LTH Iea 1065

EMC Analysis of Electric Drives
Fadi Abdallah
Licentiate Dissertation
Department of Measurement Technology and
Industrial Electrical Engineering
2012
Department of Measurement Technology and Industrial Electrical Engineering
Faculty of Engineering
Lund University
Box 118
221 00 LUND
SWEDEN
http://www.iea.lth.se
ISBN:978-91-88934-57-4
CODEN: LUTEDX/(TEIE-1065)/1-120/(2012)
Fadi Abdallah, 2012

Printed in Sweden by Tryckeriet i E-huset, Lund University
Lund 2012
To the purest, to the kindest and to the most wonderful human
being I got to know..
To my mother
i
Abstract
In order to study conducted ElectroMagnetic Interference (EMI) emissions along
a frequency band, circuit models of main components in the drive system are
considered to be effective means to understand, predict, and control the phe-
nomenon. These models should cover the capacitive and inductive coupling
behaviour of the component along the frequency band of interest. Despite the
dynamic behaviour of EMC-related issues, these detailed circuit models when
used under generic SPICE simulation softwares, are very helpful for design-
related and troubleshooting activities.
In this thesis the capacitive and inductive coupling for a winding on a lam-
inated ferro-magnetic core prototype has been investigated and analysed with
the aid of FEMM software package.
Capacitive coupling analysis resulted in the generation of complex capac-
itance network expressing a certain wiring arrangement. This is achieved by
automatically generating a netlist file which contains capacitance values and
structure to be represented under SPICE simulation software. In the netlist file
the turn-to-turn capacitances (Ctt ) and turn-to-ground capacitances (Ctg ) are
calculated from FEMM analysis through a MATLAB code and linked back to
MATLAB by a Sparse Matrix Representation in order to control the accuracy
and complexity of the resulting network.
Inductive coupling analysis has led to the proposal of a SPICE circuit
model for a single turn of copper winding on iron core. This has been ac-
complished by the quantification of the magnetic analysis output from FEMM,
creating zero-pole-gain models and transfer functions of the SPICE model com-
ponents, and then making use of the pole-zero maps and Bode plots utilities
under MATLAB to analytically propose compensation factors for the com-
plete SPICE circuit model components of a single copper turn, in order to fulfill
low- and high- frequency behaviour.
Given the capacitive and inductive coupling of the winding turns, the re-
sulting SPICE circuit model output is compared against measurements of the
prototype taken by Hewlett Packard (hp) impedance analyzer 4194A where it
showed very good agreement.
ii
iii
Acknowledgements
First and Foremost I would like to express my sincere thanks and gratitude to
my supervisor Professor Mats Alaküla for giving me this opportunity and for
making this thesis work possible. Mats has always been a source of inspiration
and guidance to me. His brilliant ability of actualizing concepts and ideas into
functional systems is something one can always be proud of and thrive for. I
have indeed been privileged to be within Mats domain.
My thanks go to my co-supervisor Dr. Hans Bängtsson for the hours of
supervision he devoted for me during his regular visits to IEA in order to review,
assess and provide relevant feedback to my work. Thanks are also due to Dr.
Avo Reinap for introducing me to the world of FEM simulation, for the time
and effort he has given me despite his tight schedule usually. Without Avo’s key
insights, the magnetic modeling part of this thesis would not have been realized.
Special thanks go to my predecessor and mentor in the EMC field, Dr.
Sabine Alexandersson. Sabine was my mentor when I started my research at
IEA. She shared with me the cream of her experience and pointed me to key
areas in the field including very helpful resources and references, and many
software and hardware tools. Sabine helped me form my reference group. Sabine
has been involved in my project since the start and been supportive all the way
through by her follow ups, suggestions, regular visits and phone calls.
Dr. Per Karlsson is a key person in my research project, not least during
his visits to IEA as a consultant. Per introduced me to the world of Power
Electronics with both theory and applications in mind. Per’s unique style of
supervision made me realize the potential I have in order to search for the
information, build up knowledge and make things happen in reality.
My reference group members represented by Jan Welinder (SP Technical
Research Institute of Sweden), Kjell Attback and Håkan Berg (Volvo Cars Cor-
poration), and Johan Nilsson (Electrical Measurement Department) have influ-
enced the development of this work, not least the suggestion of building up the
simple winding on core prototype by Jan Welinder. They did not hesitate in
sharing their knowledge and experience with me which always added stimulus
and momentum to the work, as well gave the chance for a broad scrutiny and
feedback. Particularly Jan and Kjell; the time and effort you’ve devoted outside
our meeting hours are deeply appreciated!
A special ”Thank you!” goes to Lars Lindgren for his interest and engage-
iv
ment in what I’m doing, for the time and effort he devoted to review my work,
and for the valuable suggestions and tips he provided. Getachew Darge deserves
a big and special ”Thank you!” for all the help, time and effort he provided in
the laboratory. Getachew’s technical experience and problem-solving abilities
are well acknowledged.
As well the effective organization and management of our department, rep-
resented by Dr. Ulf Jeppsson and the rest of the team, deserve an acknowledge-
ment for keeping things in order and making everything smooth for us in order
to focus on our research and teaching duties, not to mention the thoughtful
gestures they always have in mind when it comes to private or public occasions.
Special thanks go to each and everyone of my colleagues and friends at IEA
especially my roommates Yury Loayza (Abu Vincent) and Dan Hagstedt (Abu
Oline) for making the work place a fun place to be at and for all the good and
enjoyable time we had inside and outside the school.
I’m also indebted to all of my friends outside the academic world for remind-
ing me of the enjoyable part of life outside the work domain and for helping me
keep the balance. Many names should be mentioned here but I choose not to
in case I forget any of you! My neighbours in Sakförarevägen-Lund are also
included. I owe you all lots of thanks and appreciation, and because of you life
can have a special and sweet taste.
Last but not least my family deserves the biggest thanks and credits for
their sacrifice, patience and understanding. I wish I could trade all what I have
with couple of days in my mother’s life! My mother Nawal, who passed away
while I was doing this project, my father Wajeeh and my brother Morsi have
given me lots of support and encouragement. With all the love and appreciation
to my family I end these lines.
Fadi Abdallah
P.S. This work is part of the Integrated Charger project which is funded by
the Swedish Hybrid Vehicle Centre (SHC) under the thematic area Electrical
Machines and Drives. This financial support is gratefully acknowledged.
Contents
Abstract i
Acknowledgements iii
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Scope and Goal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Study Roadmap and Thesis Structure . . . . . . . . . . . . . . . 8
1.5 Contribution of This Study . . . . . . . . . . . . . . . . . . . . . 10
2 Analysis Tools 11
2.1 Slot Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 “Macro Level” and “Micro Level” Machine Models . . . . . . . . 15
2.3 Layer-to-Layer Approach for Slot Analysis . . . . . . . . . . . . . 17
2.4 Turn-to-Turn Approach for Slot Analysis . . . . . . . . . . . . . 18
2.5 Simple Winding-on-Core Prototype . . . . . . . . . . . . . . . . . 19
3 Capacitive Coupling Matrix 23

3.1 Electrostatic FEMM Modelling . . . . . . . . . . . . . . . . . . . 23
3.1.1 Capacitance Matrix for Multi-Conductor System . . . . . 24
3.2 How the Capacitive Coupling Matrix is created? . . . . . . . . . 25
3.3 Sparse Matrix Representation . . . . . . . . . . . . . . . . . . . . 31
3.4 From Sparse Matrix to Capacitance Network . . . . . . . . . . . 33
4 Inductive Coupling Matrix 41

4.1 How the Inductive Coupling Matrix is Created? . . . . . . . . . . 52
v
vi CONTENTS
4.2 Complex-Numbered Inductance Concept . . . . . . . . . . . . . . 54

4.3 Single Turn Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Single Turn Model along a Frequency Sweep . . . . . . . . . . . . 68
5 Model Validation 89
5.1 Capacitively Coupled Optimized Validation . . . . . . . . . . . . 93
5.2 Inductively Coupled Optimized Validation . . . . . . . . . . . . . 98
6 Conclusions and Future Work 101
Glossary 103
Acronyms 105
References 110
Appendices 111
A Decibel 113
List of Figures
1.1 Parasitic capacitances in an electric drive system . . . . . . . . . 3

1.2 A photo of the second inhouse-built electric drive labsetup . . . . 4
1.3 Triangular wave and reference voltages, generated Pulse-Width
Modulation (PWM) signals and a varying 2-step/3-step Com-
mon-Mode (CM) voltage waveform . . . . . . . . . . . . . . . . . 6
1.4 The resulting high-frequency damped-sinusoid CM current wave-
form occuring at each CM voltage step . . . . . . . . . . . . . . . 7
2.1 A photo of induction machine stator showing Integral Slot, Single

Layer Winding of Slots per Pole per Phase (q)=3 (36 slots, 3
phases & 4 poles) as well as the End Turns arrangement . . . . . 13
2.2 A side-view drawing of induction machine stator winding showing
Integral Slot, Single Layer Winding of q=3 (36 slots, 3 phases &
4 poles) as well as its End Turns arrangement . . . . . . . . . . . 14
2.3 Macro(slot)-level circuit model of induction machine with Single
Layer Winding of q=3 . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 An unfolded view of the End Turns region approximating the
interface between two adjacent phases in the machine stator . . . 17
2.5 Layer-to-layer approach for sensitivity analysis in frequency do-
main for a 3-layer winding represented by 3 RLC-sections . . . . 18
2.6 Analysis development of the winding portion modelling inside a
slot. First the winding is considered as one complete unit, then
it’s divided into layers and finally is represented by turns . . . . 19
2.7 The 24 conductors arrangement for electrostatic FEMM analysis 20
2.8 3-layer 24 turns winding-on-core model in both distanced and
condensed wiring arrangements . . . . . . . . . . . . . . . . . . . 21
vii
viii LIST OF FIGURES
3.1 The 24 conductors arrangement for electrostatic FEMM analysis

(repeated) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Electrostatic FEMM analysis of 24 conductors on core - con-
densed arrangement . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Electric field lines connecting conductors 11 and 12 in the inner
most layer, the Second Layer . . . . . . . . . . . . . . . . . . . . 28
3.4 Linear scale representation of Capacitive Coupling Matrix for the
system of 24 conductors on iron core - condensed arrangement . 29
3.5 Turn-to-ground capacitance in pico farads [pF] for conductors
in the condensed wiring arrangement, represented by the main
diagonal elements in the Capacitive Coupling Matrix . . . . . . . 30
3.6 Adjacant turn-to-turn capacitance in pico farads [pF] for con-
ductors in the condensed wiring arrangement, represented by the
first diagonal above (or below) the main one in the Capacitive
Coupling Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.7 Logarithmic scale representation of Capacitive Coupling Matrix
for the system of 24 conductors on iron core - condensed arrange-
ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8 Examples of sparse matrices with different threshold values for
the simple winding prototype - condensed arrangement . . . . . . 34
3.9 Sparse matrices with substantially lower threshold limits for the
simple winding prototype, the condensed wiring arrangement . . 35
3.10 The sparse matrix chosen to be fed to the transmission line model
in LTspice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.11 The upper half of the sparse matrix to be fed automatically to
the transmission line model in LTspice . No redundancy included 36
3.12 Capacitance network equivalent to the code in the netlist file
above, drawn manually in LTspice . . . . . . . . . . . . . . . . 38
4.1 Penetration/Skin depth for different types of materials along a

frequency sweep from 0 Hz - 1 MHz . . . . . . . . . . . . . . . . 42
4.2 Frequency versus the corresponding wavelength (λ) in free space
for frequency range from 50 Hz - 30 MHz . . . . . . . . . . . . . 44
4.3 One of the magneto-static analysis runs for the FEMM model of
the winding arrangement on iron core prototype . . . . . . . . . 46
4.4 Iron block property details for AC time-harmonic magnetic anal-
ysis used for the core configuration . . . . . . . . . . . . . . . . . 47
LIST OF FIGURES ix
4.5 Zoomed-in post-processor views of magneto-static and AC time-

harmonic magnetic FEMM analyses, for the 24 conductors on
core - condensed arrangement - at different frequencies . . . . . . 49
4.6 Zoomed-in post-processor views of AC time-harmonic magnetic
FEMM analysis for the 24 conductors on core - condensed ar-
rangement - at several “high” frequencies . . . . . . . . . . . . . 50
4.7 Zoomed-in post-processor views of magneto-static FEMM analy-
sis, showing the top 50% of magnetic field flux lines . . . . . . . 51
4.8 Linear scale representation of Inductive Coupling Matrix run at
0 Hz frequency for the system of 24 conductors on iron core -
condensed arrangement . . . . . . . . . . . . . . . . . . . . . . . 56
1 kHz frequency for the system of 24 conductors on iron core -
2 kHz frequency for the system of 24 conductors on iron core -
4.17 Complex inductance components along the frequency sweep ex-
tracted from the FEMM model of a single conductor in a 24-
conductor winding-on-core prototype . . . . . . . . . . . . . . . . 67
x LIST OF FIGURES
4.18 A logarithmic scale of absolute value of the complex-numbered

inductance in [μH] vs frequency in [Hz] . . . . . . . . . . . . . . . 68
4.19 Different types of losses associated with a single turn in a 24-turn
winding-on-core prototype, along a frequency sweep . . . . . . . 69
4.20 A logarithmic scale of hysteresis and eddy current iron core losses
in [mW] vs frequency in [Hz] . . . . . . . . . . . . . . . . . . . . 70
4.21 A logarithmic scale of the resistive losses in the copper conductor
in [mW] vs frequency in [Hz] . . . . . . . . . . . . . . . . . . . . 70
4.22 A ladder model for a single turn’s ac inductance and core losses
covering a frequency sweep starting from first critical frequency
(fcr1 = 316 Hz) up to 100 MHz . . . . . . . . . . . . . . . . . . . 74
4.23 Pole-zero plots for the RL-branch ladder representing a single
turn’s inductance and core losses, and covering a frequency sweep
starting from 1 kHz up to 100 MHz - Labels in the legend indicate
the start of the decade . . . . . . . . . . . . . . . . . . . . . . . . 76
4.24 Bode plot for the system of RL-branch ladder representing a sin-
gle turn’s inductance and core losses, and covering a frequency
sweep starting from 1 kHz up to 100 MHz - Labels in the legend
indicate the start of the decade . . . . . . . . . . . . . . . . . . . 77
4.25 Pole-zero plots for the system of RL branches and their equiva-
lence representing a single turn’s inductance and core losses, and
covering a frequency sweep of 5 decades starting from 1 kHz up
to 100 MHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.26 Bode plot for the system of RL branches and their equivalence,
representing a single turn’s inductance and core losses, and cov-
ering a frequency sweep of 5 decades starting from 1 kHz up to
100 MHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.27 Bode plot for the system of RL branches and their equivalence
after inserting compensation factors in the first branch . . . . . . 83
4.28 An improved ladder model for a single turn’s ac inductance and
core losses covering a frequency sweep including and starting from
1 kHz up to 100 MHz . . . . . . . . . . . . . . . . . . . . . . . . 84
4.29 A single turn of copper winding on an iron core model covering
a frequency sweep from dc up to 100 MHz . . . . . . . . . . . . . 86
4.30 Bode plot for the system of RL branches and their equivalence
after inserting compensation factors in the first branch . . . . . . 87
LIST OF FIGURES xi
5.1 Measurements of impedance magnitude for the winding-on-core

prototype along a frequency band from 0 Hz up to 40 MHz -
single and tripple layers . . . . . . . . . . . . . . . . . . . . . . . 91
5.2 Measurements of impedance magnitude for the winding-on-core
prototype along a frequency band from 0 Hz up to 40 MHz -
single layer, two layers and three layers . . . . . . . . . . . . . . . 92
5.3 Layout of simulation model optimized towards capacitive cou-
pling version-1 and its agreement with measurement data . . . . 95
5.6 Layout of simulation model optimized towards inductive coupling
and its agreement with measurement data . . . . . . . . . . . . 99
A.1 Difference in power levels (dB) versus power ratios . . . . . . . . 116

xii LIST OF FIGURES
List of Tables
2.1 Measurements access points in the 3-layer 24-winding-on-core

prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1 Key frequencies versus their corresponding wavelength (λ) in free

space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Key results out of Inductive Coupling Matrices for different fre-
quency values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3 Frequency occurrence of a penetration depth corresponding to
“half” the lamination thickness for different relative permeability
values μr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4 Key results out of a single turn analysis along a frequency sweep 71
4.5 Quantification factors of key parameters corresponding to fre-
quency steps - single turn analysis . . . . . . . . . . . . . . . . . 73
4.6 Quantification factors for core losses increase and inductance
magnitude decrease for 5 decades of frequency - single turn anal-
ysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7 Summary of the compensation factors for a single turn’s complete
model along a frequency sweep from dc up to 100 MHz . . . . . . 86
A.1 Power ratios and voltage/current amplitude ratios and their cor-
responding Decibels (dBs) . . . . . . . . . . . . . . . . . . . . . . 115
A.2 dB units describing absolute quantities and their corresponding
reference values . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
xiii
Chapter 1
Introduction
1.1 Background
Introducing electric propulsion in a large scale to the automotive industry

through hybrid and electrical vehicles, in addition to the associated develop-
ment of charging techniques, creates a unique and challenging situation that
was not a norm before. The relatively high-power electric drives on board along
with already established small-power electrical loads make the co-existence be-
tween systems and components in the vehicle environment more challenging.
The switching process of semiconductor elements in high-power (10’s of
kiloWatts) electric drives and at relatively high frequency rates (few kHz of
switching frequency), involve high voltage- and high current-derivatives. For
example, semiconductor elements switching at 4 kHz in a 30 kW electric drive
system which is supplied from a floating Energy Storage System (ESS) of 300
9
volts can produce a voltage derivative ( du
dt ) of more than a billion (10 ) [V/s],
di 9
and a comparable current derivative ( dt ) in the order of 10 [A/s], a fact that
can result in conducted and radiated emissions. The confined space and the
surrounding magnetic/electrical media will make intervention between different
systems/sub-systems/components easier and will place more challenges on an
electromagnetically compatible vehicular system.
The importance of the ElectroMagnetic Compatibility (EMC) field in the
1
2 CHAPTER 1. INTRODUCTION
automotive industry comes from many key contributors among which are:
i. The pressing need and public awareness to cheaper, cleaner and more effi-
cient mobility solutions which can be achieved through electrical propulsion,
ii. The nature of high-scale production in the automotive industry,
iii. The importance of modular design [2], and
iv. The desirable minimum development time with minimum design iterations
[4][5].
Traditionally EMC optimization comes as a later stage after the functional
and safe design in a way that will not compromise any, but being able to predict
the EMC behavior of the system at an early stage will only contribute positively
to the development time and hence productivity and cost.
Another interesting fact to mention is that there are no clear and defini-
tive EMC standards developed for the individual electrical components used in
building up the powertrain in a hybrid or electrical vehicle, despite the fact that
standards exist for the vehicle as a whole unit [6]. This reflects the broadness
and freedom in fulfilling EMC-related standards, but at the same time distin-
guishes between manufacturers interested in “getting under the skin” of EMC
challenges in order to fulfill the standards, and those interested in quick-fix
solutions.
1.2 Problem Formulation
Conducted and radiated ElectroMagnetic Interference (EMI) emissions is the

downside of power electronics systems development towards increasing power
density and decreasing size and losses. Unlike three phase grid power-supplied
machines, Pulse-Width Modulation (PWM)-Inverter-Fed machines inherently
result in Common-Mode (CM) voltage on the connection point of the machine
phase windings, which has a value at any point in time as shown in figure 1.3
[25][32][3][28][1][26].
On the other hand, parasitic capacitances-to-ground exist literally all over
the electric drive system [18][19], as shown in figure 1.1 and 1.2, starting from
the battery pack’s (ESS) positive and negative poles to ground, passing through
power cables, each end of dc-link capacitor or line Differential-Mode (DM) filters
1.2. PROBLEM FORMULATION 3
to ground, each junction module of switching elements to ground, and ending

up in the electric machine parasitics to ground.
...................
a~
stator winding
+ __
Battery b~
c
...................
~
Motor Chassis
19 Lead Acid Batteries

placed on metalic sheet
Power electronics rack Motor placed on metalic sheet
Figure 1.1: Parasitic capacitances in an electric drive system
Parasitic capacitances are usually of low orders of magnitude which make

them negligible at low frequency analysis. Investigation out of the inhouse built
electric drive labsetup showed a range from 100’s of pico (p) Farads (Fs) up
to few nano (n) Fs. However, with the development of Insulated Gate Bipo-
lar Transistor (IGBT) technology and the increase of switching frequency and
rated power, not least with Silicon Carbide (SiC)-based technology, these capac-
itances interact with the high-frequency CM voltage waveform and can provide
a low-impedance path for the leakage current. Capacitive impedance path de-
pends on the combination of capacitance value and CM voltage frequency. The
high-frequency leakage current takes the the form of a damped sinusoid [28],
Figure 1.2: A photo of the second inhouse-built electric drive labsetup

1.3. SCOPE AND GOAL 5
at every instant a change in the varying 2-step/3-step CM voltage waveform

occurs, which is estimated to be 4-6 times the carrier wave or CM voltage wave-
form frequency, as shown in figures 1.3 and 1.4, assuming carrier wave based
modulation.
Depending on the machine type and size the parasitic capacitances of the
machine can vary [8][13][9][27][22], but common and fundamental capacitances
among all machines are capacitances among winding copper conductors in a
stator slot, and capacitances between winding copper conductors in the slot and
stator frame. Induction machines can have additional parasitic capacitances due
to their construction [1].
The manifestation of this problem can take different forms among which
many are allready spread in the industry. Examples are:
(a). Frequent Earth Leakage Protection (ELP) trip faults when connecting
floating ESS powered electric drive to the power grid for charging.
(b). Electric Discharge Machining (EDM) currents causing premature motor
bearings failure or what is known to the industry as bearing currents.
(c). Gradual degradation and breakdown of copper conductors’ insulation within
the first few turns of a machine stator winding.
Since many factors contribute to this phenomenon, many mitigation tech-
niques can be used to control the problem, as summarized in [1]. They tend to
be reactive measures and not proactive ones taken at an early stage.
1.3 Scope and Goal
The scope of this research project focuses on analysis and simulation of electric
drive systems from an EMC perspective, in order to limit the conducted EMI
emissions observed at high frequencies up to 30 MHz [15][33].
The ultimate goal of this work is to contribute to a detailed CM-model of
an electric machine that is valid from dc up to 30 MHz, and can be used in a
circuit simulation environment in order to predict high-frequency CM current.
The Current at high frequencies can use parasitic capacitances to ground, which
exist all over the electric drive system, as low-impedance paths to ground and
back to its source in the power electronics converter.
Triangle and reference voltage signals + PWM signals (u,v,w)

1
−1
3.5 4 4.5 5 5.5 6
PWM signal (u)
200
100
0
3.5 4 4.5 5 5.5 6
PWM signal (v)
200
100
0
3.5 4 4.5 5 5.5 6
PWM signal (w)
200
100
0
3.5 4 4.5 5 5.5 6
Common−Mode voltage
200
100
0
3.5 4 4.5 5 5.5 6
Figure 1.3: Triangular wave and reference voltages, generated PWM signals and
a varying 2-step/3-step CM voltage waveform
1.3. SCOPE AND GOAL 7
0.01
current waveform
0.008
0.006
0.004
0.002
−0.002
−0.004
−0.006
−0.008
−0.01
0 0.5 1 1.5 2 2.5 3
t x 10
−5
Figure 1.4: The resulting high-frequency damped-sinusoid CM current waveform

occuring at each CM voltage step
The study aims at investigating capacitive and inductive coupling of a ma-

chine winding in order to characterize the high frequency behavior of the ma-
chine. The outcome of this study and the methodology adopted will represent
an input to the slot level analysis of electrical machines along the frequency
sweep of interest.
The developed knowledge and recommendations is expected to add value
when integrated in a full-scale electric machine design environment which will
enable the machine designer, at an early stage of the design, to predict and
influence stray capacitances of the machine.
Defining the scope of this project should not be confused with the basic
criteria about EMC behavior for any electric drive system, that is; EMC is a
very dynamic field and EMC optimization is a system-based attribute in which
any part of the system can contribute to its compatibility. By “attribute” one
means the feature that cannot be tangibly sensed by the user, however it is
necessary for safe and compliant functionality [6].
1.4 Study Roadmap and Thesis Structure
The study starts by investigating a heavy vehicle electrical system which in-
cludes a motor drive with an inverter connected to a floating ESS through
a junction box. The system is studied and model drafts are developed in an
LTspice simulation software. The target is to develop a detailed and comprehen-
sive model of the electrical system in a “Simulation Program with Integrated
Circuit Emphasis” (SPICE) environment where conducted emissions can be
viewed and studied through “what if” scenarios.
Due to ease of access and in order to perform research-oriented activities, an
experimental electrical system that mimics the above mentioned is built inhouse.
Two laboration setups are made. They included two induction machines (1.5
kW & 2.2 kW), a power electronics converter (43 kW), a shielded power cable
and a metalic sheet as a ground plane. In the first lab setup, the floating ESS is
emulated by the direct connection to the power grid and then passing through
a galvanic isolation stage of a 3-phase power transformer with a ratio primary-
to-secondary of 1:1, and then amplitude variation stage by another 3-phase
variable transformer and finally, rectification and filtering stage using 3-phase
bridge rectifier and a parallel capacitor. In the second lab setup, the ESS is
1.4. STUDY ROADMAP AND THESIS STRUCTURE 9
built from 19 lead-acid battery cells, each rated at 12V, 20 Ah corresponding

to ≈ 4.4 kWh.
The investigation reveals that low-order of magnitude capacitances existing
between winding turns and between turns and stator frame in the electrical ma-
chine winding, are key elements in the analysis of high frequency current paths
in the whole system, therefore lots of effort and attention is given to develop a
detailed model of the machine winding including parasitic capacitances which
covers a wide frequency band considering the complete conducted emission fre-
quency range up to 30 MHz as per [15].
For that purpose, an induction machine of 1.5 kW is opened up and stud-
ied extensively, and a prototype of a winding-on-core model composed of 24
turns/3 layers (8 turns in each layer) in distanced and condensed arrangements,
is built and investigated to study the turn-to-turn capacitance, layer-to-layer
capacitance and the iron core impact on the inductance in high frequency. An
impedance analyzer Hewlett Packard (hp) 4194A from the Electrical Measure-
ment Department (Elmät) is used (and thankfully acknowledged) to perform
tests and measurements for the required frequency range.
The study is supported by a steering group of professionals from the academia
and industry. Besides my supervisor Professor Mats Alaküla and my co-supervisors
Dr. Avo Reinap and Dr. Hans Bängtsson, the group consists of: Jan Welin-
der from SP Technical Research Institute of Sweden, Dr. Per Karlsson from
Emotron then and currently an Associate Professor at IEA-Helsingborg Cam-
pus, Håkan Berg and Kjell Attback from Volvo Cars Corporation (VCC), Dr.
Johan Nilsson from Electrical Measurement Department, Dr. Sabine Alexander-
sson and Jan-Olof Brink from Kockums. Their guidance, support and feedback
are highly appreciated and have had a significant impact on the development of
this project.
The Thesis structure proceeds as follows:
- Chapter 2: Analysis Tools.
- Chapter 3: Capacitive Coupling Matrix.
- Chapter 4: Inductive Coupling Matrix.
- Chapter 5: Model Validation.
- Chapter 6: Conclusions and Future work.
1.5 Contribution of This Study

This study offers knowledge and tools targeting electrical machine designers,
personnel working in electrical system integration, and those fulfilling EMC
requirements through:
- Highlighting the capacitive and inductive coupling paths which the CM cur-
rent takes throughout its way back to the converter, and representing them
in a circuit simulation software. The perspective is enhancing immunity of
electrical machines against conducted EMI emissions.
- Generating a capacitance network with pre-determined accuracy (complexity)
levels, for the turn-to-turn copper conductors of a winding-on-core. This helps
understand and establish parasitic capacitance network for a winding inside
a slot in electrical machines. Parasitic capacitances in electrical machines are
viewed as major coupling paths that impact leakage current in motor drive
systems of PWM input.
- Using freely available software tools, like FEMM and LTspice , to quickly
but accurately predict capacitive and inductive coupling in electrical machines,
and the resulting high-frequency behavior.
Chapter 2
Analysis Tools
FEMM and LTspice along with MATLAB are used as a complete analysis
package in this project. This combination is used for drawing the desired geom-
etry, perform Finite Element Method (FEM) analyses, store results in arrays
and structure arrays (structs), establish agreement between FEMM outcome
and SPICE models and finally propose models and compare their validity and
agreement to measurements performed using the impedance analyzer hp 4194A.
The application environment for the proposed models is LTspice . As a
time-domain circuit analysis tool, LTspice can represent switching behaviour
in detail. Proposed models demonstrating capacitive and inductive coupling
paths of pre-assigned threshold limits in presence of power electronics converter
(inverter), ESS, cables and other PWM-fed driven loads enables the prediction
of CM circulating current.
In the power electronics converter model, transient analysis can be studied
using either ideal switches or manufacturer’s commercial models of switches
available from the manufacturers’ websites. In such a model, different features
can be investigated like switching frequency, switching dead-time, slew rate (rise
and fall times), modulation index and others.
Another alternative to the above-mentioned detailed representation of elec-
tric drive system, which only becomes significant when having access to a system
or lab setup with the power electronics converter, is to record measurements at
the terminals of the power converter in real-life conditions by Digital Storage
11
12 CHAPTER 2. ANALYSIS TOOLS
Oscilloscope (DSO). Then, these measurements can be configured as a voltage

source that would drive the subsystem/component of interest under SPICE.
This approach is simpler and optimized towards the subsystem/component of
interest as it simulates “real-life” stationary conditions. The advantages are:
(i.) It saves simulation time substantially, especially when incorporating more

than a single PWM-fed driven load of different ratings in the drive system.
(ii.) It gives the freedom of measuring high-frequency leakage currents without

the need of having Rogowski coil or other specialized/expensive high-
frequency current meters of a certain bandwidth.
2.1 Slot Analysis
On the way to develop the complete EMC machine model, the stator slot is
taken as a basis for developing the model and highlighting the model’s main
parameters, namely the capacitances, resistances and inductances. The portion
of the winding inside the stator slot represents the highest capacitance-to-ground
contribution due to the proximity of copper conductors to the iron frame inside
the slot, in presence of wire insulation and air. In the 1.5 kW openned Induction
Machine (IM), the lumped value of one slot capacitance-to-iron frame (ground),
with simple analytical approach, can range from Csg = 0.30 − 0.70 nF. The
outcome of the slot analysis can then be extended to the rest of stator slots
taking in consideration the number of slots, number of phases, number of poles,
end-turns layout and the winding arrangement. An Integral Slot (full-pitch)
Single Layer Winding arrangement is considered for the current study where
the winding is placed completely in two stator slots as shown in Figure 2.1 &
Figure 2.2 below. The number of Slots per Pole per Phase (q) in this machine
= 3, but from a slot perspective this does not make any difference as long
as the winding arrangement is Integral Slot and Single Layer Winding, which
means that the findings of the slot analysis for this particular machine are easily
applicable to other machines holding the stated conditions.
2.1. SLOT ANALYSIS 13
Figure 2.1: A photo of induction machine stator showing Integral Slot, Single
Layer Winding of q=3 (36 slots, 3 phases & 4 poles) as well as the End Turns
arrangement
Figure 2.2: A side-view drawing of induction machine stator winding showing

Integral Slot, Single Layer Winding of q=3 (36 slots, 3 phases & 4 poles) as well
as its End Turns arrangement
2.2. “MACRO LEVEL” AND “MICRO LEVEL” MACHINE MODELS 15
2.2 “Macro Level” and “Micro Level” Machine

Models
In order to link the slot level of analysis to the level of the complete EMC
machine model, the so called “micro-level” analysis of the slot will represent a
fundamental input to the “macro-level” model proposed in figure 2.3.
The model in Figure 2.3 represents an Integral Slot stator winding with
q = 3. The associated capacitances-to-ground of the winding conductors inside
the slots are expressed as lumped quantities on both sides of the coil wind-
ing inductance. The capacitances between phases in the End Turns region
are approximated to capacitance values between “large conductors” of a width
equivalent to the slot width, since each “large conductor” represents the total
number of conductors inside a slot, as shown in Figure 2.4.
In this particular machine we have 4 electromagnetic poles, so the Pole
Pitch (τp ) is 90◦ . The interface area between the group of “large conductors”
of adjacent phases will be expressed as follows:
5 3 1
Interf aceArea(A) = ( ∗ τp + ∗ τp + ∗ τp ) ∗ sw = τp ∗ sw (2.1)
9 9 9
The Pole Pitch τp = π2 ∗ r; where r = 65.5 mm is the radius from the

center up to the slot depth (yoke limit) of the stator in the radial direction,
and sw = 4.5 mm is the slot width. The capacitance can be expressed with the
simple parallel-plate capacitance formula [10] as:
◦ ∗ r ∗ A
C= (2.2)
d
where ◦ is the permitivity constant of free space, r is the relative per-

mitivity (dielectric constant) for insulation paper and d=0.9 mm is the paper
thickness between the two “large conductors”.
1
36∗π ∗ 10−9 ∗ 3 ∗ π
2 ∗ 65.5 ∗ 10−3 ∗ 4.5 ∗ 10−3
= ≈ 13.7pF (2.3)
0.9 ∗ 10−3
Cph_ph1
{2*Cpp}
Cph_ph3 Cph_ph2
{2*Cpp} {2*Cpp}
C_slot_left Lcoil C_slot_right
C_slot_left12 Lcoil7 C_slot_right12 C_slot_left6 Lcoil1 C_slot_right6
{Cs} {Lc} {Cs}
{Cs} {Lc} {Cs} {Cs} {Lc} {Cs}
inner_most_coil
inner_most_coil inner_most_coil Rcoil
Rcoil12 Rcoil6
{Rc}
{Rc} {Rc}
C_slot_left1 Lcoil C_slot_right1

{Cs} {Cs} {Cs} {Lc} {Cs}
{Cs} {Cs}
{Lc} {Lc}
second_coil
second_coil second_coil
Rcoil13 Rcoil7 Rcoil1
{Rc} {Rc} {Rc}
outer_most_coil outer_most_coil C_slot_left2 Lcoil C_slot_right2

C_slot_left14 C_slot_right14 C_slot_left8 C_slot_right8
Lcoil9 Lcoil3 outer_most_coil
{Cs} {Cs} {Cs} {Lc} {Cs}
{Cs} {Cs}
{Lc} {Lc}
Rcoil2
Rcoil14
Rcoil8 {Rc}
{Rc}
{Rc}
.param Cs 0.686n;
.param Cpp 13.7p;
.param Lc 86m;
.param Rc 0.348;

{Cs} {Lc} {Cs}
inner_most_coil inner_most_coil
Rcoil15 Rcoil9 inner_most_coil
Rcoil3
{Rc} {Rc}
{Rc}

{Cs} {Lc} {Cs}
second_coil
Rcoil16 second_coil
Rcoil10
Rcoil4
{Rc} {Rc}
{Rc}
second_coil
{Cs} {Lc} {Cs}
outer_most_coil outer_most_coil
Rcoil17 Rcoil11 outer_most_coil
Rcoil5
{Rc} {Rc}
{Rc}
Cph_ph6 Cph_ph5
{2*Cpp} {2*Cpp}
Cph_ph4
{2*Cpp}
Figure 2.3: Macro(slot)-level circuit model of induction machine with Single

Layer Winding of q=3
2.3. LAYER-TO-LAYER APPROACH FOR SLOT ANALYSIS 17
Figure 2.4: An unfolded view of the End Turns region approximating the inter-
face between two adjacent phases in the machine stator
End Turns depend heavily on winding configuation and can be varied from
one machine to another. For this machine one can notice from Figure 2.2 that
two overlaps occur between any two adjacent phases in the End Turns region
at each side of the stator. This is expressed by multiplying the capacitance
value (13.7 pF) by a factor of “2”. Then the capacitances between phases has
to reflect the presence of End Turns in the front and back of the machine stator
and that explains why we have the same capacitance representation at each end
of the phases in the “macro-level” model in Figure 2.3.
It is worth noticing that “macro-level” modeling is based on established
quantities in the machine design field like number of slots, number of phases,
number of poles, winding arrangement, etc.; however the “micro-level” slot
analysis, whose outcome is to be fed to the “macro-level” model, is analyzed
according to the transmission line theory to reflect the frequency dependence
phenomena of the conductors inside the slot.
2.3 Layer-to-Layer Approach for Slot Analysis
Considering a winding in a slot model on a layer-to-layer basis, the frequency-

domain analysis in LTspice can be considered as a rough and quick approach
for developing transmission line models of low orders (few RLC-sections) as
shown in Figure 2.5. Layer-to-layer approach has been adopted in [14]. With
relatively fewer sections and standardised repeated parameters in each section,
one can propose the parameter value based on rough analytical assumptions
and then perform sensitivity analysis while fixing the rest of parameters. Sen-
sitivity analysis can be done by running the simulation for different values of
the parameter of concern in both directions above and below the roughly esti-
mated value. This can give the user an idea about the parameter value tendency
and its influence on matching the model output with the measurement results;
however this approach has limitations since it is based on rough assumptions
and depends on trial-and-error methodology which is not very sound neither
accurate. So manual representation in frequency-domain circuit analysis can be
used as a comparison tool to measurement data, but generating the model and
selecting its parameters has to be based on finer assumptions and performed in
a systematic way which will be more scientifically motivated.
C5
{c_ww}
C4
{c_ww}
C1 C2 C3
{c_ww} {c_ww} {c_ww}

C6
{c_ww}
R13 L13 L1 R14 L14 L2 R15 L15 L3
IN
V6
{r} {l} {l_lamda} {r} {l} {l_lamda} {r} {l} {l_lamda}
SINE(0 1 50) C14 C16

C15
AC 1
Rser=0 {c_wc} {c_wc}
{c_wc}
C8
L12
{ground_c}
{l}
R5
{r_core}
Figure 2.5: Layer-to-layer approach for sensitivity analysis in frequency domain

for a 3-layer winding represented by 3 RLC-sections
2.4 Turn-to-Turn Approach for Slot Analysis
The turn-to-turn analysis level is adopted in modelling the winding portion in

a stator slot. Turn-to-turn approach has been followed in [20][21][35][24]. As a
result the transmission line model is of a high order due to the number of turns
and this shows the significance of generating such a model automatically. The
represented parameters concerning each turn in the model are proposed to be:
2.5. SIMPLE WINDING-ON-CORE PROTOTYPE 19
(i.) Resistance of each winding turn (Rt ).

(ii.) Capacitances between the winding turns (Ctt ),
(iii.) Capacitances between each winding turn and the ground (Ctg ),
(iv.) Inductance of each winding turn (Lt ),
(v.) Mutual inductance between winding turns represented by coupling fac-
tors.
Magnetic core losses composed of hysteresis and eddy current losses are
represented in a so called “ladder diagram” that represents the frequency de-
pendence behaviour of the turn resistance and inductance (see section inductive
coupling matrix 4 later). The conductor insulation losses existing between ad-
jacent turns and between each turn and the ground are not represented in the
model.
Figure 2.6: Analysis development of the winding portion modelling inside a slot.
First the winding is considered as one complete unit, then it’s divided into layers
and finally is represented by turns
2.5 Simple Winding-on-Core Prototype

On the turn-to-turn level of modelling, the representation of a complete ma-
chine winding inside a slot becomes very complex and a tedious job to do, so a
simplification is proposed by one of the reference group members [31], to form

a common starting point in order to develop the model and then test it and
validate it for the frequency band of interest. A prototype of a simple winding
on an iron core is built. It has 24 turns and 3 layers, in both distanced and
condensed wire arrangements as shown in figure 2.8 below. In the condensed
arrangement the conductors are wound close to each other as the case would be
in a machine winding, while in the distanced arrangement the conductors are
wound 3 mm apart, both on the layer level among the turns, and among the
turns in different layers. The impedance of each arrangement at four different
access points are measured and analysed along the frequency range of interest.
Access points represent the start and end terminals of the winding, in addition
to the start of each layer as per the following table 2.1.

Figure 2.7: The 24 conductors arrangement for electrostatic FEMM analysis
Access Points Location

1st Winding input
2nd Second Layer Input
3rd Third Layer Input
4th Winding output
5th Iron core
Table 2.1: Measurements access points in the 3-layer 24-winding-on-core proto-

type
As in sections 2.3 and 2.4 before, frequency-domain circuit analysis is used,

but this time automatically, to generate and verify the validity of the proposed
2.5. SIMPLE WINDING-ON-CORE PROTOTYPE 21
Figure 2.8: 3-layer 24 turns winding-on-core model in both distanced and con-
densed wiring arrangements
RLC transmission line model of the built prototype in comparison to the mea-
surements from the impedance analyzer hp 4194A. The parameters are moti-
vated from FEMM analysis. Capacitance (C) parameters are taken from the
outcome of FEMM electrostatic analysis where a Capacitive Coupling Matrix
is produced. Resistance and inductance (R & L) parameters are taken from
FEMM magnetostatic and time-harmonic magnetic analysis where an Inductive
Coupling Matrix is produced. The entire process is automated and centralised
from MATLAB as a single platform.
Chapter 3
Capacitive Coupling Matrix
This chapter explains the background and details behind FEM electrostatic
analysis of the prototype model. It puts the reader in the steps of creating the
Capacitive Coupling Matrix. It displays results and the link to the capacitance
network in a circuit simulation environment.
3.1 Electrostatic FEMM Modelling
The built prototype model is modelled in FEMM electrostatic analysis as 24

copper conductors in a 2D planar problem for the condensed wiring arrange-
ment, where one turn is represented as a conductor with depth of (44*2+25*2
= 138 mm) inside the plane, which is equivalent to one turn’s length. The
conductor diameter including insulation is measured to be 0.648 mm while the
copper diameter itself is 0.589 mm [16].
Conductors are drawn with their insulation and placed close to each other
in an orderly manner the same way they are wound in reality. Numbering the
conductors start from left to right in odd layer numbers and from right to left in
even layers numbers, just as it is done in reality; so conductor number one starts
from bottom left closest to iron core and conductor number 8 is far right in the
First Layer, conductor number 9 starts from the right side of Second Layer and
continues till number 16 far left at the end of it and so on.
23
24 CHAPTER 3. CAPACITIVE COUPLING MATRIX
Besides the conductor insulation a dielectric of 3 mm high is drawn and

configured above the iron core. A semi-circle of a radius of 30 mm is drawn on
the iron core outside the Bobbin area, and used as mixed boundary condition
where electric field lines tend to vanish. The iron core is considered as a ground
and drawn with a single line that is assigned a zero voltage potential. The
drawing process is done through MATLAB code that uses the Lua Scripting
language. The conductors arrangement is shown in figure 3.1 below.

Figure 3.1: The 24 conductors arrangement for electrostatic FEMM analysis
(repeated)
The problem boundary is configured with a “mixed” boundary condition

according to equation 3.1
∂V
◦ ∗ r ∗ + c◦ ∗ V + c1 = 0 (3.1)
∂N
The “mixed” boundary condition can be achieved by setting c1 = 0 and
solving for c◦ in equation 3.1 above, the condition then would approximate
an unbounded solution region inside the border it is assigned for, in a FEMM
electrostatic problem [23].
3.1.1 Capacitance Matrix for Multi-Conductor System
The objective of the analysis is to find the capacitance matrix for the 24 conduc-
tors where mutual capacitances between any two conductors can be predicted,
meanwhile it would be possible to express the capacitance-to-ground for each
conductor.
3.2. HOW THE CAPACITIVE COUPLING MATRIX IS CREATED? 25
Referring to Cheng’s Capacitances in Multi-conductor Systems [10], the ca-

pacitance system matrix elements consist of diagonal elements (cii ) and non-
diagonal elements (cij ) where (i = j & i, j = 1, 2, 3 · · · n, “n” is the number of
conductors in the system).
Diagonal elements (cii ) are called the coefficients of capacitance for the
th
i conductor in the system. They are positive constants and represent, in
physical terms, the total capacitances between the ith conductor and every other
conductor in the system, as well as between the ith conductor and its ground.
On the other hand, non-diagonal elements (cij ) where (i = j) are called the
coefficients of induction [10], and exist between the ith and jth conductor. They
are negative constants that represent the mutual partial capacitance between
the ith and jth conductor in the system.
Adding the negative constants representing mutual capacitances to the di-
agonal elements representing the total capacitances of a conductor gives the
capacitance-to-ground value for the ith conductor of concern.
3.2 How the Capacitive Coupling Matrix is cre-

ated?
The electrostatic FEMM analysis for the prototype model is done by assigning
a voltage value of 1 V as a circuit property to one conductor of the 24 conduc-
tors while the rest are assigned zero voltages, then the problem is analyzed and
solved for the charge Q on each conductor in the system. Results are saved in
a matrix and the same procedure is repeated, in an automated way by help of
MATLAB , for every conductor in the wiring arrangement. The result repre-
sents a capacitance system matrix of order 24 by 24, given that the capacitance
value is the same as the charge value for a voltage of 1 V since Q = C ∗ V .
The Capacitive Coupling Matrix is basically a post-processed capacitance
matrix for a multi-conductor system where its diagonal elements are positive
constants that represent the capacitance-to-ground for each conductor, while
its non-diagonal elements are positive constants that represent the mutual ca-
pacitance for the corresponding conductors as shown in figure 3.4.
As can be observed from figure 3.2 above, the highest values for capacitance-
to-ground (Ctg ) occur in the First Layer interfacing the core, with maximum
(a) 1st (b) 2nd (c) 3rd (d) 4th
(e) 5th (f) 6th (g) 7th (h) 8th
(i) 9th (j) 10th (k) 11th (l) 12th
(m) 13th (n) 14th (o) 15th (p) 16th
(q) 17th (r) 18th (s) 19th (t) 20th
(u) 21st (v) 22nd (w) 23rd (x) 24th
Figure 3.2: Electrostatic FEMM analysis of 24 conductors on core - condensed

arrangement
values on the edges since the electric field lines can couple directly to the core
without any barriers. Second Layer capacitance-to-ground (Ctg ) is the lowest
with absolute values in the order of 10−19 Fs which represent 7 orders of magni-
tude lower than the first layer’s capacitance-to-ground (Ctg ). This is due to the
electric field being confined between two copper layers which represent electro-
static shields banning the electric field to couple to the core. This is not fully
true for the conductors on the edges, conductor 9 & 16.
It can be also observed that the most intense electric field potential lines lie
between two adjacent conductors, especially in the outer layers, i.e. the First
and Third Layers, and that reflects on the adjacent turn-to-turn capacitance
values (Ctt ) to be the highest.
By comparing the effect of the same stimulus applied to the conductors of
the Third Layer as shown in sub-figures 3.2q to 3.2x, to the one applied to the
conductors of the Second Layer as shown in sub-figures 3.2i to 3.2p one can
observe that in the Third Layer case, which is exposed to the air from one side,
electric field lines are more distributed and spread linking larger interface ar-
eas of the conductors through the lossless medium of air, than the case of the
Second Layer where electric field lines are mainly linked to the neighbouring
conductor through smaller interface area between the two adjacent conductors.
This will result in higher adjacent turn-to-turn capacitance between conductors
in the Third Layer than the Second Layer. Considering the confined electric
field between the Second Layer and the First and Third Layers, we can state
that electric field lines between two adjacent conductors in the Second Layer
does not exceed half the circumference of each conductor, as shown in the fig-
ure 3.3. The above mentioned observations can also be read from figure 3.4 and
subfigures 3.8a & 3.8b.
The conclusion is that conductor layers represent electrostatic shields be-
tween inner and outer layers, consequently the First Layer interfacing the core
would represent an electrostatic shield between inner layers of conductors and
the ground.
When this behaviour is to be extended to a complete machine winding
inside a slot, where iron core encircles the copper conductors from all sides,
and following the same reasoning for the electric field lines, one needs only to
consider the contribution of the First Layer of conductors interfacing the iron for
calculating the capacitance-to-ground. For adjacent turn-to-turn capacitance
considering the somewhat random process in forming the winding, the approach
would be to develop the Capacitive Coupling Matrix for the conductors inside

Figure 3.3: Electric field lines connecting conductors 11 and 12 in the inner
most layer, the Second Layer
the slot considering an ideal situation where each turn is followed by the next
in an orderly manner, and then multiply it by a permutation matrix expressing
the unpredictable positioning of some turns in the slot. This implementation
will be made in the next phase of the project.
Since the Capacitive Coupling Matrix for the simple winding prototype is of
order 24 x 24 = 576 elements, the best way to view it is in a 3D plot and 2D image
as shown in figure 3.4. Linear scale is used in figure 3.4 which reflects differences
in absolute values of capacitances in the multi-conductor system. The two
parallel red spike lines around the main diagonal in figure 3.4a correspond to the
turn-to-turn capacitances of adjacent conductors while spike lines perpendiular
to the diagonal represent the adjacancy between the layers; one between the
First and Second Layers, and the other between the second and Third Layers.
It should be noted that the matrix is symmetric around the diagonal so there
is information redundancy above and below the diagonal elements.
Note the steep valley in figure 3.4 represented by the main diagonal elements
which shows the difference between turn-to-ground and turn-to-turn capacitance
values. Turn-to-ground capacitance values can also be seen in figure 3.5 while
turn-to-turn capacitance values in figure 3.6.
The same behaviour can be seen in the 2D image of the matrix in figure 3.4b,
with a clearer view about the difference in the magnitude between the elements.
X & Y represent the position of the element in the matrix. First main diagonal
element starting top left in the matrix corresponds to the first conductor in the
wiring arrangement, while the last element ending bottom right correponds to
the last conductor.
Capacitive Coupling Matrix − condensed arrangement
12
10
capacitance in [pF]
8
20
10
0
6
25
20 25
4
20
15
15
10 2
10
5
Y 5
X 0
0 0
(a) 3D plot of the Capacitive Coupling Matrix

Image Display of the Capacitive Coupling Matrix − condensed arrangement
12
5 10
8
10
Y
15
4
20
2
0
5 10 15 20
X
(b) 2D image of the Capacitive Coupling Matrix in (pF). Note the highest
turn-to-turn capacitance values among the first 8 conductors (First Layer)
Figure 3.4: Linear scale representation of Capacitive Coupling Matrix for the
system of 24 conductors on iron core - condensed arrangement
Turn−to−ground capacitance in the condensed wiring arrangement

2.5
1.5
0.5
0
0 5 10 15 20 25
Figure 3.5: Turn-to-ground capacitance in pico farads [pF] for conductors in

the condensed wiring arrangement, represented by the main diagonal elements
in the Capacitive Coupling Matrix
Turn−to−turn capacitance in the condensed wiring arrangement Turn−to−turn capacitance in the condensed wiring arrangement
12.8
12
12.6
10 12.4
12.2
Capacitance value [ pF ]
Capacitance value [ pF ]
8
12
11.8
6
11.6
4 11.4
11.2
2
11
0 10.8
0 5 10 15 20 25 0 5 10 15 20 25
Conductor numbers Conductor numbers
(a) Y-axis starts from zero (b) zoomed to show the difference between val-
ues
Figure 3.6: Adjacant turn-to-turn capacitance in pico farads [pF] for conductors
in the condensed wiring arrangement, represented by the first diagonal above
(or below) the main one in the Capacitive Coupling Matrix
3.3. SPARSE MATRIX REPRESENTATION 31
Note the highest values for turn-to-turn adjacent capacitance among the
first 8 conductors (First Layer) followed by the last 8 conductors (Third Layer)
compared to the rest. This can also be seen from figure 3.6. Values are higher
in the First Layer than in the Third Layer due to the proximity of the First
Layer to the iron core, which in turn is configured with ground potential where
electric field is zero. This tends to force electric field lines to be more intense
near the conductors facing the dielectric towards the core in the First Layer.
On the other hand, conductors in the Third Layer face air and ultimately the
outer boundary of the problem at farther distance, which will not influence the
electric field lines distribution around the conductors of the Third Layer.
Note also that the highest turn-to-ground capacitance values are for con-
ductors in the First Layer, however these values are significantly lower than
turn-to-turn capacitance values due to the longer electric field path coupling
the conductor to the core (ground) compared to the path coupling the conduc-
tor to the neighbouring conductor.
3D plot and 2D image for the Capacitive Coupling Matrix are represented
in a logarithmic scale in figure 3.7 in order to cover all capacitance values in the
system matrix meanwhile express the small differences. The logarithmic plot
is the best in telling the relative difference among the plotted elements. In fig-
ure 3.7 there is more than 440 Decibels (dBs) difference between the values (For
dB see Appendix A [34]) which corresponds to 22 orders of magnitude. Lowest
values being for mutual capacitance between conductors surrounded totally by
other conductors from all sides, like the capacitance between conductors 10 & 15
in the Second Layer, while the highest, as stated before, is between conductors
in the First Layer.
3.3 Sparse Matrix Representation
To systematically link the outcome of the previous capacitance analysis to the

transmission line modelling of the winding on an iron core, a sparse matrix
is generated from the Capacitive Coupling Matrix by introducing a threshold
according to the level of accuracy the designer wishes to represent. The sparse
matrix only contains the values in the Capacitive Coupling Matrix which are
above or equal to the stated threshold. All other elements are represented by
zeros. Figure 3.8 below shows some examples of sparse matrix representations
with different threshold values. It also indicates the number of nonzero elements
Capacitive Coupling Matrix − condensed arrangement (colourbar in dB)
−250
−200
−300
−400
−350
Z
−400
−600
−450
−800
−500
25
20 −550
15 −600
10 −650
5 25
20 −700
Y 15
10
0 5
0 X
(a) 3D Logarithmic plot of the Capacitive Coupling Matrix

Capacitive Coupling Matrix − condensed arrangement (colourbar in dB)
−250
−300
5
−350
−400
10
−450
Y
−500
15
−550
−600
20
−650
−700
5 10 15 20
X
(b) 2D Logarithmic image of the Capacitive Coupling Matrix
Figure 3.7: Logarithmic scale representation of Capacitive Coupling Matrix for

the system of 24 conductors on iron core - condensed arrangement
3.4. FROM SPARSE MATRIX TO CAPACITANCE NETWORK 33
in each example. The different threshold values in figure 3.8 are derived from
the maximum value in the Capacitive Coupling Matrix, which is 12.58 pF and
corresponds for the adjacency capacitance between the first two conductors on
each side of the first layer, i.e. capacitance value between conductors 1 & 2, and
between conductors 7 & 8 as shown in figure 3.6b.
Figure 3.9 shows more examples of sparse matrices with substantially lower
threshold values. These patterns show the outcome of finer limits set to the
capacitance values to be modelled. Such accurate patterns would be impossible
to predict without numerical modelling, i.e. with only anayltical techniques, or
with analytical techniques coupled with trial-and-error sensitivity analysis men-
tioned earlier in sub-sub-section 2.3 “Layer-to-Layer Approach for Slot Analysis”
and shown in figure 2.5. Threshold limits for capacitance values in the Capaci-
tive Coupling Matrix can be looked at as the minimum capacitance value to be
represented in the transmission line model.
3.4 From Sparse Matrix to Capacitance Net-

work
For a given threshold value that reflects the level of accuracy desired, the sparse
matrix is used to generate the LTspice netlist that represents the capacitance
values in the transmission line model of the winding. The transmission line
model is subsequently analyzed under LTspice frequency-domain analysis.
Below in figure 3.10 a threshold limit of 5% of the maximum value in the
Capacitive Coupling Matrix is chosen. The pattern shows, and is used to model,
the adjacent turn-to-turn capacitance for all conductors in the wiring arrange-
ment, meanwhile it shows the First Layer conductor-to-ground capacitances.
As previous patterns with lower threshold levels have suggested, and ac-
cording to the conclusion realized before that only the first copper conductor
layer facing the core would count in the capacitance-to-ground values; it is not
worthy to lower the threshold further and include the backward conductor lay-
ers in the model, since that only increases the complexity of the model without
adding any substantial accuracy.
Since the matrix is symmetric around the main diagonal in figure 3.10, only
one half of the matrix is considered when using it as an automatic input to the
Capacitive Sparse Matrix Capacitive Sparse Matrix

0 0
5 5
10 10
15 15
20 20
25 25
0 5 10 15 20 25 0 5 10 15 20 25
nonzeros=14 (2.4%) nonzeros=36 (6.3%)
(a) Threshold is 94% of the matrix maximum (b) Threshold is 88% of the matrix maximum
value. Pattern is maintained from 94%-98% of value. Pattern is maintained from 88%-93% of
maximum value maximum value

0 0
5 5
10 10
15 15
20 20
25 25
0 5 10 15 20 25 0 5 10 15 20 25
(c) Threshold 20% max. Maintained from 20%- (d) Threshold is 11% of max value. Pattern is
87% of max value, the broadest pattern maintained from 11%-19% of max value
Capacitive Sparse Matrix

0
10
15
20
25
0 5 10 15 20 25
nonzeros=82 (14.2%)
(e) Threshold is 5% of max value. Pattern is

maintained from 05%-08% of max value
Figure 3.8: Examples of sparse matrices with different threshold values for the
simple winding prototype - condensed arrangement

0 0
5 5
10 10
15 15
20 20
25 25
0 5 10 15 20 25 0 5 10 15 20 25
(a) Threshold is set to be of two orders of mag- (b) Threshold is set to be of three orders of
nitude below the matrix maximum value magnitude below the matrix maximum value
Figure 3.9: Sparse matrices with substantially lower threshold limits for the
simple winding prototype, the condensed wiring arrangement
Capacitive Sparse Matrix

0
10
15
20
25
0 5 10 15 20 25
nonzeros=82 (14.2%)
Figure 3.10: The sparse matrix chosen to be fed to the transmission line model
in LTspice
transmission line circuit model. As shown in figure 3.11 only the upper half is
taken.
Triangular Capacitive Sparse Matrix (after deleting redundant values)

0
10
15
20
25
0 5 10 15 20 25
nonzeros=45 (7.8%)
Figure 3.11: The upper half of the sparse matrix to be fed automatically to the
transmission line model in LTspice . No redundancy included
The resulting netlist file for the capacitance network in the transmission line
model is shown below. It is generated automatically from the processed sparse
matrix through a MATLAB code.
1 C 1−0 N001 0 2.420272e−012

2 C 1−2 N001 N002 1.257473e−011
3 C 2−0 N002 0 1.274958e−012
4 C 2−3 N002 N003 1.237769e−011
5 C 3−0 N003 0 1.135148e−012
6 C 3−4 N003 N004 1.236700e−011
7 C 4−0 N004 0 1.088643e−012
8 C 4−5 N004 N005 1.236629e−011
9 C 5−0 N005 0 1.088575e−012
10 C 5−6 N005 N006 1.236658e−011
11 C 6−0 N006 0 1.133048e−012
12 C 6−7 N006 N007 1.237622e−011
13 C 7−0 N007 0 1.271504e−012
14 C 7−8 N007 N008 1.257776e−011
15 C 8−0 N008 0 2.411273e−012
16 C 8−9 N008 N009 1.176068e−011
17 C 7−10 N007 N010 1.097555e−011
18 C 9−10 N009 N010 1.097355e−011
19 C 6−11 N006 N011 1.097403e−011

20 C 10−11 N010 N011 1.097453e−011
21 C 5−12 N005 N012 1.097580e−011
22 C 11−12 N011 N012 1.097313e−011
23 C 4−13 N004 N013 1.097410e−011
24 C 12−13 N012 N013 1.097628e−011
25 C 3−14 N003 N014 1.097770e−011
26 C 13−14 N013 N014 1.097429e−011
27 C 2−15 N002 N015 1.097599e−011
28 C 14−15 N014 N015 1.097765e−011
29 C 1−16 N001 N016 1.176263e−011
30 C 15−16 N015 N016 1.097829e−011
31 C 16−17 N016 N017 1.178963e−011
32 C 15−18 N015 N018 1.097689e−011
33 C 17−18 N017 N018 1.181399e−011
34 C 14−19 N014 N019 1.097381e−011
35 C 18−19 N018 N019 1.171499e−011
36 C 13−20 N013 N020 1.097684e−011
37 C 19−20 N019 N020 1.171617e−011
38 C 12−21 N012 N021 1.097569e−011
39 C 20−21 N020 N021 1.171870e−011
40 C 11−22 N011 N022 1.097463e−011
41 C 21−22 N021 N022 1.172336e−011
42 C 10−23 N010 N023 1.097594e−011
43 C 22−23 N022 N023 1.172585e−011
44 C 9−24 N009 N024 1.179031e−011
45 C 23−24 N023 N024 1.181752e−011
After the row number on the left side, the first column in the netlist file
represents the capacitance labels with subscripts indicating turn numbers. The
second and third columns represent the first and second nodes (turns) numbers,
respectively, which the capacitance links. The numbers are written in the format
that LTspice accepts (NXXX). The last column is the capacitance value in
Farads.
Figure 3.12 below shows the capacitance network equivalent to the generated
netlist file above, but drawn manually in LTspice .It is observed from the ca-
pacitance network that turn-to-ground capacitance values in the First layer are
lower than turn-to-turn capacitances. On average they are 13% lower; However
they will represent the dominant path to ground from any turn in the arrange-
ment. The equivalent of any series-connected capacitances is dominated by the
1
lower capacitance value which corresponds to higher impedance as Xc = ωC .
The significance of this method is appreciated more when applied to complex
situations with excessive number of turns, like the stator slot case with high fill
C_1-16 C_9-24
11.76p 11.79p
C_2-15 C_10-23
10.98p 10.98p
C_3-14 C_11-22
10.98p 10.97p
C_4-13 C_12-21
10.97p 10.98p
C_5-12 C_13-20
10.98p 10.97p
C_6-11 C_14-19
10.97p 10.97p
C_7-10 C_15-18
10.98p 10.98p
C_1-2 C_2-3 C_3-4 C_4-5 C_5-6 C_6-7 C_7-8 C_8-9 C_9-10 C_10-11 C_11-12 C_12-13 C_13-14 C_14-15 C_15-16 C_16-17 C_17-18 C_18-19 C_19-20 C_20-21 C_21-22 C_22-23 C_23-24
12.6p 12.38p 12.37p 12.37p 12.37p 12.38p 12.58p 11.76p 10.97p 10.97p 10.97p 10.98p 10.97p 10.98p 10.98p 11.79p 11.81p 11.71p 11.72p 11.72p 11.72p 11.73p 11.82p
C_1-0 C_2-0 C_3-0 C_4-0 C_5-0 C_6-0 C_7-0 C_8-0

2.42p 1.28p 1.14p 1.09p 1.09p 1.13p 1.27p 2.41p
Figure 3.12: Capacitance network equivalent to the code in the netlist file above,
drawn manually in LTspice
factor. Automating the drawing process then, by generating the netlist file,
becomes of high value since one is dealing with high-order transmission line
model representing adjacent turn-to-turn capacitances.
By automatically generating the netlist file, one avoids the burden of man-
ually drawing a complex capacitance network of high-order, which is a difficult
task for humans to do, consume lots of time and effort, and is usually and easily
prone to mistakes. A feeling of how complex the capacitance network from a
stator slot case might get, can be sensed from figure 3.12 above, which only rep-
resents the case of 24 turns in the winding arrangement prototype after being
subjected to a threshold limit of 5% of the maximum capacitance value in the
corresponding multi-conductor system.
Chapter 4
Inductive Coupling Matrix
After being able to investigate capacitance-to-ground and mutual capacitance

between conductors in a multi-conductor system from the previously introduced
Capacitive Coupling Matrix in chapter 3, a similar approach is followed when
investigating the inductance in the system.
The Inductive Coupling Matrix tells about self inductance for each conduc-
tor, as well as mutual inductance between any two conductors in the multi-
conductor system.
Here a special attention is given to the fact that unlike capacitance, in-
ductance is highly dependent on frequency, since the penetration depth of the
magnetic field inside various materials and media is dependent on its frequency
but also the properties of the medium being subjected to the field and pen-
etrated, namely the medium’s conductivity and permittivity according to the
formula 4.1 below [10][12]:

1
δ= (4.1)
πf μσ
where δ is the penetration depth of the magnetic field in the material, in

meters [m], π is the mathematical constant 3.142, f is the frequency in [Hz], μ
is the permeability of the material in [H/m] and σ is the material conductivity
in [S/m].
41
42 CHAPTER 4. INDUCTIVE COUPLING MATRIX
Figure 4.1 below shows the penetration depth (skin depth) in [mm] for dif-
ferent types of materials along a frequency sweep from 0 Hz to 1 MHz. Note
the general behavior that penetration depth in these materials decreases as fre-
quency increases. Note also that metal conductors like aluminum, gold and
copper have relatively the highest penetration of magnetic fields while ferro-
magnetic materials like iron has the lowest. This is mainly due to the higher
relative permeability (μr ) for iron compared to other elements. Although the
relative permeability (μr ) is a desirable characteristic which contributes to su-
perior magnetic behavior for the ferromagnetic material, it also contributes to
lower penetration depth. This can even be seen for different relative perme-
ability values within iron itself in figure 4.1. This fact makes us consider thin
laminations and low-frequency effectiveness for high permeability iron.
Skin depth vs frequency

2
10
Aluminum
Gold
copper
1
10 Silver
Iron ur=1000
Iron ur=3000
Iron ur=5000
Skin depth [mm]
0
10
−1
10
−2
10
−3
10
0 1 2 3 4 5 6
10 10 10 10 10 10 10
Frequency [Hz]
Figure 4.1: Penetration/Skin depth for different types of materials along a fre-
quency sweep from 0 Hz - 1 MHz
The 24-turn winding arrangement prototype is being modeled under FEMM

magnetostatic and time-harmonic magnetic analyses, in a way that application
of static and quasi-static solvers is justified by maintaining the cross-sectional
dimensions of the problem small compared to the wavelength of the analysis
frequency. By ’small’ one means the dimension is below ( λ4 ); the limit at which
the reflection effect of the electromagnetic wave can be noticeably observed
43
λ
[29][17]. A range for the dimension (d) where 10 < d < λ5 is acceptable in
the modelling depending on accuracy and simulation time sought.
The wavelength (λ) of an electromagnetic wave at a certain frequency, is
the distance in meters [m] which this wave travels in the medium (usually free
space/air) in one period of its frequency. In free space, electromagnetic waves
travel with the velocity of light c ≈ 3 ∗ 108 [m/s]. In power cables, the electro-
magnetic waves travel in the range of 40-60% of light speed [17], as the relative
dielectric constant (permitivity) for typical cable insulations varies from 3 to 8
[12][10].
Given the frequency of the wave to be (f) and the free space as the traveling
medium, then the wavelength (λ) in meters is expressed as:
c
λ= (4.2)
f
Table 4.1 and figure 4.2 below show some common frequencies versus their
corresponding wavelength in free space [7]. It covers the frequency range from
0 Hz - 30 MHz over which the conducted emission problem is most severe [15].
Frequency [Hz] (λ) in free space [m]

50 6 000 000
1 000 300 000
10 000 30 000
1 000 000 (1 MHz) 300 (AM radio band)
10 000 000 (10 MHz) 30
30 000 000 (30 MHz) 10
Table 4.1: Key frequencies versus their corresponding wavelength (λ) in free
space
Given the converter output voltage rise time (tr ), and the electromagnetic
wave propagation velocity in the conductor or cable of interest (vpropagation ),
then the wavelength corresponding to the critical conductor length (λcritical ) is
expressed as [17]:
λcritical = tr · vpropagation (4.3)

Frequency vs Wavelength in free−space

7
10
6
10
5
10
Wavelength [m]
4
10
3
10
2
10
1
10
2 4 6 8
10 10 10 10
Frequency [Hz]
Figure 4.2: Frequency versus the corresponding wavelength (λ) in free space for
frequency range from 50 Hz - 30 MHz
The built prototype model for the condensed wiring arrangements is mod-
elled as 24 copper conductors in a 2D planar problem. As in the electrostatic
analysis, the conductor (and core) depth inside the plane is 138 mm which is
equivalent to one turn’s length. The conductor diameter including insulation is
0.648 mm while copper diameter itself is 0.589 mm. Conductors are drawn with
their insulation and placed close to each other where center of conductors of
different layers lie on the same vertical axis. They are numbered in an orderly
manner as they were wound in reality. See figure 3.1 in the previous section.
They start from left to right in odd layer numbers and from right to left in even
layer numbers. The Bobbin on which the winding is wound, is represented by a
3 mm high dielectric material above the iron core.
The iron core in the model is represented by two parallel lines with a length
that is equal to the centerline length of the core’s peripheral, and a thickness
of 10 mm that corresponds to half its measured thickness due to magnetic
symmetry. An iron lamination is approximated to be 250μm and modeled in
the x-axis orientation under the magneto-static analysis, despite part of the core
laminations are oriented in the y-direction, as can be seen from figure 2.8 shown
45
previously. The reason being is that x-axis oriented laminations represent the
iron core part where the Bobbin and winding lie, in addition the approximation
of unfolding the core length to be represented in the 2D plane is taken from the
x-axis orientation perspective for the laminations sheets. The core depth inside
the plane is the same as one turn’s length which is 138 mm.
Between the two ends of the unfolded core, an arc with a radius of 160
mm is drawn to represent the outer boundary limit that is assigned “mixed”
boundary condition to model an unbounded problem [23]. Both ends of the core
are configured with “Periodic” boundary conditions [23] to reflect that they are
connected in reality and the resulting magnetic field will be the same at both
of them. The top line of the core is configured with local element size that is
dependent on the analysis frequency, and the bottom line is configured with a
prescribed zero boundary condition where the magnetic vector potential, A, has
a zero value. As the normal flux component is equal to the tangential derivative
of the magnetic vector A along the boundary, the magnetic flux lines vanish
when they approach this prescribed boundary. A screenshot of the described
FEMM magnetic problem is shown in figure 4.3 below.
The iron core is configured with an iron block property which has a linear
B-H relationship, as a simple case is considered to start with, besides there
was no reliable information available about this particular core. The relative
permeability (μr ) considered for iron is 5000 both in the x- and y- directions.
The laminations are considered to be laminated in x-direction for magneto-static
analysis and in into-the-plane direction for AC time-harmonic magnetic analysis
as the solver could not solve the proplem when having laminations oriented in
either x- or y- directions under time-harmonic magnetic analysis, the reason
being that on-edge laminations are not supported by the AC analysis solver
[23]. The iron block property details is shown in figure 4.4 belwo.
The analysis is performed by assigning each conductor in the system a cir-
cuit property with a current value of 1 A, while the rest of conductors are
assigned circuit properties of zero current values. The problem is solved for
the self-inductance value for the conductor of concern, i.e. the one which has
the stimulus current of 1 A, meanwhile it is also solved for the mutual induc-
tance between the conductor of concern and every other conductor in the system
during the same analysis run.
For magneto-static problems the analysis frequency is zero and so an un-
limited theoretical penetration depth is considered as per equation 4.1. For AC
time-harmonic magnetic problems, the analysis frequency is reflected by mak-
(a) A general meshed view of the FEMM magneto-static problem
(b) A zoomed-in view of the same FEMM magneto-static problem
(c) A zoomed-in view of one post-processor analysis run for

the same FEMM problem. The lines inside the core repre-
sent the magnetic flux lines (φ) or magnetic vector potential
(A) while the arrows around the conductors represent the
magnetic field intensity “H-field” vector plot
Figure 4.3: One of the magneto-static analysis runs for the FEMM model of the
winding arrangement on iron core prototype
47
Figure 4.4: Iron block property details for AC time-harmonic magnetic analysis
used for the core configuration
ing the element size defining a certain geometry, smaller than its corresponding
magnetic field penetration depth (δ), not least when drawing the iron core. The
same goal can be also achieved by making the mesh size of a certain geometrical
area smaller than its corresponding (δ).
Sub-figure 4.5a below shows the outcome of magneto-static analysis while
sub-figures 4.5b to 4.5h and figure 4.6 show the outcome of AC time-harmonic
magnetic analysis of the winding on core arrangement prototype for the con-
densed wiring arrangement. Magnetic flux lines are shown inside the iron core.
For magneto-static analysis the lines in the core represent the magnetic flux
lines while in AC analysis the lines represent real part of the vector magnetic
potential (A), as a complex number of (A) with both real and imaginary parts
would result from the analysis.
The significance of using the vector magnetic potential (A) in the analysis
is that its line integral around any closed path equals the total magnetic flux
passing through the area enclosed by that path as
ˆ ˆ ˛
Φ= B · ds = (∇ × A) · ds = A · dl [W b] (4.4)
S S C
The vector magnetic potential (A) itself can be found knowing the current
I which flows in a closed path in a conductor according to the equation [10]:
˛
μ0 I 1
A= dl [W b/m] (4.5)
4π R
C
where dl is the differential length element for the conductor and R is the distance
from the conductor at which A is calculated.
The magnetic flux lines shown in figure 4.5 and 4.6 are generated by a stim-
ulus current of 1 A in the last conductor of the condensed wiring arrangement,
i.e. conductor number 24, for all the cases shown. The plot options are set to
19 contour plot lines with a lower limit of 0 [Wb] and an upper limit of 200
[μWb] in the case of magneto-static analysis, while in the AC magnetic analysis
over 300 Hz a lower limit of -80 [μWb/m] and an upper limit of 200 [μWb/m]
were considered for the real part plot of the vector magnetic potential (A). The
introduction of a minus sign to the lower limit of the (A) plot is to include the
contribution of the core losses in the analysis plot. Core losses result in complex
49
(a) Magnetostatic @ 0 Hz (b) AC time-harmonic magnetic @ 1

Hz
(c) AC time-harmonic magnetic @ 10 (d) AC time-harmonic magnetic @

Hz 100 Hz
(e) AC time-harmonic magnetic 200 (f) AC time-harmonic magnetic @

Hz 300 Hz
(g) AC time-harmonic magnetic @ (h) AC time-harmonic magnetic @

400 Hz 500 Hz
Figure 4.5: Zoomed-in post-processor views of magneto-static and AC time-

harmonic magnetic FEMM analyses, for the 24 conductors on core - condensed
arrangement - at different frequencies
(a) AC time-harmonic magnetic @ 1 kHz (b) AC time-harmonic magnetic @ 2 kHz
(c) AC time-harmonic magnetic @ 5 kHz (d) AC time-harmonic magnetic @ 10 kHz
(e) AC time-harmonic magnetic @ 20 kHz
Figure 4.6: Zoomed-in post-processor views of AC time-harmonic magnetic

FEMM analysis for the 24 conductors on core - condensed arrangement - at
several “high” frequencies
51
number solution for (A) and a decrease of the resultant absolute value of (A)
as frequency increases.
In general terms, the magnetic field flux lines are highest in magnitude and
more intense within the upper part of the core, closer to the current-carrying
conductor, for each analysis case. Flux lines tend to vanish when they approach
the prescribed zero potential boundary condition assigned to the bottom line of
the core. Figure 4.7 below shows that for the magneto-static analysis case the
top 50 percent value of the magnetic flux lines exist within the upper half of the
core.
Figure 4.7: Zoomed-in post-processor views of magneto-static FEMM analysis,

showing the top 50% of magnetic field flux lines
From sub-figure 4.5a to 4.5f we note that in the magneto-static case and
also at “low” frequencies, the iron core acts as an attractive medium for the
magnetic field flux lines to pass through, in order to complete their path, linking
the current-carrying conductor to the core. This is due to the high permeability
of the ferro-magnetic core at low frequencies, and the lamination thickness being
smaller than the corresponding penetration depth. This contributes to higher
values of inductance at low frequencies.
As frequency increases for AC time-harmonic magnetic problems, the ferro-

magnetic iron core tends to loose this attractive feature of containing all the
magnetic field flux lines within. This is due to the associated core losses and
the nonlinear behaviour of the iron magnetic permeability (μr ) that affects the
penetrtaion depth of the magnetic field in iron at higher frequencies. This will
cause the inductance to decrease with the frequency increase as inductance is
magnetic flux divided by current as:
φ
L= (4.6)
i
Due to skin depth effect, the higher the frequency the less magnetic flux
lines tend to penetrate the core laminations. So the challenge is to predict, on
a case-by-case basis, at which point/range the frequency starts to influence the
inductance value and to quantify how much that would be compared to the dc or
“low” frequency value for a certain geometry and material characteristics. Here
the ferro-magnetic core plays a very important role since iron generally has the
lowest penetration depth among other elements and metals, as can be seen in
figure 4.1, due to the fact that it has high magnetic permeability (μr ) values and
the penetration depth is inversely proportional to the magnetic permeability as
per equation 4.1.
This explains why iron cores of high permeability values are effective only in
dc and “low” frequency range, while their characteristics degrade substantially
when considering “high” frequency operation. In our case of the winding-on-
core prototype and according to the FEMM analysis, a substantial decrease
of the inductance value to 15% lower than its corresponding dc value, can be
observed within the range of 300-400 Hz.
This frequency range can vary for different lamination thicknesses and for
iron additives and composites forming the lamination, but can still be expected
to fall within a frequency range of few 100’s of Hz to few kHz. One of the
thinnest iron core laminations observed is 100 [μm] and the range of relative
permeability (μr ) values considered are from 1000 to 9000 [11].
Flux lines inside the core will gradually decrease beyond this critical fre-
quency range or “knee frequency” untill eventually at very high frequencies
they tend to close their paths in the air without the need to pass through the
iron core. Iron will effectively behave like air for very high frequency magnetic
fields.
4.1 How the Inductive Coupling Matrix is Cre-

ated?
For a given frequency, the FEMM model of the winding-on-core prototype has
been run number of times equal to the number of conductors in the wiring
4.1. HOW THE INDUCTIVE COUPLING MATRIX IS CREATED? 53
arrangement. The reason being is that the stimulus current has been assigned to
one conductor at a time to find the self-inductance and the mutual-inductances
due to that stimulus, then, in the next run, the stimulus current is shifted to the
neighbouring conductor in the multi-conductor system. For each run this gives
a number of results equal to the number of conductors in the system, i.e. 24
results in our case (1 self-inductance and 23 mutual-inductance values). This is
repeated automatically until eventually creating a matrix system of 24×24 result
values, in which the main diagonal elements represent the self-inductance and
the off-diagonal elements represent the mutual-inductance values with symmetry
and redundancy around the main diagonal.
The self-inductance for the conductor of stimulus is calculated for each anal-
ysis run, according to the formula:
´
(A · J) · dV
Lself = (4.7)
i2
Where “A” is the magnetic vector potential in [Wb/m](or can also be re-
garded as magnetic flux (Φ in [Wb]), “J” is the current density in [A/m2 ],
“dV” is a volumetric differential element in [m3 ] and “i” is the stimulus current
through the conductor in [A]. The integrand part in equation 4.7 is configured as
a block integral which can be carried over a selection area in the FEMM post-
processor. This makes it possible to calculate the self-inductance for known
values of current [23].
Mutual inductance between the conductor of stimulus and every other con-
ductor in the system is calculated according to the relation:
´
(A1 · J2 ) · dV2
Lmutual = (4.8)
i1 · i2
Where “A1 ” is the magnetic vector potential produced by the conductor

of stimulus (the source conductor), “J2 ” is the current density resulted in the
other conductor in the system (passive conductor with no stimulus), “dV2 ” is a
volumetric differential element of the other conductor in [m3 ], “i1 ” is the current
of stimulus in the source conductor, and “i2 ” is the resulting induced current in
the other conductor in the system.
Considering that the total current in any coil is represented by the number
of turns times the current in a single turn of the coil, which is also equivalent
to the current density times the coil’s cross-sectional area. This is expressed, in
mathematical notation, for the other conductor in our system as:
n 2 · i 2 = J2 · a 2
Solving for the other conductor’s current density over current ( Ji22 ) in equation
4.8, we have:
ˆ
n2
Lmutual = ( )· A1 · dV2 (4.9)
a2 · i 1
The integrand part in equation 4.9 is defined as a block integral that can
be carried over a selection area in the FEMM post-processor. This enables the
calculation of mutual-inductance for known values of source current, number of
turns and cross-sectional area over which the integration is performed [23].
Figures 4.8 to 4.16 below show 3D plots and 2D images for the Inductive
Coupling Matrix resulted from the FEMM analysis under different frequency
values ranging from dc (0 Hz) up to 2 kHz. For AC analysis the resulting value
for the inductance will be a complex number. The imaginary component is
expressing the core losses. In the above mentioned plots the absolute values of
the complex-numbered inductance have been considered.
4.2 Complex-Numbered Inductance Concept

From the relationship in 4.6 when the magnetic flux lags the current in time-
harmonic problems, due to core losses of eddy currents and hysteresis, it will
result in complex-numbered inductance of which the imaginary part is a lumped
representation of the core losses.
When L is the complex-numbered inductance, it is represented as:
L = Lr − jLi (4.10)
√where Lr is the real part of the inductance, Li is the imaginary part and
j= −1 is the imaginary operator.
Considering the impedance of inductance to be: Z(ω) = jωL, where ω is
the angular frequency in [radians/second]. We will have:
4.2. COMPLEX-NUMBERED INDUCTANCE CONCEPT 55
Z(ω) = jωLr + ωLi (4.11)
The imaginary part of complex-numbered inductance contributes to a real

part of the frequency-dependent impedance Z(ω), associated with core losses.
One could interpret the (ωLi ) term as a frequency-dependent resistance which
will be used later to develop a circuit model for the core losses along a frequency
sweep. The real part of the inductance contributes to the imaginary part of Z(ω)
which is associated with inductive energy storage.
Table 4.2 show some key results out of the Inductive Coupling Matrices run
at various frequencies. It shows the maximum, the minimum and the mean value
of the matrix elements. It also shows the difference in absolute value between
the maximum and minimum inductance in the matrix, for each frequency. As
well it shows the coupling factor for each run which is defined as the average
inductance value divided by the maximum (self-inductance) value.
f [Hz] Lmax [μH] Lmin [μH] Laverage [μH] ΔL [nH] coupling factor
DC (0 Hz) 26.76 26.67 26.71 93 0.9978
10 Hz 26.76 26.66 26.70 93 0.9978
100 Hz 26.15 26.06 26.09 92 0.9978
200 Hz 24.54 24.45 24.49 88 0.9978
300 Hz 22.46 22.38 22.41 83 0.9977
400 Hz 20.33 20.25 20.28 79 0.9976
500 Hz 18.38 18.31 18.34 75 0.9975
1 kHz 12.24 12.17 12.20 66 0.9966
2 kHz 8.12 8.05 8.08 66 0.9949
Table 4.2: Key results out of Inductive Coupling Matrices for different frequency
values
By investigating figures 4.8 to 4.16, and table 4.2 above, we observe the
following:
(i.) The matrix elements maintain the same behaviour and have similar pat-
terns in all frequency runs, i.e. the highest values are for self-inductances
which lie in the main diagonal and then followed by mutual inductances
between adjacent conductors on the same layer or at different layers; First
off-diagonal elements above and below the main diagonal represent adja-
cancy on the same layer (symmetry and redundancy around the main
Inductive Coupling Matrix [μH] − @ 0 Hz − condensed arrangement

Difference between max and min 93.3 nH
Coupling factor: Average mean/Max value = 0.9978
26.76
26.75
26.8
26.74
Inductance [ μH ]
26.75 26.73
26.72
26.7 26.71
26.7
26.65
30 26.69
20 30
20 26.68
10
10
Y 0 0 X
(a) 3D plot of the Inductive Coupling Matrix at DC (0 Hz)

Image Display of Inductive Coupling Matrix [μH] − @ 0 Hz − condensed arrangement
26.76
26.75
5
26.74
26.73
10
26.72
Y
26.71
15
26.7
26.69
20
26.68
5 10 15 20
X
(b) 2D image of the Inductive Coupling Matrix at 0 Hz
Figure 4.8: Linear scale representation of Inductive Coupling Matrix run at 0 Hz

frequency for the system of 24 conductors on iron core - condensed arrangement

26.75
26.8 26.74
26.73
Inductance [ μH ]
26.75
26.72
26.71
26.7
26.7
26.65 26.69
30
30 26.68
20
20
10 26.67
10
Y 0 0 X
(a) 3D plot of the Inductive Coupling Matrix at 10 Hz

26.75
5 26.74
26.73
10 26.72
Y
26.71
15 26.7
26.69
20 26.68
26.67
5 10 15 20
X
Figure 4.9: Linear scale representation of Inductive Coupling Matrix run at 10

Hz frequency for the system of 24 conductors on iron core - condensed arrange-
ment

26.14
26.2 26.13
26.12
26.15
Inductance [ μH ]
26.11
26.1
26.1
26.05
26.09
26 26.08
30
30 26.07
20
20
10 26.06
10
Y 0 0 X

26.14
5 26.13
26.12
10 26.11
Y
26.1
15
26.09
26.08
20
26.07
26.06
5 10 15 20
X
Figure 4.10: Linear scale representation of Inductive Coupling Matrix run at

100 Hz frequency for the system of 24 conductors on iron core - condensed
arrangement

Difference between max and min 88 nH
24.54
24.53
24.6
24.52
24.55
Inductance [ μH ]
24.51
24.5 24.5
24.45 24.49
24.48
24.4
30
24.47
20 30
20
10 24.46
10
Y 0 0 X

24.54
24.53
5
24.52
24.51
10
24.5
Y
15 24.49
24.48
20 24.47
24.46
5 10 15 20
X

arrangement

22.45
22.5
22.44
Inductance [ μH ]
22.45 22.43
22.42
22.4
22.41
22.4
22.35
30
30 22.39
20
20
10 22.38
10
Y 0 0 X

22.45
5
22.44
22.43
10
22.42
Y
15 22.41
22.4
20 22.39
22.38
5 10 15 20
X

arrangement

20.32
20.4
20.31
20.35
Inductance [ μH ]
20.3
20.3
20.29
20.25
20.28
30 20.27
20 30
20.26
20
10
10
Y 0 0 X

20.32
5
20.31
10 20.3
Y
20.29
15
20.28
20.27
20
20.26
5 10 15 20
X

arrangement

18.38
18.45 18.37
18.36
Inductance [ μH ]
18.4
18.35
18.35
18.34
18.3 18.33
30
20 30 18.32
20
10
10 18.31
Y 0 0 X

18.38
5 18.37
18.36
10
18.35
Y
15 18.34
18.33
20
18.32
18.31
5 10 15 20
X

arrangement
Inductive Coupling Matrix [μH] − @ 1e3 Hz − condensed arrangement

12.23
12.26
12.22
12.24
Inductance [ μH ]
12.22 12.21
12.2
12.2
12.18
12.19
12.16
30
20 30 12.18
20
10
10
Y 0 0 X
(a) 3D plot of the Inductive Coupling Matrix at 1 kHz

Image Display of Inductive Coupling Matrix [μH] − @ 1e3 Hz − condensed arrangement
12.23
5
12.22
10 12.21
Y
12.2
15
12.19
20
12.18
5 10 15 20
X
(b) 2D image of the Inductive Coupling Matrix at 1 kHz

1 kHz frequency for the system of 24 conductors on iron core - condensed ar-
rangement
Inductive Coupling Matrix [μH] − @ 2e3 Hz − condensed arrangement

8.11
8.14
8.1
8.12
Inductance [ μH ]
8.1 8.09
8.08
8.08
8.06
8.07
30
20 30 8.06
20
10
10
Y 0 0 X
(a) 3D plot of the Inductive Coupling Matrix at 2 kHz

Image Display of Inductive Coupling Matrix [μH] − @ 2e3 Hz − condensed arrangement
8.11
5
8.1
10 8.09
Y
8.08
15
8.07
20
8.06
5 10 15 20
X
(b) 2D image of the Inductive Coupling Matrix at 2 kHz

2 kHz frequency for the system of 24 conductors on iron core - condensed ar-
rangement
4.3. SINGLE TURN ANALYSIS 65
diagonal) while the two lines perpendicular to the main diagonal repre-
sent adjacancy between first and second layers for one line, and between
second and third layers for the other.
(ii.) Unlike the previously studied Capacitive Coupling Matrix, the Inductive
Coupling Matrix has a smooth transition concerning the absolute values
of adjacent elements.
(iii.) The coupling factor for all frequency runs was high and averaged to 0.9973
which makes it fair to conclude a very high coupling factor for an orderly-
wound condensed wiring arrangement.
(iv.) An interesting obesrvation was the substantial decrease in absolute values
of the Inductive Coupling Matrix elements when going from dc analysis
to 2 kHz, compared to the difference between maximum and minimum
elements in the matrix (ΔL); As it can be seen, for example, for Laverage
which has dropped from 26.71 [μH] at dc to 8.08 [μH] at 2 kHz, with a
70% decrease, while the correponding (ΔL) has dropped from 93 [nH] to
66 [nH] with a 29% decrease.
From the above, and despite the limitation of running only time-harmonic
magnetic field analysis with no account for electromagnetic fields, we conclude
that for simplification purposes it is sufficient to calculate the inductance of a
single conductor in the arrangement, and then consider a high coupling factor
between the conductors of the orderly-wound condensed wiring arrangement.
4.3 Single Turn Analysis
As per the observations and conclusions obtained from the Inductive Coupling
Matrix by studying self-inductance and mutual inductances for each conductor
in the multi-conductor system, the attenntion is focused now on studying the
inductance and resistance behaviour of a single turn, in the 24-turn winding-
on-core prototype, along a frequency sweep.
The objective is to make use of the results found in the previous section
in order to have a clearer picture about the frequency impact on basic prop-
erties on the turn level, to simplify the calculations with fair assumptions and
develop a corresponding circuit model covering the frequency range of interest.
A circuit model representing the turn’s frequency-dependent inductance and re-
sistance can be used later, along with the capacitance network (see figure 3.12)
for building up the transmission line model of the winding. Transmission line
model will be extended to the machine case for studying conducted emissions
in the electric drive.
The same FEMM model which has been used previously to generate the
Inductive Coupling Matrix is used here along the conducted-emission frequency
range. The simplification of extracting the properties of only one conductor from
the model has been implemented. These properties include self inductance with
its magnitude, loss angle, real and imaginary parts as the inductance becomes
a complex number for time-harmonic magnetic analyses. As well resistive, hys-
teresis and eddy current losses in the copper conductor, the iron core and the
dielectric Bobbin. Figure 4.17 below show some key results out of this simplified
analysis.
By comparing the loss angle and the imaginary inductance behaviour along
the frequency sweep, we notice that the loss angle starts from zero at dc and
increases with the frequency until it reaches a maximum peak of 45◦ at 1 kHz,
after which it starts to decrease again, while the imaginary inductance part has
a similar behaviour, however the peak of 11 [μH] occurs earlier at 500 Hz. This
is due to the decrease in magnitude of the complex-numbered inductance after
what is known as “knee frequency” which is in our case falls between 300 and
500 Hz. This can be seen clearly in the logarithmic scale plot of the inductance
magnitude against frequency in figure 4.18.
Magnitude decrease of inductance is related to lower skin/penetration depth
compared to “half” the given dimension of a lamination sheet in the iron core.
For a thickness of 250 μm for one lamination, a change in the magnetic field flux
pattern is expected to happen when the penetration depth becomes less than
half the dimension, as magnetic field is expected to penetrate the lamination
from both sides. A change of the flux pattern will affect mainly the inductance.
For the given 250 μm thickness of an iron lamination, and for several com-
mon permeability values (μr ), the penetration depth corresponding to 125 μm
occurs at following frequencies as per Table 4.3
Figure 4.19 show different types of losses associated with a single turn in
the 24-turn wiring arrangement. For the iron core, sub-figure 4.19a, we notice
that the dominant type of losses along the frequency sweep are hysteresis and
eddy current losses which will reflect on the inductance value. The resistive
losses in the iron are negligible. Losses in the iron core are generally negligible
4.3. SINGLE TURN ANALYSIS 67
Absolute value of Inductance vs Frequency Loss angle (degrees) vs Frequency

30 50
45
25
40
Absolute Inductance [μH]
35
Loss Angle [degrees]

20
30
15 25
20
10
15
10
5
5
0 0
0 2 4 6 8 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]
(a) Absolute value of inductance vs frequency (b) Loss angle vs frequency
Real Inductance vs Frequency Imaginary Inductance vs Frequency

30 12
25 10
Imaginary Inductance [μH]
Real Inductance [μH]
20 8
15 6
10 4
5 2
0 0
0 2 4 6 8 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10
(c) Real inductance part vs frequency (d) Imaginary inductance part vs frequency
Figure 4.17: Complex inductance components along the frequency sweep ex-
tracted from the FEMM model of a single conductor in a 24-conductor winding-
on-core prototype
Iron relative permeability (μr ) Penetration depth [μm] Frequency [Hz]

1000 126 1600
3000 125 540
5000 126 320
Table 4.3: Frequency occurrence of a penetration depth corresponding to “half”

the lamination thickness for different relative permeability values μr
logarithmic scale of absolute value of Inductance vs Frequency

2
10
Absolute Inductance [μH] 1

10
0
10
−1
10
0 2 4 6 8
10 10 10 10 10
Frequency [Hz]
Figure 4.18: A logarithmic scale of absolute value of the complex-numbered

inductance in [μH] vs frequency in [Hz]
at dc and very low frequency until they exceed the 20 [mW] limit around 316
[Hz] according to our analysis, as can be seen clearly in figure 4.20.
For copper conductor losses 4.19b we notice that resistive losses are the
dominant ones while hysteresis and eddy current losses are negligible. This is
mainly due to relatively high electrical conductivity and diamagnetic properties
of the copper. Copper losses are significant at DC level with a value of 8.7
[mW] in the copper conductor and maintains constant level along low frequency
up to the limit where skin depth in the copper becomes significant, where they
will increase substantially. This is observed to be around 50 [kHz] where losses
in one conductor reach the limit of 9 [mW] as can be seen in figure 4.21. At
50 [kHz] the penetration depth in copper is equivalent to 0.296 [mm], as per
equation 4.1, which roughly accounts for half the copper conductor dimension
of 0.589 [mm] in diameter.
4.4 Single Turn Model along a Frequency Sweep
In order to develop a circuit model for a single turn along a frequency sweep of
interest, three main quantities need to be investigated:
(i.) The inductance value of the winding at dc and very low frequencies, when
and how its magnitude will be substantially affected by frequency increase.
4.4. SINGLE TURN MODEL ALONG A FREQUENCY SWEEP 69
PFe vs Frequency PCu vs Frequency

Total losses ≈ HE losses Total losses ≈ Resitive losses
12 0.14
HE losses Fe HE losses CU
resistive losses Fe 0.12
10 Resistive losses Cu
total losses Fe Total losses Cu
0.1
8
Power Losses [W]
Power Losses [W]
0.08
6
0.06
4
0.04
2
0.02
0 0
0 2 4 6 8 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10
(a) Iron core losses vs frequency (b) Copper conductor losses vs frequency
Losses corresponding to imaginary part of Inductance vs Frequency

12
10
8
Losses [W]
0
0 2 4 6 8
10 10 10 10 10
Frequency [Hz]
(c) Losses corresponding to imaginary induc-

tance part (ωLi ) vs frequency
Figure 4.19: Different types of losses associated with a single turn in a 24-turn
winding-on-core prototype, along a frequency sweep
LOG plot of PFe vs Frequency

5
10
HE losses Fe
4
10
3
10
Power Losses [mW]
2
10 X: 316.2
Y: 20.65
1
10
0
10
−1
10
−2
10
−3
10
0 2 4 6 8
10 10 10 10 10
Frequency [Hz]
Figure 4.20: A logarithmic scale of hysteresis and eddy current iron core losses
in [mW] vs frequency in [Hz]
LOG plot of PCu vs Frequency

3
10
Resistive losses Cu
2
10
Power Losses [mW]
1
10
X: 5e+004
Y: 8.916
0
10
0 2 4 6 8
10 10 10 10 10
Frequency [Hz]
Figure 4.21: A logarithmic scale of the resistive losses in the copper conductor
in [mW] vs frequency in [Hz]
(ii.) The resistive copper losses at dc and low frequencies and then at which
frequency theses losses increase significantly.
(iii.) Core losses and at which frequency they become significant and how would
they develop with the frequency increase.
In order to do that we make a table, and record the above mentioned quan-
tities for dc and very low frequency, and then for each decade of frequency
increase as it shows in table 4.4.
f rCu [mΩ] rcore [mΩ] L [μH]
DC (0 Hz) 8.7 0 26.76
10 Hz 8.7 3 · 10−2 26.76
316 Hz 8.7 21 22.11
1 kHz 8.7 54 12.24
10 kHz 8.7 159 3.72
100 kHz 9.5 508 1.28
1 MHz 23.4 1648 0.50
10 MHz 73.6 5767 0.26
30 MHz 134.5 11 925 0.21
Table 4.4: Key results out of a single turn analysis along a frequency sweep
To quantify the change and relate it to the frequency increase, two criti-
cal frequencies are observed; one is related to both core losses and inductance
magnitude, and the other is related to resistive copper losses.
The frequency related to the resistive copper losses is determined by the
skin depth in the material compared to half the given dimension of the copper
conductor, and also the resolution of frequency points taken to develop the
model. In our case the frequency corresponding to the skin depth in copper is
50 kHz but since the resolution is taken every one decade after the 1 kHz limit,
the critical frequency corresponding to copper losses becomes 100 kHz for the
sake of developing the model.
The frequency related to both inductance magnitude and core losses (fcr1 )
is determined by the skin depth in the fundamental magnetic core unit in com-
parison to “half” its dimension. In our case fcr1 = 316 Hz corresponding to
half an iron lamination sheet of a thickness 250 [μm] and a relative magnetic
permeability (μr = 5000). Obviously this frequency can be shifted according to
the geometrical dimensions and material characteristics of the basic core unit
used; whether it is a bulk iron core, an iron lamination sheet or even a small tiny
magnetic particle of magnetic powder used in SMC and SM 2 C. The general
trend would be the smaller the dimension of the basic core unit, the higher the
critical frequency (fcr1 ) would be.
The selection of the first critical frequency (fcr1 ) is chosen as per the maxi-
mum core losses for the minimum change in the inductance value. Referring to
equation 4.10 defining the complex inductance concept, fcr1 is chosen according
to the relation:
Rcore ωLi
fcr1 = max( ) = max( ) (4.12)
|L| Lr
where Lr is the real part & Li is the imaginary part of the complex-numbered
inductance resulting from the FEMM analysis. fcr1 is physically related to the
magnetic field penetration of the iron core.
For the sake of developing the model, it is recommended to choose (fcr1 ),
as close as possible to the penetration depth frequency in the basic core unit,
and not to be governed by the frequency resolution considered when developing
the model, even if it falls beyond the 1 kHz limit. Unlike resistive copper losses,
which will occur further in the frequency scale and will be offset by the core
losses, critical frequency related to inductance and core losses (fcr1 ) tends to
occur in low frequency range, i.e. between few 100’s Hz to few kHz.
Later we quantify the factors contributing to the magnitude change of the
three basic quantities when stepping from each key frequency to the next. we
make table 4.5.
Starting from the third frequency step at the critical frequency (fcr1 ), the
complex-numbered inductance will be represented by two passive elements in
the circuit model; an inductance and a resistance (see figure 4.22). The real part
of the complex-numbered inductance (Lr ) is expressed by an inductive energy
storage element of inductance 22.11 μH, while the core losses are lumped and
expressed by a resistive element of a 21 mΩ resistance corresponding to ωLi ,
where Li is the imaginary part of the complex-numbered inductance as was
shown in equations 4.10 and 4.11.
Then from the next decade point in the frequency scale and up to the
highest (1 kHz – 10 MHz), we relate the real inductance part and core losses
to their corresponding values at (fcr1 ), our starting limit. This is done by
Factors for (rcu ) Factors for Factors for (L)

Frequency steps
change (rcore ) change change
DC (0 Hz) – 10 Hz 1 ∞ 1
10 Hz – 316 Hz 1 700 0.8262
316 Hz – 1 kHz 1 2.5714 0.5536
1 kHz – 10 kHz 1 2.9444 0.3039
10 kHz – 100 kHz 1.0920 3.1950 0.3441
100 kHz – 1 MHz 2.4632 3.2441 0.3906
1 MHz – 10 MHz 3.1453 3.4994 0.5200
10 MHz – 30 MHz 1.8274 2.0678 0.8077
Table 4.5: Quantification factors of key parameters corresponding to frequency

steps - single turn analysis
proposing two constants “a” and “b”; “a” to be multiplied by the resistive
part expressing core losses magnitude increase, and “b” to be divided by the
inductive part expressing the inductance magnitude decrease for each decade
increase in frequency. Constants “a” and “b” are shown in table 4.6.
Frequency- Factors for (rcore ) Factors for (L) Reciprocal of (L)

decades increase decrease decrease factors
316 Hz – 1 kHz 2.5714 0.5536 1.8064

1 kHz – 10 kHz 2.9444 0.3039 3.2906
10 kHz – 100 kHz 3.1950 0.3441 2.9061
100 kHz – 1 MHz 3.2441 0.3906 2.5602
1 MHz – 10 MHz 3.4994 0.5200 1.9231
Table 4.6: Quantification factors for core losses increase and inductance magni-
tude decrease for 5 decades of frequency - single turn analysis
Factor “a” is then taken to be the average of all (rcore )-icrease factors for
successive frequency decade shifts (a = 3.10 in our case). Factor “b” is chosen
according to the relation:
10
b= (4.13)
a
So that we can add one additional RL branch for each decade of frequency.
Factor “b” from equation 4.13 = 3.23, which is not far from the reciprocal
average of FEMM result for inductance decrease (2.50). The deviation in in-
ductance magnitude decrease will be compensated by the frequency term in the
impedance which will be relatively high.
The next step will be to multiply the resistive part representing the core
losses with the factor “a”, and to divide the inductive part by the factor “b”, in
the circuit model for each successive branch. The resulting model will be a ladder
of RL parallel-connected branches where the number of branches indicates how
many frequency decades the model covers. The model acts like concentric shells
in which the inner most represents the lowest frequency and the outer most the
highest. The model is shown in figure 4.22 below:
.param r_core 21m ;

.param L_ac 22.11u ;
.param a 3.10 ;
.param b 3.24 ;
R1 L1 From First critical frequency (f_cr1) to 1 kHz
{r_core} {L_ac}
R2 L2
From 1 kHz to 10 kHz
{a*r_core} {L_ac/b}
R3 L3
{a*a*r_core} {L_ac/(b*b)}
R4 L4
From 100 kHz to 1 MHz
{a*a*a*r_core} {L_ac/(b*b*b)}
R5 L5
From 1 MHz to 10 MHz
{a^4 * r_core} {L_ac / b^4}
R6 L6
{a^5 * r_core} {L_ac / b^5}
Figure 4.22: A ladder model for a single turn’s ac inductance and core losses
covering a frequency sweep starting from first critical frequency (fcr1 = 316 Hz)
up to 100 MHz
The reason why we have used equation 4.13 in deriving factor “b” from “a”
is that we are interested in having the corner frequency ωk for each RL branch to
be one decade, in the frequency scale, further from the previous one, according
to the relation:
Rk
ωk = = 10 · ωk−1 (4.14)
Lk
where ωk is the corner frequency in [rad/sec] for the k th RL-branch in the

system, and represents a zero of that branch in the pole-zero plot. k=1,2,...,n
where n is the number of RL-branches. Figure 4.23 and 4.24 show the pole-
zero plot and the bode plot, respectively, for the six RL parallel-connected
branches of the system. Please note that for bode plot the decibel calculation for
magnitude on the y-axis is calculated according to the relation 20·log10 | Z(jw) |
, where | Z(jw) | is the frequency response for the branch.
Notice the occurrence of corner frequency for each successive branch fur-
ther in the frequency than its preceding one. The higher resistance value con-
tributes to higher low-frequency response and the lower inductance value for
lower high-frequency response. The response of the equivalent combination of
all the branches besides the response of the branches themselves are shown in
figures 4.25 and 4.26
We notice that the equiuvalent combination of the 6 branches is in a good
agreement with the high-frequency behaviour for each branch after its corre-
sponding corner frequency. In the highest frequency range it agrees with the be-
haviour of the outer most RL branch, the purple line representing Z{10 MHz ↔ 100 MHz} .
However, we see that there is a clear difference between the equivalent system of
branches and the original branch Z{316 Hz ↔ 1 kHz} at the low frequency range
before its corner frequency.
A need to compensate for that rises, by introducing compensation factors for
the resistance and inductance parts in the first RL branch Z{316 Hz ↔ 1 kHz}
to account for the deviation of the equivalent branch system in the dc and
low-frequency region [12].
Following the approach presented in [30], for modelling skin effect in air-
surrounded conductors using passive circuit elements, and extending that to
our case for the frequency range of interest, we can represent each branch in the
prallel combination as:
Zk = Rk + j · Xk (4.15)
Pole−Zero Map
1
1 1 1 1 3.06e−
Z316 Hz
0.8 Z
1 kHz
Z
10 kHz
0.6 Z100 kHz
Z1 MHz
0.4
Z
10 MHz
Imaginary Axis
0.2
8 8e+007 6e+007 4e+007 2e+007

0
System: Z_{10 MHz} System: Z_{1 MHz}
−0.2 Zero : −9.5e+007 Zero : −9.5e+006
Damping: 1 Damping: 1
Overshoot (%): 0 Overshoot (%): 0
−0.4 Frequency (rad/sec): 9.5e+007 Frequency (rad/sec): 9.5e+006
−0.6
−0.8
1 1 1 1 3.06e−
−1
−10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0
Real Axis 7
x 10
(a) A general view of the pole-zero map showing all the zeros for the system
of RL branches, separately
−3 Pole−Zero Map
x 10
2 1 1 1 4.59e−009 Z
316 Hz
Z1 kHz
1.5 Z
10 kHz
Z100 kHz
1 Z1 MHz
System: Z_{1 kHz}
Zero : −9.5e+003 Z10 MHz
0.5 Damping: 1
Imaginary Axis
Overshoot (%): 0
Frequency (rad/sec): 9.5e+003
0
2e+005 1.5e+005 1e+005 5e+004
System: Z_{10 kHz} System: Z_{316 Hz}
Zero : −9.5e+004 Zero : −950
−0.5 Damping: 1 Damping: 1
Frequency (rad/sec): 9.5e+004 Frequency (rad/sec): 950
−1
−1.5
−2
1 1 1 4.59e−009
−2 −1.5 −1 −0.5 0 0.5 1
Real Axis 5
x 10
(b) A zoomed-in view of the pole-zero map showing the first three zeros for
the system of RL branches
Figure 4.23: Pole-zero plots for the RL-branch ladder representing a single turn’s
inductance and core losses, and covering a frequency sweep starting from 1 kHz
up to 100 MHz - Labels in the legend indicate the start of the decade
Bode Diagram
100
80
60
Magnitude (dB)
40
20
0
System: Z_{100 kHz}
−20 Frequency (rad/sec): 9.5e+005
Magnitude (dB): −1.06
−40
90
Z
316 Hz
Z
1 kHz
Phase (deg)
Z
10 kHz
45 Z
100 kHz
Z
1 MHz
Z
10 MHz
0
2 4 6 8 10
10 10 10 10 10
Frequency (rad/sec)
Figure 4.24: Bode plot for the system of RL-branch ladder representing a single
turn’s inductance and core losses, and covering a frequency sweep starting from
1 kHz up to 100 MHz - Labels in the legend indicate the start of the decade
Pole−Zero Map
1
1 1 1 1 3.06e−
Z316 Hz
0.8 Z
1 kHz
Z10 kHz
0.6 Z100 kHz
Z1 MHz
0.4
Z10 MHz
Zeq
Imaginary Axis
0.2
8 8e+007 6e+007 4e+007 2e+007

0
System: Zeq System: Zeq
−0.2 Zero : −9.5e+007 Pole : −3.53e+007
−0.4 Frequency (rad/sec): 9.5e+007 Frequency (rad/sec): 3.53e+007
−0.6
−0.8
1 1 1 1 3.06e−
−1
−10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0
Real Axis 7
x 10
(a) A general view of the pole-zero map showing all the poles and zeros for
the system of RL branches and their equivalence
−3 Pole−Zero Map
x 10
1 1 1 1 1 1.35e−009
Z
3 316 Hz
Z
1 kHz
2.5 Z10 kHz
1
Z100 kHz
2
Z
1 MHz
1.5 Z10 MHz
Zeq
Imaginary Axis
0.5
1e+005 8e+004 6e+004 4e+004 2e+004

0
System: Zeq System: Z_{316 Hz}
−0.5 Zero : −9.5e+004 Zero : −950
−1 Overshoot (%): 0 Overshoot (%): 0
Frequency (rad/sec): 9.5e+004 Frequency (rad/sec): 950
−1.5
−2
1 1 1 1 1 11.35e−009
−10 −8 −6 −4 −2 0
Real Axis 4
x 10
(b) A zoomed-in view of the pole-zero map showing the first three zeros
(circles) and the first pole (x) for RL branches and their equivalence
Figure 4.25: Pole-zero plots for the system of RL branches and their equivalence
representing a single turn’s inductance and core losses, and covering a frequency
sweep of 5 decades starting from 1 kHz up to 100 MHz
Bode Diagram
100
80
60
Magnitude (dB)
40
20
−20
−40
90
Z
316 Hz
Z1 kHz
Phase (deg)
Z10 kHz
45 Z100 kHz
Z1 MHz
Z
10 MHz
Zeq
0
2 4 6 8 10
10 10 10 10 10
Frequency (rad/sec)
Figure 4.26: Bode plot for the system of RL branches and their equivalence,
representing a single turn’s inductance and core losses, and covering a frequency
sweep of 5 decades starting from 1 kHz up to 100 MHz
where k = 1,2,..n, is the branch number, and n=6 is the number of branches in
the system for the frequency sweep of interest
1
Yk =
Rk + j · Xk
Rk − j · Xk
=
Rk 2 + Xk 2
F or Xk Rk
1 Xk
Yk −j· (4.16)
Rk Rk 2
As we are interested in the low frequency range just before the corner fre-
quency of the first original branch Z{316 Hz ↔ 1 kHz} , then it is fair to assume
that Xk Rk in equation 4.16. Adding all the real parts of admittances (con-
ductances) for the branch system to give the conductance of the first original
branch with compensation factor, we have:
(Y1 + Y2 + Y3 + Y4 + Y5 + Y6 ){316 Hz ↔ 1 kHz} ≡ Y1 {316 Hz ↔ 1 kHz}
For the real parts we have:
1 1 1 1 1 1 1
( + + + + + ) ≡
R1 R2 R3 R4 R5 R6 R1
Considering Rk = ak−1 · R1 , where (k = 1, 2, ..., 6)
1 1 1 1 1 1 1
( + + 2 + 3 + 4 + 5 ) ≡
R1 a · R1 a · R1 a · R1 a · R1 a · R1 R1
Introducing “c1 ” as a constant to be multiplied with the resistive part of the

first RL branch in the parallel combination, so that the branch system behaves
like the original first branch at low frequency. Expressing R1 = Rf cr , the core
loss equivalent resistance at the first critical frequency, we have:
1 1 1 1 1 1 1
(1 + + 2 + 3 + 4 + 5 ) =
c1 · Rf cr a a a a a Rf cr
1 1 1 1 1
c1 = 1 + + + 3+ 4+ 5
a a2 a a a
1
n−1
c1 = (4.17)
ak
k=0
where n = 6 is the number of branches in the system. Given that a =

3.10 ⇒ c1 = 1.4745
From equation 4.17, we observe that the compensation factor for the resistive
part is dependent on factor “a”, the resistance increase factor which reflects core
losses increase. The resistance part (rcore ) in the branch system of figure 4.22
shall be updated to c1 · rcore = 1.4745 · rcore .
For the imaginary parts we have:
X1 X2 X3 X4 X5 X6 X1
( 2 + 2 + 2 + 2 + 2 + 2) ≡
R1 R2 R3 R4 R5 R6 R1 2
Considering that Rk = ak−1 · R1 & Xk = bX 1

k−1 , where (k=1,2,...,6) and putting
in the constant “c1 ” we just found from equation 4.17, we have:
X1 X1 X1 X1 X1
1 X1 b2 b3 b4 b5 X1
· ( 2 + b
+ + + + ) ≡
c1 2
R1 (a · R1 )2 2
(a · R1 )2 3
(a · R1 )2 4
(a · R1 )2 5
(a · R1 )2
R1 2
Introducing “c2 ” as a constant to multiply the left-hand side with, so that

we get the right-hand side equivalence. Taking note that (R1 = Rf cr ), we have:
X1 1 1 1 1 1 X1
(c2 · ) · (1 + 2 + + 2 + 2 + 2 )=
c1 2 · Rf2 cr (a · b) (a2 · b)2 (a · b)3 (a · b)4 (a · b)5 Rf cr 2
c1 2
c2 = 1 1 1 1 1
1+ (a2 ·b) + (a2 ·b)2 + (a2 ·b)3 + (a2 ·b)4 + (a2 ·b)5
n−1 1 2
c1 2 ( k=0 ak
)
c2 = n−1 1
= n−1 1
(4.18)
k=0 (a2 ·b)k k=0 (a2 ·b)k
where n is the number of branches in the system, “a” is the increase factor of
resistance & “b” is the decrease factor of inductance in each successive branch,
and “c1 ” is the compensation factor for the resistive part in the first branch.
Given that a = 3.10, b = 3.23 & c1 = 1.4745 ⇒ c2 = 2.1041
From the relation 4.18 we observe that the compensation factor for the
inductive part is not only dependent on the inductance decrease factor “b”, but
also the resistance increase factor “a”.
Generating bode plot for the branch system with compensated factors in
the first branch, along with the original fisrt branch we have:
So from figure 4.27 we can see an accomplishment of a satisfactory high-
frequency behaviour for the system of parallel branches expressing the induc-
tance decrease and core losses, meanwhile we see a very good agreement between
the equivalent branch system and the original branch representing the first crit-
ical frequency, fcr1 , (i.e. the first branch before introducing the compensation
factors) in the low-frequency range.
We also notice that the high frequency behaviour is even lower than the
response of the outer most branch (see figure 4.28). This makes sense as we
parallel-connect all the branches in the system and the resulting equivalent
impedance/response will be lower than the response for any branch.
We observe that the high frequency behaviour of the branch system over 107
[rad/sec] has been reduced from about 60 dB to 12 dB compared to the case
of the originial branch expressing (fcr1 ). On the other hand the low-frequency
behaviour has a mismatch of 1.3 dB compared to the same branch. From the
fourth decade up to the eighth (104 up to 108 ), the mismatch has been between
4–6 dB with less than 4 dB difference in the high frequency range above the 8th
decade [rad/sec]. This is not the complete model as we will continue developing
the model to include the phenomena at dc and very low frequency.
The next and final step in developing the complete model for the single turn
along a frequency sweep, is to add the resistive copper losses to the model along
with the ac part that was developed earlier. At very low frequency close to dc
the resistive copper losses are dominant and the inductance effect is negligible.
Bode Diagram
100
80
60
Magnitude (dB)
40
20
−20
−40 Z316 Hz
90
comp−Z316 Hz
Z1 kHz
Phase (deg)
Z
10 kHz
45 Z
100 kHz
Z1 MHz
Z
10 MHz
Zeq
0
2 4 6 8 10
10 10 10 10 10
Frequency (rad/sec)
Figure 4.27: Bode plot for the system of RL branches and their equivalence
after inserting compensation factors in the first branch
R1 L1 From First critical frequency (f_cr1) to 1 kHz
{c1 * r_fcr} {c2 * L_fcr}
R2 L2
.param r_fcr 21m ; From 1 kHz to 10 kHz
.param L_fcr 22.11u ; {a * c1 * r_fcr} {c2 * L_fcr / b}
.param r_cu 8.7m ;
R3 L3
.param a 3.1;
.param b {10/a}; {a*a* c1 * r_fcr} {c2*L_fcr /( b*b)}
R4 L4
.param c1 1.4745 ; From 100 kHz to 1 MHz
.param c2 2.1053 ; {a^3 * c1 * r_fcr} { c2*L_fcr / b^3 }
R5 L5
{a^4 * c1 * r_fcr} { c2*L_fcr / b^4 }
R6 L6
{a^5 * c1 * r_fcr} {c2*L_fcr / b^5}
Figure 4.28: An improved ladder model for a single turn’s ac inductance and
core losses covering a frequency sweep including and starting from 1 kHz up to
100 MHz
The inductance at low frequency will be taken care of by the first branch in the
proposed ladder model. A mismatch between the proposed model and the FEM-
M/measurement data could rise for the frequency range where the inductance
is not associated with significant core losses, which is usually limited along the
frequency band. In that case a simple RL model with no compensation factor
can fill the gap, in which the resistance expresses the dc copper resistive losses
(RCu ), and the inductance is the FEMM magnetostatic inductance value.
The challenge is to mainatin the dc and low frequency behaviour of the
whole combination, meanwhile not affecting the accomplished high-frequency
behaviour. This is the reason why we will introduce a third and final compen-
sation factor “c3 ” for the resistive copper losses (RCu ) in the model.
Following the same approach in the system equations 4.17 & 4.18, and this
time for the real part of the whole combination, we have:
RCu + Req ≡ RCu
c3 · RCu + Req = RCu

Req
c3 = 1 −
RCu
but
1
Req = 1 1 1 1 1 1
c1 ·Rf cr + R2 + R3 + R4 + R5 + R6
Considering that Rk = ak−1 · c1 · Rf cr , where (k = 1,2,...,n & n = 6)
1
Req = 1 1 1 1 1 1
c1 ·Rf cr + a·c1 ·Rf cr + a2 ·c1 ·Rf cr + a3 ·c1 ·Rf cr + a4 ·c1 ·Rf cr + a5 ·c1 ·Rf cr
1
= 1 = Rf cr
c1 ·Rf cr · c1
=⇒
Rf cr
c3 = 1 − (4.19)
RCu
From equation 4.19, we observe that the compensation factor for the resistive
copper losses “c3 ” is dependent on:
- The resistive copper losses value itself (Rcu ).
- The resistance of the first branch of the ac inductance model (Rf cr ), which
stands for the maximum core loss corresponding to minimum change in ac
inductance.
- Compensation factor “c1 ” introduced to the first branch resistance.
So the compensation factors “a” and “b” along with “c1 ”, “c2 ”, and “c3 ”,
which are needed to cover the whole frequency sweep for a single turn model,
are summarized in the following table:
The complete model for a single turn of copper winding on an iron core, after
including the core loss model and introducing the three compensation factors
Constant Formula Value
a fromm FEMM analysis 3.10

b b = 10 3.23
n−1 a 1
c1 k=0 ak 1.4745
2
c2 n−1c1 1
2.1053
k=0 (a2 ·b)k
Rf cr
c3 1− RCu -1.5414
Table 4.7: Summary of the compensation factors for a single turn’s complete
model along a frequency sweep from dc up to 100 MHz
to cover the whole frequency sweep, will look like:
R7 R1 L1 From First critical frequency (f_cr1) to 1 kHz
{c3 * r_cu} {c1 * r_fcr} {c2 * L_fcr}
R2 L2
.param r_fcr 21m ; From 1 kHz to 10 kHz
.param L_fcr 22.11u ; {a * c1 * r_fcr} {c2 * L_fcr / b}
.param r_cu 8.7m ;
R3 L3
.param a 3.1;
.param b {10/a}; {a*a* c1 * r_fcr} {c2*L_fcr /( b*b)}
R4 L4
.param c1 1.4745 ; From 100 kHz to 1 MHz
.param c2 2.1053 ; {a^3 * c1 * r_fcr} { c2*L_fcr / b^3 }
.param c3 -1.5414 ;
R5 L5
{a^4 * c1 * r_fcr} { c2*L_fcr / b^4 }
R6 L6
{a^5 * c1 * r_fcr} {c2*L_fcr / b^5}
Figure 4.29: A single turn of copper winding on an iron core model covering a
frequency sweep from dc up to 100 MHz
As can be seen from figure 4.29, we need 3 fundamental quantities to be fed

from the FEMM analysis to the circuit model, which are; the resistive copper
losses in the copper conductor (Rcu ), the maximum core losses corresponding
to minimum change in inductance value (Rf cr ), and the inductance value at
that frequency point (Lf cr ). In addition to the constant ”a” that indicates the
increase of core losses for each successive decade of frequency, 3 more secondary
constants will be needed. Constants “c1 ”, “c2 ” and “c3 ” are derived from the
fundamental quantities as per equations 4.17, 4.18 & 4.19, respectively.
Plotting Bode-plot for the complete turn’s model reposnse (red curve) along
with the response of the dc (RCu Ldc ) branch (yellow stars), the first critical
frequency (Rf cr Lf cr ) branch (black) , and the ac model for inductance and core
losses (green) will give:
Bode Diagram
100
Magnitude (dB)
50
−50
90
Phase (deg)
Z
dc
45
Z
fcr
Zcore losses
Zcomp−Rcu + core losses
0
2 4 6 8 10
10 10 10 10 10
Frequency (rad/sec)
Figure 4.30: Bode plot for the system of RL branches and their equivalence
after inserting compensation factors in the first branch
Notice the very good agreement both in low and high frequencies. Ob-
serve that the high frequency behaviour reflecting core losses is now not directly
proportional to the frequency, rather it counts for high frequency losses.
Chapter 5
Model Validation
Different measurements have been taken for the winding-on-core prototype

shown in figure 2.8 by the hp impedance analyzer 4194A. Measurements of
impedance frequency response along a frequency band from dc up to 40 MHz
for different acces points of the prototype have been performed. The following
measurements have been recorded:
(1.) Excitation of the measurement device was set between the winding input
at the 1st access point, and the winding output at the 4th access point
measuring the three conductor layers response.(see table 2.1)
(2.) Excitation was set between the winding input at the 1st access point, and
the second layer input at the 2nd access point measuring the first layer
response.
(3.) Excitation was set between the winding second layer input at the 2nd access
point, and third layer input at the 3rd access point measuring the second
layer response.
(4.) Excitation was set between the winding third layer input at the 3rd access
point, and the winding output at the 4th access point measuring the third
layer response.
the winding third layer input at the 3rd access point measuring the first
two layers response.
89
90 CHAPTER 5. MODEL VALIDATION
point, and the winding output at the 4th access point measuring the last
two layers response.
the magnetic iron core at the 5th access point measuring the winding input-
to-ground response.
point, and the magnetic iron core at the 5th access point measuring the
winding first layer-to-ground response.
(9.) Excitation was set between the winding third layer input at the 3rd access
point, and the magnetic iron core at the 5th access point measuring the
winding first and second layer-to-ground response.
(10.) Excitation was set between the winding third layer output (winding output)
at the 4th access point, and the magnetic iron core at the 5th access point
measuring the winding first,second and third layer-to-ground response.
(11.) Short-circuiting the winding input at the 1st access point, and output at
the 4th access point and then setting the excitation between the common
short circuit point and the magnetic iron core at the 5th access point in
order to cancel the inter-winding capacitances and highlight the lumped
sum of all capacitances to ground contributed by each turn in the winding.
(12.) Short-circuiting the winding input at the 1st access point, second layer in-
put at the 2nd access point, third layer input at the 3rd access point, and
winding output at the 4th access point and then setting the excitation be-
tween the common short circuit point and the magnetic iron core at the 5th
access point in order to further cancel any inter-winding capacitances and
parasitic capacitances due to acces points conductors. This measurement
also highlights the lumped sum of all capacitances to ground contributed
by each turn in the winding.
Figures 5.1 and 5.2 show the response for different layers in the wiring
arrangement where excitation was set between different access points of the
winding. The behaviour is inductive in all the measurements up until the first
resonance frequency at 5-6 MHz. An interesting observation when scaling a sin-
gle layer measurement to fit a double layer’s as in figure 5.2 or a tripple layer’s
as in figure 5.1; is that there’s very good agreement between the scaled mea-
surement and the actual one up to the first resonance frequency. The capacitive
91
Impedance Magnitude
80
First Layer
70 Second Layer
Third Layer
Impedance magnitude [20log10(|Z|)] [dB]
60 All three layers

2
First Layer * (# of layers = 3)
50
40
30
20
10
−10
−20
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]
Figure 5.1: Measurements of impedance magnitude for the winding-on-core pro-

totype along a frequency band from 0 Hz up to 40 MHz - single and tripple layers
Impedance Magnitude
80
First Layer
70 Second Layer
Third Layer
60 All three layers

First Two Layer
50 Last Two Layers
First Layer * (# of layers = 2)2
40
30
20
10
−10
−20
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]
Figure 5.2: Measurements of impedance magnitude for the winding-on-core pro-

totype along a frequency band from 0 Hz up to 40 MHz - single layer, two layers
and three layers
5.1. CAPACITIVELY COUPLED OPTIMIZED VALIDATION 93
behaviour occuring after the first resonance has significantly increased (3 dB

difference on average). This can be seen by the lower negative slope line on the
logarithmic impedance-frequency scale which indicates higher capacitance value
1
as we are plotting (Xc = ωC ) vs. frequency.
This is partly due to the fact that capacitances associated with a single
layer are not scaled as the inductance by multiplying with the number of layers
squared. The other reason is that considering the higher value of turn-to-turn
capacitance (Ctt ) compared to the turn to ground’s (Ctg ) as it was found in
the capacitive coupling analysis earlier; having fewer turns will result in higher
lumped equivalent capacitance especially when viewing the turn-to-turn capac-
itances to be series-connected between the ends of the layer. The equivalent
capacitance for series-connected capacitances is given by:
1
Ceq = n 1 (5.1)
k=0 ck
where n is the number of series-connected capacitances.

In the experimental verification of the model two model versions were used,
one optimized towards capacitively coupled model, and the other optimized
towards the inductively coupled model.
5.1 Capacitively Coupled Optimized Validation

In the case of the model optimized towards capacitive coupling, 3 versions of
the model were used where the capacitance network on the turn level, generated
in chapter 3 has been used in all of them. The three versions of the model are:
(1.) The capacitance network is used to represent turn-to-turn capacitances
(Ctt ) and turn-to-ground capacitances (Ctg ). The turn resistance and in-
ductance are chosen to be the dc quantities of the analysis in section 4.
(2.) In addition to the capacitance network in the model, the turn inductance
is represented by the the first 2 decades of the ac inductance and core loss
model developed in section 4.4.
(3.) In addition to the capacitance network in the model, the turn inductance
is represented by the the first 3 decades of the ac inductance and core loss
model developed in section 4.4.
Measurement for a single layer containing 8 turns is taken as the mea-

surement reference in the validation process, since it represents the minimum
section measured within the wiring arrangement. As we have noticed earlier
in figure 5.2, single, double and tripple layers have identical behaviour (given
the proper scaling factor) up until the first resonance frequency, and similar
behaviour after the first resonance.
The simulation model has been subjected to the same exitation which was
set in the measurement, i.e. for the first layer to be between the winding input
at the 1st access point, and the second layer input at the 2nd access point. The
model measuremenet layout and measurement results are shown in the figures
below:
From figure 5.3 we observe that there is a significant gain differenece between
the two curves in the range of (20 - 30 dB) during the inductive behaviour of the
simulation model, as no core loss effect is represented; however the resonance
frequency occurence of the simulation model is at 4.3 MHz which is very close
to the measurement’s at 5.2 MHz.
From figures 5.4 and 5.5, we see a tendency to reduce the gain difference
between the two curves as we include more decades of the core loss model. This
is done on the expense of shifting the difference between the first resonance
ocurrences of the two curves further.
In figure 5.4 we see a difference in the range of (12 - 15 dB) between the
two curves during the inductive behaviour, indicating a relative reduction in the
difference since a core loss representation covering two decades of frequency on
the turn level has been inserted; however the first resonance frequency of the
simulation model has become 9 MHz, increasing the difference in first resonance
frequency occurrence to be almost 4 MHz.
On the other hand we notice a very good agreement in figure 5.5 between
the simulation model output and measurements during the inductive behaviour,
when a core loss model of 3 branches representing three frequency decades has
been inserted; however this comes with the expense of even further distance be-
tween the first resonance frequency occurrences for the two curves as it becomes
more than 12 MHz.
C_1-16 C_9-24
11.76p 11.79p
C_2-15 C_10-23
10.98p 10.98p
C_3-14 C_11-22
10.98p 10.97p
C_4-13 C_12-21
10.97p 10.98p
C_5-12 C_13-20
10.98p 10.97p
C_6-11 C_14-19
10.97p 10.97p
.ac dec 400 1 30e6

C_7-10
C_15-18
.param Rdc 8.7m ; 10.98p
10.98p
.param Ldc 26.76u ;
R7 L7 R6 L6 R5 L5 R4 L4 R3 L3
R1 L1 R2 L2 R8 L8
{Rdc} {Ldc} {Rdc} {Ldc} {Rdc} {Ldc} {Rdc} {Ldc} {Rdc} {Ldc}
{Rdc} {Ldc} {Rdc} {Ldc} {Rdc} {Ldc}
V1
SINE(0 1 50)
AC 1
Rser=0
C_1-0
C_2-0 C_3-0 C_4-0 C_5-0 C_6-0 C_7-0 C_8-0
2.42p
1.28p 1.14p 1.09p 1.09p 1.13p 1.27p 2.41p
C_1
2p
(a) Layout of the simulation model optimized towards capacitive coupling where turn resis-
tance and inductance are the dc quantities
Impedance Magnitude −− Measured vs Simulated
120
Measurement: First Layer

100 Simulation: Equivalent Spice Model
80
60
40
20
−20
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]
(b) Agreement between measurement data and simulation model output - Ca-
pacitively coupled optimized model version-1
Figure 5.3: Layout of simulation model optimized towards capacitive coupling

version-1 and its agreement with measurement data
C_1-16 C_9-24
11.76p 11.79p
C_2-15 C_10-23
10.98p 10.98p
C_3-14 C_11-22
10.98p 10.97p
C_4-13 C_12-21
10.97p 10.98p
C_5-12 C_13-20
10.98p 10.97p
C_6-11 C_14-19
10.97p 10.97p
.ac dec 400 1 30e6

C_7-10
C_15-18
10.98p
.param r_fcr 21m ; 10.98p
.param L_fcr 22.11u ;
.param a 3.1;
.param b {10/a};
.param c1 1.4745 ;
.param c2 2.1053 ; C_1-2 C_2-3 C_3-4 C_4-5 C_5-6 C_6-7 C_7-8 C_8-9 C_9-10 C_10-11 C_11-12 C_12-13 C_13-14 C_14-15 C_15-16 C_16-17 C_17-18 C_18-19 C_19-20 C_20-21 C_21-22 C_22-23 C_23-24
R51 L51
R22 L22 R27 L27 R31 L31 R37 L37 R42 L42 R5 L5 R16 L16
{c1 * r_fcr}
{c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr}
{c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr}
{c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr}
V1 R49 L49
R20 L20 R2 L2 R7 L7 R35 L35 R6 L6 R3 L3 R1 L1
{a * r_fcr}
SINE(0 1 50) {a * r_fcr} {a * r_fcr} {a * r_fcr} {a * r_fcr} {a * r_fcr} {a * r_fcr} {L_fcr / b} {a * r_fcr}
{L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b}
AC 1
Rser=0
C_1-0
C_2-0 C_3-0 C_4-0 C_5-0 C_6-0 C_7-0 C_8-0
2.42p
1.28p 1.14p 1.09p 1.09p 1.13p 1.27p 2.41p
C_1
2p
(a) Layout of the simulation model optimized towards capacitive coupling where a core loss
model covering 2 decades on the turn level has been inserted
140
120
Measurement: First Layer (8 turns)

Equivalent Simulation Model
100
80
60
40
20
−20
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]

C_1-16 C_9-24
11.76p 11.79p
C_2-15 C_10-23
10.98p 10.98p
C_3-14 C_11-22
10.98p 10.97p
C_4-13 C_12-21
10.97p 10.98p
C_5-12 C_13-20
10.98p 10.97p
C_6-11 C_14-19
10.97p 10.97p
.ac dec 400 1 30e6

C_7-10
C_15-18
10.98p
10.98p
.param r_fcr 21m ;
.param L_fcr 22.11u ;
.param a 3.1;
.param b {10/a};
.param c1 1.4745 ;
.param c2 2.1053 ;
R51 L51
R27 L27 R31 L31 R37 L37 R42 L42 R5 L5 R16 L16
R22 L22 {c1 * r_fcr}
{c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c1 * r_fcr} {c2 * L_fcr} {c1 * r_fcr}
{c1 * r_fcr} {c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr} {c2 * L_fcr} R49 L49 {c2 * L_fcr}
R2 L2 R7 L7 R35 L35 R6 L6 R3 L3 R1 L1
{c2 * L_fcr}
V1 R20 L20 {a * r_fcr}
{a * r_fcr} {a * r_fcr} {a * r_fcr} {a * r_fcr} {a * r_fcr} {L_fcr / b} {a * r_fcr}
{a * r_fcr} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b} {L_fcr / b}
R50 L50
SINE(0 1 50) {L_fcr / b} R26 L26 R8 L8 R36 L36 R41 L41 R4 L4 R15 L15
AC 1 R21 L21 {a*a* r_fcr}
Rser=0 {a*a* r_fcr} {a*a* r_fcr} {a*a* r_fcr} {a*a* r_fcr} {a*a* r_fcr} {a*a* r_fcr}
{L_fcr / (b*b)}
{a*a* r_fcr}
C_1-0 {L_fcr / (b*b)} {L_fcr / (b*b)} {L_fcr / (b*b)} {L_fcr / (b*b)} {L_fcr / (b*b)} {L_fcr / (b*b)}
{L_fcr / (b*b)} C_2-0 C_3-0 C_4-0 C_5-0 C_6-0 C_7-0 C_8-0
2.42p
1.28p 1.14p 1.09p 1.09p 1.13p 1.27p 2.41p
C_1
2p
(a) Layout of the simulation model optimized towards capacitive coupling where a core loss
model covering 3 decades of frequency has been inserted
140
Measurement: First Layer
120 Simulation: Equivalent Spice Model
100
80
60
40
20
−20
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]

5.2 Inductively Coupled Optimized Validation

For the single turn simulation model with core loss representation which is
optimized towards inductive coupling, the model has two capacitances to ground
(Ctg ) connected on each end of the model and a one turn-to-turn capacitance
(Ctt ) connected across the model, since the turn is considered to be in the first
layer. The model layout and the comparison between simulation outcome and
measurement is shown below.
In figure 5.6b the measurement data representing 8 turns is compared against
the simulation model output of one turn. The model output is scaled as per the
number of turns in the first layer, and also as per the number of turns giving
the best fit to the measurements. The following observations can be noticed:
(1.) Inductively optimized turn model has no indication of any resonance fre-
quency behaviour despite the representation of capacitances to ground
(Ctg ) and turn-to-turn capacitance (Ctt ) in the model.
(2.) The single turn model doesnt scale to the 8-turn-layer measurement, in the
inductive part of the measurement data. The best fit in our case indicates
5 turns instead of 8. This strongly suggests that capacitive coupling has
prominent effect resulting in effectively bypassing 3 turns out of 8 in order
to get the best fit within the inductive region.
5.2. INDUCTIVELY COUPLED OPTIMIZED VALIDATION 99
.param r_fcr 21m; .param c1 1.4745;

.param a 3.1; .param ctt 12.3p; .param c2 2.1053;
.param L_fcr 22.11u; .param b {10/a};
.param r_cu 8.7m; .param ctg 1.1p; .param c3 -1.0940;
C3
{ctt}
R7 R1 L1
{c3 * r_cu} {c1 * r_fcr} {c2 * L_fcr}
V1
R2 L2
{a * r_fcr} {L_fcr / b}
SINE(0 1 50)
AC 1
R3 L3
Rser=0
{a*a* r_fcr} {L_fcr / (b*b)}
C2 C1
{ctg} R4 L4 {ctg}
{a*a*a* r_fcr} {L_fcr / (b*b*b)}
R5 L5
{a^4 * r_fcr} {L_fcr / b^4}
R6 L6
{a^5 * r_fcr} {L_fcr / b^5}

.ac dec 400 1 30e6
C4
2p
(a) Layout of the simulation model of a single turn with core loss
model and optimized towards inductive coupling
100
Measurement: First Layer (8 turns)
80 Simulation: Single Turn
Simulation: Single Turn * 82

Simulation: Single Turn * 52
60
40
20
−20
−40
2 3 4 5 6 7 8
10 10 10 10 10 10 10
Frequency [Hz]
(b) Agreement between measurement data and simulation model output
Figure 5.6: Layout of simulation model optimized towards inductive coupling

and its agreement with measurement data
Chapter 6
Conclusions and Future

Work
Given the possibility of representing high frequency phenomena with SPICE

circuit models and the methodology followed throughout this thesis in proposing
the circuit model elements, the following conclusions are observed:
- The number of branches representing the frequency decades in the core loss
model on the turn level seems to be oversized when including it with the
capacitance network in the capacitively coupled model (subsection 5.1) for
model version-2 (sub-figure 5.4a) and version-3 (sub-figure 5.5a). It appears
that due to capacitive coupling among turns, 2 or 3 branches of the core loss
model is adequate in representing the frequency dependence of the core loss
and consequently the inductance along the frequency sweep of interest.
- Capacitive coupling among the winding layers are more influential in having
a very good agreement both for the resonance behaviour of the winding and
the gain difference within the inductive region.
- When scaling one turn model to fit a single layer measurement of a certain
number of turns, the study suggests only portion of the number of turns may
be considered. In our case 58 ≈ 63% of the actual turns. This suggests that
capacitive coupling between the turns effectively bypass some turns and so
affecting the gain within the inductive region of the model.
101
102 CHAPTER 6. CONCLUSIONS AND FUTURE WORK
There is no clear cut single model when it comes to representing the high
frequency behaviour of a winding under SPICE. We could optimize the model
representation towards capacitive or inductive coupling.
When optimized towards capacitive coupling, only dc quantities of the turn
resistance and inductance seems to be adequate in telling the resonance fre-
quency behaviour of the model. For having good agreement within the inductive
region of the measurement, three branches of the core loss model on the turn
level is enough when inserted in the capacitance network of the capacitively
coupled model.
When optimized towards inductive coupling, we can only have agreement
within the inductive region of the measurement data, as there is no indication
about the frequency resonance behaviour of the simulation model output. In
order to get good agreement in that region, only portion of the turns may be
considered as capacitive coupling between the turns is effectively bypassing some
of them.
Further investigation about the best methodology of combining capacitive
and inductive coupling outcomes of the model, in an automated way and in-
troducing desired thresholds and accuracy levels, may be needed. Capacitive
or inductive coupling by itself can not be calculated from first principles as we
have noticed throughout this thesis. We have seen the need of using simulation
software package like FEMM to perform these analyses. So the combination pro-
cess would rather account for similar accuracy and complexity levels in order to
produce narrow and more specific conclusions.
Future work may also include the extension of the winding-on-core study
to the stator machine case, since more details about the ferromagnetic mate-
rial of the laminations could be available from the manufacturer. Magnetic
characteristics of the lamination proved to be crucial in the inductive coupling
behaviour of the winding and the machine generally. Analysis on the turn level
of a winding inside a machine stator slots, and the comparison to measurements
on numbered and pre-determined access points, can give more insights about
the high-frequency behaviour of a winding inside machine slots.
Glossary
Bobbin A cylinder or spindle on which the wire, thread or film is wound. It

can be with or without flanges and can be typically found in wire rolls,
sewing machines, cameras, and within electronic equipment. 39
End Turn The winding portion which lies outside the active length of the
machine stator. It does not contribute to torque production while it con-
tributes to the losses. It can extend up to 13 of the active axial length on
each side of the stator especially for small machines. iii, 11, 12, 14
Integral Slot The number of slots per pole per phase is an integer. iii, 10–12
Pole Pitch The mechanical angular distance between consecutive (electro)magnetic

poles expressed in mechanical units; For an electric machine of N-poles,
◦
the pole pitch τp in mechanical degrees will be 360
N . 12
Single Layer Winding a winding configuration where one side of the winding
totally occupies one slot. iii, 10, 11, 13
103
104 Glossary
Acronyms
ANSI American National Standards Institute. 101
CISPR International Special Committee on Radio Interference. 101

CM Common-Mode. 4, 6
dB Decibel. viii, 27, 105–108

DM Differential-Mode. 2
DSO Digital Storage Oscilloscope. 9
EDM Electric Discharge Machining. 6

ELP Earth Leakage Protection. 5
EMC ElectroMagnetic Compatibility. 1, 2, 6, 8, 10, 12, 100, 103
EMI ElectroMagnetic Interference. i, 2, 6, 8, 101
ESS Energy Storage System. 1, 2, 6, 7, 9
F Farad. 3, 4
FEM Finite Element Method. 9, 19
hp Hewlett Packard. i, 7, 9, 17, 83
IEC International Electrotechnical Commission. 101
105
106 Acronyms
IGBT Insulated Gate Bipolar Transistor. 4
n nano. 4
p pico. 3
PWM Pulse-Width Modulation. 2, 8–10, 100–102
q Slots per Pole per Phase. iii, 10–13
struct structure array. 9
VCC Volvo Car Corporation. 8, 100

References
[1] J. Adabi, F. Zare, G. Ledwich, and A. Ghosh. Leakage current and common
mode voltage issues in modern ac drive systems. In Power Engineering
Conference, 2007. AUPEC 2007. Australasian Universities, pages 1 –6,
December 2007. 2, 5
[2] M. Alaküla, April 2009. A professor in industrial electrical engineering at

Lund Institute of Technology (LTH) - Lund University, and also a Senior
Scientific Advisor at Volvo AB with emphasis on hybrid vehicles. 2
[3] M. Alaküla and P. Karlsson. Power Electronics Devices, Converters, Con-

trol and Applications. Department of Industrial Electrical Engineering and
Automation, Lund University, 2003. 2
[4] S. Alexandersson, M. Alaküla, and H. Bängtsson. Prediction of parasitic

components in an automotive environment. In Electromagnetic Compat-
ibility, 2008. EMC 2008. IEEE International Symposium on, pages 1 –7,
August 2008. 2
[5] S. Alexandersson and H. Bängtsson. Analytic prediction of electromagnetic

behaviour. In International Symposium on Power Electronics, Electrical
Drives, Automation and Motion, 2008. SPEEDAM 2008, pages 338 –343,
June 2008. 2
[6] H. Berg, May 2009. Electronic Engineer at VCC with a focus on EMC in
cars. 2, 8
[7] P. A. Chatterton and M. A. Houlden. EMC: Electromagnetic Theory to

Practical Design. Wiley, Chichester, 1992. 43
107
108 REFERENCES
[8] S. Chen, T.A. Lipo, and D. Fitzgerald. Source of induction motor bearing
currents caused by PWM inverters. Energy Conversion, IEEE Transactions
on, 11(1):25 –32, March 1996. 5
[9] Sh. Chen and T.A. Lipo. Bearing currents and shaft voltages of an induc-
tion motor under hard and soft switching inverter excitation. In Industry
Applications Conference, 1997. Thirty-Second IAS Annual Meeting, IAS
’97., Conference Record of the 1997 IEEE, volume 1, pages 167 –173 vol.1,
October 1997. 5
[10] D. K. Cheng. Field and Wave Electromagnetics. Addison-Wesley series in

electrical engineering. Addison-Wesley, 1989. 15, 25, 41, 43, 48
[11] Cogent Typical Data. Non oriented electrical steel, June 2012. Suraham-
mars Bruks AB, PO Box 201, SE-735 23 Surahammar, Sweden. 52
[12] K. R. Demarest. Engineering Electromagnetics. Prentice-Hall International,

London, 1998. 41, 43, 75
[13] J.M. Erdman, R.J. Kerkman, D.W. Schlegel, and G.L. Skibinski. Effect of
PWM inverters on ac motor bearing currents and shaft voltages. Industry
Applications, IEEE Transactions on, 32(2):250 –259, March/April 1996. 5
[14] O. Henze, Z. Cay, O. Magdun, H. De Gersem, T. Weiland, and A. Binder.

A stator coil model for studying high-frequency effects in induction motors.
In International Symposium on Power Electronics, Electrical Drives, Au-
tomation and Motion, 2008. SPEEDAM 2008, pages 609 –613, June 2008.
17
[15] Specifications in ANSI C63.2 and/or CISPR 16-1-1:2006. American Na-

tional Standards Institute (ANSI), International Special Committee on
Radio Interference (CISPR) - International Electrotechnical Commission
(IEC). 5, 9, 43
[16] DASOL Product Information. Round enamelled winding wire of copper,

solderable, class 155, April 2011. AB DAHRNTRÅD, Jonslund, SE-465 80
Nossebro, Sweden. 23
[17] J. Luszcz. Motor cable effect on the converter fed ac motor common-
mode current. In Compatibility and Power Electronics (CPE), 2011 7th
International Conference-Workshop, pages 445 –450, June 2011. 43
REFERENCES 109
[18] J. Luszcz and K. Iwan. Modelling conducted EMI in inverter-fed ac mo-

tor. In Compatibility in Power Electronics, 2005. IEEE, pages 209 – 212,
January 2005. 2
[19] J. Luszcz and K. Iwan. Conducted EMI propagation in inverter-fed ac
motor. In Electrical Power Quality and Utilisation Magazine, Vol II, No.
1, pages 47 – 51, 2006. 2
[20] O. Magdun, A. Binder, C. Purcarea, and A. Rocks. High-frequency in-
duction machine models for calculation and prediction of common-mode
stator ground currents in electric drive systems. In Power Electronics and
Applications, 2009. EPE ’09. 13th European Conference on, pages 1 –8,
September 2009. 18
[21] O. Magdun, A. Binder, A. Rocks, and O. Henze. Prediction of common-
mode ground current in motors of inverter-based drive systems. In Electrical
Machines and Power Electronics, 2007. ACEMP ’07. International Aegean
Conference on, pages 806 –811, September 2007. 18
[22] P. Maki-Ontto, H. Kinnunen, and J. Luomi. Three-phase model for the cir-
cuit simulation of common-mode phenomena and shaft voltages in ac motor
drive systems. In Electric Machines and Drives, 2005 IEEE International
Conference on, pages 437 –443, May 2005. 5
[23] D. Meeker. Finite Element Method Magnetics - User’s Manual, Version
4.2. [email protected], October 2010. 24, 45, 53, 54
[24] V. Mihaila, S. Duchesne, and D. Roger. A simulation method to predict the
turn-to-turn voltage spikes in a PWM fed motor winding. Dielectrics and
Electrical Insulation, IEEE Transactions on, 18(5):1609 –1615, October
2011. 18
[25] N. Mohan, T. M. Undeland, and W. P. Robbins. Power Electronics: Con-
verters, Applications and Design. Wiley, New York, 1995. 2
[26] B. Muralidhara, A. Ramachandran, R. Srinivasan, and M.C. Reddy. Ex-
perimental measurement of shaft voltage and bearing current in an inverter
fed three phase induction motor drive. In Electronics Computer Technology
(ICECT), 2011 3rd International Conference on, volume 2, pages 37 –41,
April 2011. 2
[27] R. Naik, T.A. Nondahl, M.J. Melfi, R. Schiferl, and Jian-She Wang. Circuit
model for shaft voltage prediction in induction motors fed by PWM-based
110 REFERENCES
ac drives. Industry Applications, IEEE Transactions on, 39(5):1294 – 1299,

September-October 2003. 5
[28] S. Ogasawara and H. Akagi. Modeling and damping of high-frequency
leakage currents in PWM inverter-fed ac motor drive systems. Industry
Applications, IEEE Transactions on, 32(5):1105 –1114, September-October
1996. 2, 3
[29] T. Rybak and M. Steffka. Automotive Electromagnetic Compatibility
(EMC). Kluwer Academic Publishers, 2004. 43
[30] B.K. Sen and R.L. Wheeler. Skin effects models for transmission line struc-
tures using generic spice circuit simulators. In Electrical Performance of
Electronic Packaging, 1998. IEEE 7th Topical Meeting on, pages 128 –131,
October 1998. 75
[31] J. Welinder, January 2011. A Deputy Manager at SP Technical Research
Institute of Sweden, with background and 24+ years of experience in EMC
laboratory measurements and testing. 20
[32] B.W. Williams. Principles and Elements of Power Electronics: Devices,
Drivers, Applications, and Passive Components. Barry W Williams, 2006.
2
[33] T. Williams. EMC for Product Designers. Newnes, Oxford, 2007. 5
[34] R.A. Witte. Spectrum and Network Measurements. Prentice Hall, 1993.
31, 117
[35] M.T. Wright, S.J. Yang, and K. McLeay. General theory of fast-fronted in-
terturn voltage distribution in electrical machine windings. Electric Power
Applications, IEE Proceedings B, 130(4):245 –256, July 1983. 18
Appendices
111
Appendix A
Decibel
dB is a logarithmic unit expressing a ratio. The ratio can describe two power
quantities, currents or voltages. The difference in dBs between two power quan-
tities is defined as ten times the logarithm to base 10 for the ratio of the two
power quantities.
In mathemetical notation:
P2
10 ∗ log( ) (A.1)
P1
Note the factor of 10 in the definition which explains the term ”deci” in
”decibel”, so a ”decibel” is one-tenth of a ”bel” which is in turn a rarely-used
unit.
If the power quantity P2 is twice as P1 , the difference in dBs would be
P2
10 ∗ log( ) = 10 ∗ log(2) = 3dB (A.2)
P1
If P2 is the same as P1 , the difference in dBs would be
10 ∗ log(1) = 0dB (A.3)
113
114 APPENDIX A. DECIBEL
If P2 is 10 times larger than P1 , the difference in dBs would be
10 ∗ log(10) = 10dB (A.4)
Besides power, dBs can also describe current and voltage quantities.
Since P ∝ I 2 , V 2 , then:
P2 I2 2 I2
dB f or current = 10 ∗ log( ) = 10 ∗ log( 2 ) = 20 ∗ log( ) (A.5)
P1 I1 I1
P2 V2 2 V2
dB f or voltage = 10 ∗ log( ) = 10 ∗ log( 2 ) = 20 ∗ log( ) (A.6)
P1 V1 V1
This can be seen for different values of power and current/voltage ratios as in
table A.1 below, as well for different values of power ratios as shown in figure A.1.
115
dB Power ratio Current/Voltage ratio

100 10 000 000 000 100 000
90 1 000 000 000 31 620
80 100 000 000 10 000
70 10 000 000 3 162
60 1 000 000 1 000
50 100 000 316.2
40 10 000 100
30 1 000 31.62
20 100 10
10 10 3 .162
0 1 1
-10 0 .1 0 .316 2
-20 0 .01 0 .1
-30 0 .001 0 .031 62
-40 0 .000 1 0 .01
-50 0 .000 01 0 .003 162
-60 0 .000 001 0 .001
-70 0 .000 000 1 0 .000 316 2
-80 0 .000 000 01 0 .000 1
-90 0 .000 000 001 0 .000 031 62
-100 0 .000 000 000 1 0 .000 01
Table A.1: Power ratios and voltage/current amplitude ratios and their corre-
sponding dBs
116 APPENDIX A. DECIBEL
Difference in Power Levels in dB vs Power Ratios

15
10
5
dB
−5
−10
0 5 10 15 20
Power ratio (P2 / P1)
Figure A.1: Difference in power levels (dB) versus power ratios
dBs may also describe absolute quantities above or below a certain refer-
ence, which makes it necessary to know the reference value used. Examples for
commonly used absolute quantities and their corresponding reference values are
shown in table A.2 below
Unit Symbol Unit Type Reference Value

dB Power 1W
dBm Power 1 mW
dBV Voltage 1V
dBmV Voltage 1 mV
dBμV Voltage 1 μV
Table A.2: dB units describing absolute quantities and their corresponding

reference values
The advantage of using dBs can be summarized in the following points:

- dB plots are best in describing very large windows which yet contain small
variations. As one can see from table A.1 above, a 1 volt signal and a 10
117
microvolt signal can be both represented on a plot with a range of 100 dBs.[34]
- dBs can represent very large or very small ratios in moderate size numbers.
It is easier for the mind to compare 2-3 digit numbers than comparing up to
10 digit numbers.
- Gain and loss calculation is simpler with dBs; Power ratios expressed in dBs
are simply additive. For example, if we have a system with a gain of 13 dBs
(corresponding to a power ratio of 20 times) followed by another with a loss of
3 dBs (power ratio of 12 times), the overall gain/loss would be simply 13+(-3)
= 10 dBs which corresponds to a gain with a power ratio of 10.

LTH Iea 1065

Uploaded by

Copyright:

Available Formats

LTH Iea 1065

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

LTH Iea 1065

Uploaded by

Copyright:

Available Formats

EMC Analysis of Electric Drives

Fadi Abdallah, 2012

3 Capacitive Coupling Matrix 23

4 Inductive Coupling Matrix 41

4.2 Complex-Numbered Inductance Concept . . . . . . . . . . . . . . 54

6 Conclusions and Future Work 101

1.1 Parasitic capacitances in an electric drive system . . . . . . . . . 3

2.1 A photo of induction machine stator showing Integral Slot, Single

3.1 The 24 conductors arrangement for electrostatic FEMM analysis

4.1 Penetration/Skin depth for diﬀerent types of materials along a

4.5 Zoomed-in post-processor views of magneto-static and AC time-

4.18 A logarithmic scale of absolute value of the complex-numbered

5.1 Measurements of impedance magnitude for the winding-on-core

A.1 Diﬀerence in power levels (dB) versus power ratios . . . . . . . . 116

2.1 Measurements access points in the 3-layer 24-winding-on-core

4.1 Key frequencies versus their corresponding wavelength (λ) in free

Introducing electric propulsion in a large scale to the automotive industry

1.2 Problem Formulation

Conducted and radiated ElectroMagnetic Interference (EMI) emissions is the

to ground, each junction module of switching elements to ground, and ending

19 Lead Acid Batteries

Power electronics rack Motor placed on metalic sheet

Figure 1.1: Parasitic capacitances in an electric drive system

Parasitic capacitances are usually of low orders of magnitude which make

Figure 1.2: A photo of the second inhouse-built electric drive labsetup

at every instant a change in the varying 2-step/3-step CM voltage waveform

1.3 Scope and Goal

Triangle and reference voltage signals + PWM signals (u,v,w)

Figure 1.4: The resulting high-frequency damped-sinusoid CM current waveform

The study aims at investigating capacitive and inductive coupling of a ma-

1.4 Study Roadmap and Thesis Structure

built from 19 lead-acid battery cells, each rated at 12V, 20 Ah corresponding

1.5 Contribution of This Study

Oscilloscope (DSO). Then, these measurements can be conﬁgured as a voltage

(i.) It saves simulation time substantially, especially when incorporating more

(ii.) It gives the freedom of measuring high-frequency leakage currents without

2.1 Slot Analysis

Figure 2.2: A side-view drawing of induction machine stator winding showing

2.2 “Macro Level” and “Micro Level” Machine

The Pole Pitch τp = π2 ∗ r; where r = 65.5 mm is the radius from the

where ◦ is the permitivity constant of free space, r is the relative per-

C_slot_left1 Lcoil C_slot_right1

outer_most_coil outer_most_coil C_slot_left2 Lcoil C_slot_right2

C_slot_left15 Lcoil10 C_slot_right15 C_slot_left9 Lcoil4 C_slot_right9

C_slot_left16 Lcoil11 C_slot_right16 C_slot_left10 Lcoil5 C_slot_right10

Figure 2.3: Macro(slot)-level circuit model of induction machine with Single

2.3 Layer-to-Layer Approach for Slot Analysis

Considering a winding in a slot model on a layer-to-layer basis, the frequency-

{c_ww} {c_ww} {c_ww}

SINE(0 1 50) C14 C16

Figure 2.5: Layer-to-layer approach for sensitivity analysis in frequency domain

2.4 Turn-to-Turn Approach for Slot Analysis

The turn-to-turn analysis level is adopted in modelling the winding portion in

(i.) Resistance of each winding turn (Rt ).

2.5 Simple Winding-on-Core Prototype

simpliﬁcation is proposed by one of the reference group members [31], to form

Access Points Location

Table 2.1: Measurements access points in the 3-layer 24-winding-on-core proto-