can be expressed as in (4.32). Finally, the
Nusselt number (N u) also appears in any heat transfer process and relates the
heat transfer coefficient hconv in the solid surface with the thermal conduc-
tivity of the fluid per unit length (4.33). In our specific problem it takes also
the form given in (4.34). The coefficients A, m and n are taken depending on
60 Chapter 4. Thermal model

the value of the Reynolds and Prandlt numbers. Combining (4.33) and (4.34)
allows to calculate the heat transfer coefficient from the cylinder.

q̇ac
Uac = (4.30)
zyc · wdct
ρ·U ·x
Re = (4.31)
µ
µ · cp
Pr = (4.32)
κ
hconv · x
Nu = (4.33)
κ
Nu = A · Rem L · Pr
n
(4.34)

A similar formulation was used to calculate the heat transfer coefficient from
the surface of the end-plates. The length Lplt is approximated as the dashed line
shown in Figure 4.7 (4.35), and the area Aplt corresponded to half the end-plate
surface. The properties of the water were calculated at the film temperature,
where Tsf is now the temperature at the surface of the end-plate. The increase
in the cooling water temperature was calculated as in (4.29), updating the coef-
ficients. The speed of the flow is calculated as in (4.36), where q̇dc = Q̇/8. The
Reynolds, Prandlt and Nusselt numbers were calculated as above, using Lplt as
the geometric characteristic. A summary of the values obtained applying these
formulas to the cylinder and the plate is shown in Table 4.2.

Lplt
Aplt

Figure 4.7: Length and cooling area in the end-plate

4.4. Heat transfer 61

Table 4.2: Parameters for the heat transfer coefficient calculation

Coefficient Cylinder End-plate
Tm [o C] 48 47
∆ Tw [o C] 9 10
U [m/s] 0.022 0.014
Re 12076 6663
Pr 3.7 3.8
Nu 112 84
A 0.664 0.664
m 0.50 0.50
n 0.33 0.33
hconv [W/mK] 229 193

1
Lplt = (ros − rir) + zdx + 2π(ros + rir) (4.35)
4
q̇dc
Udc = (4.36)
Aplt /Lplt · wdct

Heat transfer model

In this section, the heat transfer between the different parts of the model will be
described. The thermal resistances are a function of the area and the length of
the transmission path. The explicit derivation of these components are included
in Appendix C. The calculations for the path following ‘Pac’ in Figure 4.2
will include all the a.c. copper losses and all the iron losses in the back core
and in the teeth. Therefore, the whole cylindrical surface of the core will be
used for cooling. The path for the heat dissipation from the two d.c. coils is
identical, and therefore the calculations will be done only for one of them. The
temperatures are calculated starting from the heat exchange between the surface
of the motor and the coolant. The temperature in the surface of the core is given
by (4.37), where the iron losses from all the rotor claw-poles (P f erC ) have been
62 Chapter 4. Thermal model

included in the power dissipated. The temperature in the surface of the end-
plates is given by (4.38), which includes the iron losses in the corresponding
side of the rotor attached to the claw-poles (P f erT ). The areas Asfac and
Asfdc correspond to the cooling surface area of the back core and one end-
plate respectively.

P cuac + p · (P f e1ac + P f e2ac ) + P f erC

T sfac = + T∞1(4.37)
Asfac · hconvac
P cudc /2 + P f edc + P f erT
T sfdc = + T∞2 (4.38)
Asfdc · hconvdc

The temperature in the inner surface of the core was calculated as in (4.39),
where Lscac is the thickness of the yoke, and Ascac is the area around the
cylinder through the middle of the yoke. The temperature in the surface of the
wall insulation in contact with the iron is calculated using (4.40) and (4.41) for
the a.c. and d.c. circuits respectively. For the calculation of the temperature in
the teeth walls, the copper losses are reduced by 20% and only the iron losses
in the teeth and in the claw-poles are considered. The length Lcwac is half the
distance between the teeth tips (at the width of the body) and the center of the
connection with the core, as shown in Figure 4.8(a). The equivalent distance is
the arithmetic average between half Lt1 and half Lt2 . At each side, the area for
the transmission was selected as the average between the wall area (At1 ) and the
corresponding triangle at the top (Ac1 ). The total area Acwac was calculated
adding the value for the four sides and multiplying by the number of teeth.
The area in one tooth is actually equivalent to the arithmetic average between
the area in the four walls together and the connection area with the core. For the
d.c. circuit Lswdc is calculated as the average distance from the surfaces of the
iron in contact with the wall insulation to the surface of the machine, as shown
in Figure 4.8(b) (Lp1 , . . . , Lp4 ). The area Aswdc was calculated between the
wall insulation and the surface at the level indicated by the dotted line in the
figure.

(P cuac + p · (P f e1ac + P f e2ac ) + P f erC )Lscac

T crac = +
Ascac · κf e
+ T sfac (4.39)
1
(p · (P cuac + P f e1ac ) + P f erC )Lcwac
T wlac = + T crac (4.40)
Acwac · κf e
4.4. Heat transfer 63

(P cudc /2 + P f edc + P f erT )Lswdc

T wldc = + T sfdc (4.41)
Aswdc · κf e

Lp1
Ac2 Ac1
Aswdc
Aihdc

At1 H
At2 Lp2

Lt1
Lt2
Lp3
Lp4

(a) Tooth element (b) Plate element

Figure 4.8: Parametrization of the tooth and plate thermal elements.

The temperature in the surface of the a.c. and d.c. windings closest to the
iron is calculated using (4.42) and (4.43). The length of the path is the insu-
lation thickness. For the a.c. winding the area Awiac is similar to Acwac , but
now through the middle of the insulation. In the same way, for the d.c. winding
the area Awidc is proportional to Aswdc through the middle of the insulation,
but now it also includes the insulation area between the top of the coil and the
core. Only the copper losses are transferred through these paths.

p · P cu1ac · insl
T isac = + T wlac (4.42)
Awiac · κkapton
P cudc /2 · insl
T isdc = + T wldc (4.43)
Awidc · κkapton
The hot spot of the a.c. winding is located at the midpoint along the the
external surface of the coil which is located in the center of the slot. The temper-
ature at this point was calculated using (4.44), where Lihac is the geometrical
distance and Aihac corresponds to half the surface area of the coil at the level
64 Chapter 4. Thermal model

Table 4.3: Estimated temperature distribution [o C]

Part a.c. path d.c. path
Tsf 65 63
Tcr 72
Twl 100 82
Tis 133 126
Thp 140 140

of the main insulation, multiplied by the number of teeth. The hot spot of the
d.c. winding is located at the midpoint along its interior surface, as shown in
Figure 4.8(b) (H). The temperature is calculated using (4.45), where Lihdc is
the arithmetic average between the axial and half the radial lengths of the coil,
and Aihdc is the area through the middle of the coil as shown in Figure 4.8(b).

P cu1ac · Lihac
T hpac = + T isac (4.44)
Aihac · κcoil
P cudc /2 · Lihdc
T hpdc = + T isdc (4.45)
Aihdc · κcoil

The results at the different parts of the machine at the thermal limit are
summarized in Table 4.3. The Ampere turns in one a.c. coil were 839 A, and
in one d.c. coil 1195 A. A proposed method to cool the machine in practice is
shown in Figure 4.9. The parts are attached on top of the stator core and the
end-plates. The water flow is similar to the pattern explained in this chapter,
although now the fluid is constrained into tubes rather than flowing through a
continuous surface. The calculated heat transfer coefficient is not applicable for
this case since a different formulation should be used, namely forced convection
inside a tube. The number of ducts and their diameter can be optimized at will.
This system may not be straight forward to manufacture, and practical measures
should be taken to ensure that the water flow is evenly distributed through
all the ducts. However, it can provide an acceptable cooling with reasonable
robustness.
4.4. Heat transfer 65

Cold water

Hot water

Pump
Exchanger

Figure 4.9: An alternative cooling system.

66 Chapter 4. Thermal model

4.5 Conclusions
A thermal network model for the machine has been described together with the
formulations for the heat transfer between the different parts of the machine.
The total copper losses are 568 W (209 W a.c. and 358 W d.c. ) while the iron
losses calculated in the previous chapter at load were 193 W. For a peak torque
of 15.4 Nm, the efficiency of the machine at 1500 r.p.m. is then 76%. This low
efficiency is mainly due to the high copper losses, which account for 75% of the
total losses in the active parts of the machine. Mechanical and stray losses have
not been included in the calculations since they are very difficult to predict, but
they will further decrease the value of the efficiency.
Chapter 5

Alternatives for leakage reduction

5.1 Introduction

In claw-pole structures the flux has a tendency to flow between the lateral faces
of adjacent poles. One way of reducing the leakage is to introduce permanent
magnets between the claw-poles with opposite polarization to the direction of
the leakage, and this idea has already been presented by Taniguchi (2000), and
Henneberger et al. (1996). However, this solution assumes that the leakage
from the tip of the claw-pole to the end-plate is negligible, which is not true
in the case of having magnetically conducting end-plates. The purpose of this
chapter is to analyse how this principle can be applied to our electrically magne-
tized claw-pole machine, based on an in-depth study of the leakage paths in this
novel structure by means of FEM simulations. The study compares machines
with 12, 16, 20 and 24 poles.

5.2 Leakage paths in the rotor

The flux balance is evaluated in a claw-pole as indicated in Figure 5.1(a). This

is a north pole according to the convention that the flux lines leave a north pole
and enter a south pole. The paths considered for the flux flow are represented by
arrows. The flux entering the pole is referred to as ‘Fluxin’. The area through
which this flux is measured comprises not only the surface spanned the same
angle as the main claw-pole, but also the surface below and next to the adja-
cent cavity. The flux from the claw-pole crossing the radial airgap is referred
to as ‘Fluxout’. There are three leakage paths, referred to as ‘Fluxbtw1’ and

67
68 Chapter 5. Alternatives for leakage reduction

‘Fluxbtw2’ for the leakage between the claw-poles, and ‘Fluxtip’ for the leakage
from the tip of the claw-pole to the opposite end-plate and rotor side. The areas
used for the measurement of the flux through the different paths are shown in
Figure 5.1(b).

(a) Flux paths (b) Measurement surfaces

Figure 5.1: Flux paths and measurement surfaces in the rotor.

The leakage in the machine is expressed as the percentage of the flux flowing
through the end-plates which does not enter the main body of the stator teeth.
These quantities are measured at the radius corresponding to the middle of the
teeth. It should be pointed out that the leakage cannot be directly calculated
as the ratio between Fluxout and Fluxin, since a part of Fluxout will be driven
through the leakage between the teeth. In order to reduce the leakage, ferrite
and/or neodium-iron-boron (NdFeB) magnets will be incorporated in the orig-
inal design. The ferrite magnet has a remanence of 386 mT, and a coercivity of
191 kA/m. The NdFeB magnet has a remanence of 1.12 T and a coercivity of
781 kA/m. Their BH curve is shown in Figure 5.2.

5.3 Topologies for leakage reduction

The analysis of the leakage flux flow in the original machine without magnets
indicated that the main leakage path is from the tips of the claw-poles to the
opposite end-plate (Fluxtip). Although perhaps unexpected, this fact is not sur-
5.4. Flux distribution results 69

1 .2
F e rrite
N d F eB
1

0 .8
B [T ]
0 .6

0 .4

0 .2

0
− 8 e5 − 6 e5 − 4 e5 − 2 e5 0
H [A /m ]

Figure 5.2: Magnet BH curves.

prising due to the proximity of the magnetically conducting end-plate. In order

to decrease the leakage, seven designs with different alternatives for the location
of the magnets around the claw-poles have been studied. The configurations
are referred to with a number: number 1 is the original design without magnets
(Figure 5.3(a)); number 2 is the design with only ferrite magnets in the cavities
opposite the tips of the claw-poles (Figure 5.3(b)); number 3 is the design with
only ferrite magnets between the claw-poles (Figure 5.3(c)); number 4 is the de-
sign with ferrite magnets in the tips and between the claw-poles (Figure 5.3(d));
number 5 is the design with only NdFeB magnets in the tips; number 6 is the
design with NdFeB magnets in the tips and ferrite magnets between the claw-
poles; and number 7 is the design with NdFeB magnets both in the tips and
between the claw-poles. The arrows in the magnets indicate the direction of
magnetization, which is opposite to the direction of the leakage in the original
design.

5.4 Flux distribution results

The topologies presented in the previous section have been implemented in

machines with different number of poles. It would be expected that the torque
response increases with the pole number. However, as the number of poles in-
creases, the claw-poles are closer to each other, which in turn increases the leak-
age between them. The ratio of three stator teeth per pole pair was maintained,
independent of the number of poles. The angular measures were adapted in
70 Chapter 5. Alternatives for leakage reduction

(a) No magnets: Design 1 (b) Magnets only on tips: Designs 2,5

(c) Magnets only in-between: Design 3 (d) Magnets overall: Designs 4,6,7

Figure 5.3: Topologies for leakage reduction.

5.4. Flux distribution results 71

proportion to the change in the pole angle, while the radial and axial measures
were kept constant. The study of the flux distribution was performed at no
load.

Design with 12 poles

The results for the flux flow in the 12-pole design are summarized in Table 5.1.
It can be observed that in the original design without magnets, Fluxtip is about
twice Fluxbtw1 and Fluxbtw2 together. When ferrite magnets are included
in the tips, Fluxtip decreases by 14% while Fluxbtw1 and Fluxbtw2 are in-
creased by 10%. The leakage is marginally reduced by 2%. Introducing only
ferrite magnets between the claw-poles changes the direction of Fluxbtw1 and
Fluxbtw2, which now contribute to increase Fluxout, and the leakage is de-
creased by 10%. The change in direction in the fluxes is represented with a
negative sign in front of its value in the table. The differences between De-
sign 3 and Design 4, with ferrite magnets both in the tips and between the
claw-poles, are small. When only NdFeB magnets are included in the tips, the
leakage is considerably improved compared to Design 2, since Fluxin is little
changed whereas Fluxout is increased by 18%. However, although Fluxtip is
decreased by 33%, the leakage between the claw-poles increases by 60%. This
is probably due to the fact that the stator teeth are already saturated (1.4 T),
what makes them a more difficult path to follow than the way to the adjacent
claw-pole.
With ferrite magnets between the claw-poles and NdFeB in the tips, Flux-
btw1 and Fluxbtw2 shift direction and contribute to increase Fluxout by 40%
compared to Design 1. Fluxtip is almost unchanged, and the leakage is re-
duced to less than half. Finally, including NdFeB magnets overall increases
enormously the contribution of Fluxbtw1 and Fluxbtw2 to Fluxout, which is
now 63% bigger than in Design 1. Fluxin is however decreased by 62%. These
results indicate that the NdFeB magnets between the claw-poles are more effec-
tively increasing the total flux in the machine, and driving it in the tangential
direction. In fact, the leakage calculated as the percentage of the flux entering
the teeth compared to the flux flowing in the end-plates is now negative, which
implies that there is a source which drives the flux through an alternative path.
Design 7 is basically a permanent magnet machine where the field coils can be
used to adjust the level of magnetization.
72 Chapter 5. Alternatives for leakage reduction

Table 5.1: Summary of flux flow [wb]· 10−4 for the 12 pole design
Design Fluxin Fluxtip Fluxbtw1 Fluxbtw2 Fluxout leakage
1 7.53 2.24 0.55 0.59 4.44 71.6%
2 8.22 1.93 0.61 0.65 4.60 69.5%
3 6.20 3.50 -1.71 -1.66 5.54 62.0%
4 6.79 3.23 -1.66 -1.61 5.60 60.1%
5 8.41 1.28 0.90 1.10 5.45 48.3%
6 6.90 2.20 -1.00 -0.91 6.24 27.1%
7 2.84 3.51 -4.70 -4.57 7.22 <0%

Designs with higher pole number

The results for the flux flow in the designs with 16, 20 and 24 poles are sum-
marized in Table 5.2, Table 5.3, and Table 5.4 respectively. It can be observed
how the leakage increases with the number of poles. For a given pole number,
the leakage decreases continuously from Design 1 to 7. In Design 1, the differ-
ence between the flux leaking between the poles and Fluxtip decreases with the
pole number. In the machine with 24 poles, Fluxtip, Fluxbtw1 and Fluxbtw2
are quite similar. This means that as the pole number increases, the leakage
between the claw-poles is becoming more important than the leakage from the
tips. This is reasonable, since the angular distance between the poles decreases
while the distance from the tip to the opposite end-plate is constant.
When ferrite magnets are included in the tips (Design 2), Fluxtip decreases
by up to 24%, but Fluxout increases less than 7% in all the machines. However,
Fluxbtw1 and Fluxbtw2 also increase by up to 20%. Therefore, part of the
contribution from the magnet is lost through the leakage between the claw-
poles. As indicated above, this is probably due to the saturation in the teeth
combined with the low strength of the ferrite magnet. The decrease in the
leakage with this solution is around 2% for all the machines.
Introducing only ferrite magnets between the claw-poles (Design 3) changes
the direction of Fluxbtw1 and Fluxbtw2 in all the machines. Fluxtip is now
increased from 56% in the 12-pole machine to 87% in the 24-pole machine.
However, Fluxout is also increased by around 30%, and the leakage is decreased
5.4. Flux distribution results 73

Table 5.2: Summary of flux flow [wb]· 10−4 for the 16 pole design
Design Fluxin Fluxtip Fluxbtw1 Fluxbtw2 Fluxout leakage
1 5.88 1.53 0.63 0.65 3.26 74.8%
2 6.42 1.28 0.72 0.75 3.40 72.6%
3 5.07 2.60 -1.16 -1.12 4.31 65.9%
4 5.56 2.35 -1.07 -1.03 4.36 64.2%
5 6.79 0.77 1.04 1.21 4.11 56.2%
6 6.05 1.42 -0.25 -0.16 4.76 45.2%
7 2.08 2.70 -4.04 -3.92 5.95 <0%

Table 5.3: Summary of flux flow [wb]· 10−4 for the 20 pole design
Design Fluxin Fluxtip Fluxbtw1 Fluxbtw2 Fluxout leakage
1 4.86 1.13 0.68 0.71 2.48 77.5%
2 5.33 0.91 0.78 0.89 2.63 75.0%
3 4.40 2.01 -0.69 -0.65 3.41 69.5%
4 4.88 1.76 -0.58 -0.53 3.46 68.0%
5 5.72 0.46 1.08 1.12 3.20 61.5%
6 5.40 0.94 0.16 0.25 3.74 55.0%
7 1.74 2.12 -3.62 -3.51 5.14 <0%
74 Chapter 5. Alternatives for leakage reduction

Table 5.4: Summary of flux flow [wb]· 10−4 for the 24 pole design
Design Fluxin Fluxtip Fluxbtw1 Fluxbtw2 Fluxout leakage
1 4.10 0.86 0.72 0.74 1.93 80.4%
2 4.49 0.65 0.82 0.84 2.07 77.7%
3 3.87 1.61 -0.41 -0.37 2.74 73.2%
4 4.25 1.33 -0.30 -0.25 2.78 71.7%
5 4.91 0.26 1.11 1.14 2.54 66.6%
6 4.78 0.64 0.41 0.49 2.99 62.1%
7 1.55 1.70 -3.18 -3.07 4.54 <0 %

by around 8% in all the machines. The differences between Design 3 and

Design 4, with ferrite magnets both in the tips and between the claw-poles, are
small for all the machines. Fluxout is indeed very similar, and the difference in
leakage between these two designs is less than 2%.
When NdFeB magnets are included in the tips (Design 5), Fluxtip decreases
by up to 50%, and Fluxout increases between 22% and 32%. These values are
clearly better than in Design 2, using ferrite magnets. However, the leakage be-
tween the claw-poles increases now between 55% and 75% compared to Design
1, due to the higher strength of the magnet. The benefits of including NdFeB
magnets in the tips to reduce the leakage decreases with the pole number, from
23% for 12 poles, to 14% for 24 poles. In fact, the magnitude of the increase
in Fluxout is four times the increase in Fluxin for the 12-pole design, while it is
almost the same for the 24-pole design.
With ferrite magnets between the claw-poles and NdFeB in the tips (Design
6), Fluxbtw1 and Fluxbtw2 do not shift direction in the designs with 20 and 24
poles. The magnet capability to drive the flux in a certain direction is directly
proportional to its thickness. When the pole number is increased, the space
available for the magnet in its axial direction is considerably reduced between
the claw-poles, while it is unaffected in the cavities. Therefore, in this case the
contribution from the NdFeB magnets in the tips is stronger than that from the
ferrite magnets on both sides of the claw-poles. The effectiveness of Design 6 to
improve the leakage decreases with the pole number, from 45% in the 12-pole
design, to 18% in the 24-pole design. While the increase in Fluxin between
5.5. Torque response 75

Design 1 and 6 is more or less constant in all the machines, the increase in
Fluxout is reduced with higher pole numbers, up to half in the 24-pole machine
compared to the machine with 12 poles.
Finally, when NdFeB magnets are included overall (Design 7), the effect
of the magnets between the claw-poles prevails over the magnets in the tips.
It can be observed that the directions of Fluxbtw1 and Fluxbtw2 are opposite
to the original direction of the leakage in all the machines. The increase in
Fluxout between Design 1 and 7 is almost the same for all the machines, while
the decrease in Fluxin is almost half in the 24-pole design compared to the
design with 12 poles. This trend could be interpreted as a lower capability of
the magnets between the claw-poles to drive the flux in the tangential direction,
as the pole number increases, due to the reduction of their thickness.
It can be observed that the leakage in Design 7 is negative in all the ma-
chines. Indeed, the NdFeB magnets between the claw-poles open an important
path to the linking flux along the circumferential direction, as an alternative
to the axial path. This new linking flux is not accounted for in the total flux,
which is measured through the stator end-plates. Therefore the leakage results
for Design 7 are not really representative of the leakage of the machine and can-
not be compared to the rest of the designs under the same conditions. However,
it is clear that the linking flux is importantly increased in Design 7, and this will
eventually lead to a considerable increase in the output torque of the machine.

5.5 Torque response

The torque response is calculated at the thermal limit, and the iron losses were
obtained from the MEC model, as explained in Chapter 2. As the pole number
increases, there is less cooling surface per tooth, and the resistance in the a.c.
coils increases. This is due to the less space available in the slots if the insulation
thickness in the wall and between the coils is kept constant. This implied that
the total a.c. current for the machines with 16, 20 and 24 poles was reduced
by 14%, 26% and 34% respectively, compared to the design with 12 poles. To
achieve thermal equilibrium, the d.c. currents also had to be adjusted, although
the decrease was less than 5%.
The static torque response through the positive electrical pole for the ma-
chines with different pole numbers is shown in Figure 5.4. The results for
Design 5 have been omitted in order to provide a clearer plot. This response
is indeed very close to the response from Design 4, with a slightly higher peak.
76 Chapter 5. Alternatives for leakage reduction

However, this peak is always below the peak in Design 6 and usually half the
way between Design 4 and 6. The responses from Designs 1 and 2, and Designs
3 and 4, have been assigned the same curve, since they are almost literally on
top of each other.

3 5 3 5
D 1 & 2 D 1 & 2
3 0 D 3 & 4 3 0 D 3 & 4
D 6 D 6
m e c h a n ic a l to rq u e [N m ]

m e c h a n ic a l to rq u e [N m ]
2 5 D 7 2 5 D 7

2 0 2 0

1 5 1 5

1 0 1 0

5 5

0 0
0 4 5 9 0 1 3 5 1 8 0 0 4 5 9 0 1 3 5 1 8 0
e le c tric a l a n g le [d e g ] e le c tric a l a n g le [d e g ]

(a) 12 pole machine (b) 16 pole machine

3 5 3 5
D 1 & 2 D 1 & 2
3 0 D 3 & 4 3 0 D 3 & 4
D 6 D 6
m e c h a n ic a l to rq u e [N m ]
m e c h a n ic a l to rq u e [N m ]

2 5 D 7 2 5 D 7

2 0 2 0

1 5 1 5

1 0 1 0

5 5

0 0
0 4 5 9 0 1 3 5 1 8 0 0 4 5 9 0 1 3 5 1 8 0
e le c tric a l a n g le [d e g ] e le c tric a l a n g le [d e g ]

(c) 20 pole machine (d) 24 pole machine

Figure 5.4: Static torque response.

It can be observed from the figures that, for a given pole number, the peak
torque increases continuously from Design 1 to 7. This agrees with the flux
flow results presented in the previous section, which indicate that Fluxout in-
5.5. Torque response 77

creases in the same manner. The peak torque in the original design without
magnets is very similar for all the machines, around 15 Nm, although it slightly
decreases for the 24-pole case. The torque delivered by Design 3&4 increases
by 13%, 16%, 20% and 22% for the machines with 12, 16, 20 and 24 poles
respectively, compared to the original design with no magnets. When imple-
menting Design 6, the torque increase reaches 25%, 27%, 30% and 33% in
the same pole number order. Finally, when implementing Design 7 the peak
torque increases by 53%, 69%, 88% and 113% respectively. The configuration
delivering the maximum torque is Design 7 with 24 poles, which gives 30 Nm.
For all machines, it can be observed that the difference is very significant when
NdFeB magnets are used overall, compared to the rest of the designs.

Economical considerations
The solutions presented in this chapter to reduce the leakage are based in the
use of permanent magnet material, which is considerably more expensive than
iron or copper. Therefore, the improvement in the performance of the machine
needs to be compared to the cost of the solution. Another implication is that the
inertia of the rotor will be increased due to the mass added when the magnets
are incorporated. Table 5.5 presents a summary of the amount of each magnet
material which is needed for the different configurations, the change in the
inertia of the rotor and the cost.
The mass calculation was performed as a function of the volume, with a
density of 7500 kg/m3 for NdFeB magnets and 4800 kg/m3 for the ferrite
material. The mass of magnet material and iron is independent of the number
of poles. The mass of iron in the machine is 9.73 kg. This is distributed in
1.42 kg in the rotor, 4.35 kg in the stator core and teeth, and 3.96 kg in the
end-plates. The mass of copper in the d.c. coils is constant for all the machines,
1.12 kg. However, the mass of copper in the a.c. coils decreases with the pole
number, 1.71 kg for 12 poles, 1.39 kg for 16 poles, 1.20 kg for 20 poles, and
1.07 kg for 24 poles.
The per unit material cost was calculated as a function of the mass of the
active parts of the machine. The iron was assigned 1 p.u., the copper 5 p.u.,
the ferrite magnet 10 p.u., and the NdFeB magnet 50 p.u. The cost due to
the difference in the weight of the a.c. coils for different pole numbers was not
considered. The values in the table correspond to the 12-pole design, but the
ratio of change between different designs will be similar independently of the
78 Chapter 5. Alternatives for leakage reduction

Table 5.5: Summary of the properties of the designs

Design ferrite [kg] NdFeB [kg] inertia [kg · m2 ] cost [p.u.]
1 0 0 4.0 · 10−3 23.9
2 0.185 0 4.6 · 10−3 25.7
3 0.377 0 5.3 · 10−3 27.7
4 0.563 0 5.9 · 10−3 29.5
5 0 0.290 5.0 · 10−3 38.4
6 0.377 0.290 6.2 · 10−3 42.2
7 0 0.880 6.9 · 10−3 67.9

pole number. In general, an average increase in the torque by 18% (Design

3&4), 29% (Design 6) and 81% (Design 7) implies an approximate increase in
the cost of the machine by 20%, 77% and 284% respectively. The maximum
torque increase (Design 7, 24 poles) corresponds to a increase in the cost around
19% per extra Newton-meter.

5.6 Conclusions

In this chapter, it has been shown how permanent magnets can be used to en-
hance the performance of a claw-pole machine. Although the main leakage
path in our particular machine is through the tip of the claw-poles to the op-
posite end-plate, placing ferrite magnets between the claw-poles proves to be
more effective than using them in the tips. This is mainly due to the higher
space available on both sides of the poles compared to the cavity in front of
the tip, which allows to use thicker magnets. However, if high density NdFeB
magnets are used only in the tips, the leakage is considerably reduced, although
the peak torque only increases with less than 6% compared to the case with
only ferrite magnets in-between. Combining NdFeB magnets in the tips and
ferrite magnets in-between further improves the leakage and torque response,
although not as much as using NdFeB magnets overall. In this case, the pat-
tern of magnetization of the machine is significantly altered, since it is basically
shifted from the axial to the tangential direction. The torque response increases
between 53% for the machine with 12 poles up to 113% for the 24-pole case.
5.6. Conclusions 79

However, in this design the mass of NdFeB reaches 7% of the total mass of the
active components in the machine, which in turn would increase its cost signif-
icantly. Finally it should be pointed out that the possibilities of achieving field
weakening by reducing the current in the d.c. coils are reduced when stronger
magnets are incorporated.
80 Chapter 5. Alternatives for leakage reduction
Chapter 6

Comparison with three alternative

machines

6.1 Introduction

The purpose of this chapter is to compare the performance of the machine

treated so far with other topologies. The designs considered are an outer rotor
claw-pole machine, and two more conventional designs of synchronous ma-
chines with electrical and permanent magnet magnetization. The machines are
compared for similar cooling conditions. Therefore, a similar calculation of the
iron losses and the thermal properties will be carried out as in Chapter 3 and
Chapter 4. The a.c. coils form a similar concentrated winding in all the designs,
and the thermal calculations will emphasize the new considerations specific for
each design. The losses at load are higher in all the designs compared with
the no load case, and they are used in the thermal models. All the machines
have 12 poles, and the characteristics of the 12-pole version of the EMSM ana-
lyzed so far are included for comparison. The new machines have been directly
optimized in FEM, refining the original measures from the EMSM for each
particular case. In principle it was desired to keep constant the total volume for
comparison. However, it was observed that a considerable amount of material
could be removed from certain topologies in the axial direction. The constrain
was then chosen to be a constant outer and inner diameter only. With these
specifications, and for the given cooling system, the aim was to achieve the
best possible design for each topology in order to make a fair comparison. The
dimensions of the machines are included in Appendix D.

81
82 Chapter 6. Comparison with three alternative machines

6.2 EMSM

The characteristics of the 12-pole version of the machine analyzed so far are
summarized in this section, and the design will be referred to as ‘Design 1’.
The iron losses in the stator and in the rotor at no load and load are shown
in Figure 6.1(a) and Figure 6.1(b) respectively. The losses at load are around
30% higher than in the no load case. Compared to the 16-pole machine, the
losses at no load are only slightly reduced (3%). At load the losses are decreased
by around 10% in the 12-pole design, mainly due to the reduced stator losses.
Table 6.1 shows the distribution of the alternating and rotating losses through
the different parts in the machine. The same trends as in the 16-pole design
are observed, although a considerable reduction in the alternating losses can be
appreciated in the teeth body.

T o t: 1 3 4 , S t: 1 1 5 , R t: 1 9 W T o t: 1 7 3 , S t: 1 4 5 , R t: 2 8 W
8 0 8 0
s ta to r s ta to r
ro to r ro to r
6 0 6 0
Iro n lo s s e s [W ]

Iro n lo s s e s [W ]

4 0 4 0

2 0 2 0

0 0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r H a rm o n ic n u m b e r

(a) No load (b) Load

Figure 6.1: Distribution of the iron losses in Design 1 as a function of the har-
monic number.

The thermal model is exactly the same as the one presented in Chapter 4,
now adapted to the new dimensions of the machine. The parameters obtained
for the heat transfer calculation are shown in Table 6.2, and they are very sim-
ilar to the design with 16 poles. In particular, the heat transfer coefficient is
identical for the end-plate, and only reduced by two units in the cylinder. This
is mainly due to the 2o C decrease in the film temperature, which affects slightly
the properties of the coolant.
6.2. EMSM 83

Table 6.1: Summary of the predicted iron losses in Design 1. [W]

No Load Load
Part Alternation Rotation Alternation Rotation
back-core 4.3 27.4 7.6 40.8
teeth body 18.4 6.7 28.4 5.8
teeth tip 1.7 23.0 3.2 26.1
end-plates 1.0 32.2 2.7 30.0
rotor sides 0.1 3.3 0.4 4.2
claw-poles 0.5 15.2 1.1 22.6
Total 26.0 107.8 43.4 129.5

Table 6.2: Parameters for heat transfer calculation in Design 1

Coefficient Cylinder End-plate
Tm [o C] 46 46
∆ Tw [o C] 8 10
U [m/s] 0.022 0.014
Re 11697 6605
Pr 3.8 3.8
Nu 112 84
A 0.664 0.664
m 0.50 0.50
n 0.33 0.33
hconv [W/mK] 227 193
84 Chapter 6. Comparison with three alternative machines

Table 6.3: Temperature distribution in Design 1 [o C]

Part a.c. path d.c. path
Tsf 61 62
Tcr 67
Twl 96 81
Tis 130 126
Thp 140 140

The temperatures in the different regions of the machine are summarized

in Table 6.3. The temperatures in the d.c. circuit are practically identical as in
the 16-pole design, while in the a.c. circuit they are reduced by around 4o C for
the same temperature rise in the coils. This can be explained considering the
lower losses in the teeth and core, together with the fact that the total current
loading is reduced by 15% in the 12-pole machine. Adding the a.c. Ampere
turns in all the slots around the machine, a figure of 6.4·104 A/m is obtained
for the 16-pole machine, and 5.6·104 A/m for the 12-pole machine, expressed
as a function of the circumferential length along the surface of the machine.
The static and dynamic torque response at the thermal limit are shown in
Figure 6.2. The maximum torque is 14.4 Nm. and the ripple is 33% of the
peak torque. The Ampere turns in one a.c. and d.c. coil are 969 A and 1199 A
respectively. The copper losses are 545 W (184 W a.c. and 361 W d.c.). The
efficiency at 1500 r.p.m. is 76%. The copper losses account for 76% of the total
losses in the active parts of the machine.

6.3 Outer rotor EMSM

The second topology studied is shown in Figure 6.3, and it will be referred to as
‘Design 2’. This design is an outer rotor variant of Design 1, and the claw-pole
rotor is now sandwiched between the stator core and the stator ring, where the
field coil is placed. In the following, the stator core in Figure 6.3 will be referred
to as ‘stator 1’, and the stator ring as ‘stator 2’. There are two radial airgaps,
between stator 1 and the rotor, and between the rotor and stator 2. With this
topology the slip-rings are also removed, but the mechanical coupling from the
rotor to the shaft is more complicated. A mould could be used to fit both sides
6.3. Outer rotor EMSM 85

1 5

m e c h a n ic a l to rq u e [N m ] 1 0

− 5

− 1 0
s ta tic
d y n a m ic
− 1 5
0 9 0 1 8 0 2 7 0 3 6 0
e le c tric a l a n g le [d e g ]

Figure 6.2: Torque response in Design 1.

of the rotor into one piece, which is then joined to the shaft using a set of radial
bars on one side of the machine. On the other side the stator ring and the stator
core are coupled together mechanically.
From FE simulations it could be observed that the length of the core in
stator 1 (zyc) could be reduced to 41% of the original value in Design 1. The
thickness of the core (dy) is however increased to almost 3 times. The only
reason for this is to provide the mechanical coupling between stator 1 and the
shaft (through bearings). The dimensions of the teeth body and tips were ba-
sically unchanged. Only the slot opening was decreased to 24% of the tooth
pitch, compared to 33% in Design 1. In the rotor, the thickness of the sides
attached to the claw-poles (ztr) was increased by 50%, matching the width of
the closest iron parts in stator 2. This was done to decrease the reluctance of the
magnetization path, and in fact the total length of the machine is directly af-
fected by this measure. The radial height of these sides was kept to a minimum
since they penalize the diameter of the machine. However, they can not bee too
small since they also provide the mechanical coupling between the claw-poles
in the absence of the attachment to the shaft. The angle span at the base of
the claw-poles was maintained constant. The angle at the tips was increased
by around 50%. This decreased the reluctance in the claw-poles and had little
effect on the leakage between them, since their radial length at the base (dr) was
halved. Finally, the thickness of the yoke in stator 2 (dy2) was maximized to
86 Chapter 6. Comparison with three alternative machines

stator ring

d.c. coil a.c. coil

rotor stator core

Figure 6.3: Design 2.

reduce the reluctance along the magnetization path. However, it also reduces
the space available for the d.c. coil since it was assumed that this coil was con-
strained within stator 2. But in fact, the coil could be extended further into
the rotor, between the sides of the claw-poles, provided a suitable mechanical
support system.

Iron losses
The ratio between the alternating and rotating losses for the fundamental in
the stator is shown in Figure 6.4(a), and the corresponding flux density in Fig-
ure 6.4(b). As in Design 1, the alternating losses are concentrated in the body
of the teeth (1.2 T), while the rotational losses are mainly located in the tips.
Rotational losses are also expected in the parts of stator 2 close to the rotor (0.3
T), and the core in stator 1 close to the teeth (0.6 T). However, the flux density
is very little in most regions of the core, especially those close to the shaft, and
there is almost no variation in their value which is the reason for the unity ratio
in these elements. The ratio and the flux density for the third harmonic in the
rotor are shown in Figure 6.4(c) and Figure 6.4(d). The losses are mainly due
to rotation in the claw-poles, although the flux density is low (0.2 T).
A summary of the losses in the stator and in the rotor at load and no load
including the harmonics is presented in Figure 6.5(a) and Figure 6.5(b) respec-
6.3. Outer rotor EMSM 87

(a) Stator RB , fundamental (b) Stator B, fundamental

(c) Rotor RB , 3rd harmonic (d) Rotor B, 3rd harmonic

Figure 6.4: Design 2 RB and flux density distribution for the fundamental in the
stator and the third harmonic in the rotor, no load.
88 Chapter 6. Comparison with three alternative machines

tively. In both cases, the highest losses are due to the fundamental component
of the stator. The overall losses are increased by 50% when the machine is
loaded, and there is also an important increase of the losses due to the rotor
third harmonic. This contributes to almost triplicate the losses in the rotor at
load, while in the stator alone the losses are increased by 31%. Table 6.4 shows
a summary of the alternating and rotating loss components calculated at differ-
ent parts of the machine. The teeth body stand for around 75% of the total
alternating losses at load and no load. The teeth tips are the most important
contributor to the rotational losses, around 30% at no load and load. The core
of stator 1 and the claw-poles experiment the most significant increase in the
rotational losses at load, while these losses in stator 2 are decreased by around
35%. The rotational losses are the most important both at no load and load,
corresponding to 83% and 76% of the total losses respectively.

T o t: 7 5 , S t: 6 4 , R t: 1 1 W T o t: 1 1 2 , S t: 8 4 , R t: 2 8 W
8 0 8 0
s ta to r s ta to r
ro to r ro to r
6 0 6 0
Iro n lo s s e s [W ]

Iro n lo s s e s [W ]

4 0 4 0

2 0 2 0

0 0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r H a rm o n ic n u m b e r

(a) No load (b) Load

Figure 6.5: Distribution of the iron losses in Design 2 as a function of the har-
monic number.

Thermal model
The paths considered for the dissipation of the losses in Design 2 are shown in
Figure 6.6. Only the sides of the core in stator 1 can be used to cool ‘Pac’, since
the shaft passes through the centre of the core. The rotor is not completely
enclosed as in Design 1 and therefore it was assumed that the rotor losses were
6.3. Outer rotor EMSM 89

Table 6.4: Summary of the predicted iron losses in Design 2. [W]

No Load Load
Part Alternation Rotation Alternation Rotation
St. 1 core 1.2 9.4 2.4 15.3
St. 1 teeth body 10.0 5.8 18.9 5.7
St. 1 teeth tip 1.2 18.7 2.9 27.6
St. 2 core 0.2 7.4 0.1 5.1
St. 2 sides 0.1 9.8 0.0 5.8
rotor sides 0.1 2.2 0.2 5.3
claw-poles 0.2 8.9 2.0 20.4
Total 13.0 62.2 26.5 85.2

dissipated directly through the airflow caused by the rotor rotation. This airflow
could be in turn cooled through the external surfaces of the water channels, al-
though the temperature increase in the water due to this effect was not taken
into consideration in the calculations. The area available to cool the field wind-
ing is considerably reduced compared to Design 1, and this will constrain the
field current. The iron losses in stator 1 were grouped in a similar way as in
Design 1, whereas the iron losses in stator 2 were separated into those in the
core above the d.c. coil and on its sides.
As in Design 1, it was assumed that the coolant would flow through a duct
of exactly the same shape as the cooled surface and a thickness of 3 mm. The
water flow of 1.2 liter/minute (Q̇) at 27o C was split equally for the a.c. and d.c.
cooling. The heat transfer coefficient was calculated using a similar formulation
as in Design 1, i.e. forced convection over a plate. Symmetry was exploited,
and this formulation was applied to only one side of the core in stator 1, and to
half of the cylinder defined by stator 2. The coefficient for the cooling surface
in stator 1 was calculated considering a flow of q̇ac = Q̇/8 through half the
surface in one side Aplt (6.1), and the length Lplt corresponding to this path
(6.2). This length is the geometric characteristic used to calculate the Reynolds
and Nusselt numbers. The speed of flow is obtained from (6.3). For stator
2, the water flow through one half of the cylinder was q̇dc = Q̇/4. The sides
90 Chapter 6. Comparison with three alternative machines

Pdc

Pcu2
X

X O

Pac Pac
Pcu1

Figure 6.6: Thermal dissipation in Design 2.

of stator 2 will also contribute to the cooling, and their area were added to
the cylindrical surface, forming an equivalent duct. For simplicity, the length of
this duct was taken as half the circumferential length of the machine at the outer
radius. A summary of the coefficients obtained for the heat transfer calculation
in stator 1 and 2 is shown in Table 6.5. The heat transfer coefficient from stator
2 is very similar to the cylindrical surface in Design 1, while the coefficient for
stator 1 is considerably increased. This is mainly due to the shorter length of
the cooling path.
1
Aplt = π(ros12 − ris12 ) (6.1)
2
1
Lplt = ros1 − ris1 + 2π(ros1 + ris1) (6.2)
4
q̇ac
Uac = (6.3)
Aplt /Lplt · wdct

The temperatures are calculated starting from the heat exchange between
the surface of the motor and the coolant. For ‘Pac’ a similar thermal circuit as
the one proposed in Chapter 4 is used. The temperature in the cooled surface of
stator 1 is given by (6.4), where Asfac is the surface on both sides of the core.
6.3. Outer rotor EMSM 91

Table 6.5: Parameters for heat transfer calculation in Design 2

Coefficient Stator 1 Stator 2
Tm [o C] 49 45
∆ Tw [o C] 5 8
U [m/s] 0.033 0.022
Re 8390 11711
Pr 3.7 3.9
Nu 93 113
A 0.664 0.664
m 0.50 0.50
n 0.33 0.33
hconv [W/mK] 421 229

The losses dissipated through this path are the total losses in the a.c. winding
and the iron losses in stator 1. The temperature in the surface of stator 2 was
calculated using (6.5), where Asfdc is the total cooling area around the surface.
The losses in this case corresponded to the copper losses in the d.c. coil and
the iron losses in stator 2. The copper losses in the coils were calculated as in
Design 1.
P cuac + p · (P f e1ac + P f e2ac )
T sfac = + T∞1 (6.4)
Asfac · hconvac
P cudc + P f edc
T sfdc = + T∞2 (6.5)
Asfdc · hconvdc
The temperature in the core of stator 1 was calculated using (6.6). The
length Lscac corresponds to half the length from the center of the core surface
attached to the teeth to the cooling surface at the level of the shaft, as indicated
in Figure 6.7(a). The area Ascac is calculated as the arithmetic average between
the whole core surface at the level of the teeth and the area of the cooling sides
Asfac . The temperature in the teeth wall is calculated using (6.7), in the same
way as it was done for Design 1, but now excluding the iron losses in the claw-
poles. The rest of the equations for ‘Pac’ up to the hot spot in the a.c. winding
92 Chapter 6. Comparison with three alternative machines

are identical to Design 1. The temperature in the wall of stator 2 is calculated

using (6.8). The length Lswdc is the arithmetic average between Lp1 , Lp2 and
Lp3 in Figure 6.7(b), and the area Aswdc is the average between the cooled
surface and that in contact with the wall insulation, as indicated in the figure.

(P cuac + p · (P f e1ac + P f e2ac ))Lscac

T crac = + T sfac (6.6)
Ascac · κf e
p · (P cu1ac + P f e1ac )Lcwac
T wlac = + T crac (6.7)
Acwac · κf e
(P cudc + P f edc )Lswdc
T wldc = + T sfdc (6.8)
Aswdc · κf e

X O
Aswdc
Lp1

Lp2 H Lp3
Asfac Asfac
2 2 · Lscac 2 Aihdc

(a) Stator 1 yoke element (b) Stator 2 element

Figure 6.7: Parametrization of the stator 1 and stator 2 thermal elements.

The temperature in the surface of the d.c. winding closest to the iron is
calculated using (6.9). The length of the path is the insulation thickness insl.
The area Awidc is proportional to Aswdc , but now taken through the middle
of the insulation. The hot spot in the coil was located at its geometrical center,
as shown in Figure 6.7(b) (H). The temperature at this point was calculated
using (6.10). The length Lihdc is the arithmetic average between the distance
from the hot spot to the longest and shortest sides of the coil (in the radial and
axial directions). The area Aihdc was calculated as half the area corresponding
to the longest outer side of the coil, plus two times half the area corresponding
to the shortest side, as indicated in the figure. The temperatures in the different
6.4. Conventional EMSM 93

Table 6.6: Temperature distribution in Design 2 [o C]

Part a.c. path d.c. path
Tsf 66 60
Tcr 90
Twl 110 67
Tis 136 122
Thp 140 140

parts of the machine are summarized in Table 6.6.

P cudc · insl
T isdc = + T wldc (6.9)
Awidc · κkapton
P cudc · Lihdc
T hpdc = + T isdc (6.10)
Aihdc · κcoil

The static and dynamic torque response at the thermal limit are shown in
Figure 6.8. The maximum torque is 7.8 Nm and the ripple is 28% of the peak
torque. The Ampere turns in the one a.c. coil are 670 A, and in the d.c. coil
1525 A. The total copper losses are 449 W (137 W a.c. and 312 W d.c.). The
copper losses account for 80% of all the losses in the active parts of the machine.
The efficiency of the machine at 1500 r.p.m. is 69%.

6.4 Conventional EMSM

The third topology studied is shown in Figure 6.9, and it will be referred to
as ‘Design 3’. This configuration resembles the conventional design of an elec-
trically magnetized synchronous machine, where the field coils in the rotor are
accessed through slip-rings. This is a radial machine that could be entirely built
from iron laminations. However, iron powder allows both the stator yoke and
the teeth tips to be extended above the teeth body, which leaves more space for
the winding. For the same reason iron powder was used in the stator core in
Design 2, where the flux flow is only radial. It can be argued that the massive
yokes in the stator core in Design 2 and the rotor core in Design 3 could be
built from laminations. However, it was tested in FEM that the difference in
94 Chapter 6. Comparison with three alternative machines

1 5

m e c h a n ic a l to rq u e [N m ] 1 0

− 5

− 1 0
s ta tic
d y n a m ic
− 1 5
0 9 0 1 8 0 2 7 0 3 6 0
e le c tric a l a n g le [d e g ]

Figure 6.8: Torque response in Design 2.

the torque response is negligible, since the linking flux is still constrained by the
lower permeability of the iron powder used in the teeth.

a.c. coil
d.c. coil

stator rotor

Figure 6.9: Design 3.

The stator of the machine is very similar to that in Design 1. The only
differences are the length (zyc), which is reduced by 44% and the slot opening
(wos), which is reduced by 28% to compensate for the ripple. The rotor has
the shape of an inverted stator and the teeth have the same axial dimensions as
those in the stator. The length of the rotor yoke was minimized, and it matched
6.4. Conventional EMSM 95

the axial length of the teeth tips (ztps2). The slot opening in the rotor (wos2),
as in the stator, was adapted for an optimal torque response.

Iron losses
The ratio between the alternating and rotating losses for the fundamental in the
stator is shown in Figure 6.10(a), and for the third harmonic in the rotor in Fig-
ure 6.10(c). As in the previous designs, the alternating losses are concentrated
in the body of the teeth, where the flux density is around 1.2 T. The rotational
losses dominate at the teeth tips (0.8 T) and parts of the yoke located around
the teeth connection. In the rotor, rotational losses are expected mainly in the
tips, although the flux density is only around 0.2 T. Little losses are expected at
no load in the body of the rotor teeth and the yoke, where the flux density is
too small.
The losses per harmonic component in the stator and the rotor at load
and no load are shown in Figure 6.11. At load the stator losses are almost
doubled and the rotor losses are triplicated. The most important contribution
to the stator losses is still due to the first harmonic, and for the rotor the third
harmonic dominates clearly at load. A summary of the contribution to the
total losses from the alternating and rotating components in different parts of
the machine is shown in Table 6.7. At no load, 70% of the alternating losses
are located in the stator teeth body. At load, the alternating losses are increased
overall, but still the stator teeth body stands for 63% of the total. The rotational
losses at no load are dominated by the stator teeth tips and back core, and their
values are almost doubled at load. In the rotor, the teeth tips contribute to
around 20% of the total rotational losses in the machine. Overall, the rotational
losses represent 71% and 66% of the total losses at no load and load respectively.

Thermal model
The paths considered for the dissipation of the losses in Design 3 are shown in
Figure 6.12. The path for the dissipation of ‘Pac’ is identical to that in Design 1,
although the cooling surface is reduced together with the length of the machine.
The copper losses from the d.c. coils mounted in the rotor (‘Pcu2’) must be air-
cooled. The coolant volumetric flow in the stator is the same as in the stators in
the previous designs, and therefore a smaller pump could be purchased, which
in turn will also compensate for the additional cost of the air-cooling system.
96 Chapter 6. Comparison with three alternative machines

(a) Stator RB , fundamental (b) Stator B, fundamental

(c) Rotor RB , 3rd harmonic (d) Rotor B, 3rd harmonic

Figure 6.10: Design 3 RB and flux density distribution for the fundamental in
the stator and the third harmonic in the rotor, no load.
6.4. Conventional EMSM 97

T o t: 5 1 , S t: 4 4 , R t: 7 W T o t: 1 0 7 , S t: 8 2 , R t: 2 5 W
8 0 8 0
s ta to r s ta to r
ro to r ro to r
6 0 6 0
Iro n lo s s e s [W ]

Iro n lo s s e s [W ]
4 0 4 0

2 0 2 0

0 0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r H a rm o n ic n u m b e r

(a) No load (b) Load

Figure 6.11: Distribution of the iron losses in Design 3 as a function of the har-
monic number.

Table 6.7: Summary of the predicted iron losses in Design 3. [W]

No Load Load
Part Alternation Rotation Alternation Rotation
St. back-core 3.0 11.8 6.6 23.4
St. teeth body 10.2 4.4 22.9 4.8
St. teeth tip 0.9 13.6 3.4 21.1
Rt. back-core 0.0 0.6 0.2 2.2
Rt. teeth body 0.1 0.4 0.9 5.9
Rt. teeth tip 0.2 5.8 2.3 13.8
Total 14.4 36.6 36.3 71.2
98 Chapter 6. Comparison with three alternative machines

The iron losses in the stator were grouped as in Design 1. The iron losses in the
rotor were assumed to be dissipated directly to the air, as in Design 2, thus not
contributing to heat the coils.

Pac

Pcu1

X O

Pcu2 X O Pcu2

Figure 6.12: Thermal dissipation in Design 3.

In principle, it was assumed that no losses were transferred between the

stator and the rotor through the airgap. The open rotor construction allows
the heat to be removed directly from the surface of the d.c. coils by an axial
airflow if enough space is left between these coils in the slots. It was assumed
that a separate fan would be used to provide an airflow of 10 m/s along the
lateral surfaces of each coil, as shown in Figure 6.13. The fan could be attached
to the rotor if the cooling could be compromised at low speeds, otherwise it
should be mounted separately. A typical fan of 80 mm diameter and 15 mm
length produces at 1500 r.p.m. a flow of 15 CFM (cubic feet per minute),
which is equivalent to 0.42 cubic meter per minute. Assuming that all this
flow is driven through an area equivalent to all the ducts between the coils,
with an average thickness of 1.75 mm each, a speed of 21.5 m/s is obtained.
However, the resistance of the body affects the actual flow, which was therefore
assumed to move with around half the ideal speed. In machines where the rotor
losses are cooled through the stator, such as small asynchronous machines, the
6.4. Conventional EMSM 99

conduction/convection through the airgap is usually calculated using a Nusselt

number of 2 (Bjorklund and Kays, 1959), (Becker and Kaye, 1962), (Aoki et
al., 1967). This coefficient was obtained empirically for smooth surfaces and
laminar flow. Gazley (1958) carried out a similar study considering a more
realistic airgap with the slotted stator, and it was shown that the heat conduction
decreases because of the slots. However, according to Gazley, a convective heat
transfer component appears, and a value of N u = 2.5 was obtained for laminar
flow.

Figure 6.13: Air flow around the rotor coils.

The heat transfer coefficient from the stator cylindrical surface was cal-
culated in the same way as in Design 1. The water flow (Q̇) is now 0.6
liter/minute, and the flow through half the cylinder is q̇ac = Q̇/2. The pa-
rameters for the heat transfer calculation in the stator are summarized in Ta-
ble 6.8. Compared to Design 1, the speed of the flow is increased since the
cooling surface is reduced. This results in an increase in the Reynolds and Nus-
selt numbers, and the heat transfer coefficient is 33% higher than in Design 1.
The coefficient from the surface of the rotor coils is calculated using a similar
formulation. The cooling area corresponds to the external surface of the coils
(parallel to the teeth body), and the length of the path to half the perimeter of
one coil, as described by the arrow lines in Figure 6.13. The calculations of the
dimensions of the d.c. coils is similar to those in the stator described in Chap-
ter 4. The cross-sectional coil area Adc (6.11) is however reduced by 15% in
order to allocate space between the coils for the axial airflow, as described above.
100 Chapter 6. Comparison with three alternative machines

This affected obviously the thickness and the resistance of the coil.

Adc = min(Aaxidc , Atandc ) · 0.85 (6.11)

The inlet temperature of the cooling air was selected to be equal to that of
the water. This ensures that there is no heat exchange between the cooling water
and the airflow, maintaining two separate circuits. The film temperature was
selected as the average between the inlet and outlet temperatures, as explained
in Chapter 4. For the calculation of the temperature increase in the rotor coils
only the copper losses were considered, and the air volumetric flow was selected
as the equivalent of 15 CFM. The coefficients obtained for the rotor are also
summarized in Table 6.8. The high temperature in the surface of the coils
increases considerably the film temperature. A 16% higher Reynolds number
is obtained compared to the water cooled circuit, although the Prandlt and
Nusselt numbers are significantly decreased. This results in a low heat transfer
coefficient, around one sixth of the value for water cooling.
It is assumed that the hot spot of the d.c. coils is located at the center of
the contact surface between the coils and the teeth insulation. The temperature
in the cooled surface of the coils is calculated using (6.12), where Acsdc is the
area equivalent to the surface of all coils, and P cudc corresponds to all the
copper losses in the rotor. The temperature in the hot spot is calculated using
(6.13), where Lihdc corresponds to the thickness of a d.c. coil and Aihdc to half
Acsdc . The temperatures in the different parts of the machine are summarized
in Table 6.9.

P cudc
T sfdc = + T∞2 (6.12)
Acsdc · hconvdc
P cudc · Lihdc
T hpdc = + T sfdc (6.13)
Aihdc · κcoil

The static and dynamic torque response at the thermal limit are shown in
Figure 6.14. The maximum torque is 8.0 Nm and the ripple is 25% of the peak
torque. The Ampere turns in one a.c. coil are 960 A, and in one d.c. coil 473
A. The total copper losses are 282 W (181 W a.c. and 101 W d.c.). The copper
losses account for 72% of all losses in the active parts of the machine at load.
The efficiency of the machine at 1500 r.p.m. is 76%.
6.4. Conventional EMSM 101

Table 6.8: Parameters for heat transfer calculation in Design 3

Coefficient Stator Rotor
Tm [o C] 47 84
∆ Tw [o C] 6 12
U [m/s] 0.039 10.0
Re 20897 24235
Pr 3.8 0.7
Nu 149 92
A 0.664 0.664
m 0.50 0.50
n 0.33 0.33
hconv [W/mK] 303 54

Table 6.9: Temperature distribution in Design 3 [o C]

Part a.c. path d.c. path
Tsf 63 136
Tcr 72
Twl 96
Tis 130
Thp 140 140
102 Chapter 6. Comparison with three alternative machines

1 5

m e c h a n ic a l to rq u e [N m ] 1 0

− 5

− 1 0
s ta tic
d y n a m ic
− 1 5
0 9 0 1 8 0 2 7 0 3 6 0
e le c tric a l a n g le [d e g ]

Figure 6.14: Torque response in Design 3.

6.5 Surface mounted PMSM

The fourth topology studied is shown in Figure 6.15, and it will be referred to
as ‘Design 4’. The rotor in Design 3 has been replaced with a surface mounted
permanent magnet rotor, while the stator is practically unchanged. Only the
slot opening was reduced to 8% of the tooth pitch to reduce the torque ripple.
Iron-powder is used everywhere although in practice the rotor yoke could be
laminated. However, that will have little effect in the results, as pointed out
above for Design 2 and 3.
Today most permanent magnet machines use neodymium-iron-boron (Nd-
FeB) magnets due to their high energy density, high remanence and good co-
ercitivity at relatively high temperatures (Sadarangani, 2000). However, they
are still more sensitive to very high or low temperatures compared to samarium-
cobalt magnets (SmCo), and they are also more sensitive to humidity, which
leads to corrosion problems. However, NdFeB magnets are approximately 20%-
30% cheaper than SmCo magnets, and the latter are mainly used in special
applications such as military equipment and robots. The permanent magnet
material selected for Design 4 is of the NdFeB type (Vacodym677), and its
demagnetization characteristics are shown in Figure 6.16.
Techniques to reduce the torque ripple in permanent magnet machines are
well reported in the literature, and a review is presented by Jahns and Soong
6.5. Surface mounted PMSM 103

a.c. coil
magnet

stator rotor

Figure 6.15: Design 4.

Figure 6.16: Demagnetization curves for the permanent magnet.

104 Chapter 6. Comparison with three alternative machines

(1996). From the design point of view, only skewing of the magnets and mini-
mizing the slot opening were explored. The magnets were skewed by half a slot
pitch. It was found in FEM that an optimal response was obtained choosing a
magnet arc of (k + 0.22) · τs , where k is the optimum integer number of teeth
under a pole (‘1’ in our case), and τs is the slot pitch. This value is slightly higher
than the one proposed in the literature, where the pole arc is usually selected as
(k + 0.14) · τs or (k + 0.17) · τs (Li and Slemon, 1988), (Ishikawa and Slemon,
1993). However, the optimal PM arc width depends also on other parameters
such as the magnetisation type, the airgap thickness, and the slot opening. The
airgap thickness was selected as 1 mm above the magnet to account for manu-
facturing tolerances of the magnets, glue thickness and the bandaging used for
protection and centrifugal forces. As pointed out above, the slot opening was
significantly reduced compared to the other designs, which has the drawback of
increased leakage between the teeth tips. This results in a lower average torque
at the cost of reduced torque ripple. The height of the magnet was adjusted in
order to obtain the same fundamental airgap flux density as in Design 1 at no
load (0.55 T), which corresponded to a height of 1.5 mm. It should be noted
that this value is higher than that for Design 2 (0.37 T) and Design 3 (0.39 T).

Protection against demagnetization

Permanent magnets can be partially or totally demagnetized if a strong magnetic
field is applied in the opposite direction of magnetization. In electrical machines
this phenomena is produced by the armature reaction, i.e. the field produced
by the currents in the armature coils. The point when demagnetization starts
varies with temperature, and it is given by the ‘knee’ in the demagnetization
curve as shown in Figure 6.16. The thickness of the permanent magnet has to
be sufficient to withstand these fields without suffering from demagnetization.
This minimum thickness can be easily calculated from a simplified MEC model
for the flux flow through a magnet, as shown in Figure 6.17 (Andersson, 2000).
The MEC models a flux-tube (Ostović, 1989) closing the path in the machine
through two adjacent magnets, although only one is represented in the figure.
The flux-tube has an infinitely small cross-sectional area, A, and both the link-
ing flux φ and the flux density Bm are constant along the tube. The magnet
is modelled as a m.m.f. in series with a reluctance. The m.m.f. is a function of
the magnet height hm and the coercivity Hc . The m.m.f. drop in the airgap is
given by Rg , and the m.m.f. from the armature reaction is Fi . In our case, at the
6.5. Surface mounted PMSM 105

g
Rg = µ0 A
φ = Bm A

Rm = hm
µr µ0 A

Fi = N I Fc = hm Hc

Figure 6.17: MEC model for magnet flux.

position of maximum torque, the contribution from each phase is Fa = I · N ,

Fb = −I/2 · N and Fc = −I/2 · N . The total contribution along the path
closed by both magnets is therefore Ft = 2N · I, and considering only one
magnet gives Fi = N · I. Solving the magnetic circuit yields (6.14).

N I + Bm µg0
hm−min = (6.14)
Hc − Bm
µr µ0

Assuming that the magnet working temperature is 150o C at most, the de-
magnetization point corresponds to a negative flux density (-0.2 T), and it was
selected that Bm = 0 to give certain margin before reaching this point. The
minimum magnet height calculated was 1.23 mm. In practice, and given the
lower permeability of the iron powder, the m.m.f. drop in the iron also should
be included, Rf e = lf e /(2 · µf e µ0 A), where lf e is the total length along the
iron of the flux tube defined by the two magnets, and µf e is the permeability
in the iron, selected at double the airgap flux density (1.1 T). The contribution
from this reluctance term is added to the one from the airgap and both are mul-
tiplied by Bm , which in this case is zero and therefore there is no difference in
the minimum calculated magnet height. It can be therefore concluded that the
magnet will be protected against demagnetization from the armature reaction.
In general, permanent magnet machines are designed for an airgap flux density
around 0.8 T, and the thickness of the magnet is considerably increased. The
106 Chapter 6. Comparison with three alternative machines

only reason for keeping the fundamental flux density at 0.55 T in this study is
for the sake of comparison mainly with Design 1.

Iron losses
The ratio between the alternating and rotating losses for the fundamental in
the stator is shown in Figure 6.18(a), and for the third harmonic in the rotor
in Figure 6.18(c). The results for the stator are very similar to those in De-
sign 3, although the flux density is now increased. The alternating losses are
concentrated in the body of the teeth, where the flux density is around 1.6 T.
The rotational losses are mainly located in the teeth tips (1.2 T) and in the
back core, where the decrease in the flux density (0.4 T) is compensated by a
greater iron volume. The losses in the permanent magnets were not calculated.
However, eddy-current losses are expected in sintered magnets since the insula-
tion between particles is minimized, and they may be problematic in high-speed
machines (Polinder and Hoeijmakers, 1997).
The losses per harmonic component in the stator and the rotor core at load
and no load are shown in Figure 6.19. The pattern is similar to Design 3,
with the highest losses due to the stator fundamental component. The rotor
losses are very small, and their main contribution is due to the third harmonic
at load. The total losses are increased by 45% when the machine is loaded. A
summary of the contribution to the total losses from the alternating and rotating
components in different parts of the machine is shown in Table 6.10. At no load
most of the alternating losses are due to the teeth body, and the rotational losses
are mainly distributed between the back-core and the teeth tips. At load, 69%
of the alternating losses are still located in the teeth body. Also, the rotational
losses increase considerably in the back-core of the rotor and the stator, while the
losses in the tips are slightly decreased. Overall, the rotational losses represent
69% and 62% of all the losses in the machine at no load and load respectively.

Thermal considerations
The thermal model for the stator is identical to that in Design 3. The only dif-
ference in the stator structure in Design 4 is the size of the slot opening, which
has no effect in the cooling. Therefore, the stator current and the temperature
distribution are the same as in Design 3. As pointed out above, the losses in
the magnets were not considered, and the losses in the rotor core are very little.
6.5. Surface mounted PMSM 107

(a) Stator RB , fundamental (b) Stator B, fundamental

(c) Rotor RB , 3rd harmonic (d) Rotor B, 3rd harmonic

Figure 6.18: Design 4 RB and flux density distribution for the fundamental in
the stator and the third harmonic in the rotor, no load.
108 Chapter 6. Comparison with three alternative machines

T o t: 6 4 , S t: 6 0 , R t: 4 W T o t: 9 3 , S t: 8 2 , R t: 1 1 W
8 0 8 0
s ta to r s ta to r
ro to r ro to r
6 0 6 0
Iro n lo s s e s [W ]

Iro n lo s s e s [W ]
4 0 4 0

2 0 2 0

0 0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r H a rm o n ic n u m b e r

(a) No load (b) Load

Figure 6.19: Distribution of the iron losses in Design 4 as a function of the har-
monic number.

Table 6.10: Summary of the predicted iron losses in Design 4. [W]

No Load Load
Part Alternation Rotation Alternation Rotation
St. back-core 4.4 14.4 7.2 22.8
St. teeth body 14.5 6.4 24.4 5.6
St. teeth tip 1.3 19.6 3.8 17.7
Rt. back-core 0.0 3.6 0.0 11.2
Total 20.2 44.0 35.4 57.3
6.6. Discussion 109

Therefore it was assumed that the rotor losses could either be neglected or as-
sumed to be dissipated through the airflow produced by a fan, as in Design 3.
The magnet selected shows a robust behaviour against temperature changes up
to 150o C, with the demagnetization point located in the negative flux density
region. The demagnetization curve was anyway selected at 150o C to account
for the maximum temperature that the magnet is expected to reach.
The static and dynamic torque response at the thermal limit are shown in
Figure 6.20. It can be observed that the static shape is smoother than in the
previous designs, and that the torque ripple is importantly reduced to about
9% of the peak torque, which is 10.5 Nm. As in Design 3, the Ampere turns
in one a.c. coil are 960 A, and the copper losses are 181 W. The copper losses
account now for 66% of all the losses in the active parts of the machine at load
(excluding the magnets). The efficiency of the machine at 1500 r.p.m. is 86%.

1 5

1 0
m e c h a n ic a l to rq u e [N m ]

− 5

− 1 0
s ta tic
d y n a m ic
− 1 5
0 9 0 1 8 0 2 7 0 3 6 0
e le c tric a l a n g le [d e g ]

Figure 6.20: Torque response in Design 4.

6.6 Discussion

A summary of the properties of all the topologies studied in this chapter is

shown in Table 6.11. The average torque will be used for comparison, and it
is calculated from the values of the peak torque and the torque ripple. This
ripple is highest in Design 1 and lowest in Design 4. The mass of the active
parts of the machine is also listed, together with that of the bakelite mould used
110 Chapter 6. Comparison with three alternative machines

to fix the claw-poles. It can be observed that Design 1 outputs most torque,
although its weight and volume are around 80% and 65% higher than in the
other designs respectively. This is mainly due to the end-plates used for the
magnetization. Design 1 presents also the highest iron and copper losses. The
rotor inertia in Design 1 and 2 was calculated including the bakelite mould,
with a density of 1.5 kg/m3 . This lower density helps Design 1 to keep the
lowest inertia. The inertia in Design 3 was calculated including the rotor coils,
and in Design 4 including the magnets. However, it should be noted that in a
commercial machine, the rotor in Design 4 would be probably built with holes
in the core to decrease its weight and inertia, but this has not been taken into
consideration in the comparison. The efficiency is calculated from the averaged
torque values, and it is highest in the PM machine. The per unit material cost
was calculated as a function of the mass of the active parts of the machine. The
price of the iron powder varies between 1 and 1.5 euro/kg. It was assumed that
the cost of the iron powder pressed into shape was 2 euro/kg (1 p.u.). The price
of copper was approximated as 5 p.u. (10 euro/kg), and the price of the NdFeB
magnet was 50 p.u. (100 euro/kg). With these figures it can be observed that
Design 4 is the most expensive machine, mainly due to the permanent magnets,
closely followed by Design 1. The ratio of average torque per weight is similar in
Design 1, 2 and 3, while it is around 40% higher in Design 4. Considering the
torque per volume, Design 4 leads by about 50%. The cost per torque output
is however also highest in Design 4, closely followed by Design 3, and around
32% higher than in Design 1 and 2.

6.7 Conclusions

Among the three electrically magnetized machines, the conventional design

(Design 3) presents the highest average torque per weight and per volume. The
difference with Design 1 is however less than 8%. Design 3 is also the most
expensive of the three topologies. Design 1 and 2 have a similar cost per torque,
which is about 18% lower than in Design 3. The PM machine (Design 4)
presents the best performance of all the machines regarding efficiency, and the
ratio of torque per weight and torque per volume, leading by around 12%, 40%
and 50% respectively. It is however also the most expensive machine consider-
ing the cost per output torque. Design 2 is the less expensive of all the machines
per output torque, closely followed by Design 1. The cost of Design 4 is about
32% higher. It should be noted that the cost of the slip-rings has not been
6.7. Conclusions 111

Table 6.11: Summary of the properties of all designs

Component Design 1 Design 2 Design 3 Design 4
Temp. rise [o C] 100 100 100 100
Outer diameter [mm] 200 200 200 200
Length [mm] 76.0 50.0 42.7 42.7
Peak torque [Nm] 14.4 7.8 8.0 10.5
Torque ripple [%] 33 28 25 9
Average torque [Nm] 12.0 6.7 7.0 10.0
Mass iron [kg] 9.7 5.5 4.7 5.6
Mass copper [kg] 2.8 1.5 2.4 1.7
Mass magnet [kg] 0 0 0 0.236
Mass mould [kg] 0.5 0.3 0 0
Total mass [kg] 13.0 7.3 7.1 7.5
Iron losses [W] 173 112 107 93
Copper losses [W ] 545 449 282 181
Total volume [lit] 2.38 1.57 1.34 1.34
Inertia [kg m2 ] 5.1e-3 9.8e-3 6.0e-3 6.9e-3
Efficiency [%] 72 65 74 85
Material cost [p.u.] 23.7 13.0 16.7 25.9
Torque/weight 0.92 0.92 0.99 1.33
Torque/volume 5.04 4.27 5.22 7.46
Cost/torque 1.98 1.94 2.39 2.59
112 Chapter 6. Comparison with three alternative machines

included in Design 3, neither the cost of the machine assembly, which may be
more problematic for the PM topology.
It is therefore shown that the PM machine presents superior electromag-
netic performance compared to the electrically magnetized machines at normal
operation. In fact, the main advantages of using electrical magnetization would
be at field weakening operation, not included in the comparison.
Chapter 7

Prototype and measurements

7.1 Introduction

In this chapter, the prototype construction process will be described, as well as

the results from the measurements. The parts of the prototype were manufac-
tured at ProEngCo AB, in Sweden. Some adaptations had to be introduced to
the original model to facilitate the manufacturing and the assembling process.
They will inevitably affect to some extent the performance of the machine.
A new cooling system was designed, which implied that the model for the
heat removal from the surface had to be adapted accordingly, and it is described
in this chapter. Differences in the material used for the main insulation and the
diameter of the conductors also affected the current loading at thermal limit,
so the iron losses had to be re-calculated. The control system was implemented
in dSPACE, and vector control was used to control the currents feeding the
prototype machine. Finally, an idea consisting of feeding the field winding
from the rectified three phase armature currents at the neutral point has also
been tested.

7.2 Construction of the prototype

The machine structure was divided into a series of components that were pro-
duced separately and then assembled together after inserting the coils. These
components were machined from blocks of pressed iron powder, which was
preferred for prototyping rather than developing the specific pressing tools for
each component. The iron powder Somaloy500 with the lubricating binder

113
114 Chapter 7. Prototype and measurements

0.6% LB1 was selected since it could be more easily machined compared to
the same powder using the lubricant 0.5% Kenolube, although at the expense
of a lower maximum permeability (300 vs. 500) and higher hysteresis losses.
On the other hand, LB1 gives higher transverse rupture strength (100 MPa vs.
50 MPa TRS) and higher resistivity. The powder was warm compacted at 800
MPa and 120o C, which increases the strength after pressing (green strength)
and improves the magnetic properties, mainly due to the higher density com-
pared to conventional compaction. The material was then cured at 250o C for
60 minutes to melt the lubricant and further increase the strength. It should be
pointed out that if Kenolube was used instead, then the heat treatment would
be at a higher temperature (500o C), which removes the stresses and decreases
the hysteresis losses. The temperature for the heat treatment should be kept
within the specified limits, since otherwise at higher temperatures the coating
between the particles is destroyed and the eddy current losses become too high.

Stator core
The stator core was divided through the middle into two rings with slots on
their common edge to allocate the teeth and to insert the cables connecting the
coils, see Figure 7.1. Each ring consisted of six modules glued together using
a high temperature epoxy glue. The layer of glue between the modules was
0.02 mm thick. The teeth were machined into one piece each. The top of
the part of the teeth inserted into the core had a salient edge to withstand the
attraction forces in the airgap and hold the teeth in position. This edge did
not extend over the actual width of the teeth in order to allow the insertion
of the pre-pressed coils from the top. However, the idea of pressing the coils
was later abandoned and the coils were directly wound around the teeth. The
wound teeth were inserted into one of the rings, and thereafter the second ring
was placed on top. The length of the gap between the core back and tooth was
controlled to be within 0.05±0.02 mm, and would therefore have negligible
influence upon the magnetic performance. It should be noted that the teeth
and the core were not glued together. This was convenient since they could be
easily dismantled during assembly.

End-plates
The end-plates were built from six modules and they were attached to an alu-
minium plate used for cooling, see Figure 7.2. The end-plates expanded until
7.2. Construction of the prototype 115

Figure 7.1: Stator core sides and teeth mounted.

the outer diameter of the machine, and they were screwed into the stator hous-
ing. At the level of the stator core, the surface was machined with a 45o slope,
matching the corresponding core surface when mounted. The d.c. coils were
accessed through two axial slots machined in the stator core.

Rotor

All the claw-poles with the same polarity formed a structure as shown in Fig-
ure 7.3 (built from four modules each). The two structures were glued onto
a bakelite mould constructed to match their geometry. The finished rotor is
shown in Figure 7.4. It should be noted that the bakelite mould extended be-
yond the inner radius of the claw-pole structures, providing magnetic insulation
between the claw-poles and the shaft in order to minimize the homopolar flux.
Two nuts screwed on the sides of the shaft were used to adjust and fix the axial
position of the rotor when mounted.
116 Chapter 7. Prototype and measurements

Figure 7.2: End-plate.

Figure 7.3: Rotor parts.

7.2. Construction of the prototype 117

Figure 7.4: Finished rotor.

Windings

The number of turns in the a.c. winding were adjusted to keep the maximum
d.c. link voltage below 300 V, as required for the laboratory equipment. The
linking flux from one phase at no load and load calculated in FEM is shown in
Figure 7.5. For 54 turns per coil, the corresponding phase and phase to phase
voltages at 1500 r.p.m. are shown in Figure 7.6.
Each one of the 54 turns around a tooth consisted of three parallel wires
with a bare diameter of 0.63 mm each, and a coating of 0.015 mm. For the
insulation in the slots DMD was used, which consists of three polyester layers
(fibre - film - fibre). The thickness of the insulation layer in the body of the
teeth was 0.16 mm, and in the back core 0.40 mm. The practical slot fill
factor was 48%. It should be noted that the slots could not be completely
filled since the conductors had to be accessed radially from outside the stator
back-core due to the space limitation in the axial direction. The holes were
machined at the middle of the core between two teeth for the sake of symmetry,
but this complicated a great deal the insertion of the cables. Therefore, only
one cable was inserted through each hole, to connect either the beginning or
118 Chapter 7. Prototype and measurements

− 3
x 1 0
n o lo a d
2 lo a d

1
F lu x [V s ]

− 1

− 2

0 9 0 1 8 0 2 7 0 3 6 0
E le c tric a l a n g le [d e g ]

Figure 7.5: Linking flux in one phase from FEM.

3 0 0 3 0 0
n o lo a d n o lo a d
lo a d lo a d
2 0 0 2 0 0

1 0 0 1 0 0
V o lta g e [V ]

V o lta g e [V ]

0 0

− 1 0 0 − 1 0 0

− 2 0 0 − 2 0 0

− 3 0 0 − 3 0 0
0 1 e− 3 2 e− 3 3 e− 3 4 e− 3 0 1 e− 3 2 e− 3 3 e− 3 4 e− 3
T im e [s e c ] T im e [s e c ]

(a) Phase voltage (b) Phase to phase voltage

Figure 7.6: Induced voltage derived from FEM.

7.2. Construction of the prototype 119

the end of the tooth coil, while the connection to the next coil in the phase
was done inside the machine. In this way, half of the connections were made
outside the machine and the other half inside. It was preferred to avoid long
loose connections between the coils inside the machine since there is virtually
no air-flow to cool these cables.
The number of turns in the d.c. coils was adjusted so as to make possible
to connect the machine with series magnetization (Alaküla, 2004). Each end-
plate was wound with 74 turns of a single wire with 1.20 mm bare diameter and
a coating of 0.04 mm. The main insulation was 0.24 mm thick. The slot fill
factor for the d.c. winding was 58%. The a.c. and d.c. coils were impregnated by
soaking with Ultimeg 2000 enamel and cured for 6 hours at 130o C. A picture
of a finished d.c. coil attached to the end-plate is shown in Figure 7.7.

Figure 7.7: Finished end-plate.

A thermocouple sensor was inserted in six a.c. coils (two per phase) and the
two d.c. coils to measure the temperature. The a.c. coils equipped with sensors
were evenly located every four teeth around the stator. The sensor was a cable
consisting of two conductors welded at the measuring point, which was located
in the middle of the coils. The sensor cable was of the type K manufactured by
Pentronic AB in Sweden.
120 Chapter 7. Prototype and measurements

Cooling system

A pump was used to provide a water coolant flow of approximately 4 liter/minute.

The coolant in turn was refrigerated by water in a heat exchanger (Alfa Laval
DOC14-20H) before being recirculated into the machine. The water was taken
directly from the tap and was disposed after leaving the heat exchanger. An alu-
minium cover was screwed on the side of the aluminium housing mounted on
an end-plate. The coolant circulated in the cavity between the cover and the
housing and was then conducted through axial ducts machined in the housing
of the stator core, one above each tooth. The flow coming out from the other
side of the ducts circulated around the cavity of the second end-plate before
leaving the machine through a valve on the top. A CAD plot of the machine
and the cooling system is included in Figure 7.8. The finished stator with the
housing is shown in Figure 7.9, where the axial ducts for the cooling are visi-
ble. It can be seen also the cables introduced radially through the core for the
connection of the coils.

Figure 7.8: Plot of the cooling system in the machine.

7.3. Adaptation of the thermal model 121

Figure 7.9: Finished stator with housing.

7.3 Adaptation of the thermal model

The thermal model for the heat removal from the surfaces of the stator core and
end-plates had to be adapted to the new cooling system used. The model for the
heat transfer from the center of the coils to these surfaces is however identical
to that described in Chapter 4.

Heat transfer coefficient

The heat transfer coefficient from the surface of the end-plates was calculated
using the formulation for forced convection over a plate. Figure 7.10 shows the
end-plate and half the number of outlets from the axial ducts, represented by
small holes located around the periphery of the plate. Symmetry was exploited,
and only half the surface was considered. The idea was to define an equivalent
duct with an area corresponding to the cooling area divided by the number
of ducts. The equivalent length was approximated as the mean average length
from the center of the holes to the machine outlet, located at the level of the
upper hole. The dashed lines in the figure indicate the path for the water flow
122 Chapter 7. Prototype and measurements

from some of the holes to the outlet. It can be observed that, for example,
for hole number 13 the path is circular when it arrives at the inner radius of
the plate, so the length was increased accordingly. The coolant flow through
the equivalent duct corresponded simply to the flow from one axial duct. Its
cross-sectional area was calculated dividing the equivalent cooling area by the
equivalent length of the duct and multiplying by the thickness of the cavity
where the coolant flows between the end-plate and the cover.

L4 outlet

L7

Apt
L10
L13

Figure 7.10: Cooling of the end-plate.

The heat transfer coefficient was calculated similarly in both end-plates, the
only difference being the mean temperature of the coolant, which affects its
properties. The temperature increase was calculated when the coolant leaves
the first end-plate, the axial ducts and the second end-plate. The increase in
temperature at each stage was calculated using the formulation in (4.29). The
inlet temperature was known (11o C), and applying the formulation to the first
plate resulted in the temperature of the coolant entering the axial ducts. In the
same way, applying the formulation to these ducts yielded the coolant temper-
ature entering the second end-plate, and finally the outlet temperature. The
properties of the coolant using (4.29) were calculated at the inflow temperature
at each stage. The properties of the coolant used to calculate the heat transfer
coefficient were calculated at the average temperature between the inflow and
the outflow at each stage.
The heat transfer coefficient from the axial ducts expanding through the sta-
tor back-core was calculated using the formulation for forced convection inside
a tube. The geometric characteristic was the diameter of the duct D (4 mm). It
7.3. Adaptation of the thermal model 123

was found that the flow in the ducts was laminar, and the formulation in (7.1)
(Sieder and Tate, 1936) was used to calculate the Nusselt number. In this equa-
tion L is the length of the tube and µs is the viscosity of the fluid evaluated at
the temperature of the interior surface of the tube. Obviously this temperature
was a priori unknown, so an iteration process had to be implemented.

1/3 0.14
D µ
N u = 1.86 · (ReD · P r)1/3 · · (7.1)
L µs

The experiments to check the thermal model were carried out at 500 r.p.m.
The Ampere turns in one a.c. coil were 827 A, and in one d.c. coil 984 A. A
summary of the values obtained for the heat transfer coefficient calculation is
shown in Table 7.1. The predicted iron losses are shown in Figure 7.11.

Heat transfer model

The heat transfer model up to the surface of the stator core and the end-plates
was identical to that presented in Chapter 4. Only the diameter of the copper
conductors and their insulation was updated, as well as the thickness and the
thermal properties of the main insulation (DMD). According to the manufac-
turer, the thermal conductivity of DMD was 0.041 W/mK, which was about
one third than that for Kapton (0.12 W/mK). This negative effect in the ther-
mal performance of the machine was partially compensated by the reduction in
the insulation thickness, from 0.5 mm assumed in the models to 0.15 mm in
the teeth and 0.24 mm in the end-plates of the prototype.
The temperature at the inner surface of the axial ducts was calculated using
(7.2). All the losses from the iron-powder and the copper are cooled through
this path. No losses were assumed in the aluminium casing. The area Aduct cor-
responded to the interior area off all the ducts in the casing together, and T∞d is
the average temperature of the coolant in the ducts. Similarly, the temperature
in the surface of the aluminium casing attached to the end-plate was calculated
using (7.3). The area Acas corresponded to the surface area of the casing, and
T∞c was the average temperature of the coolant in the corresponding plate. The
value of hconvdc was selected depending on the plate analyzed.

P cuac + p · (P f e1ac + P f e2ac ) + P f erC

T alac = + T∞d (7.2)
Aduct · hconvac
124 Chapter 7. Prototype and measurements

Table 7.1: Parameters for the heat transfer calculation, 500 r.p.m.
Coefficient End-plate 1 Back-core End-plate 2
Tm [o C] 11.3 12.3 13.3
∆ Tw [o C] 0.54 1.48 0.54
U [m/s] 0.15 0.22 0.15
Re 25170 726 26457
Pr 7.89 8.63 7.46
Nu 208 13 209
hconv [W/mK] 667 1947 674

T o t: 6 0 , S t: 5 1 , R t: 9 W
3 0
s ta to r
2 5 ro to r
Iro n lo s s e s [W ]

2 0

1 5

1 0

0
1 2 3 4 5 6 7 8 9 1 0 1 1
H a rm o n ic n u m b e r

Figure 7.11: Predicted iron losses at load, 500 r.p.m.

7.3. Adaptation of the thermal model 125

P cudc /2 + P f edc + P f erT

T aldc = + T∞c (7.3)
Acas · hconvdc
The temperature in the surface of the back-core was calculated using (7.4),
where Lldac is the average length between the top and the bottom distance of
the duct to the core surface, following the edge of the duct. The value of Aldac
corresponds to the average between the surface of the back-core (Asfac ) and the
inner surface of all the ducts together (Aduct ). The value of κal is the thermal
conductivity of the aluminium (165 W/mK). For the end-plates (7.5) was used,
were Llcdc is simply the thickness of the casing attached to the end-plates (5
mm), and Alcdc is the area of the cooling surface in the aluminium casing. The
results at the different parts of the machine are summarized in Table 7.2, for the
current used in the experiments.

(P cuac + p · (P f e1ac + P f e2ac ) + P f erC )Lldac

T sfac =
Aldac · κal
+T alac (7.4)
(P cudc /2 + P f edc + P f erT )Llcdc
T sfdc = + T aldc (7.5)
Alcdc · κal
The end-plate that is cooled first in the thermal circuit will be referred to as
the first end-plate, and that where the outlet is located will be referred to as the
second end-plate. From the experiments, the measured temperature in the coil
attached to the first end-plate was 149o C, while the temperature in the coil in
the second end-plate was 126o C. The d.c. coils were wound radially onto the
end-plates, and they were fixed axially using enamel. One of the two ends in
each coil was inserted in a narrow slot machined in the end-plate to facilitate
the winding process. During assembly, the insulation of this wire in the first
end-plate was damaged, originating a short-circuit. Therefore, the coil had to
be separated from the plate for reparation and then glued back. This process
probably affected the fitting of the coil into the plate, and in turn the thermal
path. Actually, this coil should present a lower temperature than the second
d.c. coil, since it is cooled by the water coming directly from the pump at a
lower temperature. Compared to the prediction, the measured temperature in
the second coil is 19% higher. It should be noted that the copper is not pressed
naturally into the side of the plate during the winding process. The fact that a
thermal sensor was included in the coil did not allow to keep an ordered layer
of wires on top of each other and the proper fit into the side of the plate. The
126 Chapter 7. Prototype and measurements

Table 7.2: Simulated temperature distribution [o C]

Part End-plate 1 Back-core End-plate 2
Tal 19.6 19.8 21.5
Tsf 19.8 20.2 21.7
Tcr 28.6
Twl 38.7 63.8 40.7
Tis 89.0 103.7 91.0
Thp 104.4 113.0 106.3

layer of enamel used for this purpose provides further insulation, which in turn
decreases the efficiency of the heat transfer from the coils to the iron.
The temperature from the six armature coils equipped with sensors dif-
fered between each other by up to 10o C at steady state when the machine was
loaded. The average temperature was 138o C, which is 22% higher than pre-
dicted. However, the models assume that 20% of the copper losses from the
armature coils are transferred directly to the back-core through the top of the
coil. When the machine was assembled, an empty space around 1 mm was
left between the top of the coils and the back-core in order to avoid squeezing
loose wires when the second core ring was mounted. This, added to the thicker
insulation on the back-core (0.4 mm) effectively isolates the top of the coil.
If 100% of the copper losses are transmitted through the teeth, then the pre-
dicted temperature is 133o C instead of 113o C, which is less than 4% difference
compared to the average measured value. The accuracy in the iron loss calcula-
tion also affects the difference between the predicted and measured temperature
distribution. However, for example the iron losses in the a.c. path need to be
increased by 20% in order to increase 1.3o C the temperature in the armature
coils, so their influence in this case is minor.

7.4 Experimental set-up and instrumentation

The prototype was loaded with a d.c. machine through a torque transducer of
the type Fast Technology TM-HR-Rd-20, see Figure 7.12. The nominal power
of the d.c. machine was 2.2 kW at 1500 r.p.m. Also shown in the picture
7.5. Control 127

are four of the eight multimeters used to measure the temperature from the
sensors. The field winding in the machine was fed from a d.c. supply of the
type SM 70-22 manufactured by Delta Elektronika. A Tektronix TDS 640 A
oscilloscope was used for the voltage measurements. An inductance meter of the
type ELBRO LC-6043 was used to measure the inductance in the tooth coils.
Each machine was connected to a converter, where the switches were controlled
from a digital card connected to a PC using dSPACE. Analog modules were
used to estimate the speed and the flux in the machines and to measure their
current. Figure 7.13 shows the hardware used to control the machines.

Figure 7.12: Test-bench used for the measurements.

7.5 Control
The control system was implemented in S IMULINK, see Figure 7.14, and then
compiled into dSPACE for the real-time control. The measured values are in-
troduced in the model through the ‘Feedback’ block. The output from the
PWM modulators are connected to the converters through the dSPACE inter-
face. It was possible to connect any of the signals from the S IMULINK model
into the dSPACE user interface for visualization and real-time adjustments of,
for example, the references and controller parameters. The d.c. machine was
128 Chapter 7. Prototype and measurements

Figure 7.13: Set-up for the control equipment.

7.5. Control 129

speed controlled, while the prototype was torque controlled. A PI controller

was used to set the reference for the torque in the d.c. machine depending on
the difference between the measured and reference speeds. The speed controller
parameters were calculated using symmetric optimum (Alaküla, 2002). The
proportional gain was Kpw = J/(a · Tf w ), where J is the inertia, Tf w is the
speed filter time constant, and a is a constant controlling the relative damping
of the poles. With a = 3, the speed control system will be on the limit of com-
plex poles. The integral time constant was Tiw = a2 ·Tf w , and the integral gain
was Kiw = 1/Tiw . The torque reference output from the speed controller was
transformed into current (ia ) dividing by the nominal flux in the machine, and
another PI controller was used to control this current. The current controllers
for the d.c. machine and the prototype were deduced using predictive dead beat
control (Alaküla, 2002). For the d.c. machine, the controller equation is given
by (7.6), where the proportional and integral components can be distinguished
clearly, and the e.m.f. (ea (k)) represents the so-called feed forward term. The
values of La and Ra correspond to the inductance and resistance of the wind-
ing in the d.c. machine, and Ts is the sampling time (corresponding to 5 kHz).
The reference voltage u∗a is modulated with a triangular carrier wave, which sets
the switching pattern in the converter connected to the machine.

∗ La Ra Ts
ua = + · (i∗a (k) − ia (k)) + ·
Ra 2 La /Ra + Ts /2
n=k−1

· (i∗a (n) − ia (n)) + ea (k) (7.6)
n=0

Vector control was used to control the currents in the prototype. The theory
of vector control is well reported in the literature (Vas, 1990) (Novotny and
Lipo, 1996), and the basics are explained in the second part of this thesis. It is
based on the concept that alternating currents in the machine can be expressed
as a rotating vector. From a synchronous coordinate system attached to the
stator or the rotor flux, the rotating vector is expressed as the combination of
two d.c. current components along the x (isx ) and y (isy ) axis of the rotating
frame, which can be controlled independently. In the prototype, the equations
for the PI controllers governing the current are given in (7.7) and (7.8). It can
be observed how the reference voltage for one axis is dependent on the current
in the other axis through the feed forward term, where ωr is the rotor speed,
and Ψm is the linking flux in the machine. The location of the rotating frame is
 130
57,'DWD GFOLQNYROWDJH

8GF

8GF

Z Z 7 7 LD LD
XD XD
VSHHG 7!,D Z
UHIHUHQFH
3,VSHHGFRQWURO LD 3:0'&0
DUPDWXUHFXUUHQW
FRQWUROOHU 3,(FRQWUROOHU VDPSOHGDUPDWXUHFXUUHQW
DGMXVWPHQW

DGM

8GF
 8GF
8GF
LV[ DGM D] YD
OLPLW
D
LV SK
XV E E] YE
Z
SK F
UHVROYHU
F] YF
 YROWDJHYHFWRU LV
LV
6\PPHWUL]H 3:03060
LV\ OLPLW 9HFWRUFRQWURO

8GF

Figure 7.14: Structure of the control system.

FXUUHQWYHFWRU LD

VDPSOHGDOSKDDQGEHWD SK
LSKD
FRPSRQHQWVRIVWDWRUFXUUHQWV
SK
LSKE

VDPSOHGFRVDQGVLQFRPSRQHQWV
UHVROYHU
RIHOHFWULFDQJXODUSRVLWLRQ
VDPSOHGDQGILOWH 
Z
UHGDQJXODUVSHHG V
)HHGEDFN
VSHHGILOWHU
Chapter 7. Prototype and measurements
7.5. Control 131

measured using a resolver mounted in the machine. The actual values of isx and
isy are obtained from the measured three phase currents. These currents are first
transformed to the two phase stationary α/β system and then to the rotating
x/y frame using the angle from the resolver. The conversion from three to
two phases was done using the power invariant transformation (Alaküla, 2002).
The output reference voltage from the controller is transformed back to three
phases, which are symmetrized using third harmonic injection. This increases
the modulation depth by 15%. Finally, the resultant voltages are modulated
using a triangle carrier wave, which sets the switching pattern in the converter
connected to the prototype.

Lsx Rs Ts
u∗sx = + · (i∗sx (k) − isx (k)) + ·
Ts 2 Lsx /Rs + Ts /2
n=k−1

· (i∗sx (n) − isx (n)) − ωr · Lsy · isy (k) (7.7)
n=0

Lsy Rs Ts
u∗sy = + · (i∗sy (k) − isy (k)) + ·
Ts 2 Lsy /Rs + Ts /2
n=k−1

· (i∗sy (n) − isy (n)) + ωr · (Ψm + Lsx · isx (k)) (7.8)
n=0

The values of the inductances Lsx and Lsy where derived from the linking
flux at load and no load calculated in FEM (Figure 7.5). The general equation
for the stator flux is given in (7.9), where if is the field current. At load, Ψs
corresponds to the maximum value of the curve for the load flux. From (7.9),
the particular case when there is no armature current can be expressed as in
(7.10), from where Lsx is calculated. The value of Ψnl corresponds to the
maximum value of the curve for the no load flux. The equation when the no
load flux is zero is given in (7.11). The value of Ψs0 corresponds to the load
flux at the point when the no load flux is zero. Finally (7.12) expresses the x/y
currents as a function of the phase current vector. Combining (7.9), (7.11) and
(7.12), and knowing the value of Lsx from (7.10) yields the value of Lsy , isx
and isy . The values obtained are Lsx = 5.54 · 10−3 H, Lsy = 6.41 · 10−3 H,
isx =-6.2 A, and isy =17.70 A. The currents correspond to a total Ampere turns
of 827 A in one a.c. coil (1082 A in one d.c. coil), as used for the FEM results
in Figure 7.5. Finally, is should be noted that the value of the resistance of one
132 Chapter 7. Prototype and measurements

phase was Rs = 0.8 Ω.

Ψs 2 = (Lsx (if + isx ))2 + (Lsy · isy )2 (7.9)

Ψsnl = Lsx · if (7.10)
2 2 2
Ψs0 = (Lsx · isx ) + (Lsy · isy ) (7.11)
2 2 2
is = isx + isy (7.12)

7.6 Measurements
Coil inductance
The inductance in the coil in every tooth was measured with the teeth standing
alone and when they were mounted in one core ring (half core). The results
are shown in Figure 7.15. The reference for the level of the half core values
is zero. The difference in the standard deviation between the maximum and
minimum inductance of the teeth alone was 8%. When they were mounted in
half the core this difference was 14%. Around half the coils were unwound and
the number of turns was controlled to be 54 in all of them. When they were
re-wound, they had the same old inductance. Therefore, the differences should
be due to the teeth rather than the winding. It is possible that the bulk material
from where the teeth were machined (six teeth per bulk) was not completely
homogeneous, which in turn would affect the magnetic properties of the teeth.

If the teeth were mounted in the order shown in Figure 7.15, the standard
deviation of the total inductance of phase A, B and C would be 0.3%, -3.9%
and 3.6% respectively. Simply by swapping teeth number 2 and 21, this devia-
tion was reduced to 0.3%, 0.0% and -0.3%. In order to get a smoother change
in the inductance of every phase and between the phases around the machine
the tooth pair 7-10 was swapped in phase A, 14-17 in phase B, and 6-15 and
15-21 in phase C. The goal was to distribute the coils so that the average be-
tween two adjacent positions in one phase was as close as possible to the mean
average of the eight positions in that phase. Also, that the addition of the in-
ductance of the eight groups of three adjacent coils (corresponding to different
phases) is as constant as possible. Only six of the eight teeth in every phase
could be freely swapped, since the other two were provided with a thermal sen-
sor to measure the temperature at a particular position around the machine, and
they only could be swapped with each other. Figure 7.16 shows the distribution
7.6. Measurements 133

0 .5
A lo n e
0 .4 5 H a lf c o re

0 .4

0 .3 5
In d u c ta n c e [H ]

0 .3

0 .2 5

0 .2

0 .1 5

0 .1

0 .0 5

0
1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 2 1 2 2 2 3 2 4
T o o th n u m b e r

Figure 7.15: Inductance measurements in the teeth coils.

of the inductance in the eight consecutive teeth in each phase before and after
the teeth swapping. The total inductance of each group of three adjacent teeth
around the machine is shown in Figure 7.17.

Induced voltage

The induced voltage at no load was measured with the machine rotating at
nominal speed (1500 r.p.m.) driven by the d.c. machine. The voltage was
measured in an oscilloscope and saved for further post-processing. The three
phases produced the same induced voltage, although it was observed that the
peak voltage was 20% lower than predicted. However, the shape resembled
reasonably the simulated results, as shown in Figure 7.18. In this figure the
measured voltage has been amplified by 20% in order to match the amplitude
from the simulated curve. The reasons for the lower induced voltage could be
a lower permeability in the iron-powder material, and/or a distortion of the
magnetic circuit. It was observed that the shaft was magnetic, which indicates
the existence of homopolar flux. The rotor claw-pole structures were isolated
134 Chapter 7. Prototype and measurements

0 .4 0 .4

0 .3 8 0 .3 8
In d u c ta n c e [H ]

In d u c ta n c e [H ]
0 .3 6 0 .3 6

0 .3 4 0 .3 4

0 .3 2 P h ase A 0 .3 2 P h ase A
P h ase B P h ase B
P h ase C P h ase C
0 .3 0 .3
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
T o o th p o s itio n T o o th p o s itio n

(a) Original inductance. (b) Inductance after swapping.

Figure 7.16: Distribution of the inductance along the eight coils in each phase.

1 .2

1 .1
In d u c ta n c e [H ]

0 .9

o rig in a l
s w a p p in g
0 .8
1 2 3 4 5 6 7 8
T o o th p o s itio n

Figure 7.17: Sum of the inductances in the eight groups of three coils.
7.6. Measurements 135

from the shaft through a 5 mm layer of bakelite. The space between the inner
radius of the end-plates and the surface of the shaft was 6.5 mm. It is possible
that there is a leakage flux to the shaft directly from the end-plates, but it is more
likely to travel through the bearings. This leakage path was not considered in the
simulations, and it is likely the cause of the disagreement between the predicted
and measured induced voltage.
There are other constructional details that can affect the magnetic circuit.
First of all, the holes between the teeth used to introduce the conductors have
a dimension of 5x5 mm and they were not included in the model. These holes
affect the linking flux flowing between the teeth through the core in the circum-
ferential direction. However, their influence in the axial magnetization path is
probably not too significant since they are located exactly in the middle of the
core. Another source of disagreement could be the possible air gaps that appear
between the core rings and between the core and the end-plates if they are not
adjusted properly. However, with a proper mechanical design these gaps will be
considerably too small to have a significant effect. In general, a lower induced
voltage indicates a decrease in the linking flux, which in turn will also reduce
the torque delivered by the machine.

1 5 0 2 0 0
s im u la te d s im u la te d
m e a su re d 1 5 0 m e a su re d
1 0 0
1 0 0
5 0
5 0
V o lta g e [V ]

V o lta g e [V ]

0 0

− 5 0
− 5 0
− 1 0 0
− 1 0 0
− 1 5 0

− 1 5 0 − 2 0 0
0 1 e− 3 2 e− 3 3 e− 3 4 e− 3 5 e− 3 0 1 e− 3 2 e− 3 3 e− 3 4 e− 3 5 e− 3
T im e [s e c ] T im e [s e c ]

(a) Phase voltage (b) Phase to phase voltage

Figure 7.18: Comparison between the simulated and measured induced voltage,
1500 r.p.m. Measured voltage amplified by 20%.
136 Chapter 7. Prototype and measurements

Average torque
The average torque from the prototype was measured with a torque sensor. It
could also be calculated by multiplying the current in the d.c. machine by its
magnetizing flux (1.35 Vs). The accuracy of this procedure was within 1 Nm
compared to the value from the transducer. The torque was measured at four
levels of magnetization up to the nominal value, and different combinations of
the isx and isy armature currents. The maximum value of these currents was
selected so that the resultant current in all the combinations did not exceed the
value at thermal limit by more than 10%. The results are shown in Figure 7.19.
It can be observed that the shape is very similar in all the cases, although the
torque increases for higher field currents. The output torque from the prototype
is governed by (7.13) (Alaküla, 2002). In the prototype, Lsy is higher than Lsx .
For a given isy , the absolute value of the torque increases for a negative isx and
decreases for a positive isx , as can be observed from the experimental results.
T = Ψm · isy + (Lsx − Lsy ) · isx · isy (7.13)
The average torque measured for isx =-6.2 A, isy =17.7 A, and if =14.6 A was
10.4 Nm. The simulated torque response for these currents is shown in Fig-
ure 7.20. The average value of the torque considering the ripple is 11.88 Nm,
which means that the measured torque is 14% lower than the estimated from
FEM. This is not unexpected, since the linking flux is also reduced compared
to the measurements as indicated above.

Iron losses
The core losses were measured using the calibrated d.c. machine as a prime
mover. The mechanical losses in the system were measured with the two ma-
chines coupled together and with no excitation in the synchronous machine.
The input power to the d.c. machine was measured for speeds up to 1500 r.p.m.
in steps of 37.5 r.p.m. The speed was determined from the electrical frequency
measured from the resolver output signals. This procedure was then repeated
with field excitation in the synchronous machine (if =14.6 A). The difference
between the input power from both experiments minus the resistive copper
losses in the d.c. machine yielded the core losses at no load. The measured
mechanical and core loss energy are shown in Figure 7.21. For comparison, the
predicted losses in FEM are also included in the plot. The loss energy is simply
obtained dividing the losses (W) by the electrical frequency (Hz).
7.6. Measurements 137

3 6
2 4
T o rq u e [N m ]

1 T o rq u e [N m ] 2
0 0
− 1 − 2
− 2 − 4
− 3 1 5 − 6 1 5
1 4 1 0 1 4 1 0
1 0 5 5 1 0 5 5
0 0 0 0
− 5 − 5 − 5 − 5
− 1 0 − 1 4 − 1 0 is x [A ] − 1 0 − 1 4 − 1 0 is x [A ]
is y [A ] − 1 5 is y [A ] − 1 5

(a) Field current 3.7 A (b) Field current 7.3 A

8
9
6
4 6
T o rq u e [N m ]
T o rq u e [N m ]

2 3
0 0
− 2 − 3
− 4
− 6
− 6
− 9
− 8
1 0 1 5
1 5
1 4 1 0 1 4 1 0
1 0 5 5 5 5
0 − 5 − 5
0 0 − 5 − 1 0 − 5 0
is y [A ] − 1 0 − 1 4 − 1 5
− 1 0 is x [A ] is y [A ] − 1 4 − 1 5− 1 0 is x [A ]

(c) Field current 11.0 A (d) Field current 14.6 A

Figure 7.19: Measured average torque for different field currents.

138 Chapter 7. Prototype and measurements

1 5

1 0
m e c h a n ic a l to rq u e [N m ]

− 5

− 1 0
s ta tic
d y n a m ic
− 1 5
0 9 0 1 8 0 2 7 0 3 6 0
e le c tric a l a n g le [d e g ]

Figure 7.20: Simulated torque response for isx =-6.2 A, isy =17.7 A, if =14.6 A.

0 .8
e n e rg y p e r re v o lu tio n [J ]

0 .6

0 .4

0 .2
p re d ic te d c o re lo s s e n e rg y
m e a s u re d c o re lo s s e n e rg y
m e a s u re d m e c h a n ic a l lo s s e n e rg y
0 0 1 2
1 0 1 0 1 0
fre q u e n c y [H z ]

Figure 7.21: Measured and predicted losses at no load.

7.6. Measurements 139

It can be observed how the mechanical loss energy increases more rapidly
with the frequency, while the predicted and measured core loss energy follow a
more similar pattern. However, the overall measured core loss is considerably
reduced compared to the predictions. The hysteresis energy loss is independent
on the frequency, and the measured value is 70% lower than the estimated (at 0
Hz). The dynamic loss energy (eddy current and anomalous) is a function of the
frequency, and the estimated value is 45% lower than the measured at 200 Hz.
It should be pointed out that the predicted response in Figure 7.21 has been ob-
tained adapting the coefficients in the loss formulation to the lubricating binder
LB1. Indeed, the coefficients presented by Guo et al. (2003) (Chapter 3) are
for the lubricant Kenolube. Skarrie (2001) measured the coefficients for the
alternation losses in a powder core inductor, and they were used to estimate the
losses due to alternation in the machine, see Table 7.3 (Cha , Cea , Caa , h). The
hysteresis losses due to rotation where calculated using Guo’s formulation and
then multiplied by the coefficient Cf f which is the ratio between the alternat-
ing hysteresis component from Skarrie and Guo. The eddy current rotational
component (Cer ) was already zero in Guo’s formulation, and the anomalous
component (Car ) was also set to zero since it could not be deduced from the
available data.
One of the reasons contributing to the mismatch in the losses is the fact
that the real machine is actually less magnetized compared to the model in the
simulations, as it was observed from the induced voltage measurements. The
hysteresis losses are proportional to B h , where B is the flux density. If the
linking flux in the prototype was increased by 20% to match that in the FEM
model, and assuming that this increase is evenly distributed in all parts of the
machine, then the measured hysteresis losses would be increased by 34%. Nev-
ertheless, the hysteresis losses would still be 60% lower than the estimated. On
the other hand, the dynamic losses would also be considerably increased since
the eddy current losses are proportional to B 2 and the anomalous losses to B 1.5 .
This would increase the curvature of the measured response at higher frequen-
cies, indicating that the dynamic losses would be even more underestimated.
However, in the formulation for the estimation of the dynamic losses only the
anomalous losses due to alternation are included. The eddy current losses due
to alternation and rotation as well as the rotational anomalous losses are ne-
glected for LB1, which is probably the reason for the lower predicted dynamic
losses. Bulk eddy current losses are also a function of the geometry, as indicated
by Lange et al. (2000) and Hofer (2004). The weight of the SMC parts in the
140 Chapter 7. Prototype and measurements

Table 7.3: Coefficients for the iron loss calculation

Coefficient Adapted value
Cha 0.16
Cea 0
Caa 4.38· 10−3
h 1.6
Cf f 0.16/0.14
Cer 0
Car 0

machine presented by Guo et al. (2003) is around half compared to the pro-
totype in this thesis, and therefore higher dynamic losses would be expected in
the measurements compared to the estimations.
Finally, it should be noted that the harmonic loss formulation proposed in
this thesis does not consider the bias field of the d.c. excited core. Therefore, all
the harmonics are assumed to follow a symmetrical major hysteresis loop (with
different amplitudes) around the center B=0, H=0. However, if the harmonic
has a bias d.c. component, there will be an asymmetric behaviour, and even
minor loops as shown in Figure 7.22. This could be the case in the regions
around the surface of the end-plates close to the rotor, where there is a strong
bias d.c. component (the field is not reversed), which is added to the partial
rotation of the flux vector with the rotor rotation. In general, it is evident that
the core loss prediction is a complicated process where many variables have to
be taken into consideration.

7.7 Series Magnetization

The concept of series magnetization in a three phase electrically magnetized

synchronous machine was developed by Alaküla (2004). It is based on that the
stator winding is connected through a three-phase diode rectifier to the field
winding, as shown in Figure 7.23. In this way the additional power supply
for the field current can be suppressed. Studies on commercial machines re-
adapted for series magnetization have been reported earlier by Tonin (2003)
7.7. Series Magnetization 141

Figure 7.22: Major hysteresis loop and minor loops. Source (Skarrie, 2001).

and Hagstedt (2004).

Figure 7.24 shows a sample of three sinusoidal currents shifted 120 degrees
in time, and the rectified curve corresponding to the field current. It can be
observed that this current has a ripple with a frequency of six times the funda-
mental electrical frequency. The average field current can be calculated using
(7.14), where |is | is the modulus of the phase current. The number of conduc-
tors in the d.c. coils for series magnetization is calculated using (7.15), where
Itdc and Itac are the Ampere turns in the field and armature coils, and N cac
is the number of conductors in an armature coil. The number of turns in the
field winding of the prototype were calculated using this method.


π/2+π/6
6  3
īf = | is | sin θdθ = | is | (7.14)
2π π
π/2−π/6
Itdc
N cdc−series = (7.15)
3/π · Itac /N cac
The measured average torque for a given combination of isx and isy is
shown in Figure 7.25(a). The corresponding measured field current is shown in
Figure 7.25(b). It can be observed that when isy is positive, the torque increases
with negative isx , which indicates that Lsy > Lsx according to (7.13). How-
ever, when isy is negative, the torque slightly increases negatively for positive
values of isx . This is very likely due to a slight misalignment of the resolver,
142 Chapter 7. Prototype and measurements

v fp

u a v 1
v a
u b v 2
v b u
u c v
f
3
v c

v fm

Figure 7.23: Connections for the series magnetized synchronous machine.

fie ld c u rre n t
C u rre n t [A ]

b c a

T im e [s e c ]

Figure 7.24: Rectified field current.

7.7. Series Magnetization 143

as can be observed from the torque values corresponding to isy =0, which are
not exactly zero all the time. In fact, before these measurements were carried
out, the machine had to be dismantled since the rotor needed to be repaired as
the connection to the shaft became loosen. When the machine was mounted
back it is possible that the resolver was not aligned with the same accuracy as
before. Indeed, from Figure 7.19 it can be observed that the line at isy = 0 is
much more straight, and these measurements were performed before the repa-
ration. According to the results for series magnetization at isy = 0, the torque
curve should be shifted so that the values corresponding to isx > 0 are slightly
lifted up, and those for isx < 0 are slightly pushed down. Then the relation
Lsy > Lsx would be valid for all the current combinations.
The average torque measured for isx =-6.2 A and isy =17.7 using series mag-
netization was 10.6 Nm, which is very similar (only 2% higher) to the value
obtained from separate magnetization. Of course, the slight misalignment of
the resolver influences the results, but its effect decreases at lower absolute val-
ues of isx , as observed from Figure 7.25(a). In fact it was measured that the field
current for the given combination of isx and isy above was around 5% higher
than the nominal value for separate magnetization. If the output torque is re-
duced proportionally, then the difference between the torque values obtained
for separate and series magnetization would be within only a few percent.

2 0
1 2
8 1 5
T o rq u e [N m ]

4
If [A ]

0 1 0
− 4
5
− 8
− 1 2
1 4 1 0 1 5 0 1 5
1 0 5 5 1 4 5 1 0
0 − 5 − 1 0 − 5 0 1 0 5 0 − 5 − 5 0
is y [A ] − 1 4 − 1 5− 1 0 is x [A ] − 1 0 − 1 4 − 1 5− 1 0 is x [A ]
is y [A ]

(a) Average torque. (b) Field current.

Figure 7.25: Measured average torque and field currents for series magnetization.
144 Chapter 7. Prototype and measurements

7.8 Conclusions

The practical details on the construction of the prototype have been presented.
The thermal model had to be re-adapted to the new cooling system. The pre-
dicted temperature in the d.c. coils was underestimated by 19%. This could
be due to the improper fit of the coil into the side of the plate, and the added
insulation from the layer of enamel used to fix it. In the a.c. coils the accuracy
was within 4% assuming that the heat transfer from the top of the coils to the
back-core is negligible. Therefore it can be concluded that the thermal models
are reasonable.
It was observed that the inductance from different tooth coils varied by up
to 14%. This is probably due to inhomogeneities in the bulk material from
where the teeth were machined. The teeth were distributed so that the in-
ductance from the three phases was the same, and this was confirmed from
the induced voltage measurements. However, compared to the predictions, the
measured induced voltage was 20% lower, although the shape resembled well
that from FEM. The reason for the lower voltage is probably the leakage path
between the end-plates through the shaft. The holes between the teeth for the
winding connections also affected the magnetic circuit.
The coefficients for the iron loss calculation were not available for the lu-
bricant LB1. Instead they were roughly adapted from studies performed in
inductors using the same material. The measured hysteresis losses were around
60% lower than the estimations, assuming that the flux density is uniformly in-
creased by 20% to match that in the model. The estimated dynamic losses are
at least 45% lower than the measured. A source of error is the fact that the har-
monic loss formulation does not consider the bias field of the d.c. excited core,
neglecting therefore minor hysteresis loops. In general, the iron loss estimation
is a complicated issue, since the material parameters are difficult to obtain and
some of them depend on the machine size. The flux variation in different parts
of the machine can be estimated from FEM, but the proper adaptation of the
iron loss formulation is more complicated. It should be noted that in the FEM
model the effects of the eddy currents on the field distribution is not taken
into consideration. This can therefore introduce inaccuracies in the model that
affect the prediction of the iron losses and the output voltage.
The measured average torque at thermal limit was 14% lower than pre-
dicted, which is related to the reduction of the linking flux. The idea of series
magnetization has been tested. It is shown that the output torque for the given
7.8. Conclusions 145

design current is very similar to that with separate magnetization. It is there-

fore demonstrated how traditional disadvantages of electrically magnetized syn-
chronous machines, such as slip-rings and the power source for the field, can be
overcome with the prototype machine.
146 Chapter 7. Prototype and measurements
Chapter 8

Conclusions

In this first part of the thesis a novel topology of electrically magnetized syn-
chronous machine has been designed to solve the main drawbacks in the more
conventional design: the slip-rings, and the power source for the d.c. wind-
ing. The slip-rings are suppressed, since the field winding is removed from the
rotor and placed in magnetically conducting end-plates. This requires three-
dimensional flux flow, which is achieved using SMC. The windings in the ma-
chine have been designed to exploit the concept of series magnetization. This
consists of feeding the field winding directly from the rectified three phase ar-
mature currents at the neutral point, therefore eliminating the need for the d.c.
power source.
In this chapter a summary of the results obtained is presented, followed by
some topics suggested for future research in this field.

8.1 Summary

MEC
A Magnetic-Equivalent-Circuit model has been developed including a method
to account for non-linearities, which is based on the linearization of the BH
curve. The method provides very fast and accurate results for the flux distri-
bution in the machine at load and no load compared to FEM. Modifications
to accomplish forward and backward rotation were included in the model. A
simplified method was used to assess the shape of the torque response. It was
observed that a change in the area of the airgap reluctances affected significantly

147
148 Chapter 8. Conclusions

the amplitude of the torque response, while the shape was properly predicted.

Leakage reduction
From the FE simulations it was estimated that the leakage in the machine was
around 70%. The main leakage path is from the tips of the claw-poles to the op-
posite end-plate. A study was carried out to assess the effectiveness of including
permanent magnets around the claw-poles to reduce the leakage. Using ferrite
magnets the leakage is marginally decreased by up to 10%. However, combin-
ing NdFeB and ferrite magnets, the leakage is reduced by up to 50%. Including
NdFeB magnets all around the claw-poles modifies the pattern of magnetiza-
tion and increases the output torque between 53% and 113%, for designs with
12 and 24 poles respectively.

Comparison with other topologies

The prototype design has been compared by simulation with three alternative
topologies: an outer-rotor claw-pole machine, a conventional electrically mag-
netized machine, and a surface permanent magnet machine. Among the electri-
cally magnetized machines, the conventional design has a slightly higher average
torque per weight and per volume (8%) compared to the prototype design. The
latter has however around 18% lower material cost per output torque. The PM
machine presents superior electromagnetic performance compared to the elec-
trically magnetized machines. The torque density is around 40% higher. It
is however also the most expensive machine considering the cost per output
torque, which is around 32% higher than in the prototype design.

Measurements
The shape of the measured induced voltage was almost identical to that pre-
dicted in FEM. However, the amplitude of the flux was around 20% lower
than predicted, and the average torque was reduced by 14%. This could be
due to the leakage path between the end-plates through the shaft, not included
in the models. Constructional details such as the holes between the teeth to
connect the coils can also contribute to reduce the linking flux.
The thermal network model predicted reasonably the temperature rise in
the windings. In the armature coils the temperature was predicted with an
accuracy of 4% assuming that there is no heat transfer from the top of the coils
8.2. Future work 149

to the core. This is reasonable since a gap was left in this region to facilitate
the mounting process. The temperature in the field coils was underestimated
by 19%. This could be due to the improper fit of the coil into the side of the
plate, and the added insulation from the layer of enamel used to fix it. These
details introduce uncertainties in the model that are very difficult to assess, but
which would contribute to increase the estimated temperature.
Despite the efforts to model precisely the field variation in the machine,
the iron loss prediction was rather inaccurate. This is partially due to the lack
of proper coefficients for the lubricant LB1 to be used in the loss formulation.
Nevertheless, the results indicated that the measured hysteresis losses were 60%
lower than predicted, while the estimated dynamic losses were at least 45%
lower than the measured.

8.2 Future work

Regarding the MEC model, the formulation for mesh rotation could be fur-
ther improved, including more sophisticated methods to calculate the torque
response.
The constructional details of the prototype affect to some extent the perfor-
mance of the machine compared to the predictions in the models. The coupling
between the end-plates, the bearings and the shaft could be included in the elec-
tromagnetic finite element model to assess the amount of flux leaking through
the shaft. The holes in the back-core for the connections of the windings should
also be included.
The thermal characteristics of the machine could be more accurately mod-
elled in thermal FEM. However, although the geometry is better discretized in
FEM, there are still uncertain parameters that have to be input, such as the de-
gree of contact between different materials at the boundaries, the heat transfer
coefficient, etc. These parameters can affect significantly the output results.
The proper parameters for the iron loss formulation using the lubricant LB1
should be measured. The loss calculation should be also improved. Although it
is instructive to calculate the iron losses as a function of the harmonic number,
simply adding them to calculate the total losses is not adequate in non-linear
systems. However, this procedure is frequently used in the literature, mainly
due to the lack of more advanced models. A proper model to calculate the iron
losses should take into consideration minor hysteresis loops, including the bias
field of the d.c. excited core.
150 Chapter 8. Conclusions

An algorithm should be implemented in the control system to compensate

for the high torque ripple in the machine. This can be achieved by adding a
harmonic component to the currents with the proper frequency and phase.
A sensitivity study should be carried out to assess the change in the mag-
netic properties of the teeth when they are machined from iron-powder bulks
of different size.
The comparison of the four machines indicated that the PM machine clearly
presents superior electromagnetic performance at normal operation. However,
a similar comparison should be carried out at field weakening operation, where
the electrically magnetized machines operate more advantageously.
Bibliography

Alaküla, M. (2002). Power electronic control. Dept of Industrial Electrical Engi-

neering and Automation, Lund University, Sweden.
Alaküla, M. (2004). International Patent Application PCT/SE2004/000396.
Synchronous machines.
Andersson, S. (2000). Optimization of a servo motor for an industrial robot ap-
plication. Licentiate Thesis, Dept of Industrial Electrical Engineering and
Automation, Lund University, Sweden.
Aoki, H., Nohira, H., Arai, H. (1967). Convective heat transfer in an annulus
with an inner rotating cylinder. Bulletin of JSME, vol. 10, no. 39, pp. 523-
532.
Arkkio, A. (2002). Thermal analysis of high-speed electrical machines. Postgradu-
ate Seminar on Electromechanics, Laboratory of Electromechanics, Helsinki
University of Technology, Finland.
Becker, K.M., Kaye, J. (1962). Measurements of diabatic flow in an annulus
with an inner rotating cylinder. Journal of Heat Transfer, 84, pp. 97-105.
Bjorklund, I.S., Kays, W.M. (1959). Heat transfer between concentric rotating
cylinders. Journal of Heat Transfer, 81, pp. 175-186.
Blarigan, P.V., Paradiso, N., Goldsborough, S. (1998). Homogeneous charge
compression ignition with a free piston: a new approach to ideal Otto cycle
performance. Society of Automotive Engineers, SAE paper nr 982484.
Boglietti, A., Cavagnino, A., Staton, D.A. (2004). Thermal sensitivity analysis
for TEFC induction motors. Proceedings of IEE International Conference on
Power Electronics, Machines and Drives, Edinburgh, UK, pp. 160-165.

151
152 Bibliography

Bottauscio, O., Canova, A., Chiampi, M., Repetto, M. (2002). Iron losses in
electrical machines: influence of different material models. IEEE Transactions
on Magnetics, vol. 38, no. 2, pp. 805-808.
Cros, J., Viarouge, P. (2002). Synthesis of high performance PM motors with
concentrated windings. IEEE Transactions on Energy Conversion, vol. 17, no.
2, pp. 248-253.
Driesen, J., Belmans, R.J.M., Hameyer, K., Stephenson, D. (2001). Finite-
element modeling of thermal contact resistances and insulation layers in elec-
trical machines. IEEE Transactions on Industry Applications, vol. 37, no. 1, pp.
15-20.
Erlandsson, O. (1998). Alvar engine, an engine with variable compression ratio,
experiments and tests. MSc Thesis, Dept of Heat and Power Engineering,
Division of Combustion Engines, Lund University, Sweden.
Erlandsson, O., Lundholm, G., Söderberg, F., Johansson, B., Wong, V.W.
(1998). Demonstrating the performance and emission characteristics of a
variable compression ratio, Alvar-cycle engine. Society of Automotive Engi-
neers, SAE paper nr 982682.
Gazley, C. (1958). Heat transfer characteristics of rotational and axial flow
between concentric cylinders. Transactions of the ASME, vol. 80, no. 1, pp.
79-90.
Guo, Y., Zhu, J.G., Zhong, J.J., Wu, W. (2003). Core losses in claw pole
permanent magnet machines with soft magnetic composite stators. IEEE
Transactions on Magnetics, vol. 39, no. 5, pp. 3199-3201.
Hagstedt, D. (2004). Analysis of a series magnetized Lundell alternator. MSc The-
sis, Dept of Industrial Electrical Engineering and Automation, Lund Univer-
sity, Sweden.
Hamdi, E.S. (1994). Design of small electrical machines. John Wiley & Sons
Ltd., UK.
Henneberger, G., Küppers, S., Ramesohl, I. (1996). Numerical calculation,
simulation and design optimisation of claw pole alternators for automotive
application. IEE Colloquium on Machines for Automotive Applications, Digest
No 1996/166.
Bibliography 153

Hofer, U. (2004). High speed motor design issues using soft magnetic compos-
ites. Invited paper, The UK Magnetics Society, Seminar on SMC and their
use in electrical machines, Newcastle, UK, Nov. 2004, 4 pp.

Holman, J.P. (1992). Heat transfer. 7th edition, McGraw-Hill.

Höganäs AB S-26383, Höganäs, Sweden, SOMALOYT M 500, SMC 97-1, AB

Ruter Press.

Höganäs AB S-26383, Höganäs, Sweden, SMC School, 27-28 November, 2001.

Ishikawa, T., Slemon, G.R. (1993). A method of reducing ripple torque in

permanent magnet motors without skewing. IEEE Transactions on Magnetics,
vol. 29, no. 2, pp. 2028-2031.

Jack, A.G. (2003). Exploring design opportunities offered by SMC for electrical
machines. IEEE International Electric Machines and Drives Conference, SMC
seminar tutorial, 1 June, 2003.

Jack, A.G., Mecrow, B.C., Dickinson, P.G., Stephenson, D., Burdess, J.S.,
Fawcett, N., Evans J.T. (2000). Permanent-magnet machines with powdered
iron cores and prepressed windings. IEEE Transactions on Industry Applica-
tions, vol. 36, no. 4, pp. 1077-1084.

Jahns, T.M., Soong W.L. (1996). Pulsating torque minimization techniques

for permanent magnet AC motor drives - A review. IEEE Transactions on
Industrial Electronics, vol. 43, no. 2, pp. 321-330.

Jokinen, T., Saari, J. (1997). Modelling of the coolant flow with heat flow con-
trolled temperature sources in thermal networks. IEE Proceedings on Electric
Power Applications, vol. 144, no. 5, pp. 338-342.

Kazuaki, S., Kaoru, K., Toshiaki, K., Yuji, H. (2000). Development of electric
motors for the Toyota hybrid vehicle PRIUS. 17th International Electric Vehi-
cle Symposium & Exposition, Montréal, Canada, Record from the CD-ROM.

Kylander, G. (1995). Thermal modelling of small cage induction motors. PhD

Thesis, School of Electrical and Computer Engineering, Chalmers University
of Technology, Sweden.
154 Bibliography

Lange, A., Canders, W.R., Mosebach, H. (2000). Investigation of iron losses

of soft magnetic powder components for electrical machines. Proceedings of
International Conference on Electrical Machines, Helsinki, Finland, pp. 1521-
1525.

Li, T., Slemon, G.R. (1988). Reduction of cogging torque in permanent magnet
motors. IEEE Transactions on Magnetics, vol. 24, no. 6, pp. 2901-2903.

Luomi, J. (1993). Electromagnetic field analysis. Summer seminar for postgrad-

uates, Tammsviks Kursgård, Bro, Sweden, 15-21 August 1993.

Ma, L., Sanada, M., Morimoto S., Takeda Y. (2003). Prediction of iron loss
in rotating machines with rotational loss included. IEEE Transactions on
Magnetics, vol. 39, no. 4, pp. 2036-2041.

Mellor, P.H., Roberts, D., Turner, D.R. (1991). Lumped parameter thermal
model for electrical machines of TEFC design. IEE Proceedings -B on Electric
Power Applications, vol. 138, no. 5, pp. 205-218.

Miller, T.J.E. (1993). Switched reluctance motors and their control. Oxford Uni-
versity Press, UK.

Moallem, M., Dawson, G.E. (1998). An improved magnetic equivalent circuit

method for predicting the characteristics of highly saturated electromagnetic
devices. IEEE Transactions on Magnetics, vol. 34, no 5, pp. 3632-3635.

Novotny, D.W., Lipo, T.A. (1996). Vector control and dynamics of AC drives.
Oxford Science Publications, Wisconsin, USA.

Ostović, V. (1989). Dynamics of saturated electric machines. Springer-Verlag,

USA.

Ostović, V., Miller, J.M., Garg, V.K., Schultz, R.D., and Swales, S.H. (1999).
A magnetic-equivalent-circuit-based performance computation of a Lundell
alternator. IEEE Transactions on Industry Applications, vol. 35, no 4, pp. 825-
830.

Ottosen, N., Petersson, H. (1992). Introduction to the Finite Element Method.

Prentice Hall Europe, UK.
Bibliography 155

Polinder, H., Hoeijmakers, M.J. (1997). Eddy-current losses in the permanent

magnets of a PM machine. Proceedings of IEE International Conference on
Electrical Machines and Drives, Cambridge, UK, pp. 138-142.
Sadarangani, C. (2000). Electrical machines - Design and analysis of induction
and permanent magnet motors. Dept of electric power engineering, electrical
machines and power electronics, Royal Institute of Technology, Sweden.
Sieder, E.N., Tate, C.E. (1936). Heat transfer and pressure drop of liquids in
tubes. Ind. Eng. Chem., vol. 28, no. 4, p. 1429.
Skarrie, H. (2001). Design of powder core inductors. Licentiate Thesis, Dept of
Industrial Electrical Engineering and Automation, Lund University, Sweden.
Stenlåås, O. (2004). On the impact of piston motion and in-cylinder charge com-
position on energy release, auto ignition and emission formation in premixed
charge internal combustion engines. PhD Thesis, Dept of Heat and Power
Engineering, Division of Combustion Engines, Lund University, Sweden.
Stewart, M. (1997). Design and analysis of a variable-compression-ratio internal-
combustion engine - The Alvar engine concept. MSc Thesis, Dept of Mechani-
cal Engineering, Massachusetts Institute of Technology, USA.
Taniguchi, C. (2000). Rotor of rotary machine having claw poles and auxiliary
permanent magnets. United States Patent number 6011343.
The Royal Commission on Environmental Pollution (2000). 22nd Report, En-
ergy - The changing climate, Chapter 2: On causes and effects of climate change.
http://www.rcep.org.uk/newenergy.htm, UK.
The Swedish Energy Agency (2003). Energy in Sweden 2003, Chapter 4: Energy
use. http://www.stem.se, Sweden.
Tonin, D. (2003). A novel synchronous drive using series magnetized synchronous
machine with a special position sensing. MSc Thesis, Dept of Industrial Elec-
trical Engineering and Automation, Lund University, Sweden.
Vas, P. (1990). Vector control of AC machines. Oxford University Press, UK.
Vong, P.K., Rodger, D. (2003). Coupled electromagnetic-thermal modeling
of electrical machines. IEEE Transactions on Magnetics, vol. 39, no. 3, pp.
1614-1617.
156 Bibliography

Wong, H.Y. (1977). Handbook of essential formulae and data on heat transfer for
engineers. Longman Inc., New York, USA.

Wong, V.W., Stewart, G., Lundholm, G., Höglund, A. (1998). Increased power
density via variable compression/displacement and turbocharging using the
Alvar-cycle engine. Society of Automotive Engineers, SAE paper nr 981027.

Zhou, P. (1993). Numerical analysis of electromagnetic fields. Springer-Verlag,

New York, USA.

Zhu, J.G., Ramsden, V.S. (1998). Improved formulations for rotational core
losses in rotating electrical machines. IEEE Transactions on Magnetics, vol.
34, no 4, pp. 2234-2242.
Appendix A

Equations for the MEC network

The permeances (P ) of the elements used in the MEC model are listed below.
They are easily converted into reluctances (R = 1/P ). The equations are
parametrized according to the geometry, and these parameters are described in
Appendix B. The permeability of each iron element is a function of the interval
in the BH curve where its flux density is located during the iteration process.
The equations have not been completely simplified in order to allow the reader
to understand the geometrical considerations used in the calculations.

2π
α = (A.1)
Ns
 
α/2 · ros2 − (ros − dy)2
P yx1 = µ0 µpyx1 (A.2)
zyt − zdx + zdx/2 + zts/8
(2π(ror + ros − dy)/2) /Ns · zdx
P ss1 = µ0 µpss1 (A.3)
ros − dy − ror + dy/2
2π(ror − dr + ror)/2 dr
P sr1a = µ0 µpsr1a zdx/ (A.4)
Ns 2
α/2   1
P sr1b = µ0 µpsr1b ror2 − rir2 /zdy (A.5)
2 2
wot · ris · ztps/4
P sc1 = µ0 µpsc1  (A.6)
(wot/2 · (ris + dos) − wt/2)2 + dos2
P ps1 = µ0 µpps1 · wt · zts/ds (A.7)

157
158 Appendix A. Equations for the MEC network

α/2  2 
P cr1a1 = µ0 µpcr1a1 ror − rir2 /ztr (A.8)
2
(α − (2π/Np − wtr1))/2
P cr1a2 = µ0 µprtt ·
2
ror2 − (ror − dr)2
· (A.9)
(ztpr/2)2 + (dr/2)2
zts · dy
P yp12 = µ0 µpyp12 (A.10)
(2π(ros − dy + ros)/2)/Ns
wot
P gr1 = µ0 ror · ztps/g1 (A.11)
2
α/2  2 
P ge1 = µ0 ror − rir2 /g2 (A.12)
2
(α − (2π/Np − wtr3))/2
P sr1σ = µ0 ·
2
ror2 − (ror − dr)2
· (A.13)
zrt − ztpr − ztr + g2
P sf 1σ = µ0 · wot · ris · dos/(ztr + g2) (A.14)
2π(ris + dos + ris)/2
P cs12σ = µ0 · ztps · dos/ −
Ns
ris + dos + ris
− wot (A.15)
2
  dr
P cr14σ = µ0 ztpr2 + (dr − htr3)2 + ztpr / (A.16)
 2 
ror − dr + ror 2π 2π/Np − wtr3
/ − wtr1 +
2 Np 2
α − (2π/Np − wtr1)/2
P cr2a2 = µ0 µprtt ·
2
ror2 − (ror − dr)2
· (A.17)
(ztpr/2)2 + (dr/2)2
wtr3 − (α − (2π/Np − wtr3))/2
P sr2σ = µ0 ·
2
ror2 − (ror − htr3)2
· (A.18)
zrt − ztpr − ztr + g2
 159

µ
P sc2 = P sc1 µpsc1
psc2
P ge4 = P ge1
µpps2
P ps2 = P ps1 µpps1 P cr4σ = P cr1σ
µ
P yp23 = P yp12 µpyp23
pyp12
P sr4σ = P sr1σ
µ
P cr3a2 = P cr2a2 µprtbprtt
P yx2 = P yx3 = P yx1
µpsc3
P sc3 = P sc2 µpsc2 P ss2 = P ss3 = P ss1
µ
P ps3 = P ps2 µpps3
pps2
P sr2a = P sr3a = P sr1a
µ
P yp31 = P yp23 µpyp31
pyp23
P sr2b = P sr3b = 2P sr1b
µ
P cr4a2 = P cr1a2 µprtbprtt
P cr2a1 = P cr3a1 = 2P cr1a1
P lea12 = 2µrtt P cs1σ P gr2 = P gr3 = 2P gr1
P lea13 = 2µrtb P cs1σ P ge2 = P ge3 = 2P ge1
P sr3σ = P sr2σ P sf 2σ = P sf 3σ = P sf 1σ
P cr4a1 = P cr1a1 P cs2σ = P cs3σ = P cs1σ
P sr4b = P sr1b P cr2σ = P cr3σ = P cr1σ
P gr4 = P gr1
160 Appendix A. Equations for the MEC network
Appendix B

Parametrization of the machine

360/(Np/2)

zyc 360/Ns

dy

zyt

ztps
ros
ds

dos
wos wot
ris

Figure B.1: Stator teeth and yoke.

161
162 Appendix B. Parametrization of the machine

360/(Np/2)
wtp2
wtp1

wtr1
wtr3
htr3
ztpr
ztr
ror

ror-dr

rir

Figure B.2: One rotor claw-pole structure.

wt 360/(Np/2)
zdx

zts ros-dy
wc1

wc2
zdy

ror-dr
ztip

rir

(a) Stator tooth (b) Stator end-plate

Figure B.3: Details of the stator tooth and end-plate.

 163

zrt

Radial
Airgap
g1

g2 Axial g2
Airgaps

Figure B.4: Claw-pole rotor and end-plates.

164 Appendix B. Parametrization of the machine

Table B.1: Dimensions of the prototype [mm]

Dimension Value Dimension Value
Np 16 ztpr 30.16
Ns 24 ztip 7.97
L 76.04 zdx 10.00
ros 100.00 zdy 6.67
ris 67.50 zrt 41.70
ror 66.90 wt 9.05
rir 19.00 wc1 1.13
dy 11.30 wc2 1.13
ds 18.27 wos 5.00o
dos 2.93 wot 10.00o
dr 22.49 wtp1 24.75o
htr3 2.18 wtp2 18.00o
zyc 76.04 wtr1 18.00o
zyt 30.36 wtr3 10.81o
zts 15.32 g1 0.60
ztps 31.26 g2 0.50
ztr 5.22
Appendix C

Equations for the thermal

network

The equations used for the thermal model are listed below. The parameters
in the equations represent parts of the geometry, and they are defined in Ap-
pendix B.
Asfdc = π(ros2 − rir2 ) + 2π(ror − dr) · zdx (C.1)
Asfac = 2π · ros · zyc (C.2)
1
Ascac = 2π(ros + dy + ros) zyc (C.3)
2
1
Lcwac = (ds + dos)2 + (wt/2)2 +
2
1 1
+ (ds + dos)2 + (zts/2)2 (C.4)
2 2
1
Acwac = (2 · ds(wt + zts) + zts · wt) Ns (C.5)
2

1  
Lswdc = (zdx + zdy/2)2 + dy 2 + zdx + zdx2 + dr2 +
4
1
+(dr + (ror − dr + rir)) (C.6)
2
1 1
Aswdc = 2π (ror − dr + ror) · (zdx + zdy) +
2 2 !
ros − dy + ros 2 ror − dr + ror 2
+π − (C.7)
2 2

165
166 Appendix C. Equations for the thermal network

hth = ds − 2 · insl (C.8)

Awiac = 2 · hth ((wt + insl) + (zts + insl)) · Ns (C.9)
1
Awidc = 2π(ros − dy − insl)zdy +
2
1 1
+2π(ror + insl)(zdy − insl) +
2 2
1 1
+π((ros − dy − insl) − (ror + insl)2 )
2
(C.10)
2 2
1 1
Lihac = (2π(ris + dos + ds)/Ns − wt) (C.11)
2 2
1
Aihac = 2 · hth ((wt + 2 · insl) + (zts + 2 · insl)) Ns (C.12)
2
Lvdc = (ros − dy − insl) − (ror + insl) (C.13)
1 1
Lihdc = ((zdy − insl) + Lvdc ) (C.14)
2 2
1 1
Aihdc = 2π(ros − dy − insl − Lvdc ) · (zdy − insl) +
4 2
1 1
+2π(ror + insl + Lvdc ) · (zdy − insl) +
4 2
1
+π((ros − dy − insl − Lvdc )2 −
4
1 2
−(ror + insl + Lvdc ) ) (C.15)
4
Appendix D

Dimensions of the alternative

machines

In this appendix the dimensions of the topologies analyzed in Chapter 6 are

summarized. For Design 1, which is a 12-pole version of the prototype, only
the circumferential measures are included, since the rest are exactly the same as
in the prototype, and can be found in Appendix B.
The measures for Design 2, the outer rotor claw-pole machine, can be ex-
pressed as a function of those of the prototype, only changing the definition of
some components (marked with ‘*’). The value of ‘ris2’ and ‘ros2’ correspond
to the inner and outer radius of the stator part above the claw-poles. The thick-
ness of its yoke is given by ‘dy2’. Similarly, ‘ris1’ and ‘ros1’ correspond to the

Table D.1: Dimensions of Design 1 [mm]

Dimension Value Dimension Value
Np 12 wot 13.33o
Ns 18 wtp1 33.00o
wt 12.06 wtp2 24.00o
wc1 1.50 wtr1 24.00o
wc2 1.50 wtr3 14.40o
wos 6.66o

167
168 Appendix D. Dimensions of the alternative machines

Table D.2: Dimensions of Design 2 [mm]

Dimension Value Dimension Value
Np 12 ztpr 30.16
Ns 18 ztip 7.97
L 50.00 ros1∗ 72.25
ros2∗ 100.00 ris1∗ 19.00
ris2∗ 88.00 zrt 50.00
ror 87.40 wt 12.10
rir 72.85 wc1 1.51
dy2∗ 6.90 wc2 1.51
ds 18.27 wos 4.80o
dos 2.93 wot 15.20o
dr 11.63 wtp1 33.00o
htr3 2.61 wtp2 24.00o
zyc 31.26 wtr1 24.00o
zyt 7.97 wtr3 21.84o
zts 15.32 g1 0.60
ztps 31.26 dy1∗ 32.06
ztr 7.83

inner and outer radius of the stator below the claw-poles. The thickness of its
yoke is given by ‘dy1’. The airgap ‘g1’ is the same between the rotor and the
upper and the lower stator parts.
The measures of the more conventional synchronous machine without claw-
poles, Design 3, are given in Table D.3. The parametrization of the stator is
similar to that of the prototype and Design 1. Actually, the rotor has a similar
shape but inverted, and the parameters expressing the same measure as in the
stator are marked with ‘*’.
 169

Table D.3: Dimensions of Design 3 [mm]

Dimension Value Dimension Value
Np 12 zyt2∗ 7.97
Ns 18 ztip 7.97
L 42.70 ztip2∗ 7.97
ros 100.00 zts2∗ 15.32
ris 67.50 ztps2∗ 31.26
ror 66.90 wt 12.06
rir 19.00 wc1 1.50
dy 11.30 wc2 1.50
ds 18.27 wos 4.80o
dos 2.93 wot 15.20o
ds2∗ 15.75 wt2∗ 16.75
dos2∗ 2.93 wc12∗ 1.50
zyc 42.70 wc22∗ 1.50
zyt 13.70 wos2∗ 8.0o
zts 15.32 wot2∗ 22.0o
ztps 31.26 g2 0.60
zyc2∗ 31.26
170 Appendix D. Dimensions of the alternative machines
 Part II

Acoustic Noise in Electrical

Machines

171
Chapter 9

Introduction

9.1 Background
Major contributions on acoustic noise in adjustable-speed a.c. electric motor
drives have been made during the last 20 years in connection with the ad-
vent of more and more cost effective and powerful frequency converters. The
new installations of adjustable-speed drives in industry and commercial build-
ings are today mainly embodied as a.c. induction motor drives, i.e. as low-
maintenance, brush-less drives, pushing traditional d.c. brush-commutator mo-
tor drives aside. The level of motor noise radiated increased considerably, as a
consequence, due to the current harmonics introduced by the switching tech-
nology in the converter. This thesis focuses on the embodiment of the frequency
converter what regards modern voltage and airgap flux modulation techniques.
The literature referenced here is therefore limited to recent publications on
adjustable-speed a.c. electric motor drives although the author is well aware
that the history of a.c. drives extends long before the solid-state era, with key
engineering developments in the late nineteenth century and the first half of the
twentieth century (Jahns and Owen, 2001).

Noise sources
In general, three different sources of noise can be distinguished in electric ma-
chines: mechanical, aerodynamic and electromagnetic. The mechanical compo-
nent is mainly influenced by the way the stator sheets are mounted into the core
of the motor, the coupling to the load, rotor unbalance, as well as the vibration
of the bearings and the amount of damping material present in the structure.

173
174 Chapter 9. Introduction

The aerodynamic component is only dependent on the fan used in air cooled
machines. The electromagnetic component is due to the action of magnetic
forces in the airgap. These forces are influenced by the harmonics present in the
voltage feeding the machine, slot combinations, saturation, magnetostriction
and eccentricity among others.
After all, noise is produced due to the mechanical vibrations that take place
in the machine. Hence, it is important to evaluate how the different compo-
nents of the structure influence the noise emissions. One factor to take into
consideration is that the machine usually presents some changes compared to
the original design, due to the tolerances introduced during the manufacturing
process, which become more important in mass production. Starting from the
same prototype design, it is not unusual to find machines with a rather differ-
ent acoustical behaviour, as analyzed by Zhu et al. (1997), and Lo et al. (2000).
The main variations are found in the eccentricity of the rotor relative to the
stator, which is one important source of noise. By using acoustical simulation,
Ramesohl et al. (1999) studied the effects of different manufacturing tolerances
in mass produced electric machines. Electromagnetic and structural models us-
ing the finite element method (FEM) were used to calculate the force densities
which were applied to an acoustic model using the boundary element method
(BEM).
Eccentricity is one important factor that has to be taken into consideration
when studying the acoustic response of an electric machine. There are namely
two types of eccentricity: static and dynamic. Static eccentricity appears when
there is an incorrect positioning of the rotor with respect to the stator bore. In
this case, when the rotor rotates the position of minimal radial airgap length is
fixed in space. Dynamic eccentricity occurs when the center of the rotor is not
at the center of the rotation, and the position of minimum airgap rotates with
the rotor. Dynamic eccentricity is usually checked in the manufacturing stage
and compensated by means of dynamic balancing. However, static eccentricity
is hardly checked during the production process and its contribution to the
generation of audible noise has been studied by Delaere et al. (1998).
One of the optimization parameters in electric machine design is the size
of the motor, and thus the amount of iron needed to achieve a certain per-
formance. Reducing the size of the core decreases the stiffness of the stator
structure, which in turn becomes more sensitive to forces acting on it. There-
fore, there is a compromise between cost and noise. Dealing with this issue,
an optimization process was developed by Hadj Amor et al. (1995) where, for
9.1. Background 175

randomly selected dimensions of the structure, the noise is estimated and in

this way the dimensions of the motor producing the least noise can be found.
The results obtained suggested the existence of only one minimum, which not
surprisingly corresponded to a very heavy and expensive motor.
The stiffness of the stator is also directly influenced by the amount of damp-
ing material present on it, as explained by Verma and Balan (1998). Damping
materials in the stator are the windings, laminations and insulation. Experi-
ments carried out with a stator shell, both with no laminations and a laminated
stator, confirmed that the noise level was around 20 dB lower in the case of
the laminated stator, due to its higher amount of damping. Besides, the peaks
along the frequency scale in the noise spectrum were more flattened. However,
the stiffness of the laminated stator is lower than the stator shell with no lam-
inations, which means that there will be several more resonances falling in the
audible frequency range. This behaviour is further affected by the fact that the
damping at these frequencies is reduced, as it is proportional to the resonant
frequency and the damping ratio, i.e. the benefits of damping are less accentu-
ated at lower frequencies, but still the overall acoustic emissions are lower when
damping is present.
The stiffness of the machine is also increased when coupling the motor to
the ground (foot mounting) and the load (shaft mounting). The effect on the
acoustical emissions of this coupling was analyzed by Zhu et al. (1999) who
showed that, although additional natural frequencies are added to the system,
the coupling restricts the stator vibration, as its stiffness is increased. The re-
sult is that the acoustic noise produced by the machine is reduced. Further,
it was shown that even if two similar motors present quite a different acoustic
behaviour at no load, when loaded not only the acoustic noise is reduced, but
also the acoustic behaviour of both machines becomes much more similar.
In fact, the noise produced by one induction machine can change consider-
ably depending on the ratio between the stator and rotor slots. In general, odd
rotor slot numbers should be avoided. A more complete study was presented
by Kobayashi et al. (1997), where transient magnetic field FEM and harmonic
analyses using Fourier series expansion were used together to compute the forces
for a given number of stator and rotor slots. It was found that the level of acous-
tic noise is dependent on the first space harmonic order, which was not always
predicted by analytical calculations based on the slot ratio of the machine.
While the sources responsible for the acoustic noise in electric machines are
widely accepted, there is a lack of consensus about the physical mechanisms of
176 Chapter 9. Introduction

noise production in the machine. The most accepted theory is that the stator
vibrations are produced by the radial components of the electromagnetic air-
gap forces, whereas tangential forces are usually neglected. However, tangential
forces can play an important role in large machines, as discussed by Garvey and
Le Flem (1999). The influence of magnetostriction is not yet clear. It is a phe-
nomenon that results from the change in the dimensions of the iron when a
magnetic flux passes through it. The iron tends to contract along the direction
of the flux and tends to expand in the other two directions, and this process
increases the level of vibration in the machine. Belmans and Hameyer (1998)
considered that although the effect of magnetostriction in theory can contribute
to the audible noise, in practice its contribution is negligible. However, a me-
thod was presented by Garvey and Glew (1999) to compute the effects of mag-
netostriction in the stator core and to show its importance. By conducting a
series of experiments, Låftman (1997) found that the magnetostrictive forces
are about the same magnitude as the Maxwell forces in medium sized induc-
tion machines (a few tens of kilowatts) and therefore they would have a major
importance in the production of noise in electrical machines.
Finally, the effects of the geometrical and constructional irregularities were
presented by Chang and Yacamini (1996). It was found that these irregulari-
ties introduce asymmetrical behaviour, which results in an enriched frequency
response spectrum. The asymmetry frequencies are caused mainly by the sta-
tor winding and frame irregularities which means that the design details of the
winding and the frame will play a very important role in the mechanical be-
haviour of electrical machines and should not be treated simply as an additional
mass. Hence, electrical machines should be designed and constructed as sym-
metrical as possible, in particular the winding overhang.

Electromagnetic force computation

In order to predict the noise behaviour of the machine, it is important to evalu-
ate the airgap flux distribution, which is directly linked to the harmonic pattern
of the electromagnetic forces acting on the stator. Many theories have been
developed for their analytical computation and approximations need to be in-
troduced in order to model the numerous details in the machine affecting the
airgap forces such as saturation, slot shape, etc. Some of the theories proposed
provide a quick and relatively accurate prediction of the expected airgap force
frequencies, although it is not simple to calculate their magnitude. A modi-
9.1. Background 177

fied one-dimension theory was used by Gerling (1994) taking into account the
changes in the reluctance along the airgap due to the stator slots, while no rotor
slots were considered. The main idea is to use the geometrical airgap in the re-
gion of a stator tooth, and enlarge it only in the region along a slot opening by
a factor dependent on the average length of the field lines penetrating through
the slot. This procedure shows more accurate results than considering a general
enlargement of the geometrical airgap through the whole surface to account
for the effects of the stator slot openings. A new correction factor associated
with the depth of the slot openings improved considerably the accuracy when
predicting the amplitudes of the harmonics.
Another approach accounting for the effects of stator and rotor slotting, ec-
centricity, iron saturation and the interactions between the various stator and
rotor current components was presented by Vandevelde and Melkebeek (1994),
using the rotating field theory and the permeance wave technique. The air-
gap flux density components were then obtained by multiplying the permeance
waves, in which the effects of slotting, eccentricity and saturation were mod-
elled, by the magnetomotive force. A more accurate model based on a mag-
netic equivalent circuit method (MEC) was also elaborated for the calculation
of the flux components and the resulting radial magnetic forces (Vandevelde
and Melkebeek, 1997).
The effects of both stator and rotor slot openings under load and no load
conditions were studied analytically by Cho and Kim (1998), taking into con-
sideration the effects of skewing and the rotation of the rotor, as well as eccen-
tricity. At no load the model was accurate in predicting the frequencies of the
harmonics but not the amplitudes. Besides, at load there were slight discrepan-
cies in the magnitude of the main flux. Hubert and Friedrich (2002) presented
a semi-analytical method to identify the harmonic components of the airgap
electromagnetic force and the resonance frequencies of the stator in a small
induction machine. Noise was reduced simply by missmatching these compo-
nents with a suitable change in the fixed switching frequency in the modulator.
Although the analytical approaches present certain limitations, their short
computation time proves to be helpful for a quick assessment of the force spec-
trum in the machine. However, for more accurate calculations, the use of finite
element methods is more appropriate, although the computation time is con-
siderably increased. The developments in computing technology through the
years has contributed to spread the use of FEM packages for electromagnetic
an mechanical analyses, among others. Since the noise problem in electrical
178 Chapter 9. Introduction

machines is a consequence of the interaction between electromagnetic and me-

chanical forces, both packages need to be used together. In this way, Ishibashi et
al. (1998) presented an algorithm to predict the shape of the vibration modes of
a machine at no load, combining electromagnetic and mechanical FEM analy-
ses. It was found that the vibration response of one mode depends on the mode
of the nearby natural frequency, and not on the mode of the electromagnetic
force wave.
The use of FEM is mainly limited to low frequencies, since the number of
elements increases in a quadratic way. For higher frequencies, a new method
suitable for calculations of noise and vibration in electrical machines was devel-
oped by Wang et al. (2002). The method was called statistical energy analysis
(SEA), commonly used for ship and aircraft structures but not earlier applied to
electrical machines. The main idea is to divide the whole structure into groups
of components presenting similar modal energy storage and then estimate the
vibration energy of each subsystem by solving the energy balance equations
which relate the power flow from one subsystem to another. In the case of in-
duction machines these structures correspond to those presenting similar modes
of vibration.

Influence of the modulation strategies

The electromagnetic noise component acquired a great importance with the
arrival of converters. Many research efforts were directed towards the manipu-
lation of the switching pattern so as to mismatch between the new harmon-
ics introduced in the currents and the resonant frequencies of the machine
structure, which proved to be the cause of increasing noise in the new drives
(Belmans et al., 1991). One of the most widespread modulation techniques
is pulse width modulation (PWM), where the states of the switches are selected
by comparing the actual value of the command signal, which is output from
a regulator, with a constant frequency carrier wave. Increasing the switching
frequency above the audible spectrum range (20 Hz - 20 kHz) would indeed re-
duce the noise perceived from the motor, but drawbacks appear in the converter
instead. Switching faster involves an increase of the power lost in the switches
as well as the amount of electromagnetic noise produced by the converter. In
order to avoid these side effects, the switching frequency should be maintained
between the range from 3 kHz up to 12 kHz, which is within the sensitive
human hearing range. Increasing the switching frequency may also contribute
9.1. Background 179

to the appearance of strong subharmonics, causing more single tones, which

are perceived as highly disturbing by the human ear (Belmans and Hameyer,
1998). An alternative to PWM is pulse frequency modulation (PFM), where the
frequency of the carrier wave is changed continuously in order to spread the
harmonic components over the whole spectrum. The results obtained using
this technique showed a better acoustical response compared to PWM, as de-
scribed by Stemmler and Eilinger (1994). The frequency can be changed in a
random fashion (Habetler and Deepakraj, 1991), (Covic and Boys, 1998), or
among some predetermined values, selected in a way that they do not interact
with any of the resonant frequencies of the motor (Garcia-Otero and Devaney,
1994), (Bologniani et al., 1999).
In the last decade new control strategies have been developed in order to
achieve better performance in the drives. As a consequence, the switching pat-
tern in the converter is modified, which in turn results in a different acoustical
response. The noise emissions from a direct torque controlled (DTC) drive are
analyzed by Xu et al. (2000). It was shown that DTC controlled drives have
similar spread spectrum features as in the case of random PWM modulation,
which is generally regarded as less irritating. However, the wideband of their
harmonic spectra is more likely to induce mechanical resonances, which some-
times may lead to higher noise emissions. The noise level was found to increase
with the load, although an optimal flux level for minimal noise was also found.

Noise reduction techniques

A means to reduce the noise by arranging the winding in the machine was
presented by Hupe and Kennel (1989). Two inverters were used to drive a
large induction machine equipped with two separate windings. The noise and
the current ripple were reduced by triggering two PWM inverters with inverse
pulses so that the switching frequency harmonic voltages are in antiphase. This
method is only practical for very high power rating induction machine drives.
For smaller motors a similar configuration was described by Chau et al. (2000)
where an increase in the power and a reduction of the noise was achieved by
purposely selecting the phase of the two carriers of the dual inverters, depending
on the number of poles used for different machine operations.
The advantages of using magnetic bearings over conventional ones in order
to reduce the noise emissions has been reported by Maliti (2000). Active noise
control can also be applied for noise reduction in electrical machines (Belkhayat
180 Chapter 9. Introduction

et al., 1997). In order to suppress some magnetic noise components, it is pos-

sible to inject into the stator a number of current harmonics, and control their
magnitudes and phases with their frequencies linked to the rotation speed. If
one of the force waves has the same magnitude, frequency and mode number
as the natural one but is in opposite phase, the two force waves will nullify and
the corresponding magnetic noise will disappear. However, the process of active
noise reduction is considerably expensive since it requires sensors and advanced
control tools for real time processing.
Classical books treating different aspects of acoustic noise and vibration in
electrical machines are published by Yang (1981) and Timar (1989).

9.2 Objectives and contributions

Although the acoustic influence of different modulation strategies such as DTC,
PWM and PFM is available in the literature, there is a lack of studies treating
the acoustic problem in vector controlled drives. Since the use of this control
strategy is widely spread in electrical drives, therefore the importance of exam-
ining its noise emissions. Hence, the major objective of this part of the thesis
is to investigate the mechanisms responsible for the noise emissions when the
vector control strategy is used in induction motor drives. With this purpose, a
digital drive system has been developed for a 2.2 kW induction motor, and this
tool has been used for the experimental evaluation of the noise emissions when
the flux and/or the torque are modulated with high frequency noise signals.
The study does not take into consideration the harmonics introduced by the
switching in the converter. These harmonics are moved outside the frequency
range for the measurements, by using a 20 kHz switching frequency.
Another objective of this work is to develop a process for the prediction of
the noise emissions of electromagnetic origin, by means of numerical simula-
tions. In this way, the acoustic behaviour of a machine using a certain control
strategy could be studied and optimized already at the design stage. The de-
terministic method is used for the noise prediction, where the electromagnetic
forces are calculated using a model based on FEM. The vibration characteris-
tics are determined using a structural model, also normally based on FEM. The
results from these two models are combined into an acoustic model, based on
BEM, where the radiated noise is calculated.
The nature of vector control links the modulation of the currents to a char-
acteristic force distribution in the machine, where the influence of radial and
9.3. Outline 181

tangential forces may be weighted in order to develop suitable structural changes

to reduce the noise emissions. A study of some structural changes in the con-
ventional stator of an electrical machine has been conducted with this objective.
In particular, the effect of introducing air gap layers around the outer part of
the stator core in order to interfere with the natural transmission path of the
vibrations has been analyzed.

9.3 Outline
The material in this part of the thesis is divided into seven chapters. In this
Chapter, a general literature review dealing with relevant aspects of acoustic
noise in electrical machines has been presented.
Chapter 10 deals with the practical implementation of the control system
used for the experiments. Its dynamic capabilities are demonstrated inducing
the machine to vibrate with high frequency music signals.
In Chapter 11 the results from the measurements are presented. Sound
pressure and sound power measurements were conducted inside an anechoic
chamber at no load and load. The sound power measurements were performed
using the sound intensity measurement technique.
Chapter 12 presents an analytical computation of the airgap force in the
machine. The method assumes ideal iron, and the slotting effects are included.
The purpose of this study is to validate the electromagnetic FEM model used
for the noise prediction.
Chapter 13 deals with the noise prediction process, based on numerical
simulations using FEM and BEM. The available structural models are described
first, followed by the electromagnetic and acoustic calculations. The results are
compared to the measurements in Chapter 11.
Chapter 14 presents a study of the effect that peripheral air gaps in the stator
core have in the overall acoustic emissions. The study was performed using
FEM and BEM models and serves as a complement to previous experimental
work using a similar concept.
Finally, in Chapter 15 a summary of the most important results and con-
clusions of this part of the thesis is presented. It also contains some suggestions
for possible future work in this area.
182 Chapter 9. Introduction
Chapter 10

Control system

10.1 Introduction to Vector Control

Electrically driven industrial and household applications most often require the
flexibility of adjusting the speed of the process depending on the demands. This
is achieved by means of rapid changes in the shaft torque of the electrical ma-
chine incorporated in the drive. In the past, d.c. motors were used extensively
in variable-speed drives since their flux and torque could be easily controlled by
the field and armature currents respectively. However, the commutator and the
brushes in a d.c. motor lead to additional maintenance costs. This is a major
drawback compared with alternating current machines, which present the ad-
vantage of having smaller dimensions and therefore higher output ratings for
low weight and low rotating mass. The developments in the fields of power
electronics and computing capability in the past 20 years brought cheaper and
more sophisticated converters and digital signal processors (DSP), which made
possible the incorporation of a.c. motors into variable speed drives, where new
control strategies could be implemented.
The conventional way of controlling the voltages fed into the motor from
a frequency converter is to maintain constant the ratio between the amplitude
of the voltage and its frequency, i.e. the flux in the machine. With these drives,
more commonly known as ‘v/f drives’, the ratio of the supply frequency to the
motor voltage can be increased in the converter, which also increases the slip.
This in turn leads to a higher induced voltage in the rotor and subsequently
to a larger rotor current. This process is to a large extent dictated by the large
rotor time constant, which implies that rapid torque changes are difficult to

183
184 Chapter 10. Control system

implement using a conventional controller.

In order to achieve better transient response, another control strategy known
as ‘vector control’ was introduced. This theory is well reported in the literature
(Vas, 1990), (Novotny and Lipo, 1996). The basic principle is that the alter-
nating currents in the machine can be expressed as a rotating vector, and by
controlling the amplitude and the phase of this vector, both the amplitude of
the flux and the torque can be changed independently. Hence, the dynamics of
the machine are set directly by controlling the currents, as in the case of the d.c.
motor. However, in an a.c. machine the process is more complicated since not
only the modulus of the current vector but also its phase has to be controlled.
The control problem of an a.c. motor would then be reduced to the same case
as for the d.c. motor if the current vector could be seen as stationary from a
proper coordinate system.
The current vector rotates with the synchronous speed ωs , and the flux in
the machine rotates with the same frequency although ‘delayed’ in phase with
respect to the current under a certain load. An observer that is oriented towards
the flux vector will see a constant current vector, which can be expressed as
a combination of two d.c. components, one pointing in the same direction
as the flux vector and the other perpendicular to this direction, as shown in
Figure 10.1.

q s d
is Ψ
isq
isd θs

Figure 10.1: Stator flux oriented model.

Two different coordinate systems can be distinguished. On one hand the

αβ system which is stationary and is fixed to the stator structure of the motor.
On the other hand the dq system, rotating with the speed ωs , where the direct
10.1. Introduction to Vector Control 185

axis d coincides with the stator flux vector (or the rotor flux vector, depending
on the control strategy) and the quadrature axis q lies perpendicular to this
vector. The stator current is has two components: isd along the d-axis, which
is the magnetizing component of the current that controls the magnitude of the
flux in the machine; and isq along the q-axis which controls the phase change
and thus the speed of the flux vector. In most cases the direct current will be
constant, except in the field weakening region. In general, the torque produced
in any a.c. machine is expressed as in (10.1) (Krause, 1986).
3p
Te (t) = [ψsd (t)isq (t) − ψsq (t)isd (t)] (10.1)
2K 2
In this equation, p represents the number of pole pairs and K is a scaling con-
stant of the space vector. In an asynchronous machine ψsq is zero, while ψsd is
maintained constant, so the torque can directly be controlled by changing isq .
While in v/f drives the torque and flux response are coupled with each other,
using vector control allows independent control of the currents governing the
torque and the flux in an induction motor, and the transient response charac-
teristics become similar to those of separately excited d.c. machines. In terms of
control, the ‘d.c.’ nature of the direct and quadrature currents in an a.c. motor
suggests some sort of similarity with the control of d.c. machines. In fact, isd
and isq correspond to the field and the armature currents respectively in a d.c.
motor, and the stationary field imposes a fixed current distribution in the rotor,
as shown in Figure 10.2(a). A similar current distribution appears in the rotor
of an induction motor under load, with the difference that this distribution is
shifted continuously, driven by the time varying stator flux, see Figure 10.2(b).
However, an observer placed in the rotating dq system will see a stationary dis-
tribution, as in the case of the d.c. motor. Under zero slip conditions, the rotor
current is negligible in an induction motor, and only a direct axis current dis-
tribution will appear, as shown in Figure 10.3(a). When a load is applied, the
slip increases and a quadrature current is induced in the rotor, which must be
compensated by an additional stator current, as indicated in Figure 10.3(b), and
the sum of the current from the rotor and the stator for the quadrature axis is
ideally zero. The resultant current distribution in the stator winding is found
by the superposition of the direct and quadrature current distributions. In or-
der to change the torque, the instantaneous three phase currents are changed so
that a certain quadrature current is produced, while the direct current remains
unchanged. In case that a transient error appears in isd , that will not result in
186 Chapter 10. Control system

(a) D.C. motor (b) Induction motor

Figure 10.2: D.C. and Induction motor structures.

any significant flux transient, since the flux dynamics are slow. This decoupled
response between flux and torque control constitutes the main advantage of
vector controlled drives compared to v/f drives, improving the response under
transient conditions. However, the behaviour of both drives should be simi-
lar under steady state operation. Vector drives are most frequently used with a
speed control loop where the torque reference is continually adjusted, in what
could be defined as a quasi-continuous transient operation. The advantages
of vector control both in ease of control and faster transient response have in-
creased the range of applications where vector drives are being used. As a result
of this market trend, the implications of using this type of drive in terms of
acoustic noise emissions must be therefore examined.

10.2 Noise emissions in vector controlled drives

The electromagnetic noise emissions in an electrical machine are directly deter-
mined by the harmonic content of the relevant forces in the airgap of the ma-
chine, where the electro-mechanic energy conversion takes place. Every struc-
ture tends to vibrate at certain frequencies, called resonance frequencies, with a
10.2. Noise emissions in vector controlled drives 187

(a) Direct axis current distribution. (b) Quadrature axis current distribution.

Figure 10.3: Field oriented current distributions in the induction motor.

characteristic deformation, called mode of vibration. When the frequency of one

harmonic in the airgap force coincides with one of the resonance frequencies
of the machine, it vibrates with that frequency in a certain fashion producing
noise. The noise emissions at other frequencies depend on their proximity to a
resonance frequency and their damping.
The harmonic content of the airgap forces is directly determined by the
harmonics present in the motor currents as well as other construction factors
such as the effects of the slots, eccentricity, etc. It is not the aim of this work to
analyze the influence of the harmonics introduced by the converter or the dif-
ferent construction parameters. Instead, the objective is to experimentally de-
termine how the motor responds acoustically to different current excitations of
the direct and/or quadrature currents using vector control, and verify these mea-
surements with the aid of computational tools. The direct current (d-current)
controls the flux, and is responsible for the radial forces, while the quadrature
current (q-current) controls the torque, and is responsible for the tangential
forces. Therefore, apart from measuring the difference in the motor acoustic
behaviour when the flux and/or the torque are modulated, the study would
188 Chapter 10. Control system

provide information about the role that radial and tangential forces play in the
overall acoustic emissions of small electrical machines.
The desired excitation can be achieved by adding to the reference direct
and/or quadrature currents an external a.c. signal, that could be a sine wave
with a certain amplitude and with a frequency corresponding to one of the res-
onance frequencies of the stator. The flux is modulated by adding a noise har-
monic dmod to the reference for the direct current, as shown in Figure 10.4(a).
The amplitude of the current vector will vary in a sinusoidal fashion along the
direction of the flux, providing that the quadrature current is set to zero, and
therefore exciting the stator with radial forces. The torque is modulated by
adding the noise harmonic qmod to the reference for the quadrature current, as
shown in Figure 10.4(b). This reference is either zero, at no load, or the cor-
responding load current. The harmonic qmod will produce a phase shift of the
flux vector, which will excite the stator with tangential forces.

q qmod
q

Ys is
Ys
d d
Qs iq
is Qs
dmod
id
a a

(a) Direct current modulation (b) Quadrature current modulation

Figure 10.4: Modulation of the direct and quadrature currents.

Figure 10.5 illustrates how the stator is excited and the force pattern in the
air gap when modulating the torque and the flux. The oval shape corresponds
to the force distribution around the airgap for a given current. When only the
flux is modulated, the oval shape will shrink and expand in the radial direction.
When only the torque is modulated the oval shape will be rotated around the
center of the machine, while keeping its amplitude.
10.3. Stator flux control 189

Figure 10.5: Airgap stress distribution and stator excitation for flux and torque
modulation.

10.3 Stator flux control

The electrical behaviour of an asynchronous motor can be simplified in terms of

an electrical circuit and there are namely two ways of representing the equivalent
diagram of an induction motor, depending on the number of parameters taken
into account. The five parameter model is presented in Figure 10.6, where the
steady state value Rr /s has been replaced by the rotor resistance in series with
a voltage source which corresponds to the back EMF. In the equivalent four
parameter model the rotor and stator leakage inductances are grouped together
in the stator side, as shown in Figure 10.7.

is Rs Lsλ Lrλ Rr ir

+ ism
us s
Ψ Lm s
Ψ s
jwm Ψ
r r
−

Figure 10.6: Five parameter model.

190 Chapter 10. Control system

is Rs Lσ RR iR

+ is
M
us s
Ψ LM s
Ψ s
jwm Ψ
R R
−

Figure 10.7: Four parameter model.

The parameters in both models are related in the following way:

2
Lm
RR = · Rr (10.2)
Lr
L2m
LM = (10.3)
Lr
Lσ = σLs (10.4)
L2m
σ = 1− (10.5)
Ls Lr

The four parameter model will be used to implement the control system
and to model the machine. For the control system, it can be assumed that
either the rotor or the stator flux vector coincides with the direction of the direct
axis. Since measuring the stator currents and estimating the stator flux is more
convenient than the rotor flux and currents, the stator flux model represented in
Figure 10.1 will be followed. In the stationary αβ coordinate system the stator
voltage can be written in vector form as in (10.6).

sαβ (t)
dψ
uαβ
s (t) = Rs
i αβ
s (t) + (10.6)
dt

This equation can be transformed into the synchronous dq reference frame

where the direct axis is aligned with the stator flux vector, which sets the speed
of rotation ωs . The rotor will rotate with the electrical speed ωm and the slip
frequency becomes ωsl = ωs − ωm . The stator voltage in synchronous coordi-
nates is expressed as in (10.7).

sdq (t)
dψ  dq (t)
s (t) = Rs is (t) + + jωs ψ
udq  dq
s (10.7)
dt
10.3. Stator flux control 191

The voltage can be separated into its direct and quadrature components,
introducing that ψsq = 0 and ψsd = ψs for the coordinate system selected, as
in (10.8) and (10.9) respectively.
dψs (t)
usd (t) = Rs isd (t) + (10.8)
dt
usq (t) = Rs isq (t) + ωs ψs (t) (10.9)
Based on these equations, the direct and quadrature current controllers can
be deduced separately, following the strategy of ‘predictive dead beat control’
(Alaküla, 2002). The detailed deduction of these controllers is included in
Appendix E. It is shown that the controller for the direct current is given by
(10.10). This corresponds to a PI controller with a proportional gain equal to
LM + Rs2Ts and an integral gain equal to Rs Ts , where Ts is the sampling time.
The reference signals are denoted by the superscript ‘*’. The actual value of the
direct current is obtained from the flux level in the machine, which is difficult
to measure. As an alternative, this flux can be estimated as will be described
later in this chapter.
j=k−1
Ts u∗sd (tk ) = Rs Ts [i∗sd (tj ) − isd (tj )] +
j=0
Rs Ts
+ LM + [i∗sd (tk ) − isd (tk )] (10.10)
2
The controller for the quadrature current can also be implemented as a
PI controller
 (10.11).
 Now the  proportional and integral gains are equal to
Lσ + Rs + LM RR Ls Ts
2 and Rs + LM
RR Ls
Ts , respectively. There is also a
feed forward term ωm (tk )ψs (tk )Ts which depends on the speed of the ma-
chine. This controller assumes as an input the reference value of the quadrature
current, which can be easily obtained from the torque equation in (10.12),
where only the flux has to be estimated as in the case of the direct controller.
j=k−1
RR Ls " #
Ts u∗sq (tk ) = Rs + Ts i∗sq (tj ) − isq (tj ) +
LM
j=0
 
RR Ls Ts " ∗ #
+ Lσ + Rs + isq (tk ) − isq (tk ) +
LM 2
+ ωm (tk )ψs (tk )Ts (10.11)
192 Chapter 10. Control system

Figure 10.8: Controller structure.

Te∗ (tk )
i∗sq (tk ) = (10.12)
ψs (tk )

10.4 Implementation
The complete controller structure is presented in Figure 10.8. The set point
value for the direct current results from the addition of two signals: the mag-
netizing current, obtained as the reference flux divided by the stator induc-
tance, and an external noise signal defined as musl . The reference value for the
quadrature current is set by a speed controller, and another noise signal musr
is added. The signals musl and musr correspond in fact to dmod and qmod .
In this chapter the experiments will be performed with music signals, there-
fore the denomination. The measured three phase currents in the motor are
input into the controller in stationary coordinates and then transformed to the
synchronous frame. This transformation is achieved by estimating the angle of
rotation θ̂s from the flux vector using the relation in (10.13).
 ψsα ψsβ
|ψs | = ψsα 2 + ψ2 cos θ̂s = sin θ̂s = (10.13)
sβ |ψs | |ψs |
10.4. Implementation 193

Figure 10.9: Speed observer.

The noise signals can contain passages with a d.c. offset, which added to
the quadrature current component will cause the rotor to accelerate since a non
zero average torque will be produced. This phenomenon is avoided with the
speed control loop, where the actual speed is estimated from the flux vector,
as shown in Figure 10.9. The speed of rotation of the flux vector is approx-
imated as the rate of change of the quadrature component of the flux during
one sampling interval with respect to its modulus. Figure 10.10 shows the path
followed by the flux vector between two consecutive samples. At time tk−1
the synchronous frame is rotated so that the direct axis dk−1 is aligned with
the position of the flux sampled at that time, ψk−1 . The flux vector will con-
tinue rotating but its actual position will not be read until the next sample at
time tk , when the synchronous frame is again rotated so that the new direct
axis dk coincides with the new sampled flux ψ k . An observer placed in the old
synchronous frame (d/q)k−1 will see that the flux vector was rotated by an ad-
ditional quadrature component ψqk . The flux at time tk can be expressed in the
old coordinate system as the sum of a direct component |ψ k−1 | ≡ |ψk |, and a
quadrature component ψqk . The angle of rotation between the two samples is
approximated as in (10.14).

ψqk
δ = θk − θk−1 = (10.14)
k |
|ψ

The speed of rotation of the flux vector is calculated as the derivative of δ,

which in a sampled system can be approximated as Tδs , where Ts is the sampling
194 Chapter 10. Control system

time. This is actually the process described in Figure 10.9, where the quadra-
ture component of the flux is obtained using in the transformation the angle
corresponding to the old synchronous frame, which is easily stored working in
a digital system.

β
qk−1 dk
qk
k ψqk
ψ
k−1
ψ dk−1
θk θk−1
α

Figure 10.10: Flux vector path.

0 αβ
The flux is estimated by substituting the stator current iαβ
s
ψs
Ls into the
stator equation (10.6), leading to the expression of a low pass filter in (10.15).
The three phase voltages are transformed into the stationary frame and (10.15)
is implemented in an analog circuit as shown in Figure 10.11.

sαβ
Rs  αβ dψ
uαβ
s = ψ + (10.15)
Ls s dt

The control system and the process were implemented in S IMULINK, and
the overall structure is presented in Figure 10.12. The voltage time area output
from the controller in stationary coordinates is transformed into three phases
which are symmetrized in order to increase the modulation depth of the drive,
by centring the reference voltages with respect to the bus voltage window (Hol-
mes, 1997). The switches in the converter are controlled by the modulator,
where regularly sampled sine-triangle modulation is used. Since the model
10.5. Musical drive test 195

Figure 10.11: Flux observer.

of the induction motor is based on the stator and rotor equations in station-
ary coordinates, the three phase voltages output from the converter have to be
transformed back to stationary coordinates in the simulation. From the motor
model, the measured current is fed back to the controller, and from the applied
voltage the flux and the speed are estimated in the observer. The external noise
signals for the direct and the quadrature current are the controller inputs musl
and musr respectively.

10.5 Musical drive test

An indication of the torque response to a given music sequence was obtained
from the simulated results using the structure showed in Figure 10.12. In this
case the audio signal is fed into the torque modulation input musr , while musl
is set to zero. The input signal and simulated torque response are given in
figures 10.13(a) and 10.13(b). The results provide an overall indication in terms
of the ability of the motor torque to track the sound signal, which is quite
acceptable.
The performance of the drive has also been experimentally tested. The drive
developed consists of a commercial inverter which was re-fitted with a new con-
troller and pulse width modulation boards. The control, vector transformation
and symmetrization modules from Figure 10.12 are implemented in software
and reside on a dedicated DSP board. The modulator unit, which has a carrier
frequency of 19.582 kHz, produces regularly symmetrically sampled PWM sig-
196 Chapter 10. Control system

Figure 10.12: Structure of the drive system.

10.5. Musical drive test 197

(a) Music input sequence (b) Torque response

Figure 10.13: Music input and torque response for musr (t), musl = 0.

nals which are connected to the IGBT modules in the converter. A 2.2 kW two
pole asynchronous motor was used, which was not connected to any external
load. The line currents and the motor voltage are measured and used by the
controller and the combined speed/stator flux estimator respectively. The drive
was tested by providing either one or both channels of a stereo input amplifier
to the drive. As predicted by the simulation, the torque and/or flux ripple fol-
lowed the musical torque and/or flux reference up to high input frequencies,
which is possible using a switching frequency in the modulator of almost 20
kHz. This ripple induces vibrations in the motor structure with the frequency
of the music and the machine behaves as a loudspeaker. If the motor is placed
on a surface the vibration from the motor structure will be transferred to that
surface and further sound amplification will be obtained. Interesting is also to
compare the different music quality that was obtained from different surfaces,
being the most suitable a wooden box, which best resembles the commercial
loudspeakers. When only the torque was modulated, a listener could perceive
more low frequency components in the music emitted, whereas the high fre-
quency passages were more accentuated when pure flux modulation was used.
As a consequence, the music quality as perceived by the listener was found to
be at its ‘best’ when both flux and torque inputs were used simultaneously. An
indication of the audio sound capabilities from this setup was achieved by mea-
198 Chapter 10. Control system

suring the ‘A’ weighted sound power over the audio frequency range. In this
case only the audio input musr (which modulates the torque reference) was
connected to a sinusoidal voltage source whereby the frequency was varied from
200 Hz to 10 kHz. The obtained results, measured in a reverberating room,
are shown in Figure 10.14. Included in this figure, by way of reference, is a
loudspeaker transfer function. These results indicate that the drive sound levels
at opposing ends of the frequency band are lower than those obtained from the
loudspeaker. The low frequency sound level fall off is caused by the radiation
efficiency of the motor, i.e. the sound radiation efficiency is reduced at lower
frequencies (Wang et al., 2002). The high frequency fall off is due to the band-
width of the current control loop. For a fixed PWM frequency, the number of
carrier wave pulses in each period decreases for an increasing frequency of the
reference signal and the resolution is then affected by this factor. Also, readily
apparent from Figure 10.14 are two motor resonance peaks found near the fre-
quencies 1.3 kHz and 2.7 kHz. It should be pointed out that the machine used
for the musical drive presented in this chapter is not the same as the machine
used for the experiments in Chapter 11.

10.6 Conclusions

Vector control provides faster transient response compared to conventional v/f

drives, and this is achieved by an efficient decoupling of the currents controlling
the flux and the torque in the machine. These currents are easily controlled since
they are regarded as d.c. components from a rotating coordinate system linked
to the stator flux vector. A simulation program including the control system and
simplified models for the power electronics, motor and observers has been pre-
sented. The control system has been implemented in a DSP card, which in turn
is connected to a PWM card where the switching frequency was set to 20 kHz.
The performance of the drive was tested by modulating music signals as noise
references added to the direct and the quadrature currents. The motor structure
resembled the behaviour of the core of a loudspeaker, reproducing efficiently
the noise signals. The musical motor application served to demonstrate the ca-
pabilities of vector controlled drives implemented in advanced digital systems,
leading to a suitable tool for acoustic research in electrical machines. This tool
will be used in this thesis to analyze more in detail the acoustic radiation from a
similar size induction motor using different combinations of direct and quadra-
ture current excitation at key frequencies, so as to ascertain the mechanisms of
10.6. Conclusions 199

Figure 10.14: Measured sound power of the ‘musical’ drive for musr (t),
musl = 0.

noise production in small electrical machines.

200 Chapter 10. Control system
Chapter 11

Acoustic measurements

11.1 Introduction

In this chapter the results of the sound pressure and sound power measurements
are presented for different excitations of the direct current (radial force) and the
quadrature current (tangential force). The measurements were carried out both
at no load and load conditions in an anechoic chamber. Sound measurement
techniques have been well reported in the literature (Norton, 1989), (Harris,
1991), and a summary of the most commonly used techniques is presented in
Appendix F, together with some basic acoustic terminology. Since the sound
intensity measurement technique was used for the experiments, it is described
before dealing with the experimental set-up and results.

11.2 Sound intensity measurement technique

Traditionally, the only way of measuring how much sound power a machine
radiates was to place the machine in an anechoic or reverberation room and
measure the sound pressure, which is related to sound power depending on the
nature of the room. However the sound power can also be determined by means
of sound intensity measurements (Fahy, 1995), where the power level is mea-
sured directly in situ. Sound intensity measurement systems became available
in the middle of the 1980’s, and practical applications of the technique have
since increased at a very fast pace.
The intensity at one point is expressed as the product of the pressure and
the particle velocity at that point and is related to the sound power as shown in

201
202 Chapter 11. Acoustic measurements

(11.1) (Brüel & Kjær, 1986).

F orce Distance Energy P ower

I = pu = · = = (11.1)
Area T ime Area · T ime Area
Sound intensity measurements are conducted by placing the noise source of
interest inside an imaginary surface enclosing the sound source. For simplicity,
the enclosure is usually selected to be a rectangular frame. The average intensity
flowing through each surface is measured with a sound intensity probe, and the
information is stored in a spectrum analyzer. Multiplying this intensity by the
area of the surface gives directly the power. The total sound power radiated by
the source is calculated by adding the contributions from all the surfaces.
The advantage of this technique is that the measurements can be performed
in situ, even with steady background noise and in the near field of machines.
The requirement is that no other noise source apart from the one subject to
study must be located inside the enclosure. An external noise source will not
produce any net contribution to the calculated intensity, since the energy will
flow in and out of the enclosure and its overall effect will cancel when the power
contributions of the different sides are added. It is important to note that this
only holds if the external source is stationary, i.e. its noise does not vary in
time, and if there is no absorption material within the box as otherwise some
background noise will not flow out of the box again.
Sound pressure can be measured with a single microphone, but particle
velocity is more complicated. The equation governing the motion of the fluid
particles is the Navier-Stokes equation. Neglecting viscous effects and in the
absence of mean flow and turbulence it can be expressed as in (11.2), where r
represents the direction of the flow.

δur δp
ρ =− (11.2)
δt δr
This equation can be re-written using finite-difference as in (11.3).

1
ur = − (pB − pA )dt (11.3)
ρ∆r

Hence, the particle velocity can be determined by measuring the pressures

pA and pB detected by two microphones closely spaced at a distance ∆r, as
shown in Figure 11.1.
11.2. Sound intensity measurement technique 203

90o -∞ dB

r A spacer B 0o 0 dB

∆r

Figure 11.1: Sound intensity microphone and dipole characteristic.

Sound pressure is a scalar quantity, i.e. it is the same whatever the angle
of incidence. However, sound intensity is a vector quantity. When a two-
microphone probe is used, only the component of the vector along one direc-
tion, the probe axis, is measured. The dashed circumferences in Figure 11.1
show an approximation to the dipole behaviour of the sound intensity probe.
For sound incident at 90o from the axis there is no component along the probe’s
axis, as there will be no difference in the pressure signals from the two micro-
phones. Hence, the reduction in the measured intensity level (in dB) will be
infinite because the particle velocity, and hence the intensity is zero. For sound
incident at 0o , there will be zero reduction and the probe will measure the cor-
rect value. For sound incident at any other arbitrary angle θ from the axis, the
intensity component along the axis will be reduced by a factor of cos θ.
The choice of spacer depends on the frequency range to be covered, the
higher the frequency the shorter the spacer should be for the finite-difference
approximation in (11.3) to hold. However, accuracy at low frequencies also
depends on the spacer used. In all analyzing systems there will be a small time
delay between the two channels which introduces a small phase change, influ-
encing the results. This is a bias error called phase mismatch error. For a given
probe and analyzer combination, this error can be estimated to be around ±
0.3o (Brüel & Kjær, 1986). For an accuracy to within 1 dB the phase change
over the spacer distance should be more than five times the phase mismatch.
In order to obtain a negligible high frequency error, the wavelength must be at
least six times the spacer distance. Thus, for a given spacer the inherent phase
mismatch of the equipment will cause a significant error in the measured inten-
204 Chapter 11. Acoustic measurements

sity at low frequencies. Since no single spacer can cover a wide frequency range,
the spacer should be selected depending on the frequencies of interest.
When the sound is incident at an angle θ from the probe’s axis, the phase
change that is detected is reduced by cos(θ), and this will imply that the phase
mismatch error will become more significant and the measurement range will be
further reduced. One indicator of the accuracy of sound intensity measurements
is the pressure-intensity index, LK , which is measured simply as the difference
between the measured sound pressure level and the measured sound intensity
level (Fahy, 1995). By measuring the pressure-intensity index the phase change
across the spacer can be determined, and the influence of the phase mismatch in
the accuracy of the results can be calculated. In a general sound field, the phase
change varies from point to point, but an average global value is often sufficient
to give an idea of the accuracy. In an ideal free field, the difference between the
sound pressure and sound intensity levels is zero. However, the phase mismatch
causes a small difference between the two signals, as if there were some intensity
along the spacer. Hence, apart from the errors introduced by the operator dur-
ing the measurements, which are reported through the pressure-intensity index,
there is another source of error inherent to the equipment that cannot be further
reduced, and it is called the residual pressure-intensity index, LKO . This index
can be determined by using calibrating equipment to feed the same signal to
the two microphones and measuring the difference between the sound pressure
level and the sound intensity level. The dynamic capability LD of the sound
intensity measuring system is defined in (11.4).

LD = LK0 − K (11.4)

The measured pressure-intensity index LK should be less than the dynamic

capability for an acceptable accuracy in the measurements. For an accuracy to
within 1 dB, the constant K is equal to 7 dB, which corresponds to a phase
change in degrees along the spacer distance of five times the phase mismatch.

11.3 Experimental set-up and instrumentation

The machine used in the experiments is a three-phase four-pole 415 V 50 Hz

2.2 kW, three phase, induction motor, with 36 and 44 stator and rotor slots
respectively, see Figure 11.2. The casing diameter is 163 mm, the total length
278 mm, and the lamination stack 92 mm. This machine was used in purpose
11.3. Experimental set-up and instrumentation 205

since there was detailed information available about its vibrational behaviour
(Wang and Lai, 1999).

Figure 11.2: The test motor. Source (Wang and Lai, 1999).

The internal dimensions of the anechoic room used for the measurements
are 3.5 m x 3.5 m x 3.5 m, and the cut-off frequency is 150 Hz. The motor
was mounted on a bench in the center of the room, and the seat was attached to
an aluminium plate through the four isolators. This plate was softly connected
to another 10 mm thick aluminium plate inserted between the test motor and
the load, as shown in Figure 11.3, in order to minimize the noise contribution
from the load motor during the measurements.
The equipment for running the machines was placed outside the anechoic
room, see Figure 11.4. The load motor was fed by a commercial inverter with
a fixed switching frequency of 8 kHz. The test motor was driven from an-
other converter connected to a control system module. This module contained
hardware for the estimation of the flux from the three phase measured voltage,
the measurement of the currents and the scaling of the d.c. link voltage in the
converter. These signals were fed into a DSP card, which had 16 analog inputs,
and a 50 pin connector cable was used for the connection between the DSP and
PWM cards. The switching frequency set by the modulator was 20 kHz. The
206 Chapter 11. Acoustic measurements

Figure 11.3: Experimental set-up inside the anechoic room.

reference values for the direct and the quadrature main currents in the DSP were
adjusted with two potentiometers, and the added noise currents were provided
by two separate signal generators.
For the sound pressure level measurements a B&K type 4134 microphone
was used, fed from a B&K type 2804 power supply. The signal from the micro-
phone was amplified using a B&K type 2639 amplifier and thereafter sent to
the analyzer. The microphone was calibrated using a B&K type 4220 piston-
phone, taking into consideration the corrections that should be applied depend-
ing on the temperature and pressure measured inside the anechoic chamber.
The acoustical response of the motor to random noise excitation in the range
from 0 Hz to 10 kHz was tested using the white noise signal from a separate
B&K type 4205 sound power source. The microphone used for the measure-
ments was connected to a Hewlett Packard (HP) 3569-A real time frequency
analyzer, where the spectrum of the measured noise was stored for further post-
processing.
The noise components added from the signal generators have the same fre-
quency, and different amplitude combinations are selected for the direct and
quadrature currents. The maximum amplitude of the added noise current is
11.3. Experimental set-up and instrumentation 207

Figure 11.4: Experimental set-up for the control equipment.

referred to as ‘2m’, and its value was determined taking into consideration the
saturation limits in the machine. The mean value of the direct current was re-
duced so that, when the noise current was added, the maximum value of the
resultant waveform did not exceed the nominal value of the direct current, see
Figure 11.5. The effective magnetizing current in the motor was 2.21 A, which
when using the power invariant transformation gives a peak direct current of
3.82 A. The maximum amplitude of the noise excitation (‘2m’) was selected to
be 0.5 A, and subtracting this value from the peak current gives a mean value
of the direct current corresponding to 87% of the nominal. Two other values
were used for the noise current, corresponding to half and one quarter of the
maximum, i.e. 0.25 A and 0.125 A, and they are referred to as ‘m’ and ‘m/2’
respectively. When the machine was loaded, the effective quadrature compo-
nent of the main current was 3.6 A, which corresponded to a peak value of 6.23
A. This current was reduced to 5.7 A in order to avoid exceeding the nominal
value when the noise current was added, in the same way as it was done with
the direct excitation. Thus all the load tests were conducted at 92% of the rated
load.
The values of the total direct and quadrature currents calculated in the DSP
were available at its analog outputs. Two multimeters were used to measure their
208 Chapter 11. Acoustic measurements

5
n o is e c u rre n t n o m in a l
2 m d ire c t c u rre n t
d ire c t c u rre n t [A ] 4

2
re d u c e d d ire c t c u rre n t

0
0 0 .0 0 5 0 .0 1 0 .0 1 5 0 .0 2
T im e [s e c ]

Figure 11.5: Modulation of the direct current.

d.c. components, and an ONO SOKKI CF-350 real time spectrum analyzer
was used to check their frequency content. A ‘Hanning’ window was used and
the spectrum was averaged from 256 samples with 50% overlap each time the
currents had to be adjusted for the measurements. As expected, it was observed
that the current spectrum was completely flat except at the frequency set with
the signal generators.
For the sound intensity measurements, a HP 35230-A two microphone
sound intensity probe was used, equipped with two 0.5 inch condenser mi-
crophones of the type HP 35237-A. The spacer used was 12 mm long, cov-
ering 1/3 octave center frequencies from 125 Hz up to 5 kHz. The measure-
ments were performed according to ISO 9614-2 (ISO 9614-2, 1996), which is
a draft international standard for measurement of sound power via surface scan-
ning of sound intensity. One requirement is to measure the residual pressure-
intensity index for the entire measurement system. This value is directly related
to the dynamic capability LD , by subtracting 7 dB, and should be compared to
the pressure-intensity index obtained from the measurements. If the pressure-
intensity index is greater than LD in any 1/1 or 1/3-octave band, the sound
power measurement is inadequate in that band. The residual pressure-intensity
index of the measurement system was determined using a B&K ZI0055 piston-
phone mounted on a B&K UA0914 sound intensity calibrator, and it is shown
11.3. Experimental set-up and instrumentation 209

in Figure 11.6.

2 0

1 9

1 8
R e s id u a l In te n s ity In d e x [d B ]

1 7

1 6

1 5

1 4

1 3

1 2

1 1
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 11.6: Measured residual pressure-intensity index.

Sound pressure as well as sound intensity measurements were conducted in

order to assess the sound pressure level (SPL, Lp ) and the sound power level
(PWL, LW ) from the machine. All the measurements were conducted with the
motor placed inside the anechoic chamber. The sound pressure measurements
were carried out first with the machine at stand still, at four different positions.
The response was measured for two different excitation frequencies, 490 Hz and
1105 Hz. These frequencies correspond to two of the resonance frequencies of
the stator-casing coupling measured experimentally in (Wang and Lai, 1999).
Various combinations in the amplitude of the direct (D) and quadrature (Q)
noise current components were selected as shown in Figure 11.7. The case when
white noise is used is referred to as ‘wn’. The same experiments were conducted
with the machine rotating at 70 r.p.m. This low speed was selected in order
to minimize the effects of the aerodynamic and the mechanical noise. The
sound intensity measurements require considerably more effort, and a smaller
number of combinations were tested, as shown in Figure 11.8. At load, the
same experiments as at no load were conducted, with the exception of the stand
still tests.
210 Chapter 11. Acoustic measurements

D Q

0 0

m/2 m/2

m m

2m 2m

0 0

wn wn
Figure 11.7: Diagram of combinations tested with SPL measurements.

D Q

0 0

m m
Figure 11.8: Diagram of combinations tested with PWL measurements.
11.4. No load tests results and analysis 211

11.4 No load tests results and analysis

Sound pressure tests at stand still and no load.
The measurements were conducted with the rotor fixed at four different posi-
tions, spaced 90 degrees between each other. In order to accurately determine
the position of the rotor, a mechanical system was attached to the shaft and
fixed to the test frame, as shown in Figure 11.9. Four holes were drilled around
the wheel attached to the shaft, and by coupling them with the bar resting on
the test frame the rotor was rotated to the desired position.

Figure 11.9: Set-up for the stand still tests.

The SPL was measured with a single microphone placed inside the anechoic
chamber. The microphone was positioned at half the length of the machine,
and the distance to the closest point at the top of the machine was 337 mm in
the horizontal direction and 54 mm in the vertical direction. The resolution
for the narrow band analysis in the spectrum analyser was 1600 lines in a range
from 0 Hz to 6.4 kHz. These results were post-processed in order to obtain a
single representative value for each measurement, which would serve as a basis
for their comparison. This was achieved by adding the values at the 1600 point
frequencies from the spectrum, using (F.2), where Lp is the total SPL and Lpi
is the SPL at each frequency i.
212 Chapter 11. Acoustic measurements

Figure 11.10 shows the results obtained at the four positions when a noise
harmonic of 490 Hz and amplitude ‘m’ is added to both the direct and quadra-
ture currents. In this figure and similar in this chapter, the modulation of the
excitation will be defined in brackets. The value associated with ‘d’ and ‘q’ corre-
sponds to the amplitude of the noise current added to the direct and quadrature
main currents respectively. Above each plot, the overall value of SPL is included
both in dB and dB(A). A comparison between all the cases at the four different
positions is shown in Figure 11.11. This comparison is based on the total SPL
in dB for each measurement, which was preferred to dB(A) since it gives more
appropriate information about the real sound emitted from the motor, without
being modified by the frequency response of the human ear.

Figure 11.10: SPL at no load, stand still, four positions. Case [d=m, q=m], f=490
Hz. Rotor rotated 0 degrees (P1), 90 degrees (P2), 180 degrees
(P3) and 270 degrees (P4).
11.4. No load tests results and analysis 213

Figure 11.11: Summary of SPL at no load, stand still, four positions.

214 Chapter 11. Acoustic measurements

In Figure 11.11 and similar figures in this Chapter, the different cases are
represented as a combination of two terms separated by a coma along the hori-
zontal axis. The values of the curve corresponding to the ‘d’ current in the leg-
end should be read as the case where the direct current corresponds to the first
term and the quadrature current to the second term of the combination. Simi-
larly, for the values of the curve corresponding to the ‘q’ current in the legend,
the quadrature current corresponds to the first term, and the direct current to
the second term of the combination. For example, in the combination ‘m/2,0’
the value of the curve associated with the ‘d’ in the legend corresponds to the
combination [d=m/2, q=0], whereas the value of the curve associated with the
‘q’ corresponds to the combination [d=0, q=m/2]. Also note that the combina-
tions have been selected so that the value of the modulation of the first term is
higher than the second value. This means that the ‘d’ curve represents the cases
where the modulation of the direct current is greater than the quadrature cur-
rent and this will be referred to as direct modulation. The ‘q’ curve corresponds
to the cases where the modulation of the quadrature current is greater than that
of the direct current, and this will be referred to as quadrature modulation.
From Figure 11.11, it can be observed that for every case the total Lp for
the direct modulation is greater than the corresponding quadrature modulation.
However, the values for the same case at different rotor positions may vary
considerably. For example, in Figure 11.10 the acoustic response at different
rotor positions varies from 69.2 dB to 75.7 dB. This indicates that the airgap
force exciting the structure is modified. Since the applied current is constant in
all positions, the phenomenon should be due to a change in the airgap geometry.
This is possible if the airgap is irregular and the position of the minimum airgap
length around the machine is changed when the rotor is rotated, i.e. if there is
dynamic eccentricity. The airgap forces are then re-distributed, affecting the
excitation of the structure and the noise radiation pattern at different rotor
positions. This leads to a change in the measured value at a fixed point.
The results obtained by averaging the values between the four different po-
sitions are presented in Figure 11.12. As expected, the overall Lp increases with
the level of modulation, from ‘m/2’ to ‘2m’. The graphs also show that, in all
the cases, the level is higher when modulating the direct current than in the case
of the quadrature current. At 490 Hz, the differences are around 7 dB when
one of the noise currents is set to zero and the other one is modulated with
different amplitudes. At 1105 Hz, the difference at these points changes from
4 dB to 9 dB. These increments are maintained at 1105 Hz when both currents
11.4. No load tests results and analysis 215

are modulated, while at 490 Hz the direct and quadrature response are closer
to each other. Comparing the results at both frequencies, the Lp at 490 Hz is
around 10 dB higher than at 1105 Hz.
The main indications from the stand still tests are the higher noise emissions
obtained in the case of direct current modulation, as well as the influence of
the dynamic eccentricity in the measurements. It was concluded that it was
preferable to perform the measurements with the rotor rotating, in order to
obtain an average value from all the positions.

Figure 11.12: Averaged no load SPL results at stand still.

Sound pressure tests at low speed and no load.

A low rotational speed of 70 r.p.m. was selected so as to minimize the mechan-
ical noise caused by the vibration of the machine, and the aerodynamic noise
since the machine is cooled by means of a fan attached to the rotor. Never-
theless, the case when no noise current is added to the fundamental was also
216 Chapter 11. Acoustic measurements

measured, see Figure 11.13. The narrow band plot was post-processed and con-
verted into a 1/3 octave band plot, in order to improve its visual inspection.
The center frequency was calculated using (F.6) and the single Lp values for the
frequencies included in each band were added using (F.2). The total Lp is re-
ferred to as Lpav and it is included above the narrow band plot. Due to the high
switching frequency of 20 kHz used, the contribution from the electromagnetic
noise caused by the harmonics in the converter might be neglected and the SPL
spectrum in Figure 11.13 could be considered as that from the mechanical and
aerodynamic noise together.

L p a v = 4 1 .3 5 d B , 3 9 .5 5 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]

4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.13: SPL at no load, 70 r.p.m. Case [d=0, q=0].

The results when the direct and the quadrature noise currents were fed al-
ternatively from a white noise source are presented in Figure 11.14 and Fig-
ure 11.15. It can be observed that the SPL is increased in the order of 20 dB,
which implies that the contribution of the mechanical and aerodynamic noise
components is negligible compared to the electromagnetic noise caused by the
induced harmonics.
A summary of the results for the different current excitations both at 490
Hz and 1105 Hz is presented in Figure 11.16. It can be appreciated that the
direct modulation still produces more noise than the quadrature modulation.
The difference is more accentuated at 490 Hz, where the average difference in
sound pressure level is 5 dB, whereas at 1105 Hz this difference decreases to
2 dB. Generally speaking, the overall SPL increases with the increase in the
amplitude of the modulation. The overall sound pressure level is slightly higher
at 1105 Hz compared to that at 490 Hz. The increment is more clearly visible
11.4. No load tests results and analysis 217

L p a v = 6 6 .1 3 d B , 6 1 .7 8 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]
4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.14: SPL at no load, 70 r.p.m. Case [d=wn, q=0].

L p a v = 6 0 .8 d B , 5 8 .7 3 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]

4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.15: SPL at no load, 70 r.p.m. Case [d=0, q=wn].

in the quadrature modulation, with an average difference of 6 dB, than in the

direct modulation, where the average difference is 3 dB. In general, the response
at 490 Hz and 1105 Hz is different since these two frequencies excite different
modes of vibration with different sound radiation patterns.
Compared to the averaged results from the stand still tests, it can be appre-
ciated that the SPL is lower when the machine rotates. Also, the noise level from
the 490 Hz harmonic is higher than that from the 1105 Hz harmonic when the
machine rotates, while at stand still the opposite is true. Actually, the values
from the stand still tests are not a good representation of the average around the
whole machine. It might be the case that the values at the four positions give
an average higher result compared to other positions around the circumference.
218 Chapter 11. Acoustic measurements

Figure 11.16: Summary of SPL at no load, 70 r.p.m.

11.5 Load tests results and analysis

Sound pressure tests at low speed and load.

The machine was loaded with another identical asynchronous motor. The influ-
ence of the noise from the load was minimized by covering it with eight layers
of barium loaded vinyl, with a thickness of 1 mm each approximately. The
noise recorded when the machines were rotating at 70 r.p.m. without adding
noise harmonics is shown in Figure 11.17. The results when white noise was
added to the direct and the quadrature currents in the test motor are presented
in Figures 11.18 and 11.19. Adding this noise resulted in an increase in the
SPL of around 25 dB, so the contribution of the aerodynamic and mechanical
noise components can be neglected. Compared to the results obtained at no
load, the noise is reduced by between 4 dB and 5 dB. A summary of the re-
sults obtained for various combinations of noise currents at the frequencies of
11.5. Load tests results and analysis 219

L p a v = 3 7 .3 3 d B , 3 4 .0 6 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]
4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.17: SPL at load, 70 r.p.m. Case [d=0, q=0].

L p a v = 6 0 .7 2 d B , 6 0 .3 3 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]

4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.18: SPL at load, 70 r.p.m. Case [d=wn, q=0].

490 Hz and 1105 Hz is presented in Figure 11.20. As in the no load case, the
noise emissions are increased with direct modulation compared to the quadra-
ture modulation, now by an average of 5 dB and 2 dB at 490 Hz and 1105 Hz
respectively. The SPL at 1105 Hz is slightly higher than at 490 Hz, with an in-
crement of 1 dB and 3 dB for the direct and quadrature currents respectively. A
comparison between the no load and load results for all the combinations of the
excitation is included in Figure 11.21. Above each plot ‘(d)’ and ‘(q)’ indicate
that the combinations in the graph are for direct and quadrature modulation re-
spectively. The noise is decreased for direct and quadrature modulation at 490
Hz by an average 6 dB and 4 dB respectively, whereas at 1105 Hz the decrease
is in average 8 dB and 7 dB for direct and quadrature current excitation.
220 Chapter 11. Acoustic measurements

L p a v = 5 6 .9 1 d B , 5 6 .4 5 d B ( A ) 1 /3 o c ta v e b a n d
7 0 7 0
6 0 6 0
5 0 5 0
S P L [d B ]

S P L [d B ]
4 0 4 0
3 0 3 0
2 0 2 0
1 0 1 0
0 0
0 1 k 2 k 3 k 4 k 5 k 6 k 7 k 1 0 0 5 0 0 1 k 2 .5 k 5 k
fre q u e n c y [H z ] fre q u e n c y [H z ]

Figure 11.19: SPL at load, 70 r.p.m. Case [d=0, q=wn].

Figure 11.20: Summary of SPL at load, 70 r.p.m.

11.5. Load tests results and analysis 221

490 Hz (d)
80

70
SPL [dB]

No Load
60 Load

50

40
m/2, 0 m/2,m/2 m,0 m,m/2 m,m 2m,0 2m,m/2 2m,m 2m,2m
combination

490 Hz (q)

80

70
SPL [dB]

No Load
60 Load

50

40
m/2, 0 m/2,m/2 m,0 m,m/2 m,m 2m,0 2m,m/2 2m,m 2m,2m
combination

1105 Hz (d)
80

70
SPL [dB]

No Load
60 Load

50

40
m/2, 0 m/2,m/2 m,0 m,m/2 m,m 2m,0 2m,m/2 2m,m 2m,2m
combination

1105 Hz (q)

80

70
SPL [dB]

No Load
60 Load

50

40
m/2, 0 m/2,m/2 m,0 m,m/2 m,m 2m,0 2m,m/2 2m,m 2m,2m
combination

Figure 11.21: Comparison of SPL at no load and load, 70 r.p.m.

222 Chapter 11. Acoustic measurements

11.6 Sound intensity measurements

Sound pressure measurements provide quick information about the level of

noise at one particular point inside the anechoic room. However, due to the
directionality of the sound radiation in an electrical machine, it would be more
appropriate to quantify the acoustic noise radiation by means of the sound
power. The radiated sound power from the machine was determined by sound
intensity measurements both at no load and load for the cases presented in Fig-
ure 11.8. A wired cubic reference frame was placed around the test motor. This
frame was attached to the base plate, where the motor rested, and to the vertical
plate placed between this machine and the load (Figure 11.3). Therefore only
the sound intensity flowing through four sides of the cube had to be measured,
since the noise was reflected from the other two sides covered by the plates, and
the full results for the no load case are presented in Appendix G, and for the load
case in Appendix H. These appendices present the values of the intensity mea-
sured through each side of the frame, as well as the measured pressure-intensity
index. In the plot of the pressure-intensity index there are two dashed lines (the
top line being LKO and the bottom line being LKO -7), which define the dy-
namic capability LD of the system. With the exception of some low frequencies
in the load measurements, it can be observed that the pressure-intensity index
is contained between the limits of the dynamic capability, which supports the
validity of the sound intensity measurements.
From the measured sound intensity level in each side, the intensity can
be obtained from (F.5), which multiplied by the area of that side leads to the
sound power. Adding the contributions of all the free sides provides the total
power radiated, and the sound power level is obtained using (F.3). The no load
results are shown in Figure 11.22, and the load results in Figure 11.23. The
overall PWL is referred to as LW total above each plot. A comparison between
the total PWL at 490 Hz and 1105 Hz both at load and no load is shown in
Figure 11.24. These results confirm that the direct modulation produces more
noise than the quadrature modulation. At no load the difference is 4 dB at
490 Hz and 7 dB at 1105 Hz. At load this difference is reduced to 1 dB and
3 dB at 490 Hz and 1105 Hz respectively. It should be noted that the results
from single point pressure measurements also indicated that direct modulation
produces more noise than quadrature modulation. However, the differences are
greater for 490 Hz (5 dB at no load, 4 dB at load) than for 1105 Hz (2 dB at
load and no load).
11.6. Sound intensity measurements 223

(a) d=0, q=m, 490 Hz (b) d=0, q=m, 1105 Hz

(c) d=m, q=0, 490 Hz (d) d=m, q=0, 1105 Hz

(e) d=0, q=0

Figure 11.22: Summary of PWL results at no load, 70 r.p.m.

224 Chapter 11. Acoustic measurements

(a) d=0, q=m, 490 Hz (b) d=0, q=m, 1105 Hz

(c) d=m, q=0, 490 Hz (d) d=m, q=0, 1105 Hz

(e) d=0, q=0

Figure 11.23: Summary of PWL results at load, 70 r.p.m.

11.6. Sound intensity measurements 225

Figure 11.24: Comparison of sound intensity measurement results.

These results highlight the importance of quantifying noise from electrical

machines in terms of sound power rather than single point pressure measure-
ments (Wang et al., 2004), because the directionality of the acoustic radiation
for a 490 Hz excitation is different from that at 1105 Hz. Figure 11.25 shows
the mode of vibration at 1105 Hz simulated in A NSYS. The mode at 490 Hz
was unfortunately not available. From the figure, it can be observed that the
vibration is a combination of a radial and a bending mode. On one hand, the
stator expands and contracts radially in anti-phase along the X and Y axis of an
imaginary XY plane parallel to the laminations. On the other hand it can be
observed that the mode shape changes along the length of the motor. This kind
of mode is more easily excited from excitation in the radial direction, which
supports the experimental results.
In general, the PWL at 1105 Hz is higher than at 490 Hz. At no load
the increment for the direct current excitation is 5 dB and for the quadrature
226 Chapter 11. Acoustic measurements

Figure 11.25: Mode of vibration at 1105 Hz predicted in A NSYS. Source (Wang,

1998).

current excitation 2 dB. At load this increment is 0.5 dB for the direct current,
and for the quadrature current the response at 490 Hz is 2 dB higher than at
1105 Hz. Finally, it is confirmed that the PWL decreases when the machine
is loaded. The decrease at 490 Hz when the direct current is modulated is 9
dB, and 6 dB in the case of the quadrature current excitation. At 1105 Hz this
decrease is 13 dB and 9 dB for the direct and quadrature currents respectively.
The reduction in the noise at load must have an electromagnetical origin,
since the mechanical and aerodynamic noise components were found to be neg-
ligible. The loading modifies the amplitude and phase of the current flowing in
the conductors and the slip. This influences in turn the amplitude and the fre-
quency of the exciting airgap force, while the main flux and leakage reactances
are modified as a consequence of saturation. Also, the coupling with the load
affects the resonance frequencies, and all these factors affect the noise radiation
from the machine. Considering that only a part of the changing parameters
tends to increase the noise, the other part may increase or decrease it alike, so
it cannot be stated in general that loading increases or decreases noise (Timar,
1989). Stephen and Davies (1970) stated that there is not much difference
in the sound radiation of small machines at load and no-load, which may be
due to the fact that small machines have generally higher resonance frequen-
11.7. Conclusions 227

cies than those in the exciting force, and they are poor radiators due to their
small size. Brozek (1973) found an important increase in the noise at load in
large machines, which may be explained by the smaller resonance frequencies
of the structure and their better radiating capability. Zhu et al. (1999) showed
that in a small machine (1.5 Kw) the foot-mounting and shaft-coupling can sig-
nificantly affect the natural frequencies and consequently the radiated acoustic
noise. These two factors increase the stiffness of the machine and hence can
reduce the emitted acoustic noise.

11.7 Conclusions
The no load stand still tests show a considerable change in the total sound pres-
sure level at four different positions, indicating that the dynamic eccentricity
is an important factor to consider in the test machine. Therefore, an average
value of the noise emitted at a certain position is obtained by operating the ma-
chine at low speed. This low speed also reduces the influence of the mechanical
and aerodynamic noise in the measurements, which are found to be negligible
compared to the noise produced by the electromagnetic forces.
The sound pressure measurements reveal that the noise emitted increases
with the level of modulation almost linearly. The results from direct and quadra-
ture modulation are closer at 1105 Hz compared to the 490 Hz case, and the
level of noise measured at 1105 Hz is greater than at 490 Hz. This is due to
the fact that these two frequencies excite different modes of vibration. The di-
rectional characteristic of the noise radiation implies that single point sound
pressure measurements are not representative of the overall acoustic behaviour
of the machine. Therefore, a better way of quantifying the acoustic noise from
electrical machines is to measure the sound power. The sound power results
determined from sound intensity measurements show that direct modulation
produces more noise than quadrature modulation, as well as the decrease in the
noise emissions when the machine is loaded.
Although the direct and the quadrature excitations have the same ampli-
tude and frequency, it has been found that the excitation in the radial direction
(direct) generates up to 7 dB higher sound power level than the excitation in
the tangential direction (quadrature). This is probably due to the radial shape
of the modes of vibration, which are much more easily excited by radial than
tangential forces.
In general, it has been observed that the noise emissions are considerably
228 Chapter 11. Acoustic measurements

reduced when the machine is loaded (by up to 13 dB). This should be due to
the change in the electromagnetic component of the noise, since the mechanical
and aerodynamic components were found to be negligible. It is however diffi-
cult to generalize these results since many factors influence the noise emissions
at load compared to the no load case, which in turn may affect in different ways
the noise radiation from different machines.
Chapter 12

Analytical force computation

12.1 Introduction

In this chapter an analytical expression will be deduced for the estimation of

the radial air gap magnetic force in the motor used in the experiments. The
procedure, which is based in the one presented by Bakhuizen (1973), is ex-
tended from a one-phase winding to a three-phase winding distribution, and
the slotting effects are also taken into account. The results will serve to check
the FEM calculations, which will be used for the noise prediction presented in
Chapter 13.
The conductors in the machine are distributed in the slots forming a com-
bined single and double layer winding as shown in Figure 12.1. Each slot con-
tains the side of either one short pitched coil with 88 conductors or two fully
pitched coils with 44 conductors each. Thus, the number of conductors in each
slot is constant and equal to 88, although each one of the two 44-conductor
coils sharing the same slot contributes to the flux in a different pole.

12.2 One phase winding distribution

Figure 12.2 shows the distribution of one phase along a quarter of the stator.
Each one of the three-slot groups has a winding distribution as shown in Fig-
ure 12.1. A Fourier analysis can be developed on the applied MMF in one
quarter of the machine exploiting symmetry. For convenience, a reference mag-
netic coordinate system is defined with its horizontal axis coinciding with the
line of maximum airgap MMF, at an angle ζ with the horizontal axis of the

229
230 Chapter 12. Analytical force computation

N2
N1 N1
N2

Figure 12.1: Winding distribution in the slots.

machine as shown in the figure. The 10 degrees angle between two slots will
be referred to as ξ and the applied MMF in each slot will be considered to be
homogeneously distributed along an angle 2E. The absolute value of the in-
finitesimal applied MMF along one slot will be defined as N25·ia , and the total
contribution of one slot to the airgap MMF will be N · ia , where N is the total
number of conductors in one slot. With this nomenclature the applied MMF,
which will be referred as Fa (α), can be expressed as follows:

N · ia
− −ζ − ξ − E < α < −ζ − ξ + E (12.1)
2E
N · ia
− −ζ − E < α < −ζ + E (12.2)
2E
N · ia
− −ζ + ξ − E < α < −ζ + ξ + E (12.3)
2E
N · ia
ζ −ξ−E<α<ζ −ξ+E (12.4)
2E
N · ia
ζ −E<α<ζ +E (12.5)
2E
N · ia
ζ +ξ−E<α<ζ +ξ+E (12.6)
2E
0 elsewhere (12.7)
12.2. One phase winding distribution 231

2E
dα

ζ
α
F
F
F ξ

Figure 12.2: One phase winding distribution along one quarter of the machine

In order to examine the harmonic components in the airgap MMF, it is

useful to develop the Fourier analysis of equations (12.1) to (12.7), defining Fa
as in (12.8),
ν=∞
Fa (α) = Cν · e−jpνα (12.8)
ν=−∞

where p is the number of pole pairs and ν is the number of the harmonic. From
this equation, and for p=2, Cν is defined as in (12.9).
π
2
1
Cν = Fa (α) · ej2να dα (12.9)
π
− π2

Combining the MMF in (12.1)-(12.7) and (12.9), and after some mathe-
matical transformations, Cν can be expressed as in (12.10).
N · ia sin(2νE)
Cν = 2j [sin(2ν(ζ + ξ)) + sin(2ν(ζ − ξ)) +
π 2νE
+ sin(2νζ)] (12.10)
A so called slot factor fg,2ν is introduced, which is defined as in (12.11).
sin(2νE)
fg,2ν = (12.11)
2νE
232 Chapter 12. Analytical force computation

This factor can be approximated to ‘1’, considering the small slot opening
and a reasonable number of ν harmonics. A closer look into (12.10) shows that
the number of slots is multiplied by a factor of three if a concentrated winding
is considered, i.e. if ξ is zero. This is reasonable since the winding is distributed
in three slots with N conductors each. If the winding was concentrated in one
slot instead, the number of conductors would be three times bigger.
The relation between the applied and airgap MMF can be deduced apply-
ing Ampere’s law to an infinitesimal sector of the machine enclosing a certain
applied MMF, as shown in Figure 12.3.

C B
Fa

H(α + dα) H(α)

D A
dα
α

Figure 12.3: Infinitesimal sector of the machine.

The sign of the MMF has been selected to be positive in the outward direc-
tion. Considering that the applied MMF is homogeneously distributed within
the sector, Ampere’s law can be expressed as in (12.12).

Hdl = Fa (α)dα (12.12)

Evaluating the line integral along the path ABCD, and neglecting the tan-
gential components, leads to (12.13) for the radial magnetic field strength H
across the airgap g.

[H(α) − H(α + dα)] · g = Fa (α)dα (12.13)

The second term on the left hand side of this equation can be rewritten
12.2. One phase winding distribution 233

using the first order Taylor expansion in (12.14),

f  (a)
f (x) = f (a) + (x − a) + ... (12.14)
1!
where f = H, a = α, and x = α + dα in this case. With these modifications
(12.15) is obtained.
∂H(α)
H(α + dα) = H(α) + dα (12.15)
∂α
Combining this equation with (12.13) leads to (12.16),

1
H(α) = − Fa (α)dα (12.16)
g
where H(α) · g corresponds to the airgap MMF, which will be referred to as Fδ .
Hence, the relation between the applied and the airgap MMF is (12.17).

Fδ = − Fa (α)dα (12.17)

Introducing (12.8) into (12.17) yields (12.18) for a four pole machine.
ν=∞ 
Fδ = − Cν · e−j2να dα (12.18)
ν=−∞

Developing the integral gives (12.19).

ν=∞
1
Fδ = Cν · e−j2να (12.19)
ν=−∞
2jν

Introducing the expression for Cν from (12.10) into (12.19) yields (12.20).
ν=∞
N · ia
Fδ = fg,2v · e−j2να [sin(2ν(ζ + ξ)) + sin(2ν(ζ − ξ)) +
ν=−∞
πν
+ sin(2νζ)](12.20)

This expression is valid for one phase, with ia defined as in (12.21).
ia = î · ejωt (12.21)
234 Chapter 12. Analytical force computation

This represents a current vector rotating with an angular frequency ω. For

the case of a concentrated winding, i.e. ξ = 0, the sum of the three sine terms in
(12.20) is reduced to 3 · sin(2νζ). The value of sin(2νζ) is equal to zero for all
even harmonics. For a given pair of odd harmonics ±ν, the sum of sin(2νζ)
will be equal to ±2. Thus, the amplitude of the airgap MMF for the different
harmonics is given by (12.22).

2 · N î 3
(12.22)
π |ν|

In this equation {|ν| = 1, 3, 5, 7, ...}. The direction of rotation will al-

ternate between two adjacent odd harmonics. Dividing (12.20) by the airgap
length g and taking the real part yields the magnetic field strength H. Assuming
infinite permeability in the iron, the airgap flux density Bδ is directly obtained
multiplying H by µ0 . Figure 12.4 shows Bδ over one half of the motor when
a current with an amplitude of 1.82 A (which corresponds to the no load phase
current in the real motor) and a frequency of 50 Hz is used.

0 .5
B [T ]

− 0 .5

− 1
0 .0 2
0 .0 1 5 3
2 .5
0 .0 1 2
1 .5
0 .0 0 5 1
0 .5
tim e [s e c ] 0 0
a n g le [ra d ]

Figure 12.4: Airgap B for the one phase concentrated winding.

In the case of the distributed winding, it can be shown that the sine terms
inside the brackets in (12.20) still add zero for all even harmonics, leading to
12.2. One phase winding distribution 235

the expression for the amplitude of the airgap MMF in (12.23).

2 · N î sin(2ν(ζ + ξ)) + sin(2ν(ζ − ξ)) + sin(2νζ)

(12.23)
π |ν|

For the odd harmonics, the sum of the sine terms decreases and increases
periodically for {|ν| = 1, 3, 5, 7, 9, 11, ...}, but the contribution of higher
order harmonics is little since |ν| in the denominator increases a well. This is
also true for the concentrated winding, but there the factor introduced from
the sine term is constant and equal to 3, whereas in the distributed winding it
decreases for the first 9 odd harmonics. Hence, the influence of these harmonics
in the distributed winding is less compared to the concentrated winding, as
can be appreciated in Figure 12.5. The amplitude of the fundamental airgap
MMF decreases when the winding is distributed, although the difference is
rather small. For the winding distribution in the test motor, the sine terms in
equation 12.23 add 2.968 for |ν| = 1, which represents a negligible decrease of
1.07 % with respect to the concentrated winding.

0 .5
B [T ]

− 0 .5

− 1
0 .0 2
0 .0 1 5 3
2 .5
0 .0 1 2
1 .5
0 .0 0 5 1
0 .5
tim e [s e c ] 0 0
a n g le [ra d ]

Figure 12.5: Airgap B for the one phase distributed winding.

236 Chapter 12. Analytical force computation

12.3 Three phase winding distribution

The airgap MMF for one phase can easily be extended for the case of a three
phase machine where the current vector has the form in (12.24).
 
ia,b,c = î 1 + ejν 2π
3 · e
−j 2π
3 + e
jν 4π
3 · e
−j 4π
3 ejωt (12.24)

In one hand, this expression takes into account the electrical phase differ-
ence between the currents from the three phases. This phase difference will be
120 degrees for the fundamental, and a multiple of this number for the corre-
sponding harmonics. On the other hand it considers also the fact that the coils
are shifted in space around the stator of the motor by an angle of 120 degrees.
Introducing this three phase current into (12.20) and taking the real part, the
total airgap MMF in the machine can be written as in (12.25).
ν=∞
N î
Fδ = fg,2ν (sin(2ν(ζ + ξ)) + sin(2ν(ζ − ξ)) + sin(2νζ)) ·
ν=−∞
πν
2π 4π
·(cos(ωt − 2να + (ν − 1)) + cos(ωt − 2να + (ν − 1)) +
3 3
+ cos(ωt − 2να)) (12.25)

The effect of introducing three phases and distributing the winding around
the motor results in a further decrease of the winding harmonics, as shown in
Figure 12.6.

12.4 Modelling the slots

So far, a constant airgap g around the motor has been assumed, which should be
modified by the Carter factor to take into consideration the higher reluctance
in the slot openings. Equation (12.26) has been proposed by Lipo (1996) for
the calculation of the Carter factor, based on the slot dimensions.
τs
Kc = b20
(12.26)
τs − (5g+b0 )

In this equation τs is the slot pitch, and b0 corresponds to the slot opening.
In the test motor, the 10 degrees angle ξ between two slots, at an inner stator
12.4. Modelling the slots 237

1
B [T ]

− 1

− 2
0 .0 2
0 .0 1 5 3
2 .5
0 .0 1 2
1 .5
0 .0 0 5 1
0 .5
tim e [s e c ] 0 0
a n g le [ra d ]

Figure 12.6: Airgap B for the three phase distributed winding.

radius of 47.66 mm gives a length of 8.32 mm. For a slot opening of 2.54 mm
and an airgap equal to 0.3 mm the Carter factor becomes 1.2375.
However, it is possible to calculate the slotting harmonics by considering
the inverse airgap function between two slots, as shown in Figure 12.7. The
value of he corresponds to the actual length of the path that the flux follows
through the air along one slot opening before penetrating one of the adjacent
stator teeth. This value has been estimated based on the work presented by
Gerling (1994). A ‘magnetically effective slot depth’ he is defined to be equal
to the geometrical slot depth h if the slot opening is greater or equal to double
the geometrical slot depth. Otherwise he = π·b 0
8 , as it is the case in the test
motor. For a geometrical slot depth of 20.085 mm, he becomes 0.997. Based
on Figure 12.7, the inverse airgap function is expressed as:

ĝ − 2ξ < α < −E (12.27)

ǧ −E < α < +E (12.28)
ĝ +E < α < + 2ξ (12.29)

Following the same procedure as for the applied MMF, a Fourier expansion
238 Chapter 12. Analytical force computation

2E

stator

1
ĝ = g
1
ǧ = he

rotor

− 2ξ 0 ξ
2

Figure 12.7: Inverse airgap function.

of equations (12.27)-(12.29) leads to (12.30),

n=∞
q(α) = Cn e−jnNs α (12.30)
n=−∞

where n is the number of the slot harmonic and Ns is the number of stator
slots. From this expression Cn is defined as in (12.31).
π
Ns
Ns
Cn = g(α)ejnNs α (12.31)
2π
− Nπ
s

Developing the integral term a final expression for Cn is given by (12.32),

where the new slotting factor has the form in (12.33), and can be approximated
to ‘1’.
Ns · E sin(nENs ) sin(nπ)
Cn = · (ǧ − ĝ) + ĝ (12.32)
π nENs nπ

sin(nENs )
fg,n = (12.33)
nENs
12.4. Modelling the slots 239

Introducing (12.32) into (12.30) and taking the real part yields (12.34).
n=∞
Ns · E sin(nπ)
q(α) = fg,n (ǧ − ĝ) + ĝ cos(nNs α) (12.34)
n=−∞
π nπ

Finally, the radial magnetic field strength including the slotting harmonics
is obtained multiplying (12.34) by (12.25), yielding (12.35).
ν=∞ n=∞
Ns · E sin(nπ)
H(α) = fg,n (ǧ − ĝ) + ĝ cos(nNs α) ·
ν=−∞ n=−∞
π nπ
N · î
· fg,2ν (sin(2ν(ζ + ξ)) + sin(2ν(ζ − ξ)) + sin(2νζ)) ·
πν
2π 4π
·(cos(ωt − 2να + (ν − 1)) + cos(ωt − 2να + (ν − 1)) +
3 3
+ cos(ωt − 2να)) (12.35)

The airgap flux density obtained when the slotting effects are included is
shown in Figure 12.8.

1
B [T ]

− 1

− 2
0 .0 2
0 .0 1 5 3
2 .5
0 .0 1 2
1 .5
0 .0 0 5 1
0 .5
tim e [s e c ] 0 0
a n g le [ra d ]

Figure 12.8: Airgap B for the three phase distributed winding with slots.
240 Chapter 12. Analytical force computation

12.5 FEM validation

It is instructive to use the procedure presented in this chapter to compare the

analytical solution to the numerical using the finite element method (FEM).
The software used for the calculations is O PERA 2D, version 7.5. Symmetry
could be exploited, and Figure 12.9 shows one quarter of the geometry, indicat-
ing the rotor, stator and surrounding air. The option of defining air around the
stator was preferred rather than imposing the Dirichlet condition in the outer
surface of the stator (vector potential equal to zero). Although the number of
elements in the model increases, the fact that leakage flux can appear in the
surface of the motor under very heavy saturation was taken into consideration.
The thickness of the layer of air surrounding the stator was made equal to the
radius of the machine.

Figure 12.9: Plot of the geometry modelled in O PERA 2D.

Since this model will be also used for transient calculations, where the rotor
mesh is rotated with respect to the stator, a new boundary condition needs to
be imposed between the stationary and the rotating parts. This was achieved
by defining a ‘rotating airgap’. In order to obtain a very fine calculation of the
airgap forces, five layers of elements were modelled in the airgap of the machine.
12.5. FEM validation 241

Two of them were attached to the rotor mesh, other two to the stator mesh and
the rotating airgap was set in-between. In this way, the number of elements
in both sides of the boundary condition was optimized, and it was also ensured
that the calculation of the forces could be done through the middle of the airgap
without crossing any node, since the field there is multivalued. Defining the
rotating airgap also had the advantage that the rest of the boundary conditions
were set automatically so as to match both the magnetical and mesh properties
in any of the solvers available. Figure 12.10 shows the element distribution in
the area surrounding the airgap. The mesh was automatically generated using
triangular elements, although it could be adapted defining the facets in a proper
way and changing the number of subdivisions at the edges.

OPERA-2d
Pre and Post-Processor 7.506

Figure 12.10: Mesh generated in the surroundings of the airgap.

The model was checked using a harmonic analysis at no load and full load.
The number of elements, especially in the airgap, the rotor slots and the sur-
rounding regions, was increased until the change in the torque delivered was less
than 1%. The total number of elements used in the model corresponding to one
quarter of the machine is 8314, comprising 1254 elements distributed along
three layers in the airgap, and 86 elements per rotor slot. The full load value
obtained was 10% higher than the rated torque specified in the data sheets. The
difference could be attributed to an inaccurate BH curve, and/or the fact that
the inductance and the resistance in the aluminium end-rings were neglected in
the model.
242 Chapter 12. Analytical force computation

The flux density distribution through the middle of the airgap was calcu-
lated performing a magnetostatic solution at no load. Since the analytical solu-
tion assumes infinite permeability for the iron, a first calculation was carried out
defining a very high linear isotropic permeability for the iron. The results from
both FEM and the analytical calculations are shown in Figure 12.11, where a
similar behaviour can be observed. It should be remarked that the phases of the
curves in this plot have been displaced in order to improve their visual inspec-
tion.

1 .5

0 .5
B [T ]

− 0 .5

− 1

− 1 .5
F E M
T h e o ry
− 2
0 1 2 3 4 5 6
a n g le [ra d ]

Figure 12.11: Comparison between analytical and FEM flux densities for ideal
iron, phase displaced.

When the non-linearities in the iron are considered in the FEM calculation,
the airgap flux density drops to a more reasonable level around 1 T and its shape
resembles a square wave, as shown in Figure 12.12. Besides, more irregularities
appear due to the saturation effects which are more predominant in the tip of
the teeth.
The radial stress in the laminated stator stack can be implemented applying
Maxwell’s stress equation for radial forces, given by (12.36).

Bδ2 (α)
σr (α) = (12.36)
2µ0
12.6. Conclusions 243

1 .5

0 .5
B [T ]

− 0 .5

− 1

− 1 .5
0 1 2 3 4 5 6
a n g le [ra d ]

Figure 12.12: FEM solution considering non-linear iron.

The result from the analytical calculation is shown in Figure 12.13, where the
slotting effects have not been included in order to make the picture more clear.
The stress in the airgap of the machine was also calculated in FEM and the
results when using non-linear iron are presented in Figure 12.14.

12.6 Conclusions
An analytical expression has been deduced for the radial airgap forces in the
test motor. The method is based in the Fourier analysis of the applied MMF.
The expressions have been gradually developed from a one phase concentrated
winding to the three phase distributed winding setup in the machine. The
slotting effects have been incorporated in the model, developing the Fourier
analysis of an inverse airgap function. Although the method assumes ideal iron,
it is useful as a check of the results obtained from an ideal finite element model
of the machine, and it could be appreciated that the results using both methods
showed a good agreement.
244 Chapter 12. Analytical force computation

5
x 1 0
1 2

1 0

8
]
2
s tre s s [N /m

0
0 .0 1
1 .5
0 .0 0 5 1
0 .5
tim e [s e c ] 0 0
a n g le [ra d ]

Figure 12.13: Analytical radial Maxwell stress, ideal iron.

5
x 1 0
5

4
]

3
2
S tre s s [N /m

0
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0
A n g le [d e g ]

Figure 12.14: FEM Maxwell stress, non-linear iron.

Chapter 13

Acoustic noise prediction

13.1 Introduction

The analytical model presented in Chapter 12 introduced an understanding on

how the linear airgap force distribution can be calculated based on the winding
distribution in the machine. However, in order to perform an accurate calcula-
tion of the electromagnetic airgap forces, both the iron non linearities and the
details of the motor geometry must be taken into consideration, and this can be
achieved using numerical methods. The mechanical properties of the machine
structure also influence the noise emissions, and numerical packages are also
available for mechanical and acoustic calculations. The aim of this chapter is to
develop a process for the prediction of the sound power radiation in an electrical
machine, based on numerical calculations. This would allow implementation
of appropriate noise control measures at the design stage.
The noise radiated from electric machinery may be predicted analytically or
numerically. In the analytical approach, the motor acoustic behaviour is usually
approximated using methods based on spherical or finite-cylindrical acoustic
models (Zhu and Howe, 1994). However, analytical methods are insufficient
when dealing with complex topologies or taking into account the end-plates.
For variable speed operations, it is difficult to calculate in practice the magni-
tude of the noise peaks, although the frequencies at which they occur can be
deduced from the interaction of the harmonics present in the inverter output
(Yang, 1981). An algorithm to calculate the lowest limit of noise radiated from
a variable speed drive was presented by Timar and Lai (1994), and Timar and
Lai (1996).

245
246 Chapter 13. Acoustic noise prediction

In the numerical approach, the deterministic method is traditionally used

for the noise prediction in induction motors, although statistical approaches
appear to be promising, especially for high frequencies and large motors (Wang
et al., 2002). The deterministic method has been used by Wang et al. (2004)
for the prediction of the sound power from a PWM inverter driven induction
motor. However, the influence of the stator slot harmonics on the rotor could
not be considered since the finite element package used did not allow currents
to be induced in the rotor, which instead had to be estimated (Astfalck et al.,
1997). The software used for the electromagnetic calculations presented in this
thesis allows currents to be induced in the rotor, and the noise prediction process
has been tested using vector control. In the process, the sensitivity of the noise
prediction to uncertainty factors such as the damping of the structure and the
accuracy of the BH curve has also been assessed.
In the deterministic method, the electromagnetic forces acting on the mo-
tor structure are calculated from the input currents (or voltages) using an elec-
tromagnetic model based on FEM. The commercial package O PERA 2D/3D
version 7.5 was used for this purpose. The vibration characteristics are usu-
ally simulated using a structural model also normally based on FEM, and the
commercial package used was A NSYS version 5.1. A detailed description of the
structural simulations with the motor used in this part of the thesis was pre-
sented by Wang (1998), Wang and Lai (1999), and a summary will be included
in this chapter. By using the vibration velocities on the motor structure pre-
dicted from the structural model, the radiated sound power level can then be
calculated using an acoustic model. This model was implemented in another
commercial package, S YSNOISE version 5.2, based on the boundary element
method (BEM). The main advantage of using BEM is that only the boundary
of the problem domain needs to be discretized, i.e. there is no interior or exte-
rior mesh. Although the mathematics involved in BEM are more difficult, the
modelling effort is lower compared to FEM and it also presents the advantage of
solving both interior (inside the defined model) and exterior (in the surrounding
fluid) problems, with different possibilities of fluid-structure interaction (Zhou,
1993). An overview of the prediction process is shown in Figure 13.1. The elec-
tromagnetic excitation force was imported into S YSNOISE from O PERA, and
the modal information as well as the geometry were taken from A NSYS.
The following sections describe the details of the different finite/boundary
element models, and the calculations performed within the structural, electro-
magnetic and acoustic analyses involved in the process shown in Figure 13.1.
13.2. Structural modal analysis 247

Electromagnetic Force ✲ Acoustic BEM ✲

(S YSNOISE)
Noise
FEM (O PERA)
✻
Geometry
Mode shapes
Structural
FEM (A NSYS)

Figure 13.1: Numerical prediction process.

13.2 Structural modal analysis

The purpose of the structural modal analysis is to determine the modes of vi-
bration, which define the vibro-acoustic characteristics of the structure. This
is achieved through the so-called free response synthesis, which can be performed
experimentally or by means of numerical simulations. These two procedures
will be described in this section. They were used to determine the character-
istics of the test machine by Wang and Lai (1999), and the main results and
conclusions from this work will be summarized.

Experimental free response synthesis

Resonance occurs when the dynamic forces in a process excite the natural fre-
quencies in the structure, and a suitable way to identify them is to analyze the
frequency response function (FRF) of the system. The FRF is expressed as the
ratio of the output vibration (X) to the input force (F) spectra over a defined
frequency range (ω).
X(ω)
H(ω) = (13.1)
F (ω)
Vibration can be measured as displacement, velocity or acceleration of the
particles in the structure and the FRF will be known as receptance, mobility and
inertance respectively. A resonance is represented in the FRF plot of the sys-
tem as a peak. Since this excitation force is usually difficult to measure in most
248 Chapter 13. Acoustic noise prediction

industrial processes, the identification of the natural frequencies of a structure

is better done in the laboratory, where the applied force can be selected con-
veniently. Modal analysis is a common tool used in the laboratory not only
to identify the natural frequencies but also the way the structure vibrates at
those frequencies (mode shapes) as well as the rate at which the vibrations decay
(modal damping). These two components, together with the modal frequency,
constitute the modal parameters of a structure and they fully characterize the
vibration of single degree of freedom (SDOF) models. In fact, each resonance
peak can be considered as the response of a SDOF system. Since a FRF con-
sists of a collection of resonance peaks, it is important to understand how these
systems can be modelled in the physical, time and frequency domains.

SDOF models
A SDOF model can be described as a combination of a point mass m, a massless
linear spring with stiffness k, and a linear viscous damper c, and the constraint
that it can move in only one direction (x) is assumed. The mathematical model
in the time domain when a force f is applied is easily obtained from Newton’s
second law:

mẍ(t) + cẋ(t) + kx(t) = f (t) (13.2)

An analytical model can also be developed in the frequency domain, where

the behaviour of each component is expressed mathematically as a function of
the frequency of the system: k1 for the spring, jωc
1
for the damper and −ω12 m
for the mass (Ewins, 1984). At low frequencies the response of the system is
dominated by the spring and it is in phase with the excitation. When the fre-
quency increases, the inertial force of the mass becomes more important, until
a point is reached where the mass and the spring terms cancel each other. At
this frequency, called undamped natural frequency, the response is controlled
only by the damping term and it lags the excitation by 90 degrees. If the fre-
quency increases even further, the mass term takes control over the system and
the response lags the excitation by 180 degrees.
The values of the mass, stiffness and damping used in the analytical model
are usually unknown in the real structure. However there is an experimental
way of modelling the vibration of a SDOF system, which indirectly leads to
these parameters (Brüel & Kjær, 1988). The properties of the system are now
defined in terms of the pole location p and the residue R, which are quantities
13.2. Structural modal analysis 249

that can be measured from the FRF plot of the system. The pole location
consists of a real part equal to half the -3 dB bandwidth of the FRF peak and an
imaginary part which$ is $directly the modal frequency. The peak amplitude of
the mode is equal to $ R $
σ , and this yields the value of the residue since the real
part of the pole is the value of the damping σ. The modal frequency is defined
as the damped natural frequency ωd , and (13.3) shows how it is related to the
undamped natural frequency ω0 , which in turn is expressed as in (13.4). The
damping ratio ζ is also related to these frequencies through (13.5), and the mass
is linked to the value of the residue as in (13.6). Knowing the mass, damping
and damping ratio, the stiffness is easily obtained from (13.7).

ω02 = ωd2 + σ 2 (13.3)


ω0 = k/m (13.4)

ωd = ω0 1 − ζ (13.5)
1
R = −j (13.6)
2mωd
σ
ζ = √ (13.7)
2 k·m

Two of the modal parameters are contained in the pole location itself. The
third one, the mode shape, is related to the residue. When a structure is ex-
cited with a force, each SDOF experiences a relative modal displacement. The
residue of a particular mode is proportional to the product of the modal dis-
placement at the response and excitation SDOF. The combination of the rela-
tive displacements of all the SDOF of the structure gives the mode shape for a
given frequency.
In practice, the FRF is measured using a dual-channel analyzer, which in
turn is connected to a computer where the data is stored for further post-
processing. For each mode, the pole location and the residue need to be es-
timated from this data, but for each FRF measurement the analyzer provides
around 800 complex values. With this enormous amount of data the use of
a computer is essential, where advanced mathematical algorithms for the esti-
mation are implemented. From the estimated parameters a new analytical FRF
curve is obtained, which is superimposed on the measured data so that the op-
erator can evaluate how they fit. This process is known as ‘curve fitting’.
250 Chapter 13. Acoustic noise prediction

Curve fitting
The aim of curve fitting is to extract reliable modal data from the measurements.
There are different types of curve fitting techniques that are used depending on
the application, and they contain algorithms to reduce the effects of random
noise in the measurements and the bias errors, such as any leakage or phase er-
rors, which would cause erroneous parameter estimations. SDOF curve fitters
can be used with lightly coupled modes, i.e. when the interaction between dif-
ferent modes is not important. Under these conditions a SDOF behaviour can
be assumed around the modal frequencies. The user specifies a band around
each mode containing the relevant data that will be used in the curve fitting
process. The selection of the band is often a compromise between including
as much data as possible and avoid moving far away from the resonance to
minimize the effects of nearby modes, as otherwise the assumption of SDOF
behaviour would be invalid. When the modes are heavily coupled, multiple
degree of freedom (MDOF) curve fitters are used instead. In this case, the op-
erator specifies a region over which the curve fitter will seek the parameters and
the number of modes contained in that region. The results using this technique
depend to a large extent on the user’s skill and experience in specifying the cor-
rect number of modes for the model. An example of a plot of the FRF function
is presented in Figure 13.2, where the cursors define the band for one of the
modes.

Figure 13.2: Frequency Response Function plot.

13.2. Structural modal analysis 251

After the curve fitting process, each group of the resulting modal parameters
is associated with the corresponding mode number. The different modes of
vibration contain important information about the inherent dynamic properties
of the free structure, i.e. when no forces are acting. The modal analysis of the
motor used for the experiments had already been performed by Wang and Lai
(1999). It is necessary to describe this process though, since it will provide
useful information about how the modal properties that will be used in future
calculations have been determined.
The free vibration of the machine structure was achieved using two kinds
of excitation: random excitation, provided from a shaker, and impact excita-
tion, using a hammer with a steel tip but without extra mass. The frequency
band for the analysis spanned from 0 to 3.2 kHz. The motor was tested resting
on the test bench, through four isolators, and also without isolators, suspended
together with the support and the base plate by soft rubber bands. In order to
examine the contribution to the overall vibration of the different parts of the
motor, the response was measured with five different motor configurations: the
whole motor structure; without the rotor; without isolators; without rotor and
isolators; and finally, without rotor, isolators and end-shields. Only acceleration
in the radial direction was measured and the modes of vibration were deter-
mined with reasonable accuracy, except perhaps those caused by the isolators.
In order to verify the modal results, the sound power spectra obtained from
the machine when driven by three inverters under no load was also determined.
It was observed that almost corresponding to each peak in the sound power
spectra, a vibration mode was identified when the whole motor or the motor
without the rotor was tested, indicating that the modal testing results obtained
were reliable. The end-shields had also a great influence on the vibration of
the motor, since they impose boundary conditions on the motor casing. Be-
sides, the modes determined without the rotor were similar to those with the
rotor, implying that experimental modal testing or numerical modelling may be
conducted without the rotor if low-frequency modes are not of interest.

FEM free response synthesis

Any structural model developed using the finite element technique has to be val-
idated using experimental modal tests. While FEM will provide all the modes of
vibration, the less significant are likely to be missing in the experimental results.
Nevertheless, all the measured modes should correspond to modes calculated
252 Chapter 13. Acoustic noise prediction

with FEM. Wang and Lai (1999) used the experimental results obtained with
the motor to help develop a structural model using the modal analysis solver
in A NSYS. In order to investigate the effects of the various substructures on
the overall vibration behaviour, six motor models were developed taking into
consideration the teeth of the stator, casing, slots on the casing, support and
end-shields, as shown in Figure 13.3. It is important to describe these models
and the results obtained, in order to justify the selection of one of them for the
acoustic calculations presented in this thesis, over the other models.
Each steel lamination in the stator has a density of 7800 kg/m3 , and the
Young’s modulus was determined to be 13.6 x 1010 N/m2 . The total mass of
the stator and the windings is 6.6 and 2.88 kg respectively. The material of the
stator was first assumed to be isotropic. The element types were Brick eight-
node Solid45 and Tetrahedron Solid72 and the boundary conditions of the
stator were set to be free at both ends. The effect of the winding was examined
first. The windings were taken into account by treating them as an additional
mass uniformly distributed in the stator, and this was done by changing the
density of the material to 11193 kg/m3 in model A. It was observed that the
windings affect the natural frequencies of the stator very much. Hence, the mass
of the windings must be taken into account in developing a structural model
with reasonable accuracy.
The effect of the teeth was investigated next. For this purpose, the stator
without teeth was considered, model B, which corresponds to the non-teeth re-
gion of model A. Both cases with and without windings were also investigated,
again by changing the density of the iron material. It was observed that without
windings the teeth had a strong influence on the calculated natural frequencies,
with variations up to 18%. Also, the natural frequencies with teeth were lower
than without teeth, which suggests that the mass of the teeth influences more
the results, rather than their stiffness. When the winding was considered, the
differences in frequency variations compared to model A were less than 10% in
general. The reason for this behaviour is that the total mass of the stator and
windings (9.48 kg) is much greater than that of the teeth (2.5 kg) and conse-
quently, the vibration behaviour is dominated by the cylindrical shell, which in
this case turns out to be a good approximation of the stator structure when the
windings are considered. However, it was indicated that such an approximation
might not be valid for other stators, where the different number of teeth and
dimensions could lead to a different vibration behaviour of the stator structure.
13.2. Structural modal analysis 253

Figure 13.3: Finite element structural models: (a), the actual stator model; (b),
a cylindrical shell stator model; (c), a two cylindrical shell model of
the motor; (d), a two cylindrical shell model of the motor with slots
on the outer casing; (e), a two cylindrical shell model of the motor
with slots, the support and the base plate; (f ), a two cylindrical shell
model of the motor with slots, the support, the base plate and end-
shields. Source (Wang and Lai, 1999).
254 Chapter 13. Acoustic noise prediction

The effect of the casing coupled to the stator and the end-shields was ana-
lyzed as well. Previous studies cited in the paper concluded that its effect could
be neglected. However, the different physical properties of the material used, as
well as the different ratio between the length of the casing and the stator in the
test motor, suggested that the effect of the casing might not be negligible in this
case. A two cylindrical shell model, which included the mass of the windings in
the stator, was developed as model C. The material was first defined as isotropic,
but a closer study of the results suggested that the material in the stator should
be orthotropic instead. The laminations are compressed in the axial direction
and, although the density of the material in the stator remains approximately
the same, the stiffness of the stator in the axial direction would not be equal
to the one of the stator made in solid material. As it is difficult to determine
the pressure between the laminations, a typical value of Young’s modulus in the
axial direction was chosen, corresponding to 2% of the Young’s modulus in the
circumferential direction. The results obtained with this modification showed
improved agreement with the experimental results. Hence it is essential to con-
sider the material of the stator to be orthotropic. From the results, it could be
observed that the vibration modes of the stator and the casing were influenced
by the coupling between them. The modes of vibration of the casing were dif-
ferent from those without the coupling, as it would be expected since now the
part of the casing attached to the stator is constrained while the rest of the cas-
ing is not. The effect of the coupling in the stator was to increase the natural
frequencies in the low frequency range, corresponding also to an increase of the
stiffness of the stator. At high frequencies the outer casing acts as a mass load to
the stator, thus decreasing the natural frequencies. It was concluded therefore
that the outer frame of the stator can play an important role in the vibration of
small and medium-sized motors, where the mass of the frame is not negligible
compared to that of the stator, and this effect would be further emphasized if
the dimensions of the two structures are also quite different.
In order to further improve the calculated frequencies, the ventilation slots
in the motor casing were included in model D. The reduction in the mass
of the casing, and thus the load on the stator, increased the frequencies of the
stator modes marginally, while the change in the stiffness of the casing decreased
the frequency of its modes substantially, showing a better agreement with the
experimental results.
The motor structure could be more accurate if the support and base plate
are included, as shown in model E. The inclusion of these parts reduces the
13.3. Electromagnetic analysis 255

frequencies of the casing due to the increase of its effective mass. The effect of
the stator teeth was again examined with this new model, and the results were
within 5% compared to the model without teeth. Hence, the teeth of the stator
can be neglected in the vibration analysis.
Finally, the end-shields made of aluminium by casting are incorporated in
model F. Since their real geometry is rather complicated and the interest is not
in their vibration itself but in their influence on the vibration of the casing, an
equivalent end-shield with the form of a circular flat plate of the same mass,
0.36 kg, and a thickness of 6 mm was considered. The end-shields are coupled
to the casing by coupling the corresponding nodes in all degrees of freedom.
From the analysis of the results, most of the vibration modes could be predicted,
although the results show more discrepancy than the previous model. This is
due to the complicated coupling between the end-shields and the casing, not
all the points along the edges of the two structures are coupled and those who
are, present different coupling strengths from each other. These details are too
complex to consider in the model and nevertheless, the natural frequencies of all
the modes identified in the calculations were within 10% of the experimental
results, concluding that the model may be acceptable for acoustic prediction.
Further studies showed that the details of the motor structure such as slots,
support, base plate and end-shields affect the vibration behaviour of the motor
very much, so they need to be modelled.
The geometry and modal results obtained from model F are the ones that
were used in S YSNOISE for the noise calculations in this chapter.

13.3 Electromagnetic analysis

The purpose of this section is to explain how the electromagnetic models were
built and how the airgap forces were calculated and applied to the S YSNOISE
model. The cases studied are those corresponding to [d=m, q=0], [d=0, q=m]
and [d=0, q=0] in the experiments. Only the harmonic frequency of 490 Hz was
considered, and the machine rotated at 70 r.p.m, at no load. The different cases
were simulated using the drive system presented in Chapter 10 (Figure 10.12).
An example of the three phase currents for direct and quadrature modulation is
presented in Figure 13.4. These currents were applied to the coils in the FEM
electromagnetic model, where the airgap stress was calculated.
256 Chapter 13. Acoustic noise prediction

2 2

1 .5 1 .5

1 1
P h a s e c u rre n t [A ]

P h a s e c u rre n t [A ]
0 .5 0 .5

0 0

− 0 .5 − 0 .5

− 1 − 1

− 1 .5 − 1 .5

− 2 − 2
0 0 .0 5 0 .1 0 .1 5 0 .2 0 0 .0 5 0 .1 0 .1 5 0 .2
T im e [ s e c .] T im e [ s e c .]

(a) Direct modulation currents (b) Quadrature modulation currents

Figure 13.4: Simulated phase currents for direct and quadrature modulation.

The 3D model

The aim was to simulate the motor with a 3D model in order to take into
consideration end-winding effects, and also the influence of skewing in the ro-
tor slots, which contributes to decrease the torque ripple and thus the noise
emissions. This model is shown in Figure 13.5, where the modulus of the flux
density is plotted at no load. The aluminium in the rotor has been removed
from the picture for the sake of clarity. Part of the conductors are visible on the
surface of the stator along the extrusion direction, but this is only due to a dis-
play problem in the version of O PERA used for the calculations. The number
of nodes and elements was 44020 and 32352 respectively.
The solution presented in Figure 13.5 was obtained from a harmonic analy-
sis. In this case the model could only be solved for single a.c. excitation frequen-
cies at a time, and for a given slip. The currents from the S IMULINK model
could be uploaded in a transient solver, where the rotor was rotated with a user-
defined speed. However, the FE simulations could not be performed, since they
failed due to convergence problems in the solver. Besides, although symmetry
could be exploited in the extrusion direction, it was not possible to apply it
13.3. Electromagnetic analysis 257

Y100.0

Y80.0

Y60.0

Y40.0 Z-120.0
Z-100.0
Z-80.0
Y20.0 Z-60.0
X-120.0X-100.0 Z-40.0
X-80.0 X-60.0 Z-20.0
X-40.0 X-20.0
X20.0 X40.0
Z20.0 X60.0 X80.0
Z40.0 X100.0 X120.0
Z60.0
Z80.0
Z100.0 Y-20.0
Z120.0

Y-40.0

Y-60.0

Y-80.0

Component: BMOD
0.0486666 0.90006 1.751454 Y-100.0
OPERA-3d
Post-Processor 7.1

Figure 13.5: 3D FEM model of the test motor.

along the circumferential direction, which increased dramatically the number

of elements in the model and hence the computation time. These problems
contributed to the transition towards the 2D software.

The 2D model
The 2D model corresponds to that described in Chapter 12. The currents in
Figure 13.4 could only be loaded as tables using the transient analysis module.
This solver included the effects eddy currents. The airgap forces were calculated
at 1952 time steps along one period of the fundamental current, which provided
a resolution of 9.3 samples per noise cycle. The fact that the maximum number
of output cases per analysis was limited to 99 implied that 20 analyses had to be
carried out in order to obtain the results at the 1952 points within one cycle of
the fundamental. Each analysis was re-started from the last value in the previous
solution. A static analysis was performed before starting to rotate the rotor, in
order to reduce starting transients. In fact, no transients were observed in the
first fundamental cycle of the current. These results were the same as those
at the end of the third cycle, after one rotor turn. Therefore it was concluded
that the forces could be extracted already from the first cycle of the fundamental.
The calculations were carried out in a PC pentium 1.2 GHz processor equipped
258 Chapter 13. Acoustic noise prediction

Y [mm]
70.0

60.0
19 1112125
11223
31113
50.0 3329
114
2735
68 53
86 101 56 62
40.0 95 89
92 59
98 65
50 115
83 17
77
80 44
47 9
10067525561
948885 41 116
74
71 91
97 58
6449 38
30.0 82
76 43 11
79 4640 15
73
70 37
117
20.0 108 104 13
5
118

10.0 7
3
119

0.0 1
-30.0 -10.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0
X [mm]
Component: BMOD
0.0 1.0 2.0
OPERA-2d
Pre and Post-Processor 7.506

Figure 13.6: Flux density distribution in the 2D FEM model.

with 500 Mb RAM, and the 1952 simulations for each case were completed in
60 hours, including post-processing. A picture showing the flux density in the
motor at one position is presented in Figure 13.6.

Force computation
The shell in model F was divided into 72 elements along the circumferential
direction and 5 in the longitudinal direction. The forces in the electromagnetic
model were calculated at 13 points along each equivalent element in the shell,
and the frequency spectrum of the forces in Cartesian components was applied
to the nodes located along the inner surface of this shell. The results in the
time domain were translated into the frequency domain using the fast Fourier
transform function (FFT) in M ATLAB. The nature of the 2D force calculation
implied that all the nodes along the extrusions in the shell were assigned the
same force as their corresponding node in the first layer. The force applied at
one node represents the one that should be applied in the middle of the adjacent
element. Since there is actually one more layer of nodes than of elements, the
nodes around both edges of the cylinder were assigned half of the force that
is applied at the inner nodes. The force densities in the airgap were obtained
from the electromagnetic FEM simulation, implementing (13.8) and (13.9)
13.3. Electromagnetic analysis 259

through the middle of the airgap. The values of Hx and Hy correspond to

the Cartesian components of the airgap magnetomotive force, Bx and By are
the flux densities and Nx and Ny are unit length vectors corresponding to the
x and y components around the motor. The force densities, Fx and Fy were
transformed into force by multiplying with the area of an imaginary element
at the airgap, spanned through the same angle as one element in the back core
of the shell, and with the same length. In this way, the total force applied is
maintained constant, although the force density in the back core is reduced
since the area of the elements is increased.
Hm
Fx = Bx Hn − Nx (13.8)
2
hm
Fy = By Hn − Ny (13.9)
2
Hn = Hx Nx + Hy Ny (13.10)
Hm = Bx Hx + By Hy (13.11)

There are several ways of integrating the force in the electromagnetic model.
One extrusion in the acoustic model consists of 72 elements and there are 36
teeth around the stator. Therefore, half of the elements could be assigned the
force in the teeth and the other half the force in the slots, alternating between
adjacent elements. In practice, part of the force in the sides of the teeth was also
applied to the slot element, since the width of the teeth and the slots is not the
same. The integration performed in this way was referred to as the odd function
integration. The other possibility is to do the integration of the forces around
groups of half a tooth and half a slot, and vice-versa for the adjacent group, and
this was referred to as the even function integration. An example of the forces
calculated with both methods is presented in Figure 13.7. The odd function
integration was used in the simulations since it resembled the force distribution
around the airgap of the real machine. However, it could be argued that the
gradient of the forces in the back core between adjacent elements would be
small and therefore the even function should be used instead. It was observed
that the acoustic results obtained using the odd function integration correlated
better with the measurements.
Two tables were created for each one of the Cartesian components of the
forces that should be applied at each node in the acoustic model. Each table
consisted of three columns, the first one specifying the frequency and the second
and the third ones specifying the real and imaginary components of the Fourier
260 Chapter 13. Acoustic noise prediction

45 45
te e th te e th
40 s lo ts 40 s lo ts
35 35
30 30
F o rc e [N ]

F o rc e [N ]
25 25
20 20
15 15
10 10
5 5

0 0
0 0 .1 0 .2 0 .3 0 .4 0 0 .1 0 .2 0 .3 0 .4
T im e [s e c ] T im e [s e c ]

(a) Even integration (b) Odd integration

Figure 13.7: Force at the teeth and slots for the even and odd integration func-
tions.

transform. The spectrum of the total force was computed using (13.12) in order
to compare the applied force for the different cases simulated. This was done
by adding the absolute value of the spectrum at the 72 nodes around one layer
of elements. The expression for FN and its components are given in (13.13)-
(13.15).
72
FT = FN (13.12)
N =1

FN = 2 + F2
Fnx (13.13)
ny

Fnx =  (fnx )2 +  (fnx )2 (13.14)

Fny =  (fny )2 +  (fny )2 (13.15)
In this way, the values of the total force were calculated when the machine
rotated at 70 r.p.m. at no load. Figure 13.8 shows the results of the total force in
one extrusion for direct current modulation ([d=m, q=0]) and quadrature cur-
rent modulation ([d=0, q=m]) with a noise frequency of 490 Hz, as well as the
13.4. Acoustic analysis 261

case with no modulation ([d=0, q=0]). For comparison, the narrow band results
were transformed into 1/12 octave bands by linearly adding all the components
within each band, see Figure 13.8(d). This plot shows a substantial increase in
the total force due to either direct or quadrature current modulation.
Figure 13.6 shows that a considerable amount of iron in the machine is
saturated. The flux density reaches 1.75 Tesla in the rotor teeth and 1.6 Tesla in
the stator teeth and in the stator back core. It is reasonable to assume that the
saturated part of the BH curve at best can be determined experimentally within
10% accuracy. A sensitivity analysis was performed by increasing the saturated
part of the BH curve by 10% for the case [d=m, q=0] (see Figure 13.9(a)). It
can be seen from Figure 13.9(b) that the total force was increased by up to 25
N at some frequencies around 500 and 1000 Hz.

13.4 Acoustic analysis

The acoustic simulations were performed for a frequency range between 80 Hz
and 2250 Hz, in steps of 10 Hz. The sound power results at these frequencies
were stored and converted into 1/3 octave bands for comparison with the mea-
surements. The sound power level (PWL) without noise excitation is shown in
Figure 13.10(a), and the results when a 490 Hz harmonic is added to the di-
rect or quadrature currents are shown in Figure 13.10(b), and Figure13.10(c),
respectively. A structural damping of 0.01 for all the modes was used in these
simulations. Another way of comparing the predictions with measurements
is to plot the difference in the sound power level between the case when one
of the currents is modulated and the case when there is no modulation at all.
These results are presented in Figure 13.10(d) and Figure 13.10(e), and a com-
parison between the direct and the quadrature current modulation is shown in
Figure 13.10(f ).
In general, from these plots it can be observed that there are some discrep-
ancies below 400 Hz between the simulated and the experimental results. This
may be attributed to the rotor not being included in the acoustic model (Wang
and Lai, 1999). It should be noted that when no extra noise is added to the cur-
rents the sound power level at the 1 kHz, 1.25 kHz and 2 kHz centre frequen-
cies is considerably higher compared to the measurements, see Figure 13.10(a).
These levels are maintained when any of the currents is modulated. At the exci-
tation frequency of 490 Hz the difference in the measured PWL between direct
and quadrature modulation is 4 dB, whereas this difference reaches 30 dB in
262 Chapter 13. Acoustic noise prediction

3 3
10 10

2 2
10 10
T o ta l fo rc e [N ]

T o ta l fo rc e [N ]
1 1
10 10

0 0
10 10

−1 −1
10 10

−2 −2
10 10
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
F re q u e n c y [H z ] F re q u e n c y [H z ]

(a) Case [d=m, q=0] (b) Case [d=0, q=m]

3 4
10 10
d0q0
dm q0
10
2 d0qm
3
10
T o ta l F o rc e [N ]
T o ta l fo rc e [N ]

1
10
2
10
0
10

1
−1 10
10

−2 0
10 10
0 500 1000 1500 2000 2500 0 500 1000 1500 2000
F re q u e n c y [H z ] F re q u e n c y [H z ]

(c) Case [d=0, q=0] (d) 1/12 octave band comparison

Figure 13.8: Spectrum of the total force for the different simulation cases.
13.4. Acoustic analysis 263

2 .5 4
10
o rig in a l
1 0 % in c re a s e
2
3
10

T o ta l fo rc e [N ]
1 .5
[T ]

2
10
B

1
10
0 .5
o rig in a l
1 0 % in c re a s e 0
0 10
0 1000 2000 3000 4000 5000 0 500 1000 1500 2000
H [A ] F re q u e n c y [H z ]

(a) BH curve increase (b) Force comparison for [d=m, q=0]

Figure 13.9: Effect of a BH curve change in the calculated airgap force.

the simulations, see Figure 13.10(b) and Figure 13.10(c). The difference in the
measurements is because the acoustic noise emission is more readily enhanced
by radial forces (direct current) than tangential forces (quadrature current).
From the incremental plots, it can be observed that while the general trend
is similar for the predictions and measurements, there are some noticeable quan-
titative differences for the frequency range between 400 Hz and 1 kHz. The
agreement between predictions and measurements at higher frequencies (above
1 kHz) is rather promising. This seems to indicate that the acoustic model
does not reproduce accurately the behaviour of the experimental motor below
1 kHz, and this was confirmed through a detailed modal analysis. The modes
of vibration up to 2 kHz and the part of the motor they are associated with
are presented in Table 13.1, both for the simulations and the experiments. It
can be appreciated that below 1 kHz only three modes are calculated in the FE
simulation for the structural model used in the acoustic simulations, and all of
them are associated with the seat. However, the experiments show ten modes of
vibration in the same frequency range, associated with the stator, rotor, casing
and seat. Also, the mode close to the 490 Hz excitation frequency is related
to the seat instead of to the stator/casing as in the experiments. Hence, the re-
264 Chapter 13. Acoustic noise prediction

m e a su re d m e a su re d
80 s im u la te d 80 s im u la te d

60 60
P W L [d B ]

P W L [d B ]
40 40

20 20

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
F re q u e n c y [H z ] F re q u e n c y [H z ]

(a) Case [d=0, q=0] (b) Case [d=m, q=0]

m e a su re d 5 0
80 m e a su re d
s im u la te d s im u la te d
4 0
60 3 0
P W L [d B ]

P W L [d B ]

2 0
40
1 0
20
0

0 − 1 0
0 500 1000 1500 2000 0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0
F re q u e n c y [H z ] F re q u e n c y [H z ]

(c) Case [d=0, q=m] (d) Inc. [d=m,q=0]-[d=0,q=0]

5 0 5 0
m e a su re d m e a su re d
4 0 s im u la te d 4 0 s im u la te d

3 0 3 0
P W L [d B ]

P W L [d B ]

2 0 2 0

1 0 1 0

0 0

− 1 0 − 1 0
0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0 0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0
F re q u e n c y [H z ] F re q u e n c y [H z ]

(e) Inc. [d=0,q=m]-[d=0,q=0] (f ) Inc. [d=m,q=0]-[d=0,q=m]

Figure 13.10: Comparison of simulated and experimental acoustic results at 1/3

octave band centre frequencies.
13.4. Acoustic analysis 265

Experiment Simulation Experiment Simulation

294 rotor 1105 stator 1123 stator
375 seat 1240 seat 1097 seat
392 st/casing 1270 casing
490 st/casing 1307 casing
567 seat 485 seat 1393 seat 1353 seat
664 rotor 1480 seat 1559 seat
717 seat 695 seat 1535 casing
770 casing 1655 casing 1572 casing
816 casing 1743 stator 1682 stator
928 rotor 1795 stator 1933 stator
961 casing 1869 casing 1897 casing
1057 stator 1121 stator 1919 casing

Table 13.1: Comparison between the simulated and the experimental modes of
vibration.

sponse of the structure at this frequency would be expected to be quite different

from the measured response. Above 1 kHz, there is better agreement between
the experimental and the simulated results, not only in the frequencies but also
in the part of the motor the modes are associated with, hence the better agree-
ment above 1 kHz between the simulated and measured sound emissions. The
peaks observed at 1 kHz, 1.25 kHz and 2 kHz are related to rotor harmonics.
In fact, skewing was not included in the 2D electromagnetic model, so the rotor
harmonics are not appropriately damped at these frequencies.
The results due to a change in the damping coefficient of the material, from
0.01 to 0.05, are presented in Figure 13.11(a). It can be observed that the sound
power level could be reduced by up to 10 dB at 500 Hz, 800 Hz, 1250 Hz and
2000 Hz, whereas the effect is negligible at other centre frequencies. Finally,
the results from a change in the properties of the material, by increasing 10%
the saturated part of the BH curve, are presented in Figure 13.11(b). It can be
observed that the variations in the sound power level are around 3 dB or less up
to 2 kHz.
Finally, the sound pressure distribution in the motor and the surround-
ings obtained from S YSNOISE at 490 Hz is presented in Figure 13.12, for the
case with no modulation. The results when the direct and the quadrature cur-
266 Chapter 13. Acoustic noise prediction

90 90
0 .0 1 d a m p in g o rig in a l
80 0 .0 5 d a m p in g 80 1 0 % in c re a s e
70 70
60 60
[d B ]

[d B ]
50 50
L

L
40 40
P W

P W
30 30
20 20
10 10
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
F re q u e n c y [H z ] F re q u e n c y [H z ]

(a) Change in damping (b) Change in BH curve

Figure 13.11: Effect of a change in the damping and in the BH curve on the
sound power level for the case [d=m, q=0].

rents are modulated are presented in Figure 13.13 and Figure 13.14 respectively.
These plots actually confirm that at this frequency the vibration of the motor in
the acoustic model is associated with the seat instead of with the stator, where
no displacement is recorded on the surface of the casing. It should be noted that
the scales used for these plots are different since otherwise the cases [d=0, q=0]
and [d=0, q=m] can not be distinguished compared to the case [d=m, q=0].
The lower limit of the scales is 35 dB for all the plots, and the upper limit is
40 dB, 85 dB and 70 dB for the cases [d=0, q=0], [d=m, q=0] and [d=0, q=m]
respectively.

13.5 Conclusions
The aim of the noise prediction is to be able to model the acoustic radiation
from a machine without actually building it, in order to implement appropriate
noise control measures at the design stage. The deterministic approach using
FEM/BEM has been adopted to predict the noise emissions from a vector con-
trolled induction motor.
13.5. Conclusions 267

Figure 13.12: SPL distribution at 490 Hz, case [d=0, q=0].

Figure 13.13: SPL distribution at 490 Hz, case [d=m, q=0].

268 Chapter 13. Acoustic noise prediction

Figure 13.14: SPL distribution at 490 Hz, case [d=0, q=m].

It was shown that the disagreement at low frequencies was due to the in-
accuracy in the structural modal model. The number of modes of vibration
predicted in A NSYS in this region was considerably less compared to the mea-
surements. Also, these modes were associated with other parts of the machine,
specially at the 490 Hz frequency used for the prediction. Including the ro-
tor and perhaps the teeth in the structural model might improve these results.
However, the complexity of the model would be considerably increased, espe-
cially since it is difficult to assess how the windings constrain the mobility of
the teeth.
However, the behaviour of the model resembled that of the machine above
1 kHz, as could be observed from the modal results and the noise prediction.
Now the results are affected by the force calculation, since the skewing of the
rotor, which minimizes the rotor harmonics, has not been modelled. This could
explain why the levels of noise were considerably higher above 1 kHz in the
simulations, when the currents were not modulated with extra noise. Skewing
could be accurately modelled using a 3D electromagnetic model. However,
there are still important limitations in 3D transient FEM calculations, both in
complexity and computation time.
The influence of uncertainties in the damping of the structure and the sat-
13.5. Conclusions 269

urated part of the BH curve has also been assessed. Increasing five times the
overall damping applied to all the modes did not affect the response at differ-
ent frequencies in the same way. Basically, some of them remained unchanged
whereas others experienced a reduction of 10 dB. Hence, the damping is a crit-
ical factor for an accurate noise prediction. However, it is difficult to estimate a
priori in electrical machines, due to the particular assembling characteristics of
different machines and the modifications introduced by the winding. A change
in the material properties corresponding to a 10% increase in the saturated part
of the BH curve was associated with a change of 3 dB or less in the acoustic sim-
ulations. Therefore, compared to the damping, the inaccuracies in the saturated
part of the BH curve have relatively low impact in the noise prediction.
270 Chapter 13. Acoustic noise prediction
Chapter 14

Acoustic properties of stators with

a layer of air gaps

14.1 Introduction
Adjustable-speed drives in industry and commercial buildings embodied as con-
verter fed a.c. induction motor drives have a major drawback in increased mag-
netic noise. This is due to the strong interaction between the fundamental
airgap flux and harmonic components in the voltages and currents from the
converter. To deal with this problem, numerous studies have been conducted
to investigate possible changes in the converter switching pattern that would re-
duce these harmonics, or move them to a frequency range where the human ear
is less sensitive, as pointed out in Chapter 9. However, not enough effort has
been directed towards reducing noise through a change in the structure of the
motor. The aim of this chapter is to evaluate the influence that several structural
changes in the stator have in the acoustic behaviour of the machine.

Previous work
Andersson and Skarrie (1997) carried out a theoretical study using a structure
where a layer of air gaps was introduced around the stator periphery to reduce
the noise emission. The goal was to isolate the outer surface of the stator, and
therefore the surrounding air, from the vibrations transmitted through the teeth
into the core. The air gaps in the new mass-spring-mass system (inner core -
air gaps - outer core) were optimized so as to reduce the vibrations of the new
structure. A prototype was built and tested, but there was no improvement

271
272 Chapter 14. Acoustic properties of stators with a layer of air gaps

in the acoustic behaviour compared to the original structure. However, no in-

depth investigation was carried out into why such a structure did nod lead to
the desired improvements. The work presented in this chapter provides further
insight into understanding the mechanisms of noise production in this novel
design solution using combined structural FEM and acoustic BEM simulations.
Other structures with different air gap design are also studied, as well as the
ideal case where the inner core and the outer core are completely separated by a
continuous airgap layer.

14.2 Basic model

The vibrations produced by the airgap forces are transmitted through the teeth
into the core, and then into the surrounding air. The teeth can contribute to
the vibration of the stator since their stiffness in the tangential direction is less
than that in the core. This contribution to the noise production becomes more
important if the tangential airgap forces are big enough to induce a tangential
movement of the teeth. However, the influence of the radial forces in the teeth
is negligible, and the teeth mainly serve as a path for the transmission of this
excitation into the core. The simulations in this chapter are only concerned
with the transmission of the vibrations from the core into the surrounding air,
independent of whether or not the teeth contribute to these vibrations.
The modes of vibration in the machine structure were calculated using A N -
SYS version 5.1. The core was modelled as a cylinder using Solid45 elements,
see Figure 14.1. The laminated nature of the structure was simulated by setting
different values for the Young’s modulus in the circumferential and in the ax-
ial directions, 2.72·109 and 1.36·1011 respectively. The element type and the
physical properties of the cylinder correspond to those used in the stator of the
motor by Wang and Lai (1999), which provided a good correlation between the
experimental and numerical modal analysis results. The cylinder had an inner
radius of 68 mm, an outer radius of 93 mm and a length of 60 mm, and the
number of nodes and elements was 4032 and 3024 respectively. The size of the
elements was adapted to achieve good accuracy when the model was used for
the acoustic simulations, which required a minimum of six elements per wave-
length. The upper frequency for the acoustic analysis was 5 kHz, and thus the
maximum length allowed for any of the three dimensions of the elements was
11 mm. In order to meet this specification, the number of divisions along the
circumferential and axial directions were 72 and 6, respectively. With this dis-
14.2. Basic model 273

Figure 14.1: Cylindrical model of the core.

tribution, the area of each element in the inner circumference became 2.1·10−5
m2 . In the radial direction, the structure was meshed in a series of layers graded
towards the surface of the cylinder (Figure 14.1), where the modifications in the
form of air gaps will be introduced.
The modal analysis of the structure was performed using the subspace me-
thod in A NSYS, and all the modes of vibration up to 5 kHz were extracted.
These modes, together with the geometry, were imported into S YSNOISE. In
this package, a finite element (FE) model was used first to calculate the vibration
response in the surface of the structure. The magnitude of the applied stress was
5·104 N/m2 , which corresponded to a force of 1.05 N at each element located
in the inner surface of the cylinder. The components of the applied force at the
nodes were set so that the magnitude of the force was the same for all the nodes
and the direction pointing outwards in the radial direction. The vibration re-
sponse was then used to calculate the radiated sound power using a boundary
element (BE) model. This model was solved using the BE indirect variational
approach (Zhou, 1993).
The mechanical input power in a structure is defined as the product be-
tween the force and the vibration velocity at the driving point. In general, this
274 Chapter 14. Acoustic properties of stators with a layer of air gaps

power is converted into two components: the vibration input power, which in
turn is converted into acoustic power depending on the radiation efficiency, and
the power dissipated by structural damping. In the following, the vibration in-
put power will be referred to simply as input power. In the BE indirect method,
the radiation efficiency is calculated as the ratio of the output power (Wo ) to
the input power (Wi ), which are defined as in (14.2) and (14.1),

1 pσ ∗
Wo = dS (14.1)
2 −ipω


Wi = ρc v̄n2 dS (14.2)

where ρ is the density of the fluid (air in this case), c is the speed of sound, v̄n2
is the space average mean-square
√ normal velocity of the surface, S is the area of
the radiating surface, i = −1, p is the acoustic pressure, σ ∗ is the complex
conjugate of the single layer potential, and ω is the circular frequency. It should
be noted that the radiation efficiency calculated in this way is equivalent to the
definition found in the specialized literature (Harris, 1991), which is the ratio
of the sound power radiated by the structure to the sound power radiated by a
piston (which is large compared with the acoustic wavelength) vibrating with
the same space-average mean-square velocity as the structure, see (14.3) . If the
spatial velocity is uniform then the behaviour is like in the case of the piston.
If the spatial velocity is not uniform, then the behaviour is similar to a piston
with a uniform velocity equal to the spatial averaged velocity of the structure.

Wo
σ= (14.3)
ρcSv̄n2

Equations (14.2) and (14.1) are evaluated through all the radiating sur-
faces, and the result for the active sound power is determined over a surface
enclosing the source, as it is done experimentally by means of sound intensity
measurements. The coupling between the structure and the surrounding fluid
was defined as ‘weak’. This is true when the fluid is air, since the modes of vibra-
tion of the structure are not modified by the surrounding fluid. Nevertheless,
simulations were also performed using a ‘strong’ coupling and the changes in
the results were negligible.
14.3. Zig-zag gap model. 275

The geometry analyzed contains a cavity, and when used to calculate ex-
terior radiation problems, the solution is not unique. This means that there
is more than one solution at a given frequency, but it is not possible to know
whether the solution has converged to the right answer. The unreal solutions
correspond to the resonance frequencies of the cavity and when a solution is ob-
tained for one of these frequencies, the only way of ensuring that the solution
converges to a correct answer is to ensure that the sound pressure field inside
the cavity does not become too large because of resonance. When using a BE
direct method this can be achieved specifying a minimum number of points
(known as overdetermination points) inside the cavity, where the sound pres-
sure is constrained to be zero. In this case the solution is ‘overdetermined’, since
there are initially m equations and m unknowns but, by specifying additional
points inside the cavity, new constraints are introduced overspecifying the prob-
lem. When using a BE indirect method, the treatment to overcome this ‘non-
uniqueness’ problem is to add some impedance (absorption) inside the cavity.
However, according to the S YSNOISE manual, impedance elements should not
be used inside closed volumes in a BE indirect model which is linked to a struc-
ture. In this case, ‘singularities’ in the response will be due to physically-correct
behaviour, the interaction of internal resonances of the volumes through the
flexible structure to the exterior. Since all the models in this thesis are simulated
by using a BE indirect model which is linked to a structure, the ‘non-uniqueness’
problem will not be present and no remedy is needed for this phenomenon.
The superposition method was used for the acoustic analysis, and it is based
on a linear summation of the contributions from the different modes. Each
mode can be represented as a curve with a resonance at a certain frequency.
The response at the resonance frequency is dominated by the mode at that
frequency. At any other frequency, the response depends on the location of
the nearby resonance frequencies and the number of modes involved in the
addition. The results obtained from the basic model (Figure 14.1) are compared
in the following sections with other models where some changes are introduced
in the core structure.

14.3 Zig-zag gap model.

The shape of the zig-zag gap is shown in Figure 14.2, and it resembles the one
studied by Andersson and Skarrie (1997). The gaps were distributed in two
adjacent layers and shifted in the circumferential direction. These gaps were
276 Chapter 14. Acoustic properties of stators with a layer of air gaps

located close to the stator surface, at 95% of the outer radius, in order to reduce
their interaction with the natural paths followed by the flux. The number of
elements in the model was reduced to 2448, while the number of nodes was
unchanged.

Figure 14.2: Zig-zag gap model.

The fact that some iron material is replaced by air results in a decrease in the
weight of the structure, which in turn affects the modal frequencies. Thus two
cases were considered. In the first case, the density of the new model was kept
the same as the basic model, which implied a reduction in its total weight. In
the second case, the density of the new model was increased so that the weight
of the basic model was maintained. Figure 14.3 shows a comparison between
the total sound power level (PWL) radiated by the three different structures,
when a material damping of 0.01 was used. It can be observed that the frequen-
cies of the zig-zag gap model with constant weight have slightly different peak
frequencies compared to the case with constant density, but their response is
still very similar up to 3 kHz. Above this limit the curve of the constant weight
model is slightly shifted towards lower frequencies, as it would be expected since
the modal frequencies decrease when the mass of the system is increased.
The acoustic response of the zig-zag gap model is actually worse compared
to the basic model. The PWL is increased below 2 kHz. Between 2 kHz and 3
kHz the response is characterised by a sudden change in the slope of the PWL.
The radiation efficiency in this region is significantly more accentuated in the
zig-zag model, as shown in Figure 14.4. Above 4 kHz the PWL is slightly higher
14.3. Zig-zag gap model. 277

9 0

8 0

7 0
P W L [d B ]

6 0

5 0

4 0
n o g ap s
3 0 s a m e d e n s ity
s a m e w e ig h t
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.3: PWL of basic and zig-zag gap models, damping 0.01.

than in the basic model, even though the radiation efficiency is considerably
lower in the zig-zag model. Although the force applied in the different models
is the same, the input power is different (Figure 14.5) because the structures are
different. The structure with gaps is less stiff, so it would be expected to have
a higher response velocity and therefore higher input power. In general, the
increase in the PWL is a result of higher input power and, to a certain extent,
higher radiation efficiency in the zig-zag model, as shown in the figures.
The influence of the material damping was tested for the constant weight
case. The radiated sound power level for a material damping of 0.001, 0.01 and
0.1 is presented in Figure 14.6. A decrease in the damping implies an increase
in the PWL at the resonance frequencies above 3 kHz. When the damping is
increased from 0.001 to 0.1, the PWL at the peak frequencies above 3 kHz is
reduced by 20-40 dB. This behaviour is explained by the input power being also
reduced above 3 kHz due to the lower response velocity of the structure when
the damping is increased, as shown in Figure 14.7. The radiation efficiency
in this region is also decreased when the damping is increased, as shown in
Figure 14.8.
The results for the zig-zag gap structure agree with those for the proto-
type presented by Andersson and Skarrie (1997), see Figure 14.9. That is, the
278 Chapter 14. Acoustic properties of stators with a layer of air gaps

1 .2 n o g ap s
s a m e d e n s ity
s a m e w e ig h t
1
R a d ia tio n e ffic ie n c y

0 .8

0 .6

0 .4

0 .2

0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.4: Radiation efficiency of basic and zig-zag gap models, damping 0.01.

acoustic response is not improved when the gaps are introduced around the sta-
tor, and it is characterized by a shift in the resonance frequencies. However, it
should be noted that the results are not directly comparable to the basic model,
since the experimental stator included teeth and windings. Its vibrations were
also constrained by the casing and the end-plates attached to the stator as in any
conventional machine, and the exciting force was also different.
The idea of introducing the gaps in the back of the stator core was based
on the assumption that the vibrations are mainly transmitted through iron and
that the air barriers would contain these vibrations within the inner part of the
core, thus isolating the outer part and also the surrounding air. The unsatis-
factory results were initially attributed to the fact that the vibrations could still
be transmitted through the thin iron paths located between the two air gap lay-
ers. Hence, reducing the amount of these paths would contribute to a better
isolation, and this led to the study of a new geometry.
14.3. Zig-zag gap model. 279

1 0 0

9 0

8 0
In p u t P o w e r [d B ]

7 0

6 0

5 0 n o g ap s
s a m e d e n s ity
s a m e w e ig h t
4 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.5: Input power of basic and zig-zag gap models, damping 0.01.

1 1 0
1 0 0
9 0
8 0
P W L [d B ]

7 0
6 0
5 0
4 0
D a m p in g = 0 .0 1
3 0 D a m p in g = 0 .0 0 1
D a m p in g = 0 .1
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.6: PWL of zig-zag gap model with constant weight for different damp-
ing coefficients.
280 Chapter 14. Acoustic properties of stators with a layer of air gaps

1 2 0

1 0 0
In p u t P o w e r [d B ]

8 0

6 0

4 0 D a m p in g = 0 .0 1
D a m p in g = 0 .0 0 1
D a m p in g = 0 .1
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.7: Input power of zig-zag gap model with constant weight for different
damping coefficients.

D a m p in g 0 .0 1
1 .2
D a m p in g 0 .0 0 1
D a m p in g 0 .1
1
R a d ia tio n e ffic ie n c y

0 .8

0 .6

0 .4

0 .2

0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.8: Radiation efficiency of zig-zag gap model with constant weight for
different damping coefficients.
14.4. Four gap model 281

Figure 14.9: Experimental results from Andersson and Skarrie (1997). The con-
tinuous line corresponds to the prototype without air gap layers, and
the dashed-dotted line to the prototype with air gap layers.

14.4 Four gap model

This model was built by replacing the iron with air between the gaps and be-
tween the two air gap layers in the zig-zag gap structure. Only four small areas of
connection remained between the inner and outer parts of the core, as shown in
Figure 14.10. The number of nodes and elements decreased to 3136 and 1800
respectively. A comparison between the PWL obtained from the new structure
and the basic model is shown in Figure 14.11. Within the same frequency range,
there are seven peak frequencies for the PWL in the four gap model compared
with only two in the basic model. For the four gap model, the peak frequencies
below 2.5 kHz are lower for the constant weight case compared with the con-
stant density case, because of its heavier mass. The radiation efficiencies for the
four gap model are considerably higher than those of the basic and zig-zag gap
models for frequencies above 2.5 kHz, as shown in Figure 14.12. The input
power, however, is considerably lower than the radiated acoustic power at peak
frequencies above 2.5 kHz, as shown in Figure 14.13.
282 Chapter 14. Acoustic properties of stators with a layer of air gaps

Figure 14.10: Four gap model.

The influence of a change in the damping of the structure was also studied
and the results are presented in Figure 14.14. The response below 2.5 kHz
at the peak frequencies is considerably reduced as the damping is increased.
This behaviour is explained by the lower input power below 2.5 kHz when the
damping is increased, as shown in Figure 14.15. The radiation efficiency, shown
in Figure 14.16, is very similar for the three cases, except around 2 kHz, where
it increases with the material damping.
The reduction in the number of iron paths between the inner and the outer
core in the four gap model does not yield a reduction in the radiated sound
power level. Actually, the acoustic behaviour of the structure is worsened,
mainly due to the increase in the number of resonance frequencies within the
same frequency range. A theoretical study where there are no connections be-
tween the inner and the outer parts of the core would determine the maximum
achievable level of isolation.
14.4. Four gap model 283

9 0

8 0

7 0
P W L [d B ]

6 0

5 0

4 0
n o g ap s
3 0 s a m e d e n s ity
s a m e w e ig h t
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.11: PWL of basic and four gap models, damping 0.01.

2
1 0

1
1 0
R a d ia tio n e ffic ie n c y

0
1 0

− 1
1 0

− 2
1 0

− 3
1 0 n o g ap s
s a m e d e n s ity
− 4 s a m e w e ig h t
1 0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.12: Radiation efficiency of basic and four gap models, damping 0.01.
284 Chapter 14. Acoustic properties of stators with a layer of air gaps

1 2 0

1 0 0
In p u t P o w e r [d B ]

8 0

6 0

4 0 n o g ap s
s a m e d e n s ity
s a m e w e ig h t
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.13: Input power of basic and four gap models, damping 0.01.

1 0 0

9 0

8 0
P W L [d B ]

7 0

6 0

5 0 D a m p in g = 0 .0 1
D a m p in g = 0 .0 0 1
D a m p in g = 0 .1
4 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.14: PWL of four gap model with constant weight for different damp-
ing coefficients.
14.4. Four gap model 285

1 4 0
D a m p in g = 0 .0 1
1 3 0 D a m p in g = 0 .0 0 1
D a m p in g = 0 .1
1 2 0
In p u t P o w e r [d B ]

1 1 0

1 0 0

9 0

8 0

7 0

6 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.15: Input power of four gap model with constant weight for different
damping coefficients.

2
1 0
D a m p in g 0 .0 1
1
D a m p in g 0 .0 0 1
1 0 D a m p in g 0 .1

0
R a d ia tio n e ffic ie n c y

1 0

− 1
1 0

− 2
1 0

− 3
1 0

− 4
1 0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.16: Radiation efficiency of four gap model with constant weight for
different damping coefficients.
286 Chapter 14. Acoustic properties of stators with a layer of air gaps

14.5 Ideal gap model

The ideal model was built with a continuous airgap separating completely the
inner and outer parts of the core, forming two separate rings, as shown in Fig-
ure 14.17. The number of nodes and elements was 3528 and 2160 respectively.
This increase in the number of elements compared with the four gap model is
because one of the thin layers close to the surface was not replaced by air. Al-
though there is no physical connection between the two rings, their vibration
modes were calculated by A NSYS. It should be noted that in the BE acoustic
model using S YSNOISE, the link between the inner and outer rings is still pro-
vided by the air, through which the vibrations are transmitted. Both cases where
the fluid-structure interaction was defined as ‘weak’ and ‘strong’ were simulated,
and the change in the results was negligible. This is reasonable since the load
imposed by the air on the structure is not big enough to change its modes of
vibration.

Figure 14.17: Ideal gap model.

Since the difference in the previous models between the constant weight and
constant density simulations is small, only the structure with the same weight
as the basic model was simulated. The result is shown in Figure 14.18. The
overall PWL is still higher compared to the basic model, mainly in the region
between 2 kHz and 3 kHz. The peak PWL occurs around 2.7 kHz, and at
this frequency the radiation efficiency of the ideal model is considerably more
pronounced, as shown in Figure 14.19.
14.5. Ideal gap model 287

9 0

8 0

7 0
P W L [d B ]

6 0

5 0

4 0

3 0 n o g ap s
id e a l g a p
2 0
1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.18: PWL of ideal gap and basic models, damping 0.01.

2
1 0

1
1 0

0
R a d ia tio n e ffic ie n c y

1 0

− 1
1 0

− 2
1 0

− 3
1 0
n o g ap s
− 4 id e a l g a p
1 0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0
F re q u e n c y [H z ]

Figure 14.19: Radiation efficiency of ideal gap and basic models, damping 0.01.
288 Chapter 14. Acoustic properties of stators with a layer of air gaps

To investigate the difference in the noise radiation between 2 kHz and 3

kHz, the modal characteristics of the ideal gap model and the basic model were
compared in that region. The basic model has three modal frequencies, at 2176
Hz, 2357 Hz and 2753 Hz. The modes of the two rings in the ideal model
were calculated separately in A NSYS. The inner ring has exactly the same three
modes of vibration as the basic model, but the frequencies were slightly higher
(2246 Hz, 2357 Hz and 2846 Hz), as it would be expected since the mass of
the structure is decreased. Hence, the inner ring and the basic model can be
considered to be acoustically equivalent between 2 kHz and 3 kHz. The outer
ring has 9 vibration modes in this region. Furthermore, it was observed that the
modes of vibration calculated for the inner ring and the outer ring separately
coincided with all the modes of vibration calculated in A NSYS for the ideal gap
model.
Table I shows a comparison of the sound power levels calculated at the
modal frequencies for the basic and the ideal gap models. The values with the
superscript ‘*’ in the ideal gap model correspond to the modes associated with
the inner ring. The value with the superscript ‘**’ corresponds to a modal fre-
quency present in both the inner and the outer rings. The remaining modes
calculated for the ideal gap model are associated with the outer ring. From Ta-
ble I, it is clear that the outer ring is excited, and also that it is responsible for
the higher PWL of the ideal model compared to the basic model. The sound
emitted at the frequencies corresponding to the inner ring modes is higher com-
pared to the same modes in the basic model. This is due to the influence of the
nearby outer ring modes. Also, it should be noted that the peak PWL response
occurs at 2721 Hz, which is associated with the vibration mode of the outer
ring.
The results from the simulations indicate that not only the iron but also
the air are the media conducting the vibrations from the inner part to the outer
part of the core. Even if the transmission coefficient through the air is lower
compared to the iron, the fact that the stiffness in the outer part of the cylinder
is considerably decreased by removing iron material is a key factor in the acous-
tical behaviour of the new structures. On one hand the outer core introduces
more modal frequencies within the same frequency range. On the other hand,
the outer core is also more easily deformed, which makes it more sensitive to
the vibrations transmitted from the inner core. This behaviour can be seen in
Figures 14.20, 14.21 and 14.22, which show some of the vibration modes cal-
culated in A NSYS for the different structures. It is illustrated how the degree of
14.6. Conclusions 289

Table 14.1: Modal frequencies for the basic and ideal gap models
Basic model Ideal gap model
2076 Hz: 50.7 dB
2176 Hz: 52.9 dB 2246 Hz: 56.7 dB∗
2278 Hz: 57.7 dB
2336 Hz: 59.7 dB
2357 Hz: 53.9 dB 2357 Hz: 60.4 dB∗∗
2604 Hz: 71.9 dB
2675 Hz: 78.0 dB
2721 Hz: 83.7 dB
2805 Hz: 82.1 dB
2842 Hz: 72.9 dB
2753 Hz: 58.5 dB 2846 Hz: 79.0 dB∗

deformation experienced in the outer part of the core is more accentuated than
in the inner part, and this contributes to the radiation of noise, which explains
the unsatisfactory results obtained from the models where the surrounding air
gaps are introduced.

14.6 Conclusions

In this chapter the effect of some structural changes in the stator of an electrical
machine with the aim of reducing the noise emissions has been discussed. The
structural change consisted of introducing some air gaps around the periphery
of the stator to control the transmission path for the vibrations. Various ge-
ometries for the air gaps have been investigated, together with the ideal case,
where the outer core is not physically connected through iron with the inner
core. Results indicate that the idea of introducing air gaps in the periphery of
the stator core does not reduce the noise emissions. Replacing some iron by air
reduces the stiffness in this area and makes the structure more sensitive to vibra-
tion excitation. However, it has also been shown that increasing the damping
290 Chapter 14. Acoustic properties of stators with a layer of air gaps

(a) Mode 81, 3397 Hz (b) Mode 89, 3546 Hz

Figure 14.20: Modes of vibration in the ideal model.

(a) Mode 6, 848 Hz (b) Mode 58, 2918 Hz

Figure 14.21: Modes of vibration in the four gaps model.

14.6. Conclusions 291

(a) Mode 35, 3640 Hz (b) Mode 37, 3716 Hz

(c) Mode 68, 4618 Hz (d) Mode 69, 4659 Hz

Figure 14.22: Modes of vibration in the zig-zag gap model.

292 Chapter 14. Acoustic properties of stators with a layer of air gaps

of the material decreases the noise emissions. An increase in the damping could
be achieved by filling the slots with additional materials, preferably with high
thermal and magnetic conductivity.
Chapter 15

Conclusions

This second part of the thesis deals with topics related to the acoustic noise
radiated from vector controlled induction motor drives, the noise prediction in
an electrical machine, and structural changes in the stator to reduce the noise
emissions. First, a summary of the obtained results is given. Second, some
topics for future research regarding experimental and numerical studies of the
acoustic noise in electrical machines are provided.

15.1 Summary of results

Noise from vector controlled drives

The acoustic noise emissions from a vector controlled induction motor drive
have been analyzed. Vector control provides faster transient response compared
to conventional v/f drives, and this is achieved with an efficient decoupling of
the currents controlling the flux and the torque in the machine. Measurements
were carried out when the direct and the quadrature currents were modulated
with high frequency noise signals, exciting the structure with radial and tangen-
tial forces.
The directional behaviour of the noise radiation from electrical machines
implies that the most proper way of quantifying their acoustic noise emissions
is to measure the sound power. Results from sound intensity measurements
indicated that direct modulation produced up to 7 dB more sound power than
quadrature modulation. This is likely due to the radial/bending modal shape
of the excited mode of vibration, which is more easily excited by radial than

293
294 Chapter 15. Conclusions

tangential forces.
When the machine was loaded, it was observed that the overall sound power
emissions decreased by up to 13 dB. This should be due to a change in the
electromagnetic component of the noise, since the mechanical and aerodynamic
components were found to be negligible. It is however difficult to generalize
these results for other machines, since the factors influencing the noise emissions
at load could affect in different ways the noise radiation from one machine to
another.

Noise prediction
A method for the prediction of the sound radiation from an electrical machine
has been analyzed, based on the combination of structural and electromagnetic
finite element models, together with an acoustic boundary element model. It
was found that the level of agreement between the noise calculations and the
measurements was sensitive to a few factors. From the structural point of view,
it is mandatory to calculate properly the modes of vibration in the structural
analysis over the whole frequency range of interest, to ensure a proper corre-
lation with the real machine. It was observed that a change in the damping
influenced the response at some frequencies considerably, while at others the
behaviour remained basically unchanged. From the electromagnetical point of
view, using a two-dimensional model for the induction machine implies that
the skewing of the rotor bars cannot be precisely modelled. If skewing is not
modelled at all, the rotor harmonics are not damped, which in turn affects the
force calculations. Finally, a change in the material magnetic properties, con-
sisting of a 10% increase in the flux density along the saturated part of the BH
curve, produced an increase of 3 dB or less in the simulated acoustic response.

Stator structural changes

A layer of air gaps of different geometries was introduced around the periph-
ery of the stator core, in order to interfere with the path for the vibrations and
reduce the noise emissions. The idea was tested by means of finite and bound-
ary element models. It was observed that the stiffness around the gap area is
decreased due to the material removal. This makes the structure more sensi-
tive to vibration excitation, and it actually radiates more noise than the original
model without air gaps. These results agree with earlier experimental tests with
a machine built using a similar concept.
15.2. Future work 295

15.2 Future work

Concerning the experiments, other modal frequencies could be tested in order
to assess if the direct and the quadrature current modulation affect differently
the noise emissions. Regarding the acoustic noise prediction, the FEM modal
analysis referred in this thesis needs to be improved below 1 kHz for a better
agreement with the real machine. The prediction process could also be tested at
higher frequencies, for example at 1105 Hz, where the predicted and measured
modes of vibration show a better agreement than at 490 Hz. However, if the
sampling frequency in the electromagnetic calculations is maintained, then the
number of simulations at higher frequencies is considerably increased. Further-
more, for a more accurate force calculation the airgap forces should be extracted
from a three-dimensional FE electromagnetic model, in order to take into ac-
count the influence of skewing and end-windings. This would however also
increase dramatically the computation time.
The results from Chapter 14 indicated that an increase in the damping of
the material resulted in a decrease of the noise emissions. New ways of increas-
ing homogeneously the damping in the stator structure could be investigated.
In principle, iron powder cores would present a higher damping factor in the
radial and circumferential directions compared to a similar axially laminated
structure, due to the distributed insulation around the iron particles. However,
the damping in the axial direction is probably higher in the laminated core, due
to the thicker layer of insulation between the laminations. Nevertheless, the
study of the acoustic properties of iron-powder structures is needed, in order to
assess the acoustic characteristics of machines built from this material compared
to their laminated counterparts.
296 Chapter 15. Conclusions
Bibliography

Andersson, C., Skarrie, H. (1997). Mekanisk ljudreducering i en asynkron-

maskin. MSc Thesis, Dept of Industrial Electrical Engineering and Automa-
tion, Lund University, Sweden. (In Swedish).

Alaküla, M. (2002). Power electronic control. Dept of Industrial Electrical Engi-

neering and Automation, Lund University, Sweden.

Astfalck, A., Pulle, D.W.J., Lai, J.C.S. (1997). Prediction of transient stator
force distribution of inverter driven induction motors. Proceedings of AU-
PEC/EECON’97, The University of New South Wales, vol. 1, pp. 1-6, Aus-
tralia.

Bakhuizen, A.J.C. (1973). A contribution to the development of stepping motors.

PhD Thesis, Eindhoven Technical University, The Netherlands.

Belkhayat, D., Roger, D., Brudny, J.F. (1997). Active reduction of magnetic
noise in asynchronous machines controlled by stator current harmonics. Pro-
ceedings of IEE International Conference in Electrical Machines and Drives, pp.
400-405, London, UK.

Belmans, R., Verdyck, D., Geysen, W., Findlay, R.D. (1991). Electro-
mechanical analysis of the audible noise of an inverter-fed squirrel-cage in-
duction motor. IEEE Transactions on Industry Applications, vol. 27, no. 3, pp.
539-544.

Belmans, R., Hameyer, K. (1998). Impact of inverter supply and numerical

calculation techniques in audible noise problems. International seminar on
vibrations and acoustic noise of electric machinery, pp. 9-23. Béthune, France.

297
298 Bibliography

Bologniani, S., Di Bella, A., Zigliotto, M. (1999). Random modulation and

acoustic noise reduction in IM drives: a case study. Proceedings of IEE Interna-
tional Conference on Electrical Machines and Drives, pp. 137-141, Canterbury,
UK.

Brozek, R. (1973). No-load to full-load airborne noise level change on high-

speed polyphase induction motors. IEEE Transactions on Industry Applica-
tions, IA-9, 2, pp. 180-184.

Brüel & Kjær (1986). Sound Intensity booklet.

Brüel & Kjær (1988). Structural testing. Part 2: Modal analysis and simulation.

Chang, S.C., Yacamini, R. (1996). Experimental study of the vibrational be-

haviour of machine stators. IEE Proceedings on Electric Power Applications,
vol. 143, no. 3, pp. 242-250.

Chau, K.T., Jiang, S.Z., Chan, C.C. (2000). Reduction of current ripple and
acoustic noise in dual-inverter pole-changing induction motor drives. Record
of the 31th Annual IEEE Power Electronics Specialists Conference, vol. 1, pp.
67-72.

Cho, D.H., Kim, K.J. (1998). Modelling of electromagnetic excitation forces

of small induction motor for vibration and noise analysis. IEE Proceedings on
Electric Power Applications, vol. 145, no. 3, pp. 199-205.

Covic, G.A., Boys, J.T. (1998). Noise quieting with random PWM AC drives.
IEE Proceedings on Electric Power Applications, vol. 145, no. 1, pp. 1-10.

Delaere K., Hameyer K., Belmans, R. (1998). Static eccentricity as a cause for
audible noise of induction motors. Proceedings of International Conference on
Electrical Machines, vol. 1, pp. 502-506, Istanbul, Turkey.

Ewins, D.J. (1984). Modal testing: theory and practice. Research Studies Press
Ltd.

Fahy, F.J. (1995). Sound Intensity. E & FN Spon, London, UK.

Garcia-Otero, S., Devaney, M. (1994). Minimization of acoustic noise in vari-

able speed induction motors using a modified PWM drive. IEEE Transactions
on Industry Applications, vol. 30, no. 1, pp. 111-115.
Bibliography 299

Garvey, S.D., Glew, G.N. (1999). Magnetostrictive excitation of vibration in

machines - a modal approach. Proceedings of IEE International Conference on
Electrical Machines and Drives, pp. 169-173, Canterbury, UK.

Garvey, S.D., Le Flem, G.D. (1999). Tangential forces matter. Proceedings of

IEE International Conference on Electrical Machines and Drives, pp. 174-178,
Canterbury, UK.

Gerling, D. (1994). The audible noise of induction machines compared for

different methods of calculating the air-gap field. Proceedings of International
Conference on Electrical Machines, vol. 3, pp. 409-414, Paris, France.

Habetler, T.G., Deepakraj, M.D. (1991). Acoustic noise reduction in sinusoidal

PWM drives using a randomly modulated carrier. IEEE Transactions on Power
Electronics, vol. 6, no. 3, pp. 356-363.

Hadj Amor, A., Timar P.L., Poloujadoff M. (1995). Induction squirrel cage ma-
chine design with minimization of electromagnetic noise. IEEE Transactions
on Energy Conversion, vol. 10, no. 4, pp. 681-687.

Harris, C.M. (1991). Handbook of acoustical measurements and noise control. 3rd
edition, McGraw-Hill, USA.

Holmes, D.G. (1997). A generalized approach to the modulation and control

of hard switched converters. PhD Thesis, Dept of Electrical and Computer
Systems Engineering, Monash University, Australia.

Hubert, A., Friedrich, G. (2002). Influence of power converter on induction

motor acoustic noise: interaction between control strategy and mechanical
structure. IEE Proceedings on Electric Power Applications, vol. 149, no. 2, pp.
93-100.

Hupe, R., Kennel, R. (1989). Increase of power and reduction of noise and
current ripple of inverter fed induction machines. Proceedings of European
Conference on Power Electronics and Applications, pp. 47-50, Aachen, Ger-
many.

Ishibashi, F., Noda, S., Mochizuki, M. (1998). Numerical simulation of elec-

tromagnetic vibration of small induction motors. IEE Proceedings on Electric
Power Applications, vol. 145, no. 6, pp. 528-534.
300 Bibliography

ISO 9614-2 (1996). Acoustics - Determination of sound power levels of noise

sources using sound intensity - Part 2: Measurement by scanning.

Jahns, T.M., Owen, E.L. (2001). AC adjustable-speed drives at the millenium:

How did we get there? IEEE Transactions on Power Electronics, vol. 16, no. 1,
pp. 17-25.

Kobayashi, T., Tajima, F., Ito, M., Shibukawa, S. (1997). Effects of slot com-
bination on acoustic noise from induction motors. IEEE Transactions on
Magnetics, vol. 33, no. 2, pp. 2101-2104.

Krause, P.C. (1986). Analysis of Electric Machinery. McGraw-Hill.

Lipo, T.A. (1996). Introduction to AC machine design. WisPERC UW, Madison

Wisconsin, USA.

Lo, W.C., Chan, C.C., Zhu, Z.Q., Xu, L., Howe, D., Chau, K.T. (2000).
Acoustic noise radiated by PWM-controlled induction machine drives. IEEE
Transactions on Industrial Electronics, vol. 47, no. 4, pp. 880-889.

Låftman, L. (1997). The contribution to noise in an induction machine, fed

from a pulse width modulated inverter. Proceedings of European Conference
on Power Electronics and Applications, vol. 2, pp. 2483-2489, Trondheim,
Norway.

Maliti, K.C. (2000). Modelling and analysis of magnetic noise in squirrel-cage

induction motors. PhD Thesis, Dept of Electric Power Engineering, Electrical
Machines and Power Electronics, Royal Institute of Technology, Sweden.

Norton, M.P. (1989). Fundamentals of noise and vibration analysis for engineers.
Cambridge University Press, UK.

Novotny, D.W., Lipo, T.A. (1996). Vector control and dynamics of AC drives.
Oxford Science Publications, Wisconsin, USA.

Ramesohl, I., Kaehler, C., Henneberger, G. (1999). Influencing factors on

acoustical simulations including manufacturing tolerances and numerical
strategies. Proceedings of IEE International Conference on Electrical Machines
and Drives, pp. 142-146, Canterbury, UK.
Bibliography 301

Stemmler, H., Eilinger, T. (1994). Spectral analysis of the sinusoidal PWM

with variable switching frequency for noise reduction in inverter-fed induc-
tion motors. Record of the 25th Annual IEEE Power Electronics Specialists
Conference, vol. 1, pp. 269-277.
Stephen, D.D., Davies, R.J. (1970). Discussion on acoustic noise and vibration
of rotating electrical machines. IEE Proceedings, vol. 117, no. 1, pp. 127-129.
Timar, P.L. (1989). Noise and vibration of electrical machines. Elsevier Science
Publishers, Amsterdam, The Netherlands.
Timar, P.L., Lai, J.C.S. (1994). Acoustic noise of electromagnetic origin in
an ideal frequency converter driven induction motor. IEE Proceedings -B on
Electric Power Applications, vol. 141, no. 6, pp. 341-346.
Timar, P.L., Lai, J.C.S. (1996). Analysis of the effects of time harmonics on the
radiated sound power of induction motors. Proceedings of Internoise 96, Book
1, pp. 445-448, Liverpool, UK.
Vandevelde, L., Melkebeek, J.A.A. (1994). Theoretical and experimental study
of radial forces in relation to magnetic noise of induction motors. Proceedings
of International Conference on Electrical Machines, vol. 3, pp. 419-424, Paris,
France.
Vandevelde, L., Melkebeek, J.A.A. (1997). Calculation of radial magnetic forces
for the analysis of noise and vibrations of squirrel-cage induction motors.
Proceedings of IEE International Conference in Electrical Machines and Drives,
pp. 86-90, London, UK.
Vas, P. (1990). Vector control of AC machines. Oxford University Press, UK.
Verma, S.P., Balan, A. (1998). Experimental investigations on the stators of
electrical machines in relation to vibration and noise problems. IEE Proceed-
ings on Electric Power Applications, vol. 145, no. 5, pp. 455-461.
Wang, C. (1998). Vibro-acoustic analysis of inverter driven induction motors.
PhD Thesis, School of Aerospace and Mechanical Engineering, University
College, The University of New South Wales, Australia.
Wang, C., Lai, J.C.S. (1999). Vibration analysis of an induction motor. Journal
of Sound and Vibration, 224(4), pp. 733-756.
302 Bibliography

Wang, C., Lai, J.C.S., Astfalck, A. (2004). Sound power radiated from an
inverter driven induction motor II: numerical analysis. IEE Proceedings on
Electric Power Applications, vol. 151, no. 3, pp. 341-348.

Wang, C., Lai, J.C.S., Pulle, D.W.J. (2002). Prediction of acoustic noise from
variable speed induction motors: deterministic versus statistical approaches.
IEEE Transactions on Industry Applications, vol. 38, no 4, pp. 1037-1044.

Xu, L., Zhu, Z.Q., Howe, D. (2000). Acoustic noise radiated from direct
torque controlled induction motor drives. IEEE Proceedings on Electric Power
Applications, vol. 147, no. 6, pp. 491-496.

Yang, S.J. (1981). Low noise electrical machines. Oxford University Press, New
York, USA.

Zhou, P. (1993). Numerical analysis of electromagnetic fields. Springer-Verlag,

New York, USA.

Zhu, Z.Q., Howe, D. (1994). Improved methods for prediction of electromag-

netic noise radiated by electrical machines. IEE Proceedings on Electric Power
Applications, vol. 141, no. 2, pp. 109-120.

Zhu, Z.Q., Xu, L., Howe, D. (1997). Acoustic noise radiated by nominally
identical induction machines fed from sinusoidal and non-sinusoidal power
supplies. Proceedings of European Conference on Power Electronics and Applica-
tions, vol. 2, pp. 2489-2494, Trondheim, Norway.

Zhu, Z.Q., Xu, L., Howe, D. (1999). Influence of mounting and coupling
on the natural frequencies and acoustic noise radiated by a PWM controlled
induction machine. Proceedings of IEE International Conference on Electrical
Machines and Drives, pp. 164-168, Canterbury, UK.

Åstrom, K.J., Wittenmark, B. (1997). Computer controlled systems: theory and

design. 3rd edition, Pentrice Hall, Inc., NJ, USA.
Appendix E

Current controllers

E.1 Controller for the direct current

Predictive dead beat control assumes that the reference value for one signal
(denoted in the following by the superscript ‘*’) set at one discrete time tk is
achieved by the actual value of the signal with one sampling time delay, i.e. at
time tk+1 . For the voltage signal this implies that u∗sd (tk ) = usd (tk+1 ). The
average voltage at time tk+1 can be expressed as a moving average filter of the
form in (E.1).
tk+1
1
usd (tk+1 ) = usd (τ )dτ (E.1)
Ts
tk

Introducing (10.8) from Chapter 10 into this equivalence and multiplying

by the sampling time Ts leads to (E.2).
tk+1
1
Ts u∗sd (tk ) = Rs Ts isd (τ )dτ + ψs (tk+1 ) − ψs (tk ) (E.2)
Ts
tk

In this equation the derivative of the flux has been approximated using a
forward Euler approximation (Åstrom and Wittenmark, 1997). The average
value of the current within one sampling interval is denoted by isd (tk+1 ), fol-
lowing the same equivalence as for the voltage in (E.1). Assuming that the
current varies linearly within one sampling interval, it can be approximated as

303
304 Appendix E. Current controllers

in (E.3).
isd (tk+1 ) + isd (tk )
isd (tk+1 ) (E.3)
2
If the leakage inductance is neglected, the flux is expressed as ψs isd LM .
For dead beat control isd (tk+1 ) = i∗sd (tk ) and introducing these changes into
(E.2) yields (E.4).
Rs Ts ∗
Ts u∗sd (tk ) = [isd (tk ) + isd (tk )] + i∗sd (tk ) − isd (tk ) (E.4)
2
This expression can be transformed using (E.5), which leads to (E.6).
Rs Ts Rs Ts
= Rs Ts − (E.5)
2 2

Rs Ts
Ts u∗sd (tk ) = Rs Ts isd (tk ) + LM + [i∗sd (tk ) − isd (tk )] (E.6)
2
Assuming dead beat control, the current at one point in time tk can be
written as the sum of all the set point changes up to the time tk−1 , as in (E.7).
j=k−1
isd (tk ) = [i∗sd (tj ) − isd (tj )] (E.7)
j=0

and the final expression for the direct current controller takes the form in (E.8).
j=k−1
Ts u∗sd (tk ) = Rs Ts [i∗sd (tj ) − isd (tj )] +
j=0
Rs Ts
+ LM + [i∗sd (tk ) − isd (tk )] (E.8)
2

E.2 Controller for the quadrature current

Equation (10.9) for the quadrature current can be transformed into (E.9) using
the equivalence that the synchronous speed ωs is the sum of both the slip and
rotor electrical frequencies, ωsl and ωm respectively.

usq = Rs isq + ωsl ψs + ωm ψs (E.9)

E.2. Controller for the quadrature current 305

In order to deduce an expression for the slip frequency, it is helpful to

consider the rotor equation, which in stationary coordinates takes the form in
(E.10).

αβ αβ
 αβ
dψ

−jωm ψR + iR RR +
 R
=0 (E.10)
dt
This equation is translated into the synchronous dq reference frame yielding
(E.11), where the rotor flux and current in synchronous coordinates are given
by (E.12) and (E.13) respectively.
 dq
dψR  dq + idq RR = 0
+ j(ωs − ωm )ψ (E.11)
R R
dt

 dq = ψ
ψ  dq − idq Lσ (E.12)
R s s
 dq
idq = ψR − idq s (E.13)
R
LM
The rotor current can be expressed in terms of stator variables substituting
(E.12) into (E.13), and introducing the equivalence Ls = LM + Lσ , where Ls
is the stator inductance, LM is the magnetizing inductance and Lσ is the com-
bined leakage inductance from the stator and the rotor in the four parameter
model, and (E.14) is obtained.
 dq
idq = ψs − Ls idq (E.14)
R
LM LM s
Introducing (E.12) and (E.14) into (E.11) yields (E.15).
sdq
dψ didq  
s  dq − idq Lσ +
− Lσ + jωsl ψ s s
dt dt
RR  dq RR Lsdq
+ ψ − i =0 (E.15)
LM s LM s
This expression can be divided into a real and an imaginary part, still con-
sidering that ψsq = 0 for the reference system selected, as in (E.16) and (E.17)
respectively.
dψs disd RR RR Ls
− Lσ + ωsl isq Lσ + ψs − isd = 0 (E.16)
dt dt LM LM
306 Appendix E. Current controllers

disq RR Ls
−Lσ + ωsl (ψs − Lσ isd ) − isq = 0 (E.17)
dt LM
From (E.17) the slip frequency is obtained, assuming that ψs −Lσ isd ψs
since the leakage flux can be neglected compared to the linkage flux in the
machine, giving (E.18).
disq
RR Ls
LM isq + Lσ dt
ωsl = (E.18)
ψs
Substituting (E.18) into (E.9) leads to (E.19).

RR Ls disq
usq = Rs + isq + Lσ + ωm ψs (E.19)
LM dt

In the same way as for the direct voltage, u∗sq (tk ) = usq (tk+1 ) where the
quadrature voltage is averaged using a moving average filter of the form shown
in (E.1). Introducing (E.19) in the expression for the moving average filter and
multiplying by the sampling time Ts yields (E.20).
tk+1
RR Ls 1
Ts u∗sq (tk ) = Rs + Ts isq (τ )dτ +
LM Ts
tk
tk+1
1
+Lσ [isq (tk+1 ) − isq (tk )] + Ts ωm (τ )ψs (τ )dτ (E.20)
Ts
tk

In this equation the derivative of the current has been approximated with a
forward difference, using Euler’s method. The integral of the quadrature current
divided by the sampling time becomes isq (tk+1 ) which can be approximated as
in (E.3) for the direct current. The approximation in (E.21) is applied to the
speed dependent term.
tk+1
1
ωm (τ )ψs (τ )dτ ωm (tk )ψs (tk ) (E.21)
Ts
tk

If the main flux and the speed change, they do it slowly and these compo-
nents can be assumed to be constant during one sampling period. Assuming
E.2. Controller for the quadrature current 307

dead beat control for the quadrature current leads to (E.22).

j=k−1
" #
isq (tk ) = i∗sq (tj ) − isq (tj ) (E.22)
j=0

Taking into consideration all these assumptions, (E.20) can be rewritten as

in (E.23).
j=k−1
RR Ls " #
Ts u∗sq (tk ) = Rs + Ts i∗sq (tj ) − isq (tj ) +
LM
j=0
 
RR Ls Ts " ∗ #
+ Lσ + Rs + isq (tk ) − isq (tk ) +
LM 2
+ ωm (tk )ψs (tk )Ts (E.23)
308 Appendix E. Current controllers
Appendix F

Acoustic terminology

In acoustics it is important to distinguish between sound pressure and sound

power. Sound power relates to the rate at which a sound source radiates energy,
i.e. energy per unit time. A noise source radiating a certain amount of sound
power will produce a change in the sound pressure depending on the surround-
ing environment. This concept can be better understood by considering its
analogy with a heater placed in a room. The heater radiates a certain amount
of heat into the room (sound power), but the actual temperature (sound pres-
sure) will depend on the dimensions of the room, materials, other heat sources,
insulation, distance from the heat source, etc. Sound pressure is therefore the
physical quantity that is perceived by the ear. The unit of pressure is Pascal (Pa),
and 1 Pascal is equivalent to 1 N/m2 . The human ear is sensitive to a range of
frequencies from 20 Hz to 20 kHz and a range of acoustic pressure from the
threshold of hearing of 20 µPa to the threshold of pain of 200 Pa (Norton,
1989). Owing to the wide range of pressure (10 million times) that causes the
sensation of hearing, a logarithmic scale is used to express the sound pressure
level (SPL, Lp ) in decibels (dB). The SPL is defined by (F.1), where p is the
root-mean-square sound pressure in Pa and p0 is the reference sound pressure,
which is 20 µPa for air.

2
p
Lp = log10 (F.1)
p0

In general, for noise due to a composite of sounds with different frequencies,

309
310 Appendix F. Acoustic terminology

the total sound pressure level is given by (F.2).

n
Lp = log10 100.1Lpi (F.2)
i=1

The amount of noise produced by a machine cannot be necessarily quanti-

fied by measuring the sound pressure. The sound power has to be determined
instead, which is a better descriptor of the noise producing capacity of a sound
source. The sound power level (PWL, LW ) of a source is defined in dB by (F.3),
where W0 is the reference power, which is 10−12 W for air.

W
LW = log10 (F.3)
W0
Related to the sound power is the concept of sound intensity, which is the
sound power in watts per unit area in m2 , and is defined as the average rate of
energy flow through a unit area normal to the direction of propagation, as given
by (F.4), where u is the particle velocity. The instantaneous sound intensity,
pu, is a vector quantity whose magnitude and direction at any point in space
may vary with time. Note also that energy travelling back and forth does not
contribute to the averaged intensity, and thus there will be some net intensity
flow only if there is a net energy flow through the area considered.

T
1
I = pudt (F.4)
T
0

The sound intensity level, LI , of a source is defined in dB by (F.5), where

I0 is the reference intensity which is 10−12 W/m2 for air.

I
LI = log10 (F.5)
I0
The sound pressure or sound intensity levels can be displayed and stored
in a spectrum analyzer for further post-processing. Frequency analyzers may be
based on either constant bandwidth or proportional bandwidth devices. The
constant bandwidth device is basically a tunable narrow-band filter with con-
stant bandwidth ∆f = fu − fl , where fl and fu are the lower and upper
half-power frequencies. The proportional bandwidth device consists of a series
 311

of relatively broadband filters such that the ratio fu /fl is constant. The centre
frequency fc of a proportional bandwidth filter is calculated as in (F.6). The
most common proportional bandwidth filters are in the form that satisfies that
fu /fl = 2x , where x=1 for octave band filters and x=1/3 for 1/3-octave band
filters.

fc = fl fu (F.6)

In order to reflect the response of the ear to different frequencies, a weight-

ing function (known as A-weighting function) approximating an inverted 40-
phon loudness contour is usually applied to the measured sound pressure level
and the result is expressed as dB(A). There are other weighting functions, as
shown in Figure F.1. The B and C functions approximate the human response
at loudness levels of 70 phon and 90 phon respectively while D is a standarized
function for use in aircraft noise measurements. The A-weighting scale is the
most widely used since it best approximates the human response. Below 1 kHz
and over 6 kHz the A-scale is negative, representing a reduction in the sound
level in order to compensate for the human ear poorer sensitivity in the lower
and higher frequency ranges.

2 0
D
1 0
C A
0
R e la tiv e re s p o n s e [d B ]

B (B + C )
− 1 0
D
− 2 0

− 3 0 A

− 4 0

− 5 0

− 6 0 2 3 4
1 0 1 0 1 0
F re q u e n c y [H z ]

Figure F.1: The frequency weighting functions (Harris, 1991).

312 Appendix F. Acoustic terminology

Sound measurement techniques

Sound power can be related to sound pressure only under carefully controlled
conditions, where special assumptions are made about the sound field. The
‘free field’ describes sound propagation in idealized free space where there are
no reflections. These conditions are held in open air far from the ground or in
an anechoic room, where all the sound striking the walls is absorbed. In a free
field the intensity I is related to the root-mean-square pressure p by (F.7), where
c is the speed of sound.

p2
I= (F.7)
ρc
Introducing this equation into (F.5) leads to (F.1), implying that in a free
field the sound intensity level is equal to the sound pressure level. Hence, the
sound power of a source can be determined by measuring the averaged sound
pressure level over a hypothetical surface enclosing the sound source in an ane-
choic room, and is given by (F.8), where Lp is the averaged surface sound pres-
sure level in dB and S is the area of the test surface in m2 .

LW = Lp + log10 S (F.8)

In a ‘diffuse field’ sound is reflected so many times that it travels in all

directions with equal magnitude and probability, and this field is approximated
in a reverberant room. In a diffuse field, the relation between the intensity and
the pressure is given by (F.9).

p2
I= (F.9)
4ρc
The sound power level of a source can be determined by measuring the
average sound pressure level inside a reverberation room. It can be calculated
from (F.10), where T is the reverberation time of the room in seconds, V is the
volume of the room in m3 , S is the total surface of the room in m2 and λ is the
wavelength at the centre frequency of the frequency band in meters.

Sλ
LW = Lp − log10 T + log10 V + log10 1 + (F.10)
8V
Appendix G

No load sound intensity

measurements
314 Appendix G. No load sound intensity measurements

Figure G.1: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=0].
 315

Figure G.2: Measured sound intensity and residual index over the four free sides
of the frame, case [d=m, q=0], f=490 Hz.
316 Appendix G. No load sound intensity measurements

Figure G.3: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=m], f=490 Hz.
 317

Figure G.4: Measured sound intensity and residual index over the four free sides
of the frame, case [d=m, q=0], f=1105 Hz.
318 Appendix G. No load sound intensity measurements

Figure G.5: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=m], f=1105 Hz.
Appendix H

Load sound intensity

measurements
320 Appendix H. Load sound intensity measurements

Figure H.1: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=0].
 321

Figure H.2: Measured sound intensity and residual index over the four free sides
of the frame, case [d=m, q=0], f=490 Hz.
322 Appendix H. Load sound intensity measurements

Figure H.3: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=m], f=490 Hz.
 323

Figure H.4: Measured sound intensity and residual index over the four free sides
of the frame, case [d=m, q=0], f=1105 Hz.
324 Appendix H. Load sound intensity measurements

Figure H.5: Measured sound intensity and residual index over the four free sides
of the frame, case [d=0, q=m], f=1105 Hz.