i

An FRA Transformer Model with Application

on Time Domain Reflectometry

Hanif Tavakoli

Doctoral Thesis in Electromagnetic Engineering

Stockholm, Sweden 2011

ii

Royal Institute of Technology (KTH)

School of Electrical Engineering
Division of Electromagnetic Engineering
Teknikringen 33
SE 100 44 Stockholm, Sweden

TRITA-EE 2011:071
ISSN 1653-5146
ISBN 978-91-7501-134-9

Akademisk avhandling som med tillstnd av Kungliga Tekniska Hgskolan framlggs till
offentlig granskning fr avlggande av teknologie doktorsexamen torsdagen den 15 december
2011 klockan 10.00 i sal F3, Lindstedtsvgen 26, Kungliga Tekniska Hgskolan, Stockholm.
Hanif Tavakoli, oktober 2011
Tryck: Universitetsservice US AB

iii

To my mother

iv

Abstract
Frequency response analysis (FRA) is a frequency-domain method which is used to detect
mechanical faults in transformers. The frequency response of a transformer is determined by
its geometry and material properties, and it can be considered as the transformers fingerprint.
If there are any mechanical changes in the transformer, for example if the windings are moved
or distorted, its fingerprint will also be changed so, theoretically, mechanical changes in the
transformer can be detected with FRA. A problem with FRA is the fact that there is no
general agreement about how to interpret the measurement results for detection of winding
damages. For instance, the interpretation of measurement results has still not been
standardized.
The overall goal of this thesis is to try to enhance the understanding of the information
contained in FRA measurements. This has been done in two ways: (1) by examining the FRA
method for (much) higher frequencies than what is usual, and (2) by developing a new
method in which FRA is combined with the ideas of Time Domain Reflectometry (TDR). As
tools for carrying out the above mentioned steps, models for the magnetic core and the
winding have been developed and verified by comparison to measurements.
The usual upper frequency limit for FRA is around 2 MHz, which in this thesis has been
extended by an order of magnitude in order to detect and interpret new phenomena that
emerge at high frequencies and to investigate the potential of this high-frequency region for
detection of winding deteriorations.
Further, in the above-mentioned new method developed in this thesis, FRA and TDR are
combined as a step towards an easier and more intuitive detection and localization of faults in
transformer windings, where frequency response measurements are visualized in the time
domain in order to facilitate their interpretation.
Index terms: complex permeability, lumped circuit model, frequency response analysis, time
domain reflectometry, high frequency modeling, transformer diagnosis, reluctance network
method.

vi

vii

Acknowledgements
This doctoral thesis is based on results within the research group of Electrotechnical
Modeling, at the Department of Electromagnetic Engineering, School of Electrical
Engineering, Royal Institute of Technology (KTH). The project which resulted in this thesis
was funded by the Swedish Center of Excellence in Electric Power Engineering (EKC2). The
financial support is gratefully acknowledged.
I would like to thank my supervisor Professor Gran Engdahl for his guidance during this
project, interesting and useful discussions about transformers and transformer modeling.
Also, I would like to thank Dr. Dierk Bormann for his ideas and support, his fruitful
collaborations with me, and his reviewing and correcting of my papers and models during this
project. Without his help, most of the achievements in this project would not have been
possible.
I also thank Professor Roland Eriksson and Professor Rajeev Thottappillil, former and
present head of the department, respectively, for trusting me enough to employ me for this
PhD project.
Furthermore, I acknowledge Associate Professor Martin Norgren for kvalitetsgranskning
(quality review), and Peter Lnn for technical support with computer hardware and software.
Also thanks to Carin Norberg for administration support.
I also thank Dr. David Ribbenfjrd, Johanna Rosenlind and Assistant Professor Patrik
Hilber for our cooperation, and the rest of the people at the Electromagnetic Engineering Lab
for friendship, discussions, lunches and refreshing coffee breaks.
And last but not least I would like to thank my mother for her endless love to me and for
her patience during lifes trials and tribulations.
Hanif Tavakoli
Stockholm, Sweden, October 2011

viii

ix

List of Publications

I. H. Tavakoli, D. Bormann, G. Engdahl, D. Ribbenfjrd, Comparison of a Simple

and a Detailed Model of Magnetic Hysteresis with Measurements on Electrical
Steel, COMPEL: The International Journal for Computation and Mathematics in
Electrical and Electronic Engineering, Vol. 28 No. 3, pp. 700710, 2009.
II. H. Tavakoli, D. Bormann, G. Engdahl, High Frequency Oscillation Modes in a
Transformer Winding Disc, Studies in Applied Electromagnetics and Mechanics,
Vol. 34, pp. 329337.
III. J. Rosenlind, H. Tavakoli, P. Hilber, Frequency Response Analysis (FRA) in the
Service of Reliability Analysis of Power Transformer, Proceedings of International
Conference on Condition Monitoring, Diagnosis and Maintenance - CMDM 2011.
IV. D. Bormann, H. Tavakoli, Reluctance Network Treatment of Skin and Proximity
Effects in Multi-Conductor Transmission Lines, accepted for publication in IEEE
Transactions on Magnetics.
V. H. Tavakoli, D. Bormann, G. Engdahl, Time Domain Reflectometry for Fault
Localization in Transformer Windings, submitted to International Journal of
Applied Electromagnetics and Mechanics.
VI. D. Bormann, H. Tavakoli, Reluctance Network Theory of Skin and Proximity
Effects in Strongly Coupled Multi-Conductor Transmission Lines, submitted to
IEEE Transactions on Power Delivery.
VII. H. Tavakoli, D. Bormann, G. Engdahl, Fault Localization in Transformer Windings
Using Time-Domain Representation of Response Functions, submitted to European
Transactions on Electrical Power.

xi

Contents
ABSTRACT
ACKNOWLEDGMENTS
LIST OF PUBLICATIONS
1

INTRODUCTION ....................................................................................................................................... 1
1.1
1.2

BACKGROUND AND AIM ....................................................................................................................... 1

OUTLINE OF THE THESIS ....................................................................................................................... 2

FREQUENCY RESPONSE ANALYSIS (FRA) OF POWER TRANSFORMERS .............................. 3

2.1
FREQUENCY RESPONSE MEASUREMENTS ............................................................................................. 3
2.1.1
Impulse Response Method............................................................................................................... 3
2.1.2
Frequency Sweep Method ............................................................................................................... 4
2.2
MECHANICAL FAULTS IN A TRANSFORMER .......................................................................................... 4
2.3
DIAGNOSING MECHANICAL FAULTS IN A TRANSFORMER WITH THE HELP OF FRA .............................. 4

CORE MATERIAL MODEL AND MEASUREMENTS ........................................................................ 7

3.1
3.2
3.3
3.3.1
3.3.2
3.4
3.5

HIGH FREQUENCY WINDING MODEL ............................................................................................ 17

4.1
4.2
4.3
4.4
4.5
4.6

IMPEDANCE MEASUREMENT DEVICE AND DIMENSIONING OF THE EXPERIMENTAL SETUP ................. 27

IMPEDANCE MEASUREMENT RESULTS ................................................................................................ 28

MODEL VERIFICATION ....................................................................................................................... 31

6.1

TRANSFORMER WINDING MODEL FOR HIGH FREQUENCY APPLICATIONS .......................................... 17

THREE DIFFERENT RESOLUTIONS OF THE MODEL............................................................................... 17
CALCULATION OF THE CAPACITANCES ............................................................................................... 19
CALCULATION OF THE INDUCTANCES AND RESISTANCES ................................................................... 20
STATE SPACE MODEL FOR THE WINDING ........................................................................................... 21
COMPARISON BETWEEN THE THREE MODELS ..................................................................................... 25

FREQUENCY RESPONSE MEASUREMENTS................................................................................... 27

5.1
5.2

BACKGROUND ...................................................................................................................................... 7
COMPLEX-PERMEABILITY MODEL ........................................................................................................ 8
DETAILED HYSTERESIS MODEL ............................................................................................................ 9
Static Hysteresis .............................................................................................................................. 9
Excess Losses ................................................................................................................................ 11
MEASUREMENTS AND MODEL ADJUSTMENTS .................................................................................... 11
COMPARISON BETWEEN MODEL AND MEASUREMENTS ...................................................................... 13

COMPARISON OF MODEL WITH MEASUREMENTS ................................................................................ 31

INTERPRETATION................................................................................................................................. 35
7.1
EXPLANATION OF THE DIFFERENT OSCILLATION MODES ................................................................... 35
7.1.1
Radial Resonance Modes .............................................................................................................. 35
7.1.2
Azimuthal Resonance Modes......................................................................................................... 37
7.2
RADIAL AND AZIMUTHAL RESONANCES FOR A WINDING WITH MORE THAN ONE DISC..................... 38

xii
8
RELUCTANCE NETWORK METHOD FOR INCLUSION OF THE SKIN AND PROXIMITY
EFFECTS............................................................................................................................................................. 41
8.1
SKIN AND PROXIMITY EFFECTS .......................................................................................................... 41
8.2
USING RELUCTANCE NETWORK METHOD (RNM) TO CALCULATE THE FREQUENCY DEPENDENT
INDUCTANCE AND RESISTANCE MATRICES ....................................................................................................... 42
8.3
RNM FOR HIGH FREQUENCIES ........................................................................................................... 43
8.3.1
Principle of the Reluctance Network Description ......................................................................... 44
8.3.2
Calculation of the Reluctances for High Frequencies................................................................... 46
8.4
EXTENSION OF RNM TO LOW FREQUENCIES ...................................................................................... 49
8.4.1
Summary of the Generalized Approach......................................................................................... 49
8.4.2
Exact Solution of a Single-Slab Problem ...................................................................................... 51
8.5
EXTENSION TO VARIABLE GAP WIDTH ............................................................................................... 54
8.6
COMPARISON WITH FEM CALCULATIONS .......................................................................................... 56
8.6.1
Results for Rectangular Conductors ............................................................................................. 57
8.6.2
Results for Round Conductors....................................................................................................... 60
8.7
CONCLUSIONS ABOUT RNM ............................................................................................................... 62
9

TIME DOMAIN REFLECTOMETRY (TDR)....................................................................................... 63

9.1
9.2
9.3
9.4
9.5
9.6
9.7

10

COMBINATION OF FRA AND TDR ...................................................................................................... 63

BASIC IDEAS BEHIND THE TDR TECHNIQUE ....................................................................................... 63
SIMPLIFIED MODEL OF A TRANSFORMER WINDING AND THE EFFECT OF DISPERSION ........................ 64
THE TRANSFORMER WINDING MODEL USED FOR TDR SIMULATIONS ............................................... 66
THE EFFECT OF CHOICE OF RISE-TIME FOR THE APPLIED VOLTAGE ................................................... 66
SIMULATIONS WITH AND WITHOUT WINDING DAMAGES .................................................................... 68
CONVERSION OF IN-IMPEDANCE MEASUREMENTS TO DTDR SIGNALS .............................................. 75

SUMMARY, CONCLUSIONS AND FUTURE WORK........................................................................ 77

REFERENCES.................................................................................................................................................... 79
APPENDIX A ...................................................................................................................................................... 85
APPENDIX B ...................................................................................................................................................... 89
LIST OF SYMBOLS .......................................................................................................................................... 95
LIST OF ACRONYMS .................................................................................................................................... 103

1 Introduction
In this Chapter, a short background information about power transformers is given, and the
aim, outline and structure of the thesis are presented.

1.1 Background and Aim

A power transformer is an electric device employed in electric power systems for
transmission and distribution of electric power, and it is used when there is a need for a
voltage transformation. In a transformer, electric energy is transferred between different
electrical circuits by the use of electromagnetic induction.
Power transformers are very large and expensive so preventive replacement of old ones
with new ones, in order to increase the system reliability, is often not economically justified;
therefore, they are supposed to be used for maximum number of years, and they are also
supposed to be in continuous operation for reliable and uninterrupted power supply. These
goals can only be achieved through proper maintenance and condition monitoring, so that
deteriorations in the transformer, which may lead to a failure, can be detected at an early
stage.
A number of different methods are routinely used for condition monitoring and diagnostics
of transformers, for instance oil analysis, winding resistance measurements, winding transfer
ratio measurements, dielectric response measurements (DFR) and frequency response analysis
(FRA). This thesis concentrates on the method of FRA, which is one of the more recent and
advanced diagnostic methods. A fundamental problem with FRA is the fact that despite the
advancements and improvements of this method since its invention in 1978, there is still no
general agreement about how to draw reliable conclusions from it regarding the transformer
health. In particular, the interpretation of FRA measurement results has still not been
standardized.
So the goal of this thesis is to try to improve the understanding of the information
contained in FRA measurements. This has been done in two ways: (1) by considering the
FRA method for (much) higher frequencies than what is standard, and (2) by developing a
new method in which FRA and Time Domain Reflectometry (TDR) are combined.
The usual upper frequency limit for FRA measurements is around 2 MHz, and this limit
has been increased here by a factor of ten in order to investigate new high frequency
phenomena and to examine whether this high frequency region has the potential to facilitate
the detection of winding deformations. It was found that, for the disc geometry employed
here, two classes of internal resonance modes exist for higher frequencies, so-called radial
and azimuthal oscillation modes, and measurements show that these modes are very
sensitive to small changes in the disc geometry.
Time Domain Reflectometry (TDR) is a method traditionally used for detection and
localization of changes in transmission lines. Time-domain diagnostic methods have scarcely
been applied to power transformers at all in the literature since the late 1970s, after the lowvoltage impulse method [1] had declined in popularity. In the new time domain method
described in this thesis, principles of FRA and TDR have been combined to create an easy and
intuitive approach for detection and localization of winding faults, where frequency response

2
measurements are visualized in the time domain in order to facilitate their interpretation.
Thereby mechanical changes in transformer windings can be identified and located in a
simple manner.
As tools for carrying out the steps mentioned above, models for the magnetic core and the
winding have been developed and verified by comparison to measurements.
It may be mentioned that during the work on this thesis an attempt has been made also to
combine the method of FRA with the formal reliability analysis of fleets of power
transformers or individual units (published as paper no. III cited in the List of Publications on
page ix above). However, this work is only loosely connected to the main body of this thesis
and therefore not included here.

1.2 Outline of the Thesis

The thesis is structured as follows:
Chapter 2 presents a general overview of FRA and measurement techniques, mechanical
faults in transformers and the way these faults can be detected by FRA.
In Chapter 3, which deals with the core material model, the developed simple complex
permeability model, a detailed hysteresis model and the measurements are presented and
compared to each other. The complex permeability model has been developed by the author
with the support of Dr Bormann, the detailed hysteresis model has been developed by Dr
Ribbenfjrd, and the measurements have been performed by the author.
Chapter 4 presents the developed lumped element winding models, the formulas for the
lumped element parameters, and the state space equation which is used for calculating the
currents and voltages in the model. This Chapter is concluded by a comparison between the
impedance of three models with different levels of discretization in the frequency spectrum.
The winding models have been developed by the author with the support of Dr Bormann.
Chapter 5 presents the winding measurement set-up and the performed measurements
(carried out by the author supported by Dr Bormann), and in Chapter 6, the winding models
are verified by comparison with measurements.
Chapter 7 deals with the explanation and interpretation of the different oscillation modes of
a single disc (accomplished by the author in cooperation with Dr Bormann), and is concluded
by analyzing the state of the modes when a winding consists of more than one disc.
In Chapter 8, for calculation of frequency dependent inductances and resistances, a
reluctance network method, which includes the impact of the skin and proximity effects, is
deduced and compared to finite element calculations. The theory behind the reluctance
network method has to a considerable extent been developed by Dr Bormann with the support
of the author, and the implementation of it in MATLAB and creation of finite element models
for the purpose of comparison have been carried out by the author.
In Chapter 9, the time domain method TDR, its benefits, and the way it can be combined
with FRA for diagnosis of transformer winding faults are presented. Also, the developed
winding models are used in the time domain to simulate the detection, localization and
identification of some different kinds of winding faults. This has been performed by the
author, supported by Dr Bormann.
And finally, Chapter 10 contains a summary, conclusions and some suggestions for future
work.

2 Frequency Response Analysis (FRA) of Power

Transformers
In this Chapter, the method of FRA and the measurement techniques are explained. Further,
the different mechanical faults in transformers and the way they can be detected by FRA are
reviewed.

2.1 Frequency Response Measurements

Frequency Response Analysis (FRA) is a powerful method for characterizing a system by
analyzing its frequency response, which is uniquely defined by the system parameters; this
means that FRA can be used to either design a system or to analyze an existing one. It is the
phase and magnitude response of a system when subjected to sinusoidal inputs, and it has
become a popular method for evaluating the mechanical condition of the windings and the
clamping structure of power transformers [216].
The first technical work describing the possibility of using the FRA technique for
diagnosing mechanical faults inside power transformers was published by Dick and Erven
[17] in 1978. Since then, FRA has been gaining popularity among researchers and utilities as
a potential method to detect mechanical changes inside power transformers.
The frequency range for FRA is generally from 10 Hz up to 2 MHz and the evaluation is
based on the fact that the frequency response of a transformer is defined by its capacitance
and inductance distributions, which are determined by the geometrical construction of the
transformer and characteristics of materials used. Therefore, mechanical deformations change
the capacitive and inductive parameters, yielding deviations in the FRA spectrum. This means
that FRA is basically a comparative method, in which a fingerprint measurement taken at an
earlier stage is compared with a measurement taken at a later stage, perhaps after relocation or
during a maintenance operation. Then the changes in characteristics of the response are
analyzed to detect mechanical changes inside the transformer.
Frequency response can either be measured directly by sweeping the frequency (sweep
frequency method) or be estimated from impulse response measurements. Both methods have
advantages and disadvantages. For example, the impulse response method needs less
measuring time, but it is very noise sensitive. On the other hand, the frequency sweep method
takes a little longer time for the measurements, but it is not so noise sensitive.
2.1.1

Impulse Response Method

In the impulse response method, an impulse voltage that has adequate frequency content is
applied to the test object and both the applied voltage and some resulting response voltage or
current are simultaneously measured. This method is based on the definition of the transfer
function which says that the transfer function doesnt depend on the applied signal when the
system is linear and time invariant. Then both of the measured signals are numerically
transformed into the frequency domain using Fast Fourier Transform (FFT). The ratio

4
between the FFT of the response signal and the applied signal is the frequency response of the
corresponding transfer function.
This method has been used by many researchers for diagnosing mechanical faults in power
transformers [5]. Limitation of the excitation source that can produce enough energy in the
whole frequency band of interest, reduced energy level of injected impulse at higher
frequencies limiting the upper limit of the calculated frequency response, and the need for
good noise prevention techniques are some of the disadvantages of the impulse response
method.
2.1.2

Frequency Sweep Method

In this method, a sinusoidal exciting voltage is applied and the magnitude and phase of the
response voltage or current are measured at different frequencies. This means that this is a
direct method for determining the frequency response, since the final result is already
available after sweeping the predefined frequency range. This is the method which has been
used in this thesis.

2.2 Mechanical Faults in a Transformer

A transformer can be damaged due to a variety of reasons. Some of these reasons are:
insulation degradation, transportation, installation, and the forces inside it due to the
interaction of the current and the (leakage) magnetic flux density according to:

F IB

(1)

where F is the force, I is the current in the winding and B is the magnetic flux density.
According to Eq. (1) there will be heavy mechanical stresses in the transformer in case of a
sudden short circuit fault, as the current flowing through the winding at that time is enormous.
Eq. (1) also means that there will be two kinds of force vectors generated by the axial
component of the leakage flux density (radial force) and by the radial component of the
leakage flux density (axial force). The radial forces tend to squeeze the inner winding and
expand the outer winding, resulting in circlet buckling in the former and circlet deformation in
the latter due to imbalance of the radial pressure. In contrast, the axial forces tend to displace
the windings axially in relation to each other, perpetrating pressure on the clamping structure.
It may bend conductors between rigid axial spacers, and during winding movements the
insulation between the turns could be abraded, which can lead to short circuiting and
damaging of the windings in the same layer, the same disc, different layers or different
windings. Short circuit faults can cause great harms, because if the clamping pressure is not
capable to counteract the involved forces, significant winding deformation or even break
down of windings can happen almost immediately, often convoyed with shorted turns.

2.3 Diagnosing Mechanical Faults in a Transformer with the Help of FRA

It is generally said that FRA has the capability of identifying faults of the types: core
movement, winding deformation, winding movement, broken or loose winding or clamping
structure, partial collapse of the winding and short-circuited turns or open circuit windings.
Interpretation of FRA results, when searching for mechanical changes inside a transformer,
has neither been standardized nor fully agreed among researchers yet. Therefore, one may

5
find several different types of transfer functions (in-impedance, transfer-impedance, voltage
transfer ratio, etc.), different measurement techniques and different ways of interpreting FRA
results. But as mentioned before, FRA is essentially a comparative method and therefore, a
fingerprint response measurement of the same transformer which is going to be diagnosed or
a sister unit should be available for comparison with the present measurement. When the
measurement is compared with the reference set, then the changes in the frequency response
which could be identified as mechanical faults are as follows:

Abnormal shifts in the existing resonances

Emergence of new resonances or evaporation of the existing ones
Considerable changes in the overall shape of the frequency response.

By comparison of the new and the old measurements, or by comparison of the measurement
results from different phases of the same transformer (since normally not all phases are
affected in the same way by the fault [18]), an expert can identify possible faults.

3 Core Material Model and Measurements

In this Chapter, which is largely based on [19], a simple core material model is developed. For
efficient magnetic field calculations in electrical machines and transformers, the hysteresis
and eddy current losses in laminated electrical steel must be modelled in a simple and reliable
way. Therefore, in this chapter, a frequency dependent complex permeability model and a
more detailed model (describing hysteresis, classical eddy current effects, and excess losses
separately) are compared with single sheet measurements. It is discussed under which
circumstances the simple complex- model is an adequate substitute for the more detailed
model.

3.1 Background
Recent research has resulted in detailed models of the magnetic hysteresis and loss
mechanisms in a wide frequency range [2022]. Although these models provide a good
description of magnetic material properties or of simple reluctance circuits based on them,
they are too demanding numerically to be incorporated into a full-scale magnetic field
simulation of a realistic geometry, as with a FEM or FDM calculation tool. In other words,
while such a detailed simulation of the H-B relation of a single or a few interacting cells is
still perfectly feasible, simulating thousands or ten thousands of them simultaneously may be
inconvenient or impossible.
Moreover, in many practical situations a detailed description is not required either (as is
the case in this thesis). Often the goal is to obtain a good estimate of some local or global
quantity containing much less information than the detailed local H-B relation, such as the
local losses causing dangerous hot spots, or simply the total losses in a machine relevant for
cooling or economic reasons. For such applications it is desirable to use a simple model of
magnetic hysteresis and losses, which can easily be incorporated in field calculation tools but
which at the same time is sufficiently close to reality, within the frequency range of interest
for the specific application. Such a model is the description of laminated magnetic materials
by a suitable frequency dependent complex permeability, which is the most general linear
description of a local and isotropic H-B relation. If desired, it can easily be extended to a
nonlocal and/or anisotropic H-B relation by turning from a scalar function of position into a
distance dependent integral kernel and/or tensor, respectively [23].
In this thesis, it is discussed to which extent results from measurements on strips of
electrical steel, obtained with a single-sheet tester, can reliably be described by a simple
complex permeability function of frequency. Both the resulting H-B curves and the effective
complex permeability are compared to the measured data at different frequencies. For
comparison, simulation results obtained with a much more detailed model of the magnetic
hysteresis, eddy current and excess losses are also reported.

3.2 Complex-Permeability Model

Reduced to its simplest terms, hysteresis introduces a time phase difference between B and H.
B is assumed to lag H by a constant angle h called the hysteresis angle. In such a description,
harmonics introduced by saturation are ignored, and the hysteresis loop becomes an ellipse
whose major axis is inclined by an angle h relative to the H-axis. Using complex field
components B and H , a low-frequency complex permeability including hysteresis can be
defined as
B
h 0 r e jh .
(2)
H
In addition to this, eddy currents in the lamination sheets introduce frequency dependence.
The well-known procedure [24] for deriving the effective frequency dependent complex
permeability is briefly sketched below. Faradays law

B
t

and Amperes law

H J

(3)

D
,
t

(4)

in combination with the constitutive relations

J E , D E , B h H ,

(5)

and time-harmonic assumption lead to

2 H 2 H j h j H .

(6)

For lower frequencies when wave propagation can be ignored (i.e., ), one
has 2 j h .
x

2b

H z ( x)

Fig. 1: Laminate infinite in z direction, with a width in y direction much larger than its thickness 2b, exposed to
a H field in z direction.

For analysis of the magnetic field in a laminate, the simple geometry illustrated in Fig. 1 is
appropriate. The magnetic field is applied in the z direction, hence the only component of the
magnetic field strength is Hz which varies only in the x direction, Hz = Hz (x). In one
dimension, Eq. (6) reduces to

9
2 H z
2 H z ,
2
x

(7)

H z ( x) A1e x A2 e x .

(8)

which has the general solution

The field strength on the both sides of the laminate is assumed to be H0. For the reason of
symmetry the following condition is obtained
H z (b) H z (b) H 0 .

(9)

The final expression for the magnetic field strength then becomes

H z ( x) H 0

cosh x
.
cosh b

(10)

The effective, complex permeability of a lamination is given as the average magnetic flux
density B in the laminate normalized to the surface magnetic field strength H0 ,
jeff

eff eff

B
1
tanh( b)

.
H
x
x
(
)d
h
h
z
b
H 0 H 0 2b b
b

(11)

This expression accounts for the effect of hysteresis without saturation, and the effect of eddy
currents. It is assumed here that additional (or excess) losses are either negligible or have a
similar frequency dependence so that they can be incorporated in the expression (11) for eff .

3.3 Detailed Hysteresis Model

Later in this Chapter, some results obtained with a more detailed model of the magnetic
hysteresis, eddy current and excess losses will be reported, so therefore, this more detailed
model, which has been developed by Dr. David Ribbenfjrd in [21], is described here in short
(for a thorough description and explanation, see [21]).
The total hysteresis is a combination of three different phenomena, namely, static
hysteresis, eddy current effects and excess eddy currents. For the detailed hysteresis model,
the following approach has been used. The static hysteresis is modeled using Bergqvists lag
model [25, 26], the classical eddy currents are modeled using Cauer circuits [22, 27], and the
excess losses are modeled using an approach by Bertotti [20].

3.3.1 Static Hysteresis

The Bergqvists lag model of static hysteresis starts from the idea that the magnetic material
consists of a finite number of pseudo particles np, i.e., volume fractions with different
magnetization. The total magnetization is then a weighted sum of the individual
magnetization of all pseudo particles.

10

H
2k

Fig. 2: An-hysteretic curve (left), play operator (middle), and resulting hysteresis curve (right); figure taken
from [21].

The hysteresis curve for one particle is introduced by applying a play operator with a play
equal to the pinning strength k (which will determine the width of the hysteresis curve) on
the an-hysteretic curve, see Fig. 2, where m is the magnetization of the actual pseudo particle,
and is the back field i.e. the field that will give the magnetization m if no hysteresis is
present.
Using a population of pseudo particles with different pinning strengths allows constructing
minor loops. An individual pinning strength ik is assigned to every pseudo particle, where k
is the mean pinning strength, and i is a dimensionless number for particle i. The total
magnetization is then given by a weighted superposition of the contributions from all pseudo
particles (Fig. 3).

H
21k

22k

23k

Fig. 3: Weighted superposition of the contributions from pseudo particles describes a minor loop; figure taken
from [21].

The expression
M an ( H )

H
M s arctan

2M s
2

(12)

is used for the an-hysteretic magnetization, where Ms is the magnetization saturation and is
the susceptibility at H = 0. For infinite number of pseudo particles, the total magnetization of
the material is then given by

M cM an ( H ) M an P k ( H ) ( )d ,

(13)

where c is a constant that governs the degree of reversibility, and the integral describes the
hysteretic behaviour (irreversible part). Pk is a play-operator with the pinning strength k, and
() is a density function describing the distribution of the pseudo particles. Finally, the
magnetic flux density is obtained from B = 0(H+M).

11

3.3.2 Excess Losses

Excess losses are caused by microscopic eddy currents induced by local changes in flux
density due to domain wall movements. For the detailed model, an approach described by
Bertotti [20] is used. In this approach, a number of active correlation regions are assumed
randomly distributed in the material. The correlation regions are connected to the microstructure of the material like grain size, crystallographic textures and residual stresses. In
Bertottis model, the resulting contribution to the magnetic field strength is given by
H excess

n0V0
4 G 2bw dB
dB
1

1
sign
,
2
2
dt
n0V0
dt

(14)

where w is the width of the laminate and 2b, as before, its thickness. G is a parameter
depending on the structure of the magnetic domains. n0 is a phenomenological parameter
related to the number of active correlation regions when the frequency approaches zero,
whereas V0 determines to which extent micro-structural features affect the number of active
correlation regions.
The parameters n0 and V0 are by definition frequency independent, but they are expected in
reality to depend on the amplitude of the B field [28]. Since the precise form of this
dependence is unknown, their values are usually adjusted empirically for a given amplitude.
In the simulations reported here one set of (empirically determined) values is used, although
the amplitude of the B field varies slightly in the measurements.

3.4 Measurements and Model Adjustments

The magnetic measurements were carried out using a Single Sheet Tester. It consists of two
equal U-shaped yokes placed face-to-face to each other (Fig. 4). The magnetic sheet to be
tested is placed between the yokes and most of the flux is forced through it due to its high
permeability. For the measurement of the flux in the test material, a coil is surrounding the
strip which is connected to a flux meter. The magnetic field strength is measured with a Hall
probe placed close to the surface of the sample and connected to a Tesla meter. A sinusoidal
H field was applied to the sample; the H and B field values were measured for 100 periods
and numerically filtered. Thereafter, the mean values at different phase angles of the B and H
fields were calculated and used for the experimental verification of the complex- model.

Fig. 4: Cross sectional view of the Single Sheet Tester.

12

The measured (mean-value) H-B curve is then approximated with a complex- ellipse
characterized by the permeability meas . This is done by matching both the peak values Hp , Bp
and the area A of the ellipse to the measured results (i.e. the measured H-B curve and the
complex- ellipse will have the same area and peak values). This is of course appropriate as
long as the shape of the measured H-B curve is close to an ellipse, i.e., if saturation effects are
not too pronounced. The power loss per cycle, the area A within the H-B curve, is given by the
integral
T
dH meas
(15)
A Bmeas dH meas Bmeas
dt ,
dt
0
where Bmeas and Hmeas are time dependent fields (correlated trough meas ) defined as
approximations of the time dependent measured B and H fields, respectively, and T is the
duration of a period. If the measured H field is assumed to vary sinusoidally, then one can use
the approximation
H meas (t ) Re H p e jt H p cos(t ) ,
(16)

with the derivative

dH meas (t )
H p sin(t ) .
dt

(17)

The approximation of the measured B field becomes then

Bmeas (t ) Re meas H pe jt

jmeas
H p e jt
Re meas

cos(t ) meas
sin(t ) .
H p meas

(18)

By inserting Eq. (18) and (17) into Eq. (15) one gets

meas

A
.
H p2

(19)

Furthermore, from the relation meas H p Bp one obtains

meas

meas meas
2

Bp

H
p

(20)

which implies

meas

Bp

H
p

and meas
are functions of frequency.
Both meas

2
.
meas

(21)

13

1.5

1
0.8

0.6
0.4

B [Tesla]
B

B
B [Tesla]

0.5

0.2
0
-0.2

-0.5

-0.4
-0.6

-1

-0.8
-1.5
-400

-300

-200

-100

100

200

300

400

-1
-400

-300

-200

-100

HH[A/m]

0
H

100

200

300

400

H [A/m]
(a)
(b)
Fig. 5: H-B curves from measurements (blue) and complex- model (green) with meas meas
jmeas
, for
(a) f = 50 Hz and (b) f = 400 Hz.

Fig. 5 compares the measured H-B curves with complex- ellipses, generated with the
adapted meas at frequencies f = 50 Hz and 400 Hz.
eff as defined in Eq. (11) is a function of frequency and of a vector x = (r , h , b2)
containing the model parameters. It is adjusted to measured data by numerically minimizing
the expression
N

i 1

eff

( x, i ) meas (i )

(22)

with respect to x. meas (i ) are the measured complex permeability values, defined by
Eq. (19) and (21), at N different frequencies i = 2fi, i = 1, , N. Measurements at N = 9
different frequencies ranging from 50 Hz to 2 kHz (see Fig. 6 and Fig. 7 below) were
performed on a 100 mm 3.2 mm strip of the non-oriented magnetic material M600 with a
thickness of 2b = 0.5 mm.

3.5 Comparison between Model and Measurements

Since the measurement setup was quite sensitive to noise, the measurements had to be
numerically filtered. Adjusting eff to the filtered data using Eq. (22), the following model
parameter values were obtained: r = 3366, h = 0.477 rad, and b2 = 0.243 Sm, i.e.,
= 3.89106 S/m which is somewhat larger than the true DC conductivity
dc = 3.33106 S/m since excess losses were included in the classical phenomenological
form (11). In Fig. 6, the real and imaginary parts of the measured complex permeability are
compared at different frequencies with the adjusted eff .

14

'eff /0

3000

''eff /0
'meas /0

2500

''meas /0
2000

1500

1000

500

0
0
10

10

10
Frequency [Hz]

10

10

Fig. 6: Real and imaginary parts of the measured complex permeability (symbols) and of the fitted
permeability function (curves), normalized by 0.

The agreement is quite satisfactory considering the simplicity of the model, especially at

at the lowest frequencies is

higher frequencies. The deviation between meas
and eff
probably due to saturation effects which are not properly taken into account by the
expression (11) for eff , see for instance the measurement at 50 Hz (Fig. 5(a)), where the
amplitude had to be chosen large enough for the signal not to be covered by noise.
Below, the H-B hysteresis curves are shown for all measured frequencies. Measurement,
simple model, and detailed model are represented by solid green lines, dashed blue lines and
dotted red lines, respectively.
f=100
1.5

0.5

0.5

B [Tesla]

B
B [Tesla]

f = 50 Hz
1.5

-0.5

-0.5

-1

-1

-1.5
-400

-300

-200

-100

0
H
H [A/m]

100

200

300

400

-1.5
-400

-300

-200

-100

H [A/m]

100

200

300

400

15
f = 200 Hz

f = 400 Hz

1.5

1
0.8

1
0.6
0.4
0.2

B [Tesla]
B

B
B [Tesla]

0.5

0
-0.2

-0.5
-0.4
-0.6

-1

-0.8
-1.5
-400

-300

-200

-100

100

200

300

-1
-400

400

-300

-200

-100

H
H [A/m]

100

200

300

400

100

200

300

400

100

200

300

400

H [A/m]
f = 800 Hz

f = 500 Hz

0.8

1
0.8

0.6

0.6

0.4
0.2

0.2

B
B [Tesla]

B
B [Tesla]

0.4

0
-0.2

0
-0.2

-0.4

-0.4
-0.6

-0.6

-0.8
-1
-400

-300

-200

-100

100

200

300

-0.8
-400

400

-300

-200

-100

H [A/m]

H [A/m]
f = 1000 Hz

f = 1250 Hz

0.8

0.6

0.6

0.4

0.4
0.2

B [Tesla]

B [Tesla]
B

0.2
0

-0.2

-0.2
-0.4
-0.4
-0.6

-0.6
-0.8
-400

-300

-200

-100

100

200

300

-0.8
-400

400

-300

-200

-100

0
H
H [A/m]

H [A/m]
f = 2000 Hz
0.6

0.4

B
B [Tesla]

0.2

-0.2

-0.4

-0.6

-0.8
-500

-400

-300

-200

-100

100

200

300

400

500

H [A/m]

Fig. 7: H-B curves from measurements (solid green line), detailed model (dotted red line) and complex-
model (dashed blue line) with calculated from expression (11), at different frequencies ranging from 50 Hz to
2 kHz.

16

The above way of defining a best fit of ellipses to the more complicated H-B hysteresis
relations approximately preserves both H and B amplitudes and magnetic losses in the whole
frequency range. This is illustrated in the Fig. 7, where the measured H-B curves are
compared with the corresponding complex- ellipses and the detailed model at different
frequencies. As can be seen, the simple model agrees very well with the measurements as
long as saturation is not too strong, which means for low amplitude fields and/or for
frequencies higher than about 200 Hz.

17

4 High Frequency Winding Model

In this Chapter, a winding model based on a lumped element approach with three different
levels of discretization has been developed. The developed models are analyzed using state
space analysis in the frequency domain and the impedances for the three different models are
compared to each other to detect the new phenomena emerging for higher frequencies as the
model discretization is made finer and finer. Parts of this chapter are based on [29].

4.1 Transformer Winding Model for High Frequency Applications

In the field of transformer winding modeling, various approaches and tools are available.
Among the most common tools are lumped element circuits [5, 3032]. Usually all the turns
of one or two discs are lumped together into one inductive element (segment) of the model,
which leads to a decreased computation time but also to a reduction of the models upper
frequency limit, typically to values around some 100 kHz.
In this thesis, lumped element models with much higher resolution (up to 4 segments per
turn) are used, in order to increase the upper frequency limit.
Each lumped element represent a part, a section of the physical geometry with similar
quantities like magnetic flux, electric potential, resistance and etc, and these lumped elements
are connected together to represent the whole geometry. For power transformers, the windings
are divided into finite sections represented by lumped resistance, inductance, and capacitance,
where each section should be small enough so that it can be assumed that the current through
it is constant and not influenced by the displacement current which will be noticeable at
higher frequencies.
Up to a few hundreds of kilo hertz, the displacement current will not be so remarkable and
can be approximated to zero, so that a winding can merely be modeled by means of its self
and mutual inductances and resistances alone. But at higher frequencies, the aforementioned
approximation is no longer valid, and the displacement currents from a section to other
sections or to conductive bodies have to be accounted for, to make the model realistic, and this
is done by means of capacitors. The total capacitance for a particular section is then split into
two halves and located at both ends of the section. As mentioned above, the complete winding
model is made by connecting all the sections together.
In this thesis, the transformer winding is a single phase continuous disc winding (with the
order of turns as shown in Fig. 9), composed of quadratic discs as in Fig. 8. Also, since the
low voltage winding is on a much lower voltage than the high voltage winding, it (the LVwinding) is replaced by ground in the models developed here.

4.2 Three Different Resolutions of the Model

Three different model resolutions are studied: in the models labeled 1, 2, and 3, each turn in
the discs is modeled by one, two, or four segments, respectively. This implies that there are
one, two, or four turn-to-turn capacitances between any two neighboring turns in a particular
disc of the HV winding, respectively (see Fig. 8 from left to right). Further, there are one, two,

18
or four disc-to-disc capacitances between any two turns facing each other in two neighboring
discs, respectively. Also, at the innermost and outermost turn of every disc, there are in
addition one, two, or four capacitances to ground, respectively (these capacitances are not
shown in Fig. 8 due to the lack of space).
n+1
n

mutual
inductances

2n-1

mutual
inductances

resistance

mutual
inductances

self
inductance

4n
2n+1

9
5

3
1

4n+1

U
2
7

4n-1

2n

4n-2

model 3
model 2
model 1
Fig. 8: The three different levels of discretization with the nodes numbered in an increasing sequence from one
winding end to the other. The resolution increases from model 1 to model 3, and the disc-to-disc capacitances
between neighboring discs are oriented perpendicular to the plane and are not shown here.

Ground

Ground
Plane of
symmetry

ks
htot

di

do
i /2
o / 2
Fig. 9: The cross section of the continuous disc winding.
Lself

Rseg

Fig. 10: One segment and its electrical circuit equivalence.

It has been argued in the literature on the basis of simulations and measurements that the
impulse voltage distribution in transformer windings, and the winding impedances for higher
frequencies can be analyzed using air-core self and mutual inductances neglecting the iron
core without serious error (see for instance [3334]). This implies that the complex
permeability model developed in Chapter 3 does not need to be included in the calculations of

19
inductances and losses. In order to further facilitate the calculation of self and mutual
inductances, the discs have been chosen to be of quadratic shape here. The whole HV winding
thus consists of straight segments which are all either parallel or perpendicular to each other,
so that the self and mutual inductances can be calculated by simple analytic formulas [35].
Every segment is modeled by one resistance in series with one self inductance, as it is
depicted in Fig. 10, and there are mutual inductances between any two parallel segments of
the whole winding. For simplicity, self inductances and resistances are shown in Fig. 8 on one
segment only, and some of the mutual inductances are indicated by arrows. In case of model
1, also the connection of the voltage source for impedance measurement for one single disc is
shown.
Finally, all the inductance and capacitance parameters (i.e., all model parameters except for
the damping resistances) are estimated from the physical winding geometry, and not fitted to
measurements, and they are discussed in Sections 4.3 and 4.4. The models are analyzed by
solving their state space equations [3032] in both the frequency- and time domain, which is
discussed in Section 4.5 of this chapter and in Chapter 9, respectively.

4.3 Calculation of the Capacitances

The capacitance depends on the geometry of the conductors and on the permittivity of the
dielectric material between them. Here, the relation for the capacitance between two planar
surfaces, capacitance = permittivity area / distance, is used [36]. The total capacitance
between two turns Ctt in the quadratic disc is then approximately given by
Ctt 4 i 0

i o h 2 i
,
2
i

(23)

where h is the height of the conductor, i is twice the insulation thickness, i is the relative
permittivity of the conductor insulation, i is the inner length of one side of the square
discs, o is the outer length of one side of the square discs, and the addition of 2i to h
accounts for the fringing effect [37]. The total capacitance between two discs, if they are close
enough to each other and if air is used as insulation between them, is approximately given by
Cdd 4 air 0

( o / 2) 2 ( i / 2) 2
i ks

(24)

where air is the relative permittivity of air (=1) and ks is the distance between two discs. The
total capacitance between one disc and the outer ground wall is given by
Cog 4 air 0

o htot
do K

(25)

and the total capacitance between one disc and the inner ground wall is given by

Cig 4 air 0

i htot
di K

(26)

20
where K is the total number of discs in the winding, htot is the total height of the winding, and
d0 and di are the outer and inner distances between the winding and the ground respectively.

4.4 Calculation of the Inductances and Resistances

For the calculation of the self and mutual inductances, formulas in [35] are used. The self
inductance Lself of each straight segment with the length , height h and width w in the disc is
Lself

0
2

2
ln
1 .

0.2235(
w
h
)

(27)

The mutual inductance M between two segments which are perpendicular to each other is
zero, whereas for two parallel segments of length , separated by a distance x (as in Fig. 11),
it is given by
2
2
0
x

ln 1 1 .
M
(28)
2 x
x
x

Fig. 11: Two filaments with negligible cross section area with same lengths.

When the segments are parallel but have different lengths, as in Fig. 12, the mutual
inductance is given by
2M M m p M m q M p M q ,
(29)

m
Fig. 12: Two filaments with negligible cross section area with different lengths.

where for example Mm+p is the mutual inductance between two straight wires both having the
length m+p and being placed relative to each other as in Fig. 11, and which for the symmetric
case p = q reduces to
M M m p M p .
(30)

21
Of course, in the formulas for the mutual inductances it is assumed that the conductors have
very small cross section areas, which is just an approximation. The resistance Rseg of each
segment is assumed to be of the form
1
1
0 f
Rseg

,

wh 2( w h)

(31)

where is the conductivity of the conductor and f is the frequency. The first term is the DC
resistance [38] (which has a vanishing effect for the frequencies dealt with here) and the
second term accounts for the skin effect at higher frequencies [3940]. Since proximity losses
are not included in the model, a numerical factor > 1 has been introduced and adjusted so
that a realistic level of resonance damping is obtained.

4.5 State Space Model for the Winding

The circuit model for the three different winding models in Fig. 8 and Fig. 9 of the single
winding consists of Kni winding sections resulting in K(ni + 1) nodes and Kni inductive
branches and associated capacitances and resistances, where ni = 2i1n, for i = 1, 2, 3 for
model 1, 2 and 3 respectively, and where n is the number of turns in one disc. K is the total
number of discs used in the winding and hence one will arrive at the following two matrix
equations by considering the voltage difference between the nodes of inductive branches and
the current conservation at the nodes:
d
I C V
,
(32)
dt
d
T V L I R I .
(33)
dt

Here V and I are the vectors containing the voltages at the nodes and the currents in the
inductive branches, respectively
V1
I1
I
V
2
2
V
, I
.
(34)

I K ( ni 1)
VKni
I

Kni
K ( ni 1)
K ( ni 1)1

Kni 1

The matrix connects the currents and voltages and consists of 1, 1 and 0, and T is the
transpose of
S 0 0
1 0 0
0 S 0 0
1 1 0 0

, where S 0
0

S 0
0 1 1
0 0 S
0 0 1
K ( ni 1)ni
( ni 1)ni

(35)

22
and 0 is a zero matrix. The resistance matrix for the whole winding R is a diagonal matrix
R disc
0

0
0

0
R disc

0 R disc
0

0
0
,

0
R disc Kn Kn
i
i

(36)

composed of the resistance matrix for individual discs Rdisc , where

R disc

Rseg,1
0

0
0

Rseg,2 0

0 Rseg,ni -1

0
Rseg,ni
ni ni
0
0

(37)

is composed of the resistances of all segments in one disc (for example, Rseg,j is the resistance
of segment j (see Eq. (31))). The inductance matrix for the whole winding L is composed of
smaller matrices
L disc L12 L13 L1K
L

21 L disc L 23 L 2 K
,
(38)
L

L disc


L K 1 L K 2 L disc
Kn Kn
i

where the off-diagonal matrices Lkj are nini matrices for the mutual inductance between disc
k and j, and the matrix in the diagonal Ldisc is the inductance matrix for a single disc being
composed of the mutual inductances M (Eqs. (28)(30)) and self inductances Lself (Eq. (27))
of the segments in one disc

L disc

Lself,1
M
21

M
K1

M 12

M 13

Lself,2

M 23

Lself,ni 1

MK2

M 1K
M 2 K

Lself,ni

(39)

ni ni

(for example, Mkj and Lself,j are the mutual inductance between segment k and j, and the self
inductance of segment j, respectively). The total capacitance matrix C in Eq. (32) is

23
i)
i)
C(i ) C(DD
C(DD

i)
(i )
C(i ) 2C(DD
CDD
C

i)
C(DD
(i )
DD

i)
i)
C(DD
C(i ) 2C(DD

(i )
(i )
(i )
CDD
C CDD K ( n 1)K ( n 1)
i

(40)

where, as mentioned before, i = 1, 2, 3 for model 1, 2 and 3 respectively. C(i) is the specific
capacitance matrix for model i, and for model 1, the roughest model, it is
1 (1)
1 (1)
(1)
0

2 Ctt Cig 2 Ctt

3 (1)
1 C (1)
Ctt
0
Ctt(1)

tt

2
2

0
0
Ctt(1) 2Ctt(1) Ctt(1)

(1)
C

(1)
(1)
0
Ctt 2Ctt
Ctt(1)

3 (1)
0
Ctt
Ctt(1)

0
0
Ctt(1)

1
Ctt(1)
2

1 (1)
(1)
Ctt Cog
2
( n1)( n1)

(41)

where Ctt(1) Ctt , Cig(1) Cig , Cog(1) Cog , (Ctt is the turn-to-turn capacitance (Eq. (23)) and
Cog /Cig are capacitances to ground (Eqs. (25)(26))). For model 2, the next finer model, it is
1
1 (2)
(2)
0
0

Ctt(2)
2 Ctt Cig
2

0
0
0

Ctt(2) Cig(2)
Ctt(2)

1 (2)
3 (2)
0
0
0

Ctt
Ctt(2)
Ctt
2
2

0
Ctt(2)
0
2Ctt(2)
0
Ctt(2)

(2)
C

0
Ctt(2)
0
2Ctt(2)
0
Ctt(2)

3 (2)

0
Ctt(2)
0
Ctt
0

0
Ctt(2)
0
Ctt(2) Cog(2)

Ctt(2)
0
0
0

(42)

1
Ctt(2)

1 (2)
(2)
Ctt Cog
(2n1)(2n1)
2
0

where Ctt(2) Ctt / 2, Cig(2) Cig / 2, Cog(2) Cog / 2 . For model 3, the finest model, it is

24
1
1 (3) (3)

Ctt(3)
0
0
0
0
0
2 Ctt Cig

(3)
(3)
(3)

Ctt
Ctt Cig
0
0
0
0
0

Ctt(3) Cig(3)
0
0
0
0
0 Ctt(3)
0

Ctt(3) Cig(3)
0
0
0
0
0
0 Ctt(3)
0

1
3
(3)
(3)
(3)
C

Ctt
Ctt
0
0
0
0
0
0
0
tt
2

2
(3)
(3)
(3)

Ctt
Ctt
0
0
0
0
2Ctt
0
0
0
0

(3)

(43)

Ctt(3)
Ctt(3)
0
0
0
0
2Ctt(3)
0
0
0
0

3 (3)
1

Ctt(3)
Ctt(3)
Ctt
0
0
0
0
0
0
0

2
2

Ctt(3)
Ctt(3) Cog(3)
0
0
0
0
0
0
0

(3)
(3)
(3)

Ctt Cog
0 Ctt
0
0
0
0
0

Ctt(3) Cog(3)
0 Ctt(3)
0
0
0
0

1 (3)
1 (3)
(3)

C
C
C
0
0
0
0
0
tt
tt
og

2
2
(4n1)(4n1)
i)
in Eq. (40) accounts for the
where Ctt(3) Ctt / 4, Cig(3) Cig / 4, Cog(3) Cog / 4 . The matrix C(DD
capacitive coupling between two neighbouring discs and it has the form

i)
C(DD

1 (i )
2 Cdd

Cdd(i ) 0

0 Cdd( i )

0
1 (i )
Cdd
( ni 1)( ni 1)
2

(44)

where Cdd(i ) Cdd / ni for all three models i.e. i = 1, 2, 3 (Cdd is the disc-to-disc capacitance
according to Eq. (24)).
The pre-factors 1/2 in the capacitance matrices are due to the fact that the capacitances
connected to the first and the last nodes in a disc account only for a half segment, and the prefactors 1, 2 and 3/2 in the diagonals are due to the fact that, if there is no capacitance to
ground, the sum of the elements in a row/column must be equal to zero.
When an external voltage source is connected to a node (k), its node voltage is no longer
unknown (k = 1 here). The voltage at that node Vk and its time derivative dVk /dt should
therefore be separately inserted in (32) and (33) as inputs accompanied by the corresponding
columns of matrices C and T. Eq. (32) and (33) are then transformed to
d
d
V O Vk
dt
dt
d
P Vk T V L I R I
dt

I C

(45)
.

(46)

Here, O consists of one column taken out from the C matrix corresponding to index k, and P
consists of the k:th column taken out from T (transpose of ). In addition, appropriate
changes should be made according to the connection of the terminals of the discs and node
grounding [41]. By rearranging the terms in these equations and putting them in one matrix
equation, Multi Input Multi Output (MIMO) state space model of the lumped parameter
circuit can be formulated as:

25
d
X A X B Vk
dt

(47)

where

0
V
X , A 1 T
I
L

d
1
C
O

C1

dt .
and B

L1 R
1
L P

(48)

The state vector X consists of all the nodal voltages (except the applied one) and inductor
currents of the lumped circuit. By taking the Fourier transformation of the equation system
and selecting all state variables as outputs, one will arrive at
TF ( )

X( )
1
j II A B ,
Vk ( )

(49)

where is the angular frequency, II is the identity matrix with the same size as A, and
C1 O j
B
.
1
L P

(50)

TF() contains all the transfer functions of the nodal voltages (except the applied one) and
inductor currents with respect to the applied voltage Vk.

4.6 Comparison between the Three Models

The first comparison between the three models is made for a single disc winding, and the new
phenomena that emerge with increasing model resolution are studied. The disc in the models
consist of n = 10 turns of varnished copper wire with the conductivity = 5.8107 S/m, the
conductor height h = 7 mm and width w = 3 mm. The inner sides of the square disc have a
length of i = 1.2 m, and the gap between any two neighboring conductors (turns) i.e. twice
the insulation thickness is i = 0.4 mm, and there is no ground wall which means that do and di
are set to infinity in the calculations (see Fig. 9 for a geometrical illustration of the
parameters). These are also the dimensions of the experimental setup which will be described
in Chapter 5. The reason for this choice of dimensions will be explained in the same Chapter.
The magnitudes and phases of the calculated in-impedances Z(f)=U(f)/I(f) (see Fig. 8) are
compared to each other in Fig. 13(a) and (b), respectively. It can be seen that several
resonances occur: the first, pronounced impedance maximum is the fundamental resonance of
the winding, due to the total inductance and series capacitance of the whole disc. As it will be
argued in Section 7.1, the three following resonances can be interpreted as radial resonance
modes, and the two after that, which form a pronounced impedance minimum above 10 MHz
and do not appear in the lowest-resolution model 1, as azimuthal resonance modes.
As can be seen in Fig. 13, for model 1 the impedance becomes purely capacitive i.e. it
becomes of the form Z = (jChf)-1 after the radial modes i.e. for frequencies higher than about
6 MHz (Chf is the winding model capacitance at high frequencies). This is not a physical
reality and it means that model 1 is for sure not valid for that part of the frequency spectrum.
For model 2 and 3, the impedance becomes purely capacitive in the end of the frequency
spectrum after the azimuthal modes, which as expected would mean that with finer
discretization the model becomes valid for higher frequencies.

26

model 1
model 2
model 3

Impedance amplitude []

fundamental coil
resonance
4

10

10

radial modes
0

10

azimuthal modes
6

10
Frequency [Hz]

10

(a)

Impedance phase [degrees]

100
model 1
model 2
model 3

50

-50

-100

10
Frequency [Hz]

10

(b)

Fig. 13: (a) Impedance amplitude and (b) impedance phase for the different models.

27

5 Frequency Response Measurements

In this Chapter, the measurement device and the experimental setup are introduced and the
choice of dimensions for the setup is explained. Also, the different measurements performed
are described.

5.1 Impedance Measurement

Experimental Setup

Device

and

Dimensioning

of

the

A network analyzer Bode 100 from Omicron Electronics [42] (frequency range 1 Hz
40 MHz) was used for the impedance measurements (see Fig. 14).

Fig. 14: Impedance measurement device Bode 100 from Omicron Electronics.

As mentioned before, the shape of the discs is chosen quadratic so that all the self and mutual
inductances can be calculated by simple analytic formulas from [35]. The location of the
resonances of the winding in the frequency spectrum depends on the physical geometry and
material properties of the winding, and generally, larger dimensions of the winding leads to
larger inductance and capacitance values, which in turn leads to the resonances occurring for
lower frequencies.
The measurement device can measure up to the frequency of 40 MHz, but the
measurements will be more sensitive to the effects of environmental noise and measurement
cables for the higher part of this frequency range. Consequently, the measurements will be
more disturbed and unreliable for this high-frequency part of the frequency range, and due to
this fact, the geometrical size of the discs had to be chosen so that all the interesting
phenomena and resonances occur for frequencies below approximately 20 MHz.
The most effective way to satisfy this requirement is to design the square discs with large
side lengths. So the constructed discs consist of n = 10 turns of varnished copper wire with
rectangular cross section (7 mm 3 mm). The inner sides of the square discs have a length of

28
1.2 m, and the gap between any two neighboring conductors (turns) is varying between
0.4 mm (= twice the insulation thickness) and about 1 mm because of manufacturing
irregularities. The cross section dimensions of the varnished copper wires (7 mm 3 mm) are
regular wire dimensions used in power transformers, and the number of turns can not be
chosen too high (the discs will be to heavy and impractical to handle), or too low (there will
be too few resonances), so n = 10 turns seemed reasonable and was chosen.
10 units of these discs were manufactured by the transformer manufacturing company
Nordtrafo AB [43] (see Fig. 15).

Fig. 15: Separate disc units manufactured by Nordtrafo AB.

5.2 Impedance Measurement Results

The first measurements were performed on each of the ten different single disc units
separately (Fig. 16).

Fig. 16: Frequency response measurement on a single disc.

By comparison between the impedances of the different discs, it could be affirmed that due to
the manufacturing irregularities mentioned above, the impedances of the discs differ from

29
each other more and more as the frequency is increased. This seems consistent, because for
higher frequencies, the capacitances between the turns in the discs play a larger part in
shaping the impedance. Since a capacitance is per definition sensitive to small distance
changes between two conductors, and since (as mentioned earlier) the gap between any two
neighboring conductors (turns) is varying between 0.4 mm and about 1 mm (because of
manufacturing irregularities), hence the discs have different capacitive features for higher
frequencies, leading to different high frequency impedances. In Fig. 17, the measured
impedance for four of the discs is plotted to illustrate the differences.

fundamental coil
resonance

Impedance amplitude []

10

10

radial modes
0

10

azimuthal modes
6

10
Frequency [Hz]

10

Fig. 17: Impedance measurement for four different discs.

It can from Fig. 17 be seen that for frequencies up to approximately 300 kHz, the discs have
the same impedance value. This is the so called inductive regime where the impedance
behaves as Z = jLlf (where Llf is the inductance of a disc for low frequencies), and since the
inductance is not so sensitive to small irregularities in the physical geometry, hence the discs
have all the same impedance. But as the frequency is approached to the fundamental coil
resonance frequency, the capacitance starts to play in, and it can be seen that from now on the
impedances do not coincide with each other. The radial and azimuthal modes are still
there for all of the discs, but they do not have exactly the same location, shape and amplitude.
Measurements were then performed on units with a varying number of discs connected
together. The distance between two neighbouring discs ks was changed between
approximately 1 mm to 5 cm, and different types of connections between the discs were tried,
and all these for comparison between models and measurements. It was found that as the
distance between the discs is decreased, the models deviate more and more from the
measurements, and the reason for this will be discussed and explained in Chapters 6 and 8.
Next, the model simulations will be compared to measurements for verification.

30

31

6 Model Verification
In this Chapter, model 3 is verified by comparison with measurements. The model is
compared with measurements for a winding with one, three, six and nine discs respectively,
and it is shown that the resonances predicted by the model also occur in the measurements.

6.1 Comparison of Model with Measurements

The impedance magnitudes of model 3 and measurement for one single disc are compared in
Fig. 18, where it can be seen that they have an overall satisfactory agreement. Also, the
comparison shows that the radial modes and the high-frequency azimuthal modes, which
model 1 is unable to produce and which model 2 produces partially, are no model artifacts but
real physical phenomena.

Impedance amplitude []

simulation
measurement
4

10

10

faz
5

10

10
Frequency [Hz]

10

Fig. 18: Comparison between measurement and model 3 for a single disc.

In Chapter 7, the physical meaning of the radial and azimuthal resonances will be explained
and discussed.
In the measurements, the radial modes are shifted toward somewhat higher frequencies
compared to those in the simulations, but they are fully recognizable. Such a shift is expected
to occur due to the proximity effect which has not been taken into account in the model
calculations reported here, and which is going to be discussed in Chapter 8.

32
Next, the impedance magnitudes of model 3 and measurements for a three, six and nine
disc winding, with the distance between two discs ks being approximately 6 mm, are
compared in Fig. 19 to Fig. 21, respectively.

Impedance amplitude []

simulation
measurement
4

10

10

10

10
Frequency [Hz]

10

Fig. 19: Comparison between measurement and model 3 for a three disc winding.

simulation
measurement

Impedance amplitude []

10

10

10

10

10
Frequency [Hz]

10

Fig. 20: Comparison between measurement and model 3 for a six disc winding.

33

simulation
measurement

Impedance amplitude []

10

10

10

10

10
Frequency [Hz]

10

Fig. 21: Comparison between measurement and model 3 for a nine disc winding.

Even here in Fig. 19Fig. 21, the comparisons show an overall satisfactory agreement
between model 3 and measurements, confirming the general accuracy of the model. But, as
was the case in Fig. 18, there are shifts between some of the resonances reproduced by the
model and the measured resonances. The reason for this, as for the resonance shifts for one
single disc, is expected to be the proximity effect which is neglected in the models and which
is going to be discussed in Chapter 8.

34

35

7 Interpretation
This Chapter, which is mainly based on [29], deals with the radial and azimuthal resonances.
The features of these resonances are explained, and it is shown that they are present also in a
winding with several discs.

7.1 Explanation of the Different Oscillation Modes

As it was mentioned in Section 4.6, the three resonances somewhere between 1 MHz to
8 MHz in Fig. 13 for a single disc are called radial resonances and the ones somewhere
between 10 MHz and 20 MHz are called azimuthal resonances, and it could be seen in Fig. 18
that these resonances also occur in the measurement, which means that these are physical
realities and not model artifacts, and now the physical meaning of these resonances will be
investigated and explained.
7.1.1

Radial Resonance Modes

Those resonances whose node voltages vary rapidly in the radial direction, but slowly in the
azimuthal direction are called radial. By radial and azimuthal directions, the and
directions in polar coordinates are meant, respectively (see Fig. 22 for model 3).

= 3/2

=0
4n

4n+1

1
4

5
1

0
3

2
6

4n-1

4n-2

= /2

Fig. 22: Geometry of the disc, showing the definitions of coordinates and .

In Fig. 23 Fig. 25, instantaneous node voltages for a sinusoidal excitation voltage U are
depicted for different resonance frequencies, each at two different instants of time during an
oscillation period, obtained from simulations of model 3.

36
A linear voltage profile along the whole winding (which is the low-frequency limiting
behavior) is subtracted, so that the values at both end nodes of the winding are equal to zero.
The green line depicts the geometry of the winding disc and the location of the nodes, and the
thin horizontal red line shows the zero level of the voltage as a reference. Black lines connect
voltage levels in radial direction, and vertical blue lines indicate the correspondence between
voltage levels and nodes.
The voltage distribution in the disc for the first, second and third radial resonance of Fig.
13 is depicted in Fig. 23, Fig. 24, and Fig. 25, respectively. fk (k = 1, 2, or 3 ) denotes the
frequency for that particular resonance, and 1/fk is the corresponding period time. The radial
resonances appear as standing voltage waves which can approximately be described by the
formula

0
0 1

Vrad,k ( , , t ) cos(2 f k t ) Ak sin k 2

Bk

2
1 0

(51)

for 0 < < 1 , where 0 and 1 are the inner and outer radii of the disc, respectively (see
Fig. 22). The resonance voltage amplitudes Ak and Bk are damping dependent. Note that the
approximate expression (51) is independent of . It can be seen in Fig. 23Fig. 25 that the
approximation (51) is best for low resonance order k. The amplitude Bk is close to zero for
k = 1 and increases with increasing resonance order k.

Vrad,k

Fig. 23: Voltage profile of the first radial resonance (k = 1), at times t = 0.7/f1 (left) and t = 1.4/f1 (right).

Vrad,k

Fig. 24: Voltage profile of the second radial resonance (k = 2), at times t = 0.4/f2 (left) and t = 0.8/f2 (right).

37

Vrad,k

Fig. 25: Voltage profile of the third radial resonance (k = 3), at times t = 0.3/f3 (left) and t = 0.6/f3 (right).

7.1.2 Azimuthal Resonance Modes

For azimuthal resonances, just like the radial ones, the node voltages vary rapidly in the
direction, but the difference is that there are also significant node-voltage variations in the
direction. This pattern can be seen in Fig. 26 and Fig. 27 which depict the instantaneous node
voltages for the two dominant azimuthal resonances, appearing in Fig. 13 and Fig. 18 as
pronounced minima close to faz, at two different instants of time. Again, model 3 has been
employed and a linear voltage profile has been subtracted.
In contrast to the radial resonances which are spread out in frequency, the azimuthal
resonances are clustered (at least when viewed on a logarithmic frequency scale) around a
characteristic frequency faz slightly above 10 MHz. They cannot be described by a simple
formula like that for the radial resonances (51), but their common characteristics is
approximated by the expression

Vaz ( , , t ) Aaz cos(2 f azt ) 1 cos( ) 0

.
2

(52)

This fundamental behavior is indicated in Fig. 26 and Fig. 27 by dotted lines. Individual
azimuthal resonance modes differ from it by additionally superposed short-wavelength
modulations. The mode in Fig. 26 resembles more closely to the fundamental expression (52)
than the one in Fig. 27.

Vaz

Fig. 26: Voltage profile of the first azimuthal resonance, at times t = 0.25/faz (left) and t = 0.7/faz (right).

38

Vaz

Fig. 27: Voltage profile of the second azimuthal resonance, at times t = 0.4/faz (left) and t = 0.65/faz (right).

7.2 Radial and Azimuthal Resonances for a Winding with More Than One
Disc
As it was reported in Chapter 6 and illustrated in Fig. 19, Fig. 20 and Fig. 21, measurements
were also performed for a winding with more than one disc. In Fig. 28, measurements are
shown (again) for a winding consisting of one, three, six and nine discs respectively,
connected together in a continuous way as in Fig. 9.

1 disc
3 discs
6 discs
9 discs

Impedance amplitude []

10

10

10

Radial
modes
5

10

Azimuthal
modes
6

10
Frequency [Hz]

10

Fig. 28: Measurements on windings with one, three, six, and nine discs respectively.

It can be seen that even for a several-disc-winding (and irrespective of the number of discs in
it), the radial and azimuthal resonances exist, and occur around the same frequencies as for a
single disc (see the dotted rings in Fig. 28). This supports the picture that these resonances are
internal oscillations in every individual disc.
Also, it can be noticed that the radial resonances (at least the first two) and the azimuthal
resonances tend to get smoothed out as the number of discs in the winding is increased. This
is expected to be due to the small manufacturing differences leading to the positions, shapes
and amplitudes of the resonances for different discs being different (see Fig. 17), and when

39
the number of discs in the winding is increased, these relatively different resonances
superimpose on each other, leading to a smoothening-out. Another reason for this
smoothening-out could be the proximity losses which enter the picture when several current
carrying conductors are placed close to each other (see Section 8.1).
Furthermore, measurements and simulations show that especially the azimuthal resonances
are very sensitive to small changes in the winding geometry (e.g. mechanical winding
deformations). For instance, in the measurements on different single discs of identical design
but with small manufacturing differences (see Fig. 17), the relative strength and shape of the
two dominating azimuthal resonances varied noticeably, each disc thus having its individual
finger print.
It is also worth to mention that the new resonances which emerge in the frequency range
21052106 for a winding with more than one disc are global oscillations along the total
length of the winding between different discs, called axial resonances here . A hallmark of
the global resonances is this observation that they shift considerably in frequency when the
number of discs in the winding is increased.

40

41

8 Reluctance Network Method for Inclusion of

the Skin and Proximity Effects
In this chapter, the impact of the skin and proximity effects on the inductances and losses for
a winding is explained, and a method based on the reluctance network approach is described
to compute the frequency dependent inductance and resistance matrices. This method (which
has been developed in [4445]) employs a network of complex reluctances describing the
flux paths around the conductors. The use of this method in this chapter is limited to the
conductor arrangements of multi-conductor transmission lines surrounded by a shield, since
these are easier to start with than a winding, and its extension to windings is left for future
work. The chapter is concluded with verification of the results reproduced by this method by
comparison with finite-element method calculations. This chapter is to a large extent a
description of what has been stated in [4445].

8.1 Skin and Proximity Effects

As it was mentioned earlier in Section 4.4, the formulas for the inductances are for thin
filaments and are moreover frequency independent. This is a good approximation as long as
the distance between two conductors is much larger than their largest cross sectional
dimension, and as long as the frequency is low enough. But when the frequency is high and
the two conductors are close to each other, the so called skin effect and proximity effect will be
present respectively.

Skin effect when a time-varying current flows in a conductor it creates a time-varying

magnetic field which in turn induces eddy currents i.e. induced currents that counteract the
original current. The consequence is that the total current tends to be confined to the surface
of the conductor. This effect becomes stronger as the frequency is increased and the effective
current carrying area of the conductor becomes restricted to a thin layer below the surface,
which is called skin depth and is defined by

0 r f

(53)

The simulation in Fig. 29 (a) shows the influence of skin effect on the distribution of the
current- and magnetic flux densities. It can be noticed that due to the skin effect, the current
density becomes non-uniform in the radial direction , but it is still uniform in the azimuthal
direction when the conductor has cylindrical symmetry.

Proximity effect in contrast to the skin effect, proximity effect is about eddy currents which
are induced in a conductor due to a time-varying magnetic field produced by the currents in
the other conductors in the vicinity. Due to the proximity effect, the current- and magnetic
flux density distributions become unsymmetrical in both and directions (see Fig. 29 (b)).

42

(a)

(b)

Fig. 29: Distribution of magnetic flux density (arrow plot) and current density (surface plot) for circular
conductors; in (a) skin effect is present only, while in (b) both the skin- and proximity effects are present (for the
currents in the conductors flowing in the same direction). This figure is borrowed from [46].

The impact of skin effect on the self inductance is that the internal inductance of the
conductor decreases, but since the internal inductance is a very small part of the total self
inductance of a single conductor, the influence of skin effect on the self inductance can be
neglected. The skin effect introduces of course additional, frequency dependent losses. These
losses have already been incorporated in Eq. (31)
But the proximity effect (in combination with the skin effect) is more serious, and this is
due to the fact that when the frequency is very high and the conductors are really close to each
other, the current is not only confined to the surface of the wires but it is distributed around
the axis in conformity with the law of distribution of the charges in the corresponding
electrostatic problem [35]. This means that if current flows in opposite/parallel directions in
two parallel conductors, the current density in each conductor is a maximum at the
nearest/farthest points of the cross sections of the conductors. This has the effect of a
reduction/expansion of the effective spacing of the conductors, which means that the mean
distance between the effective current carrying areas will be smaller/larger than the distance
between the centers of the two conductors which is the distance used in the thin filament
approach (see Fig. 11 and Fig. 12). This in turn means that the actual mutual inductances for
turns close to each other will deviate from the ones calculated in expressions (28) (30). The
proximity effect will also introduce additional, frequency dependent losses, which would be
equivalent with the resistance matrices in Eqs. (36) and (37) gaining off-diagonal frequency
dependent components. It is argued here that the deviations between model and measurements
in Fig. 18 to Fig. 21 are most probably due the absence of these high-frequency phenomena in
the formulas used to calculate the inductances and resistances in Sections 4.4 and 4.5.

8.2 Using Reluctance Network Method (RNM) to Calculate the Frequency

Dependent Inductance and Resistance Matrices
The solution proposed here for calculation of the frequency dependent inductance and
resistance matrices is the so called Reluctance Network Method (RNM); this method is started
by considering the cross section of one side of a single quadratic disc embedded in a
reluctance network, as depicted in Fig. 30.

43

magnetic flux
conductor

I1

I2

In-1

In

Fig. 30: The cross section view of one side of a single disc, showing the conductors, conductor-insulation,
exaggerated air gaps and different reluctances.

The dimension of the air gaps (due to the manufacturing uncertainty and irregularities)
between the conductors is exaggerated in the figure above to make is possible to illustrate the
problem picture.
As it will be shown below, it is possible to find an easy analytical expression for the
reluctance in a gap in which the length of the channel is much larger than its width, i.e. in
between the turns in Fig. 30. But this is not the case for the reluctances outside the turns
( and ) since there is no easy way to define an area which the magnetic flux density
passes through and a length which it travels along. This means that it is easier to start the
development of this method with a configuration containing narrow gaps only. This would be
a configuration in which the conductors are surrounded by a shield; a multi-conductor
transmission line. What follows below in the rest of the chapter is a description of the
derivations and results in [4445]: the method is first developed for higher frequencies, and it
is then extended to lower frequencies. After that, it is applied to some simple example
geometries, and its accuracy is verified by comparisons with Finite-Element Method (FEM)
field calculations.

8.3 RNM for High Frequencies

Consider an arrangement of parallel conductors of length aligned in x direction, with
arbitrary cross section in the y-z plane. On these AC currents of frequency are imposed,
denoted by Ik with k = 1,,n , where n is the total number of conductors. A very simple
example with n = 2 conductors is pictured in Fig. 31. A slightly more complex and general
example is discussed further below (Fig. 32). The convention that positive current directions
always point out of the plane (red symbols) is adopted here. For clarity, is kept as a
parameter in the equations below; in order to obtain per-length expressions, one simply has to
divide by .

R1

I1

R2

m,1

I2

R3

m,2

I1

I2

(a)
(b)
Fig. 31: Cross-sectional view of n = 2 parallel conductors, oriented perpendicularly to the plane. (a) Reluctance
network with chosen branch orientations, (b) circulating mesh fluxes.

44
8.3.1 Principle of the Reluctance Network Description

Assume for the moment that the conductors are perfectly conducting (or, alternatively, that
the frequency tends to infinity), so that the skin depth vanishes and magnetic flux exists only
outside the conductors. The magnetic coupling between these conductors is described by a
network of reluctances R i (i = 1,,s), characterizing flux channels formed by the gaps
between the conductors as well as the open space around them. In other words, the R i
characterize all the available distinct magnetic flux paths [47]. The R i of the flux channels
are determined by the geometry and material parameters of individual gaps between
conductors, whereas those of the open space also depend on the availability and nature of
some external current return path. Each network branch corresponds to one of these flux paths
and reluctances, and each network mesh surrounds one of the conductors.
For each reluctance branch an (arbitrary but fixed) orientation is chosen, defining the
direction of positive flux i in the branch.
At any given moment, the currents Ik flowing through all the conductors and the branch
fluxes i in the reluctance network can be combined into vectors,
I1
1


I and .
I

n
s

(54)

Since magnetic flux is conserved locally (i.e., at each node of the network, the total inflowing
flux is zero), any possible flux state can be built up by superposition of mesh fluxes m,k
, k = 1,,n , circulating counter-clock-wise in the meshes around the n conductors (Fig. 31b,
Fig. 32b).They are combined here into a new vector m with n elements, and a s n matrix
is defined as follows:
Definition of the matrix :

(55)

For any given branch index i and conductor index k, has an element ik = +1 (or 1) if
the branch i belongs to the mesh around conductor k and is oriented in parallel (or
opposite) to the circulation direction, and ik = 0 if it does not belong to mesh k.
can be interpreted as the connectivity matrix of the graph which is dual to the reluctance
network. In the simple example of Fig. 31, the total number of reluctance branches is s = 3,
and the matrix is given by:
k 1

1 0
1 1
0 1

1
2
3

(56)

Using , the relation between the branch and mesh fluxes can be written as a simple matrix
equation,
m .
(57)
The line integral of the magnetic field H along any branch i is proportional to the flux i
along that branch, with Ri as proportionality constant. Summing up these branch
contributions around any mesh k in the network yields the line integral of H on a closed loop

45
around conductor k, which according to Ampres law is equal to the enclosed current Ik .
Expressing the reluctances R i as a diagonal matrix
0
R1
R=

,
0
R s

(58)

this can again be written as a matrix relation, I TR , where T denotes the matrix
transpose of . Together with (57) this results in
I T R m .

(59)

Finally, by virtue of Faradays law the induced voltage drops between opposite ends of the
conductors are given by the time derivatives of the mesh fluxes, U dm dt , which in
frequency domain leads to
U j m ZI .
(60)
Z j T R .
1

with

(61)

Finally, when losses in the conductors are included the reluctances become complex functions
of frequency (see next section), and the total series impedance matrix of the conductor
arrangement acquires a resistive contribution and can simply be expressed as
Z R j L j T R

(62)

In the second example shown in Fig. 32 where n = 5 and s = 8, the outermost conductor
(k = 5) forms a shield surrounding the other conductors. Its cross section is multiply
connected, so that the contour describing its mesh flux m,5 (dashed line in Fig. 32b) consists
of two parts.

i=1

k=1

k=1
3

3
2
4

5
3

2
8

(a)
(b)
Fig. 32: Cross-sectional view of a more general arrangement of n = 5 parallel conductors. (a) Reluctance
network with chosen branch orientations, (b) circulating mesh fluxes.

46
For this example, the matrix is given by:
k 1

1
1
1
0
0
0
0
0

0
1
0
1
1
0
0
0

0
0
1
0
1
1
0
0

0
0
0
0
0
1
1
0

1
0
0
1
0
0
1
1

1
2
3
4
5

(63)

6
7
8

Note that the sum over each row except the last is zero, such that any common mode
voltage (i.e., U1 = = Un) corresponds to current flow in the shield only (i.e., Ik = 0 for
k n), and the shield current obeys the relation j In RsUn where s is the index of the
outermost flux path surrounding the shield (in the present case, s 8).
On the other hand, if the shield is short-circuited (i.e., if Un = 0 is imposed) all flux outside
the shield is suppressed at nonzero frequencies, and so flux path s can be disregarded
altogether. For the calculation of reluctances of the inner flux paths the shield is treated in
the same way as the other conductors. Under these conditions, the remaining conductor
voltages U1,,Un1 and currents I1,,In1 are related to each other by a reduced Z matrix
which does not depend on Rs , and the shield acts as common current return path. The
numerical examples in Section 8.6 below will be limited to this situation (it may be called the
differential mode situation).
8.3.2 Calculation of the Reluctances for High Frequencies

At a finite frequency which is still so high that the skin depth in all conductors is small
compared to their thickness, the flux penetrates only slightly into the conductors, and the
above picture of separate flux channels which do not interfere with each other is still valid.
Eq. (62) has provided an explicit expression of the impedance matrix in terms of the gap
reluctances R i , but one still needs to calculate the latter for a given geometry. In the limit of
high frequencies, one can focus on a single gap and model its geometry locally as shown in
Fig. 33. Orienting the channel i in such a way than the positive flux direction is to the right,
all quantities related to the conductor above the gap have been labelled here with a subscript
+,i and those below the gap with ,i. Later on, conductor specific quantities will instead
sometimes be labelled by the original conductor index k, which obeys the condition ik = +1
or ik = 1 for the conductor above or below the gap, respectively, according to definition (55)
of the matrix .
It is assumed that the medium in the gap is perfectly insulating (g,i = 0) and has a finite
permeability g,i > 0. The conductors above and below the gap are given finite conductivities
,i > 0 and permeabilities ,i > 0, respectively, which leads to flux penetration into the
conductors characterized by finite skin depths ,i (see further below).

47

conductor ( ,i , ,i , ,i )
d ,i

z
y

ai
d ,i

,i

skin layer

g ,i

insulator ( g,i )

,i

skin layer

,i

,i

flux
channel

conductor ( ,i , ,i , ,i )
wi

Fig. 33: Local geometry of the gap between two conductors.

z
x

,i

R ,i

g ,i

R g,i

,i

R ,i

Fig. 34: Parallel reluctances of insulating gap and skin layers in the adjacent conductors.

Assume for the moment that the gap is straight and that its width ai is constant along its whole
length wi. The case of varying gap width will be discussed below in Section 8.5. For a long
and/or narrow gap (ai wi) and for strong skin effect (,i d,i, where d,i are the thicknesses
of the two conductors in direction perpendicular to the channel, see Fig. 33), the conductor
cross sections can be approximated locally as infinite half planes, and it can be assumed that
the current densities in the conductors and all the fields are oriented parallel to the surfaces
everywhere (in the x-y plane) and are varying in z direction only. The current density is
chosen to point in x direction and the magnetic field to point in y direction. Since the quasistatic magnetic coupling between the conductors is of interest here, solutions of Maxwells
equations where the charge density is zero everywhere are looked for. The only nonvanishing components of the electric and magnetic fields are then Ex and Hy, and Maxwells
equations reduce to the one-dimensional problem given by
dEx
j H y ,
dz
dH y
j Ex J x .
dz

(64)
(65)

Since there are no surface currents, both Ex and Hy must be continuous in z direction. Current
density and electric field are assumed to be related by Jx = Ex, so (64) and (65) can be
combined into a single equation for Hy. Neglecting the displacement current jEx (which has

48
a noticeable effect only at much higher frequencies than the ones of interest here), one ends
up with the well-known simple diffusion equation [48] for Hy inside a conductor,
d2H y
dz 2

j H y 2 H y ,

(66)

where

1 j

, with

(67)

For simplicity, denotes the total permeability here and in the following, including the
permeability of free space 0, and it may have a different value in each conductor. The
conductivity may also have a different value in each conductor. is the complex wave
number for magnetic field penetration into the conductor at frequency and the
corresponding real skin depth. Since the gap is insulating, Hy is constant in this region,
according to Ampres equation (65). The fundamental solutions of Eq. (66) within the
conductors are exponential functions e z .
In order to calculate the reluctance of the channel, one has to relate the total flux i
through it to the magnetic field Hy in the gap. One can start with the contribution of the upper
conductor, assumed for the moment to occupy the region {z > 0}, so that the solution to (66)
has the simple exponential form
z
H y ( z ) H y ,0 e ,i .
(68)
Here Hy,0 is the magnetic field at the conductor surface {z = 0} and inside the gap i (where it
is independent of z), and ,i 1 j ,i are the complex wave numbers in the conductors
above and below gap i, as defined in (67). As a consequence of (68), the magnetic flux +,i
inside the upper conductor (which is roughly confined within a distance +,i from the surface)
is given by

H
,i ,i dz H y ,i y ,0 .
(69)
0

,i

The corresponding expression for the magnetic flux ,i inside the lower conductor is
analogous. Inside the gap the magnetic field is constant and equal to Hy,0 , so that the flux in
the gap is given by g,i ai g,i H y ,0 . Together with the fluxes (69) in the adjacent skin layers
it adds up to the total flux in the channel,

i ,i g,i ,i ,i ai g,i ,i H y ,0 .
,i
,i

(70)

The reluctance R i of the channel is now defined as the proportionality constant between i
and the line integral of the magnetic field along the channel wi Hy,0 , and thus

w
R i i ,i ai g,i ,i

,i
,i

(71)

49
Note that this result can be interpreted as a parallel connection R i R 1,i R g,1i R 1,i of
1

individual reluctances
R ,i

wi ,i
,
,i

R g,i

wi 1
,
ai g,i

R ,i

wi ,i
,
,i

(72)

which refer to insulating gap and adjacent skin layers, as indicated in Fig. 34.

8.4 Extension of RNM to Low Frequencies

8.4.1 Summary of the Generalized Approach

With decreasing frequency the skin depth increases until it exceeds half the conductor
thickness, i.e., skin layers of opposite sides of the conductor start overlapping and the above
picture of individual flux channels is not applicable anymore. However, it will be argued
below that even for low frequencies a reasonable approximation can still be obtained by
generalizing Eqs. (62) and (71) in the following way:
R
Z R jL Zint Zint
j T R

(73)

where R is still a diagonal matrix, but its elements are modified by additional factors ,i to
the inverse skin layer reluctances R 1,i (compare (72)):

w
Ri i

,i

,i ai g,i ,i ,i

,i
,i

(74)

It will be shown below that these factors have the form

,i tanh(

,i d ,i
2

) ,

(75)

coming from the usual skin effect formula for slabs [48] with finite thicknesses d,i. At high
frequencies (i.e., for ,i d,i), ,i tend to 1 so that the previous expression (71) is recovered.
An alternative way to write (75), which will be more convenient further below, uses
conductor indices k instead of subscripts :
d
k ,i tanh( k k ,i ) ,
(76)
2
where dk,i denotes the thickness of conductor k perpendicular to the adjacent channel i (i.e.,
dk,i = d+,i or d,i , and k = +,i or ,i , etc. for ik = 1 or 1, respectively).
R
Zint and Zint
are diagonal matrices, which have been added since the original expression
(62) in combination with (74) leads to a wrong low-frequency limit for the internal
R
impedances of individual conductors. The element Z int,k
is the internal impedance of
conductor k, resulting under the assumption that (62) together with (74) are valid at all

50
frequencies. It is calculated from these equations by setting all permeabilities in the system
equal to zero, except for k (which physically means that conductor k is completely immersed
in a perfectly diamagnetic medium, so that all flux is confined to the interior of that
conductor). In that limit, all matrix elements of ( T R ) 1 tend to zero except for

( R )
T

kk

wi k 1

k ,i
i with ik 0 k

(77)

(which is the inverse of the sum of all the skin layer reluctances around conductor k), so that
R
Z int,
k

j k
.
wik,1i

(78)

i with ik 0

The summation is around the mesh surrounding the conductor k. (Note that the sum over the
wi without the k,1i factors is just the cumulative length of the channels which surround
conductor k, i.e., the approximate circumference of the conductor cross section.)
At low frequencies, k k,1i 2 d k ,i and so expression (78) is proportional to j, implying a
R
has
vanishing DC resistance which is physically incorrect. In order to fix this problem, Zint
been subtracted in (73) and instead a diagonal matrix Zint whose elements are the true internal
impedances Zint,k has been added . It will be argued below that a reasonable approximation for
Zint,k is provided by an expression very similar to (78),

Z int,k

j k
,
wik ,i

(79)

i with ik 0

where only k,1i is replaced by the new quantity

k ,i tanh(

k d k ,i
4

) .

(80)

R
At high frequencies (,i d,i) where both k,1i and k ,i tend to 1, Z int,k and Z int,
k cancel each
other. In the limit 0, expressions (79), (80) lead to the correct DC resistance for
conductors with rectangular or circular cross section, and therefore they are adopted here as a
general rule. For instance, for a rectangular cross section with side lengths p and q, numbering
the channels around it by i = 1,,4, one has dk,1 = dk,3 = w2 = w4 = p and
dk,2 = dk,4 = w1 = w3 = q, as well as k ,i 14 k d k ,i at low frequencies. Eq. (79) then reduces to

Z int,k

j k

,
2
k pq k pq

(81)

which is the correct DC resistance Rdc of a conductor of length with cross-sectional area pq.

51
8.4.2 Exact Solution of a Single-Slab Problem

In order to justify the modified reluctance formula (74) for conductors of finite thickness, the
particular two-dimensional geometry shown in Fig. 35 is studied. A conducting slab of
length , width w, and thickness d with conductivity and permeability (since there is only
one conductor, the conductor index is dropped) is sandwiched between two insulating sheets
of thicknesses a1 and a2 with permeabilities g,1 and g,2 , respectively. The boundary
conditions in z direction are given by perfect diamagnetic media ( = 0) and those in y
direction by perfect magnetic conductors ( = ), so that when a current is flowing in x
direction through the conductor, the magnetic flux is confined within the insulating sheets as
well as the skin layers in the conductor, and it sees zero reluctance when going around the
side faces of the slab. As a consequence, the reluctance network corresponding to this
geometry has the simple structure shown in Fig. 36.

,2 ( )

conductor ( , , )
skin layer

d 2

a2

skin layer

z
x

insulator ( g,2 )

d 2

,1 ( )

insulator ( g,1 )

a1

0
w
Fig. 35: Simple single-slab geometry.

,2

R ,2
I

,1

R ,1

g ,1

R g,1

R g,2

g ,2

0
Fig. 36: Reluctance network of the single-slab geometry.

The reluctances are calculated in a similar fashion as in Section 8.3.2 above. Now, two
surfaces of the slab have to be take into account, so the origin of z is chosen in the middle of
the slab and the functions sinh z 12 e z e z and cosh z 12 e z e z are used as

52
fundamental solutions of (64), (65) (instead of ez). The general solution in a slab of
thickness d extending from z = d/2 to z = d/2 can then be written in the form

H y ( z) C

Ex ( z )

cosh( z )
sinh( z )
C
,
cosh( d 2)
sinh( d 2)

sinh( z )
cosh( z )
C
C
,
sinh( d 2)
cosh( d 2)

(82)

(83)

with arbitrary coefficients C. These coefficients have to be determined for a given total
current I through the slab. According to (82), (83), the field values at the slab surfaces,
H y H y ( z d 2) and Ex Ex ( z d 2) , are related to the C in the following simple
way:
H y C C ,

C
C
,

(85)

sinh( d 2)
tanh( d 2) .
cosh( d 2)

(86)

Ex

with

(84)

At low frequencies, when the skin depth ( ,1 ,2 , see Fig. 35) is larger than half the
conductor thickness d, the two skin layers overlap and +,1, ,2 can no longer be calculated
separately. However, the total magnetic flux c = +,1 + ,2 in y direction inside the
conducting slab can be obtained by integration of (82),

d 2

dz H y

d 2

2C

(87)

Similarly, the total current I flowing in x direction through the slab is obtained by integration
of (83) (or, alternatively, using Ampres law (65)):

I w

d 2

dz Ex w H y H y 2wC .

(88)

d 2

The total fluxes g,1, g,2 in the insulating layers below (z < d/2) and above (z > d/2) the
slab, respectively, are given by

g,1 a1g,1H y a1g,1 C C

and

g,2 a2 g,2 H y a2 g,2 C C . (89a,b)

The voltage drop U in the slab can be obtained by integration of the electric field along a
closed contour which runs in x direction along the conductor (for instance, at one of its
surfaces) and returns in the field free region below (z < d/2 a1) or above (z > d/2 +a2) the
system. According to Faradays law, both choices of contour must lead to the same result,

53
since the total flux in y direction (both inside and outside of the slab) must vanish due to flux
conservation,
g,1 c g,2 0 .
(90)
Here, the contour below is closed and the following result is obtained
U jg,1 Ex

C
ja1g,1 C C
C

C
1

j a1g,1 C C j
.

(91)

Now new quantities are defined

w

R1 a1g,1

w

, R 2 a2 g,2

(92)

Since these expressions are special cases of (74) for the present geometry, they are interpreted
as generalized channel reluctances. Re-writing (91) as
U
C 1
w 1

R1 C C ,

(93)

and inserting (87), (89a,b) into (90), the following is obtained

0 R11 C C R 21 C C

R11 R 21 C C 2R 21C .

(94)

The second line in (94) makes it possible to express C+ C in (93) by C alone, and thereby
via (88) by the current I. The final result is
Z

U
1

j
.

I
R1 R 2
w 2

(95)

Now 11 in this simple case, which implies that T R R1 R 2 , and so (95) is of the
form (73) with
j
j 1
R
Z int

and
Z int
.
(96)
2w
2w
R
coincides with the limit g,1, g,2 0 of j ( T R ) 1
Furthermore, the expression for Z int
and as a consequence the expression for Zint is the exact internal conductor impedance for the
simple geometry of Fig. 35. It does contain the inverted factor as in (79) but, in contrast to
(80), no additional factor of 2 in the denominator of the tanh argument. The physical

54
explanation for this is that even in the limit g,1, g,2 0, the conductor in Fig. 35 is not
embedded in a perfect diamagnetic medium from all sides, but magnetic conductors to the left
and right are still present. These absorb all flux in y direction, and so there are no skin layers
on the lateral vertical faces of the conductor in this case.
The above argument can be generalized to an arbitrary number of conducting slabs,
stacked onto each other in z direction and separated by insulating sheets (see Appendix A).
Again the results appear to be consistent with Eqs. (73)(80), which shows that there is no
additional interference between the individual flux channels apart from the one described by
the factors in (73) and (74), at least in this specific geometry (and presumably in general).

8.5 Extension to Variable Gap Width

So far it has been assumed that the gap widths ai between conductors are constants. In general
they may vary along any given channel, i.e., in y direction of Fig. 33, Fig. 34. However, it
may still be possible to find simple analytical expressions for the complex reluctances R i ; for
instance, under the assumption that the length scale of this variation is large compared to the
gap width itself, the only change in the basic expression (74) is that the gap width becomes a
function ai(y) and the factor wi is replaced by an integral with respect to y.
Some basic cases are shown below in Fig. 37. For simplicity the subscript i labelling the
gap has again been omitted. Along with the earlier case (a) of parallel conductor surfaces (=
constant gap width), (b) flat surfaces which are slightly inclined with respect to each other,
and (c) locally cylindrical conductor surfaces with given radii of curvature r+ , r are
considered.

amax

y
R

R
r

amin

amin
(a)

a( y )

(b)

(c)

Fig. 37: Some basic geometries of the gap (white space) between neighbouring conductors (shaded). The
channels are vertically oriented here. The current direction (x-direction) is perpendicular to the plane, as before.

(a) Planar, parallel conductor surfaces: This is the basic case (74) discussed earlier where
a = const and so
w

R ag

(97)

55
(b) Planar but inclined conductor surfaces: Here a varies along the gap as
a(y) = amin + (amax amin)y/w for 0 < y < w, and so
1

w

1

w K2
R dy a( y ) g
ln 1

0
K 2
K1

with
K1

amin g ,

K 2 g amax amin .

(98ac)

When amin, amax are equal, this reduces to case (a).

(c) Locally cylindrical conductor surfaces with curvature radii r+ , r : The largest
contribution to the channel reluctance comes from its narrowest region, which is assumed to
be located approximately in the middle. The origin of y is chosen at that point, so that
y2 1 1
a ( y ) amin in the vicinity and
2 r r
1
w2
2arctan( w2 KK2 )

1

1
R dy
a ( y ) g


w 2
K1K 2

with
K1

amin g ,

K2

g 1

1
.
2 r r

(99ac)

One or both of the radii r+ , r may be infinite. When both are infinite, (99ac) again reduces
to case (a). Alternatively, one of the radii r+ , r may be negative, corresponding to a situation
where both surfaces are bent in the same direction (examples are the gaps between conductors
and shield in Fig. 39 below). Note that as long as the flux channel is tight enough in the
middle (i.e., for those frequencies for which |K1| w2 K2), the arctan above may be replaced
by /2 so that the total channel reluctance does not depend on the length w of the channel
anymore but is determined by width and curvature radii of this high-reluctance bottleneck
only.
From the formal derivation in Sections 8.3 and 8.5 it is apparent that the RNM should be
asymptotically exact in the limit where all lateral dimensions characterizing a given flux
channel i (i.e., the maximum gap width ai,max and the skin depths ,i) are much smaller than
both its length wi , the typical scale at which ai(y) varies, and the lateral conductor thicknesses
d,i . Roughly speaking, it is expected to be exact in the limit of high frequencies and narrow
gaps between the conductors, which is precisely the limit where FEM calculations are most
difficult to perform. This will be illustrated below for two simple example geometries.

56

8.6 Comparison with FEM Calculations

The predictions of the extended reluctance network method, summarized by Eqs. (73)(80)
and (97)(99ac) above, have been verified by comparison with finite-element field
computations, using the commercial FEM tool COMSOL.
Comparisons were made for two different test geometries, which are illustrated below. The
first example is a 33-matrix of identical rectangular conductors, surrounded by a rectangular
magnetic shield (Fig. 38). The shield thickness exceeds the magnetic skin depth within it at all
frequencies considered here (1 Hz to 10 MHz), so there never is any magnetic flux in the
outer space surrounding the shield.
The second example is a symmetric arrangement of seven identical round conductors,
surrounded by a magnetic shield with a circular inner cross section boundary, so that a
number of insulator-filled pockets are created between the conductors and towards the shield
(Fig. 39).
In both examples the values k = 6107 S/m and k = 0 for all conductors, g,i = 0 for all
gaps, and shield = 1000 0, shield = 107 S/m for the shield have been used. These values were
chosen so as to roughly represent copper conductors with an iron shield or armour, but the
precise numbers are unimportant for the points made here. In all examples shown below,
inductances or resistances per length have been plotted.

a'

w"

w'

d'

k=1

d"

a"
3

shield
Fig. 38: Cross section of rectangular-conductor arrangement (not to scale). Conductors and shield are shaded,
the reluctance network is shown in blue. In this example, a' = a" = 0.2 mm, d' = 7.3 mm, w' = d' + a' = 7.5 mm,
d" = 3.3 mm, w" = d" + a" = 3.5 mm, and the shield is much thicker than the skin depth at all considered
frequencies.

57

r"

k=1
2

3
4

r'
w'

a'

r'

a'

5
w"

a"

shield
Fig. 39: Cross section of circular-conductor arrangement (not to scale). Conductors and shield are shaded, the
reluctance network is shown in blue. In this example, a' = a" = 0.05 mm, r' = 2.6 mm, r" = 3r' + a' + a" = 7.9 mm,
w' = 1.4 r' = 3.6 mm, w" = 3.5 r' = 9.1 mm, and the shield is much thicker than the skin depth at all considered
frequencies.

8.6.1 Results for Rectangular Conductors

Fig. 40 presents a comparison between RNM and FEM results for the geometry of Fig. 38
with rectangular conductors. Graph (a) shows the self inductance and resistance of conductor
1 (divided by ), whereas (b) and (c) show its mutual inductances and resistances with
conductors 2 and 9, respectively. Since the magnetic coupling between the conductors is very
strong due to the high-permeability shield, self and mutual inductances have almost the same
values. For that reason, differences L11 L12 and L11 L19 have been plotted here (which can
be interpreted as differential mode inductances) instead of L12 and L19 themselves. As a
consequence, the plotted inductance curve displays somewhat higher values for the pair of
distant conductors in (c) than for the pair of close conductors in (b), although the actual
mutual inductance is of course larger in the latter case. Similar calculations have been carried
out for all self and mutual impedances; the graphs all look very similar and therefore only a
selection is reproduced here.
The deviation between RNM and FEM at high frequencies (above ca. 105 Hz) is caused by
an insufficient resolution of the FEM mesh, particularly in the high-permeability shield where
the skin depth is very small. The circles represent the FEM results with quadratic order
elements and a mesh size of 68912 elements (about the highest resolution which the employed
desktop computer could handle), whereas the small dot symbols in (a) and (b) represent FEM
results with a lower resolution (37860 elements). The convergence of the FEM results with
increasing mesh resolution towards the RNM prediction shows that the latter is accurate at
high frequencies.

58
At high frequencies, both diagonal and off-diagonal elements of the R matrix are
dominated by circulating currents in the short-circuited high-permeability shield, and so are
almost identical. Due to the skin effect they grow as 1 2 with increasing frequency. At very
high frequencies, self and mutual inductances level off to their asymptotic limits
corresponding to complete flux expulsion from conductors and shield.
At lower frequencies, between 102 and 103 Hz, the mutual inductance curves have a knee
caused by the factors in (74). Fig. 40 thus confirms that the extension (74) of (71) is quite
accurate at low frequencies. It is not exact, though, but only an adequate approximation in that
frequency regime. Below the knee the skin depths are larger than half the conductor
thicknesses, so that the flux completely penetrates the conductors in that frequency region.
The self inductance L11 continues to grow proportionally to 1 2 with decreasing frequency,
since it is dominated by the flux in the growing skin layer in the magnetic shield. The
resistance R11 levels off towards the DC resistance value (81) of about 0.7 m/m, whereas all
off-diagonal elements of the R matrix tend to zero.

L11/

10

R11 /

-5

-1

10
-6

deviations due to
finite mesh

10

-2

10
-7

10

Rdc /

Resistance per length ( /m)

Inductance per length (H/m)

10

-3

10

-8

10

10

10

10

10
10
Frequency (Hz)

10

10

(a)

59

10

Inductance per length (H/m)

-1

-7

10

10

deviations due to
finite mesh

-2

10

-3

Resistance per length ( /m)

R12 /

L11 L12 /

10
-8

10

10

10

10

10
10
Frequency (Hz)

10

10

(b)

10
-6

10

Inductance per length (H/m)

-1

10

deviations due to
finite mesh

-7

10

-2

10

-3

Resistance per length ( /m)

R19 /

L11 L19 /

10

10

10

10

10
10
Frequency (Hz)

10

10

(c)

Fig. 40: Comparison between RNM and FEM calculation for the geometry of Fig. 38 with rectangular
conductors. A selection of matrix elements of the L and R matrices obtained from Eq. (73) are plotted as
functions of frequency in the range 1 Hz10 MHz. Full/dotted line: resistance/ inductance per length according
to RNM. Dots/circles: results from FEM calculations with rough/fine mesh.

60
8.6.2 Results for Round Conductors

Fig. 41 presents a comparison between RNM and FEM results for the geometry of Fig. 39
with round conductors. The approximation (99ac) has been used to calculate the reluctances
for RNM. The appropriate choice of the channel lengths is not as natural as in the previous
example; the values w' = 1.4 r' = 3.6 mm and w" = 3.5 r' = 9.1 mm have been chosen here,
which are of the right order of magnitude and lead to a satisfactory agreement between RNM
and FEM. Due to the arctan in the expression for the reluctance, the results are not too
sensitive to the exact choice.
As before, graph (a) shows the self inductance and resistance of conductor 1 (divided by
), whereas (b) and (c) show its mutual inductances and resistances with conductors 2 and 7,
respectively. Similar observations as in Section 8.6.1 can be made; the overall agreement is
good also in this case, although the low-frequency agreement for the mutual inductance L12 in
(b) is somewhat inferior to that for the rectangular conductors, due to the additional
approximations made to treat variable gap widths.
The high-frequency deviations between RNM and FEM are less pronounced than for the
rectangular conductors, probably because the channels are much wider on the average than in
the previous example and therefore do not require as fine a mesh resolution. The channels are
narrow only in the vicinity of the conductor-conductor and conductor-shield near-contact
points, making it possible to achieve a higher mesh resolution in these regions with the
available computing resources. In this case, the DC resistance of a single conductor is about
0.8 m/m.
-4

10

10

R11/

L11/

-1

10

deviations due to
finite mesh

-6

10

-2

10
-7

Resistance per length ( /m)

Inductance per length (H/m)

-5

10

10

Rdc/
00

10

10

-3

10

10

10
10
Frequency (Hz)

10

10

(a)

61

Inductance per length (H/m)

R12/
(L11 L12)/

-1

10
-7

10

deviations due to
finite mesh

-2

10

Resistance per length ( /m)

10

-3

10

10

10

10

10
10
Frequency (Hz)

10

10

(b)

10

R17/

-1

10

deviations due to
finite mesh

-7

10

-2

10

Resistance per length ( /m)

Inductance per length (H/m)

(L11 L17)/

-3

10

10

10

10

10
10
Frequency (Hz)

10

10

(c)

Fig. 41: Comparison between RNM and FEM calculation for the geometry of Fig. 39 with round conductors. A
selection of matrix elements of the L and R matrices obtained from Eq. (73) are plotted as functions of
frequency in the range 1 Hz10 MHz. Full/dotted line: resistance/ inductance per length according to RNM.
Symbols: results from FEM calculations.

62

8.7 Conclusions about RNM

In this chapter, it has been shown that for calculating L and R matrices of a multi-conductor
transmission line, the extended reluctance network method is in quite good agreement with
FEM down to the lowest frequencies, and it is even superior to the latter at high frequencies.
It is expected to become exact in the limit of high frequencies and narrow gaps between the
conductors, which is precisely the difficult limit for FEM calculations.
The challenge for applying this method to a transformer winding is, as it was stated earlier,
to find an adequate reluctance representation for the areas outside the turns, i.e. the regions
where there is no well-defined gap. Examples are the area between the LV and HV windings,
or between a winding and the transformer tank. A possible solution could be a combination of
FEM and RNM where the reluctances for the outer regions are calculated using FEM and
implemented in the RNM network and this is left for future work.
First attempts to apply the RNM technique to transformer windings have indicated that the
resonance frequency shifts observed in Fig. 18Fig. 21 are mainly due to the proximity effect
discussed above, so the old description without proximity effect is still considered to be
qualitatively correct. Furthermore, there is an overall satisfactory agreement between model
and measurements in Fig. 18Fig. 21. Hence, the winding calculations in the rest of this thesis
will continue to be based on the equations derived in Section 4.4 for the calculation of
inductances and resistances.

63

9 Time Domain Reflectometry (TDR)

Time Domain Reflectometry (TDR) has been used for localization of discontinuities in
transmission lines in diverse applications. In this chapter, its potential as an alternative to
Frequency Response Analysis (FRA) for the detection and localization of mechanical changes
and damages in transformer windings is investigated. It is suggested that the idea of TDR can be
used to visualize the results of frequency response measurements in time domain, where they
are easier to interpret. What is reported in this chapter is a description of the content in [5253].

9.1 Combination of FRA and TDR

As it was stated in Chapter 2, FRA is a powerful method for characterizing a system,
analyzing its response to an electrical stimulus of varying frequency. It is a popular method
for assessing the mechanical integrity of windings in power transformers and electrical
machines. Since measurements are carried out in frequency domain, their reproducibility and
signal to noise ratio are very good. Its main disadvantage is however the fact that despite a
wealth of experimental and analytical investigations, the interpretation of the FRA results has
neither been standardized [54] nor fully agreed among researchers yet (e.g., [5, 41, 5558]).
There is no general consensus about how the deviations in the FRA spectrum are correlated to
different types of mechanical damages, and to their location along the winding (see for
instance [13, 5964]). Consequently, one finds in the literature different types of transfer
functions (in-impedance, transfer-impedance, voltage transfer ratio, etc.), different
measurement techniques, and different ways of interpreting FRA results.
Time Domain Reflectometry (TDR), on the other hand, is a well-established method for
detecting and locating faults in transmission lines and cables by observing reflections of
injected signals in time domain. Its advantage is the easy and intuitive interpretation of the
measurement data.
In this Chapter, an attempt is made to combine the advantages of both the FRA and TDR
methods, as a step towards an easier and more reliable detection and localization of faults in
transformer windings. This new method is based on the principles of TDR, in which
frequency response measurements are visualized in the time domain, in order to facilitate their
interpretation.

9.2 Basic Ideas behind the TDR Technique

TDR has traditionally been used as a technique for determining the characteristics of
transmission lines by observing reflections of injected signals. It is a powerful tool for the
analysis of electrical or optical transmission media such as coaxial cables [65] and optical fibers
[66], or for the measurement of soil characteristics in geology and soil science [67]. The
principle of TDR consists in injecting a pulse or transient signal of some kind (like a step signal
or wave packet) into one end of the transmission line, recording reflections of it at later times,
and analyzing their delay and possibly changes in shape. When the injected signal reaches the
end of the transmission line or any point of change of wave impedance along it, all or part of the

64
signal is reflected back. By analyzing the magnitude, duration and shape of the total reflected
signal, the nature of the wave impedance variation can in some cases be determined. For
instance, in case of a single impedance discontinuity, the size of the change can be determined
from the magnitude of the reflected signal, and its location from the delay of the reflection
relative to the original signal.
The potential of the TDR method for transformer winding faults diagnostics has, to the
authors knowledge, so far not been systematically analyzed in the literature. In this chapter, an
attempt is made to provide such an analysis. The main problem is the following: Since
transformer windings usually are intrinsically inhomogeneous (due to their layer or disc
structure, their division into main and tap winding, to taps and interleaving, etc.) and since signal
reflection occurs at all changes of wave impedance, TDR as defined above would not only detect
reflections from mechanical damages and deteriorations, but also those from normal geometrical
irregularities in the winding. The latter can be removed by considering the difference between
TDR measurements before and after mechanical changes have occurred. In other words, the
pathological reflections from the mechanical changes are separated from other reflections that
are normal; the difference between the TDR measurements contains information only about
the mechanical changes. This procedure is thus comparative in the same sense as the FRA
method above, and will be called differential TDR (DTDR) here.

9.3 Simplified Model of a Transformer Winding and the Effect of

Dispersion
In TDR, it is considered to be an advantage to be able to send in a narrow pulse trough the
injection-point of the system which travels forth in the conductor, reaches its end, and travels
back to the injection-point without any serious dispersion. Dispersion in this context means that
components with different wavelengths travel at different speeds and thus a pulse gets smoothed
out with time. Any dispersion in the system is undesired since the more narrow and localized the
pulse remains while travelling along the conductor all the way forth and back, the more narrow
and localized will the reflections due to any possible mechanical change (which manifest itself as
a impedance change) be, which in turn makes it easier to find the exact position of the presumed
mechanical change along the winding. Further, another possible problem with a significant
dispersion would, in addition to the smoothening of the pulse, be a pronounced loss of pulse
magnitude, which could lead to the reflected pulse getting too weak to be measurable.
When the potential of the TDR method for diagnostics of transformer winding integrity is
investigated, the differences between windings and cables have to be considered. A crucial
difference between a cable, which is a specific kind of a transmission line, and a transformer
winding is the fact that for a transmission line, ideally each infinitesimal segment only has a selfinductance and a capacitance to ground, whereas for a winding, due to the fact that the conductor
is winded tightly, there will be mutual inductances between all turns and capacitances between
neighboring turns in addition to the self inductances and ground-capacitances.
Using a lumped-element approach, a lossless transmission line can be modeled by a ladder
network consisting of self-inductances and capacitances to ground, as depicted in Fig. 42(a),
where l and cg are inductances and ground-capacitances per unit length respectively, and x is a
differential length of the transmission line.

65

lx

lx

cgx

csx
x

i
cgx

(a)

(b)

Fig. 42: Equivalent circuits of a transmission-line (a) and of a transformer winding (b).

For a transformer winding, the lumped-element network depicted in Fig. 42(a) has to be
modified so as to include the mutual inductances between turns and the capacitances between
adjacent conductors. The lumped-element network for a simplified case, where each disc in the
HV winding is lumped together into one element, can be seen in Fig. 42(b) (the LV winding is
assumed to be open and at low potential everywhere, so that it can be replaced by ground), where
cg is the ground-capacitance of each disc and cs is the capacitance between adjacent discs, both
per unit length. Further, lx is the self-inductance of each element, and the dashed lines illustrate
their mutual inductances.
The solutions for the voltage u and current i along the transmission line and the transformer
winding are derived in [68, 69] and in Appendix B, where it is shown that the wave velocities in
the transmission line and the winding are
1
1 cs 2 2
c
v
v02 s 2 2
(100)
,
and
v
lcg cg
cg
lcg
respectively ( is the length of conductor per disc). The wave velocity in the transmission line is
independent of the frequency, but for the wave velocity in the transformer winding, it can be
seen that it has the value v0 = (lcg) only for waves of very low frequencies. As the driving
frequency increases, the velocity of propagation v decreases and vanishes at a critical
frequency cr = ( 2lcs ) beyond which no propagation is possible within the winding. If the
driving frequency exceeds the critical frequency, the wave velocity v in the latter part of
Eq. (100) becomes imaginary and can be rewritten as
v j

cs 2 2 2
v0 j .
cg

(101)

The voltage in the winding will now be of the form (see [68, 69] and Appendix B)

u u0 e j t e x / .

(102)

This indicates that above the critical frequency, no wave propagation can exist in the winding.
Instead, there is merely an exponential decay of the voltage as one progresses into the winding.
It can be concluded that for a lossless transmission line, since the wave velocity is frequency
independent (up to very high frequencies) and thus the different frequency components
propagate with the same speed, the pulse used for TDR may be chosen arbitrarily narrow in a
wide frequency range; it will travel along the line, be reflected at any impedance change, and
travel back to the observer without any change in its shape. For a transformer winding, in
contrast, the situation is entirely different since the wave velocity is frequency dependent and
hence the different frequency components of a signal travel with different speeds. Components
with frequencies higher than the critical frequency will not even propagate at all.

66
As a consequence, there may be a considerable change in the shape of the TDR-pulse as it
travels along the winding; it may lose its narrowness and magnitude, which makes the
localization of an impedance change in a winding more difficult, although it may still be
possible to a certain extent if the impulse shape is suitably chosen. Further below it will be
investigated how the effects of dispersion can be minimized by an appropriate choice of the
pulse width. But before that, the computer model used for this study is presented.

9.4 The Transformer Winding Model Used for TDR Simulations

For the simulations in this chapter, the most accurate model developed in Chapter 4 (model 3)
has been used. Also, a reasonable agreement between this model and measurements up to
20 MHz was observed in Chapter 6, which means that the model should be able to correctly
predict the propagation of pulses with frequency content up to about 20 MHz, with essentially
only a time rescaling factor of order unity due to the shift of resonance frequencies.

9.5 The Effect of Choice of Rise-Time for the Applied Voltage

The above mentioned model has been used to simulate a winding consisting of 60 discs with
the same dimensions as in the experimental set-up in Chapter 5, and with the distance from
the inner turns to ground (di) being 50 mm, the outer ground wall being neglected (do ),
and the vertical distance between two neighboring discs (ks) being 6 mm (see Fig. 9). The
impedance amplitude seen from the terminal of the winding is plotted in Fig. 43. It was
previously argued in Section 9.3 that when a narrow pulse is injected into a winding, the
different frequency components of the pulse travel with different speeds, the pulse suffering
from dispersion and getting smoothed out. The state space equations (Eq. (47)) of the model
have been solved in the time domain using an ode-solver in MATLAB, with an applied
external voltage of Gaussian shape, Vapp(t) = (10 V) exp(0.5(t / t1)2), with different values
of the parameter t1 chosen in the range between 0.4 s and 2.5 s. The Fourier transform is of
the form Vapp() exp(0.5( / 1)2) with 1 = 1/t1. The resulting absolute and normalized
voltage along the winding for different values for t1 are depicted in Fig. 44 and Fig. 45,
respectively, at an instant of time where the peak of the wave has reached about the middle of
the winding. It can be observed that the pulse amplitude and width both decrease with
decreasing t1, but when t1 is decreased below somewhere around 1 s, the pulse continues to
lose in amplitude (Fig. 44) without getting markedly narrower anymore (as seen most clearly
in Fig. 45). The bell-shaped Fourier transform of the applied voltage Vapp() decreases rapidly
with frequency; the major frequency content of the applied voltage Vapp(t) is confined to
frequencies below about 21. For t1 = 1 s and 1 = 1/t1, the significant part of the frequency
content of Vapp(t) is located below about 3 105 Hz. Thus, by inspecting the in-impedance of
the winding shown in Fig. 43, it can be concluded that the regular series of resonances (the
local minima) of the impedance extending up to around 3 105 Hz are the ones building up
the travelling pulse in the winding. It can also be observed that the three or four first
resonances are the most pronounced ones.
The normalized voltage along the winding for an even shorter applied voltage pulse (which
contains more high-frequency components) is depicted in Fig. 46, where it can be observed
that the higher frequency components introduced in the system only cause local oscillations
along the winding. These are internal oscillations within individual discs which were called
azimuthal resonances in Section 7.1.2, located at frequencies above 10 MHz and causing the

67

10

10

10

10

10
10
Frequency [Hz]

10

Fig. 43: Simulated in-impedance amplitude for a winding with sixty discs.
The voltage along the winding

10
9
8

t1 = 1 s

t1 = 2.5 s

7
Voltage (V)

Impedance amplitude [ ]

most pronounced impedance minima in both simulations and measurements (see Fig. 18 to
Fig. 21 and Fig. 43).
Hence it can be concluded that when applying TDR to a transformer winding, the
frequency content of the applied voltage should preferably be confined to the frequency range
where the first few pronounced resonances of the winding in-impedance occur, to get the
pulse as steep as possible without an excessive loss of magnitude. Also, the simulation models
need to be valid only in a frequency range which covers the first group of resonances,
implying that every disc in the winding can be lumped together into one element of the model.
On the other hand, localization of damages with a higher accuracy than this minimum pulse
width (which in this case is of the order of 10 discs) will not be possible.

6
5
4

t1 = 0.4 s

3
2
1
0
0

10

20

30
disc number

40

50

Fig. 44: The voltage along the winding for different values for t1.

60

68

The normalized voltage along the winding

1

t1 = 2.5 s

Normalized voltage

0.8

t1 = 1 s

0.6
0.4
0.2
0

t1 = 0.4 s
-0.2
0

10

20

30
disc number

40

50

60

Fig. 45: The normalized voltage along the winding for different values for t1.
The normalized voltage along the winding
1
0.8

0.03
0.02

0.6
Normalized voltage

Very short and

high-frequent
applied voltage

0.01

0.4

0
0

0.2
0
-0.2
-0.4
0

10

20

30
disc number

40

50

60

Fig. 46: The normalized voltage along the winding showing azimuthal oscillations

9.6 Simulations with and without Winding Damages

The same winding configuration as in the previous section has been simulated in the time
domain, before and after a set of different damages were introduced. A step voltage
(somewhat smoothed, i.e., with a finite rise time of the order of 1 s) was applied on one end
of the winding, while the other winding end was grounded and the injection-point current was

69

decreased space

increased space

ground

increased
space

ground

recorded. The reason why a step voltage was used instead of a pulse voltage as in Section 9.5
is explained at the end of this section.
The introduced damages were of three types (radial buckling, conductor bending and loose
winding conductors), and the location of each fault was varied along the winding. Radial
buckling (RB), caused by the radial forces acting on the winding during a short circuit fault, is
a compression of the inner winding (usually the LV winding) or part of it, leading to a local
change in the ground capacitance of the HV winding [5]. In the model, this implies a decrease
in Cig for those discs of the HV winding which face the buckled region of the grounded LV
winding (Fig. 47(a)). In the examples below it is assumed that buckling is restricted to one
corner of one disc of the square winding, resulting in a doubled distance between the HV
winding and ground for that corner, which corresponds to a local decrease of Cig to half of its
original value.
Loose Winding Conductors (LWC) and Conductor Bending (CB) are instead caused by
axial forces compressing the winding, and may also occur in one or several winding sections.
The axial forces can for instance result in a temporary tilting of the conductors, leaving the
turns in one or several discs loose when the forces are removed, which in turn leads to a local
decrease in Ctt (Fig. 47(b)). They may also bend the winding conductors in the region between
the radially placed insulation spacers, resulting in an average decrease in separation distance
between two neighboring discs and hence a local increase in Cdd (Fig. 47(c)). For the damaged
sections, it is assumed that the distance between the turns in a disc is increased to half of the
conductor width for the LWC case, and that the separation distance between two discs is
reduced by a factor of for the CB case. These changes correspond to a local reduction of Ctt
to about 6% of its original value or a local doubling of Cdd , respectively, in the model. Since
these relatively small geometrical changes mainly influence the capacitive distribution, the
changes in the inductances were neglected [59].

(a)
(b)
(c)
Fig. 47: (a) Radial buckling (RB), (b) loose winding conductors (LWC), and (c) conductor bending (CB).

The locations for the three different fault types were selected as follows: for the RB case, the
buckling was chosen to be close to the 10th, 30th, or 50th disc, respectively; for the LWC case,
the turns in the 10th, 30th, or 50th disc, respectively, were chosen to be loose; for the CB case,
the distance between the 10th and 11th, 30th and 31st, or 50th and 51st discs, respectively, was
reduced. The difference I = Ibefore Iafter between the currents in the injection-point before and
after introducing the different damages is plotted in Fig. 48 to Fig. 50 as a function of the time
after applying the voltage step.

70
-6

x 10

I (A )

5
0

-5

Reflection from the change in

the 10th disc
Reflection from the
winding end

0.5

1.5
time (s)

2.5

3
-5
x 10

2.5

3
-5
x 10

2.5

3
-5
x 10

-6

x 10

I (A )

5
0

-5

Reflection from the change in

the 30th disc
Reflection from the winding
end

0.5

1.5
time (s)

-6

x 10

I (A )

5
0

-5

Reflection from the change in

the 50th disc
Reflection from the
winding end

0.5

1.5
time (s)

Fig. 48: DTDR for buckling (RB) close to the 10th, 30th, and 50th disc, respectively.

71
-5

x 10

I (A)

1
0

-1
-2Reflection from the
change in the 10th
-3disc

Reflection from the

winding end

2
time (s)

4
-5

x 10

-5

x 10

I (A)

Reflection
from the
change in the
30th disc

-1
Reflection from the
winding end

-2
0

2
time (s)

4
-5
x 10

4
-5
x 10

-5

x 10

I (A)

Reflection
from the
change in the
50th disc

-1
Reflection from the
winding end

-2
0

2
time (s)

Fig. 49: DTDR for loose turns (LWC) in the 10th, 30th, and 50th disc, respectively.

72
-5

x 10

I (A)

Reflection from the

change between the
10th and 11th discs

-2
Reflection from the
winding end

-4
0

2
time (s)

4
-5
x 10

4
-5
x 10

4
-5
x 10

-5

I (A)

x 10

Reflection from
the change
2 between the 30th
and 31st discs

-2
-4

Reflection from the winding end

2
time (s)

-5

x 10

I (A)

Reflection from the

change between the
50th and 51st discs

-2
-4

Reflection from the

winding end

2
time (s)

Fig. 50: DTDR for decreased space (CB) between the 10th and 11th, 30th and 31st, and 50th and 51st discs,
respectively.

73
For the RB case (Fig. 48), the reflections have roughly the shape of single Gaussian pulses;
the reason for this is that the damages in this case are local changes of the capacitance Cig to
ground, which implies that the reflected current pulse is proportional to the time derivative of
the incoming step voltage. The reflections from damage in the 50th disc and from the winding
end partially overlap.
In Fig. 49 and Fig. 50, which represent the LWC and CB cases respectively, the reflections
have roughly the shape of derivatives of Gaussian pulses (almost the same pattern for both
cases, but with different signs and amplitudes). Similarly as above, the reason for this is that
the damages now are local changes of the series capacitance between different parts of the
winding (Ctt or Cdd), which implies that the reflected current pulse is proportional to the
second time derivative of the incoming step voltage. As above for the RB case, the more the
damage is located near the winding end, the more the reflections from the damage overlap
with those from the winding end, which makes them harder to detect; a possible remedy of
this problem might be to perform separate measurements with signal injection from either end
of the winding.
It can be concluded that for a step-like applied voltage, a localized RB damage can be
distinguished from the two other cases by its different reflection shape. A discrimination
between the LWC and CB damages is more difficult since their reflection shape is the same.
In the simulations here, their reflections had opposite signs, but if the damage for the CB case
had been an increase instead of a decrease of the inter-disc space, they would have the same
sign and thus be indistinguishable.
If the damages are sufficiently localized, their rough location (dm) can be inferred by
comparing the time delays of the reflections from the modified disc (tm) and from the winding
end (twe). Assuming that the pulse propagation speed is roughly constant along the winding,
the location of the damage can be estimated as dm dwe tm / twe where dwe is the total length
of the winding. For example, applying this relation to the second curve in Fig. 48,
dwe tm / twe = 60 (1.41 105) / (2.78 105) = 30.4 should be roughly equal to dm = 30,
which obviously is the case.
As promising as the above appears, a severe limitation may be the smallness of the
difference between responses with and without damage. Since the magnitudes of reflections
I from the changes for all three cases are of the order of 0.5 1.8 105 A, and the total
current I in the injection-point (before or after the damage) is of the order of 5 104 A, the
ratio I / I is about 14%. This means that the reflections are of the same order as the usual
absolute measurement accuracy in transformer diagnostics, and thus may be hard to detect
unambiguously. Interpretation of frequency domain FRA spectra of course suffers from the
same problem.
Finally, the question of the applied signal shape is addressed. When examining the effect
of choice of rise-time on the propagating pulse in Section 9.5 a Gaussian voltage pulse was
chosen, whereas when simulating DTDR for damage detection in the present Section a steplike shape whose time derivative is a Gaussian pulse was chosen. Since the time evolution of
the system is linear, this means that also the reflected current I(t) in the former case is
proportional to its time derivative in the latter. Whereas the information content is exactly the
same in both cases, the step-like applied voltage thus leads to reflections with fewer
oscillations (at least for the RB case), which in the authors opinion facilitates interpretation.
This is exemplified in Fig. 51 to Fig. 53, where a selected number of DTDR simulations have
been performed again, with a Gaussian voltage pulse (rise time of the order of 1 s).
Comparing the 1st graph in Fig. 48 with Fig. 51, the 1st graph in Fig. 49 with Fig. 52, and the
1st graph in Fig. 50 with Fig. 53, it is seen that the step-like applied voltage leads to reflections
with fewer oscillations.

74
-6

x 10

Reflection from the

change in the 10th disc

I (A )

5
0

-5

Reflection from the winding end

0.5

1.5
time (s)

2.5

3
-5
x 10

Fig. 51: DTDR for buckling (RB) close to the 10th disc with a Gaussian shaped applied voltage.

-5

x 10
4

Reflection from the

change in the 10th disc

I (A )

2
0

-2
-4

Reflection from the winding end

2
time (s)

th

4
x 10

-5

x 10

-5

Fig. 52: DTDR for loose turns (LWC) in the 10 disc with a Gaussian shaped applied voltage.

-5

x 10

Reflection from
the change
between the
10th and 11th
discs

I (A )

5
0

-5

Reflection from the winding end

2
time (s)

Fig. 53: DTDR for decreased space (CB) between the 10th and 11th discs with a Gaussian shaped applied voltage.

75

9.7 Conversion of In-Impedance Measurements to DTDR Signals

As already mentioned in Section 9.1, there is no general agreement in the literature about how
to interpret changes in the response characteristics of an arbitrary winding directly in the
frequency domain. If these changes are instead converted into time domain by a Fourier
transformation, interpretation along the lines of Section 9.6 becomes possible.
The procedure is as follows. The Fourier transform of the difference between currents in
the injection-point before and after damage, () = before after , is obtained as the product of
Vapp() and (), where Vapp() is the Fourier transform of the applied voltage at the
injection-point, and () = before after is the difference between the in-admittances
(inverse of in-impedances) of the winding before and after damage, measured in frequency
domain. After that, () is inverse-Fourier transformed to provide I(t). Note that the
frequency dependent in-impedance of a winding has to be measured instead of the more
conventional end-to-end FRA transfer function. From the latter the DTDR signal cannot be
deduced.
The main advantages of this procedure are summerized:
The measurements are performed in the frequency domain, leading to better
measurement reproducibility and signal-to-noise ratio than impulse measurements.
Instead of measuring the more conventional end-to-end FRA transfer function (which
requires connections from the measurement instrument to both winding ends), the inimpedance (which requires only one connection to the signal injection point) is
measured. This in turn leads to simplified measurement procedures and thus possibly to
better reproducibility.
Simple and intuitive interpretation: the type and location of possible mechanical
damages may be determined by inspection of I(t), in the way explained in Section 9.6
above. It is not known how to achieve this directly with the frequency domain data.

76

77

10 Summary, Conclusions and Future Work

In this thesis, an attempt has been made to elevate the understanding of the information
contained in FRA measurements by evaluating the FRA method for (much) higher
frequencies than what is standard and by developing a new method in which FRA and TDR
are combined. Appropriate models have been created to accomplish these tasks:
A simple, frequency dependent complex- model of magnetic core material has been
developed and adjusted to measurements. Its real and imaginary parts were compared to
measurements in a wide frequency range. The agreement was found satisfactory, especially
for higher frequencies, which makes the complex- model a very convenient starting point for
the estimation of flux distribution and losses in complicated core geometries.
Furthermore, H-B curves from the measurements, the simple complex- model and a
detailed hysteresis model were compared for different frequencies. Again the results from the
complex- model were found to agree well with measurements at higher frequencies. At low
frequencies and high field amplitudes the complex- model deviates from measurements and
detailed hysteresis model, since it does not take saturation effects properly into account. This
is, however, not expected to affect its usefulness for loss estimation.
Further, a winding model based on the lumped element approach has been developed. The
model has three steps of discretization, and these three models can be simulated in both
frequency and time domains.
When simulated in frequency domain, the models are used to predict the impedance of the
winding in a wide frequency range (which is akin to FRA). The frequency domain simulations
have been compared to measurements and the model has been verified. Two classes of
internal resonance modes of a single disc have been identified, the radial modes at high
frequencies and the azimuthal modes at even higher frequencies (above around 10 MHz for
the coils investigated in this thesis), and have been studied by measurements and model
simulations. The radial modes are characterized by a rather constant voltage amplitude within
every turn, whereas the azimuthal modes describe electrical oscillations of different parts of
the same turn against each other and therefore can only be seen in models with more than one
segment per turn. Moreover, measurements show that these modes (especially the azimuthal
modes) are highly sensitive to small changes in the disc geometry.
The application of these mode signatures to winding fault detection could be subject of
future research, but it should be kept in mind that it is extremely difficult to achieve a
sufficient measurement reproducibility on real power transformers in that frequency range.
When simulated in time domain, the models are used to predict the response of the winding
to an injected impulse (which is akin to TDR). It has been shown that the distortion of the
injected signal, caused by the effect of dispersion, can be limited to an acceptable level if the
frequency content of the applied signal is confined within the right frequency window. Also,
three different fault types with different locations along the winding have been investigated,
and it has been shown that these faults can be detected and to some extent located, and one of
the fault types can be distinguished from the two others. Furthermore, it has been shown that
frequency dependent in-impedance measurements can be numerically converted to time

78
domain signals, which allows a simple and intuitive interpretation. In future work, it remains
to apply this method to a real damaged transformer, and to investigate its potential for
detection and localization of other fault types.
Also, since the proximity effect was not included in the calculation of losses and
inductances, a reluctance network method has been presented. Its original purpose was to
reliably compute the frequency dependent inductance and resistance matrices characterizing
the transformer winding in a wide frequency range, and as a starting point, it has been
developed here for shielded multi-conductor transmission lines. The method has been
demonstrated for some simple example geometries, where its accuracy has been verified with
harmonic FEM field calculations. It remains to extend this method to the full geometry of
transformer windings in future work.

79

References
1.
2.
3.
4.
5.
6.

7.
8.
9.
10.
11.

12.
13.

W. Lech and L. Tyminski, Detecting transformer winding damage by the Low

Voltage Impulse method, Electrical Review, Vol 179, No. 21, November 1966,
pp. 768-772 (ERA Translation).
S. A. Ryder, Methods for Comparing Frequency Response Analysis
Measurements, IEEE International Symposium on Electrical Insulation, Boston,
MA USA, April 710, 2002, pp. 187190.
S. A. Ryder, Diagnosing Transformer Faults Using Frequency Response
analysis, IEEE Electrical Insulation Magazine, Vol. 19, No. 2, pp. 1622,
March/April 2003.
L. Coffeen, J. Britton, J. Rickmann, A New Technique to Detect Winding
Displacement in Power Transformers Using Frequency Response Analysis,
Power Tech Conference Proceedings, Vol. 2, 2003.
E. Rahimpour, J. Christian, K. Feser, H. Mohseni, Transfer Function Method to
Diagnose Axial Displacement and Radial Deformation of Transformer Windings,
IEEE Transactions on Power Delivery, Vol. 18, No. 2, pp. 493505, April 2003.
K. Feser, J. Christian, C. Neumann, U. Sundermann, T. Liebfried, A. Kachler, M.
Loppacher, The Transfer Function Method for Detection of Winding
Displacements on Power Transformers after Transport, Short Circuit or 30 Years
of Service, Cigr Session Paris 2000, paper No. 12/33-04.
C. Sweetser, T. McGrail, Winding Frequency Response Analysis Using Sweep
Frequency Response Analysis (SFRA) Method, IEEE SERA specification, March
2003.
J. Bak-Jensen, B. Bak-Jensen, S. D. Mikkelsen, Detection of Faults and Ageing
Phenomena in Transformers by Transfer Functions, IEEE Transactions on Power
Delivery, Vol. 10, No. 1, pp. 308314, January 1995.
J.-W. Kim, B. Park, S. C. Jeong, S. W. Kim, P. Park, Fault Diagnosis of a Power
Transformer Using an Improved Frequency-Response Analysis, IEEE
Transactions on Power Delivery, Vol. 20, No. 1, pp. 169178, January 2005.
P. T. M. Vaessen, E. Hanique, A New Frequency Response Analysis Method for
Power Transformers, IEEE Transactions on Power Delivery, Vol. 7, No. 1, pp.
384391, January 1992.
M. Bigdeli, M. Vakilian, E. Rahimpour, A New Method for Detection and
Evaluation of Winding Mechanical Faults in Transformer through Transfer
Function Measurements, Advances in Electrical and Computer Engineering, Vol.
11, No. 2, pp. 2330, May 2011.
J. Christian, K. Feser, Procedures for Detecting Winding Displacements in Power
Transformers by the Transfer Function Method, IEEE Transaction on Power
Delivery, Vol. 19, No. 1, pp. 214220, January 2004.
E. Rahimpour, M. Jabbari, S. Tenbohlen, Mathematical Comparison Methods to
Assess Transfer Functions of Transformers to Detect Different Types of

80

14.
15.
16.
17.
18.
19.

20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.

Mechanical Faults, IEEE Transactions on Power Delivery, Vol. 25, No.4,

pp. 25442010, October 2010.
T. Leibfried, K. Feser, Monitoring of Power Transformers Using the Transfer
Function Method, IEEE Transactions on Power Delivery, Vol. 14, No. 4,
pp. 13331341, October 1999.
J.R. Secue , E. Mombello, Sweep Frequency Response Analysis (SFRA) for the
Assessment of Winding Displacements and Deformation in Power Transformer,
Electric Power Systems Research, Vol. 78, No. 6, pp. 11191128, June 2008.
M. Florkowski, J. Furgal, Modelling of Winding Failures Identification Using the
Frequency Response Analysis (FRA) Method, Electric Power Systems Research,
Vol. 79, No. 7, pp. 10691075, July 2009.
E. P. Dick, C. C. Erven, Transformer Diagnostic Testing by Frequency Response
Analysis, IEEE Transactions PAS-97, No. 6, pp. 21442150, Nov.Dec. 1978.
S. M. Islam, K. M. Coatas, G. Ledwich, Identification of High Frequency
Transformer Equivalent Circuit Using Matlab from Frequency Domain Data,
IEEE Industry Application Society, Annual Meeting October 1997.
H. Tavakoli, D. Bormann, G. Engdahl, D. Ribbenfjrd, Comparison of a Simple
and a Detailed Model of Magnetic Hysteresis with Measurements on Electrical
Steel, COMPEL: The International Journal for Computation and Mathematics in
Electrical and Electronic Engineering, Vol. 28 No. 3, 2009, pp. 700710.
G. Bertotti, Hysteresis in Magnetism, Academic Press, San Diego, 1998.
D. Ribbenfjrd, Electromagnetic Transformer Modelling Including the
Ferromagnetic Core, Doctoral Thesis in Electrical Systems, Stockholm, Sweden,
2010.
D. Ribbenfjrd, G. Engdahl, Novel Method for Modelling of Dynamic
Hysteresis, IEEE Transactions on Magnetics, Vol. 44, No. 6, pp. 854857, June
2008.
J. D. Jackson, Classical Electrodynamics, third edition, John Wiley & Sons,
Inc., New York, 1998.
R. L. Stoll, The Analysis of Eddy Currents, Clarendon press, Oxford, 1974.
A. Bergqvist, A Phenomenological Differential-Relation-Based Vector Hysteresis
Model, Journal of Applied Physics, Vol. 75, No. 10, pp. 54845486, May 1994.
A. Bergqvist, Magnetic Vector Hysteresis Model with Dry FrictionLike
Pinning, Physica B: Condensed Matter, Vol. 233, No. 4, pp. 342347, 1997.
F. de Len, A. Semlyen, Complete Transformer Model for Electromagnetic
Transients, IEEE Transactions on Power Delivery, Vol. 9, No. 1, pp. 231239,
January 1994.
E. Barbisio, F. Fiorillo, C. Ragusa, Predicting Loss in Magnetic Steels Under
Arbitrary Induction Waveform and With Minor Hysteresis Loops, IEEE
Transactions on Magnetics, Vol. 40, No. 4, pp. 18101819, July 2004.
H. Tavakoli, D. Bormann, G. Engdahl, High Frequency Oscillation Modes in a
Transformer Winding Disc, Studies in Applied Electromagnetics and
Mechanics, Vol. 34, pp. 329337.
K. Ragavan, L. Satish, An Efficient Method to Compute Transfer Function of a
Transformer from Its Equivalent Circuit, IEEE Transactions on Power Delivery,
Vol. 20, No. 2, pp. 780788, April 2005.
K. G. N. B. Abeywickrama, Y. V. Serdyuk, S. M. Gubanski, Exploring
Possibilities for Characterization of Power Transformer Insulation by Frequency
Response Analysis (FRA), IEEE Transactions on Power Delivery, Vol. 21, No.
3, pp. 13751382, July 2006.

81
32.
33.

34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.

P. I. Fergestad, T. Henriksen, Transient Oscillations in Multiwinding

Transformers, IEEE Transactions on Power Systems and Apparatus, Vol.
PAS93, No. 2, pp. 500509, March 1974.
A. Miki, T. Hosoya, K. Okuyama, A Calculation Method for Impulse Voltage
Distribution and Transferred Voltage in Transformer Windings, IEEE
Transactions on Power Apparatus and Systems, Vol. PAS97, No. 3, May/June
1978.
Y. Shibuya, Sh. Fujita, High Frequency Model of Transformer Winding,
Electrical Engineering in Japan, Vol. 146, No. 3, 2004.
F. W. Grover, Inductance Calculations, Dover Phoenix Editions, 2004.
David J. Griffiths, Introduction to Electrodynamics, 3rd Edition, Pearson
Education, 2008.
R. M. D. Vecchio, B. Poulin, P. T. Feghali, D. M. Shah, R. Ahuja, Transformer
Design Principles, Taylor & Francis, New York, 2002.
David K. Cheng, Field and Wave Electromagnetics, Addison-Wesley Publishing
Company, 1983.
Guishu Liang, Haifeng Sun, Xile Zhang, and Xiang Cui, Modeling of
Transformer Windings under Very Fast Transient Overvoltages, IEEE
Transactions on Electromagnetic Compatibility, Vol. 48, No. 4, November 2006.
C. Paul, Analysis of Multiconductor Transmission Lines, John Wiley & Sons,
New York, September 1994.
M. Florkowski, J. Furgal, Detection of Transformer Winding Deformations based
on the Transfer FunctionMeasurements and Simulations, Measurement Science
and Technology, Vol. 14, pp. 19861992, 2003.
www.omicronlab.com
www.nordtrafo.se
D. Bormann, H. Tavakoli, Reluctance Network Treatment of Skin and Proximity
Effects in Multi-Conductor Transmission Lines, accepted for publication in IEEE
Transactions on Magnetics.
D. Bormann, H. Tavakoli, Reluctance Network Theory of Skin and Proximity
Effects in Strongly Coupled Multi-Conductor Transmission Lines, submitted to
IEEE Transactions on Power Delivery.
Nilanga Abeywickrama, Effect of Dielectric and Magnetic Material
Characteristics on Frequency Response of Power Transformers, Doctoral Thesis,
Chalmers University of Technology, Gothenburg, Sweden 2007.
J. Turowski, M. Turowski, M. Kopec., Method of Three-Dimensional Network
Solution of Leakage Field of Three-phase Transformers, IEEE Transactions on
Magnetics, Vol. 26, No. 5, pp. 2911-2919, Sept. 1990.
See any theoretical textbook on electrical engineering or on electromagnetic wave
propagation in dissipative media, for instance [49] p. 268, [50] p.438, or [51]
p. 204.
Karl Kpfmller, Einfhrung in die theoretische Elektrotechnik, 6th ed.,
Springer Verlag Berlin/Gttingen/Heidelberg, 1959.
Richard B. Adler, Lan Jen Chu, Robert M. Fano, Electromagnetic Energy
Transmission and Radiation, John Wiley & Sons, Inc., New York, 1960.
A. Greenwood, Electrical transients in power systems, 2nd ed., John Wiley,
1991.
H. Tavakoli, D. Bormann, G. Engdahl, Time Domain Reflectometry for Fault
Localization in Transformer Windings, submitted to International Journal of
Applied Electromagnetics and Mechanics.

82
53.
54.

55.
56.
57.
58.

59.

60.

61.
62.
63.

64.
65.
66.
67.

H. Tavakoli, D. Bormann, G. Engdahl, Fault Localization in Transformer

Windings using Time-Domain Representation of Response Functions, submitted
to European Transactions on Electrical Power.
P. Picher et al., Final Report of Cigr WG A2.26, Mechanical Condition
Assessment of Transformer Windings using Frequency Response Analysis
(FRA), CIGRE Technical Brochure No. 342, April 2008. (67 pages) [ISBN 9782-85873-030-8]
L. Satish, S. K. Sahoo, An Effort to Understand What Factors Affect the Transfer
Function of a Two-Winding Transformer, IEEE Transactions on Power Delivery,
Vol. 20, No. 2, pp. 14301440, Apr. 2005.
Zhongdong Wang, Jie Li, D. M. Sofian, Interpretation of Transformer FRA
Responses Part I: Influence of Winding Structure, IEEE Transactions on Power
Delivery, Vol. 24, No. 2, pp. 703710, April 2009.
D. M. Sofian, Zhongdong Wang, Jie Li, Interpretation of Transformer FRA
Responses Part II: Influence of Transformer Structure, IEEE Transactions on
Power Delivery, Vol. 25, No. 4, pp. 25822589, October 2010.
J. A. S. B. Jayasinghe, Z. D. Wang, P. N. Jarman, A. W. Darwin, Winding
Movement in Power Transformers: A Comparison of FRA Measurement
Connection Methods, IEEE Transactions on Dielectrics and Electrical Insulation,
vol. 13, No. 6, pp. 13421349, December 2006.
P. Karimifard, G. B. Gharehpetian, S. Tenbohlen, Localization of Winding Radial
Deformation and Determination of Deformation Extent Using Vector FittingBased Estimated Transfer Function, European Transactions on Electrical Power,
Vol. 19, No. 5, pp. 749762, July 2009.
E. Rahimpour, D. Gorzin, A New Method for Comparing the Transfer Function
of Transformers in Order to Detect the Location and Amount of Winding Faults,
Electrical Engineering (Archiv fur Elektrotechnik), Vol. 88, No. 5, pp. 411-416,
May 2005.
S. K. Sahoo, L. Satish, Discriminating Changes Introduced in the Model for The
Winding of a Transformer Based on Measurements, Electric Power Systems
Research, Vol. 77, No. 7, pp. 851858, May 2007.
K. Ragavan, L. Satish, Localization of Changes in a Model Winding Based on
Terminal Measurements: Experimental Study, IEEE Transaction on Power
Delivery, Vol. 22, No. 3, pp. 15571565, July 2007.
P. Karimifard, G. B. Gharehpetian, S. Tenbohlen, Determination of Axial
Displacement Extent Based on Transformer Winding Transfer Function
Estimation Using Vector-Fitting Method, European Transactions on Electrical
Power, Vol. 18, pp. 423436, May 2008.
P. Karimifard, G. B. Gharehpetian, A New Algorithm for Localization of Radial
Deformation and Determination of Deformation Extent in Transformer Windings,
Electric Power Systems Research, Vol. 78, pp. 17011711, 2008.
R. Eriksson, R. Papazyan, G. Mugala, Localization of Insulation Degradation in
Medium Voltage Distribution Cables, First International Conference on
Industrial and Information Systems, pp. 167172, 811 August 2006.
Y. Sato, K. Aoyama, Optical Time Domain Reflectometry in Optical
Transmission Lines Containing InLine ErDoped Fiber Amplifiers, Journal of
Lightwave Technology, Vol. 10, issue 1, pp. 7883, January 1992.
A. Cataldo, L. Tarricone, F. Attivissimo, A. Trotta, A TDR Method for
RealTime Monitoring of Liquids, IEEE Transactions on Instrumentation and
Measurement, Vol. 56, issue 5, pp. 16161625, October 2007.

83
68.
69.

R. Rdenberg, Performance of Travelling waves in Coils and Windings, AIEE

Transactions, Vol. 59, pp. 10311040, December 1940.
R. Rdenberg, Electrical Chock Waves in Power Systems, Harvard University
Press, Cambridge, Massachusetts, 1968.

84

85

Appendix A
RNM for a Stack of Slabs
In this Appendix, the reluctance network formulas are proven to be valid also for a
configuration with several conducting slabs, stacked on each other in z direction with
insulating sheets in between. The derivations in this Appendix are borrowed from [45].
The solution for a single-slab geometry (Section 8.4.2) is with out much difficulty
generalized here to a stack containing an arbitrary number n of slabs. The slabs are labelled by
k = 1,,n and the insulating sheets by i = 0,,n, in the way shown in Fig. 54. As before, the
stack ends in a perfectly diamagnetic ( 0) layer above the uppermost insulating sheet (not
shown in the picture).
Since Eqs. (82)(89a,b) are valid separately for every slab, the interface fields H y,k and
Ex,k , the fluxes c,k and g,i, as well as the total currents Ik can again be expressed in terms of

coefficients Ck and Ck .

d2
y

a1

R ,2

g ,2

R g,2

,2

R ,2

a2

,2

I2

,1

R ,1

g ,1

R g,1

,1

R ,1

d1

a0

I1

,0

R ,0

g ,0

R g,0

I3

c,2

c,1

0
w
Fig. 54: Geometry of a stack of slabs, generalizing Fig. 35, Fig. 36. Also shown are the different flux paths in
the insulating gaps and skin layers, together with their respective reluctances.

86

Assuming as before that the connections to the conductors are made from below the stack, the
following are obtained for the voltages Uk:
U1 jg,0 Ex,1 ,

(A1)

U 2 j g,0 c,1 g,1 Ex,2 ,

(A2)

etc., so that in general

k 1

U k j g,i1 c,i jg,k 1 Ex,k .

(A3)

i 1

(By definition, the empty sum for k = 1 is equal to zero.) Using (85), (87), (89a,b) this can be
re-expressed as
k 1

U k j ai1g,i1 Ci Ci i i 2Ci
i
i 1

j ak 1g,k 1 Ck Ck j
j

k k
Ck Ck
k

k
1
Ck k .
k
k

(A4)

In the same way, the condition of zero total flux can be re-expressed as follows:
n

0 g,i1 c,i g,n

i 1

ai1g,i1 Ci Ci i i 2Ci
i
i 1

an g,n Cn Cn .

(A5)

Since the field Hy is constant across any given gap k and equal to the adjacent surface fields
H y,k and H y,k 1 , (84) implies the following matching condition for Ck and Ck :
Ck Ck H y,k H y,k 1

Ck1 Ck1 , for k 1,..., n 1 .

(A6)

Eq. (A6) allows in particular to express the terms 2Ci in the sums of (A4), (A5) in terms of
differences Ci Ci , since

2Ci Ci Ci Ci Ci

Ci Ci Ci1 Ci1 ,

so that the sums of (A4), (A5) can be re-arranged into the following forms:

(A7)

87
k

U k j w R i11 Ci Ci
i 1

j k

1
Ck k ,
k

(A8)

0 R i11 Ci Ci R n1 Cn Cn ,

(A9)

i 1

with the coefficients R i1 defined by

and

R 01

11
a0 g,0
,
w
1

R i1

ii

ai g,i i1 i1 for i 1,..., n 1 ,

w i
i1

R n1

n n
an g,n .

w n

(A10)

The R i have exactly the form of the channel reluctances defined in (74), adapted to the
particular geometry and labelling convention of the stack in Fig. 54. Furthermore, due to (88)
one can simply write
I k 2wCk .
(A11)
Finally, defining as in (96)
R
Z int,
k

j k
j k 1
k and Z int,k
,
2 w k
2 w k k

(A12)

and moving the second term on the r.h.s. in (A8) to the left, one obtains
k

R
1

U k Z int,k I k Z int,
k I k j w R i 1 Ci Ci .

(A13)

i 1

The r.h.s. of this equation is now denoted by Qk,

k

Qk j w R i11 Ci Ci for k 1,..., n ,

(A14)

i 1

and all Qk are collected in a column vector Q. In order to prove (73), it then remains to show
that

R Q j I .
T

(A15)

This can be achieved in the following way. First, it is noted that for the particular stack
geometry of Fig. 54,

88
k 1

1
1
0

0
0
1

1 0
1
0 1

0
1

(A16)

so that
R1
R 0 R1

R1 R 2
R1
0
R 2
TR

R 2

R 2 R3

R n1

.
0

R n1
R n1 R n
0

(A17)

To facilitate the argument, one can resort to the following trick. One can choose some
(arbitrary) Cn1 and Cn1 with Cn1 Cn1 Cn Cn , so that the condition of zero total flux
(A9) takes the form (A14) with Qn+1 = 0, whereby the validity of (A14) is extended to
k n 1 . Defining in addition Q0 = 0, one has for all k 0,..., n :

Qk 1 Qk j wR k1 Ck1 Ck1 .

(A18)

This in turn implies that for all k = 1,,n , the k:th component of the l.h.s. of (A15) can now
be written as
R k 1Qk 1 R k 1 R k Qk R k Qk 1
R k 1 Qk Qk 1 R k Qk 1 Qk

j w Ck Ck j w Ck1 Ck1
j w2Ck j I k ,

(A19)

where the last two equalities follow from (A6) and (A11), respectively. But the end result is
just the r.h.s. of (A15) which completes the proof of (73) for the particular stack geometry
of Fig. 54.

89

Appendix B
Wave Velocity in Transmission Lines and
Transformer Windings
In this Appendix, the relations governing the voltages and currents in a transmission line and a
transformer winding are derived and the wave velocities are obtained. These derivations are
largely borrowed from [6869]. For the transmission line ladder network in Fig. 42, if u is the
voltage in the nodes, and i is the current in the inductors, then the voltage difference between an
arbitrary node and the node before it is
u l x
which in the limit x 0 leads to

i
,
t

(B1)

u
i
l .
x
t

(B2)

Further, the difference between the currents in one inductive branch and the one before is
i cg x
leading to

u
,
t

i
u
cg
x
t

(B3)

(B4)

By differentiating Eq. (B2) and (B4) with respect to x and t respectively, and eliminating i(x,t),
one will arrive at
2u
2u
lc

,
(B5)
g
x 2
t 2
which is the wave equation for a voltage wave (a similar one can be obtained for the current
i(x,t)) propagating with the velocity
1
v
.
(B6)
lcg
For the case of a transformer winding, consider a tightly wound coil of individual pancake turns
(or discs), which are parallel and closely spaced (Fig. 55(a)).

90

cs
cg
ground

LV

cg

HV

cs
(b)

(a)

Fig. 55: Tightly wound coil consisting of individual pancake turns (a), and a transformer winding in crosssection (b).

The charging current for each element of conductor length x of the n:th turn of the winding,
which is due to the ground-capacitance and thus flows mainly to ground, is

ig,n cg x

un
,
t

(B7)

where cg is the ground-capacitance per unit length of conductor. Let cs be the disc-to-disc
capacitance per unit length between adjacent pancake discs. Then the charging current that
flows from the n:th turn to the (n 1):th turn is

is, n cs x

un un 1
un
cs x
,
t
t

(B8)

is,n cs x

un un 1
un
cs x
.
t
t

(B9)

2un

,
un un cs x
t
t

(B10)

and that to the (n +1):th turn is

The sum of these two currents,

is,n cs x

is the total disc-to-disc capacitive current per unit conductor length. Here u is the voltage
difference between adjacent turns and 2u is the difference of this difference between
successive turns. The voltage difference is thus that due to a length of conductor
x

(B11)

of one complete turn of length . Thus the second difference of voltage can be rewritten in
differential rather than difference form
2u 2

2
2u
2 u

x 2
x 2

(B12)

which is valid if the coil contains many turns and if the detailed distribution of the phenomena
along a single turn length are not of interest. The total disc-to-disc and ground capacitive

91
charging current of each turn, as given by Eqs. (B7) and (B10), is then equal the decrease in
in the current in the n:th turn. Thus, using Eq. (B12), one obtains
in cg x

un
3un
,
cs 2 x
t
tx 2

(B13)

and hence the space derivative of the current is

i
u
3u
,
cg
cs 2
x
t
tx 2

(B14)

which shows that it depends on the change of voltage not only with time but also with space.
For fast travelling pulses with steep portions, and thus pronounced changes in space along the
conductor, the second term of Eq. (B14) can become very significant, particularly if the disc-todisc capacitance cs is substantially larger than the ground capacitance cg, which is frequently the
case for tightly wound coils.
The current as well as the voltage in a long coil varies from turn to turn. If the n:th turn, of
self inductance per unit length, were separated from the rest of the coil, the voltage induced by
the current in it would be
i
u n x n .
(B15)
t
The two adjacent turns coupled by the mutual inductance m per unit length further add an
induced voltage
i i
umn mx n 1 n 1 .
(B16)
t
By subtracting in from the current in each of the adjacent turns and rearranging the terms, the
following expression for the effect of the adjacent turns is deduced:

umn mx

i
2in

.
in 1 in in in 1 2in 2mx n mx
t
t
t

(B17)

Thus the mutual inductance of the adjacent turns contributes an induced voltage that can be
thought of as consisting of two parts, one of which is proportional to effects in the n:th turn and
the other due to second differences between the currents in the turns. Each succeeding turn will
produce an induced voltage in the n:th turn that will have the same form as Eq. (B17), except
that m will become smaller with increasing distance between turns. All induced voltages
proportional to the current in in the n:th turn can be summed from Eqs. (B15) and (B17), to give

uln m x

in
i
l x n .
t
t

(B18)

Here l is the self-inductance of the coil per unit length derived from the total inductive effects
between all turns; thus it is the self-inductance of the entire coil divided by the length of wire.
In the last term of Eq. (B17) one can introduce Eq. (B11) and write

92

2i 2

2
2i
2 i

x 2
x 2

(B19)

when many turns are again used as basis for transformation from the difference to the
differential form. For this term, only the influence of the immediately adjacent turns is taken
into account, because the induced effects due to the farther ones have already been included in
Eq. (B18) via the total inductance of the coil and should not be counted again. The additional
inductive influence of the two adjacent turns is effective here, like that of the inter-disc
capacitance treated earlier. The voltage induced in the n:th turn, in accordance with Eqs. (B17)
and (B18), must be equal to the decrease un of the observable voltage difference. Using Eq.
(B19) one thus obtains
i
3in
,
(B20)
un l x n m 2 x
t
tx 2
And the voltage difference along the coil is
u
i
3i
.
l m 2
x
t
tx 2

(B21)

Again it can be seen that this depends on the variation of current with time as well as with
space. For fast travelling pulses with steep portions the third-order derivative may attain
appreciable magnitudes. However, its influence here is smaller than the corresponding
capacitive one on voltage in Eq. (B14) because the mutual inductance m between adjacent turns
is small compared with the self-inductance component l per unit turn of the entire coil. This
means that the second term in Eq. (B21) can be neglected as a simplification. This
approximation is fully explained and justified in [68] and [69]. Thus one has
3
i
u
2 u
,
cg
cs
x
t
tx 2

and

u
i
l .
x
t

(B22)

The validity of Eq. (B22) can be extended to other cases of importance. If the coil has not only
single turns as in Fig. 55(a), but a number of turns per disc as shown in Fig. 55(b), it is possible
to consider average values un and in for each disc n. Correspondingly, one can now let cg be the
ground-capacitance for each disc and cs the capacitance between adjacent discs, both per unit
length. The definition of self-inductance l is unchanged and is now the length of conductor
per disc.
By differentiating the first and second part of Eq. (B22) with respect to t and x respectively,
and eliminating i(x,t), one can write
4
2u
2u
2 u
lc
lc

0.
g
s
x 2
t 2
t 2x 2

(B23)

A similar fourth-order differential equation holds for the current i. A simple form of solution for
Eq. (B23) is
u u0 exp(j (t x / v)) ,
(B24)
By introducing Eq. (B24) into Eq. (B23) one will arrive at

93
2

lcg 2 lcs 2 2 0 .
v
v

(B25)

This shows the relation between the frequency and the velocity v of traveling waves. Eq.
(B25) gives the velocity of propagation as

1 cs 2 2
c
v02 s 2 2 .
lcg cg
cg

(B26)

94

95

List of Symbols
Symbol

Quantity

Matrix used for formulating the state space equation of the

winding model
Area of the complex- ellipse
Constant for determining the field Hz in laminate
Constant for determining the field Hz in laminate
Voltage amplitude for the Azimuthal resonances
Voltage amplitude for radial resonance k
Gap width
Gap width
Gap width
Width of gap i
Maximum width for gap i
Maximum gap width
Minimum gap width
Magnetic flux density vector
Matrix used for formulating the state space equation of the
winding model
Matrix used for formulating the state space equation of the
winding model
Magnetic flux density
Voltage amplitude for radial resonance k
Complex magnetic flux density
Average magnetic flux density in laminate
Approximaion for measured magnetic flux density
Peak value of the magnetic flux density for the complex-
ellipse
Half of the laminate thickness
Total capacitance matrix
Specific capacitance matrix for model i
Capacitance matrix for model i, for coupling between two
neighboring discs
Disc-to-disc capacitance
Disc-to-disc capacitance for model i

[F]
[F]

Winding model capacitance at high frequencies

Disc-to-ground capacitance for inner ground

[F]
[F]

A
A1
A2
Aaz
Ak
a
a
a
ai
ai,max
amax
amin
B
B

B
B
Bk
B
B
Bmeas
Bp
b
C
C(i)
i)
C(DD
Cdd
Cdd(i )
Chf
Cig

SI-Unit

[m2]
[A/m]
[A/m]
[V]
[V]
[m]
[m]
[m]
[m]
[m]
[m]
[m]
[T]

[T]
[V]
[T]
[T]
[T]
[T]
[m]
[F]
[F]
[F]

96
Cig(i )

Capacitance to inner ground for model i

[F]

Cog
Cog(i )
Ctt
Ctt(i )
C

Disc-to-ground capacitance for outer ground

Capacitance to outer ground for model i

[F]
[F]

Turn-to-turn capacitance
Turn-to-turn capacitance for model i

[F]
[F]
[A/m]

E
Ex
Ex

Constant coefficients for determining the field values in a

conductor
Constant coefficients for determining the field values in
conductor k
Degree of reversibility
Ground capacitance per unit length
Series capacitance per unit length
Electric displacement field vector
Conductor thickness
Conductor thickness
Conductor thickness
Distance between inner ground and winding
Thickness of conductor k perpendicular to the adjacent channel
i
Location of damage
Distance between outer ground and winding
Total length of winding
Thickness of the conductor above/below gap i, in perpendicular
direction
Electric field vector
x-component of electric field
x-component of electric field at slab surfaces

[V/m]
[V/m]
[V/m]

Ex,k

x-component of electric field at the surfaces of slab k

[V/m]

F
f
faz
fi
fk
G
H
H
H
Hexcess
Hmeas
Hp
Hy
H y

Force vector
Frequency
Frequency for azimuthal resonances
Frequency for measurement i
Frequency for radial resonance k
Magnetic domain structure parameter
Magnetic field vector
Magnetic field
Complex magnetic field
Magnetic field due to excess effects
Approximation for measured magnetic field
Peak value of the magnetic field for the complex- ellipse
y-component of magnetic field
y-component of magnetic field at slab surfaces

[N]
[Hz]
[Hz]
[Hz]
[Hz]
[-]
[A/m]
[A/m]
[A/m]
[A/m]
[A/m]
[A/m]
[A/m]
[A/m]

H y,k

y-component of magnetic field at the surfaces of slab k

[A/m]

Hy,0
Hz
H0
h

y-component of magnetic field at the conductor surface

z-component of magnetic field
Magnetic field strength at laminate surfaces
Conductor height

[A/m]
[A/m]
[A/m]
[m]

Ck
c
cg
cs
D
d
d
d
di
dk,i
dm
do
dwe
d,i

[A/m]
[-]
[F/m]
[F/m]
[C/m2]
[m]
[m]
[m]
[m]
[m]
[m]
[m]
[m]
[m]

97

htot
I
II
I
Iafter
after
Ibefore
before
Ij
Ik
i

[m]
[A]
[-]
[A]
[A]
[A]
[A]
[A]
[A]
[A]
[-]; [A]

ig,n
in
is,n

Total height of the winding

Current vector
Identity matrix
Current
Current in injection-point after damage
Fourier transform of Iafter
Current in injection-point before damage
Fourier transform of Ibefore
Current in segment j
Current in conductor k in a multi-conductor transmission line
Index for pseudo particles and single sheet measurements
(Chapter 3), developed winding models (Chapter 4), gaps and
reluctance branches in a multi-conductor transmission line
(Chapter 8); Current (Chapter 9)
Current in the n:th turn flowing to ground
Current in the n:th turn
Sum of is, n and is,n

is,n

Current flowing from the n:th turn to the next turn

[A]

is,n

Current flowing from the n:th turn to the previous turn

[A]

J
Jx
j
K
K1
K2
k

Current density vector

x-component of current density
Imaginary unit
Total number of discs in the winding
Reluctance factor
Reluctance factor
Pinning strength (Section 3.3.1); index for an arbitrary node
(Section 4.5), the radial resonances (Section 7.1.1), the
conductors in a multi-conductor transmission line (Chapter 8)
Inductance matrix
Inductance matrix for a single disc
Matrix for the mutual inductance between disc k and j
Self inductance of conductor k in multi-conductor transmission
line
Mutual inductance between conductor k and k in multiconductor transmission line
Inductance of a disc for low frequencies
Self inductance of a straight conductor segment
Self inductance of segment j
Inductance per unit length
Conductor length
Inner length of one side of the square disc
Outer length of one side of the square disc
Magnetization (Section 3.3.1 ); mutual inductance (Chapter 4)
An-hysteretic magnetization
Mutual inductance between segment k and j
Mutual inductance between two segments with the length p, q,
m+p, m+q, respectively
Saturation magnetization
Magnetization of pseudo particle (Section 3.3.1); mutual

[A/m2]
[A/m2]
[-]
[-]
[H]
[H]
[Am2/s]; [-]

L
Ldisc
Lkj
Lkk

Lkk
Llf
Lself
Lself,j
l

i
o
M
Man
Mkj
Mp , Mq ,
Mm+p , Mm+q
Ms
m

[A]
[A]
[A]

[H]
[H]
[H]
[H]
[H]
[H]
[H]
[H]
[H/m]
[m]
[m]
[m]
[A/m]; [H]
[A/m]
[H]
[H]
[A/m]
[A/m]; [H/m]

98

N
n
ni
np
n0
O
P
Pk
p
Q

Qk
q
R
Rdisc
Rdc
Rkk
Rkk
Rseg
Rseg,j
r
r
r
S
s
TF
T
t
t1
tm
twe
U
U
Uk
u
un
umn
u0
u n
V
Vapp(t)
Vapp()
Vaz
Vj

inductance per unit length for adjacent turns (Appendix B)

Number of measurements performed with single sheet tester
Number of turns in a disc (Chapters 4, 5, 7, 8), conductors in a
multi-conductor transmission line (Chapter 8)
Number of branches in a disc for winding model i
Number of pseudo particles
Phenomenological parameter
Matrix consisting of one column taken out from the C matrix
corresponding to index k
Matrix consisting of the k:th column taken out from T
Play-operator with the pinning strength k
Length
Column vector used to prove the validity of RNM for a stack of
slabs
k:th element of Q
Length
Resistance matrix
Resistance matrix for one disc
DC resistance
Resistance of conductor k in multi-conductor transmission line
Mutual resistance between conductor k and k in multiconductor transmission line
Resistance of a straight conductor segment
Resistance of segment j
Conductor radius
Shield inner radius
Radii of curvature
Matrix consisting of 0,1 and 1
Number of gaps and branches in the reluctance network
Transfer function vector
Duration of one period of measurement
Time
Time parameter for Vapp(t)
Time delay of the reflection from the modified disc
Time delay of the reflection from the winding end
Vector containing the voltage drops Uk
Voltage
Induced voltage drop between opposite ends of conductor k in a
multi-conductor transmission line
Voltage
Voltage in the n:th turn
Induced voltage in the n:th turn due to m
Voltage amplitude
Induced voltage in the n:th turn due to the current in itself
Vector containing the node voltages
Applied voltage
Fourier transform of Vapp(t)
Voltage distribution in a disc for azimuthal resonances
Voltage in node j

[-]
[-]
[-]
[-]
[-]
[F]
[-]
[-]
[m]
[radAH/s]
[radAH/s]
[m]
[]
[]
[]
[]
[]
[]
[]
[m]
[m]
[m]
[-]
[-]
[s]
[s]
[s]
[s]
[s]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]
[V]

99

Vk
Vrad,k
V0
v
v0
w
w
w
wi
X
x
after

before
Z
Zint
R
Zint
Z
Zint
Zint,k
R
Z int

Voltage in node k
Voltage distribution in a disc for radial resonance k

[V]
[V]

Phenomenological parameter
Wave velocity
Maximum wave velocity in a winding
Width of laminate (Section 3.3.2) and conductor (Chapter 4,
Section 8.4.2, and Appendix A )
Gap length
Gap length
Length of gap i
State vector
Vector containing parameters of eff
In-admittance of winding after damage (measured in frequency
domain)
In-admittance of winding before damage (measured in
frequency domain)
Impedance matrix for multi-conductor transmission line
True internal impedance matrix
Internal impedance matrix

[A/m]
[m/s]
[m/s]
[m]
[m]
[m]
[m]

[S]
[S]
[]
[]
[]

Impedance
True internal impedance
True internal impedance of conductor k
Internal impedance

[]
[]
[]
[]

R
Z int,
k

Internal impedance of conductor k

[]

Numerical factor for adjusting Rseg

Complex wave number for magnetic penetration in laminate
Alternative way to write v when it becomes imaginary
Matrix connecting the currents and voltages
Transpose of
Skin depth
Skin depth in the conductor above/below gap i
Permittivity
Relative permittivity of air
Relative permittivity of the conductor insulation
Permittivity of free space
Back field
Reluctance factor for skin layer
Hysteresis angle
Reluctance factor for skin layer of conductor k
Alternative way to write ,i

[-]
[m1]
[m/s]
[-]
[-]
[m]
[m]
[F/m]
[-]
[-]
[F/m]
[A/m]
[-]
[rad]
[-]
[-]

Factor similar to k ,i

[-]

Reluctance factor for the skin layer above/below gap

Reluctance factor for the skin layer above/below gap i
Propagation constant in a conductor
Propagation constant in conductor k
Propagation constant in the conductor above/below gap

[-]
[-]
[m1]
[m1]
[m1]

,i

air
i
0

h
k

k ,i
k ,i

,i

100

,i

Propagation constant in the conductor above/below gap i

[m1]

eff

eff

eff

Self inductance per unit length of one isolated turn

Number for pseudo particle i
Permeability
Effective complex permeability of lamination

[H/m]
[-]
[H/m]
[H/m]
[H/m]
[H/m]
[H/m]

g
g,i
h
k
meas

meas

meas

r
shield
0

,i

0
1

dc
g,i
k
shield
,i

i
ks

m
c
c,k

g,i

i
m,k
,i

cr
i

Real part of eff

Imaginary part of eff
Permeability of the medium in gap
Permeability of the medium in gap i
Low-frequency complex permeability including hysteresis
Permeability of conductor k
Complex permeability for approximating the measured H-B
curve with a complex- ellipse
Real part of meas
Imaginary part of meas
Low-frequency relative permeability excluding hysteresis
Permeability of shield
Permeability of free space
Permeability of the conductor above/below gap
Permeability of the conductor above/below gap i
Radial polar coordinate (Chapter 7); volume charge density
(Section 8.3.2)
Inner radius of a disc
Outer radius of a disc
Conductivity
DC Conductivity
Conductivity of the medium in gap i
Conductivity of conductor k
Conductivity of shield
Conductivity of the conductor above/below gap i
Density function
Twice the insulation thickness
Distance between two discs
Vector containing the fluxes i
Vector containing the mesh fluxes m,k
Total magnetic flux inside conducting slab
Total magnetic flux inside conducting slab k
Magnetic flux in gap i

[H/m]
[H/m]
[H/m]
[H/m]

Magnetic flux in branch i

Magnetic flux in mesh k
Magnetic flux inside the conductor above/below gap i
Azimuthal polar coordinate
Susceptibility at H = 0
Angular frequency
Critical angular frequency
Angular frequency for measurement i

[Wb]
[Wb]
[Wb]
[rad]
[-]
[rad/s]
[rad/s]
[rad/s]

[H/m]
[H/m]
[-]
[H/m]
[H/m]
[H/m]
[H/m]
[m], [C/m3]
[m]
[m]
[S/m]
[S/m]
[S/m]
[S/m]
[S/m]
[S/m]
[-]
[m]
[m]
[Wb]
[Wb]
[Wb]
[Wb]
[Wb]

101

R
R

R g,i
Ri
R ,i

T
ik
,i

0
I
()
x
()

Inverse of t1
Del operator
reluctance matrix
Reluctance
Reluctance between the turns for a single disc
Reluctance outside the turns for a single disc (placed
vertically)
Reluctance outside the turns for a single disc (placed
horizontally)
Reluctance of gap i

[rad/s]
[-]
[A/Vs]
[A/Vs]
[A/Vs]
[A/Vs]

Reluctance in gap or channel i

Skin layer reluctance in the conductor above/below gap i

[A/Vs]
[A/Vs]
[-]
[-]
[-]
[-]

connectivity matrix for the reluctance network

Transpose of
Arbitrary element of
Label for all quantities related to the conductor above (+) or
below () gap i
Zero matrix
Difference between currents in injection-point
Fourier transform of I
Differential length
Difference between the in-admittances of the winding

[A/Vs]
[A/Vs]

[-]
[A]
[A]
[m]
[S]

102

103

List of Acronyms
AC
CB
DC
DTDR
FDM
FEM
FFT
FRA
HV
LV
LWC
MIMO
ode
RB
RNM
TDR

Alternating current
Conductor bending
Direct current
Differential time domain reflectometry
Finite difference method
Finite element method
Fast Fourier transform
Frequency response analysis
High voltage
Low voltage
Loose winding conductor
Multi input multi output
Ordinary differential equation
Radial buckling
Reluctance network method
Time domain reflectometry