Sie sind auf Seite 1von 60

Digital Comprehensive Summaries of Uppsala Dissertations

from the Faculty of Science and Technology 1444

Demagnetization and Fault


Simulations of Permanent Magnet
Generators

STEFAN SJÖKVIST

ACTA
UNIVERSITATIS
UPSALIENSIS ISSN 1651-6214
ISBN 978-91-554-9733-0
UPPSALA urn:nbn:se:uu:diva-303517
2016
Dissertation presented at Uppsala University to be publicly examined in Häggsalen,
Ångströmlaboratoriet, Lägerhyddsvägen 1, Uppsala, Friday, 9 December 2016 at 09:15 for
the degree of Doctor of Philosophy. The examination will be conducted in English. Faculty
examiner: Dr. Sami Ruoho (Nordmag Oy).

Abstract
Sjökvist, S. 2016. Demagnetization and Fault Simulations of Permanent Magnet Generators.
Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and
Technology 1444. 59 pp. Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-554-9733-0.

Permanent magnets are today widely used in electrical machines of all sorts. With their increase
in popularity, the amount of research has increased as well. In the wind power project at Uppsala
University permanent magnet synchronous generators have been studied for over a decade.
However, a tool for studying demagnetization has not been available. This Ph.D. thesis covers
the development of a simulation model in a commercial finite element method software capable
of studying demagnetization. Further, the model is also capable of simulating the connected
electrical circuit of the generator. The simulation model has continuously been developed
throughout the project. The simulation model showed good agreement compared to experiment,
see paper IV, and has in paper III and V successfully been utilized in case studies. The main
focus of these case studies has been different types of short-circuit faults in the electrical system
of the generator, at normal or at an elevated temperature. Paper I includes a case study with the
latest version of the model capable of handling multiple short-circuits events, which was not
possible in earlier versions of the simulation model. The influence of the electrical system on
the working point ripple of the permanent magnets was evaluated in paper II. In paper III and
VI, an evaluation study of the possibility of creating a generator with an interchangeable rotor
is presented. A Neodymium-Iron-Boron (Nd-Fe-B) rotor was exchanged for a ferrite rotor with
the electrical properties almost maintained.

Keywords: Demagnetization, Permanent magnet, Finite Element Method, Synchronous


generators, Wind power

Stefan Sjökvist, Department of Engineering Sciences, Electricity, Box 534, Uppsala


University, SE-75121 Uppsala, Sweden.

© Stefan Sjökvist 2016

ISSN 1651-6214
ISBN 978-91-554-9733-0
urn:nbn:se:uu:diva-303517 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-303517)
Not so permanent now are you, permanent magnet?
List of papers

This thesis is based on the following papers, which are referred to in the text
by their Roman numerals.

I S. Sjökvist and S. Eriksson, "Investigation of Permanent Magnet


Demagnetization in Synchronous Machines During Multiple
Short-Circuit Fault Conditions", Submitted to IET Renewable Power
Generation, September 2016.

II S. Sjökvist, M. Rossander and S. Eriksson, "Permanent Magnet


Working Point Ripple in Synchronous Generators", Submitted to IET
Journal of Engineering, October 2016.

III S. Sjökvist, P. Eklund and S. Eriksson, "Determining Demagnetisation


Risk for Two PM Wind Power Generators with Different PM Material
and Identical Stators", IET Electric Power Applications, vol. 10, no. 7,
pp. 593-597, August 2016.

IV S. Sjökvist and S. Eriksson, "Experimental Verification of a Simulation


Model for Partial Demagnetization of Permanent Magnets", IEEE
Transactions on Magnetics, vol. 50, no. 12, pp. 1-5, December 2014.

V S. Sjökvist and S. Eriksson, "Study of Demagnetization Risk for a


12 kW Direct Driven Permanent Magnet Synchronous Generator for
Wind Power", Energy Science and Engineering, vol. 1, no. 3, pp.
128-134, September 2013.

VI P. Eklund, S. Sjökvist, S. Eriksson and M. Leijon, "A Complete


Design of a Rare Earth Metal-Free Permanent Magnet Generator",
Machines, vol. 2(2), pp. 120-133, May 2014.

Reprints were made with permission from the publishers.

Other contributions:
A The author presented parts of the work from paper V at the conference
Magnetic Materials in Electrical Machine Applications 2012 in Pori,
Finland, 13-15 June 2012.
B L. Sjökvist, J. Engström, S. Larsson∗ , M. Rahm, J. Isberg and M. Lei-
jon, "Simulation of Hydrodynamical Forces on a Buoy - a Comparison
Between Two Computational Approaches", In Proceedings of the 1st
Asian Wave and Tidal Energy Conference Series, Jeju, South Korea, 27-
30 November 2012.

∗ The author changed his surname from Larsson to Sjökvist in July 2013.
Contents

Nomenclature .................................................................................................... ix
List of Abbreviations ........................................................................................ xi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1 History of the wind power project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Aim of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 Magnetic materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Classification of magnetic materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Magnetism on a macroscopic level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.3 Soft ferromagnetic materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.4 Hard ferromagnetic materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 The finite element method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Derivation of the transient field equation . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Modeling of the B-H-curve and the recoil line . . . . . . . . . . . . . . 24
3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1 Simulating with COMSOL Multiphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Development of demagnetization model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Experimental verification of stationary demagnetization model 31
3.3.1 Experimental Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Measuring the magnetic flux density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Evaluation of the two time-dependent demagnetization models 34
3.4.1 Electrical circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Transient analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Comparison with stationary model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5 Ripple in surface mounted permanent magnets . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Comparison of ferrite and Nd-Fe-B rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 Summary of results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1 Experimental verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Time-dependent models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Ripple in permanent magnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 The ferrite alternative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5 Conclusions ................................................................................................ 47
6 Summary of Papers .................................................................................... 49
7 Svensk sammanfattning ............................................................................. 52
8 Acknowledgements ................................................................................... 54
References ........................................................................................................ 55
Nomenclature

Symbol SI Unit Quantity

A Wb/m Magnetic vector potential


Az Wb/m Magnetic vector potential, z-component
Adest
z Wb/m Magnetic vector potential on the destination boundary, z-
component
Asrc
z Wb/m Magnetic vector potential on the source boundary, z-
component
B T Magnetic flux density
BHmax J/m3 Maximum energy product
Bmin T Minimum magnetic flux density
B|| T Magnetic flux density parallel to the magnetization direction
Br T Remanence
Bnew
r T New remanence
Brprev T Remanence from the previous iteration
Bs T Saturation of the magnetic flux density
Byprev T Magnetic flux density, y-component, from the previous iter-
ation
CDC F Capacitance on the DC-link
D C/m2 Electric displacement field
E V/m Electric field
fel Hz Electrical frequency
H A/m Magnetic field
Hc A/m Coercivity
HcJ A/m Intrinsic coercivity
Hmin A/m Minimum magnetic field
Hyprev A/m Magnetic field, y-component, from the previous iteration
I A Current
ID A Diode current
Im T Magnetic polarization or intensity of magnetization
IS A Reverse bias saturation current
J A/m2 Current density
Jf A/m2 Free current density
Jm A/m2 Magnetization current density
k Vs/Am Slope of the B-H-curve
K1 m/A Knee point shape parameter
kB J/K Boltzmann constant
kBr %/K Temperature coefficient for the remanence
Continued on next page
Symbol SI Unit Quantity

kHcJ %/K Temperature coefficient for the intrinsic coercivity


L H Generator inductance per phase
M A/m Magnetization
Ms A/m Saturation magnetization
q C Magnitude of the charge of an electron
R Ω Generator resistance per phase
Rab Ω Short-circuit resistance between phase a and b
Rbc Ω Short-circuit resistance between phase b and c
RDC Ω Short-circuit resistance over the load on the DC side
RL Ω Resistance load
Rs Ω Snubber resistance
T K Absolute temperature
Tre f K Listed reference temperature
u − Solution to the ODE
va , vb , vc V Generator output line-neutral voltages of phase a, b and c
VD V Diode voltage
Ea , Eb , Ec V Generator electromotive force of phase a, b and c
φ V Electric potential
ϕm ◦ Angle between the magnetization and the applied field
µ0 Vs/Am Permeability
µr − Relative permeability
σ A/Vm Electric conductivity
χm − Susceptibility
List of Abbreviations

Abbrevation Description

2D Two dimensional
3D Three dimensional
AC Alternating current
CM COMSOL Multiphysics R

DC Direct-current
DDPMSG Direct driven permanent magnet synchronous generator
DLL Dynamic-link library
FEM Finite element method
Nd-Fe-B Neodymium-Iron-Boron
ODE Ordinary differential equation
PM Permanent magnet
PMSG Permanent magnet synchronous generator
PMSM Permanent magnet synchronous machine
RD Rolling direction
Sm-Co Samarium-Cobalt
TD Transverse direction
VAWT Vertical axis wind turbine
VCCS Voltage-controlled current source
1. Introduction

Permanent magnets (PMs) are widely used as the field source in electrical
machines and compared to electrically excited machines there are no losses
in the excitation system which results in higher efficiency and higher output
power per volume of machine [1–7]. When designing an electrical machine
the design goals are; to optimize the use of the magnetic flux density (B) by
a suitable selection of both hard and soft magnetic materials, to maximize
the interaction between the magnetic flux density and the stator windings by
optimizing the geometry, and to minimize losses such as eddy-current, hys-
teresis and frictional losses. The most recent big step in PM development was
with the introduction of the Neodymium-Iron-Boron (Nd-Fe-B) PMs in the
1980’s [8, 9], not to say that PMs have not developed since then. There is
ongoing research on the material design of PMs trying to increase both their
ability to produce high fields and withstand high fields without being partially
demagnetized [10, 11].
Since permanent magnet synchronous machines (PMSMs) are widely used
and have a long expected lifespan, a machine is therefore likely to be exposed
to one or several faults during its lifetime. Faults in PMSM can be divided
into stator faults, rotor faults, and bearing faults. If the machine is connected
to an inverter, faults in the switches and the sensors may also occur. Faults
on the rotor side may be caused, for example, by eccentricity, damage in one
or more permanent magnets, asymmetries, or mechanical looseness. For the
stator, a cause of failure may be winding insulation breakdown, asymmetries
or mechanical looseness [12–15]. There are several methods for detecting and
predicting faults, e.g. by monitoring the mechanical vibrations, the temper-
ature, the insulation degradation, the power output, and/or the current [12].
Current monitoring is a widely spread technique which can be used on ma-
chines in production environments for continuous monitoring. The spectrum
of the current is analysed to determine abnormal frequency components due
to, for example, air gap eccentricity or partial demagnetization [12, 16–18].
In any machine design process, consideration needs to be put into toler-
ances and material properties to make sure the machine can be assembled and
can manage the mechanical stresses. The same needs to be done regarding
the electromagnetic properties. In the same way the mechanical construction
needs to be able to withstand some abnormal conditions, the electromagnetic
properties needs to be maintained after, e.g. an overload situation and/or a situ-
ation of increased temperature. The demagnetization risk should be analysed,
at least, for normal working conditions and some expected overload condi-
tions. Preferably, the most common fault conditions should also be analysed.

13
The arguments for not considering all possible fault conditions is the cost,
however, if the most severe case do happen, replacement of the whole ma-
chine may be necessary anyway. [19–21]
With the increasing popularity of PM electrical machines the amount of
studies on demagnetization has increased as well. In the 1980s and 1990s,
there were several studies on the influence of grain orientation alignment on
the coercivity and the angular dependence of an applied field [22–27]. In
these studies, one focus was to try to describe and present models for the grain
orientations influence on the intrinsic coercivity and the remanence of the per-
manent magnets. Roughly, the trend was a 1/ cos ϕm relationship, where ϕm is
the angle between the magnetization (M) and the applied magnetic field (H).
However, the results do not seem to converge to a general angular dependence
relationship. There are simply too many factors involved, such as the material
composition, grain size, and different alignment field strengths and pressing
directions during the sintering process.
Since the 2000s, the research focus has shifted a bit, from the influence of
the grain orientation, to more focus on case studies on actual machines and
design optimization [21, 28–31]. However, there is still research performed
on the inclined field topic [32]. For example, Ruoho presented a demagneti-
zation model taking the angular dependence into account using an empirical
model [33]. While this is the simplest model to implement in a finite element
method (FEM) model the author has come across, since it consists only of the
intrinsic coercivity (HcJ ), a third-degree polynomial, and ϕm , it is still an em-
pirical model suitable for a limited set of Nd-Fe-B PMs. While, to the author’s
knowledge, a general model is yet to exist, this polynomial model may be the
best option for designers of PM machines.

1.1 History of the wind power project


The work presented in this Ph.D. thesis is a part of the wind power research
project at the division for electricity, Uppsala University [34]. The wind power
project started in 2002 with the vision of developing a mechanically simple
and low maintenance wind power concept. The vision resulted in an idea
with a vertical axis wind turbine (VAWT), more specific a Darrieus H-type
rotor. The H rotor is omnidirectional and can, therefore, extract energy from
the wind regardless of the wind direction making a yaw system unnecessary.
The turbine is passive stall regulated. The turbine connects, via its shaft, to a
direct driven permanent magnet synchronous generator (DDPMSG) placed at
ground level. The possibility to install the generator and other heavy compo-
nents at ground level reduces stress on the tower which allows for a slender
tower construction, giving the system a low center of mass which is preferable
in e.g. floating offshore wind applications. Installing the generator and other
components at ground level also enables easier access during maintenance.

14
Figure 1.1. A vertical axis H-type wind turbine connected to a 12 kW DDPMSG
designed and built within the wind power project.

The wind power group have to date designed and produced three full wind
power plants (turbine and generator), and one lab generator. The lab genera-
tor was recently converted from using a surface mounted NdFeB rotor to the
present spoke type ferrite rotor. Figure 1.1 shows a vertical axis wind tur-
bine connected to a 12 kW DDPMSG, which was built within the wind power
project.
Previous work from the group has focused on the wind power system and
the properties of the generator [35–37], as well as aerodynamics and sound
pollution [38–41].
Here follows a summary of the most recent studies by the wind power
group. Eklund has performed studies on the mechanical design and the pos-
sible use of rare earth metal-free alternatives to the Nd-Fe-B PMs previously
used [42]. Rossander has performed measurements of the blade forces of the
12kW prototype and studied how oscillations propagate through the turbine
as well as how passive rectification of the generator affects the turbine [43].
Olauson has in [44] studied and modeled the variability and forecasting of re-
newable energy sources, and their influence on the power system. Apelfröjd
has made extensive research on the grid connection, electrical system, and
control system for the wind turbine [45]. He has also written a comprehen-
sive review of the previous work done by the wind power group over the last
decade [34].

15
While extensive work previously has been done regarding the generator
design, a tool for studying demagnetization of permanent magnets has not
been available. Therefore, design optimization has not been possible to the
same extent as if a tool for studying demagnetization had been available. A
demagnetization model would, therefore, be beneficial for the future generator
design and generator-VAWT interaction research.

1.2 Aim of this thesis


The aim of this thesis is to study and model demagnetization of permanent
magnets by, firstly, developing an easy to use simulation model which require
little hands-on application by the user.
A secondary objective is to build the entire model in commercially avail-
able software. In the majority of the references in the Introduction section, the
simulation software was either not mentioned or proprietary. Using a com-
mercially available software may have some limitations in the user’s ability to
implement specific features. But for research progress to become easily avail-
able in the industry, it is important to prove that it is possible to implement
special features/functions in commercial available software and not only in
proprietary software.
Finally, a third goal is to demonstrate and experimentally verify the simu-
lation model.

16
2. Theory

2.1 Magnetic materials


The magnetic material used in an electrical machine is more than just the PM
material. The magnetic materials in an electrical machine include both the
PMs and all the material used to complete the machine’s magnetic circuit,
usually called the active material. This section gives a brief introduction to the
different magnetic material classifications and their properties [46–48].

2.1.1 Classification of magnetic materials


The macroscopic magnetic properties of materials are a consequence of the
magnetic moments of the electrons in individual atoms. Each electron’s mag-
netic moment originates from two sources. One is the electron’s orbital motion
around the nucleus, and the other is the electron’s spin. The magnetic moment
of the orbital motion can be considered a small current loop, having its mag-
netic moment along the axis of rotation. Each electron can either have a spin
up or spin down and the magnetic moment is pointing along the spin axis. In
each atom, the individual orbital magnetic moments of the electrons can can-
cel each other out; this is also the case for the spin magnetic moment. An
electron cancels the spin magnetic moment of an electron with a spin up, with
a spin down. The net magnetic moment of an electron is, therefore, the sum of
the contributions of all electrons of the atom. For an atom having filled inner
and outer shells, there is total cancellation of both orbital and spin magnetic
moments. These materials are not able to be permanently magnetized.
The three common types of magnetism are diamagnetism, paramagnetism,
and ferromagnetism. There are further categories which usually are consid-
ered being subclasses of ferromagnetism; these subclasses will not be dis-
cussed. All materials exhibit at least one of these three types, depending on
their electrons response to an externally applied magnetic field.

Diamagnetism
Diamagnetism is a weak form of magnetism only present when an external
magnetic field is applied. With the application of an external magnetic field,
the orbital motion of the electrons is deranged causing a change in the total
magnetic moment of the atom. The change is small, and the atom’s magnetic
field is always pointing in the opposite direction of the external field. Dia-
magnetism is present in all materials but can only be observed when all other
forms of magnetism are absent since the effect is so small.

17
Paramagnetism
For materials with an incomplete cancellation of the atoms’ magnetic moment,
a net magnetic moment of each atom will remain even in the absence of an ex-
ternal magnetic field. However, the orientation of these magnetic moments is
randomized, such that the material does not possess a net macroscopic mag-
netization. These magnetic moments are free to rotate, and when aligned by
an external field, the effect is called paramagnetism. The magnetic dipole of
each atom is individually affected and brought into line with an external field,
and there is no interaction between adjacent dipoles. Both diamagnetic and
paramagnetic materials are considered non-magnetic since an external field is
required for them to show their magnetic behavior.

Ferromagnetism
The characteristic of ferromagnetism is when a material on a macroscopic
scale can possess a net magnetization in the absence of an applied magnetic
field. Ferromagnetism is probably the most well known magnetic material
classification since its properties are easily observable without any measure-
ment equipment. The main contribution to the great permanent magnetic mo-
ment of ferromagnetic materials is the uncanceled spin contributions, as a con-
sequence of the beneficial electron structure. There is an orbital magnetic mo-
ment contribution, but it is small in comparison to the magnetic moment from
the electron spin. Further, there is a coupling factor between the atoms mag-
netic moments which allow for the alignment of the magnetic moments. This
alignment exists over relatively large areas; these regions where the magnetic
moment is aligned in the same direction are called magnetic domains.
The maximum magnetization possible, or saturation magnetization (Ms ), is
achieved when the alignment of all magnetic domains is in the same direction.
The saturation magnetization is the product of the magnetic moment of each
atom and the number of atoms present in the structure.

2.1.2 Magnetism on a macroscopic level


As briefly described in the previous section, the resulting magnetization field,
M, from all magnetic moments in a material is the sum of contributions from
all magnetic dipoles of the structure. Since each atom has a net current due to
the moving and spinning of its electrons, the current from all the atoms in a
limited space can be considered a current density. This current density can be
determined by
Jm = ∇ × M, (2.1)
where, Jm is the magnetization current density. B is determined by current
densities of all kinds, which at low frequencies (neglecting the displacement
current density) is defined by Ampere-Maxwell’s law as
∇ × B = µ0 J = µ0 (J f + Jm ), (2.2)

18
where, µ0 is the permeability and J f is the free current density. Combining
2.2 with the constitutive relation
B = µ0 (H + M), (2.3)
the following is obtained
∇×H = Jf . (2.4)
Which is the Ampere-Maxwell equation for the magnetic field. Equation 2.3
is often used to model the properties of PMs in FEM simulations. Further, M
is proportional to H as
M = χm H, (2.5)
where, χm is called the susceptibility. The susceptibility can also be written as

χm = µr − 1, (2.6)
where, µr is the relative permeability of the material. Combining Equation 2.3,
2.5 and 2.6 one eventually arrives at
B = µ0 µr H (2.7)
which is another constitutive relation, often used to describe soft magnetic
materials such as electrical steel for stator and transformer cores. Please note
that the relative permeability is not necessarily a constant.
The properties of ferromagnetic materials can vary significantly between
different materials, which make some of them ideal for electric power appli-
cations. Significant for ferromagnetic materials, compared to diamagnetic and
paramagnetic materials is their highly non-linear relative permeability. The
relative permeability and the ability to be permanently magnetized depend on
several factors, e.g. material composition, temperature and the previous value
of the applied magnetic field. The non-linearity of µr becomes clearly visible
in a B-H-curve. Ferromagnetic materials are usually characterized by their sat-
uration B-H-curve since much information can be extracted from this graph.
A generic B-H-curve is shown in Figure 2.1, where the slope of the curve at
any point is µ0 µr . The dashed part is the initial magnetization curve, i.e. the
curve of a previously non-magnetized material. When the applied magnetic
field becomes high enough the material becomes saturated. When the applied
magnetic field is removed, some magnetic domains will not rotate back, and
a resulting B will be left in the material. The remaining B is called the rema-
nence (Br ). For a ferromagnetic material subjected to a fluctuating magnetic
field like in Figure 2.1 the hysteresis effect, or hysteresis loop, is clearly vis-
ible. As can be seen in the figure, it is not sufficient to remove the applied
filed to reduce B inside the material to zero. A field in the opposite direction
is needed and the field strength needed is called the coercivity (Hc ). The area
within the hysteresis loop is a measure of how much extra energy is required to
complete a full revolution of the B-H-curve, i.e. the hysteresis losses. Losses

19
B(T )
Bs

Br

Hc
H(A/m)

Figure 2.1. A generic B-H-curve, or hysteresis loop, for a ferromagnetic material. The
initial magnetization line is dashed starting at origo.

will not be discussed further in the theory section. If the reader has a special
interest in this area, please refer to any textbook on the subject.

2.1.3 Soft ferromagnetic materials


Materials which transport the magnetic field in its path, analogous to the wires
in an electric circuit, are soft ferromagnetic materials. The main parts of an
electrical machine with this task are the stator and the rotor core. The magnetic
field in the rotor will be more or less static, or with only small variations. In the
stator on the other hand, especially in the stator teeth, the field fluctuates from
one direction to the other with the same frequency as the machine’s electrical
frequency ( fel ). Preferably, Hc and Br should have values as close to zero
as possible, since small values decrease the area of the hysteresis loop and
therefore decrease the hysteresis losses. The saturation of the magnetic flux
density (Bs ) is preferably high to allow for higher flux density in a smaller
space.

2.1.4 Hard ferromagnetic materials


Permanent magnets are considered hard ferromagnetic materials due to their
high values of Hc and Br . As previously mentioned, the coercivity is a mea-
sure of the magnitude of an external field required to reduce the magnetic flux
density within the material to zero. However, depending on the quality of the
permanent magnet material, this field strength does not necessarily reduce the
magnetization at all. The magnetic field strength needed to reduce the mag-
netization to zero is called the intrinsic coercivity (HcJ ). To visualize this in a
B-H-curve the magnetic polarization or intensity of magnetization (Im =µ0 M)
is plotted against H, which is the same as B-µ0 H vs. H. This type of plot is

20
1.5
Intrinsic 1
0.5

B or Im (T)
HcJ Normal
0
Hc
−0.5
−1
−1.5
−1.4 −1.2 −1 −0.8 −0.6 −0.4 −0.2 0
H (A/m) ·106
Figure 2.2. Example of a normal and an intrinsic B-H-curve for a permanent magnet.

usually called an intrinsic B-H-curve. An example of a normal and intrinsic


B-H-curve for a hard ferromagnetic material is shown in Figure 2.2.
All materials have some property which is temperature dependent and mag-
netic materials are no exception. The temperature coefficient for the rema-
nence (kBr ) is generally negative, i.e. Br decreases when increasing the abso-
lute temperature (T ). The temperature coefficient for the intrinsic coercivity
(kHcJ ) depends on the material. Ferrite permanent magnets have a positive
temperature coefficient, making them harder to demagnetize with increasing
temperature, whereas the opposite is true for other permanent magnet materi-
als [49].

2.2 The finite element method


One purpose of using finite element analysis on electrical machines is to cal-
culate the magnetic field distribution inside the machine. This distribution can
be used to calculate other properties such as the induced voltage of the ma-
chine. For symmetry reasons, it is usually sufficient to model a limited part of
the machine. Depending on the size of the machine and the desired accuracy
of the study, it is in many cases adequate to reduce the problem from a three
dimensional (3D) to a two dimensional (2D) problem. An example machine
of such a case is shown in Figure 2.3, which is the 12 kW permanent mag-
net synchronous generator (PMSG) studied in Paper III and V where only the
smallest symmetrical part is modeled. Solving a 3D problem in 2D has some
drawbacks, 3D effects such as eddy currents and end effects of the machine
can not be modeled with full accuracy.

21
Air
FeSi
Copper
Aluminium
Permanent magnet
Figure 2.3. 2D representation of the smallest symmetrical piece of the 12 kW PMSG.

2.2.1 Derivation of the transient field equation


Time-dependent magnetic problems can be described with the following dif-
ferential equations
∇·B = 0 (2.8)

∂B
∇×E = − (2.9)
∂t

∂D
∇×H = Jf + (2.10)
∂t
where, E is the electric field, and D is the electric displacement field. As be-
fore, if the frequencies of the source currents are low, the displacement current
density (−∂ D/∂t) is close to zero and can, therefore, be neglected and Equa-
tion 2.10 becomes the same as Equation 2.4. B and H are related through the
constitutive relation in Equation 2.7. From definition, we know that

B = ∇×A (2.11)

where, A is called the magnetic vector potential. Inserting Equation 2.11 into
Equation 2.9 and performing some vector algebra one arrives at

∂A
E=− − ∇φ (2.12)
∂t
where, φ is the electric potential. We also know that

Jf = σE (2.13)

22
where, σ is the electric conductivity. Combining Equation 2.7 with 2.10-2.13,
and letting ∇ · A = 0, one should arrive at
∂A
∇ × (µ0 µr )−1 ∇ × A = −σ − σ ∇φ . (2.14)
∂t
Assuming µr is constant, applying some vector algebra onto Equation 2.14,
we obtain
∂A
σ − (µ0 µr )−1 ∇2 A = −σ ∇φ , (2.15)
∂t
which is the equation solved in a FEM software for time-dependent magnetic
problems. For magneto-static problems all time-dependent terms are removed
and Equation 2.15 is reduced to

−(µ0 µr )−1 ∇2 A = −σ ∇φ . (2.16)

The assumption that µr is (piecewise) constant implies that hysteresis effects


are neglected.

2.2.2 Boundary conditions


In order for the field problem to be fully defined, boundary conditions have
to be defined on the boundaries of the geometry. Three different boundary
conditions have been used, which all will be given a short explanation assum-
ing simulation in 2D. The first, set on the inner and outer boundaries of the
example geometry is the homogeneous Dirichlet boundary condition

Az = 0 (2.17)

where, Az is the magnetic vector potential, z-component. A homogeneous


Dirichlet boundary condition is equivalent to an insulating boundary condi-
tion, e.g. no flux is allowed through the boundary.
In a perfectly round and concentric electrical machine, the geometry ex-
hibits a consistent feature. The size of the simulated geometry can be reduced
to the smallest repeatable part by including periodic continuity boundary con-
ditions
Asrc
z = Az
dest
(2.18)
where, Asrc
z and Az
dest represents the z-component of the magnetic vector po-

tential on the source and destination boundaries, respectively. The periodic


boundary condition relates the value of Az on two equivalent boundaries to
each other, i.e. the flux passing through the source boundary will come back
through the destination boundary.
The last boundary condition used is a sector symmetry boundary condition
or moving boundary condition, which is set on the boundary that represents
the air gap between the rotor and the stator. The sector symmetry boundary

23
Figure 2.4. The solid, dashed and dotted lines represent sector symmetry, homoge-
neous Dirichlet and periodic boundary conditions, respectively. Paper V

condition works in a similar way as the periodic boundary condition with the
addition of allowing movement (without re-meshing) along the boundary. In a
circular machine, the periodic boundary condition would be set on boundaries
with a surface normal parallel to the ϕ̂ in a standard cylindrical coordinate sys-
tem and the sector symmetry boundary condition would be set on boundaries
with a surface normal unit vector parallel to r̂.

2.2.3 Modeling of the B-H-curve and the recoil line


To successfully model the demagnetization of permanent magnets a B and
H relationship is required. Depending on how the permanent magnets are
modeled, M and the constitutive relation in Equation 2.3 might be needed as
well. Ideally, the B-H relationship is provided by the manufacturer. If not, a
B-H-curve can be modeled with an analytical function [50]

B = Br + µ0 µr H + EeK1 (K2 +H) (2.19)

where, E is a unit conversion factor of 1 T. The parameter K2 is calculated


from
ln (Br + (µr − 1) · µ0 · HcJ ) · E −1
 
K2 = − HcJ (2.20)
K1
and the K1 parameter sets the shape of the knee point in the B-H-curve, where
a larger value gives a sharper knee. The author found good agreement with
several Nd-Fe-B magnet grades with a K1 value of −1.5 · 10−4 m/A. The
magnetization can then be calculated with Equation 2.3.

24
1.5
Br
Straight •
Exponential 1

Magnetic flux density (T)


0.5

• 0
Hc

−0.5

−1
•H
cJ
−1.5
−1 −0.8 −0.6 −0.4 −0.2 0
Magnetic field (A/m) ·106
Figure 2.5. Comparison of the exponential model and the straight line model for
reproducing B-H-curves.

An alternative and more straight forward way of modeling a B-H-curve are


by using two straight lines. This approach has been tested against the method
presented above with good results [51], at least for permanent magnets with
very sharp knees, which many Nd-Fe-B permanent magnets have. With this
method a straight line with the slope

k = µ0 µr (2.21)

is drawn from the Br . The second line is a straight line through Hc and HcJ that
will coincide with the first line at the knee. A comparison of the two methods
is presented in Figure 2.5.
Since manufacturers often provide the temperature coefficient for the rema-
nence and the coercivity, it is easy to reproduce a B-H-curve for any tempera-
ture using

Br (T ) = Br (Tre f )(1 + kBr · (T − Tre f )), (2.22)


HcJ (T ) = HcJ (Tre f )(1 + kHcJ · (T − Tre f )) (2.23)

where, Tre f is the listed reference temperature for the material properties of
the permanent magnet.
When demagnetization has occurred and the applied magnetic field is re-
moved, the recoil line will bend somewhat upwards when the applied field
approaches zero [33, 50]. However, this is often neglected to simplify analy-
sis [52]. By doing so, the recoil line can be approximated by a straight line

25
with the same slope as the initial part of the B-H-curve, as in Equation 2.21.
The new remanence (Bnew
r ) can, therefore, be expressed as

Bnew
r = Bmin − kHmin (2.24)

where, Bmin and Hmin are the minimum magnetic flux density and magnetic
field, respectively. The recoil will therefore follow the line

B = k(H − Hmin ) + Bmin . (2.25)

26
3. Methods

This chapter will cover the methods used and the demagnetization model de-
veloped during this project. The generators in the case studies was of a similar
kind. All of them are modified versions of the 12 kW lab generator in Fig-
ure 2.3 i.e. slow rotating multi-pole machines with relatively low load angle,
and with surface mounted PMs, except for one which had a rotor with buried
ferrite magnets. The chapter will begin with a section about the simulation
software used in all papers.

3.1 Simulating with COMSOL Multiphysics


The following section will briefly describe the structure of the FEM software
COMSOL Multiphysics 1 R
(CM). The software includes several physics mod-
ules; each module provides a relevant set of options regarding loads, boundary
condition and initial values for that specific physics application. In this way,
the user can add all the relevant physics modules needed, and skip those which
are not necessary or considered negligible. These physics modules can, if de-
sired, be set-up to be coupled to one another. For example, leading a current
through an electrical conductor will generate heat which not will be consid-
ered if only the module for electric currents is used. However, when the heat
transfer module is added, the resistivity of the conducting material can be cou-
pled to the temperature, and now the resistivity will change depending on the
temperature of the conducting material. Further, in many cases, there are mod-
ules which are very similar. When simulating magnetic fields, for example, the
base module is called magnetic fields (mf). With the mf module, the user can
perform a vast variety of simulations on magnetic fields. However, there is
another module called rotating machinery, magnetic (rmm), this module has
some preset properties for, as the name suggests, simulating rotating electrical
machines. Everything that can be done in rmm can also be accomplished with
mf, but the user has to do more work to set-up the model properly.
In this Ph.D. project, the different modules for magnetic fields have primar-
ily been used and one for electrical circuits. The electrical circuits module is a
zero dimensional model where only electrical circuits are modeled, much like
the Simulink R
package for MATLAB 2 R
.
1 COMSOL Multiphysics is a registered trademark of COMSOL AB.
2 MATLAB and Simulink are registered trademarks of The MathWorks, Inc.

27
3.2 Development of demagnetization model
This section will cover the work put into developing the demagnetization
model and a few sidetracks which were investigated to get the model to work.
The author has always had an idea that the finished model should be easy to
use and implement in other models by others than the author.
In the study performed in Paper V, the model was only capable of modeling
the B-H-curve of a PM using the exponential model described in Section 2.2.3.
Reproducing the exponential model was done by reproducing the B-H-curve
with an M(B) interpolation function in MATLAB which was exported and
served as a lookup table in CM. The demagnetization analysis was made man-
ually by finding the areas with the lowest magnetic flux density parallel to the
magnetization direction (B|| ). Only B|| was considered because the influence of
the magnetic field’s inclination angle on the demagnetization is hard to estab-
lish, as discussed in Section 1. In [33], where Ruoho presented his empirical
demagnetization model for axially pressed Nd-Fe-B PMs, he also performed a
study comparing the inclination model with the exponential model from [50].
The result from [50] showed that the machine geometry has a strong influ-
ence on the magnetic field distribution and therefore a field inclination model
might not always be necessary. For a machine with surface mounted PMs the
demagnetization results were identical between the inclined modeled and the
exponential model. Therefore, since the initial plan in this work was to study
surface mounted PM synchronous generators the initial choice was only to
consider B|| . Further, at this point it was not obvious how to implement such
inclination features in the FEM software, since the model at this stage used
an interpolation function to calculate the magnetization. The model could be
used in time-dependent simulations but the magnetization would return to its
original value if the opposing field was removed.
After Paper V was submitted, a series of ideas were evaluated to try to
get the model to store the current magnetization. Various ways of doing
this within CM was assessed but with no success. Eventually, a persistent
MATLAB-function was linked and called by the FEM software. Variables
within MATLAB-functions are usually wiped from memory after the function
is completed but persistent functions let the user access the variables from the
previous function call, i.e., the magnetization could be stored for the next it-
eration of the solver and could be lowered if demagnetization occurred. A
proof of concept was made, dividing a PM into three parts and letting a persis-
tent function store the magnetization for each of these domains. The proof of
concept model produced satisfactory results but at rather slow computational
speeds. After the initial success, the problems started. It was discovered (the
hard way) that the connection to MATLAB was very slow, and increasing the
amount of domains inside the permanent magnet to any reasonable amount
would simply not work or even get the simulation to start. At this stage, the
model only consisted of a test geometry with a single magnet. Efforts were

28
made trying to reduce the number of domains in areas that were unlikely to be
demagnetized. Results progressed somewhat but never got to a stage where
they were considered suitable for further use.
During what turned out to be a sidestep with the persistent MATLAB func-
tion, an experiment was planned to compare the simulation model to reality.
All details about the study can be found in Paper IV and will be further dis-
cussed in Section 3.3. Small changes were made to the simulation model
for this study; the principal difference was the introduction of the recoil line,
also described in Section 2.2.3. The implementation of the recoil line was
not perfect; it did, however, open up for the ability to do a second simula-
tion to investigate how the permanent magnets would perform after they had
been partially demagnetized. The implementation calculated the remanence at
each mesh node, and used the values to model an ideal permanent magnet in
the following simulations. This limits all further simulations to the linear part
of the B-H-curve which excludes the possibility to study multiple fault con-
ditions. To get the same functionality as before demagnetization, each mesh
node value would have to be used to generate a B-H-curve for each mesh node
in the magnet. Applying a value with a variable to a particular mesh node is
to the author’s knowledge not possible in CM. This version of the model was
also used in the comparison study on the demagnetization risk in a ferrite and
Nd-Fe-B rotor with identical stators which can be read in full text in Paper III
and summarized in Section 3.6.
In parallel with the experimental study, a temperature dependence was im-
plemented. Instead of creating an interpolation curve in MATLAB, as done
before, all relations were set-up within the FEM software. The Br and the
HcJ were previously set to a static value corresponding to a predefined tem-
perature and PM grade. Now Equation 2.22 and 2.23 was implemented and
coupled to the absolute temperature in the heat-transfer module in the FEM
software. This implementation enabled studies with temperature distributions
within the PMs. Unfortunately, no obvious way of evaluating the results was
determined, and no paper was published with this feature implemented.
The recoil part of the demagnetization models presented up until this point
are not coupled to the field equation, 2.15, in any way. The demagnetization
has been checked after each simulation and can rather be considered as post
processing of the results. However, the B-H relationship is coupled to Equa-
tion 2.15 through a constitutive relation. For the most parts of the machine,
the constitutive relation in Equation 2.7 were used. The PMs was implemented
with either Equation 2.3 or
B = µ0 µr H + Br , (3.1)
since the B-H relationship was expressed as either a M(B) or Br (B) relation-
ship, as mentioned in Section 3.2. By changing constitutive relation, the field
equation will be somewhat altered as well. Equation 2.15 is derived using the
constitutive relation in Equation 2.7. By implementing Equation 2.3 the field

29
equation would be

∂A
σ − µ0 −1 ∇2 A = −σ ∇φ + Jm . (3.2)
∂t

The latest version of the demagnetization model is very closely related to


the persistent MATLAB function discussed as a sidestep above. The rea-
son it works better this time around is that CM has implemented support for
dynamic-link libraries (DLL), also known as .dll-files. DLL-files are compiled
function files, a C-function in this case, executed in the memory space of the
calling process which means that there is little overhead. This process can
be compared to the calling process with the MATLAB persistence-function
method. In that case, when the function is initiated a link is established be-
tween CM and MATLAB, CM sends the information to MATLAB where the
function is run, then the result is sent back to CM. The function uses a straight
(or ideal) B-H-curve, where the knee is modeled with two straight lines as
described in Section 2.2.3. The basic structure of the model is presented as a
flowchart in Figure 3.1.
By accessing data from the previous iteration, time-dependent studies are
possible, and this model was used in the study on multiple short-circuit events
in Paper I which is discussed further below in Section 3.4. This latest version
of the demagnetization model will from here on be referred to as the external
model.

Alternative model
At more or less the same time as the development of the external model began,
a parallel development project began as well. It was discovered that CM’s or-
dinary differential equation (ODE) model could be used to more than solving
ODEs. It was possible to use the solution of the ODE model (u) to store the
current Br of each node in the PMs and access it in the next time-step. The
procedure was similar to the external model, to begin with, the initial value of
u was set to the desired value. The main difference was the demagnetization
check; it uses the same method but the process does not affect the Jacobian
(the system matrix), and it uses the solution from the previous time-step. This
procedure will be faster than the external model since the demagnetization
will not be considered in each iteration, but only in the first iteration of each
time-step. However, this implementation requires more hands-on attention
and lacks the same structural overview as the external model.
This alternative model uses only internal functions of CM and will from
here on be referred to as the internal model.

30
Input initial
Input state variables
values from CM
(Brprev , Byprev , Hyprev )
(Br , Hc , HcJ and µr )
B input from CM

Calculate B-H-curve

Next iteration
yes Update state
Demagnetization?
variable Brprev
no
Update state variable
Byprev , Hyprev

Write Jacobian

Return Jacobian no
Convergence?
and H to CM
yes
Done!

Figure 3.1. Flowchart describing the basic structure of how the time-dependent de-
magnetization model works.

3.3 Experimental verification of stationary


demagnetization model
To evaluate the stationary demagnetization model with the implemented recoil
line an experimental set-up was built which will be further discussed in this
section. The study is fully described in Paper IV.
In this study all simulations were performed in 3D since the dimensions
were not favorable in respect of neglecting 3D-effects. 2D simulations were
evaluated, but the error was too large. The main 3D-effect present in this
study was the leakage flux around the air gap and the PM which can not be
fully evaluated in 2D.

3.3.1 Experimental Set-up


The base of the experimental set-up was an iron core consisting of a rectangu-
lar iron core with the side lengths 400 and 550 mm respectively. A schematic
drawing is shown in Figure 3.2; all dimensions are in mm, and the out of
plane thickness is 85 mm. The set-up was constructed from 1 mm sheets

31
17 95
300
550
4xR5

95
95 65
400
Figure 3.2. Schematic drawing of the iron core used in the experiments. All dimen-
sions are in mm and the out of plane thickness is 85 mm. Adapted version from
Paper IV.

of M700-100A laminated steel. The sample PM was placed in the 17 mm


opening on the right leg of the iron core. Below the opening a 423 turn coil
was wound with a 6 mm2 cable. The coil was connected to a power supply
(EA-PS-8160-170) capable of delivering a direct-current (DC) of 170 A. A
photo of the iron core is shown in Figure 3.3.
The PMs used during the experiments were of the Samarium-Cobalt (Sm-
Co) kind with a composition of Sm2 Co17 . The grade used was SM30L3 and
it is presented in detail in Table 3.1. Unfortunately, no value of the μr was
found, a value of 1 was therefore used in the simulations. The dimensions
of the PMs were 74x54x14 mm with the magnetization direction along the
14 mm direction. Given these PM dimensions, there is a resulting air gap of
3 mm in the iron core.
Table 3.1. Typical values of the Sm-Co grade SM30L at 20◦ C, provided by the
supplier3 . Paper IV.
Grade Br (T) Hc (kA/m) HcJ (kA/m) BHmax (kJ/m3 )
SM30L 1.08–1.12 -557–(-477) -716–(-493) 223–247

3.3.2 Measuring the magnetic flux density


The magnetic flux density in the air gap was measured using a direction sen-
sitive hand-held Gaussmeter (F.W. Bell 5180). To get reproducibility of the
3 Sura Magnets AB, http://www.suramagnets.se/, retrieved November 5th, 2013

32
Figure 3.3. A picture of the iron core used in the experiments to verify the stationary
demagnetization model.

results a measuring probe holder was 3D printed. The probe holder had ten
slots where the measuring probe would be placed during measurements which
also ensured that only the B|| was measured. The slots were distributed along
a line perpendicular to the air gap on the 54 mm side of the PM. The cor-
ners of the probe holder were extended to ensure a tight fit around the PM. In
this way, the measuring probe holder was placed in the middle of the air gap,
and in the same position relative to the PM in each measurement and a high
reproducibility could therefore be achieved.
A control measurement of the iron properties was performed in the iron
core measuring B as a function of the current (I) in the air gap of the iron
core. The measurement was performed 2 mm from the lower side of the air
gap in the middle of the structure. The measurements were then compared to
simulations with the B-H-curve provided by the manufacturer.
The PM properties were presented in a performance interval as can be seen
in Table 3.1; the actual value should be somewhere in those ranges. It is not
certain that all PMs are of the same quality, some could be near the upper
limits and some closer to the lower. For this reason, all PMs where control
measured and the simulation model was calibrated to each one.

Remanence
The remanence of each PM was approximated by measuring B|| in the air gap
with the measurement probe holder as described above. The remanence in
the simulation model was then adjusted until similar results were achieved.
The simulation model had previously been compared to experiments where
all parameters were known with good agreement.

33
Coercivity and intrinsic coercivity
To approximate/calibrate the knee point shape parameter K1 , Hc , and HcJ , a
PM was placed in the iron core and B|| was measured as a function of I. The
same was done in the simulations where the PM parameters were adjusted
until a good agreement was achieved.

3.3.3 Experiments
For each magnet in the study, B|| was measured in all ten positions when placed
in the iron core. After that, a current of 30 A was led through the coil, and
when the current had been returned to zero, B|| was measured again at all ten
positions. The magnetic flux density was measured four times at each position,
both before and after the current pulse, and an average of these measurements
was used.

3.4 Evaluation of the two time-dependent


demagnetization models
The idea of how to evaluate the latest version of the two demagnetization mod-
els was to simulate multiple fault conditions for a generic generator. These
fault conditions would represent a few of the possible faults that could occur
during a generator’s lifespan. The generator used in this study was a generic
modification of the 12 kW DDPMSG studied in Paper III and V. The changes
consisted of reducing the air gap from 10 to 3 mm and decreasing the PM
height from 14 to 5 mm. These changes resulted in an increase in output
power, from 12 to 20 kW. Further, the operational temperature was assumed
to be 60◦ C to make the PM more susceptible to demagnetization. The main
characteristics of the generator are presented in Table 3.2. Not only the de-
magnetization was studied, but the basic overall condition of the machine was
monitored as well, including the torque and the induced voltage.

3.4.1 Electrical circuit


The electrical circuit consisted of a full wave three phase diode rectifying
bridge connected to the generator output with a resistive load (RL ) and a ca-
pacitor (CDC ) on the DC side, as shown in Figure 3.4. The capacitor was added
to lower the voltage ripple. Ea , Eb , Ec are the generator electromotive forces
and va , vb , vc are the generator output line-neutral voltages of phase a, b and c,
respectively. L and R are the generator inductance and the internal resistance
per phase, respectively. Further, there are three resistors denoted Rab , Rbc , and
RDC which were used as switches in the short-circuit simulations. The com-
ponents in the generator block of Figure 3.4 were not included in the circuit

34
Table 3.2. The main properties of the 20 kW generator used in the multiple short-
circuit events study. Paper I.
Power 20 kW
Current, DC-load, rms 50 A
Voltage, DC-load, rms 400 V
Phase voltage, no-load, rms 189.4 V
Phase voltage, no-load, peak 262.2 V
Rotational speed 127 rpm
Electrical frequency 33.9 Hz
Air gap 3 mm
Permanent magnet height 5 mm
Machine length 224 mm
Stator radius (inner/outer) 380/437 mm

model; these components were automatically calculated by CM’s magnetic


field model. CDC had a capacitance of 0.02 F and RL had a resistance of 8 Ω.
To simplify the modeling of the diodes in CM, they were replaced by voltage-
controlled current sources (VCCSs) and snubber resistances (Rs ), as shown in
Figure 3.5. The VCCSs were controlled using the diode law
 V  
D
ID = IS e nVT
−1 (3.3)

where; ID is the diode current, VD is the voltage across the diode, IS is the
reverse bias saturation current, n is the ideality factor (1.25 in this case), and
VT is the thermal voltage
kB T
VT = (3.4)
q
where; kB is the Boltzmann constant and q is the magnitude of the charge of
an electron. The snubber resistance was added for one reason; in the FEM
software, VD had to be measured over Rs not to cause a circular dependency.
As a compromise between the leakage current through Rs and the stability of
the simulation, Rs was set to 10 kΩ.

3.4.2 Transient analysis


To test the external and internal demagnetization models a series of short-
circuits was simulated on the terminal side of the electrical circuit, using the
three resistors mentioned earlier Rab , Rbc , and RDC . The short-circuits are de-
noted with capitalized letters for the phases with a dash (-) in between, e.g. a
short-circuit between phase a and b is denoted A-B. A series of five consecu-
tive short-circuits was executed in the following order: B-C, A-B, B-C again,
A-B-C and the last over the DC-load. At the event of the short-circuit, the
resistance of the relevant resistor(s) was set to zero.

35
Generator
Ea L R
va

Eb L R Rab
vb
Cdc RDC RL
Ec L R Rbc
vc

Figure 3.4. A schematic figure of the DC rectification circuit. In the simulations each
diode was substituted for a voltage-controlled current source, VCCS, and a snubber
resistance, Rs , see Figure 3.5. Paper I.

ID Rs VD

Figure 3.5. The VCCS and Rs which replaced each diode in the DC-circuit of Fig-
ure 3.4.

Since the short-circuit currents depend on the relative position of the rotor
and the stator, the short-circuits were initiated at the same relative position
for each fault condition. The duration of each short-circuit was two electrical
periods. The model was simplified a bit by assuming that there was an unlim-
ited supply of driving torque, i.e. the generator rotated with a constant angular
velocity throughout the simulation.
After each fault, the generator was given some time to reach a new steady
state of operation. The capacitor would partially/fully discharge during the
fault conditions and needed to be recharged before the next short-circuit could
be initiated to achieve satisfactory results.
If demagnetization occurs causing a decrease in voltage, there are three
actions one can take. The first is to replace the PMs to restore the machines
fully potential, this alternative is usually very expensive. The other alternative
is to increase the current density in the windings to maintain the same output
power. This alternative will, however, result in losses and higher temperatures
if the cooling system can not handle it. The last option is to let the machine
operate at its new lower capacity and accept the loss in output power. The last
option was used when evaluating the two time-dependent models. To compare

36
the severity of the different short-circuit events, all cases were also tested on a
healthy machine.

3.4.3 Comparison with stationary model


To further evaluate the models, they were compared to the stationary model
which was verified with experiments in Paper IV and Section 3.3. The com-
parison was performed in a three-legged iron core, where the middle leg had a
1 mm air gap separating a permanent magnet and a coil of 100 turns. The am-
bient temperature was set to 20◦ C. A series of current pulses with a maximum
peak value of 60 A and a duration of 1 s was led through the coil.

3.5 Ripple in surface mounted permanent magnets


When passively rectifying a generator output voltage using diodes, there are
harmonics introduced into the current. These current fluctuations will in their
turn cause a ripple in the magnetic field. Therefore, the PMs of a PMSG will
be subjected to different magnetic loading depending not only on the electrical
loading but also the type of electrical system the generator is connected to. The
difference in working point and in particular the ripple of the working point
for the PMs in a PMSG was studied in Paper II, for three different load cases;
no-load, a purely resistive Y-connected alternating current (AC)-load, and a
diode rectified DC-load (the same circuit as in Figure 3.4). The magnetic flux
density was recorded in three positions close to the surface of the PM, shown
in Figure 3.6. The average magnetic flux density of the PM was also recorded.
The generator in this study was a 100 kW generic modification of the 12 kW
PMSG lab generator. The loads in the different load cases were adjusted until
the same output power of 100±0.1 kW was achieved from the generator. For
more details about the study, please refer to Paper II.

3.6 Comparison of ferrite and Nd-Fe-B rotor


In the aftermath of the sudden Nd-Fe-B PM price peak at the end of 2011 and
the growing debate about the environmental impact of the mining for the rare
earth metals [53, 54] used in Nd-Fe-B PMs, the idea of building a generator
with an interchangeable rotor took form [55]. The main idea was to be able
to have a generator with as similar properties as possible with both a Nd-
Fe-B and a ferrite rotor, and depending on the current price levels building
the most profitable one. There are some challenges with this; ferrites have
approximately one-third of the remanence and about one-tenth of the energy
product compared to Nd-Fe-B PMs. Therefore, about ten times more PM-
material is needed to generate the same magnetic flux density in the air gap.

37
A

Figure 3.6. Cross section of one-fourth of the geometry simulated in Paper II. The
positions A, B, and C are positioned along a line 1 mm below the surface of the
magnet. A and B are each 1 mm from the respective side, and B is placed in the
middle of the permanent magnet. The rotational direction of the rotor is clockwise.

The 12 kW generator in Paper V was used as a reference to the new ferrite


rotor generator and is further described in [56, 57]. The design procedure of
the ferrite rotor is described in Paper VI.
To further analyse and compare the two generators, the short-circuit behav-
ior and demagnetization risk under different conditions were analysed for the
two generators. Since the mass and the mass moment of inertia are very differ-
ent for the two machines, the short-circuit simulations were also modified to
include the influence of the mass moment of inertia. To expand the study even
further, the mass moment of inertia of a wind turbine was introduced into the
problem to compare how the generators behaved when connected to a wind
turbine and when the turbine was not considered.

38
4. Summary of results and discussion

4.1 Experimental verification


The results of the experimental verification study were overall satisfactory de-
spite some troubles with the PM samples and the set-up, as discussed below.
The results in this section are based on a total of three PM samples. Mea-
surement and simulation results from two samples are plotted in Figure 4.1.
The maximum deviation of B|| in the simulation results compared to the ex-
perimental results before demagnetization was 7.9%. After demagnetization
the same number was 17.8%. However, neglecting the two outer points the
maximum deviation between simulations and experiments drops down to 3%.
The cause for the bigger deviation of the results close to the edge could be a
consequence of the model only considering the magnetic flux parallel to the
magnetization direction. Figure 4.2 shows an example of the resolution that
could be achieved in the demagnetization simulations. The figure shows, an
example of, the demagnetization after a phase to phase short-circuit in the
12 kW PMSG studied in Paper V.
The results are only based three samples due to the poor sample quality.
During calibration of the PMs, it was discovered that the PMs differed very
much in quality. Some samples had very low Br , and some were not evenly

Sim. before Sim. after Exp. before Exp. after

1 1
Magnetic flux density (T)

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Position (mm) Position (mm)
Figure 4.1. The magnetic flux density as a function of the position in the PMs for two
samples. Paper IV.

39
100

90

% of Br
80

70
Figure 4.2. Example of the resolution of the demagnetization in a permanent magnet.
The generator in the figure is the 12 kW machine studied in Paper V.

magnetized. Therefore, the decision was made to only include the PMs with a
rather high and even magnetization. The values of Hc and HcJ was not within
the limits of Table 3.1. It seemed that the percentage difference for the Br was
about the same as for Hc and HcJ compared to the calibrated sample. During
the experiments it was therefore assumed that the percentage difference of Br
compared to the calibrated sample was of the same size for Hc and HcJ , i.e.
if Br was 5% lower than the calibrated sample the values of Hc and HcJ were
assumed to be 5% lower as well.

Iron core measurements


While analyzing and measuring the iron core properties, the experimental re-
sults did not coincide with the simulated results. The simulation results under-
estimated B at low currents and overestimated B at high currents. Due to these
results, some leftover pieces of the iron core material were control measured
at the manufacturer using an Epstein frame. Using the resulting B-H-curve
from that measurement in the simulations improved the results, but they were
still not perfect.
The non-perfect correlation between the simulated and experimental results
could be caused by the orientation of the steel’s rolling direction (RD) in the
experimental set-up. Steel has different properties in the RD, and the trans-
verse direction (TD) and the magnetic properties are usually better in the RD.
When measuring in an Epstein frame equal number of steel pieces oriented in
the RD and the TD are used. Measurements, therefore, result in a B-H-curve
which corresponds to an average of the two directions. In the experimental
set-up the RD was in the horizontal direction of Figure 3.2, i.e. the magnetic
field has to travel further in the TD than in the RD. At the same time, the
magnetic field in RD parts would be higher than what the simulation results

40
show. The effect of the rolling direction orientation could explain a part of the
deviation between the simulated and experimental results.

4.2 Time-dependent models


Results from the transient analysis of the two time-dependent demagnetiza-
tion models showed that the power output decreased from the original 20 kW
to 13.5 kW and 13.1 kW due to the demagnetization, for the external and inter-
nal model respectively. The capacitor voltage, which is directly proportional
to the current in the resistive load and hence the delivered power, is plotted
in Figure 4.3. The major part of the demagnetization occurred during the two
first short-circuits. There is a slight voltage drop not visible in Figure 4.3 in
the second to last fault condition, the A-B-C short-circuit. It can also be seen
in Figure 4.3 that the generator was allowed to reach a new steady state of
operation before each fault condition, as discussed earlier. The reason for the
initial increase in capacitor voltage is because the initial value of the capacitor
voltage in the simulation was set to 360 V. A more detailed plot over the short-
circuit consequences is plotted in Figure 4.4, where the average Br in all four
PMs is shown. The smallest repeatable part of this generator was of two pole
pairs, and the geometry was similar to the one in Figure 2.3 in Section 2.2.
With the PMs denoted 1 to 4 from the bottom up. The series of short-circuit
events resulted in an unsymmetrical demagnetization of the PMs as shown in
Figure 4.4. It can be noticed that the average Br is only slightly decreased
(0.8%) in PM 1 while PM 3 experienced a decline of 23.8% in average Br .
The unsymmetrical demagnetization can also be observed when comparing
the no-load voltage before and after the short-circuit events, which are plotted
in Figure 4.5. The unsymmetrical no-load voltage is a direct cause of the
unsymmetrical demagnetization of the PMs. This uneven induced voltage will
also have an influence on the torque. While performing a frequency analysis
of the no-load voltage, an overtone with a frequency of 1.5 fel was found which
is directly coupled to the torque ripple. An increase in torque ripple could lead
to increased vibrations which in turn could result in problems, especially at
lower frequencies when there is less damping by the generator and shaft dy-
namics. Low vibrational frequencies can, therefore, more easily transfer to the
turbine from the shaft of the generator [58]. As mentioned above, the majority
of the demagnetization occurred during the first two short-circuit events. If
the current had been adjusted to maintain the output power, the short-circuit
current would have been higher in the later cases leading to higher demagne-
tization in the end. Since the load was kept constant, the current reduced as a
consequence of the decreased induced voltage. The no-load voltage after the
final demagnetization was 156 V for the external model. This voltage level
is close to the no-load voltage after the worst case on a healthy machine, the
A-B-C fault. The no-load voltage after an A-B-C case on a healthy machine

41
500
Int model
400 Ext model
Voltage (V)

300

200

100

0
0 100 200 300 400 500 600 700 800 900 1000
Time (ms)
Figure 4.3. Voltage over the capacitor during all the successive fault events for both
models. Paper I.

1.3
PM 1
Average remanence (T)

PM 2
1.2 PM 3
PM 4
1.1

0.9
0 100 200 300 400 500 600 700 800 900 1000
Time (ms)
Figure 4.4. The average remanence in each magnet during all short-circuit events for
the external model. Adapted from Paper I.

was 154 V. This suggests that it is not the number of fault conditions that
determines the damage on a machine, but which cases it is exposed too.
Comparing the two time-dependent demagnetization models to the station-
ary model shows that the computational time of the external model is about
four times longer than for the internal model. However, the internal model
tends to overestimate the demagnetization slightly. For detailed results re-
garding the comparison with the stationary model, please refer to Paper I,
more specifically Table 3.

42
300
200
Voltage (V)

100
va
0 vb
vc
−100
−200
−300
0 10 20 30 40 50 60
Figure 4.5. The no-load voltage for the 20 kW generator, before (solid) and after
(dashed) the short-circuit events for the external model. Adapted from Paper I.

4.3 Ripple in permanent magnets


Results from the FEM simulations showed a small difference in working point
between the different load cases, the largest difference was at the corners. In
the no-load case the amplitude of the magnetic flux density at position A and
C were almost the same, as expected. When loading the generator the trailing
corner, A in this case, will experience a rise in the magnetic flux density while
the leading corner, C, will experience a decrease in magnetic flux density. The
ripple of the magnetic flux density at the different positions during the different
load cases is presented in Figure 4.6.
A frequency analysis of the results in Figure 4.6 revealed that the slot har-
monic at 7.5 fel was present in all loading conditions, but mainly at the surface
of the PM. An unexpected harmonic at 1.5 fel occurred, which did not seem to
arise from any harmonic in the electrical current, and was therefore assumed
to be an undertone of the slot harmonics. Small harmonics at 3, 4.5 and (for
the AC-load case) 6 fel were also assumed to be a consequence of the slotting.
The 6 fel harmonic in the DC-load case was significantly higher than in the
AC-load case. The 6 fel harmonic was caused by the current fluctuation during
the passive rectification and can be significant. In the original 12 kW machine,
the fluctuation in electric torque at nominal power was determined to be in the
range of 20-30% [58]. The value obtained for this 100 kW machine was only
6%, and the resulting magnetic field ripple was at most 1.54% for the studied
positions. The reason for the low power fluctuation in the 100 kW machine
was the high current in combination with the large generator inductance.

43
A B C Avg.
1.1 No-load
1
0.9
0.8
0.7
0.6
Magnetic flux density (T)

1.1 AC
1
0.9
0.8
0.7
0.6
1.1 DC
1
0.9
0.8
0.7
0.6
0 20 40 60 80 100 120 140 160 180 200
Time (ms)
Figure 4.6. The ripple of the average of the magnetic flux density, and in the different
points within the PM during no-load, AC-load, and DC-load at 60◦ C. Paper II

4.4 The ferrite alternative


The generator characteristics for the ferrite- and the Nd-Fe-B rotor generators
are found in Table 4.1. The electrical properties are somewhat different since
the induced voltage of the ferrite rotor generator was lower, therefore, the
armature current had to be increased to maintain the same output power. In this
case, this was not a problem since the armature current density was rather low
and increasing it would not dramatically increase the current losses. On the
positive side, the iron losses decreased as a consequence of the lower magnetic
flux density of the ferrite rotor. However, the overall electromagnetic losses
were slightly increased. Further, due to the use of ferrite PMs, the total mass
of the rotor has been increased by more than three times. This increase is not
only due to the increased weight of the PMs but also due to the weight of the
pole shoes and the extra support structure needed.
The basic electrical and mechanical properties of the ferrite rotor design is
rather similar to the Nd-Fe-B design. One big reason for the induced volt-
age not decreasing more was that the air gap width was decreased from 10 to

44
Table 4.1. A comparison of the generator characteristics for the old Nd-Fe-B and new
ferrite design. Values are at rated load and speed. All values are from simulations
except for the weights of the old design. Paper VI
Quantity NdFeB rotor Ferrite rotor
Rated power 12 kW 12 kW
Phase voltage, no load, rms 172 V 146 V
Phase voltage, rms 167 V 141 V
Armature current, rms 23.9 A 28.5 A
Armature current density, rms 1.49 A/mm2 1.78 A/mm2
Amplitude of the fundamental
air gap flux density at no load 0.79 T 0.66 T
Resistive losses 275 W 390 W
Iron losses 254 W 159 W
Electromagnetic efficiency 95.8% 95.6%
Minimum air gap 10 mm 7 mm
Mass of rotor 130 kg 407 kg
Mass of PMs 41 kg 158 kg
Moment of inertia 16.9 kgm2 34.2 kgm2
PM material grade N40 Y40
Remanence of PMs 1.27 T 0.45 T
Maximum energy product 310 kJ/m3 47.6 kJ/m3

7 mm. By doing this, the reluctance in the air gap decreased and in addition
more PM material could be fitted in the rotor. Since the Nd-Fe-B generator
was used for educational and experimental purposes the design had not been
optimized. Therefore, the same similarity of the electrical properties may be
difficult to achieve when converting other, already optimized, Nd-Fe-B ma-
chines. Another criteria is that the machine must have a rather large rotor
radius since so much ferrite PM material is required. The ferrite rotor has now
been built and has replaced the Nd-Fe-B rotor in the laboratory generator.
The results from the comparative study between the Nd-Fe-B and the fer-
rite rotor generators showed that the Nd-Fe-B generator did not suffer from
demagnetization, same as the results from Paper V. The ferrite rotor did suffer
from demagnetization in all tested cases. However, although the minimum B||
was well below the knee point, the induced no-load voltage only decreased
by <1% of all cases. As expected, the simulated short-circuit currents were
higher, and B|| was lower when the wind turbine was connected to the genera-
tors.
Figure 4.7 shows the short-circuit currents for both generators during a
three-phase-neutral fault. The short-circuit occurs after 7 ms, and the dis-
tortion visible for the ferrite rotor is probably due to numerical errors during
the simulation. It can be seen that the short-circuit currents are lower when the
wind turbine is disconnected and that the rotational velocity of the Nd-Fe-B ro-
tor has decreased significantly even in the short simulation span of Figure 4.7.

45
When the wind turbine is connected, the two generators behave similarly ex-
cept that the current is higher in the Nd-Fe-B generator. In the simulations, the
driving torque on the wind turbine was assumed to be constant, i.e. the wind
turbine efficiency was independent of its rotational speed.
The reason the for the low induced voltage drop in the ferrite generator,
although the minimum B|| was well below the knee point, is because of high
but local demagnetization of the ferrite PMs. The PMs were only damaged in
the corners which was somewhat expected due to the softer characteristics of
ferrite PMs as well as the high field concentration in the PM corners in a spoke
type configuration.

800 ia ib ic
NdFeB
400
Current (A)

-400

-800
800 Ferrite
400
Current (A)

-400

-800
0 10 20 30 40 50 60 70 80 90 100
Time (ms)
Figure 4.7. Currents for the Nd-Fe-B and ferrite generator during an A-B-C-N short-
circuit. The solid and dashed lines are with and without turbine, respectively. The
short-circuit occurred at 7 ms. Paper III

46
5. Conclusions

From the results presented in this thesis, it can be concluded that the demag-
netization model works fairly well. The model has evolved throughout the
project and, in its latest form, the model is very easy to implement, only re-
quiring a few steps to set-up in a new simulation model. However, with the
conversion to use a DLL-file for the demagnetization model some features
were left out, such as the exponential model and the temperature dependence.
These features were left out due to the time limitation of the project, but can,
however, easily be added again. The stationary model performed well in com-
parison to experiments. Although the time-dependent model never was com-
pared to experiments, the accuracy is expected to be similar to the stationary
model since their working principle are the same.
A significant improvement of the simulation model was the implementa-
tion of the electrical circuit. The model is capable of simulating the complete
electrical system of a machine.
The conversion from a Nd-Fe-B rotor to a ferrite rotor turned out well,
achieving similar electrical properties in the new machine compared to the
old. The induced voltage did, however, decrease somewhat due to a lower air
gap magnetic flux density. By increasing the armature current, the same out-
put power was maintained. The electromagnetic efficiency did only decrease
slightly due to this change.
The demagnetization comparison between the Nd-Fe-B and the ferrite ro-
tor showed that the corners of the PMs in the ferrite would be demagnetized
during the tested fault conditions. However, the induced voltage would only
decrease less than 1%, according to the simulations. Results also confirmed a
large difference in short-circuit currents when a wind turbine was connected
to the generator, adding a significant amount of mass moment of inertia to the
system. The Nd-Fe-B generator showed no sign of demagnetization during the
simulations with and without the connected wind turbine. It was also shown in
Paper V that changing the mechanical design of the Nd-Fe-B generator would
allow for the PM material to be reduced by more than 50%, maintain the same
electrical properties, and still not risk demagnetization at a short-circuit fault.
From the study on multiple short-circuit events in Paper I the conclusions
were that even at low demagnetization and a low induced voltage drop, the
induced voltage can become unsymmetrical and potentially become a problem
due to higher amount of torque ripple. Further, it is the type of the short-
circuit fault that determines the total demagnetization rather than the number
of short-circuit events. From this study, it was also concluded that the external

47
model performed almost identical to the stationary model whereas the internal
model overestimated the demagnetization compared to the stationary model.
However, the computational speed of the external model was about one-fourth
of the internal model.
Regarding the magnetic flux density ripple for the 100 kW machine, the
added harmonics in the magnetic field due to passive rectification is less than
the slot-related harmonics. However, the harmonics from the passive rectifi-
cation have a higher frequency and will, therefore, cause more eddy current
losses.

48
6. Summary of Papers

Paper I
Investigation of Permanent Magnet Demagnetization in Synchronous Ma-
chines During Multiple Short-Circuit Fault Conditions
This paper compares the demagnetization characteristics of a PMSG with
two different time-dependent demagnetization models developed in CM. The
ferrite model is based on a proprietary external function, where the material
properties are defined in an external C-file. The other internal model is based
on the ODE-module in CM. The results showed that the external model is
about four times slower but more accurate and more versatile than the internal
model. Not only did the power output decrease by more than 30% after five
successive faults but the no-load voltage had become unsymmetrical which
could be explained by unsymmetrically demagnetized permanent magnets.
The permanent magnet with the lowest decrease in average remanence had
decreased by 0.8% while the highest average reduction was 23.8% in another
magnet.
The author did the majority of the work.
The paper was submitted to IET Renewable Power Generation, September
2016.

Paper II
Permanent Magnet Working Point Ripple in Synchronous Generators
This article addresses the difference in working point and in particular the
magnetic flux density ripple of the working point for the permanent magnets
in a PMSG. The working point ripple is studied for three different load cases;
no-load, a purely resistive AC-load, and a diode-rectified DC-load. The re-
sults show a smaller than expected difference in the working point, especially
between the AC and DC-load case. The frequency spectrum analysis of the
average magnetic flux density inside a magnet showed an unexpected 1.5 fel
component, assumed to be an undertone of the slot harmonic. Also, the added
harmonics in the magnetic field due to passive rectification is less than the
slot-related harmonics.
The author contributed with the electromagnetic simulations and wrote most
of the paper.
The paper was submitted to IET Journal of Engineering, October 2016.

49
Paper III
Determining Demagnetisation Risk for Two PM Wind Power Generators
with Different PM Material and Identical Stators
This paper evaluates the demagnetization risk of the 12 kW DDPMSG in
Paper V compared to the demagnetization risk of the redesigned rotor from Pa-
per VI during several different short-circuit events. In this study the torque and
the mass moment of inertia of the rotors and wind turbines were implemented
to get more accurate results of the short-circuit currents. Results showed that
both generators performed well. Both generators did suffer from a significant
amount of demagnetization, but only when the operating temperature was well
out of normal operation intervals.
The author did the majority of the work.
The paper was published in IET Electric Power Applications, August 2016,
doi:10.1049/iet-epa.2015.0518

Paper IV
Experimental Verification of a Simulation Model for Partial Demagneti-
zation of Permanent Magnets
This article aims to experimentally verify the simulation model first pre-
sented in Paper V. Further development was made to the model since it was
originally presented. The main improvement was that a function of the mag-
netic recoil behavior was added. To experimentally verify the simulation
model an iron core with a coil capable of producing high magnetic field was
designed and built. The experimental set-up was modeled in the simulation
software and the experimental and simulation results were compared. Results
showed that the simulation model performed well and can predict demagne-
tization behavior accurately. The simulations and experiments gave similar
results for most parts of the magnet.
The author did the majority of the work.
The paper was published in IEEE Transactions on Magnetics, December
2014, doi:10.1109/TMAG.2014.2339795

Paper V
Study of Demagnetization Risk for a 12 kW Direct Driven Permanent
Magnet Synchronous Generator for Wind Power
In this paper, the first version of the simulation model analyzing demagneti-
zation of permanent magnets is presented. The paper consists of a simulation-
based study of the use of different, cheaper, permanent magnet grades in a
12 kW DDPMSG designed and constructed by the research group. Further,

50
an alternative geometry was suggested to study if less permanent magnet ma-
terial could be used without the risk of demagnetization. Results showed that
the only magnet grade that was not demagnetized during any of the assumed
cases for both geometries was the one used in the built generator.
The author did the majority of the work.
The paper was published in Energy Science & Engineering, December
2013, doi:10.1002/ese3.16.

Paper VI
A Complete Design of a Rare Earth Metal-Free Permanent Magnet Gen-
erator
This paper addresses the redesign of the rotor for a 12 kW DDPMSG. The
main objective was to design a completely new rotor with ferrite permanent
magnets replacing a rotor with Nd-Fe-B. The design includes both electro-
magnetic and mechanic design, and the electromagnetic properties should be
as close to the old design as possible. An entirely new design of a rotor with
ferrite magnets was presented. Simulations show comparable results to the old
design even though the air gap magnetic flux density was reduced. To keep
the same output power the current needs to be increased slightly.
The author was a supervisor of the master thesis the paper was based on
and was involved in the planning, execution of the project and writing of the
paper.
The paper was published in Machines, May 2014,
doi:10.3390/machines2020120

51
7. Svensk sammanfattning

För att säkerställa att en permanentmagnetiserad generator kan leverera dess


märkeffekt genom hela dess livslängd krävs framförallt ett noggrant utfört des-
ignarbete samt en god underhållsplan. Oavsett hur bra designad en maskin är
så kommer den förr eller senare behöva underhåll på ett eller annat sätt. De-
signbiten är en smått delikat uppgift, dels behöver maskinen rent fysiskt klara
av de på frestningar som den kommer att utsättas för, samtidigt får maskinen
inte bli för dyr så att den inte blir lönsam. För den strukturmekaniska delen
finns det, i alla fall inom vissa användningsområden, riktlinjer och bestäm-
melser för vilka säkerhetsmarginaler som krävs för den mekaniska konstruk-
tionen. Den elektromagnetiska delen handlar framförallt om att optimera ut-
nyttjandet av materialet på ett så bra sätt som möjligt. Det aktiva materialet i
en generator, d.v.s. det järn som leder det magnetiska flödet samt permanent-
magneterna, är en betydande del av kostnaden av maskinen. För järnet, eller
elektroplåt som det kallas, gäller det att optimera priset för materialet mot de
förluster som kommer att uppstå i plåten. Bättre material med lägre förluster
kostar mer. Vad gäller valet av permanentmagneter är det inte lika självklart.
Dels vill man att permanentmagneterna som levererar det magnetiska flödet
ska vara så starka som möjligt. Med starkare fält så ökar den inducerade spän-
ningen i maskinen och en högre uteffekt kan erhållas. Däremot, klarar inte
starka magneter så bra av motriktade fält eller höga temperaturer Detta kan
leda till att magneterna kommer att avta i styrka d.v.s. avmagnetiseras. Det
finns givetvis magneter som kan utsättas för höga motriktade magnetiska fält
utan att avmagnetiseras, dock så kommer detta att ske på bekostnad av att de
själva inte kommer kunna skapa ett lika starkt magnetiskt fält. Här måste man
välja om man vill vara säker på att magneterna kommer bibehålla sin styrka
även efter eventuella felfall av höga motriktade magnetfält eller designa för
starkare magneter och hoppas på att underhållsarbetet kommer skydda mot de
allvarligaste fel som skulle kunna skada magneterna.
Arbetet i denna doktorsavhandling behandlar framförallt avmagnetisering
av permanentmagneter samt simuleringar av olika typer av kortslutningsfelfall.
En simuleringsmodell har utvecklats för detta ändamål. Tanken med denna
simuleringsmodell är att den ska kunna fungera som ett designverktyg för att
utvärdera hur olika permanentmagnets kvalitéer fungerar i en viss maskin och
på så sätt kunna välja en magnet som man tycker är lämplig. Avmagnetiser-
ingsmodellen har, med goda resultat, jämförts med experiment som visat på
liknande nivåer av avmagnetisering. Specifika studier där modellen har an-
vänds har också genomförts däribland en jämförande studie där en rotor med

52
ytmonterade Neodym-Järn-Bor (Nd-Fe-B) magneter jämfördes med en rotor
med ferritmagneter i ekerkonfiguration. Den studien var en del av en större
studie där man ville undersöka huruvida det går att byta mellan de två ovan
nämnda rotortyperna och samtidigt bibehålla de elektriska egenskaperna hos
maskinen. Resultaten visade att man relativt bra kunde bibehålla de elektriska
egenskaperna för den specifika maskin som undersöktes i detta fall. Risken för
avmagnetisering för de olika maskinerna var också snarlika. Även om rotorn
med ferritmagneter avmagnetiserades i alla testade fall sjönk den inducerade
spänningen, enligt simuleringarna, med mindre än 1%.
Ett betydande steg i utvecklingen av avmagnetiseringsmodellen togs under
det senaste året då modellen gjordes om från grunden för att numera även vara
tidsberoende. Denna förändring tillsammans med utvecklingen av hur den
elektriska kretsen till maskinerna simulerades resulterade i att multipla felfall
nu kan studeras. Detta är viktigt då elektriska maskiner riskerar att utsättas för
flera felfall under sin livstid och det är då viktigt att kunna förutse hur mask-
inen kommer att bete sig. Resultatet från en sådan studie med fem vartefter
varandra följande kortslutningar visade, inte helt överraskande, att det inte
är antalet felfall som nödvändigtvis bestämmer hur mycket permanentmag-
neterna skadas. Utan vilken typ av fel som uppstår. Denna studie visade även
att avmagnetiseringen i magneterna inte behöver vara jämn eller symmetriskt
fördelad i magneterna. Denna asymmetri kommer ge upphov till övertoner i
både inducerad spänning och uteffekt. Vidare så kan denna asymmetri ge cyk-
liska krafter på rotorn vilket i längden skulle kunna leda till att till exempel
kullager kommer att nötas ut snabbare än förutspått.

53
8. Acknowledgements

First, I would particularly like to express gratitude to my supervisor Sandra


Eriksson for giving me this opportunity. Thank you for your patient and great
guidance during these years.

All my colleagues, past, and present, you have helped make work a fun place
to be. Special thanks to Mikael and Mårten for emphasizing that it is always
five o’clock somewhere.
To all my roommates over these past years, thank you for all the laughs,
time wasted on vaguely work-related discussions and your excellent baking
skills!
My Mom and Dad, my brother Krister with family; Marie-Louise, Carl, and
Svea, thank you for everything.
Odd and Eva, I could not ask for better parents-in-law.
I am also especially grateful to my lovely wife for always being there for
me. Our wonderful daughter Anna, you can brighten my day like no other. I
love you both!
If I have forgotten anyone, please feel free to include yourself!

Finally, I would like to thank my financiers. This project has been supported
by the Swedish Research Council (grant number 2010-3950), StandUp for En-
ergy and Uppsala University.

54
References

[1] J. Gieras and M. Wing, Permanent magnet motor technology: Design and
applications, 2nd ed. Boca Raton, FL, USA: Taylor & Francis, 2002.
[2] Y. Chen, P. Pillay, and A. Khan, “PM wind generator topologies,” IEEE
Transactions on Industry Applications, vol. 41, no. 6, pp. 1619–1626, Nov 2005.
[3] D. Trumper, W. jong Kim, and M. Williams, “Design and analysis framework
for linear permanent-magnet machines,” IEEE Transactions on Industry
Applications, vol. 32, no. 2, pp. 371–379, 1996.
[4] T.-F. Chan and L.-L. Lai, “Permanent-magnet machines for distributed power
generation: A review,” in IEEE Power Engineering Society General Meeting,
2007, Tampa, USA, June 2007, pp. 1–6.
[5] S. Hlioui, L. Vido, Y. Amara, M. Gabsi, M. Lecrivain, and A. Miraoui, “PM and
hybrid excitation synchronous machines: Performances comparison,” in 18th
International Conference on Electrical Machines (ICEM) 2008, Vilamoura,
Portugal, September 2008, pp. 1–6.
[6] A. Kiyoumarsi, “Analysis and comparison of a permanent-magnet DC motor
with a field-winding DC motor,” Journal of Electrical Engineering &
Technology, vol. 4, no. 3, pp. 370–376, September 2009.
[7] M. Hedlund, J. Lundin, J. de Santiago, J. Abrahamsson, and H. Bernhoff,
“Flywheel energy storage for automotive applications,” Energies, vol. 8, no. 10,
pp. 10 636–10 663, 2015.
[8] J. J. Croat, “High energy product rare earth-iron magnet alloys,” 1989, US
Patent 4,851,058.
[9] O. Gutfleisch, M. A. Willard, E. Brück, C. H. Chen, S. G. Sankar, and J. P. Liu,
“Magnetic materials and devices for the 21st century: Stronger, lighter, and
more energy efficient,” Advanced materials, vol. 23, pp. 821–842, 2011.
[10] J. Liu, T. Ohkubo, K. Hioki, A. Hattori, T. Schrefl, and K. Hono, “Effect of Nd
content on the microstructure and coercivity of hot-deformed Nd-Fe-B
permanent magnets,” Acta Materialia, vol. 61, pp. 5387–5399, June 2013.
[11] H. Sepehri-Amin, J. Liu, T. Ohkubo, K. Hioki, A. Hattori, and K. Hono,
“Enhancement of coercivity of hot-deformed Nd-Fe-B anisotropic magnet by
low-temperature grain boundary diffusion of Nd-Dy-Cu eutectic alloy,” Scripta
Materialia, vol. 69, pp. 647–650, July 2013.
[12] W. T. Thomson and M. Fenger, “Current signature analysis to detect induction
motor faults,” IEEE Industry Applications Magazine, vol. 7, no. 4, pp. 26–34,
July 2001.
[13] G. Singh and S. A. S. A. Kazzaz, “Induction machine drive condition
monitoring and diagnostic research—a survey,” Electric Power Systems
Research, vol. 64, no. 2, pp. 145 – 158, 2003.
[14] P. J. Tavnet, G. J. W. V. Bussel, and F. Spinato, “Machine and converter
reliabilities in wind turbines,” in The 3rd IET International Conference on

55
Power Electronics Machines and Drives 2006, Dublin, Ireland, April 2006, pp.
127–130.
[15] P. J. Tavner, “Review of condition monitoring of rotating electrical machines,”
IET Electric Power Applications, vol. 2, no. 4, pp. 215–247, July 2008.
[16] J. A. Rosero, J. Cusido, A. Garcia, J. A. Ortega, and L. Romeral, “Study on the
permanent magnet demagnetization fault in permanent magnet synchronous
machines,” in IECON 2006 - 32nd Annual Conference on IEEE Industrial
Electronics, Paris, France, November 2006, pp. 879–884.
[17] D. Casadei, F. Filippetti, C. Rossi, and A. Stefani, “Magnets faults
characterization for permanent magnet synchronous motors,” in IEEE
International Symposium on Diagnostics for Electric Machines, Power
Electronics and Drives, 2009. SDEMPED 2009, Cargèse, France, August 2009,
pp. 1–6.
[18] S. Chakraborty, E. Keller, A. Ray, and J. Mayer, “Detection and estimation of
demagnetization faults in permanent magnet synchronous motors,” Electric
Power Systems Research, vol. 96, pp. 225 – 236, 2013.
[19] S. Ruoho, J. Kolehmainen, J. Ikäheimo, and A. Arkkio, “Interdependence of
demagnetization, loading, and temperature rise in a permanent-magnet
synchronous motor,” IEEE Transactions on Magnetics, vol. 46, pp. 949–953,
April 2010.
[20] C. Ruschetti, C. Verucchi, G. Bossio, C. D. Angelo, and G. García, “Rotor
demagnetization effects on permanent magnet synchronous machines,” Energy
Conversion and Management, vol. 74, pp. 1–8, October 2013.
[21] J. D. McFarland and T. M. Jahns, “Investigation of the rotor demagnetization
characteristics of interior PM synchronous machines during fault conditions,”
IEEE Transactions on Industry Applications, vol. 50, no. 4, pp. 2768–2775, July
2014.
[22] D. Givord, P. Tenaud, and T. Viadieu, “Angular dependence of coercivity in
sintered magnets,” Journal of Magnetism and Magnetic Materials, vol. 72,
no. 3, pp. 247 – 252, 1988.
[23] G. Martinek and H. Kronmüller, “Influence of grain orientation of the coercive
field in Fe-Nd-B permanent magnets,” Journal of Magnetism and Magnetic
Materials, vol. 86, no. 2, pp. 177 – 183, 1990.
[24] D. Plusa, “Angular dependence of coercivity in sintered NdDyFeB permanent
magnets,” Journal of Magnetism and Magnetic Materials, vol. 140, pp. 1061 –
1062, 1995.
[25] D. Plusa and B. Wyslocki, “Magnetization reversal processes in sintered
NdFeBNb magnets,” Journal of Magnetism and Magnetic Materials, vol. 157,
pp. 338 – 339, 1996.
[26] W. Fernengel, A. Lehnert, M. Katter, W. Rodewald, and B. Wall, “Examination
of the degree of alignment in sintered NdFeB magnets by measurements of the
remanent polarizations,” Journal of Magnetism and Magnetic Materials, vol.
157, pp. 19 – 20, 1996.
[27] R. Gao, W. Li, C. Ji, D. Zhang, and J. Zhang, “Effects of the grain alignment on
the coercivity and its angular dependence for NdCoFeB permanent magnets,”
Chinese Science Bulletin, vol. 43, no. 2, pp. 107–111, 1998.
[28] K. C. Kim, S. B. Lim, D. H. Koo, and J. Lee, “The shape design of permanent

56
magnet for permanent magnet synchronous motor considering partial
demagnetization,” IEEE Transactions on Magnetics, vol. 42, no. 10, pp.
3485–3487, October 2006.
[29] K.-Y. Hwang, S.-B. Rhee, J.-S. Lee, and B.-I. Kwon, “Shape optimization of
rotor pole in spoke type permanent magnet motor for reducing partial
demagnetization effect and cogging torque,” in International Conference on
Electrical Machines and Systems, 2007. ICEMS, Seoul, South Korea, October
2007, pp. 955–960.
[30] K. C. Kim, K. Kim, H. J. Kim, and J. Lee, “Demagnetization analysis of
permanent magnets according to rotor types of interior permanent magnet
synchronous motor,” IEEE Transactions on Magnetics, vol. 45, no. 6, pp.
2799–2802, June 2009.
[31] G. Zhao, L. Tian, Q. Shen, and R. Tang, “Demagnetization analysis of
permanent magnet synchronous machines under short circuit fault,” in
Asia-Pacific Power and Energy Engineering Conference 2010, ChengDu,
China, March 2010, pp. 1–4.
[32] M. Katter, “Angular dependence of the demagnetization stability of sintered
Nd-Fe-B magnets,” IEEE Transactions on Magnetics, vol. 41, no. 10, pp.
3853–3855, November 2005.
[33] S. Ruoho and A. Arkkio, “Partial demagnetization of permanent magnets in
electrical machines caused by an inclined field,” IEEE Transactions on
Magnetics, vol. 44, no. 7, pp. 1773–1778, August 2008.
[34] S. Apelfröjd, S. Eriksson, and H. Bernhoff, “A review of research on large scale
modern vertical axis wind turbines at uppsala university,” Energies, vol. 9, no. 7,
p. 570, 2016.
[35] S. Eriksson, “Direct driven generators for vertical axis wind turbines,” Ph.D.
dissertation, Digital Comprehensive Summaries of Uppsala Dissertations from
the Faculty of Science and Technology 547, ISSN 1651-6214, 2008.
[36] J. Kjellin, “Vertical axis wind turbines - Electrical system and experimental
results,” Ph.D. dissertation, Digital Comprehensive Summaries of Uppsala
Dissertations from the Faculty of Science and Technology 981, ISSN
1651-6214, 2012.
[37] F. Bülow, “A generators perspective on vertical axis wind turbines,” Ph.D.
dissertation, Digital Comprehensive Summaries of Uppsala Dissertations from
the Faculty of Science and Technology 1034, ISSN 1651-6214, 2013.
[38] P. Deglaire, “Analytical aerodynamic simulation tools for vertical axis wind
turbines,” Ph.D. dissertation, Digital Comprehensive Summaries of Uppsala
Dissertations from the Faculty of Science and Technology 774, ISSN
1651-6214, 2010.
[39] A. Goude, “Fluid mechanics of vertical axis turbines: Simulations and model
development,” Ph.D. dissertation, Digital Comprehensive Summaries of
Uppsala Dissertations from the Faculty of Science and Technology 998, ISSN
1651-6214, 2012.
[40] E. Dyachuk, “Aerodynamics of vertical axis wind turbines - Development of
simulation tools and experiments,” Ph.D. dissertation, Digital Comprehensive
Summaries of Uppsala Dissertations from the Faculty of Science and
Technology 1274, ISSN 1651-6214, 2015.

57
[41] E. Möllerström, “Vertical axis wind turbines - Tower dynamics and noise,”
Licentiate thesis, Uppsala University, Department of Engineering Sciences,
ISSN 0349-8352; 340-15L, 2015.
[42] P. Eklund, “Rare earth metal-free permanent magnet generators,” Licentiate
thesis, Uppsala University, Department of Engineering Sciences, ISSN
0349-8352; 348-16L, 2016.
[43] M. Rossander, “Blade force measurements and electrical torque ripple of a
vertical axis wind turbine,” Licentiate thesis, Uppsala University, Department of
Engineering Sciences, ISSN 0349-8352; 346-16L, 2016.
[44] J. Olauson, “Modeling wind power for grid integration studies,” Ph.D.
dissertation, Digital Comprehensive Summaries of Uppsala Dissertations from
the Faculty of Science and Technology 1428, ISSN 1651-6214, 2016.
[45] S. Apelfröjd, “Grid connection of permanent magnet generator based renewable
energy systems,” Ph.D. dissertation, Digital Comprehensive Summaries of
Uppsala Dissertations from the Faculty of Science and Technology 1436, ISSN
1651-6214, 2016.
[46] J. William D. Callister, Materials science and engineering an introduction,
6th ed. New York, NY, USA: John Wiley & Sons, Inc., 2003.
[47] B. Cullity and C. Graham, Introduction to Magnetic Materials, 2nd ed.
Piscataway, NJ, USA: John Wiley & Sons, Inc., 2009.
[48] D. Jiles, Introduction to magnetism and magnetic materials, 2nd ed. Boca
Raton, FL, USA: Taylor & Francis, 1998.
[49] M. S. Walmer, C. H. Chen, and M. H. Walmer, “A new class of Sm-Tm magnets
for operating temperatures up to 550◦ C,” IEEE Transactions on Magnetics,
vol. 36, no. 5, pp. 3376–3381, September 2000.
[50] S. Ruoho, E. Dlala, and A. Arkkio, “Comparison of demagnetization models for
finite-element analysis of permanent-magnet synchronous machines,” IEEE
Transactions on Magnetics, vol. 43, no. 11, pp. 3964–3968, December 2007.
[51] S. Hamidizadeh, N. Alatawneh, R. R. Chromik, and D. A. Lowther,
“Comparison of different demagnetization models of permanent magnet in
machines for electric vehicle application,” IEEE Transactions on Magnetics,
vol. 52, no. 5, pp. 1–4, May 2016.
[52] H. C. Lovatt and P. A. Watterson, “Energy stored in permanent magnets,” IEEE
Transactions on Magnetics, vol. 35, no. 1, pp. 505–507, January 1999.
[53] B. Sprecher, Y. Xiao, A. Walton, J. Speight, R. Harris, R. Kleijn, G. Visser, and
G. J. Kramer, “Life cycle inventory of the production of rare earths and the
subsequent production of NdFeB rare earth permanent magnets,” Environmental
Science & Technology, vol. 48, no. 7, pp. 3951–3958, 2014.
[54] J. C. K. Lee and Z. Wen, “Rare earths from mines to metals: Comparing
environmental impacts from China’s main production pathways,” Journal of
Industrial Ecology, September 2016.
[55] S. Eriksson and H. Bernhoff, “Rotor design for PM generators reflecting the
unstable neodymium price,” in XXth International Conference on Electrical
Machines (ICEM) 2012, Marseille, France, September 2012, pp. 1419–1423.
[56] S. Eriksson, A. Solum, M. Leijon, and H. Bernhoff, “Simulations and
experiments on a 12 kW direct driven PM synchronous generator for wind
power,” Renewable Energy, vol. 33, no. 4, pp. 674–681, May 2008.

58
[57] S. Eriksson, H. Bernhoff, and M. Leijon, “FEM simulations and experiments of
different loading conditions for a 12 kW direct driven PM synchronous
generator for wind power,” International Journal of Emerging Electric Power
Systems, vol. 10, no. 1, February 2009.
[58] M. Rossander, A. Goude, and S. Eriksson, “Mechanical torque ripple from a
passive diode rectifier in a 12 kW vertical axis wind turbine,” To appear in IEEE
Transactions on Energy Conversion.

59
Acta Universitatis Upsaliensis
Digital Comprehensive Summaries of Uppsala Dissertations
from the Faculty of Science and Technology 1444
Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science and


Technology, Uppsala University, is usually a summary of a
number of papers. A few copies of the complete dissertation
are kept at major Swedish research libraries, while the
summary alone is distributed internationally through
the series Digital Comprehensive Summaries of Uppsala
Dissertations from the Faculty of Science and Technology.
(Prior to January, 2005, the series was published under the
title “Comprehensive Summaries of Uppsala Dissertations
from the Faculty of Science and Technology”.)

ACTA
UNIVERSITATIS
UPSALIENSIS
Distribution: publications.uu.se UPPSALA
urn:nbn:se:uu:diva-303517 2016

Das könnte Ihnen auch gefallen