Documente Academic
Documente Profesional
Documente Cultură
Hanif Tavakoli
ii
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
xi
Contents
ABSTRACT
ACKNOWLEDGMENTS
LIST OF PUBLICATIONS
1
INTRODUCTION ....................................................................................................................................... 1
1.1
1.2
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
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
10
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.
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.
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
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.
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.
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:
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.
3.1 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.
B
t
H J
(3)
D
,
t
(4)
(5)
(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)
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 .
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
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.
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)
(17)
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)
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.
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
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
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
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.
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.
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.
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)
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
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.
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)
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
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).
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
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.
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.
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
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).
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
(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
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.
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.
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
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.
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)
,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
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
and
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)
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)
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).
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)
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
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.
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 /
10
-3
10
-8
10
10
10
10
10
10
Frequency (Hz)
10
10
(a)
59
10
-1
-7
10
10
deviations due to
finite mesh
-2
10
-3
R12 /
L11 L12 /
10
-8
10
10
10
10
10
10
Frequency (Hz)
10
10
(b)
10
-6
10
-1
10
deviations due to
finite mesh
-7
10
-2
10
-3
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
-5
10
10
Rdc/
00
10
10
-3
10
10
10
10
Frequency (Hz)
10
10
(a)
61
R12/
(L11 L12)/
-1
10
-7
10
deviations due to
finite mesh
-2
10
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
(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
63
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.
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.
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
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
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
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
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
0.5
1.5
time (s)
-6
x 10
I (A )
5
0
-5
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
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)
-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
2
time (s)
-5
x 10
I (A)
-2
-4
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
I (A )
5
0
-5
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
I (A )
2
0
-2
-4
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
2
time (s)
Fig. 53: DTDR for decreased space (CB) between the 10th and 11th discs with a Gaussian shaped applied voltage.
75
76
77
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.
80
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
81
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
82
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
83
68.
69.
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)
(A2)
(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
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
(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
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
(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)
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)
(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
[F]
[F]
[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 )
[F]
Cog
Cog(i )
Ctt
Ctt(i )
C
[F]
[F]
Turn-to-turn capacitance
Turn-to-turn capacitance for model i
[F]
[F]
[A/m]
E
Ex
Ex
[V/m]
[V/m]
[V/m]
Ex,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
[A/m]
Hy,0
Hz
H0
h
[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
is,n
[A]
is,n
[A]
J
Jx
j
K
K1
K2
k
[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
[-]
[-]
[-]
[-]
[-]
[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
[]
[-]
[m1]
[m/s]
[-]
[-]
[m]
[m]
[F/m]
[-]
[-]
[F/m]
[A/m]
[-]
[rad]
[-]
[-]
Factor similar to k ,i
[-]
[-]
[-]
[m1]
[m1]
[m1]
,i
air
i
0
h
k
k ,i
k ,i
,i
100
,i
[m1]
eff
eff
eff
[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
[H/m]
[H/m]
[H/m]
[H/m]
[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]
[A/Vs]
[A/Vs]
[-]
[-]
[-]
[-]
[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