Institutionen fr systemteknik

Department of Electrical Engineering

Examensarbete
Evaluation of software using the nite element
method by simulating transformers and inductors
Examensarbete utfrt i Elektroniksystem
vid Tekniska hgskolan vid Linkpings universitet
av
David Hkansson and Jenny Larsson
LiTH-ISY-EX-ET--11/0381--SE
Linkping 2011
Department of Electrical Engineering Linkpings tekniska hgskola
Linkpings universitet Linkpings universitet
SE-581 83 Linkping, Sweden 581 83 Linkping
Evaluation of software using the nite element
method by simulating transformers and inductors
Examensarbete utfrt i Elektroniksystem
vid Tekniska hgskolan i Linkping
av
David Hkansson and Jenny Larsson
LiTH-ISY-EX-ET--11/0381--SE
Supervisor: Sievert Johansson
ETAL Group AB
Examiner: Kent Palmkvist
isy, Linkpings universitet
Linkping, 10 February, 2011
Avdelning, Institution
Division, Department
Division of Electronic Systems
Department of Electrical Engineering
Linkpings universitet
SE-581 83 Linkping, Sweden
Datum
Date
2011-02-10
Sprk
Language
Svenska/Swedish
Engelska/English

Rapporttyp
Report category
Licentiatavhandling
Examensarbete
C-uppsats
D-uppsats
vrig rapport

URL fr elektronisk version

http://www.ep.liu.se
ISBN

ISRN
LiTH-ISY-EX-ET--11/0381--SE
Serietitel och serienummer
Title of series, numbering
ISSN

Titel
Title Evaluation of software using the nite element method by simulating transformers
and inductors
Frfattare
Author
David Hkansson and Jenny Larsson
Sammanfattning
Abstract
In this bachelor thesis several software, capable of calculating and
simulating complex problems concerning the power losses in inductors and
transformers with the nite element method, have been evaluated and used to
solve test cases provided by the commissioner. The software have been evaluated
with respect to several requirements stated by the commissioner.
The aim is to be able to simulate power losses and inductance levels in complex
designs of inductors and transformers. By reading the manuals to the software, a
view of the methods and equations the dierent software use for their calculations
have been established. The enclosed tutorials have provided the knowledge for
the operations of the dierent software. By designing the test models provided
by the commissioner, a deeper understanding of the work area has been reached.
The test results provides an answer for the test models, the behaviour of the
magnetic eld has been analysed for the models and the calculated power losses
seem to correspond to the behaviour of the prototypes.
The evaluation of the software has been done with regard to the commissioners
requirements. The recommendation will be to use either FEMM 4.2 or QuickField
5.7, both software have a short training curve and an interface easy to maintain.
For problems requiring a transient analysis the recommendation is QuickField, but
the material library maintainability is better in FEMM 4.2. Regarding COMSOL
Multiphysics 3.5 and Ansys R Ansoft Maxwell Student Version 9, both software
are highly qualied for the complex calculations needed for these kind of problems.
The training curve for these software is however much longer than for the other
two software and for the commissioner to be able to fully use all the possibilities
in the software this will not be ecient.
Nyckelord
Keywords transformers, inductors, design, simulation, power losses, software, nite element
method
Abstract
In this bachelor thesis several software, capable of calculating and
simulating complex problems concerning the power losses in inductors and
transformers with the nite element method, have been evaluated and used
to solve test cases provided by the commissioner. The software have been
evaluated with respect to several requirements stated by the commissioner.
The aim is to be able to simulate power losses and inductance levels in
complex designs of inductors and transformers. By reading the manuals to
the software, a view of the methods and equations the dierent software
use for their calculations have been established. The enclosed tutorials
have provided the knowledge for the operations of the dierent software.
By designing the test models provided by the commissioner, a deeper un-
derstanding of the work area has been reached.
The test results provides an answer for the test models, the behaviour
of the magnetic eld has been analysed for the models and the calculated
power losses seem to correspond to the behaviour of the prototypes.
The evaluation of the software has been done with regard to the com-
missioners requirements. The recommendation will be to use either FEMM
4.2 or QuickField 5.7, both software have a short training curve and an in-
terface easy to maintain. For problems requiring a transient analysis the
recommendation is QuickField, but the material library maintainability is
better in FEMM 4.2. Regarding COMSOL Multiphysics 3.5 and Ansys
R Ansoft Maxwell Student Version 9, both software are highly qualied for
the complex calculations needed for these kind of problems. The training
curve for these software is however much longer than for the other two soft-
ware and for the commissioner to be able to fully use all the possibilities in
the software this will not be ecient.
v
Acknowledgements
Thank you all who have contributed to help us complete this thesis. All
the sta at ETAL Group AB, you have welcomed us and helped us a lot
through the theoretical background and discussed the problems we have
encountered. A special thank you is included for outstanding on-line sup-
port provided by Simon Tegelid and Jonas strm concerning the writing
of this document in L
A
T
E
X. Also thank you to families and friends for the
support in life in general, without you this is not possible at all.
We would also like to thank our examiner Ph.D. Kent Palmkvist as well
as our opponent Markus Keller.
Linkping - February 2011
David Hkansson and Jenny Larsson
Electrical engineering at Linkping University, Institute of technology
Linkping University
vii
Contents
1 Introduction 3
1.1 Power losses in transformers and inductors . . . . . . . . . . 3
1.2 The commissioner . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Dierent software using the nite
element method . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Aims of this thesis . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.6 Specication of requirements . . . . . . . . . . . . . . . . . 4
2 Theory 7
2.1 The nite element method . . . . . . . . . . . . . . . . . . . 7
2.1.1 The Newton iteration process . . . . . . . . . . . . . 8
2.2 Maxwells equations . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Amperes law . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 Faradays law . . . . . . . . . . . . . . . . . . . . . . 10
2.2.3 Lenzs law . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.4 Gausss law . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 Materials 13
3.1 Ferrite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 Relation between B and H . . . . . . . . . . . . . . 14
3.1.2 Resistivity . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.3 Permeability . . . . . . . . . . . . . . . . . . . . . . 14
4 Power losses calculation 17
4.1 Eddy current losses . . . . . . . . . . . . . . . . . . . . . . . 17
4.1.1 Skin eect . . . . . . . . . . . . . . . . . . . . . . . . 17
4.1.2 Proximity eect . . . . . . . . . . . . . . . . . . . . . 18
4.1.3 Core losses . . . . . . . . . . . . . . . . . . . . . . . 19
ix
x Contents
5 The software 21
5.1 COMSOL Multiphysics 3.5 . . . . . . . . . . . . . . . . . . 21
5.1.1 Calculating using the AC/DC-module . . . . . . . . 22
5.2 QuickField 5.7 . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.2.1 Calculating using the AC magnetic module . . . . . 23
5.2.2 Calculating using the Transient Magnetics module . 24
5.2.3 Heat Transfer module . . . . . . . . . . . . . . . . . 24
5.3 FEMM 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3.1 Calculating time-harmonic magnetic problems . . . 25
5.3.2 Post processing . . . . . . . . . . . . . . . . . . . . . 26
5.4 Ansys R Ansoft Maxwell Student Version 9 . . . . . . . . . . 26
5.4.1 Calculating using the Eddy Current solver . . . . . . 26
6 The models to be simulated 29
6.1 Commutation choke coil . . . . . . . . . . . . . . . . . . . . 29
6.2 Asymmetric power inductor . . . . . . . . . . . . . . . . . . 29
6.3 Flyback transformer . . . . . . . . . . . . . . . . . . . . . . 30
6.4 Planar transformer . . . . . . . . . . . . . . . . . . . . . . . 30
7 Simulations 33
7.1 Material properties . . . . . . . . . . . . . . . . . . . . . . . 33
7.2 Simulating using FEMM 4.2 . . . . . . . . . . . . . . . . . . 33
7.2.1 Building the models . . . . . . . . . . . . . . . . . . 33
7.2.2 Physics settings . . . . . . . . . . . . . . . . . . . . . 34
7.2.3 Meshing and solving . . . . . . . . . . . . . . . . . . 34
7.2.4 Post processing . . . . . . . . . . . . . . . . . . . . . 34
7.3 Simulating using COMSOL Multiphysics 3.5 . . . . . . . . . 35
7.3.1 Building the models . . . . . . . . . . . . . . . . . . 35
7.3.2 Physics settings . . . . . . . . . . . . . . . . . . . . . 35
7.3.3 Meshing and solving . . . . . . . . . . . . . . . . . . 35
7.3.4 Post processing . . . . . . . . . . . . . . . . . . . . . 35
7.4 Simulating using QuickField 5.7 . . . . . . . . . . . . . . . . 36
7.4.1 Building the models . . . . . . . . . . . . . . . . . . 36
7.4.2 Physics settings . . . . . . . . . . . . . . . . . . . . . 36
7.4.3 Meshing and solving . . . . . . . . . . . . . . . . . . 36
7.4.4 Post processing . . . . . . . . . . . . . . . . . . . . . 37
7.5 Simulating using Ansys R Ansoft Maxwell Student Version 9 37
7.5.1 Building the models . . . . . . . . . . . . . . . . . . 37
7.5.2 Physics setting . . . . . . . . . . . . . . . . . . . . . 37
7.5.3 Meshing and solving . . . . . . . . . . . . . . . . . . 38
7.5.4 Post processing . . . . . . . . . . . . . . . . . . . . . 38
Contents xi
8 Results 39
8.1 Commutation choke coil . . . . . . . . . . . . . . . . . . . . 39
8.1.1 Calculating the temperature . . . . . . . . . . . . . . 41
8.2 Asymmetrical power inductor . . . . . . . . . . . . . . . . . 41
8.3 Flyback transformer . . . . . . . . . . . . . . . . . . . . . . 44
8.4 Planar transformer . . . . . . . . . . . . . . . . . . . . . . . 45
9 Conclusion/Discussion 47
9.1 FEMM 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
9.2 COMSOL Multiphysics 3.5 . . . . . . . . . . . . . . . . . . 47
9.3 QuickField 5.7 . . . . . . . . . . . . . . . . . . . . . . . . . 48
9.4 Ansys R Ansoft Maxwell Student Version 9 . . . . . . . . . 48
9.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9.5.1 Commutation choke coil . . . . . . . . . . . . . . . . 49
9.5.2 Asymmetrical power inductor . . . . . . . . . . . . . 50
9.5.3 Flyback transformer . . . . . . . . . . . . . . . . . . 50
9.5.4 Planar transformer . . . . . . . . . . . . . . . . . . . 50
Bibliography 51
A 53
A.1 Commutation choke coil . . . . . . . . . . . . . . . . . . . . 53
A.2 Asymmetrical Power Inductor . . . . . . . . . . . . . . . . . 56
A.3 Flyback transformer . . . . . . . . . . . . . . . . . . . . . . 57
A.3.1 Power loss graphs in the windings . . . . . . . . . . 60
A.4 Planar Transformer . . . . . . . . . . . . . . . . . . . . . . . 63
List of Figures
4.1 Skin eect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Proximity eect . . . . . . . . . . . . . . . . . . . . . . . . . 19
8.1 The magnetic eld and the current density plot. Planar mode. 40
8.2 Concentration of the current density around the air gap. Pla-
nar mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
8.3 The concentration of the current density in axial symmetric
mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.4 The original design made with a small air gap at the center
leg of the E-core. . . . . . . . . . . . . . . . . . . . . . . . . 42
8.5 The resulting design to reach the desired value of the induc-
tance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.6 The nal design with an air gap on all three legs in the core. 43
8.7 The magnetic eld when the primary winding connects. . . 44
8.8 The current density when the primary winding connects. . . 45
8.9 Planar transformer with magnetizing current on the primary
side. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
A.1 The commutation choke coil graphically displayed in FEMM 53
A.2 The commutation choke coil graphically displayed in Quick-
Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
A.3 The commutation choke coil graphically displayed in Maxwell 55
A.4 Power inductor with air gap over all legs in the core. Dis-
played in FEMM . . . . . . . . . . . . . . . . . . . . . . . . 56
A.5 Power inductor with air gap over all legs in the core. Dis-
played in QuickField. . . . . . . . . . . . . . . . . . . . . . . 56
A.6 Power inductor with air gap over all legs in the core. Dis-
played in Maxwell. . . . . . . . . . . . . . . . . . . . . . . . 57
A.7 Flyback transformer, magnetic eld intensity plot in the sec-
ondary winding. Displayed in QuickField. . . . . . . . . . . 57
A.8 Flyback transformer, magnetic eld density plot in the sec-
ondary winding. Displayed in QuickField. . . . . . . . . . . 58
A.9 Flyback transformer, current density plot in the secondary
winding. Displayed in QuickField. . . . . . . . . . . . . . . 58
A.10 Flyback transformer, current density plot in secondary wind-
ing when the current switches between the primary and sec-
ondary winding. Displayed in QuickField. . . . . . . . . . . 59
A.11 Power loss in the bias winding. Displayed in QuickField. . . 60
A.12 Power loss in the primary winding. Displayed in QuickField. 61
2 Contents
A.13 Power loss in the rst secondary winding. Displayed in
QuickField. . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
A.14 Power loss in the second secondary winding. Displayed in
QuickField. . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
A.15 Power loss in the third secondary winding. Displayed in
QuickField. . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
A.16 Planar transformer. Displayed in FEMM . . . . . . . . . . . 63
Chapter 1
Introduction
1.1 Power losses in transformers and inductors
When designing power electronic components, such as transformers and
inductors, it is important to consider several dierent parameters. If the
environment where the component will work is narrow, the power losses
will be a primary parameter for the designer to consider. Usually there are
constraints from the customer to have a high degree of power eciency.
For the designer it is important to has access to good enough tools to make
a good design of the component. There is a lot of software on the market
that can simulate the behaviour of the component before producing a pro-
totype. This makes it important that the software is accurate enough for
the designer so the prototype will be within the constraints from the cus-
tomer. For some simulation software the higher eciency in the designing
process will give a 1.6 fewer prototypes at the end.
1.2 The commissioner
This bachelor thesis is assigned by ETAL Group AB, a leading supplier and
manufacturer of inductive components for telecommunications, automotive
and electronics industries. ETAL Group AB is a part of Elektronikgruppen
BK AB with the head oce in Vllingby. Design of qualied electrical
transformers is made in Sweden at ETAL R& D center in Tornby, Linkping
to a worldwide market. Manufacturing operations are situated in Southeast
Asia and in Europe. ETAL also has sales oces in Sweden, Finland, Great
Britain, India and China. [4]
3
4 Introduction
1.3 Dierent software using the nite
element method
A lot of the software on the market specializes in FEM for mechanics of ma-
terials. It is essential that the software is able to calculate problems within
the electromagnetic eld, as well as the thermal heat transfer eld. Most
of the software available have a structure with a main platform with added
modules for each problem area you might want to solve. The equations
and the calculations that the software perform varies between the dierent
software. The basic theory in the software is based on Maxwells equations
and the general magnetic eld theory. It is essential to have knowledge of
the dierent parameters the software need to use to perform the simulation
of the specic designs.
1.4 Aims of this thesis
To evaluate dierent software on the market, that uses the nite ele-
ment method, from a point of view that suits the commissioner.
To simulate and interpret the dierent test cases provided by the
commissioner using the dierent software that will be considered
appropriate for the assignment.
1.5 Method
There is a lot of software on the market that uses the nite element method
to simulate the dierent behaviour of the considered designs.
To be able to decide which software to use, a match has been made
against a list of requirements the commissioner has specied. To solve the
dierent test cases the manuals to each software has been studied, their
tutorials have been a help for the training with the software and the test
cases have been approached with help of the manuals.
1.6 Specication of requirements
Since there are a vast number of software on the market that uses the nite
element method, a list of requirements from ETAL have been provided that
the software considered shall be able to perform.
1.6 Specication of requirements 5
1. Use the nite element method to calculate and simulate the behaviour
of a variety of designs.
2. Be able to calculate a 2-D design, it is also desirable to have the
ability to calculate a 3-D design.
3. The material libraries shall be easy to edit when dening own mate-
rials. Also it is preferable if the library contains a large number of
dierent materials used in insulation, coil and layers.
4. Be able to present the results in a graphically accessible way.
5. The software must have a short training curve for the commissioner.
Chapter 2
Theory
2.1 The nite element method
The nite element method is a numerical method for solving partial dif-
ferential equations but also integral equations. The solution will be an
approximation of the result. The method will either eliminate the dif-
ferential equation and make the problem steady-state or approximate the
equations into a system of ordinary dierential equations and then use the
numerically integrating techniques provided by the standard formulas such
as Eulers method, Runge-Kutta, etc.
The nite element method originated when the need to solve complex
problems that depended on the constraints required by a non-symmetrical
function. For example a non-symmetrical current that diers over time.
This method is also widely used when calculating problems within the
solid mechanics problems. The pioneers who worked on this method were
Alexander Hrenniko (1941) and Richard Courant (1942), their approach
is
dierent from the approach today but the essential characteristic they all
share is the mesh discretization. This is done by distributing a net of sub-
domains in a continuous domain and solve each sub-domain separately.
These sub-domains are usually named elements. During the 1950s the de-
velopment of the nite element method really took o and by 1973 a math-
ematical foundation was laid for the method used today. Gilbert Strang
and George Fix published An Analysis of The Finite Element Method, this
publication is the foundation to a wide range of generalized applied mathe-
matical numerical modelling of physical systems within a lot of engineering
disciples. [9]
7
8 Theory
2.1.1 The Newton iteration process
It is preferable for the iteration method for the solver to converge rather
fast, else it might consume a lot of computer time. The Newton iteration
process is unconditionally stable and converges rapidly but will not require
a lot more computer memory. Close to the solution point the convergence
rate is, theoretically, quadratic. In other words, if the rst approximation
will achieve a solution within a factor two, one correct bit, the second ap-
proximation will produce a solution with two correct bits, the third will
produce four and so on. The quadratically convergent steps to yield full
precision on a 32-bit digital computer should not be more than ve or six,
so the approximated solution will probably not take to long time. [12, p.181]
The electromagnetic eld problem can be expressed by a set of equations
from Maxwell. These equations will be explained further down in this
chapter. These equations will dier for example when the materials will
give a non-linear relationship between the magnetic eld intensity and the
magnetic eld density.
2.2 Maxwells equations
Maxwells equations are dened by four partial dierential equations. They
were published by James Clerk Maxwell in 1865. These equations denes
the entire eld of electromagnetics. These equations are laws that the
magnetic and electric elds always must obey. [7]
2.2.1 Amperes law
A conductor that carries a current induces a magnetic eld around the
conductor. This eld is characterized by the magnetic eld intensity H.
A thumb-rule is used to nd the direction of the magnetic eld density,
described in an easy way, if the conductor is held with your right hand and
the thumb indicates the current direction, then your ngers indicate the
direction of the magnetic eld.
The magnetic eld intensity H, is dened by Amperes law. This law
says that the integral of H [A/m] around a closed path is equal to the
passing of the total current through the interior of the path.
_
l
H dl =
_
S
J dS
2.2 Maxwells equations 9
H is the eld intensity vector [A/m], dl is a vector length element
pointing in the direction of the path l[m], J is the electrical current density
vector[A/m
2
], dS is a vector area having direction normal to the surface
[m
2
], l is the length of the circumference of the contour [m] and S is the
surface of the contour [m
2
].[2, chapter 1]
Applying this to the case where the current is carried by the wires in a
coil with N turns, then
_
l
H dl =
_
S
J dS = Ni
Where i is the current and N the number of turns in the coil.
The magnetic eld intensity H will give us a resulting magnetic eld
density B. This is given by the equation:
B =
0

r
H = H
is the characteristic of the magnetic material term permeability.
0
is
the permeability of free space, it is a constant 4 10
7
H/m.
r
is the
relative permeability of the magnetic material.
Depending on the material, the value of changes, for air and electrical
conductors (copper, aluminum) it is 1. For ferromagnetic materials such
as iron, nickel and cobalt the value varies from several hundred to tens of
thousands.
In the right-hand side of Amperes law we miss a term, this term is
called displacement current and was added to the expression by Maxwell
in 1865. So with this term added the full expression of Amperes law is:
_
l
H dl =
_
S
J dS +

t
_
S
E dS
is the permittivity of the medium and E denotes the electrical eld.
This correction of Amperes law is mainly important for high-frequency
applications with low current density. For power electronics, the mag-
netic components, are expected to have a current density of at least J =
10
6
A/m
2
. Normally the second term in Amperes law is almost never more
than 10A/m
2
, then it is possible to neglect this. The exception must be
made for currents in capacitors and currents caused by parasitic capac-
itances and also for currents in transmission lines. Come to analysis of
10 Theory
power
electronics magnetic circuits this approach is called the quasi-static ap-
proach. [2, chapter 1]
2.2.2 Faradays law
Faradays law gives us the relationship between a ux , that is time-
depending, which passes through a closed loop and a voltage v(t), that is
generated by the ux. The relationship is given by the equation:
v(t) =
d(t)
dt
There are two cases where Faradays law is valid:
When a xed circuit is linked by a time-changing ux, for example a
transformer.
When a moving circuit is related to a time-stationary magnetic ux
in such a way that it produces a time-changing ux in the interior of
the circuit.
[2, chapter 1]
2.2.3 Lenzs law
The voltage v(t) generated by a fast time-changing magnetic ux (t) has
a direction to drive a current in a closed loop, this current induces a second
magnetic ux that intends to oppose the changes in the applied ux (t).
This explains a part of the eddy current eects that appears in magnetic
cores and coil conductors. This phenomena is one of the major causes for
power losses in the core and also losses in the conducting coil. [2]
2.2.4 Gausss law
Gauss dened a law for magnetic circuits, which states that for any closed
surface S with arbitrary form, the total ux entering the volume equals the
total ux coming out of the volume. The total resulting ux through the
surface is zero according to the equation:
_
S
B dS = 0
Gausss law for magnetic circuits is analogous to Kirchos current density
law for electrical circuits. [2, chapter 1]
2.3 Parameters 11
2.3 Parameters
The software uses several common parameters for calculating the power
losses, the inductance and the heat transfer. Below we state the most im-
portant that concern our problems.
B/H-curves gives the relationship between the magnetic eld and the
magnetic eld intensity of a material. It is directly connected to the
permeability of a material. The permeability will in ferrite cores be a func-
tion of the magnetic eld and the magnetic eld intensity. The magnetic
eld might vary with the current density and/or the frequency. Thus makes
the equations for calculating the problems built on the magnetic eld non-
linear.
gives the value of the electric conductivity for a material. This con-
stant need to be set to a correct value for the cores. This will be an issue
when adding custom materials to the software libraries. This constant will
aect the calculation of the core losses and if not set correct the core losses
calculated will not correspond to the real value.
gives the value of the permeability of the material. This will give the
magnetization of the material with dependence of the magnetic eld B and
the magnetic eld intensity H since the relation is given by:
=
B
H
This constant is also essential for the calculations of the power losses.
h gives the value of the heat transfer coecient for the material. It is
used for calculating the heat transfer by convection between a uid and a
solid in the software. Together with the denition of the interesting areas of
the model from a previous solution concerning the density of the magnetic
eld, this will give the temperature value in the critical parts of the design.
h =
q
A T
A is the heat transfer surface area, T is the dierence in temperature be-
tween the solid surface and the surrounding uid area and q denotes heat
ow in input or lost heat ow. [8, p.201]
Air gap is an important part of the design of the transformer and the
inductor. This will aect the B/H-curve and increase the saturation level
12 Theory
decided by this relationship. When applying an air gap to the geometry the
saturation level for H will increase signicantly while the saturation level
for B will remain almost at the same level. If the core reach saturation the
inductance in the core will drop very fast and thus the power losses will
increase. So although this is not a real parameter in the equations used for
the calculations it will aect the behaviour of the material and is of such
importance to the design of these components that it is important to have
some knowledge about it.
Chapter 3
Materials
There are three groups of magnetic materials:
Diamagnetic materials
Paramagnetic materials
Ferromagnetic materials
When designing power electronics the ferromagnetic material is the most
suitable material. Especially ferromagnetic ceramics and metals. This is
due to the constant
r
that is signicantly higher than for the other two
materials.
r
is aected by the relation between B and H in the material.
[2, chapter 1]
3.1 Ferrite
The use of ferrite in the core of transformers and inductors is most com-
mon today. This because the volume resistivity is much higher than in
other materials. When designing for high frequencies it is of importance
to calculate with the eddy current losses, they will increase approximately
with the square of the frequency. This behaviour is inversely proportional
to the resistivity in the core.
The ferrite is composed by two dierent combinations of dierent met-
als:
manganese and zinc
nickel and zinc
13
14 Materials
Nickel and zinc are appropriate for higher frequencies, over 1 MHz. While
manganese and zinc have a high permeability and a high saturation induc-
tion levels. This alloy is suitable for low frequency power electronics.
3.1.1 Relation between B and H
The relation between B[T] and H[A/m] is given by the material proper-
ties. These properties are inuenced by the manufacturing process. By
choosing dierent manufacturing processes the manufacturer can control
the saturation level and permeability of the material, which is essential to
the designer. The permeability increases with the temperature to a maxi-
mum of the Curie temperature and then the permeability of the ferrite will
drop sharply when that specic point is reached. [2, chapter 3]
3.1.2 Resistivity
The resistivity is very dependent of the temperature and the frequency. It
increases with an increasing temperature. For high frequencies the resis-
tivity will decrease. [2, chapter 1]
3.1.3 Permeability
This property is very important when calculating power losses. There are
dierent ways to interpret the relative permeability
r
it depends on the
conditions when dening and measuring.
Amplitude permeability
a
, this is the relative permeability under
alternating external eld H. The relation between the peak value of
B and H is given by:

a
=
1

B is the amplitude value over the cross-section of the material and

H
is the amplitude in the eld parallel to the surface of the material.
Initial permeability
i
, this is the relative permeability of the material
when we have a very low applied value of H. This is given by:

i
=
1

0
B
H
(H 0)
Eective permeability
e
denotes the total permeability when an air
gap is present in a closed magnetic circuit. This value will be much
3.1 Ferrite 15
lower than in the same circuit without the gap. It is dependent on
the initial permeability
i
and the dimensions of the design and the
air gap. For designs with a relatively small air gap the eective per-
meability is given by:

e
=

i
1 +
A
g

i
l
c
where A
g
is the cross-sectional area of the air gap and l
c
is the eective
length of the magnetic path.
If there is a large air gap in the design the ux will pass outside the air gap
and this will increase the eective permeability. This phenomenon is called
fringing permeability. [2, chapter 1]
Chapter 4
Power losses calculation
4.1 Eddy current losses
There are several techniques to reduce the presence of eddy currents. When
calculating the eects of eddy currents you need to know that the presence
of them is linear, voltage and currents stay proportional. Active and reac-
tive power losses are quadratic with the currents or the elds. Since the
mathematical functions are analytical, (minimum phase impedance func-
tion) you can apply Bode laws. If you, for example neglect the capacitive
eects, the resistance increases with the frequency and the inductance de-
creases with the frequency. [2, chapter 2]
4.1.1 Skin eect
The skin eect is a phenomenon known and derived for skin depth versus
frequency since the beginning of the twentieth century. Induced eddy cur-
rent causes the current to crowd into the outer skin of a conductor. The
skin eect causes copper losses by restricting the conducting area of the
wire, the magnitude of the currents owing does not change, but the cur-
rent density at the wire surface increases.
The main reason to why we have skin eect is due to the skin depth.
The skin depth is where we have the major part of the current density in the
conductor. The eddy currents press the current density towards the edges
of the coils. With a higher frequency the skin depth tend to be smaller. It
is determined by:
=

2
c

17
18 Power losses calculation
where = 2f for the frequency of the applied magnetic eld, is the
permeability of the material the conductor is made of and
c
is the elec-
trical resistivity of the material in the conductor. This diers at dierent
temperatures.
c
= 17.24 10
9
m at 25

C and
c
= 23 10
9
m at
100

C. [2, chapter 2]
Figure 4.1. Skin eect
4.1.2 Proximity eect
When designing a multilayer coil we have to calculate the proximity eect.
This is caused by the alternating magnetic eld, B, that arises from the
currents in the winding layers in the coil. The magnetic eld induces a cur-
rent in the adjacent windings. These windings do not conduct during this
period, but the result is an eect that contributes to the total power losses.
This eect is more serious than the skin eect. Because the eddy currents
caused by the magnetic elds in the adjacent layers increase exponentially
in amplitude as the number of coil layers increases, in other words the skin
eect is enhanced by the proximity eect. [2, chapter 2]
4.1 Eddy current losses 19
Figure 4.2. Proximity eect
4.1.3 Core losses
When designing these, relatively, small transformers and inductors one
must consider the core losses. They are caused by the magnetic eld that
induces eddy currents in the core. All materials that are magnetic have
some electrical conductivity. This conductivity for iron-based magnetic
materials is relatively high. The ux in the core induces an internal volt-
age d/dt, this voltage drives circulating currents around its path. When
considering a cross-sectional area of the core, the area of the cross-section
can not be large compared to the skin depth. The applied magnetic eld
is carried by the surface area and the inner of the core carries a very small
part of the magnetic eld. The AC reluctance will increase and reducing
the low reluctance path for the eld. There is a connection between the
eddy current losses in the core and the amplitude induction B
max
, the fre-
quency f and the internal resistivity of the core material. These losses
per volume are proportional to the square of the induced voltage and as
a result proportional to (fB
max
)
2
. There is an expression for the eddy
current losses in the core as following:
P
ec
= k
e
f
2
B
2
max

where k
e
is a dimensionless eddy current coecient, B
max
is the amplitude
induction of the applied eld and is the internal resistivity of the core.
This is only an approximation of the losses in the core. The impedance of
some cores is not purely resistive but also depends on the frequency. Thus
for some frequencies the dependence might be higher than f
2
at a constant
induction level.
20 Power losses calculation
When specically taking a closer look at the core losses in ferrite cores,
used in power electronics, we will look at the losses when a non-sinusoidal
voltage waveform is applied. Generally there is one equation used for cal-
culating ferrite losses, this equation is called the Steinmetz equation:
P
loss
= kf

(4.1)
where

B is the peak induction, P
loss
is the average power loss per unit
volume and f is the sinusoidal excitation.
This equation will have a decreased accuracy for square waves but it
will still be a good approximation with a duty ratio around 50 %. For
very low or high duty ratio, < 5 % or > 95 %, the approximation of the
power loss will be more than doubled compared to sine waves. For real
materials and are frequency dependent, usually they increase with the
frequency, thus k will also change. To be able to make correct calculations
the samples will be measured and the Steinmetz equation will be adjusted
with a reference frequency at 100 kHz, a reference power and induction of
0.1 T:
P = k
ref
B

= P
ref
_
B
B
ref
_

_
f
f
ref
_

The coecient is tted for the reference frequency and the reference in-
duction. is determined by using the reference values at a higher frequency
at 250 kHz. This because it lies in between the second and third harmonic.
These coecients can be obtained from the manufacturers data sheets. [2,
chapter 3]
Chapter 5
The software
There are a lot of software on the market using the nite element method.
Many addressed to the structural solid mechanics and the problems as-
sociated with those designs. Some software have a multiphysic point of
view and will include the magnetic and electrical modules into that, others
have a more direct connection to the electromagnetic area. They are usu-
ally easy to adapt considering the requirements a specic problem might
need. Many have a platform you can expand with modules adaptable to
the problems you want to solve. Another feature is the material libraries
that usually holds a wide range of materials. There is also the possibility to
add your own materials or have the support to maintain the materials you
might need. The drawback is that most of the software are expensive. The
modules added all have an additional cost, the support has a fee for ev-
ery year beyond the rst year that usually is included in the purchasing fee.
There are also free software that are focused on solving a specic prob-
lem or a specic class of problems. Their material libraries might also be
limited in number of available material constants. It is usually easy to add
specic materials but you do not have the support team as you have with
the strictly commercial software.
5.1 COMSOL Multiphysics 3.5
The COMSOL Group was founded in Stockholm 1986. They provide soft-
ware solutions for multiphysics modeling. The software has several mod-
ules, depending on your problem you might not need all of them. The
module evaluated is the AC/DC-module, the COMSOL group has also a
heat transfer module that can be of interest in some of the problems pro-
21
22 The software
vided by the commissioner.
5.1.1 Calculating using the AC/DC-module
This module is suitable for simulating dierent designs such as coils, capac-
itors, inductors, and other designs within the electromagnetic area. You
can model your design in both 1-D, 2-D and 3-D, also the possibility for
an axial symmetric model of the problem is available for 1-D and 2-D. You
can also choose between dierent stages, such as static, transient, quasi-
static and time-harmonic simulation. This module covers dierent types of
electromagnetic eld simulation. The magnetostatic problem solution has
been used in this case. The module also has a large material library with
several of the most important constants predened for dierent materials.
The module has several dierent solvers depending on the partial dier-
ential equation. There are two that applies to our kind of design problems,
the stationary and time-dependent solver. The module also takes into con-
sideration if the problem is linear or non-linear.
It is possible to couple the problems with both the electrical eld and
the magnetic eld. The equations used for solving these problems are built
on the Maxwells equations. Amperes law for static cases is used as follows:
H = J
By dening potentials for B, E and J, adding the constitutive relationship:
B =
0
(H+M)
Amperes law is rewritten as follows:
(
1
0
AM) v (A) + V = J
e
When solving for a time harmonic equation Maxwell-Amperes law includ-
ing displacement currents is used. By adding another constitutive relation-
ship D =
0
E Amperes law is again rewritten and the equations used for
solving V and A will be as follows:
((j
2

0
)Av (A) + ( + j
0
)V (J
e
+ jP)) = 0
(j
2

0
)A+(
1
0
AM)v(A)+(+j
0
)V = J
e
J
e
denotes the externally generated current density derived from the
applied current. v is the velocity of the conductor and M is the magnetiza-
tion vector in unit A/m. Comsol Myltiphysics 3.5 will use these equations
5.2 QuickField 5.7 23
to solve for the potentials. For the time harmonic equations the solutions
calculated is not the physical eld but a phasor. By specifying a phase
factor at the result page we can obtain the result for a given time. [3]
5.2 QuickField 5.7
QuickField is developed by Tera Analyses Ltd. The company was created in
1999 and has its headquarters in Denmark. QuickField has a wide range of
modules to choose between for dierent engineering problems and designs.
The AC magnetics module solves problems with regard to time-harmonic
magnetic elds inducing proximity eects, eddy currents and other param-
eters that is of interest in switched power supply.
5.2.1 Calculating using the AC magnetic module
QuickField works in this module with equations that diers if the problem is
made in a planar mode or an axial symmetric mode. The equations are also
constrained by Maxwells equations and consists of two partial dierential
equations. Through a multiple equations derives these following equations:

x
_
1

y
A
x
_
+

y
_
1

x
A
y
_
igA = j
0
for planar equations and

r
_
1
r
z
(rA)
r
_
+

z
_
1

r
A
z
_
igA = j
0
for axial symmetric equations. g stands for electrical conductivity and
x
and
y
(
r
and
z
) stands for the magnetic permeability and represents
constants within the specic model. j
0
denotes the current density and is
assumed to be constant in the planar case and vary as 1/r in the axial
symmetric case.
QuickField has the ability to handle non-linear magnetic materials in
the models and you are able to dene the B/H curves that species the per-
meability for these materials. QuickField applies the equations by ignoring
the displacement current in Amperes law (
D
t
) in the calculations made
with these equations. Assuming that the displacement current density is
not signicant until we reaches higher frequencies (< MHz).
24 The software
5.2.2 Calculating using the Transient Magnetics module
This module applies when we are calculating with a time-varying current
as input. The module uses formulas derived from Maxwells equations for
vector magnetic potential A(B = curl A) and scalar electric potential
U(E = grad U):
curl
1

curl A = j +curlH
c
j = gE = g
A
t
ggradU
where
1

is an inverse permeability tensor and g is the electrical conductivity.

In the second equation the vector j is considered a combination of the
external current source and the induced eddy current
j = j
0
+j
eddy
where j
0
= ggradU and j
eddy
= g
A
t
.
The manual walks through a few more derivations and nally the formulas
used to solve the problems are dened as follows:

x
_
1

y
A
x
_
+

y
_
1

x
A
y
_
g
A
t
= j
0
+
_
H
cy
x

H
cx
y
_
for the planar case and

r
_
1
r
z
(rA)
r
_
+

z
_
1

r
A
z
_
g
A
t
= j
0
+
_
H
cr
z

H
cr
r
_
for the axial symmetrical case.
x
and
y
(
r
and
z
) are magnetic per-
meability tensor components. H
cx
and H
cy
(H
cr
and H
cz
) are coercive
force components. All within each block of the model. j
0
is assumed to
be constant in the planar case and
1
r
for the axial symmetric case. In this
module can use custom dened B/H curves to determine the permeability
component in the magnetic materials. [11]
5.2.3 Heat Transfer module
This module is used to couple the problem rst solved in the AC magnetic
module. QuickField is able to solve problems in both linear and non-linear
temperature elds, these elds are in this thesis model cases decided by the
material in the core. For the non-linear case the equation is as follows:

x
_
(T)
T
x
_
+

y
_
(T)
T
y
_
= q(T) c(T)
T
t
5.3 FEMM 4.2 25
This equation is only for the planar case where T stands for temperature,
t denotes the time, (T) denotes the heat conductivity as a function of
temperature, q(T) denotes the volume power of heat sources, c(T) denotes
the specic heat and denotes the density of the material. [11]
5.3 FEMM 4.2
Finite Element Method Magnetics, FEMM, is a package for solving prob-
lems in 2-D, planar or axial symmetric. The package is used for relatively
low frequencies. The package is developed by Ph.D. David Meeker in 1997-
98. The package addresses some limiting cases of Maxwells equations.
The magnetics problems the package is intended for is in a relatively low
frequency area, below radio frequencies. Approximately frequencies below
300 kHz.
5.3.1 Calculating time-harmonic magnetic problems
In the cases when the magnetic eld varies with time we will probably have
induced eddy currents in materials with a non-zero conductivity. Since
Maxwells equations has to be obeyed we will have a relationship between
the electric eld intensity E and the current density J :
J = E
with induced electrical eld that obeys:
E =
B
t
These are the initial equations FEMM uses, by deriving through mul-
tiple equations FEMM will use following equation to solve the problems in
time-harmonic magnetic state:

_
1

eff
(B)
a
_
= ja +

J
src
V
where

J
src
denotes the phasor transform of the current applied sources. a
is derived from the equation : A = Re[ae
jwt
]. Interpret this as will be
constant for harmonic problems. FEMM can approximate the eects of
saturation on the phase and amplitude of the eld distribution. [5]
26 The software
5.3.2 Post processing
In the post processing mode of FEMM there are several ways to display
the result of the calculations. By default the software shows the geom-
etry with the magnetic eld displayed. The user may choose to display
dierent solutions over the whole surface. The magnetic eld density, the
current density and the magnetic eld intensity can be displayed through
a coloured graph. By choosing the surface domains of interest the user will
be able to calculate interesting values, for example the total power loss and
the inductance level.
The most relevant equations for this thesis is the integral for total power
loss and the inductance level. The equation for power loss is in FEMMs
user guide not dened but it is a total calculation of all the losses possible
over the chosen section. By choosing the block integral A.J the user will
be able to calculate the inductance level.
L
self
=
_
A JdV
i
2
The equation gives the result in the unit H/A
2
, by dividing with the
squared current the real inductance level is obtained. [5]
5.4 Ansys R Ansoft Maxwell Student Version 9
Maxwell is a part of the Ansoft product suite owned by Ansys, Inc. Ansys,
Inc. was founded in 1970 and has 1600 employees of today. Ansys develops
a wide range of simulation software for dierent environments, for example
uid mechanics, electro magnetics and structural mechanics. Ansys head-
quarters is located in Pennsylvania, U.S.A. and has channel partners in
more than 40 countries in the world. The Ansoft product suite consists
of a wide range of modules, specied for both high-performance electronic
design and electromechanical design.
5.4.1 Calculating using the Eddy Current solver
In Maxwell this solver uses equations derived from Maxwells equations.
The eddy currents in the model are calculated by using following equation:

(A) = ( + j) (jA)
where A denotes the magnetic vector potential, denotes the electric scalar
potential, is the relative magnetic permeability, denotes the angular
5.4 Ansys R Ansoft Maxwell Student Version 9 27
frequency, denotes the conductivity of the material and stands for the
relative permittivity.
Maxwell has the ability to handle a non-linear material with B- and H-
elds that varies over the frequencies. Maxwell will calculate the eective
values of B and H from the B/H-curve for the magnetic material. This is
because will dier with the frequencies according to following equation:

eff
=
B
e
H
e
Maxwell will get the eective values of B and H by these following equation
for B
e
:
B
e
=
2
3
n

i=1
f(
i
) + f
_

i
+
i+1
2
_
+ f(
i+1
)
and for H
e
:
H
e
=
2
3
_
n

i=1
f(
i
) + 4f
_

i
+
i+1
2
_
+ f(
i+1
)
_
In the set up parameters for calculating core loss the software will use
following equation for the ferrite core loss coupled with the eddy current
problem:
p = C
m
f
x
B
y
max
Where C
m
denotes the constant value determined by the calculations in
the eddy current solver, f denotes the frequency and B
max
is given by the
maximum amplitude of the eld density. The x- and y-exponents derives
from the Steinmetz equation 4.1.
Chapter 6
The models to be simulated
In this chapter the models provided by the commissioner and the theory
behind them are described.
6.1 Commutation choke coil
This is an inductor with a foil winding. Very thin layers of copper with
an even thinner layer of insulation material in between the turns. This
whole model, where the commutation choke coil is a part, were designed
to accommodate a fairly large power output, 12 kW. When calculating the
power losses in the commissioners software the result shows no issue with
power losses around the air gap. In reality after a prototype had been made
the power loss around the air gap in the copper wire became much larger
than estimated. In fact the power loss got so high that the winding melted.
The commissioner later tested to solve this problem in reality by applying
a litz wire with few strands in the winding, the nal solution became a litz
wire with more than 500 strands in the windings.
6.2 Asymmetric power inductor
This is a power inductor with only 2.5 turns in the winding. Usually the
design is made with more turns in the winding and more common is also
to use full turns. When testing this prototype the commissioner noticed a
problem with the inductance. The measured value of the inductance was
much higher than the estimated. The reason is probably the half turn and
the small number of turns in the winding. To get the correct value of the
inductance the manufacturer had to grind down the middle leg of the E-
core to get the correct value of the inductance. At the end the air gap in
29
30 The models to be simulated
the core became so large that the middle leg in the E-core were too small.
The commissioner took this core and measured the inductance value at
dierent half turns and full turns. The result from these measures were a
clear picture that half turns increased the inductance while adding another
half turn to get a number of full turns the result was an inductance level
you could expect out of sheer experience.
6.3 Flyback transformer
This is a transformer with a bias winding, a primary winding and a sec-
ondary winding that consists of three windings. The three secondary wind-
ings consist of one winding of copper with the same diameter as the primary
winding, but less turns, and two windings with signicantly smaller diam-
eter than the rst secondary winding. Also the rst secondary winding is
done in parallel. The commissioners software will calculate the problem by
adding the proximity eect losses to each other. The problem is the shape
of the current that initially is a trapezoid current. The commissioners soft-
ware will handle the current, not as a continuous current, but will divide
the current shape into two parts with the amplitude reaching zero at the
half period and then rise the current to the wanted value again. In reality
the current will never be zero, so the desired behaviour is a software that
can apply the function of the current, over time for a full period. This be-
haviour in the model is called continuous mode, the commissioners software
interpreted the behaviour as discontinuous mode.
The commissioners software overestimate the proximity eect, due to
the inability to calculate with a time-dependent function of the current.
The estimated proximity eect, with the commissioners software, is a sum
of all the power losses in the primary winding and the three secondary
windings. It does not correspond to the true eect losses since the current
wont reach zero at any time. In comparison, the commissioners software
estimates the power loss in the bias winding to more than 2 W, while
QuickField gives an estimated power loss as low as 0.2 W. This is much
more reasonable compared to the measured value of the power loss in the
transformer.
6.4 Planar transformer
A planar transformer diers from the other designs by the way to apply
the windings. They can be stand alone components, designed according
6.4 Planar transformer 31
to a stacked layer approach or they can be integrated into the printed cir-
cuit board, PCB. This allows for a low prole, the thermal characteristics
becomes very good and the leakage inductance is small. [6, p.3] The task
for this model is to evaluate if it is possible to calculate the power loss in
the transformer. The design of the model is fairly simple to CAD into the
software.
The input signal to this model is a square wave with a current amplitude
at zero in between the positive and negative part of the wave and at the end
during one period. This requires a transient analysis to be able to calculate
the correct core losses in the model.
Chapter 7
Simulations
In this chapter the simulations made with the dierent software available
are described. A brief description how to build and solve the models in
each software are also included.
The simulations made are done with design and problems from a number
of test-cases provided by the commissioner. They are built in the dierent
software chosen to be evaluated and the result are compared between the
software. This is to establish if the results are similar but also this will give
answers regarding the maintainability and usability of the software.
7.1 Material properties
In all the software some material properties has to be dened. These are
dened by the manufacturers data-sheets for the specic material and by
the design requirements from the commissioner. It is of essence to provide
the correct B/H relationship to be able to simulate without changes made
in too many variables. For the common materials such as air and copper the
softwares material library usually has predened values on these material
properties.
7.2 Simulating using FEMM 4.2
7.2.1 Building the models
The models simulated is drawn with the tools built into the software.
FEMM has 5 modes in the draw section, Point mode, Segment mode,
Arc Segment mode, Block mode and Group mode. To dene a geom-
33
34 Simulations
etry in FEMM the user states the coordinates of a cross-section of the
model intended to be simulated. This is done by using the point mode and
segment/arc segment mode.
7.2.2 Physics settings
To assign the right properties to the dierent parts of the model we will use
the block mode, this denes labels and properties for each part/block of
the model. FEMM carries a small material library with the most common
material constants. It is easy to dene a new material for the ferrite in the
core by applying the B/H curve. The user can also import the materials
for another model. This is to prefer when dierent designs might use the
same custom made material. It will not be necessary to dene the custom
made material more than once. The user need to tell the software where
the model ends, else the user will get a result with calculations made for
innity boundaries. The correct current applied is also set, all this is done
from the properties menu.
7.2.3 Meshing and solving
FEMM has an built in mesh-tool that will mesh the geometry automatic.
The highest resolution of the mesh needs to be around the critical areas
of the geometry. A small geometric detail will need a higher resolution
since this is more important to solve accurately than a large space of air
surrounding the actual model. How many elements that are made and how
many degrees of freedom the problem will be solved for is limited by the
memory capacity. FEMM solves the problem with a predened model and
for the user it is very simple, just to press the solve button in the user eld.
7.2.4 Post processing
For post processing in FEMM there is a new window opened and as default
the ux lines of the magnetic eld is shown. This gives an initial picture of
the behaviour of the design. In the post processing area the user can choose
to calculate dierent block integrals or line integrals to see what happens
from a surface point of view or a cross-section point of view. When dening
the physics settings it is preferable to group the blocks of the same property
since this makes it much easier to calculate, for example, the total power
losses in the windings.
7.3 Simulating using COMSOL Multiphysics 3.5 35
7.3 Simulating using COMSOL Multiphysics 3.5
7.3.1 Building the models
In COMSOL the geometry is built with dening dierent blocks of the
model. Instead of dening coordinates for the geometry it is built by uniting
rectangles and circles and other geometrical parts to complete a design.
The user interface consists of several modes to complete the design, Draw
mode, Boundary mode, Subdomain mode and Point mode. In draw mode
there are many tools for the user for dierent geometrical shapes, it is
easy to scale the geometry and unite dierent blocks into one solid block
representing for example the core.
7.3.2 Physics settings
COMSOL has a lot of ways to dene custom variables to solve the problems
in the model. In subdomain mode the user assign the material constants
needed. COMSOL carries a large library of dierent materials but it is also
easy for the user to dene materials of their own. The way to apply the
correct current diers from the other software. When dening the subdo-
main for a leading copper coil the user needs to apply the correct external
current density instead of the current solved for. Thus it is necessary to
calculate the correct area of the cross-section of the coil and then apply the
right value. In boundary mode it is easy to apply the correct values on the
boundaries restraining the magnetic eld area.
7.3.3 Meshing and solving
Meshing and solving is dened by default but the user is able to dene
customized size on the mesh and also can pick the dierent algorithms
to solve the problem. Again the restraint on the solving procedure i the
number of freedom degrees the linearisation algorithms solve for. COMSOL
has, compared to the other software by default a denser mesh and higher
expectations on the memory capacity. During the solving, the software has
run out of memory during the LU-factorization (Lower-Upper) the matrix.
This can be solved using a coarser mesh but the user must be aware this
might aect the accuracy of the result.
7.3.4 Post processing
In the post processing mode by default the magnetic eld density will be
plotted. Again it gives the user an initial understanding of the behaviour
36 Simulations
of the model. It is also possible to plot each subdomain with the dierent
areas of interest the user choose. The areas of interest diers compared
to the other software. The equations for solving for power losses will be
given in the unit of W/m
2
, to get the real value the user need to perform
a calculation by multiplying by the volume of the desired block area where
the power loss of interest occurs.
7.4 Simulating using QuickField 5.7
7.4.1 Building the models
In QuickField the user build the model with nodes and edges. It is also pos-
sible to insert block-shapes but coordinates for these blocks are measured
from the center of the block, compared with COMSOL that uses the coor-
dinates for the lower left corner for a rectangle. It is very easy to import
a dxf-le (drawing exchange format) from an already made CAD (Com-
puter Aided Design). This saves some time since the user can approach the
block properties mode at once since the CAD tools in QuickField demands
another way of coordinating the blocks.
7.4.2 Physics settings
QuickField does not carry a material library as the other software do. The
user has to dene all the materials used in the model. It is possible, as
in the other software, to dene the custom materials used. It is easy to
maintain the B/H curve of a ferrite and all the necessary constants are
easy to apply. To apply the correct material constants and current all is
dened in the block properties window. QuickField has another approach
to apply the material constants and the correct properties for each domain
in the model. Where the other software use domains, QuickField will name
the same area as blocks. With a double-click the window for naming each
block will appear. When named the new block will appear in the branch
data in the tree to the left in the main window. From here it is easy to
apply the properties of every block label. For the transient analysis the
user needs to dene a time-dependent signal input described as a function
or a constant signal input.
7.4.3 Meshing and solving
In Quickeld the meshing and solving area the procedure is very similar
to the other software. A built-in mesh tool is available and it is possible
7.5 Simulating using Ansys R Ansoft Maxwell Student Version 9
37
for the user to dene the resolution of the meshed net for the design. The
areas that will be meshed can not be to narrow since this will either make
it impossible to mesh or make a mesh too dense for the solver to process in
the calculation. In Quickeld the mesh size can be locally dened by the
user at specic edges, nodes and areas.
7.4.4 Post processing
QuickField will by default display the eld picture of the current density.
There is a variety of menus for the post processing and QuickField will
give the mean, peak and root mean square values of the dierent integral
values. By choosing the contour, area or eld, of interest QuickField will
be able to calculate the values of interest. For a graphically display to be
plotted, choose a contour of interest. This plot can for the inductance level
be calculated in two dierent ways. The post processing mode has a tab in
the calculation menu for the inductance level calculation where a step by
step instruction helps the user to get the value of the inductance. It will
be either the ux linkage value or the stored energy value.
7.5 Simulating using Ansys R Ansoft Maxwell Stu-
dent Version 9
7.5.1 Building the models
In the section dene model, the draw options are models or grouping ob-
jects. Important thing to do before the user start to draw is to choose a
matching draw unit for the model. In 2-D draw modeller there is a few
handy tools to the users help, straight lines, rectangular/circular object
and move/turn object. When designing the model the user can write the
coordinates or directly give the edge intended the right length. When the
design of an object is nished the user has to apply a matching name and
color.
7.5.2 Physics setting
In material set up the user assigns the material properties to the objects
created. Maxwell has a large library over the most common materials.
The user can add a custom material, dene the B/H-curve and add the
material properties. The next step is to dene the model with boundaries
and voltage or current sources. In these models the windings will be current
sources and the surrounding area will be the boundary to avoid an innite
38 Simulations
model, the user need to create a so called "balloon". To calculate the core
loss, AC resistance and the inductance, the options will be found in the
executive parameters. In the set up for total core loss the user needs to
dene some parameters, before the calculations can be executed. For the
calculation of the core loss the power ferrite needs to be chosen, there is
also some properties that has to be dened in this area.
7.5.3 Meshing and solving
The user can choose to mesh the model manually or automatic. The mesh
size can be customized as desired. To get a relatively good accuracy the size
of the nite element mesh is of importance. The frequency for the problem
is by default 20 Hz, the user is free to change the frequency to match the
model. Before solving the right source for the problem, by default it is an
initial source but the user need to change it to the current source matching
the problem intended to be calculated.
7.5.4 Post processing
It is wise to start the post processing mode by checking the convergence
data. All the steps Maxwell needed to solve the problem and how fast
the problem converged is displayed here. The energy error in percentage
and the total power loss in watt is also available here. In the solution tab,
the user will nd the results from the options in the executive parameters
area. The graphical solution can be reached from the post process tab, the
user needs to create an object list over the area that is interesting before
watching the result. This will make it easier to get a good overview in
the area of interest. For the current density J, the magnetic eld intensity
H and the magnetic eld B, there are several options to choose between.
The correct option for these models were the phase-related options. These
options are found below the plot and eld tabs. The user will reach the list
over the geometry created previously.
Chapter 8
Results
In this chapter the results from the simulations are declared, the software
are compared towards each other and the dierence between the calcula-
tions made are shown.
8.1 Commutation choke coil
The commutation choke coil had a entirely dierent problem. The losses in
the windings proved to be much higher than the estimated value calculated
with the software the commissioner uses. The real power loss became a lot
higher than expected.
The results of the simulations made by the software we choose to use,
showed that the power losses in the windings became large. The simulated
power loss seems likely to correspond to reality. The design is modelled
in both axial symmetrical and planar mode. These modes gives dierent
results. The planar mode in the software calculating for the power loss
corresponds to each other while the axial symmetrical mode gives a higher
power loss. For the planar mode the power loss became 24.6 W.
39
40 Results
Figure 8.1. The magnetic eld and the current density plot. Planar mode.
This behaviour conrms the physical result when running a prototype.
The simulation clearly states that the problem is the concentration of the
current density around the air gap. Due to this concentration the power dis-
sipation becomes very high and the design is not suitable for this problem.
In a close up the concentration is clearer, considering the thin copper layers
and this behaviour in the magnetic eld, the result is not a big surprise.
Figure 8.2. Concentration of the current density around the air gap. Planar
mode
We have also simulated an axial symmetrical model of the commutation
choke coil. The calculated power loss becomes signicantly higher, but the
behaviour of the magnetic eld conrms the result when modeling in a
planar system. The close up of the current density and magnetic eld plot
shows the same result.
8.2 Asymmetrical power inductor 41
Figure 8.3. The concentration of the current density in axial symmetric mode
The calculated power loss became as high as 34.2 W compared with
the power loss in the planar mode. This result might be an impact of the
higher core volume since the calculations will be made of the design at a
360

around the vertical axis, the volume of the copper wiring will of course
also be higher so you might expect the result will be higher in the axial
symmetric mode.
8.1.1 Calculating the temperature
The problem calculated in the AC magnetic module were coupled with the
heat transfer module to get a temperature value in the area around the air
gap. As a result the temperature were calculated to more than 500

C in
the area around the air gap. If this is plausible then the physical result in
the prototype is not a surprising behaviour.
In an early state of the thesis work it was decided to not design all the
test models in COMSOL Multiphysics, this decision was taken on the basis
of the results calculated. Compared with the other software the results for
the commutation choke coil were not conrmed in COMSOL. Though the
magnetic eld behaviour was similar all the other calculated values were
signicantly dierent.
8.2 Asymmetrical power inductor
As described earlier in the model chapter the problem for this power in-
ductor were the behaviour of the model when the winding was made only
42 Results
with a few half turns. This causes a asymmetrical behaviour in the mag-
netic eld. The commissioners software estimated an air gap around 0.5
mm, in reality the needed air gap became seven times higher to reach the
desired value of the inductance. When simulating the original design in
our evaluation software we clearly got a picture of a magnetic eld that did
not behave as expected. The magnetic eld became much denser on the
side where the extra half turn was added and the expected behaviour in
the middle leg became oddly stretched as shown in the gure below.
Figure 8.4. The original design made with a small air gap at the center leg of
the E-core.
The magnetic eld is clearly denser at the side with the extra half turn.
The magnetic eld behaviour in the middle leg is oddly shaped. The simu-
lated values of the inductance became just over three times larger than the
calculated value in the commissioners estimation.
The next simulation became the prototype with the large grind down
middle leg of the E-core. Again the result were unexpected. The magnetic
eld did not cross the air gap in the middle leg, instead it circulated in the
remaining core, creating a magnetic eld that totally enclosed the windings
and raised the inductance. The simulated inductance values became ve
and a half times higher the originally calculated value. The magnetic eld
behaviour in the gure below shows no interaction in between the windings.
8.2 Asymmetrical power inductor 43
Figure 8.5. The resulting design to reach the desired value of the inductance.
The last simulation was made with air gaps over all the legs, the air
gap at the end was calculated as half the air gap in the middle. The
result became as you would expect the magnetic eld to behave from the
beginning. As shown in the gure below the magnetic eld creates two
elds, one around each winding of the inductor.
Figure 8.6. The nal design with an air gap on all three legs in the core.
The simulated values of the inductance level became exactly as the mea-
sured values in reality, thus conrming the correctness of the simulation and
reliability of the calculations of these problems.
The software simulated almost identical values of the inductance as well
as the behaviour of the magnetic eld. Since these results with the reference
current proved to be accurate we continued to simulate with the currents
44 Results
originally intended. The result is shown with gures in appendix. For the
design with air gap overall the legs in the core, the inductance level is the
desired value, the same result appears for the design with the very large air
gap in the middle of the core.
8.3 Flyback transformer
The simulation with this model was entirely dierent from the others, since
the current is dependent on time we had to apply the parameters for a
transient analysis. This was only possible to do in QuickField since the
requirement to apply a current with the specic waveform could be done
in no other software. Since most of the software has by default a ideal sine
wave as the current waveform they did not meet the requirement.The aim
for this calculation was to see the proximity eect in the bias winding.
Figure 8.7. The magnetic eld when the primary winding connects.
The main problem with this design was the input- and output-signal,
the current never becomes zero during the whole period. This was solved by
modelling the current waveform in QuickField and simulating a transient
analysis. The calculations became very time consuming since the calcu-
lations are made by calculating the whole model for every step in time.
Compared with the initial calculated power loss in the bias winding the
8.4 Planar transformer 45
result from QuickField is a much better approximation of the power loss.
Figure 8.8. The current density when the primary winding connects.
8.4 Planar transformer
This simulation also diers from the other models. Since there will be a
current in both the primary and secondary coils simultaneously we need
to estimate the magnetizing current in the primary coil for the simulation
with FEMM. For the simulation with QuickField it was possible to either
apply a voltage level to the primary coil or apply the same magnetizing
current as in FEMM.
The initial simulation of this model gave the results of an inductance
level similar to the real values. When calculating the power losses the
copper losses seemed to match the calculations done in the commissioners
software. The core losses on the other hand became signicantly smaller
than the estimated, this is probably a because the input signal does not
consists of a pure sinusoidal.
46 Results
Figure 8.9. Planar transformer with magnetizing current on the primary side.
Chapter 9
Conclusion/Discussion
During this thesis work we have reached the intended goal of our work.
By evaluating the software from the commissioners point of view and sim-
ulating the test models provided by the commissioner some main features
regarding the dierent software has emerged.
9.1 FEMM 4.2
This software is developed from a perspective of the electromagnetic prob-
lems. It is very straight forward when dening the problem and design the
geometry of the intended problem. The library is easily maintainable and
any custom material added by the user is easy to use in related designs
with another geometry. It can only execute models drawn in 2-D but the
graphically interface in the post processing mode is very easy for the user
to understand. Overall the software is easy to work with, has a straight
forward approach to dene the problems and has short training curve for
a new user.
9.2 COMSOL Multiphysics 3.5
This software is developed in a from a multiphysical point of view. The
strength lies in the ability to couple dierent problems to the same geom-
etry. The user will be able to model the geometry in both 2-D and 3-D,
though the 3-D simulation might be very memory-intensive since the cal-
culations might be very complex. The material library is vast and it is easy
for the user to customize a material not included in the library. Once done
it is very easy to use the dened material for another design if wanted.
47
48 Conclusion/Discussion
COMSOL Multiphysics can couple the problems with the modules of in-
terest, for example the AC/DC module is able to provide data for the heat
transfer module for calculating of the temperature. The graphically inter-
face in the post processing mode is easy to understand, but the equations
used for the integral solutions might need a re-calculation by the user to
get the total power loss for example. Overall COMSOL Multiphysics 3.5
is a highly qualied work tool but will need a long period of training for
the commissioner to be able to use all the included features eciently. The
COMSOL Group has developed a new version of this software, this version
is easier to understand and more intuitive in the work area but will still
need a lot more training to be able to execute the software eciently.
9.3 QuickField 5.7
QuickField is the other software that are only capable of calculating models
in 2-D, it is straight forward when dening the geometry. QuickField is
capable of couple simple problems to each other, such as the calculating
of the magnetic eld behaviour and the thermal heating generated from
the earlier result. It is easy to maintain the customized materials and its
parameters but it can not be saved for future models. The graphically
interface in the post processing mode gives the user a lot of information on
the parameters calculated. The user will have a shorter training curve to
learn the basics of the software and to be able to eciently model dierent
test cases.
9.4 Ansys R Ansoft Maxwell Student Version 9
Maxwell is similar to QuickField, the dierent features includes electro-
magnetic model, heat transfer model but also a structural mechanics. It is
possible to couple the problems to each other to calculate a heat transfer
depending on the current density in the model. Maxwell has the possibility
to design the geometry in both 2-D and 3-D. Though similar to COMSOL
Multiphysics, the calculations might demand a lot of memory and time.
The material maintainability is easy, but the custom dened material will
only be saved locally. In other words, for another problem the custom fer-
rite must be dened again for a new problem. The post processing mode is
slightly more complicated to produce the data of interest for the user. Alto-
gether Maxwell is qualied from the commissioners point of view, but will
probably need a signicant longer training period compared to QuickField
and FEMM. To design the geometry in 3-D, to do heat transfer analysis
9.5 Discussion 49
and to do the transient analysis the full version of Maxwell is needed.
For the results in the post processing mode there is a very brief guide
available, also to view the results the user need to write a short code to
extract the correct values from the calculations.
9.5 Discussion
Comparing the software from the commissioners point of view gives the
impression that all four software evaluated is capable of at least calculating
the power losses in the models provided for the thesis. FEMM is the only
software not able to couple the dierent problems. All the software calcu-
lates the test designs in a similar way, by studying the results graphically,
all the software conrm each others result.
For the commissioner the recommendation will be either to use FEMM,
specic for the type of problems solved in this thesis it is highly qualied,
its main drawback is the lack of ability to couple the problems and only
2-D geometry. The other software to be recommended is QuickField, the
drawback with this software is similar to FEMM, it can only model the
geometry in 2-D, but it is possible to couple the problems from AC/DC-
problem solving with heat transfer to get the temperature rise in the ma-
terial. The main conclusion is that QuickField cover a wider range of the
commissioners needs. As mentioned above both Ansys R Ansoft Maxwell
and COMSOL Multiphysics are highly qualied for these kind of simula-
tions, both are able to couple the problems with dierent modules and both
can model the geometry in 3-D but the long training curve does not meet
the commissioners requirement.
The results for the simulations on the test models the commissioner
provided has been quite satisfying.
9.5.1 Commutation choke coil
When calculating this model the current density became very high around
the air gap. There are a number of theorys to elaborate on this model. The
shape of the windings might be modied in the cross-section. The shape of
the windings can be optimized by a program developed for MATLAB, this
program determines the radius for a semi-circular cut out for the winding
around the air gap. [10] The commissioner has tried a variation of this way
to design the windings in this model, but the result was not good enough.
50 Conclusion/Discussion
9.5.2 Asymmetrical power inductor
This model was the least complicated design to model in the dierent soft-
ware. All the software calculations conrms each others results. This model
was the only one to be completed, there is no more optimization to be done
and the commissioner is very satised with the result.
9.5.3 Flyback transformer
This model became the most complicated in execution for all of the mod-
els. The equations for the signal input in the transient analysis and the
time needed for the calculation. For one solving period the time consumed
extended six hours on a highly capable computer with more than one core
in the processor. The preference from the commissioner were to see if the
power loss in the bias winding was possible to calculate. The results seems
to be satisfying and QuickField are able to get a result concerning the
copper losses in the bias winding.
9.5.4 Planar transformer
These calculations are not optimized since there is a lack of time for the
thesis work. The results calculated for the inductance level and the copper
losses seem to give a fairly accurate result there were no time to execute
the transient analysis needed to get an accurate value of the core loss.
Bibliography
[1] Ansys R Maxwell Student Version 9 Users guide, Retrieved from the
software help
[2] van den Bossche, Alex & Valchev, Vencislav Cekov (2005), Inductors
and transformers for power electronics, Boca Raton : Taylor & Francis
[3] COMSOL Multiphysics 3.5 (2008), AC/DC Module Users guide,
[www] http://www.comsol.com Retrieved: 2010-10-12
[4] ETAL Group AB, [www] http://www.etalgroup.se, Retrieved:
2010-09-15
[5] FEMM 4.2 (2005), Users manual. [www] http://www.femm.info/
Archives/doc/manual42.pdf, Retrieved: 2010-10-12
[6] Ferroxcube, Design of planar power transformers. [www] http://www.
ferroxcube.com/appl/info/plandesi.pdf, Retrieved: 2010-11-01
[7] IEEE Global History Network (2008), Maxwells equations. [www]
http://www.ieeeghn.org/wiki/index.php/Maxwells_Equations,
Retrieved: 2010-10-12
[8] Nordling, Carl & sterman, Jonny (1980) Physics Handbook for Sci-
ence and Engineering Studentlitteratur, Lund Eighth edition (2006)
[9] Pelosi, Giuseppe (2007), The Finite Element Method, Part I: R.L.
Courant [Historical Corner]. [www] http://ieeexplore.ieee.org/
xpls/abs_all.jsp?arnumber=4263187&tag=1, IEEE Antennas and
Propagation Magazine. Volume 49. Issue 2; Pages 180-182, Retrieved:
2010-10-12
[10] Pollock, Jennifer D. & Sullivan, Charles R. (2005) Mod-
elling Foil Winding Congurations with Low AC and DC Re-
sistance [www] http://ieeexplore.ieee.org/xpls/abs_all.jsp?
51
52 Bibliography
arnumber=1581830&tag=1, IEEE Power Electronics Specialists Con-
ference, 2005. Issue date 2005-06-16; Pages 1507-1512, Retrieved:
2010-10-12
[11] QuickField 5.7 (2009), Users guide. [www] http://www.quickfield.
com/free_doc.htm, Retrieved: 2010-10-12
[12] Silvester, P.P. & Ferrari R.L. (1990), Finite elements for electrical
engineers. Cambridge University Press. 2nd edit.
Appendix A
A.1 Commutation choke coil
Figure A.1. The commutation choke coil graphically displayed in FEMM
53
54
Figure A.2. The commutation choke coil graphically displayed in QuickField
A.1 Commutation choke coil 55
Figure A.3. The commutation choke coil graphically displayed in Maxwell
56
A.2 Asymmetrical Power Inductor
Figure A.4. Power inductor with air gap over all legs in the core. Displayed in
FEMM
Figure A.5. Power inductor with air gap over all legs in the core. Displayed in
QuickField.
A.3 Flyback transformer 57
Figure A.6. Power inductor with air gap over all legs in the core. Displayed in
Maxwell.
A.3 Flyback transformer
Figure A.7. Flyback transformer, magnetic eld intensity plot in the secondary
winding. Displayed in QuickField.
58
Figure A.8. Flyback transformer, magnetic eld density plot in the secondary
winding. Displayed in QuickField.
Figure A.9. Flyback transformer, current density plot in the secondary winding.
Displayed in QuickField.
A.3 Flyback transformer 59
Figure A.10. Flyback transformer, current density plot in secondary winding
when the current switches between the primary and secondary winding. Displayed
in QuickField.
60
A.3.1 Power loss graphs in the windings
Figure A.11. Power loss in the bias winding. Displayed in QuickField.
A.3 Flyback transformer 61
Figure A.12. Power loss in the primary winding. Displayed in QuickField.
Figure A.13. Power loss in the rst secondary winding. Displayed in QuickField.
62
Figure A.14. Power loss in the second secondary winding. Displayed in Quick-
Field.
Figure A.15. Power loss in the third secondary winding. Displayed in QuickField.
A.4 Planar Transformer 63
A.4 Planar Transformer
Figure A.16. Planar transformer. Displayed in FEMM
Upphovsrtt
Detta dokument hlls tillgngligt p Internet eller dess framtida ersttare
under 25 r frn publiceringsdatum under frutsttning att inga extraordinra
omstndigheter uppstr.
Tillgng till dokumentet innebr tillstnd fr var och en att lsa, ladda ner,
skriva ut enstaka kopior fr enskilt bruk och att anvnda det ofrndrat fr icke-
kommersiell forskning och fr undervisning. verfring av upphovsrtten vid en
senare tidpunkt kan inte upphva detta tillstnd. All annan anvndning av doku-
mentet krver upphovsmannens medgivande. Fr att garantera ktheten, sker-
heten och tillgngligheten nns det lsningar av teknisk och administrativ art.
Upphovsmannens ideella rtt innefattar rtt att bli nmnd som upphovsman
i den omfattning som god sed krver vid anvndning av dokumentet p ovan be-
skrivna stt samt skydd mot att dokumentet ndras eller presenteras i sdan form
eller i sdant sammanhang som r krnkande fr upphovsmannens litterra eller
konstnrliga anseende eller egenart.
Fr ytterligare information om Linkping University Electronic Press se fr-
lagets hemsida http://www.ep.liu.se/
Copyright
The publishers will keep this document online on the Internet or its possi-
ble replacement for a period of 25 years from the date of publication barring
exceptional circumstances.
The online availability of the document implies a permanent permission for
anyone to read, to download, to print out single copies for his/her own use and
to use it unchanged for any non-commercial research and educational purpose.
Subsequent transfers of copyright cannot revoke this permission. All other uses of
the document are conditional on the consent of the copyright owner. The publisher
has taken technical and administrative measures to assure authenticity, security
and accessibility.
According to intellectual property law the author has the right to be mentioned
when his/her work is accessed as described above and to be protected against
infringement.
For additional information about the Linkping University Electronic Press
and its procedures for publication and for assurance of document integrity, please
refer to its www home page: http://www.ep.liu.se/
c David Hkansson and Jenny Larsson