TRANSFORMER MODELLING FOR TRANSIENT STUDIES

by

WASHINGTON L. A. NEVES

B. Eng., Universidade Federal da Paraiba, 1979

M.Sc., Universidade Federal da ParaIba, 1982

A DISSERTATION SUBMITTED iN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

TN

THE FACULTY OF GRADUATE STUDIES

DEPARTMENT OF ELECTRICAL ENGINEERING

We accept this thesis a conforming

to the required standard

THE UNIVERSITY OF BRITISH COLUMBIA

November 1994

© Washington L.A. Neves, 1994

 In presenting this thesis in partial fulfilment of the requirements for an advanced
degree at the University of British Columbia, I agree that the Library shall make it
freely available for reference and study. I further agree that permission for extensive
copying of this thesis for scholarly purposes may be granted by the head of my
department or by his or her representatives. It is understood that copying or
publication of this thesis for financial gain shall not be allowed without my written
permission.

(Signature)

Department of E (eY,.i
The University of British Columbia
Vancouver, Canada

Date
eetde ,gg4

DE-6 (2/88)
 Abstract

Transformer modelling is a concern for the utility industry. The object of this work
is to develop and investigate dynamic core models suitable for transient studies. A major
advantage of the core models developed here is that they use readily available transformer
test data as supplied by the manufacturer.

For ferroresonance and inrush current studies, core saturation needs to be

represented reasonably well. A direct approach to producing nonlinear peak flux-current
and voltage-current characteristic of the iron core, taking iron losses into account, is
presented. The algorithm is simple in concept, easy to implement, and may be useful for
electromagnetic transient programs. A crude estimation of the transformer open circuit
capacitance is also made from rated frequency data. It is useful for situations in which the
transformer exciting current experiences strong capacitive effects.

An iterative algorithm for more correctly representing the flux-linkage curve of a

delta-connected transformer, suitable for situations in which the tests are performed with a
closed delta, is developed. It uses positive sequence excitation test data as input and takes
into consideration the removal of triplen harmonics from the line current.

An approach to model frequency-dependent effects in the transformer core from

transformer no-load loss data, is presented. Hysteresis and eddy current effects in the core
are treated simultmeously. The flux-current trajectories are generated by circuit models
with no need to pre-define them.

Simulations using the developed models are compared to laboratory measurement

of inrush current and to a ferroresonance field test.

II
 Table of Contents

Abstract

Table of Contents

Viii
List of Tables

1X
List ofFigures

Acknowledgement xiii

XIV
Dedication

1 Introduction 1

1.1 Transformer Modelling 1

1.2 Thesis Outline 2

1.3 Thesis Contributions 3

2 Literature Review and Case Studies ‘I

2.1 Introduction 4

2.2 EMTP Basic Models 4

2.3 Other Models 9

2.4 Iron Losses 12

111
 2.4.1 Laboratory Measurements 18

2.4.2 Building Factor 20

2.5 Transformer Core Representation During Transients 21

2.5.1 Ferroresonance Case Study 21

2.5.2 Inrush Current Case Study 24

2.6 Summary 25

3 On Modelling Iron Core Nonlinearities 28

3.1 Introduction 28

3.2 Saturation Curves 29

3.2.1 Computation of the v i,. Curve

- 32

3.2.2 Computation of the A-i, Curve 36

3.3 Comparisons Between Experiments and Simulations 38

3.4 Ferroresonance Simulations and Field Test 41

3.5 Distribution Transformer Saturation Curves 43

3.6 Summary 45

4 Saturation Curves ofDelta-Connected Transformers From Measurements

4.1 Introduction 47

4.2 Basic Considerations 48

4.3 Saturation Curves 49

lv
 4.3.1 Computation of the v 1T4 Curve
- 51

4.3.2 Computation of the 2 -44 Curve 53

4.4 Case Study 55

4.5 Summary 57

5 Hysteresis and Eddy Current Losses in Iron Core

5.1 Introduction 59

5.2 Frequency Dependent Core Models 59

5.3 Core Loss 62

5.4 Eddy Current and Hysteresis Modelling 63

5.4.1 Construction of Y(s) from Its Real Part 64

5.4.2 Linear Network Synthesis 66

5.4.3 Iron Core Nonlinearities 69

5.4.4 Numerical Example Hysteresis

-
69

5.5 Inrush Current: Simulation and Measurements 72

5.6 Ferroresonance 76

5.7 Summary 77

6 Transformer Models Applications

-
79

6.1 Introduction 79

6.2 Basic Transformer Equivalent Circuit 79

V
 6.2.1 Single-Phase Two-Winding Transformer 80

6.2.2 Single-Phase Three-Winding Transformer 82

6.2.3 Three-Phase Transformer 83

6.3 Estimation of Transformer Parameters 86

6.3.1 Short-Circuit Tests 86

6.3.1.1 Single-Phase Transformers 86

6.3.1.2 Three-Phase Transformers 90

6.3.2 Open Circuit Tests 92

6.3.2.1 Stray Capacitances 94

6.4 Sensitivity Study 97

6.5 Field Test 98

6.6 Summary ioo

7 Conclusions 102

Refrrences 104

A Orthogonality Between ‘r and I 110

B Measurement ofthe Initial Magnetization Curve 112

C Computation of Triplen Harmonic Components ii 4

D Rational Approximation of the Real Part of Y(s) 116

D. 1 Fitting Procedure 116

vi
D.2 G(w) Obtained from Lamination Data 118

D.3 G(w) Obtained from Standard Tests 120

vii
 List of Tables

2.1 Typical distribution transformer data 26

3.1 Laboratory measurements 39

3.2 Computed v - Ir and A i curve

-
39

4.1 Three-phase transformer test data 55

5.1 Flux-current curve 71

5.2 Linear circuit parameters 74

6.1 Distribution transformers 87

6.2 Correction factors 94

6.3 Corrupted flux-current curves 96

6.4 Sensitivity study 97

viii
 List of Figures

2.1 Nonlinear element connected to linear network

7
2.2 Simultaneous solution of two equations
7
2.3 Piecewise linear inductance with two slopes
8
2.4 “Switched inductance” implementation of two slope piecewise linear inductance
8
2.5 Elementary domain hysteresis ioop
10
2.6 Ladder network for eddy current representation
11
2.7 Eddy current distribution in a ferromagnetic sheet of thickness d
14
2.8 Hysteresis Loops
16

2.9 Typical steel lamination power loss curves

17
2.10 AC core loss measurements 19
2.11 Power loss curve for commercial grain oriented steel laminations 19
2.12 Ferroresonance in a 1100kV test system 22
2.13 Thêvenin equivalent circuit 22
2.14 Nonlinear inductance characteristic referred to the 1100 kV side 23
2.15 Voltage at phase A (sensitivity study for changes in core resistance) 23
2.16 Voltage at phase A (sensitivity study due to variations in the air core inductance) 24
2.17 Energization of a transformer
27
2.18 Inrush current (sensitivity study)
27

ix
3.1 Excitation test
31
3.2 V,- Average power curve
32
3.3 Computation of the nonlinear resistance
34
3.4 A-i
curve
1 40

3.5 Computed v ir curve

-
40
3.6 Nonlinear resistance 41
3.7 Nonlinear inductance 42
3.8 Ferroresonance in a power system 42
3.9 Nonlinear core inductances 43
3.10 Nonlinear resistances 44
3.11 Newly manufactured transformer V,- I, curve 44

4.1 Delta-connected transformer positive sequence excitation test 48

4.2 Core representation 50
4.3 V,- power loss curve 52
4.4 Generating current waveform from sinusoidal flux 54

4.5 A i curve
-
56

4.6 VirCUrve 57

5.1 Eddy current representation of the core after [18] 61

5.2 Eddy current representation of the core after [16] 61

x
5.3 Eddy current representation of the core after [20]
62
5.4 Core parameters
64
5.5 Realization of RL networks
67
5.6 Frequency-dependent representation of the core 69
5.7 A i hysteresis curve
-
70

5.8 Frequency-dependent resistance 71

5.9 Flux-current loops 72
5.10 Flux-current curve of a single-phase 1 kVA transformer 74
5.11 Transformer input voltage 75
5.12 Transformer inrush current 75
5.13 Transformer inrush current 76
5.14 Ferroresonance in a power system 77

6.1 Basic transformer model 80

6.2 Two-winding transformer 81

6.3 Frequency-dependent short-circuit impedance model 82

6.4 Three-winding transformer 83

6.5 Three-phase transformer short-circuit test 84

6.6 Impedance measured on the primary side 88

6.7 Impedance measured on the secondary side 88
6.8 Short-circuit impedances of single-phase transformers 88

xl
6.9 Measured and simulated L/R curves
89
6.10 Short-circuit impedance frequency response
90
6.11150 kVA three-phase transformer (self impedances)
91
6.12 500 kVA three-phase transformer (self impedances) 91
6.13 75 kVA short-circuit impedances (self impedances) 91
6.14 500 kVA transformer L/R curve 92
6.15 Single-Phase transformers 93
6.16 V,- I, characteristic for a 75 kVA three-phase transformer 94
6.17 Newly manufactured transformer 95
6.l8DCdrivesetup 98
6.19 Measured voltage waveform 99
6.20 Measured current waveform 99
6.21 Transformer secondary line voltage 99
B. 1: Measurement of the initial magnetization curve 111

B.2: Hysteresis loop locus 112

D. 1: Frequency dependent resistance 117

D.2 Transformer core loss curve at rated flux 120

xii
 Acknowledgement

I would like to express my gratitude to those who helped me to complete this

research work. Special thanks are due to my supervisor, Dr. Hermann Dommel for his

encouragement, patience, guidance and for arranging much-needed financial support.

I wish to thank Dr. José MartI for discussions and invaluable suggestions. I am grateful to

Dr. A. E. Araijo, Dr. S. Cameiro and Dr. A. Soudack for suggestions on the reseach work and on

the presentation on the results in this thesis.

I am grateful to Dr. W. Dunford and Dr. M Wvong for their help at the early stage of this

work. Thanks are due to Mr. A.. Otter for providing financial help from TRIUIVIF (Tri-University

Meson Facility)-Vancouver, B.C., and for discussions and providing test material.

Thanks are due to The Canadian Electiical Association (CEA) and Powertech Labs Inc., for

financial assistance duiing part ofthis work and for providing transformer data from Project 267-D-

766. The support of Mr. J. Drakos, Mr. M. B. Hughes and Mr. K. Takahashi, is gratefully

acknowledged. I am grateflul to Dr. Wilsun Xu for invaluable discussions, and for help on Chapter 6.

The financial assistance from Universidade Federal da Paraiba (UFPB)-Campina Grande

Brazil, where I received the support ofthe Department ofElectrical Engineering, and from Conseiho

Nacional de Pesquisa (CNPq-Brazil), are gratefully acknowledged.

I am deeply gratefhl to my wife Cataiina and my daughter Deborah for all their love,

patience and help. I*iing this time, my wife brought light through our new born Eduardo, looked

after the family, and kept encouraging me.

Finally, to Kadi Pun-u, Claudia Lisbôa, Claudia Oliveira, Jane Armstrong and their

families, the Soudack’s and those friends who cared, my warmest thanks.

xlii
to:

Elisa Silva

and,

to the Memory of

Edison Guimaräes - “Padin”

His early blindness by the sight did not prevent him
from being a tireless learner of electricity and music.

xiv
 Chapter 1

Introduction

1.1 Transformer Modelling

The simulation of electromagnetic transients in power systems is essential for

insulation coordination studies and for the adequate design of equipment and its protection.
To carry out these studies on digital computers, mathematical models are needed for the
various components, whether with lumped or distributed parameters. To attempt to model
each component in its entirety and then to determine its interaction with the rest of the

system would be extremely difficult due to the system complexities. Models with some
simplifications, which are still accurate enough for practical purposes, are therefore usually
used.

The problem of accurately predicting the transient electrical interaction of power

system componnts has faced the electric power engineer for almost a century. There is a
large amount of research work in this area. The transformer is one of the most important
components in power systems, and because of this, it has been given special attention [1].

1
 Chapter 1. Introduction 2

Despite the large number of papers published in the area, transformer modelling still
presents substantial difficulties today. Transformer inductances are nonlinear and frequency-
dependent. The distributed capacitances between turns, between winding segments and
between winding and ground produce resonances that may affect terminal and internal
voltages [2]. The core modeffing may play a very important role for ferroresonance and
inrush current studies in transformers [3].

1.2 Thesis Outline

When power system transients are to be computed, general purpose programs such
as the EMTP (Electromagnetic Transients Program) are often used [4]. Our goal is to
advance the modelling of transformers in connection with these programs, focussing on
saturation, eddy currents and hysteresis effects in the iron core. The developed models are
intended to be applicable for situations such as ferroresonance and inrush currents in
transformers. The models are discussed as follows:
• A brief literature review of transformer models is presented in Chapter 2.
• Saturation in the core is represented by nonlinear functions obtained from the

transformer test data. The model development, measurements and simulations are
discussed in Chapter 3.
• An algorithm to produce saturation curves of delta connected transformers, from

positive sequence open circuit tests, is developed in Chapter 4.

• Frequency-dependent core models are presented in Chapter 5.

• In Chapter 6, distribution transformer models are obtained from 60 Hz parameters.

The difficulties in obtaining the transformer parameters are outlined. Comparisons

between simulations and a field test are carried out.
 Chapter 1. Introduction 3

• The main conclusions of this thesis work are summarized in Chapter 7.

1.3 Thesis Contributions

The author believes the following to be original contributions from this research work:

a) A direct method to more accurately compute saturation curves from transformer

standard test data (Chapter 3).

b) An algorithm to compute saturation curves of three-phase delta-connected

transformers in which the delta connection could not be opened for tests (Chapter 4).

c) Development of frequency-dependent core models in which eddy current and

hysteresis are treated simultaneously. For these models it is not necessary to pre-define

the trajectories ofthe dynamic hysteresis loops (Chapter 5).

d) Guidelines to model distribution transformers from rated parameters (Chapter 6).

The following publications report part of the research work developed in this thesis:

1) W. L. A. Neves and H. W. Dommel, On Modelling Iron Core Nonlinearities, IEEE

Transactions on Power Systems, Vol. 8, No. 2, May 1993, pp.417-425.

2) W. L. A. Neves, H. W. Dommel, Saturation Curves of Delta-Connected

Transformers From Measurements, to appear in IEEE Transactions on Power

Delivery. Paper 94 SM 45 9-8 PWRD presented at IEEE PES Summer Meeting, July

24-28, 1994, San Francisco, CA.

3) W. L. A. Neves, H. W. Dommel and Wilsun Xu, Practical Distribution Transformer

Modelsfor Harmonic Studies, to appear in IEEE Transactions on Power Delivery.

Paper 94 SM 406-9 PWRD presented at IEEE PBS Summer Meeting, July 24-28,

1994, San Francisco, CA.

 Chapter 2

Literature Review And Case Studies

2.1 Introduction

A brief review of various transformer representations for digital simulation of

transients in power systems, and a discussion of eddy current and hysteresis loss in magnetic
cores are presented next. It is shown that an exact model which reproduces the frequency-
dependent core losses, even at low frequencies, is very difficult to achieve. Sensitivity
analysis for ferroresonance and transformer inrush current case studies are carried out. In
these studies, the transformer core is represented by a nonlinear inductance in parallel with a
constant resistance. It is shown that the system is more sensitive to variations in the core
inductance. The system is not sensitive to small variations in the shunt resistance. However,
typical transformer data show that no-load losses at rated frequency increase quickly as the
transformer is driven into saturation. This may be significant for ferroresonance studies.

2.2 EMTP Basic Models

Guidelines to model transformers with the EMTP are presented in references [4,51.
These models are based on circuit theory. The linear behaviour of transformers can be

4
 Chapter 2. Literature Review and Case Studies 5

represented by branch resistance and inductance matrices [R] and [LI (here the exciting
current must not be ignored since its absence produces infinite elements in the inductance
matrix), or by a matrix [R] and an inverse inductance matrix [LI-
. These matrices are
1

obtained from positive and zero sequence short circuit impedances and from open circuit
impedances. Saturation effects can be simulated by appending nonlinear inductance
branches. In [4], it is suggested that these nonlinear branches should be placed across that
branch in the equivalent circuit where the integrated voltage is equal to the iron core flux.
Although this point depends on the transformer design and, in general, is not accessible in
the model, it can be approximated fairly accurately by using the branch of the winding
closest to the core (usually the lower voltage winding). Saturation curves of transformers
are often supplied as rms values of voltages and currents(V,,, —f(I)). A technique for
converting this curve to a peak flux versus peak current characteristic( A=f(i)) is supplied

by the auxiliary program CONVERT[4]. This algorithm does not take eddy currents and
hysteresis losses into account, i. e., when computing saturation curves it is assumed that the
excitation branch consists only of a nonlinear inductance. The next two chapters present
improvements on the computation of saturation curves by including the effect of transformer
no-load losses.

In the EMTP, nonlinear elements are either represented as piecewise linear or as

nonlinear with the compensation method [6]. When the compensation method is used,
nonlinear elements are simulated as current injections, which are superimposed on the linear
network solution without the nonlinear elements. As an example, consider a case where the
network contain only one nonlinear resistance between nodes k and m (Figure 2.1). The
network solution is found by the compensation theorem according to the following steps:
 Chapter 2. Literature Review and Case Studies •6

• remove the nonlinear branch between nodes k and m and calculate the
open circuit voltage v
;
0

• build the instantaneous Thèvenin equivalent circuit between nodes k and m

(to find the Thèvenin resistance, a cuffent of 1 A must be injected from

node k, and drawn out from node m);

• solve the two following equations simultaneously:

v=0
v,,, —R(t). ‘km (2.1)

= f(1icm) (2.2)

Equation (2.2) represents the nonlinear resistance characteristic. Figure 2.2 shows the

simultaneous solution of the two equations above (intersection between the two curves). For

nonlinear inductances, the nonlinear characteristic is usually known in the form:

(2.3)

The EMTP uses the trapezoidal rule of integration and converts the flux A(t) into a

linear ftinction of (t) and the network solution is found in a similar way as for a nonlinear

resistance.

The saturation characteristics of modern transformers are often represented as

piecewise linear inductances of two slopes (Figure 2.3). Such piecewise linear inductances

can be simulated with two linear inductances L

1 and L
2 in parallel (Figure 2.4). The switch
is closed whenever Aj2SATURAT10N, and is opened again as soon as
 Chapter 2. Literature Review and Case Studies 7

k km
-

linear part
of network
1
i?; Lkm

Figure 2.1: Nonlinear element connected to linear network.

Vkm

VkmO

nonlinear resistance
curve

network curve

‘km

Figure 2.2: Simultaneous solution of two equations.

 Chapter 2. Literature Review and Case Studies 8

2..

2
L
‘SATURATION

‘SATUBA11ON

Figure 2.3 : Piecewise linear inductance with two slopes.

kj

Vkm

Figure 2.4: “Switched inductance” implementation of two slope piecewise linea

inductance.
 Chapter 2. Literature Review and Case Studies 9

Magnetic hysteresis effects have been incorporated in the BPA (Bonneville Power
Administration) version of the EMTP [7]. This model uses pre-defined trajectories in the
A i plane to decide in which direction the curve will move if the flux either increases or
—

decreases. Eddy current effects in the core are represented as fixed resistances. Mork and
Rao [8] used this model to simulate ferroresonance and compared their results to laboratory
measurements. There was a large discrepancy between measured and simulated curves. A
single-valued flux-current characteristics predicted voltage and current waveforms in closer

agreement to the tests. In Chapter 3, it is shown that a nonlinear resistance may be necessary
to represent eddy current effects in transformers.

2.3 Other Models

Dick and Watson [9] described a method of saturating large power transformers and
plotting instantaneous magnetization curves. The authors used a detailed equivalent circuit
transformer model based on the principle of duality between magnetic and electric circuits
which takes the yoke saturation into account. Hysteresis loops are modeled using pre
defined trajectories constructed from a hyperbolic equation.

Germay et al. [101 studied ferroresonance effects in power systems. They represented
magnetic hysteresis by Preisach’s theory [11,12]. This theory assumes that the ferromagnetic
material is made up of elementary domains and that the magnetization characteristic of each
domain is a rectangular ioop characterized by the constants a and b (Figure 2.5), and by the
displacement field Hm representing the action of neighboring domains. It also assumes that a
distribution function, related to the probability of finding a ioop with given (a,b) is unique.
The distribution function can be computed numerically by manipulation of the saturation
loop and magnetization curve [11]. This theory has gained large acceptance. Its basic ideas
 Chapter 2. Literature Review and Case Studies 10

and evolution are presented by Mayergoys [12]. Recently, a hysteresis model based on this
theory was developed for the EMTP [13].

Santesmases et al. [14] represent transformer cores by a simple equivalent circuit

consisting of a nonlinear inductance in parallel with a nonlinear resistance. The nonlinear
elements are obtained from functions derived from the hysteresis dynamic ioops. This is
essentially the same model as proposed by Chua and Stromsmoe [15]. The resistance in the
model accounts for the energy loss due to the loops, which means that the hysteresis and
eddy current losses are assumed to have the same frequency dependence. A family of
dynamic hysteresis loops is needed to construct the nonlinear functions.
M

a H

;1

Hm

Figure 2.5 : Elementary domain hysteresis loop

A recent attempt to build a general transformer model for transient studies was
sponsored by the EMTP Development Coordination Group (DCG) [16]. The principle of
duality was used to model the magnetic flux paths in the air and in the iron parts. Frequency

dependent effects in the core were included by solving Maxwell’s equations (in the
 Chapter 2. Literature Review and Case Studies 11

frequency domain) within laminations, ignoring nonlinear effects

. As a result, a frequency-
1
dependent equivalent impedance Zeq(W) was found. Zeq(W) was matched by a ladder

network and connected in parallel with the iron core nonlinear inductance L
1 as shown in
Figure 2.6.

Ri R2 R3 R4

Figure 2.6 Ladder network for eddy current representation.

The ladder network reproduces the theoretical transformer frequency response with
an error less than 5% for frequencies below 200kHz. This model was applied in a situation
where a circuit breaker, on the low voltage side of the transformer, attempts to clear a fault
nearby [17]. The transient recovery voltage (TRV) is computed using both a frequency-
dependent model for the core and the conventional model (constant resistance in parallel
with the magnetizing inductance). The conventional model produced a more damped TRy.
2
It is difficult to know which model is correct since the authors did not show comparisons to
field measurements.

The idea of modelling power transformer eddy current effects by means of Maxwell’s
equations, has also been used in references [18,19,201. In the next section, some difficulties

1 ferromagnetic material was assumed to have constant permeability t and constant resistivity a.
2 In reference[17] it is also shown that the conventional model is accurate enough (numerical error less
than 5%) for frequencies up to 3 kHz.
 Chapter 2. Literature Review and Case Studies 12

concerning classical eddy current and hysteresis representations of ferromagnetic materials

will be discussed.

2.4 Iron Losses

Transformer cores are usually made of iron alloys. Core materials can be divided into

three major classes: non-oriented steel (hot rolled)

, grain oriented steel (cold rolled)
3 4 and

metallic glasses (amorphous material)

.
5

Most distribution and power transformer cores in service today are of grain oriented

silicon steel laminations. However, given the very low losses of amorphous alloys, the trend

may change in the near future. Today, thousands of distribution transformers and a few

power transformers made of metallic glasses are in service in the U.S.A., Japan and Canada

[22].

In the presence of a time-varying magnetic field, induced voltages, eddy currents and

hysteresis take place in the core material. Classical electromagnetic theory assumes a

uniform distribution of eddy currents when slowly time-varying magnetic fields are applied

to iron cores (Figure 2.7a). The first theoretical studies of eddy currents in iron sheets were
done by Oliver Heaviside followed by J. J. Thomson [24]. The iron was assumed to be a
homogeneous medium characterized by two constants: permeability ji and conductivity a.

Non-oriented grades of electrical sheets are designed to have the same magnetic properties in the rolling
direction as they have perpendicular to that direction. They were largely used in the past for power and
distribution transformer cores.
rolled materials were introduced to the market in 1934 by N. P. Goss [21]. Their permeabilities are
much bigger in the direction of rolling than perpendicular to that direction. Core laminations are usually cut
so that the magnetic flux is along the rolling direction for the greatest part of its path through the core.
These materials have lower losses when compared to non-oriented steels.
5 alloys were introduced to the transformer market in the U.S.A. in 1976 [22]. These alloys present
higher resistivity, when compared to grain oriented steels, and very low losses.
 Chapter 2. Literature Review and Case Studies 13

The electromagnetic theory (Maxwell’s equations) was applied to show how the magnetic
flux density B would diffbse from one part of the material to the other according to the
equation

B=
2
V (2.4)

The equation above oversimplifies a much more complicated phenomenon.

6 Core
losses, computed by this approach, are always underestimated when compared to measured
values. A detailed review of eddy current and hysteresis loss in magnetic cores is presented
in [25]. Weiss, Barkhausen, Bitter, Landau and Liftshitz made significant contributions
towards the understanding of eddy current effects in ferromagnetic materials. In 1907, the
French physicist Pierre Weiss provided the first insight into understanding the behavior of
magnetic materials. He introduced the concept of magnetic domains. In 1919 Barkhausen
had shown that the magnetization could change in a very discontinuous way (Barkhausen
effect). in 1931, Francis Bitter, working at the Westinghouse Research Laboratories proved
the existence of domains by making them visible. His technique consists of polishing the

surface of the magnetic material and spreading a colloidal suspension of magnetic powder
over the surface. The powder will be deposited in regions of higher gradient fields (domain
boundaries) and the domains are then visible through a microscope[211. In 1935, Landau
and Lifshitz introduced the ideas that magnetization could change by a movement of the
boundary between domains, and that domains magnetized in the direction of the applied field
would expand at the expense of domains magnetized against the applied field. Today, it is

6 Maxwell died in 1879, he left his theory in the form of twenty equations in twenty variables. Shortly
slier his death, the reduction of his equations to the four vectorial equations known today was done
independently by Oliver Heaviside and Heinrich Hertz [23]. At that time, very little was known about
ferromagnetism. Eddy currents in ferromagnetic materials were assumed to behave the same way as in non-
magnetic conductors.
 Chapter 2. Literature Review and Case Studies 14

generally accepted that the eddy current loss is due to the micro eddy currents produced at
the moving domain boundaries. Therefore, eddy currents will be concentrated around the
moving domain walls, as shown in Figure 2.Th. The bigger the domain the larger the eddy
currents produced around its boundaries. Eddy current distribution may not be uniform even
for very slowly time-varying magnetic fields.

(a)
1;
L
(b)

Figure 2.7: Eddy current distribution in a ferromagnetic sheet of thickness d:

(a) Classical representation;
(b) Sheet subdivided into 180° domain of width a.
 Chapter 2. Literature Review and Case Studies 15

It is very complicated to properly use Maxwell’s equations within iron core

laminations to account for eddy currents. For a precise calculation, the effect of domain wall
motion should somehow be included in the field equations [26, 27, ‘28].

The loss in a ferromagnetic sheet at a frequency f consists of hysteresis and eddy

currents. The total loss is always greater than would be expected from the sum of static

hysteresis loss and eddy current loss, calculated using classical theory [29,30]. The excess
loss, arising from the non-uniform distribution of eddy currents, is known as anomalous
7 Since, in practice, only the total loss w can be measured, assumptions must be made
loss.
based on the most probable physical behavior of the material if hysteresis and eddy current
loss are to be separated. The total loss per kilogram, per volume of magnetic material, in an
iron sample can be written as a combination of three loss components:

WWh+We+Wa, (2.5)

where Wh , We and Wa are hysteresis, eddy current and anomalous loss, respectively, in
W/kg. It is well known that the total loss w is frequency-dependent. Hysteresis loss is

attributed to domain wall movements back and forth across crystal grain boundaries, non
magnetic inclusions and imperfections [211. It is common to assume that Wh is independent
of the speed in which the domain wall moves. So, the hysteresis loss per magnetic volume at
a given frequencyf is related to the enclosed area of the DC hysteresis loop (Figure 2.8)
according to the equation:

Wh=fHdB. (2.6)

‘
Electrical Engineering textbooks, usually address the magnetic domain theory to explain the properties of
magnetic materials, but seldom relate the domain wall movements to eddy currents. Anomalous losses were
known even before the domain theory was completed. In 1927, it was already known that there is a strong
correlation between grain size and eddy current losses [31].
 Chapter 2. Literature Review and Case Studies 16

I
H

Figure 2.8: Hysteresis loops. In the B H plane, the trajectory

-

of a signal of frequency f, will encircle the hatched

area (DC hysteresis loss) f times per second.

Any increase in loss per cycle above the DC hysteresis loss has been attributed to eddy

current effects [22]. Steinmetz [28] proposed the following equation for calculation of the

hysteresis loss:

WhkhB&f, (2.7)

where

kh is the hysteresis coefficient and depends on the core material;

Bm is the maximum flux density in Teslas; and

x is the Steinmetz coefficient (ranging from 1.5 to 2.2 depending on the core material).

The principal means of controlling the core loss is to use thin laminations. For a
lamination in which its thickness Ia is much smaller than its width, the classical eddy current

loss is given by [22]:

We(2r d Bm .f)
/(6p),
2 (2.8)
 Chapter 2. Literature Review and Case Studies 17

where p is the resistivity of the material, and Id is the lamination thickness. Inserting
equations (2.7) and (2.8) into (2.5) and dividing by f, the total loss per cycle is given by:

f I(6p)+Wa If
=kh BmX +(2td Bm)
2

Atypical curve of power loss per cycle as a fianction of frequency, for a constant flux
amplitude, is shown in Figure 2.9. The anomalous loss can be very high (usually greater
than the classical eddy loss for commercial steel at power frequency [261).

a)
C.)
C.)

G)
0
U)
U)
0

a
0
0

Frequency

Figure 2.9: Typical steel laminations power-loss curve.

 Chapter 2. Literature Review and Case Studies 18

2.4.1 Laboratory Measurements

As part of this thesis project, some measurements were perfbrmed for grain oriented

steel laminations to gain some insight into how eddy currents and hysteresis loss behave as

the frequency changes. The steel samples were assembled in a standard Epstein frame [33]. /

The amplitude Bm of the sinusoidal flux density B = Bm sin wt was kept constant during each

set of measurements.

Figure 2.10 illustrates the circuit used to measure the total AC core loss for a

frequency range from a few hertz up to 80 Hz. V(t) is a frequency variable sinusoidal

voltage source connected to a power amplifier. A waveform analyzer was used to measure

the voltages at points A and B with respect to ground. The current sample waveform was
taken from a 0. 12 resistance R connected in series with the Epstein frame primary winding.

Current and voltage waveforms (512 points) were obtained and the losses were computed
using a built-in routine. The total loss per cycle, as a function of frequency for Bm =1.OT, is

shown in Figure 2.11. The laminations were 0.3mm thick, with p = 4.5 x i02. m. The solid

line, through the measured points, is a second order polynomial approximation. This curve

is extended downwards to f=0. At f0, it is assumed that the loss per cycle is the DC

hysteresis loss. The classical eddy curent loss is computed using (2.8) and added to the

hysteresis loss. The losses per cycle calculated by the classical approach are lower than the

measured ones. The hysteresis loss per cycle at 60 Hz accounts for about half of the total

loss. Reference [32] quotes measurements in grain oriented steel laminations in which the

anomalous loss could be close to an order of magnitude higher than the classical eddy

current losses for frequencies up to 1 kHz. Herzer and Hilzinger [34] show examples of
amorphous alloys with large anomalous loss (nealy 40% of total loss) at frequencies of 100

kHz.
 Chapter 2. Literature Review and Case Studies 19

It is complicated to predict losses in iron cores accurately. A better understanding of

loss mechanism in ferromagnetic materials is providing researchers with the tools to reduce
them. Theoretically, the classical methods would be applied properly if the laminations were
made of a fine domain structure. Researchers are struggling to reduce the total loss and
increase permeability of ferromagnetic steels. Very high permeability low loss steel sheets
are on the market today. Nevertheless, although total losses are low, anomalous loss is still
high when compared to the classical eddy current loss [221.

V(t)
(EZEEEEJE

Figure 2.10 : AC core loss measurements.

10

8
0)
-‘

-3
E6
ci)
C.)

0
>— Measured Losses
Cl)
U) Classical Losses
0
.J 2

20 40 60 80 10

Frequency (Hz)

Figure 2.11: Power-loss curve for commercial Grain oriented steel lamination.
 Chapter 2. Literature Review and Case Studies 20

2.4.2 Building Factor

There is also a further complication that transformer core loss per kilogram is always

greater than the nominal loss of the steel as measured in standard testers. The ratio between

the transformer per unit loss and the nominal or standard unit loss is called the “building

factor” of the core. Building factors usually range from 1.1 to 2.0 [35]. The extra loss is due

to phenomena such as:

• Non-uniform flux distribution due to difference in path lengths among magnetic

circuits;

• Distortion of flux waveform due to magnetic saturation;

• Flux directed out of the rolling direction;

• Transverse flux between layers due to joints.

The flux distribution in transformer laminations is not uniform even at low

frequencies [36,37,38]. For a sinusoidal applied flux, the flux in each lamination is not

sinusoidal, although the flux components add up to produce the sinusoidal total flux.

Advances in computer software have been used to improve the design of electrical

machinery [39]. There are several commercial programs available today [40]. They are

essentially usefhl for situations in which qualitative results are important (for instance, in

designing transformer lap joints, it is important to find the geometry of the joints that leads

to minimum losses). The accuracy of the present methods needs verification against
experiments [35].
 Chapter 2. Literature Review and Case Studies 21

2.5 Transformer Core Representation During Transients

The major nonlinear effects in transformers are saturation, eddy currents and
hysteresis. Saturation is the predominant effect [41]. In the following sections,
ferroresonance and inrush current simulations will be addressed. The transformer excitation /

branch is represented by a crude model (constant resistance in parallel with a nonlinear

inductance). Sensitivity studies are carried out to analyze how the system responds to
changes in the core model parameters.

2.5.1 Ferroresonance Case Study

Consider the BPA (Bonneville Power Administration) 1100kV test system [42]. It

comprises a generating station, a transformer bank (autotransformers) and a short three-

phase transmission line. Field tests were carried out. Ferroresonance occurred in phase A
when this phase was switched off on the low voltage side of the transformer (Figure 2.12).
Phase C was not yet connected to the transformer at that time. One can study this case,
replacing the dotted part of the network by its Thèvenin equivalent circuit, which consists of
a voltage source behind a capacitance (crude representation of the capacitive coupling to
phase A of the line). Figure 2.13 is the equivalent circuit of the system referred to the high

voltage side. The nonlinear inductance characteristic shown in Figure 2.14 (three straight
line segments) was obtained from the curve supplied by the transformer

manufacturer, using the method of [4]. Unfortunately, the transformer no-load data were
produced by exciting voltages that did not go beyond 1.1 p.u., and data at higher saturation
levels would be needed for this case. Autotransformers have typical air core inductances (the
core is completely saturated and it behaves like air) of 3 to 4 times the short circuit
inductance [4]. A straight line segment, with a slope of 4 times the short circuit inductance
was connnected to the last segment of Figure 2.14, to represent the air core inductance. The
 Chapter 2. Literature Review and Case Studies 22

Microtran® program[43J was run twice for different values of core resistance (RRc and

R0. 8R, where Rc=4.2M2 is the resistance at the rated voltage). A time step of zt=1 OOis

was used in each case. Both simulations, of the terminal voltage a’t phase A, are shown in

Figure 2.15. The two curves are almost identical.

Figure 2.12: Ferroresonance in a 1100kV test system.

152 Q 11.3f2 742L2 O.O2619F

7422 11.3L2

635. 1/Q.kV. 131.0 Z12 kV

Figure 2.13 : Thèvenin equivalent circuit.

 Chapter 2. Literature Review and Case Studies 23

3000

2000
(0
>
x
:5

1000

2 4 6 8
Current (A)
Figure 2.14 : Nonlinear inductance characteristic referred to the 1100 kV side.

RRc
1200 R=0.8Rc

800

400

I0

-400

-800

20 40 60 80 100 120 14
Time (ms)
Figure 2.15: Voltage at phase A (sensitivity study for changes in core resistance).
 Chapter 2. Literature Review and Case Studies 24

The Microtran program was run again, now with the slope of the air core inductance

made 20% smaller. The results are shown in Figure 2.16, where the solid curve is the same

curve of Figure 2.15 for R=R. It is seen that the simulated sytem is more sensitive to

variations in the nonlinear inductance than to core losses.

Correct air core inductance

1200 —
Air core inductance 20% smaller

800C

400

-400
—

-800 —

-
I I ‘ I ‘ I
20 40 60 80 100 120 14
Time (ms)
Figtue 2.16: Voltage at phase A(sexivity study due to vaiiations hthe air core inductance).

2.5.2 Inrush Current Case Study

Now consider the analysis of inrush current simulations in a single phase transformer

(same transformer data of the previous section). The aim here is to see how sensitive the

current waveforms are to changes in the core resistance and nonlinear inductance (Figure
 Chapter 2. Literature Review and Case Studies 25

2.17). The plot of the inrush current for three values of core resistance (R=R, R0.5R
0
and R=oo) and the nonlinear inductance of the previous example, is shown in Figure 2.18.

The solid curve represents in fact any of the simulations. It was asumed that no transformer
residual flux existed prior to energization at t0. In these simulations, inrush currents are not
sensitive to variations of the core resistance. However, if the slopes of the flux-current
characteristic above the rated flux were changed by 10% (dashed curve), a noticeable
difference between the current waveforms would be seen. In these simulations, inrush
currents are shown to be very sensitive to variations of the nonlinear inductance.

2.6 Summaiy

This chapter summarized the difficulties in modelling transformer cores. Their exact
representation during transients is complex since the magnetic properties of the cores are not
yet fully understood. Another major problem is the availability of data. All that is usually
available from transformer manufacturers are data obtained from tests performed at rated
frequency [46,47,48]. The information available from open circuit tests are rms voltages as a
function of rms currents, and no-load losses for a few input voltage levels.
In the simulation examples of Section 2.4, a constant resistance was used to
reproduce the transformer core loss. It was shown that the system is not sensitive to small
variations in the resistance. In practice, however, this resistance is not linear. Typical
distribution transformer correction factors are shown in Table 2.1 [44]. They should be
applied to no-load losses, at rated voltage, to give the correct losses when the transformer is
driven into saturation. The equivalent resistance, which reproduces the open-circuit losses,
decreases as the voltage level increases beyond the rated voltage (losses increase at faster
rate than the square of the voltage). For 225 kVA and above (three-phase transformers), the
open circuit equivalent resistance is equal to approximately half of the resistance at rated
 Chapter 2. Literature Review and Case Studies 26

voltage. This may make a difference during ferroresonance studies. Another complication is

that the core loss is frequency-dependent [29,38,49].

Table 2.1 - Typical distribution transformer data.

Operating Correction

Voltage (%) Factors

105 1.15 For 167 kVA and below, 1 phase

110 1.30 For 150 kVA and below, 3 phase

105 1.5 For 225 kVA and above, 1 phase

110 2.4 For225kVAandabove,3phase

 Chapter 2. Literature Review and Case Studies 27

635. 1LLkV

Figure 2.17: Energization of a transformer.

Correct air core inductance

150
Air core inductance 10% smaller
—

100 —

50-

20 40 60 80 10
Time (ms)

Figure 2.18: Inrush current (sensitivity study).

 Chapter 3

On Modelling Iron Core Nonlinearities

3.1 Introduction

Transformer manufacturers usually supply saturation curves in the form of rms

voltages as a function of rms currents. Some methods have been used to convert these
—I,, curves into peak flux peak current curves (nonlinear inductances) [4, 45, 511.
-

As shown later, these methods can be modified to take iron core losses into account, thereby

producing a nonlinear inductance as well as a parallel nonlinear resistance.

In addition, laboratory experiments were performed with a silicon iron steel core
assembled in an Epstein frame. Average power and rms current at 60 Hz were measured at
different voltage input levels. For comparison purposes, the initial magnetization curve for
the core material was measured as well.

Simulations of ferroresonance in a power system are carried out to examine the

effect of the transformer nonlinearities on its terminal voltage waveform. These simulations
are compared to a field test.

28
 Chapter 3. On Modelling Iron Core Nonlinearities 29

The algorithm is applied to distribution transformers to get their open circuit

nonlinear parameters. Measurements have shown that there are newly manufactured
transformers in which their excitation currents at rated voltag’e are smaller than their
excitation currents at lower voltages. For these transformers, improvement on the computed /
saturation curves would be accomplished by modif,’ing the algorithm to include the effect of
stray capacitances.

3.2 Saturation Curves

The cores of transformers and reactors are sometimes represented as an equivalent

circuit consisting of a nonlinear inductance (A—ii curve) in parallel with a nonlinear
resistance (v — ‘r curve) [14, 15]. The characteristics of these elements are obtained from the

dynamic hysteresis loops. The resistance in this model accounts for the energy losses due to
the loops. Chua and Stromsmoe [15] did make comparisons between simulations and
laboratory tests for a small audio transformer, and for a supermalloy core inductor as well. A
family of peak flux - peak current ioops for 60, 120 and 180 Hz sinusoidal (voltages and
current) excitations of various amplitudes were obtained. The agreement between
simulations and measurements of the loops was very good. This indicates that, for the
frequencies under consideration, a nonlinear resistance would represent hysteresis and eddy
current effects reasonably well.

The same equivalent circuit is used here. However, the nonlinear characteristics are
calculated in a simpler way directly from the transformer test data. The nonlinear resistance
(piecewise linear v—i,. curve) is found from the no-load (excitation) losses. This information

is then used to compute the current through the nonlinear inductance and to construct the
piecewise linear A I, curve.
—
 Chapter 3. On Modelling Iron Core Nonlinearities 30

Figure 3.1(a) shows a voltage source connected to a single-phase transformer

whose excitation branch is represented by a nonlinear inductance in parallel with a nonlinear
resistance. Their nonlinear characteristics are computed according to the following
assumptions:

• the Vr and ?—i, curves (Figures 3.1(b) and 3.1(c)) are symmetric with respect
to the origin (Rk and Lk are the slopes of segment k of the VIr and A—i
1 curves,
respectively);

• the no-load test is performed with a sinusoidal voltage source; the winding
resistances and leakage inductances are ignored.

The conversion algorithm works as follows:

For the construction of the VIr curve (Section 3.2.1):

• compute the peak values of the current Ir(t) point by point from the no-load

losses, and subsequently compute their rms values

For the construction of the A—i, (Section 3.2.2):

• obtain the rms values Il.-rmj of the current i, (t) through the nonlinear inductance
from the total rms I,,.. current and the applied voltage v(t);

• compute the peak values of the inductive current i,(t) point by point from their rms

values and rms voltages.

 Chapter 3. On Modelling Iron Core Nonlinearities 31

v(t)

(a)

(b) (c)

Figure 3.1: Excitation test:

(a) core representation;
(b) V ‘r characteristic;
—

(c) ). i characteristic
—
 Chapter 3. On Modeffing Iron Core Nonlmearities 32

3.2.1 Computation ofthe v Ir Curve -

Let us assume that the no-load losses I, F,..., m are ava,ilable as a function of the
applied voltage , V as shown in Figure 3.2.
...,

ms1 3
V
2
Vrrns
rms

Figure 3.2 : Vrms Average Power curve

-

From these data points we want to construct a piecewise linear resistance curve, as shown in

Figure 3.3(b), which would produce these voltage dependent no-load losses. Let us first

explain how the no-load losses can be obtained from a given v i,. curve, before describing
—

the reverse problem of constructing the v i curve from the given no-load losses at rated
—

frequency. For instance, assume that the applied voltage is and varies sinusoidally as a
function of time, as shown in Figure 3.3(a), with

(O)=V
2
v
s inO (3.1)

where 2 = V, Because of the symmetry of the v 1 curve with respect to the origin, it
—

is sufficient to observe 1/4 of a cycle, to 0 = ir /2. From Figure 3.3, it can be seen that:
 Chapter 3. On Modelling Iron Core Nonlinearities 33

(V sin 0)1R
1 if 0<01
‘r(0)ji
+(V sin 2
)/R if 01
1
0—V

In general, ‘r (0) can be found for each v(O) through the nonlinear v r characteristic, either
—

graphically (as indicated by the dotted lines in Figure 3.3), or with equations. This will give
us the curve i,-(0) over 1/4 of a cycle, from which the no-load losses are found as

. =_$V(0)i(0)d0 (3.2)

Let us now address the reverse problem, i.e., constructing the v—i,. curve from the given no-

load losses. Obtaining the points ,

V, ..., Vm on the vertical axis of Figure 3.3(a) is
simply a re-scaling procedure from mis peak values,

(3.3)

for k = 1,2,3, ..., m. For the first linear segment in the v ‘r curve, the calculation of the

peak current 1,-i, on the horizontal axis is straightforward. Since P

11I,,,,, in the linear
=V
,
case,

21
.
1
ri-v
 Chapter 3. On Modelling Iron Core Nonlinearities 34

v(e) V

3
V
V
2

Vi

(a)

01

It
2

0 (c)

Figure 3.3: Computation of the nonlinear resistance:

(a) sinusoidal voltage input signal;
(b) V r curve to be computed;
(c) output current.
 Chapter 3. On Modelling Iron Core Nonlinearities 35

For the following segments (k >2), we must use the power definition of equation (3.2), with

k sin 0 (Figure 3.3a). Then

the applied voltage v(0) = 17

=[s:1VkSiflo(0Jdo +

f 8
(J’ sin 0Irj +
SO —

]dO+ + (3.5)

sin o-
( +

The “break points” 01, 02, ..., k—I

0 in equation (3.5) are known from

0 =arcsin(VJIVk), (3.6)

for j = 1, 2, ..., k — . The only unknown in equation (3.5) is the slope Rk in the last

segment. The average power can therefore be rewritten in the form

br
Pk=ari+—-, (3.7)

with ark, brk and Pk known values. Rk is then easily computed and In is calculated from

-r
Vk-Vkl
lrklrk_I+
Ak

This computation is done segment by segment, starting with ‘n2 and ending with the

last point ira. Whenever a point Irk has been found for the horizontal axis in Figure 3 .3b, its

rms value is calculated as well, because it is needed later for the construction of the ? —

curve. 4, is found from the definition of the rms value,

 Chapter 3. On Modelling Iron Core Nonlinearities 36

I2_
J
2 i2(o)do (3.9)

i.e.,

=[J:f0 dO+
jr +
192( VksinOJI
j dO + ... + (3.10)

+ Vk Sifl O_
Ji [ir*i do].

3.2.2 Computation ofthe A i, curve -

The A — curve is computcd using the rms current information from the v I,. curve.
—

Peak voltages are converted to peak fluxes and the rms values of the current through the

nonlinear inductance are converted to peak values.

The conversion of peak voltages Vk to peak fluxes Ak is a re-scaling procedure.

Hence, for each linear segment in the A—i

1 curve,

(3.11)

where Co is the angular frequency.

Let us now compute the peak values of the inductive current. At first, their rms
values are evaluated. It can be shown that for sinusoidal input voltages, the harmonic

components of the resistive current are orthogonal to their respective harmonic components

of the inductive current (see Appendix A). Then,

 Chapter 3. On Modelling Iron Core Nonlinearities 37

5
Il_nnsJItrnL_ Ir2_rm , (3.12)

with the resistive current Ir_ already computed from equation (3.10) and the total
current known from the transformer test data. For the first linear segment in the 2
k —i,

curve,

= 1i-rmsi (3.13)

For the following segments k 0, the peak currents are obtained by evaluating I,_ for
each segment k, using equation (3.9). Thus, assuming Ak(O)=Aksin8, we have
1

=[s:i[0 dO +

SQZ [I ÷ Aks1n0_?jdO+... + (3.14)

j do]
sin 0— k-1
+

Here, similarly to the case of the v I,. curve computation, only the last segment Lk of
—

equation (3.14) is unknown. Equation (3.14) can be rewritten in the form

alkrk+bIj’k+clk =0 (3.15)

1
F or computation of the rms value of the inductive current, it does not matter what the flux phase is, owing
to the fact that the voltage (or flux) is assumed to be sinusoidal and the 2 — curve symmetric with respect
to the origin. Here, for computing purposes only, it is assumed (0) = Ak sin8. This has the advantage
that the limits of integration in equation (3.14) are the same as those in equation (3.5). The same procedure
applied in Figure 3.3 for the computation of the v ir curve can then be used for the A. i, curve
— —

computation.
 Chapter 3. On Modelling Iron Core Nonlinearities 38

with constants a,, b,k and c, known, and rk =1/Lk to be computed. It can be shown that
4
a > 0, 1i
4 > 0 and Cik <0. Since 1 must be positive, then

— —b,k +,,Jb, —4a, Cik

r’k— . (3.16)

The peak current 14 is computed from

‘1* J*-i +f’k(Ak —Ak_I).

In this fashion, the peak values of the inductive current are computed directly for every
segment in the A—i, curve.

3.3 Comparisons Between Experiments and Simulations

Laboratory experiments were performed with a silicon iron steel core assembled in

an Epstein frame [33]. No-load losses and rms current at 60 Hz, were measured for
different voltage levels (Table 3.1). For comparison purposes, the initial magnetization curve
[50] for the core material, was measured as well (Appendix B). The computed v—i,. and

A — i, points (including core losses) are shown in Table 3.2. The measured and the calculated

points (connected by straight line segments), with and without including the core losses, are
shown in Figure 3•42 The computed v—i,. points connected by straight line segments (the

first two columns of Table 3.2) is shown in Figure 3.5.

It can be seen that the computed A i, curve is closer to the measured one if we
—

consider the core losses. The V—I,. curve (Figure 3.5) is nonlinear and this may be important

when modelling transformers and reactors for transients or harmonic studies.

2 Sometimes, due to measurement errors, V I and no-load loss curves may be crooked and need to be
-

smoothed. The developed algorithm checks the presence of “noise” and, if it exists, a low pass Fourier filter
is used to remove the “noise” from the input data.
 .---1

D4
3
>PP
0
.

000G0o00
Dcoo0c0
000
----
0cooc
0)0
0Q’00Q00
ON 0000 ‘.D40 ‘
000000000o0
CD

00000000Q0 CD
‘4)
— ‘ — 0 0 0 0 0 0 0

000
0000c’-’ 0
-t
C
0
000
0p0p0pcpp
V1 I-,J t) •
0 0 0 0 0
CD —aQN40
0j-o CD
•
ON (Ji ON 000 .,
PPPpOpppopop
•00000000000
0 0 00 O’ 0 - -.4.

0000. 0
4)00 CD 0-
CD

-
- t4) — 0 0 0 0 0 0
—
I-
p p p p p p p p p p p . ON 00 0 00 - ‘0 I..) ON - 0 ci
0 M . t 0 0000 . 0)0—
D00c.’i0 CD 000000
.-‘ 0 .- —CI . t4) \O 0 0
 Chapter 3. On Modelling Iron Core Nonlinearities 40

0.10
— ____e

0 fAO
v.vo
>
—

/
S..—
...

ci
0) 0.06
- ./ 0 Losses included
0.04 1 A Losses not Included

0.02 - f • Measured points

I I I
0.20 0.40 0.60 0.80 1.00
Current (A)

Figure 3.4: A. —i
1 curve.

40.00 —

30.00 —

—.5

>

20.00 —

0
o -

>
10.00 —

111111

0.05 0.10 0.15 0.20 0.25

Current (A)

Figure 3.5: Computed v — Er curve.

 Chapter 3. On Modelling Iron Core Nonlinearities 41

3.4 Ferroresonance Simulations and Field Test

The BPA System of Section 2.4.1 is simulated again, Now, the transformer

excitation branch is represented by a nonlinear resistance (curve represented by three

piecewise-linear segments in Figure 3.6) in parallel with a nonlinear inductance (Figure 3.7)

produced by the described algorithm, from the transformer manufacturer data. A straight

line segment, with a slope of 4 times the short circuit inductance was connnected to the last

segment of Figure 3.7 to represent the air core inductance. The voltage waveforms at the

transformer terminal at phase A line side, are shown in Figure 3.8. The simulations were
made assuming the excitation resistance to be constant (Rc=4.2M2 - see dotted curve) and

assuming a nonlinear resistance represented by Figure 3.6 obtained from the algorithm

developed in this chapter (thin solid line). Simulations come closer to the field test (thicker

solid line) if the nonlinear losses are taken into account.

1000

800

>
600

0)
z 400
0
>

200

0.10 0.20 0.30 0.40

Current (A)

Figure 3.6: Nonlinear resistance.

3 field test curve was obtained using a digitizer to copy the data points from an oscilograph plot.
Simulations were also made using a transformer air core inductance of 3 times the short circuit inductance,
but the agreement with the field test was not so good.
 3000 — Chapter 3. On Modelling Iron Core Nonlmeanties 42

Current (A)

Figure 3.7 Nonlinear inductance

Field Test

Constant Resistance
1200
Nonlinear Resistance

::(
10

,
-400

-800

I I’ I ‘ I
20 40 60 80 100 120 14
Time (ms)

Figure 3.8 Ferroresonance in a power system.

 Chapter 3. On Modelling Iron Core Nonlinearities 43

3.5 Distribution Transformer Saturation Curves

In Figure 3.9, one can see the nonlinear inductances curves (2—i curves) for 3

distribution transformers (50 kVA each), from different manufacturers, computed with the

developed algorithm from tests performed at the low voltage terminals (120 V). The no-load

losses were measured as well. The nonlinear resistance curves are shown in Figure 3.10.

Transformer A has higher inductance in the unsaturated region and saturates at higher flux

level, also it has the lowest no-load losses.

The , —I,,,,.. curve for a brand new distribution transformer is shown in Figure 3.11.
The excitation rms current is not monotonic and drops as the voltage increases up to the
vicinity of the rated voltage. In the unsaturated region, the transformer core has such a high

permeability that current through stray capacitances tend to cancel out the magnetizing
current.
0.6

C,)
>0.4

A
B
0.2
C

5 10 15 20 25
Current (A)

Figure 3.9: Nonlinear core inductances.

 Chapter 3. On Modelling Iron Core Nonlinearities 44

200

150 ..•‘

..— —
— —
/ ,- —
—‘ / / 7
/ //
//
ioo- s
//
7/’ A
B
/

Current (A)

Figure 3.10: Nonlinear resistances.

150 —

>
100-
0)
7)
-I
0
>
50-

2 4
rms Current (A)

Figure 3.11: Newly manufactured transformer —J curve.

 Chapter 3. On Modelling Iron Core Nonlinearities 45

The developed algorithm will not work for this case, unless stray capacitances are

known. The algorithm is modified by inserting —WCVksinO, where C is the open circuit

capacitance, in between each parenthesis of equation (3.14). Then,

IiWk =[s:1.nb_w Cvsinoj dO +

J82[I + Ak0
— CVk sinOdO+ ... + (3.17)

+ smn
f[1k1
— a) CV sin oj do].

Equation 3.17 is rewritten in the form of 3.15, and the inductance Lk computed for
each segment k. In Chapter 6, a crude method of estimating the capacitance C is briefly

described.

3.6 Summary

A direct method for the computation of iron core saturation curve (A I,) has been
—

presented. It is based on the transformer test data. It is a modification of previous methods,

with core losses taken into account. Besides the A—i, curve, it produces a nonlinear v

curve as well. Comparisons between laboratory measurements and simulations were made.
It was shown that more accurate A i, curves can be obtained if losses are included.
—

Ferroresonance simulations were carried out. Simulations come closer to field tests if
the nonlinear v i,. curve is taken into account.
—
 Chapter 3. On Modelling Iron Core Nonlinearities 46

The — i, and v — ‘r curves can be used for modelling transformers and iron core

reactors in electromagnetic transients and harmonic loadflow programs.

 Chapter 4

Saturation Curves of Delta-Connected Transformers

from Measurements

4.1 Introduction

In the algorithm presented in last chapter (Section 3.2), we assumed all odd har
monic current components to be present in the measured values. For three-phase

transformers, the standard excitation test data available are the positive sequence Vrrns -

Irms curves, and no-load losses. In Figure 4.1, we show a symmetrical three-phase voltage
source supplying a no-load delta-connected transformer. The delta branches consist of non

linear elements. In general, excitation tests are carried out with a closed delta [52]. In that
case, ammeters, placed in series with the line, will not “detect” the triplen harmonic currents,
because these circulate in the delta connection. In the next sections, we develop a method
for generating the piecewise linear saturation curves (nonlinear resistance and nonlinear
inductance), which accounts for the fact that triplen harmonics circulate in the closed delta,

but do not appear in the measured line currents.

47
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 48

i(t)

Figure 4.1: Delta-connected transformer

positive sequence excitation test.

4.2 Basic Considerations

In the circuit of Figure 4.1, the three branch elements of the delta connection are

assumed to be nonlinear and identical. The branch currents can be written as a Fourier time

series containing odd harmonic components only. Then:

Iab(t) = I sin(o t)+ 13 sin(3 o.J t)+... + Ii,, sin(p Co t)+...

ibC(t) = I sin(oit—120
3 °)+1 sin(3(cot—12O°))+”.+I sin(p(wt—120°))-i-... (4.1)

Ica(t)= J sin(ot+120°)
3 +1 sin(3(Cot+120°))+” +I, sin(p(w t+120°))+”.,

where p is odd.

The triplen harmonic currents (13, 19, ...) are in phase (zero sequence harmonics).

The
rms current in each branch is
/12÷12++12+
(4.2)
‘A-rtnsj
2
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 49

The line currents are:

‘a (t) 1(t)Ica (t)
,1o
ib(t)i(t)—i(t) (4.3)
i
(
0 t) =Ica(t)4,.(t)

Substituting (4.1) into (4.3) gives:

5 fl(50)t+30°)+
1a(t)=S11(0)
5 ”
t3O°)+[

ibQ)=’f1l sin(wt —15O°)+’II5 sin(5wt+150°)+•• (4.4)

sin(t+ 90°)+ sin(5ofl _900)+..

Ammeters, placed into the supply line (outside the delta) read the rms current,

112+ 2+ 2
i-II’
V
p
Inns

From equations (4.1) to (4.5) one can make the following observations:

• triplen harmonic currents, although present in each branch, are not present in the
line currents.

• if triplen harmonic currents in each delta branch are removed from the mis value
(equation (4.2)) and scaled by 1, the rms line currents (equation (4.5)) are

obtained. This is the basis of the algorithm developed next.

4.3 Saturation Curves

Each delta branch in Figure 4.1 is represented by a nonlinear inductance in parallel

with a nonlinear resistance (Figure 4.2). Their nonlinear characteristics are computed with

the same assumptions made in the previous chapter.’

‘Branch currents in the delta will have a “1k” subscript and line currents will have no subscript added. For
example, ir4 and ‘r are the resistive components of the current in the delta branch and in the line,
respectively.
Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 50

jca(t)

(a)

it

(b) (c)

Figure 4.2: Core representation:

a) Nonlinear elements;
b) VlrA curve;
c) A.—i curve.
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 51

The algorithm works as follows:

1. For the construction of the v—4A curve (Section 4.3.1):

• compute the peak values of the branch current 4, i,... point by point from

the no-load losses.

2. For the construction of the A—ij curve (Section 4.3.2):

• from the v—i,,. curve, compute the rms values ‘ri, remove the triplen

harmonic currents and obtain the resistive line current Ir-r,,,;

• obtain the rms values I,_ of the line current due to the nonlinear inductance

from i,., the total line current and the applied voltage v;

• compute the peak values of the inductive current ,-A2 point by point

iteratively.

4.3.1 Computation ofthe v irA curve -

Similarly to the previous chapter, let us assume that the three-phase no-load losses
J, 1, .
..,
F,, are available as a function of the branch voltages ,, V, .
..,
V

(Figure 4.3).

If we assume that the applied voltage is sinusoidal, the conversion of rms voltages to

peak values (vertical axis of Figure 4.2b) is simply

kVrmsk%J (4.6)

for k1,2,...,m.
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 52

3
P

2
P

1
P

\/rmsi V’r,ns Vrms3 V

Figure 4.3: ,- power loss curve.

Due to symmetry reasons, voltage and current waveforms need only be evaluated

over 1/4 of a cycle. For a sinusoidal voltage v(O) =Vk sin 0, the three-phase active power ]

can be written in the form

:
2

F)=3 [JV(O)rA(O)dO) (4.7)

For the first linear segment in the v—i, curve, the current is sinusoidal. The

computation of the first peak current ‘rtSl is therefore straightforward. Since P 1 I in

1 =3V

the linear case,

(4.8)

From the second segment onwards (k2), equation (4.7) is evaluated at each
segment k, with only i being unknown, as explained in more detail in Section 3.2.1 of the

previous chaptçi. The computation of the peak current is done segment by segment,

If the voltages have no harmonics, the active power Pk is produced only by the fundamental component of
the current. This component is present in each branch as well as in each line. So, it does not matter if the
wattmeters are connected in series with the line or with each delta branch. The three-phase power readings
would be the same in both cases.
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 53

starting with 2
‘rA and ending with the last point i,. Whenever a point i is found, its rms

value is calculated as well. A Fourier program (see Appendix C for the algorithm) is used to
compute the triplen harmonics (13, 4, ...). They are then removed from to obtain

‘r-rms, which is needed later for the construction of the 2—E, curve.

4 curve
4.3.2 Computation of the 2 i, -

The conversion of the peak branch voltages k to flux Ask, is again a re-scaling

procedure. Hence, for each linear segmcnt in the 2—i curve,

(4.9)

Let us now compute the peak values of the currents i through the nonlinear

inductance. First, the rms values of the line currents 4. are evaluated with
I,_, =4(It_)2_ (1)2 (4.10)

where the line current I_ is available from the measurements, and where ‘rl1L has already

been computed from the previous section.

For the first linear segment, the computation of ii is straightforward since there are

no harmonics yet. Therefore,

qf
I
5 , _

From the second segment onwards (k2), the algorithm works iteratively as follows

(see A.—i,, curve in Figure 4.4):

1. guessig;

2. with A(O) = 2
” k sin 0, find 1/4 of a cycle of the distorted current analytically;
Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 54

3. compute the rms inductive branch current whose peak is ig;

4. use a Fourier program to find the triplen harmonic inductive currents in the delta
branch (Appendix C);

5. remove the triplen harmonics from the estimated rms branch current. Scale the
estimated result ‘lest by and compare it to Iltms in equation (4.10);

6. if the absolute value of the difference IIlr,ns ‘lest is less than the specified tol
-

erance, convergence is achieved. Otherwise, the residue d

1 is added to ig and
the iterative process is repeated from step 2 onwards.

correct curve

It
2

01

It
2

Figure 4.4: Generating current wavefor

from sinusoidal flux.
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 55

In general, convergence is achieved in less than 20 iteration steps with a tolerance of

1 0’ iz and initial guess

+
/(Ii_rms& 1—nnsk1)
1
‘g =l1_j

for every k 2. In some cases more than 20 iteration steps may be necessary.

4.4 Case Study

Consider a 50 Hz three-phase five-legged core type transformer. The following

information is known [53]:

1. rated power 750 MVA (three-phase);

-

2. rated voltages 420 kV/ 27 kV (line to line values);

-

3. wye connection on 420 kV side, delta connection on 27 kV side.

The positive sequence excitation test data, from the closed delta 27 kV side, are

shown in Table 4.1.

V, is the rms line to line excitation voltage, is the rms excitation current

(three-phase average) and P are the no-load losses (three-phase values).

Table 4.1: Three-phase transformer test data

/ Vr
(kV) (A) (kW)
22.76 8.20 206.21
24.29 11.35 240.26
25.64 15.50 270.13
27.00 21.16 311.00
27.50 24.68 323.03
28.47 31.63 355.48
29.10 38.30 385.41
32.50 — 80.97 560.00
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 56

The two computed A - curves (points connected by straight line segments) are

shown in Figure 4.5. One of them assumes that all odd harmonic components of the current
are present in the measured values, and is therefore incorrect. The other curve is the correct
one; it has been produced with the algorithm of Section 4.3.2.

160.0

140.0

120.0
(/)

>

0)
o
- 80.0
C
-j

60.0
-—9—- “incorrect”
40.0 “correct”

20.0

20.0 40.0 60.0 80.0 100.0 120.

Cu rrent(A)

Figure 4.5: - i1z curve.

It can be seen that the correct curve goes deeper into saturation. For the highest flux
value, there is a difference of approximately 14% between the peak currents of the incorrect
curve and the correct one. The piecewise linear v - Ir1 curve is shown in Figure 4.6.
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 57

50.00

40.00

30.00

3 20.00
>

10.00

2.0 4.0 6.0 8.0 10.

Cu rrent(A)

Figure 4.6:V—ir1. curve.

In the iteration scheme of Section 4.3.2, harmonics up to the th

99 order were

included, which is more than needed in practice. An average of 23.86 iteration steps was

necessary (the maximum number of iterations was 34). In order to check the numerical
accuracy of the method, the mis line currents were recomputed back from v i and A -
-

curves. Numerical errors were found to be very small (less than 0. 001%).

4.5 Summary

An approach to the computation of instantaneous saturation curves of delta-

connected transformers has been presented. It uses positive sequence excitation test data as

input, and is suitable for situations in which the tests are performed with a closed delta.

For the case study presented in Section 4.4, it was shown that a difference of

approximately 14% between the peak currents of the incorrect curve and the correct one
occurred. The last flux linkage point is around 1.2 p.u. In fact, for transient studies, it is
 Chapter 4. Saturation Curves of Delta-Connected Transformers from Measurements 58

often necessary to know peak flux-peak current curves beyond that point. The usual way is
to extend the curve up to a value necessary for the study (this extension is sometimes done
with a straight line passing through the previous to the last and the last point in the peak
flux-peak current curve). This may lead to larger errors for the peak values of the current.
The curves diverge as the flux goes up towards deep saturation. The transformer

magnetizing current would always be underestimated if triplen harmonics inside the delta
windings were not taken into account. Errors can also affect the air core reactance value. A
parametric study was done considering typical air core reactances from 0.2 p.u. to 0.5 p.u.,
connected to the last point of the correct curve of Figure 4.5. Errors on the slope of the
saturation characteristics, for this case, are between 18% and 25%, when the magnetizing
curve reaches the transformer rated current. These differences may be important in
ferroresonance or inrush current studies

For the development of the algorithm in this chapter, it was assumed that each phase
behaves independently. This is valid when the transformer is saturated only. The current in
the saturated phase is much larger than the current in the remaining phases. The saturated
phase can be considered “decoupled” from the other phases.
 Chapter 5

Hysteresis and Eddy Current Losses in Iron Core

5.1 Introduction

In this chapter, a general discussion of hysteresis and eddy losses in iron core is

presented. RL networks, in which the inductances are nonlinear and the resistances are
linear, are developed to model the nonlinear and frequency-dependent effects of the
transfonner core (saturation, eddy currents and hysteresis). It is assumed that the core loss is
known as a fhnction of frequency. Simulations are compared to laboratory measurement of
inrush current and to a ferroresonance field test.

5.2 Frequency-Dependent Core Models

Rosales and Alvarado[l 8] represented eddy current effects in the core by solving

Maxwell’s equations within the laminations assuming that the permeability and conductivity
of the material were constant. They derived expressions for the lamination impedance Zi(jo)
and admittance Yi(jo) in the forms:

Z
(
1 jw)= 1tanh(), (5.1)

* 5c
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 60

and

Y,(jo)—= (1/ 1
R
) eoth() (5.2)

where

R,=- =d%Jjo4w

and

1= lamination length;

w = lamination width;

2d = lamination thickness;

= magnetic permeability of the material;

a = electric conductivity of the material.

ns
From the expansion of the hyperbolic tangent in (5.1) into partial fractio
1
tanh=2 2’
(5.3)
kI
2
+ [r (2k — 1)/21

ion of the hyperbolic

they realized a series Foster-like linear circuit and from the expans
accuracy of the
cotangent, they realized the parallel Foster-like circuit of Figure 5.1. The
n expansion.
representation depends on the number of terms retained in the partial fractio
 Chapter 5. Hysteresis and Eddy Current Losses in fron Core 61

1
R

Figure 5.1: Eddy Current representation of the core after [18]

Tarasiewicz et al.[17] used (5.1) and expanded the hyperbolic tangent in continued fraction

form,

tanh= , (5.4)
1+
3+
5+
7+ ,
9+

and realized the ladder network of Figure 5.2, known as the standard Cauer circuit.

1
R 2
R 3
R

Figure 5.2: Eddy Current representation of the core after [17]

They observed that the number of terms to be retained in the continued fraction of (5.4) is

smaller than the number of terms to be retained in the partial fraction of (5.3) for the same

frequency range and verified that the four sections of the continued fraction model of Figure
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 62

5.2 would be enough to reproduce the exact impedance of (5.1) for a frequency range up to
200kHz with an error less than 5%. They also noticed that to achieve the same accuracy, 72
sections of the Foster-like circuit of Figure 5.1 would be required. Based on that, they
decided to work with the standard Cauer circuit. However, it seems the authors were not
aware that once the circuit parameters of Figure 5.2 are known, a Foster-like equivalent
circuit which reproduces the same frequency response with exactly the same number of
elements can be realized as well (later on in this chapter, network realizations will be
addressed) [61]. Therefore, it seems there is no apparent gain in choosing this ladder network
to represent eddy current effects in the core.

Dc Leon and Semlyen[20] suggested the Cauer circuit of Figure 5.3. The RL
parameters were obtained by a nonlinear fitting process to match (5.1). This circuit can be
interpreted as a discretization of the lamination. The inductances represents the flux paths
and the resistances produce the eddy loss. At DC excitation, the current flows through the

magnetizing inductances lying longitudinally (in the previous circuits, the DC current flows
through one inductance only). This could be interpreted as a uniform flux distribution in the
entire lamination at DC level.
1
L L

Figure 5.3 : Eddy Current representation of the core after [20]

5.3 Core Loss

In the algorithm developed in chapters 3 and 4, the excitation branches of

transformers were represented by a nonlinear resistance in parallel with a nonlinear

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 63

inductance. Thus, hysteresis and eddy current loss were assumed to have the same
frequency-dependence. It is clear that the circuit would not reproduce the core loss for
frequencies different from the rated frequency.

A more detailed core model should take into account the nonlinear and frequency-
dependent effects of saturation, hysteresis and eddy currents. It should be emphasized that
information about how eddy and hysteresis losses are split, is in fact artificial. Today, the
point of view of many researchers is that virtually all the observed losses are resistive losses
associated with micro eddy currents due to Barkhausen jumps of domain walls [64,65], no
matter how slow the magnetization ioops are traversed.

Core models in which hysteresis and eddy currents are treated simultaneously are
developed in the next sections. A Foster-like circuit is used to represent both, hysteresis and
eddy current effects. First, consideration is given to the core operating in the linear region.
Nonlinear effects are considered later.

5.4 Eddy Current and Hysteresis Modelling

There is a property of linear passive networks in which the knowledge of the real
part of an impedance Z(s) for sjw (Re (Z(jco)), completely determines Z(s) [66]. If (Re
(Z(jo)) is a given rational function of frequency w, then one can construct the corresponding

rational function Z(s) in terms of the complex frequency variable s. Once Z(s) is known, it
can be realized as a passive network [61,63,66]. The same process applies to the admittance
function Y(s), as described next.
 Chapter 5. Hysteresis and Eddy Eurrent Losses in Iron Core 64

5.4.1 Construction of Y(s) from Its Real Part

The process of constructing Y(s) or Z(s) from their real parts is well known in

network synthesis theory and only a brief discussion is presented here. For more details, the
I
reader is referred to [61,66].

Consider first a simple circuit composed of only two linear RL elements connected

in parallel (Figure 5.4a).

A

(a) (b)

Figure 5.4: Core parameters.

Its admittance Y(s) could be written in the form

Y(s) = +

It follows that

1 Y(s) + Y(—s)
G (5.5)
R 2

If the elements are frequency-dependent, G(co)= 1/ R(o) Re[YU)], or with sjw,

= Y(s) +Y(—s)
Re{Y(s)] (5.6)
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 65

Now, consider the exciting branch of a single-phase transformer to be linear (the

extension of the method to include saturation effects will be done later). One can visualize

the core operating below the knee of the saturation curve. In this region, the saturation

curve could be approximated by a linear segment (dashed line of Figure 5.4b). The exciting
branch can be represented by a frequency-dependent lossy element R(co) in parallel with the

magnetizing inductance L(w). The relation which ties R(co) to L(o) is shown below.

First represent Y(s) as a rational function containing n pairs of poles and zeroes:

N(s) (s + z
1 )(s + z
).
2 (s + Zn)
(s) K 5 7)
—

D(s) (s+ p)(s + p

1 (s + pj’
—

— —
)•
2

i and z are real and positive, i.e., the poles and zeroes of Y(s) lie on the left hand side of the
s plane. From (5.6) and (5.7), the real part of Y(s) is expressed by:

Re[Y(s)] E
1 N(s)D(—s) + N(—s)D(s)
= D(s)D(—s)

or

— U(S2) (_s2 u)•+ u)(_s2 + .

(—s + u)
Re{Y(s)] K
.

W(s2) ‘ (5.8)
2 + p)(—s
(—s 2 + p)• 2 + p)
(—s
—

— —
.

where u is also positive and real [621. Expanding (5.8) into partial fractions

K K K
Re[Y(s)]=K
,
0 , (5.9)
(
(s2 2) 2 •
2)
+ + +

Further expanding each fraction of the equation above into two partial fractions

+ K/
1 + Kn + K,, I 2p,,
Re[Y(s)] = K, +
•

(s+ p
)
1 (—s+ p
)
1 (s + P) (—s + p,,)
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 66

or

(K K /2p K?‘ 12p (K K /2p

1 K /2p“
Re[Y(s)]=I—-+ I +.“+ (5.10)
j2 )
1
(s+p (s+p)} 2 )’
1
(—s+p (—s+pj)

Comparing (5.6) and (5.10) it is clear that the admittance function is given by:

(5.11)
)
1
(s+p )
2
(s+p (s+p)

From (5.11), one can also write the admittance Y(s) in the form:

a s” i-a2 + a s+a
(5.12)
bs”+•. 2
0
+Is+b

It can be seen that if the function G(w)=Re{YC,)] is known, Y(s) is determined.’ It

should be pointed out that G(w) must be expressed as an even rational function in w
2 with
real coefficients in the form of (5.8) with w
= -s
2 2 and 0G(co)<oo for all frequencies [66]. A
curve fitting procedure to obtain G(o) or R(o)) in the required form, from one of the core

loss frequency curves supplied by steel lamination manufacturers, is explained in Appendix

-

D. It is recommended that the data be taken for an induction level, at which the lamination is

not saturated.

5.4.2 Linear Network Synthesis

From Y(s), any of the RL equivalent networks of Figure 5.5 can be realized with the

same minimum number of elements. If for any of these circuit, the parameters are known,

Y(s) or Z(s)=1/Y(s) could be determined and the remaining circuits could be realized as

1 The construction of Y(s) from its real part in the partial fraction form of (5.11) was first suggested by
Bode[58]. There is a method to compute the coefficients a and b in (5.12) directly from (5.8) due to C.M.
Gerwetz [57]
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 67

well. It must be clear that all these circuits are equivalents only with respect to their

terminals, i.e., the network elements of each circuit must have the prOper values to produce

the same terminal impedance or admittance as a function of the con4lex frequency s.

Let us describe the algebraic process for realizing the parallel Foster-like circuit. The
terminal admittance of the RL network of Figure 5.5 a is

1 1/L
1
Y(s) = + (5.13)
0
rç /L,
1
s+R
-—

1=1

This circuit could be realized directly from (5.11) with

R L and
1
‘K

1
R 1
R
R R R

1
L 1
L

a) Parallel RL Foster-like network b) Series RL Foster-like network.

L 1
L
R

c) Physical Cauer network d) Standard Cauer network

Figure 5.5 : Realization ofRL networks.

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 68

For the Cauer circuit of Figure 5. 5c Y(s) is written in the continued fraction form:

1 1
1 (5.14)
Ls+
1 1
R
s+
1
L_

1
1
1
RIs+R

Equation (5.14) could be derived from (5.12) according to the following steps:

a from (5.12) Y(s) is split into two terms by removing the constant a,/b;
-

a a s’’ +• • + a’s
2 + a’s + a’
Y(s)=--+ (5.15)
+
2
“s+•••+b
+
1
°
0 bs

or

b bs+•”+b 2 +bis+b
s
2 0
s’ + +as
1
a 2 +as+a

b - removing —
4 s from the denominator of the second term of the equation above, results
1
a_

1
(5.16)
1
.

1
a a,
s
1 ’ +• ‘+ as
2 + as + a
‘

b’ s’+•’•+b’s
+b’s+b
2 2 1

Now, the process of long division is done for the term between the brackets ( steps a and b
are repeated). For each cycle, a constant (step a) and a pole at s=oo(step b) are removed.

The process continues until the order of the polynomial in the last continued fraction is one.
For the remaining circuit realizations of Figure 5.5, the reader is referred to [61].
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 69

A remark must be made about the calculation of Y(s). Consider the

Foster-like
circuit of Figure 5. 5a with an inductance Ld placed across its terminals
(Figure 5.6). It is
clear that Ld does not affect the conductance G(o); it affects only the
susceptance. In the
process described earlier, Y(s) is the minimum admittance function that produc
es G(o). The
full admittance is given by Ld in parallel with Y(s).

Figure 5.6: Frequency-dependent representation of the core.

5.4.3 Iron Core Nonilnearities

Since the resistances and inductances in the circuit of Figure 5.6 do not represent any

physical part of the core, it is not clear how to incorporate nonlinear effects. Considering

that low frequency elements contribute more to saturation than high frequency elemen
ts,
only the inductance Ld is made nonlinear. The Foster-like circuit was chosen becaus
e it has
only two nodes, therefore, it is computationally more efficient than the remaining circuit
s of
Figure 5.5.

5.4.4 Numerical Example Hysteresis -

It is appropriate to illustrate how the model works for an example in which a core
is
assumed to experience only hysteresis loss (this is an arbitrary case since hysteresis and
eddy
losses cannot be separated). Initially, suppose hysteresis could be reproduced by a resista
nce
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 70

in parallel with a nonlinear inductance. For a sinusoidal applied voltage, at rated frequency,

the flux-current loop of Figure 5.7 is obtained. Let us keep the flux amplitude constant.
2 For
any frequency, the area inside the flux-current loops should be the same and the loss is a
function of how fast the trajectories are traversed. For a frequency equal to twice the rated i

frequency, the voltage doubles. So, hysteresis resistance would have to be twice the

hysteresis resistance at rated frequency to produce the same loss per cycle. For any
frequency, the resistance R should be replaced by an equivalent frequency-dependent

resistance R(o.) according to:

R(O))Rr, (5.17)

where cor is the rated angular frequency and 14 is the hysteresis resistance at rated

frequency.

Figure 5.7: A—i hysteresis curve.

Let us now represent the core by a two slope A-i curve, defined by points in Table 5.1, and

a resistance Rr=100 2 at rated frequency (60Hz). Using the method described in the
previous sections with Wh>>Wdy (Appendix D), the Foster-like circuit parameters of Figure

2 This could be acomplished by keeping constant the ratio between the amplitude of the voltage signal and
the frequency.
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 71

5.6 are found for a frequency range from 60 Hz to 3 kHz. The fitted frequency-dependent

resistance and its value from (5.17) are shown in Figure 5.8.

Table 5.1: Flux-current curve

Current (A) Flux (V.s)

0.0000 0.0000

20.0000 0.4000

300.0000 0.6000

5—

Exact Curve

I 31

I ‘I

Frequency (kHz)
Figure 5.8: Frequency-dependent resistance.
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 72

0.5 0.5

0.0 0.0

-0.5 -0.5

-50 0 50 -50 0 50
Current (A) Current (A)

a) Constant resistance b) Frequency-dependent resistance

Figure 5.9: Flux-current loops.

The flux-current plots shown in Figure 5.9 were obtained from time-domain simulations
using Microtran®[431. Sinusoidal flux linkages of constant amplitudes and frequencies of 60

Hz, 180 Hz and 300 Hz, were used for the simulations. In Figure 5.9a, the loops were

obtained considering the core loss represented by a constant resistance and, in Figure 5.9b,

with the frequency-dependent model developed here. In Figure 5.9b, the area of the loops

are nearly independent of the frequency, however, the flux-current loops tilt slightly

clockwise as frequency increases. In a qualitative way, one could interpret that an increase

of the core resistance is associated with a decrease in the core inductance caused by the

flux being pushed away from the center of the lamination.

5.5 Inrush Current: Simulation and Measurements

A single-phase 60 Hz, 1 kVA, 208/104/120 V laboratory transformer, in which the

core is made of non-oriented silicon steel laminations, was tested. From short-circuit tests,

with the voltage source connected’to the 104 V side:

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 73

1. AC resistance measured with the 120V side short-circuited 0.60 2; -

2. Leakage reactance with the 120 V side short circuited 0.19 2; -

3. DC winding resistance 0.40 2.

-

The points (empty circles) of the flux-current characteristic of Figure 5.10 were

obtained from no-load loss measurements and V,-I values, using the method developed

in Chapter 3. When the core is operating in the unsaturated region, its inductance can be

roughly approximated by 2.4 H. The core loss at 60 Hz is represented by a constant

resistance of 1.8 k 2. All parameters are referred to the 104 V terminal.

The Foster-like circuit linear parameters shown in Table 5.2, was calculated for a

frequency range from 60 Hz to 3 kHz assuming the ratio of hysteresis to eddy loss per cycle

to be W1
/Weddy=2 at 60 Hz [67] (it does not make any difference in the simulations if this
1

ratio is taken as 1 or 3). The transformer was initially demagnetized and later, a voltage

source was switched in the 104 V terminal. The voltage signal (Figure 5.11) was saved as an
ASCII file and used as input in Microtran®. The inrush current waveform was measured as

well. Plots of the simulated current using two core models (Foster-like circuit and a constant

resistance in parallel with a nonlinear inductance) and the measured current using a time step
of 50 l.ts, are shown in Figure 5.12. For the study, a straight line segment was connected to

the last point of the curve of Figure 5.10. Its slope was chosen to match the first peak of the

measured current, when the flux linkage reached its first peak (the flux waveform was

obtained integrating the voltage waveform of Figure 5.1 1).3

3 By extending the last segment of Figure 6.10, it was found that the first current peak in the simulations
was underestimated by nearly 5%. The winding resistance, used for the simulations, was 0.45 12 which is
slightly above the measured value.
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 74

Table 5.2: Linear circuit parameters

Foster-like circuit parameters /

Ld=2.4000H = 1722.4899
1
R

L
=
14.1518H R
=
2 22264.6946 2

L
=
2 4.6595H R= 5271.37022

In general, the agreement between measurement and simulations, for any of the used

models, is good. Figure 5.13 is a magnification of Figure 5.12 for the time interval between

0.15 and 0.20s. The two core models produce nearly the same response and are in good

agreement with measurements. Another simulation was performed representing the core by a

nonlinear inductance only. The current waveform was nearly the same as the constant

resistance curve of Figure 5.13.

0.5

0.4

0.3

0.2

0.1

5 10 15 20 25
Current (A)

Figure 5.10: Flux-current curve of a single-phase 1 kVA transformer.

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 75

160 —

120 —

80 —

40
—

0—

-40 —

-80 —

-120 —

—160—
I I I
0.05 0.10 0.15 0.20 0.2
Time(s)

Figure 5.11: Transformer input voltage.

140-

120 —

100-
Constant Resistance
Measured Curve
80 —

Foster-like Circuit

:::
60-

I I
I
0.05 0.10 0.15 0.20 0.25
Time(s)

Figure 5.12: Transformer inrush current.

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 76

8
- — — — —

- Constant Resistance
Measured Curve
6
Foster-like Circuit

0.15 0.20
Time(s)

Figure 5.13: Transformer inrush current.

5.6 Ferroresonance

The BPA system of Section 3.4 is simulated once more. Now, the transformer

exciting branch is represented by a Foster-like circuit with the parameters calculated for a

frequency range from 60 Hz to 3 kHz, assuming the ratio of hysteresis to eddy loss per

cycle to be Ww’Wdd=l at 60 Hz [70]. The simulation and field test results are shown in

Figure 5.14. The Foster-like circuit produces nearly the same response as a linear resistance

in parallel with a nonlinear inductance (see Figure 3.8 in Chapter 3).

 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 77

- Field Test
1200 - Foster Circuit

800

400 -

0-

-400--

-800-

‘ I I I I
20 40 60 80 100 120 14
Time (ms)

Figure 5.14: Ferroresonance in a power system.

5.7 Summary

An approach to model frequency-dependent effects in the transformer core from core

loss data was presented. The parallel Foster-like circuit model, in which hysteresis and eddy

current effects are treated simultaneously, was realized. Theoretically, the circuit models can

be used for any frequency range (minimum frequency must not be zero).

If the circuit model is used to represent a core in which the hysteresis loss is

dominant (e.g. amorphous core [221), the core loss per cycle is nearly independent of the
 Chapter 5. Hysteresis and Eddy Current Losses in Iron Core 78

frequency. The flux-current trajectories are generated by the circuit models with no need to

pre-define them.

An inrush current case and a ferroresonance case were used as test examples. For

both cases, there were no major improvements in representing frequency-dependent effects.

in the core. For the inrush current case, the simple model in the form of a nonlinear

inductance reproduced the measured current reasonably well. For the ferroresonance case, a

nonlinear resistance (Chapter 3) in parallel with a nonlinear inductance represented the core

reasonably well.

It is still premature to say that frequency-dependent effects in the core are not

important for all transient cases. More validation tests against other field measurements need

to be made in the fhture, to filly determine their importance for practical applications.
 Chapter 6 /

Transformer Models Applications -

6.1 Introduction

Transformer models based on parameters obtained at 60 Hz, and suitable for

frequencies up to 5 kHz, are presented. Frequency-dependent effects in the windings and
core nonlinearities are discussed. These models were developed for harmonic studies, as part
of a research project for The Canadian Electrical Association, but may be used for transient
studies as well. To validate the models, many tests were carried out on several transformer
units of different power ratings. Sensitivity studies are performed to show how the total
harmonic distortion of current and voltage behaves as the transformer parameters change. A
field test involving a transformer, as part of a power system feeding a DC drive, is described
as well, for which simulations agree reasonably well with measurements.

6.2 Basic Transformer Equivalent Circuit

Figure 6.1 is the basic transformer model. Z(jo) represents the frequency-dependent

short-circuit impedance. The excitation branch is represented by nonlinear RL elements obtained

fron measured rms values of voltages and currents(V,mf (Lmj) and no-load losses with the

79
 Chapter 6. Transformer Models Applications 80
-

conversion technique of Chapter 3. These elements should be placed at a point in the equivalent

circuit where the integrated voltage is equal to the iron core flux. This point depends on the

transformer design and is usually not accessible in the model. Howev’er, it can be approximated

fairly accurately by connecting the elements to the winding closest to the core (usually the lowest /
voltage winding) [4].

Figure 6.1: Basic transformer model.

6.2.1 Single-Phase Two-Winding Transformer

Consider initially a single pair of magnetically coupled coils as shown in Figure 6.2a.
This network can theoretically be described in terms of self and mutual impedances

l Lz, Ti1
1
rv 12
z
,
2
[‘j[z 14j
2
z 2 => [v]—[z][i}, (6.1)

or with its inverse relationship

i Ei
1
ri
IL 211=1L’21 I = [i]=[y][v], (6.2)
221’2l

where 1
Z 2 = ZM and [Y][Z]
.
1 ’ and 12’
2
V are the secondary voltage and the
secondary current referred to the primary side, respectively. The elements of [ZJ can be

found from open-circuit excitation tests. For instance,

’
2
v
=y
2
Z 1 (6.3)
12—0
 Chapter 6. Transformer Models Applications 81
-

12’ ZSH
1 12
..........., 4

v11 1v2 vu
(a) (b)
Figure 6.2: Two winding transformer:
a) Pair of magnetically coupled coils;
b) Equivalent circuit referred to primary.

Unfortunately, in practice the magnetic coupling is so high that Z

21 is approximately

equal to Zn. [Zj is then almost singular and its inversion becomes ill-conditioned (one should

note that Z
11 22 The short-circuit impedances, which are more important than the
Z
).

magnetizing impedances in most studies, get lost. One suitable way to overcome these

difficulties is to use the branch admittance matrix [Yj of equation (6.2). Although [Z]

becomes infinite for zero exciting current, the [Y] matrix exists and can be found from

standard short-circuit tests, as shown later.

For a single pair of magnetically coupled coils, only one short-circuit impedance ZSH

exists (Figure 2b). Then, the short-circuit admittance YSH can be obtained from

(6.4)

To construct [F], the four elements )‘ii, l’12, l13, and ,

14 are needed. They can be
Y
obtained directly from Figure 6.2b.’ The 2x2 [F] matrix is then found from one short-circuit

admittance element as

1 It 11 is the sum of the admittance element connected between the two nodes (YsH)
is useful to note that Y
plus the aclniittance element connected between node 1 and the reference node (since we are neglecting the
exciting current, its value is nil). We can also see that Y
1.22
Y
 Chapter 6. Transformer Models Applications 82
-

i
1
ri EH T1
1
- 1
‘SH 1’J (6.5)

ZSH is frequency-dependent and can be reproduced reasonably well by the two branch

parallel Foster-like circuit of Figure 6.3a or the series Foster-like circuit of Figure 6.3b as

recommended by CIGRE[54]. R
5 and L are the resistance and leakage inductance of the

windings at rated frequency. R in parallel with the 60 Hz leakage inductance produces the

frequency-dependent effects.

II ‘2

(a) (b)

Figure 6.3: Frequency-dependent short-circuit impedance model.

6.2.2 Single-Phase Three-Winding Transformer

Single-phase distribution transformers usually have a grounded tap at the center of

the secondary terminals (Figure 6.4a). These transformers are modelled as three-winding

, Zpr and
transformers (Figure 6.4b). The impedances Z
8 can be found from three short-

circuit tests. Unfortunately, these test data are usually not available[4]. The only test usually

available is the impedance measurement carried out with the secondary and tertiary terminals

grounded to the center terminal. Fortunately, it appears that an accurate value of the short

circuit impedance between primary and secondary is not needed for practical applications
Since the current flows mainly through the impedances Zps and ZPT (see Figure 6.4b). It is
 Chapter 6. Transformer Models Applications 83
-

therefore reasonable to assume Zps ZPT Zsr referred to the same terminal. In order to
produce the frequency-dependent effects in each winding, each branch Z
,
3 ZPT and ZST is
replaced by the equivalent circuit of Figure 6.3. The core nonline’ar elements are obtained

from typical V, I, curves, as explained in Chapter 3.

-

P Ps
S

I (a)
T

(b)

Figure 6.4: Three-winding transformer.

a) Coupled coils;
b) Equivalent circuit.

6.2.3 Three-Phase Transformer

Three-phase transformers are best represented in matrix form. It is simple to extend

the single-phase transformer formulation to three-phase two-winding transformers. Equation

(6.2) becomes a matrix equation

[YSH]=[ZSH]’. (6.6)

For a short-circuit test performed on a two-winding transformer, the voltages and

currents in each phase of the feeding terminal are related according to

[VABC} = [ZSH][IABC]

or
 Chapter 6. Transformer Models Applications 84
-

VA ZM ZAB
zAc1 ‘A
= ZBA ZBB ZBC ‘B (6.7)
V Z Zc Zj 1c /

To obtain each element of [Zszqj in equation (6.7), the short-circuit

tests are done
according to the circuit of Figure 6.5.
‘A

Three-phase
VA
Transformer

Figure 6.5: Three-phase transformer short-circuit test. -

A voltage signal is applied to one phase at a time and the current in

that phase and
the voltages in A, B and C are then measured. The first column of [ZSH] can
be obtained by:

4
J,
I=I=O;

VB
ZBA=j-
‘B’C°

and vc

The same procedure applies to the other two columns. Due to symmetry,
only six
measurements are required to construct [Zsn-j. In practice, [ZSH] is genera
lly assumed to be
balanced (Z ZAC =ZBC and ZM=ZBB=ZCC) and the number of measurement
s is reduced to
two. Positive and zero sequence short-circuit tests are the standard data
required to build
[ZSH] 2

2 transformers rated 500 kVA and below, only positive sequence tests are
available[551.
 Chapter 6. Transfonner Models Applications 85
-

As explained earlier, since the impedance matrix may become singular, it is

convenient to use admittance matrix [Y] formulation. In fact, here we use matrices 8
[R and
j

[Lp f’ or [coLp f’ (imaginary part of [Y]). These matrices are obtained from positive and

zero sequence tests and computed by the support routine BCTRAN [4]. [Rs] is a diagonal

matrix obtained from positive sequence tests - the zero sequence resistance R is assumed

to be the same as the positive sequence resistance R

÷. The manufacturer provides either the
5

R+ in ohms or the load losses from which R

+ can be obtained. Once the matrices are
5

determined, arbritary transformer connections can be realized by node renaming.

Frequency-dependent effects in the short-circuit impedance are treated the same way

as for single-phase transformers. Here the elements of Figure 6.3a are all matrices. [Rn] is
calculated according to the equation

[Re] = k [oL], (6.8)

Where k is a constant and [oL] is the reactance matrix obtained from positive and zero

sequence short-circuit reactances. For the representation of the circuits in admittance form,

the conductance matrix [G] can be used instead of [Re], where,

[G] = [oiLJ’

The proper way to represent saturation effects is to look at the transformer magnetic

circuit. Saturation is related to fluxes in the core and tank. There are several types of core

construction. To model the core, one should know what kind of geometry the transformer

has. One should know if the transformer is constructed with three different cores, if its core

is three-legged, five legged or shell-type. A zero sequence magnetizing impedance can be

3 It can be approximated as linear, because of

estimated for each type of core construction.

This probably is not applicable to five-legged transfonners constructed from four different cores [69].
3
 Chapter 6. Transformer Models Applications 86
-

fluxes usually passing through air between core elements and the tank. The saturation curve

must be obtained for each leg of the transformer and placed across each winding at the

secondary terminal to represent the nonlinear effects[4]. If detail of core design are not

known, the positive sequence saturation curve is placed across each winding.

6.3 Estimation of Transformer Parameters

Measurements were taken to determine the values of the parameters in the

transformer equivalent circuit, and to correlate them with the nameplate data. Single and

three-phase transformers were tested at Powertech Labs and BC Hydro and Power

Authority (Table 6.1). Short-circuit impedance measurements and open-circuit saturation

tests (V,.,,,=f (I,,,,.) and no-load losses), were performed for the units listed below. The bases

of the per unit (p.u.) system used for the short-circuit impedances are rated voltage and

power.

6.3.1 Short-Circuit Tests

High frequency currents produced by the signal analyzer HP3562A

4 were injected
into the transformers under test. Short-circuit frequency response impedance measurements

(real and imaginary values) were recorded as ASCII files on floppy disks.

6.3.1.1 Single-Phase Transformers

In Figure 6.6, some of the measurements of impedances made on the primary side, with
the secondary side short-circuited and the tertiary left open, are shown. R(J) and X(/) represent

4 analyzer has a built-in signal source which can produce and measure a scanning signal well above the
frequency range of interest. Its measurement range is 6411 Hz to 100 kHz.
 Chapter 6. Transformer Models Applications 87 -

the real and imaginary parts of the short-circuit impedance, respectively. Although the

transformers have the same rated power and nearly the same short-circuit impedances at 60 Hz,
there is a considerable difference in the frequency response behaviour df transformer 1 compared
to the others. Differences in coil and core constructions may be the causes of these discrepancies.

Impedances measured on the secondary side, with the tertiary short-circuited and an

open primary, are shown in Figure 6.7. It is difficult to correlate Figure 6.6 and 6.7. X(/) in

Figure 6.7 is, for the whole frequency range, slightly bigger than X(/) in Figure 6.6

(transformer 1 behaves in the opposite way). Comparing the measurements of Figure 6.6

and 6.7, it can be seen that Zps is almost the same as ZST for transformers 2 and 3. For

practical purpose, it is reasonable to assume ZPSZST.

Table 6.1: Distribution transformers

Manufacturer (date) Power Voltage ZSH* Measured 60 Hz
(kVA) (kV) (%) Impedance
Code ZSh (%) R (%) X (%)

xformerl 1975 50 14.4 1 2.18 -

1
14.4 1
1
- -

xformer2 May 1992 50 2.10 -

1
- -

xformer3 1978 50 14.4 2.10 -

- -

xformer4 1973 50 7.2 2.30 - - -

xformer5 not available 75 14.4 2.83 2.78 1.18 2.52

xformer6 not available 100 14.4 2.10 2.03 f__0.77 1.88
xformer7 not available 167 14.4 2.10 [ 2.26 0.68 2.16
xformer8 not available 10 14.4 2.20 1.91 1.10 1.56
xformer9 not available 25 14.4 2.10 2.01 0.89 1.80
xformerlo not available f 50 14.4 2.10 1.90 0.81 1.72
xformerll f May1992 10 14.4 2.20 - -
I
[ I
-

xformerl2 ** not available 150 7.2 4.21 -

I
- -

xformerl3** not available 500 7.2_[ 5.60

[ [ ZZ[_-_I
- -

[jtrmer14** not available 75 — 14.4

- -_[

* *three_phase
* nameplate short-circuit impedance transformers
 Chapter 6. Transformer Models - Applications 88

5
xfoi±er 1 15
xfornier 1
4
xformer2
:::::::J::::::::::..xftdner.3
L
xfornier3
Olin er4
I

1 23 4 5 1 2 3 4 5
Frequency (kHz)
Frequency (kHz)
Figure 6.6 : Impedance measured on the primary side

5 15 xförm
4 vfnrnwr1
———.— i’dformer2

zz r2 ,L.. 10 ..:::::::.. .::

bcormer3 -.-.-

- dfonner4/
/
------1---. bcomier4 ‘I’ : 5 ‘I
ii /_• ,,

V.-.
..,.,

/
1 —_.--i
—,

-: .-
___±iz____t_____

1 2 3 4 5 1 2 3 4 5
Frequency (kHz) Frequency (kHz)
Figure 6.7: Impedance measured on the secondary side.

0.5 2.0

0.4
1.5
0.3
‘ 1.0
02

0.1 0.5

2 3 4 5 1 2 3 4 5
Frequency (kHz) Frequency (kHz)
Figure 6.8: Short-circuit impedances of single-phase transformers.
 Chapter 6. Transformer Models Applications 89
-

In Figure 6.8 one can see the real and imaginary parts of the short-circuit impedances
(transformers 5, 6 and 7) measured the way in which manufacturers probably supply the data to

users (measurements are taken on the piimary side with the secondary’ and tertiary grounded to

the center terminal). The equivalent impedance seen by the source is Zp in parallel with Z
1 in
this case. Measured and simulated L/R curves are shown in Figure 6.9. The solid curves are

obtained from measurements. The simulated curves are produced by two parallel circuits of the

type ofFigure 6.3a.

xformer5 (75 kVA) xformer6 (100 kVA) xformer7 (167 kVA)

0.010

: : : : : : : : : :EE:: .: :.:
‘-. \

0.001
Meaure1 Measured’
1çp Rpmax
Rpniin -

1.0 1.0 1.0

Frequency (kHz) Frequency (kHz) Frequency (kHz)
Figure 6.9: Measured and simulated LIR curves.

Two values of resistances Rp are used [54]:

R = Rpm = 30p.u. and R = Rp = 13 p.u.. (6.9)

The real and imaginary parts of the computed short-circuit impedance are calculated

as follows:

R (R I 2
2)(coL)
R(f)- 6 10
2 + 2
R+(oL)
and
 Chapter 6. Transformer Models Applications 90
-

(R / 2)(coL )
2
— 2
R+(wL)

The 60 Hz resistance R is obtained from load losses. ‘It can be seen that the

simulated L/R ratios come close to the measured L/R curves (especially for R Rpmin). For

transformers 8,9 and 10 (Figure 6.10) one can see that the X(/) curves are bent upwards,

indicating a resonance frequency near the last measured frequency. Stray capacitances need

to be included in the model to produce the correct frequency response when one gets close

to resonance points. The first resonance frequency for all tested single-phase transformers is

between 6 and 60 kH.z. A similar resonance frequency range was found previously by

Ontario Hydro[56].
0.5 5

0.4 4

0.3

0.2

0.1 1

1 2 3 4 5 1 2 3 4 5
Frequency (kHz) Frequency (kHz)
Figure 6.10: Short-circuit impedance frequency response.

6.3.1.2 Three-Phase Transformers

The short-circuit impedance measurements were performed according to the circuit

of Figure 5.5. The impedance measured at each phase (self impedance) of transformers 12,

13 and 14 are shown in Figure 6.11, 6.12 and 6.13. The mutual impedances were measured

as well and found to be negligible for the 500 kVA and 150 kVA transformer (they were less
than an order of magnitude of the self impedances).
 Chapter 6. Transformer Models Applications 91
-

30 20
Xaa
Raa
74
—

10 Xbb
Rbb
20
S Ix —

0
10
-10

-
-20
1 2 3 4 5 1 2 3 4 5
Frequency (kHz) Frequency (kHz)

-.
Figure 6.11: 150 kVA three-phase transformer (self impedances).

-
1.5 6
Raa Xaa

1.0
- Rb
5 -
Xbb //
Xcc
Rcc

0.5

1 /Z
/

23 4 5 1 2 3 4 5

.
Frequency (kHz) Frequency (kHz)

Figure 6.12: 500 kVA three-phase transformer (self impedances).

1.2 0.6
I
0.8
00
‘s ---——

0.4
-0.4 ::Y
-0.6
1 2 3 4 5 1 2 3 4 5
Frequency (kHz) Frequency (kHz)
Figure 6.13: 75 kVA transformer short-circuit impedances( self impedances).
 Chapter 6. Transformer Models Applications 92
-

0.010

0.001

1.0
Frequency (kllz)

Figure 6.14: 500 kVA transformer LIR curve.

Each phase of the 500 kVA or 150 kVA transformer could therefore be
simulated as
a single-phase two-winding transformer. In Figure 6.14, simulated and measu
red (solid lines)
L/R curves for the 500 kVA transformer are shown. R is obtained from
equation (6.9).
Simulated curves come close to the one obtained from measurements
(especially for R
=Rpmax). For this example the parameter k in equation (6.8) is:

30
k=L,

where wL+ is the positive-sequence short circuit reactance at rated frequency.

For the 75 kVA transformer, the mutual impedances were of the same
order of
magnitude of the self impedances.The matrix form would therefore be more
appropriate to
represent this transformer. Unfortunately, for three-phase distribution transfo
rmers, only
positive sequence data are usually available from manufacturers.

6.3.2 Open Circuit Tests

In the open circuit tests, the low voltage winding is excited with
a nearly sinusoidal
voltage source at 60 Hz to obtain the transformer saturation characteristics.
No-load losses
 Chapter 6. Transformer Models Applications 93
-

and mis current were measured for different rins voltage levels. The x-axis of Figure 6.1 5a

is the ratio of the rms exciting current I to the mis exciting current ‘crafed at rated

voltage for single-phase transformers 1,2,3,4 and 8. The y-axis ‘is the ratio of the rrns

exciting voltage V to the rated voltage Vrad. The solid line is the average curve. Below

rated voltage, saturation curves differ very much from each other. However, for harmonic

and transient studies, the representation above rated voltage is more important and in this

region the curves come close to each other.

P.
0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5
Iexc/Iexc-rated Pexc/Pexc-rated
(a) (b)

Figure 6.15: Single-phase transformers.

a) V, I characteristics;
-

b) V, average power curves.

-

The x-axis of Figure 6.1 5b is the ratio of the no-load exciting losses P to the no-

load losses at rated voltage Vrated for the same transformers. The solid line represents the

average curve.

For three-phase transformers open-circuit tests are usually performed for positive

sequence. Figure 6.16 is the measured positive sequence excitation characteristic of the

three-phase 75 kVA transformer. one can see how the exciting current through each phase
(solid line is the average curve) behaves as the exciting voltage changes.
 Chapter 6. Transformer Models Applications 94
-

2 4 6 8 10
Iexc/Iexc-rated
Figure 6.16: V, -I characteristic for a 75 kVA three-phase transformer.

Typical distribution transformer correction factors that should be applied to no-load

losses at rated voltage to give the correct losses when the transformers are driven into

saturation, are shown in Table 6.2[441. Agreement between this table and the solid line

curves of Figures 6.15 and 6.16 is reasonably good.

Table 6.2 -Correction factors

Correction Factors
Operating Voltage (°“) No-load Exciting kVA rating
Loss Current
105 1.15 1.50 For 167 kVA and below (1)

110 1.30 2.20 For 150 kVA and below (30)

105 1.50 2.30 For 250 kVA and above(10)

110 2.40 4.60 For 225 kVA andabove (3k)

6.3.2.1 Stray Capacitances

The curve shown in Figure 6.17 was obtained from a recently manufactured

transformer (xformerl 1 Table 6.1). The rms current at 90% of the rated voltage is smaller than
-
 Chapter 6. Transformer Models Applications 95
-

the rms current at 50% ofthe applied voltage. Here, the exciting current is very
much affected by
stray capacitances.

A crude estimation of the saturation curves and the open-circuit capaci

tance can be
found iteratively. First, the algorithm generates the peak flux-current
characteristic
“corrupted” by stray capacitance effects as shown in Figure 6. 17b. The peak
current is
decreasing in the region between points A and B. Since The inductance is very
high in this
region. It can be assumed that it is infinite between turning points A and B. The
peak current
I through the open circuit capacitance C 0 is then the horizontal distance between turning
point A and any point in between A and B. For instance at point B, the open-c
ircuited
capacitance can be obtained from:
I—,
“—‘open
—
B

In Table 6.3, the “corrupted” flux-current points were obtained using Figure 6.17a as

input data.
150

>
100
0)
0
0
>
50
EI-

2 4
rms Current (A) 1
(a) (b)
Figure 6.17: Newly manufactured transformer.
a) V I,,, curve;
-

b) Corrupted flux-current curve (not to scale).

 Chapter 6. Transformer Models Applications 96
-

Table 6.3: Corrupted flux-current curve.

Flux Linkage(Vs) Current(A)

0.000000 0.0000000
0.040327 0.42628 1
0.059046 0.747657
0.075777 0.9053 18
0.095659 1.193316
0.114790 1.240726
0.151553 0.815392
0. 169560 0.396883
0.189066 0.123851
0.207823 0. 162665

The following information is needed:

peak current at turning point A = 1.240726 A;

peak current at turning point B = 0.12385 1 A;

peak flux at turning point B = 0.189066 Vs.

Then, the open circuit capacitance referred to the 120 V side is

1.240726 0.123851
Copen = 41.56pF.
—

= (2irx 60)2 xO.1 89066

If Copen is needed to be referred to the high voltage terminal,

,, 2
0
C =41.56x(120/14400) 2.89rF.

The Copen could then be split into three parts and placed between each transformer
terminal and ground of Figure 6.4b. Although the open circuit capacitances affect the

exciting current, the first resonance frequency for this transformer was above 6 kHz.
 Chapter 6. Transformer Models Applications 97
-

6.4 Sensitivity Study

The frequency-dependent short circuit impedances of transformer 5 are obtained

from nameplate data in the way explained in Section 6.3.1. The curve and no-load
losses are measured in the laboratory and the nonlinear core parameters are calculated.

Numerical simulations using the IVIHLF program[57] are carried out to see how the total

harmonic distortion(THD) of the primary current and secondary voltage are affected by

changes in the short-circuit parameter R (see equation 6.11), and in a balanced linear load

connected half to the secondary and half to the tertiary terminal.

5 Stray capacitances are
neglected. The transformer primary terminal is excited with a sinusoidal voltage and driven

into saturation. The simulation results are shown in Table 6.4. Rp has little effect in the

primary current and in the secondary voltage. From the last two rows one should note that

at flill load the nonlinear shunt elements are of minor importance (the harmonics produced

by them are of small amplitude). The load model is more important than the nonlinear

excitation branch model.

Table 6.4 - Sensitivity study.

Load Power TFID TND
Voltage Load Factor Primary Secondary
(p.u.) (p.u.) Rp(p.u.) Current (%) Voltage (%)
1.05 0.00 - 13.0 19.317 0.312
1.05 0.00 - 30.0 18.900 0.317
1.10 0.00 - 13.0 26.041 0.542
1.10 0.00 - 30.0 25.488 0.550
1.05 1.00 0.60 13.0 0.041 0.171
1.10 1.00 0.60 30.0 0.065 0.280

TFI) is the ratio between the total rins value of the harmonics and the mis value of the fundamental [71].
 Chapter 6. Transformer Models Applications 98
-

6.5 Field Test

In Figure 6.18, three three-phase 500kVA transformers, connected to a 12 kV

three-
phase power system, feed a 1500 HP DC Drive. Currents and ‘(‘oltage waveforms
were
measured at both transformer terminals. The current and voltage waveforms at the
12 kV
side are shown in Figure 6.19 and in Figure 6.20. In Figure 6.21 one can see plots
of
measurements and simulations of the line voltage at delta side when the DC drive is on low

load. Frequency-dependent effects in the windings and the nonlinear effects in the core are

neglected (the transformer is represented as a constant RL short-circuit 6

branch). Solid lines
are computed curves and dotted curves were obtained from measurements. One can see that
there is a good agreement between measurements and simulations. The difference between

curves may be attributed to measurement noise.

12 bV .k500 kVA 600 V

I4otat
P18
R._ PDC
PCA T.qve
RPW

od

Figure 6.18: DC drive set up.

Here, the measured priniaiy voltage, current and the short circuit impedance were the required data. In
6
fact, voltage and current waveforms were saved as ASCII files and converted from time-domain to phase-
domain to be used as input sources in the MHLF program.
 Chapter 6. Transformer Models Applications 99
-

0.01 0.02
Time(s)

I
Figure 6.19: Measured voltage waveform.
20

.JEfL

0.01 0.02
Time (s)

Figure 6.20: Measured current waveform.

Vab Vca Vbc

0.01 0.0
Time (s)

Figure 6.21: Transformer secondary line voltage.

 Chapter 6. Transformer Models Applications 100
-

6.6 Summary

Transformer models for harmonic studies were presented. They are derived from
60 Hz
manufacture?s data. Short-circuit impedance and open circuit saturation tests were perform
ed,
the parameters of the transformer equivalent circuits were determined and correla
ted to
nameplate data. The transformer equivalent circuits reproduce the measured short-c
ircuit
impedances reasonably well if stray capacitances are negligible for the frequency range of interes
t
in harmonic studies (60 Hz to 5 kHz). If stray capacitances are important (resonance for short-
circuit test close to 5 kHz), it is better to add shunt capacitances to the transformer model. These
capacitances are determined by low-frequency or resonance measurements, or taken as typical
values from publications or textbooks [5 8-60].

Typical saturation curves for single-phase and three-phase transformers are presented.
For situations in which the exciting current includes strong capacitive effects, a crude estimation
ofthe open circuit capacitance can be made if saturation V,,, I curves and no-load losses are
-

provided.

Sensitivity studies indicate that the load plays an important role in the harmonic contents of
terminal voltages and currents. In genera], the representation of the core is less important than the
representation ofthe short-circuit impedance and the load.

In general, at steady state, transformers usually do not experience deep saturation(flux in the
core may be only a few percent higher than the rated flux). To predict the exciting current for a
practical case is extremely difficult since, even for the same transformer rating (see Figure 6.15), the
saturation curves may differ very much from each other. However, the exciting current, although
distorted, is veiy small and will not affect the simulation results. The simplest model of the
transfbrmer was able to reproduce the field test reasonably well.
 Chapter 6. Transformer Models Applications 101
-

It is advisable to include the nonlinear effects ofthe core for situations in which resonance may
occur ( for instance, the transformer is unloaded or supplying a light load). The transformer could be
driven into a deeper saturated level. The exciting cunent should not be neg’ected here, and similarly
to
the inrush current case of the last chapter, in the saturated region, the exciting current is likely to
be ‘

predicted with reasonable accuracy provided the saturation curves are known up to deep saturation.
 Chapter 7

Conclusions

Iron core nonlinearities may play a very important role in transient studies. In the
following paragraphs, the major contributions of this research work are outlined.

A direct method for producing saturation curves from readily available transformer
test data as supplied by manufacturers, was presented. The algorithm is easy to implement
and may be useful for electromagnetic transient programs.

A method for more correctly representing the saturation curve of a delta-connected

transformer, suitable for situations in which the tests are performed with a closed delta, was
developed. It uses positive sequence excitation test data as input and takes into
consideration the removal of triplen harmonics from the line current. The algorithm is
iterative and easy to implement.

An approach to model frequency-dependent effects in the transformer core from

transformer no-load loss data, was presented. The parallel Foster-like circuit model, in which
hysteresis and eddy current effects are treated simultaneously, were realized. The flux
current trajectories are generated by the circuit models with no need to pre-define them.

102
 Chapter 7. Conclusions 103

Practical applications of transformer models for harmonic studies were presen

ted.
The transformer models were derived from manufacturer’s data. and may
be used for
transient studies as well. In general, when the transformer is not in deep saturat
ion, the exciting
current, although distorted, is very small and will not affect the simulation results
. It is advisable
to include the nonlinear effects of the core only for situations in which resonance
may occur (for
instance, when the transformer is unloaded or supplying a light load).

A method for crudely estimating the transformer open circuit capaci

tance from
saturation Vrms I curves and no-load losses, was presented. It is useful
-
for situations in
which the exciting current includes strong capacitive effects.

Two case studies involving transients were investigated: an inrush current test in
a1
kVA transformer and a ferroresonace field test. For the inrush current test, the
simple core
model (a nonlinear inductance) reproduced the measured current reasonably well.
For the
ferroresonance test, a simple core model (nonlinear resistance in parallel with
a nonlinear
inductance) also represented the core reasonably well. Theoreticaly, this core
model is
limited to a single frequency. At frequencies both lower and higher than fundam
ental the
model would not produce the frequency-dependent core behaviour. However, for
these case
studies, there was no need to include frequency-dependent effects.

For future work, there is a need to further test the core models against other
field
measurements, to fully determine the importance of frequency-dependent core
models for
practical applications. There is also a possibility of computing saturation curves
of delta
connected transformers directly without iterations. This would save compu
ter time, and
make the solution algorithm more reliable.
 References

1. P.A.Abetti, Bibliography on the Surge Performance of Power Transformers and

RotatingMachines, Trans.ATEE, vol. 77, 1958, PP. 1150-1168.

2. C. Degeneff, A Methodfor Constructing Terminal Modelsfor Single-Phase N-Winding

Transformers, Paper A78 539-9, presented at IEEE PES Summer Meeting, Los Angeles,
Calif., July 16-21, 1978.

3. L. Nakra, Behaviour of Circuits with Ferromagnetic Nonlinearity, Chapter 1, Ph. D.

Thesis, McGill University, Montreal, August 1973.

4. H. W. Dommel, Electromagnetic Transients Program Reference Manual, Section 6,

Department of Electrical Engineering-The University of British Columbia, Vancouver,
1986.

5. F. L. Alvarado, Eletromagnetic Transients Program (EIVITP), EPRI, EL-4561, Vol. 3,

The University of Wisconsin at Madison, Madison, Wisconsin, June 1989.

6. W. F. Tinney, Compensation Methodsfor Network Solutions by Optimally Ordered Tri

angular Factorization, IEEE Trans. Power App. Syst., vol. PAS-91, Jan.fFeb. 1972, pp.
123-127.

7. J.G. Frame, N. Mohan and T. Liu, Hysteresis Modeling in an Electromagnetic Tran

sients Program, IEEE, Trans. Power App. Syst.,vol. PAS-lOl, Sept. 1982, pp. 3403-
3412.

8. B. Mork and K. S. Rao, Modeling Ferroresonance with EMTP, EMTP-Newsletter,

vol. 3, No. 4, May 1983, pp. 2-7.

9. E. P. Dick and W. Watson, Transformer Models for Transient Studies Based on Field
Measurements, IEEE Trans. Power App. Syst, vol. PAS-100, No. 1, Jan. 1981, pp. 409-
419.

10.N. Germay, S. Mastero and J. Vroman, Review ofFerroresonance Phenomena in High

Voltage Power System and Presentation of a Voltage Transformer Model for
Predetermining Them, CIGRE’, Paris, paper 33-18, 1974

104
 105

11 .G. Biorci and D. Pescetti, Analytic Theory of the Behaviour of Ferromagnetic Mate
rials, Nuovo Cimento, vol. 7, No. 6, 1958, PP. 829-842.

12. I. D. Mayergoyz, Mathematical Models ofHysteresis, Springer-Verlag, 1991

13 .J.L. Binard, Hysteresis Model For Power Transformer Transient Simulation Program,
EMTP Newsletter, Vol. 7, No. 3, Sept. 1987.

14.J.G. Santesmases, J. Ayala, S. H. Cachero, Analytical Approximation of Dynamic

Hysteresis Loops and its Application to a Series Ferroresonant Circuit, Proc. TEE 117,
No. 1, January 1970, pp. 234-240.

15.L. 0. Chua and K. A. Stromsmoe, Lumped Circuit Models for Nonlinear Inductors
Exhibiting Hysteresis Loops, IEEE Trans. on Circuit Theory, vol. CT-17, No. 4, Nov.
1970, PP. 564-574.

16.E.J. Tarasievicz, A. S. Morched, Detailed Transformer Modelfor EMTP; Part 1: Eddy

Current Representation, EMTP Review, Vol. 3, No. 4, October 1989.

17.E.J. Tarasieviez, A. S. Morched, A. Narang and E. P. Dick, Frequency Dependent Eddy

Current Modelsfor Nonlinear Iron Cores, IEEE Transactions on Power Systems, Vol. 8,
No. 2, May 1993, . 597
-
588
PP.
18. J. Avila-Rosales and F. L. Alvarado, Nonlinear Frequency Dependent Modelfor Elec
tromagnetic Transient Studies in Power Systems, IEEE Trans. Power App. Syst, vol.
PAS-lOl, Nov. 1982, pp. 428 1-4288.

19.J. Avila-Rosales and A. Semlyen, Iron Core Modeling for Electrical Transients, IEEE
Trans. Power App. Syst., vol. PAS-104, No. 11, Nov. 1985, pp. 3 189-3194.

20.F. De Leon and A. Semlyen, Time Domain Modeling of Eddy current Effects for
Transformer Transients, IEEE Transactions on Power Delivery, Vol. 8, No. 1, January
1993, .
280
-
271
pp.
21 .F. Keffer, The Magnetic Properties of Materials, Scientific American, vol.217, No. 3,
pp. 222-234.

22.G. E. Fish, Soft Magnetic Materials, Proceedings of the IEEE, Vol. 78, No. 6, June
1990, pp. 947-972.

23.1.. J. Nahin, Oliver Heaviside: Sage in Solitude, IEEE Press, New York, 1987,
pp. 109.
 106

24. M. Latour, Note on Losses In Sheet Iron at Radio Frequencies, Proceedings of the
Institute of Radio Engineers, V.7, 1918, PP. 61-71

25.T. H. O’Dell, Ferromagnetodynamics, Chapters 1 and 5, Memillan Press Ltd, London,

1981. /

26.H. J. Williams, W. Shockley and C. Kittel, Studies of Propagation Velocity of a Fer

romagnetic Domain Boundary, Phys. Rev. 80, 1950, pp. 1090-1094

27.P. D. Agarwal and L. Rabins, Rigorous Solution of Eddy Current Losses in Rectangular
Barfor Single Plane Domain Wall Model, Journal of Applied Physics, Suplement to vol.
31 No. 5, May 1960, pp. 246S-248S.

28.C. W. Chen, Magnetism and Metalurgy of Soft Magnetic Materials, North Holand
Publishing Company, New York, 1977.

29.F. Brailsford, Investigation of the Eddy Current Anomaly in Electrical Sheet Steels, 3.
lEE, 1948, No. 95, Pt. 2, pp. 12 1-125.

3 O.K. J. Overshott, I. Preece and J. E. Thompson, Magnetic Properties of Grain-Oriented

Silicon Iron, Part 2 Basic Experiments on the Nature of the Anomalous Loss Using an
-

1 840-1845.
Individual Grain, Proc. lEE, Vol. 115, No. 12, December 1968, pp.

31 .T. Spooner, Properties and Testing of Magnetic Materials, McGraw-Hill, 1927.

32.C. Heck, Magnetic Materials and Their Applications, (English translation from the
German),Chapter 10, Butterworth & Co. Ltd, 1974, pp. 337.

33.S.L. Burgwin, Measurement of Core Loss andA.C. Permeability with the 25 cm Epstein
Frame. Proceedings, Am, Soc. Testing Mats., ASTEA Vol. 41, 1941 pp. 779-796.

34.G. Herzer and H. Hilzinger, Recent Developments in Soft Magnetic Materials, Physica
Scripta, Vol. T24, 1988, pp. 22-28.

35.A. J. Moses, Recent Advances In Experimental Methods For The Investigation of

Silicon Iron, Physica Scripta, Vol. T24, 1988, pp. 49-53.

36.F. J. Wilkins and A. E. Drake, Measurement and Interpretation of Power Losses in

Electrical Sheet Steel, Proc. IRE Vol 112, No. 4, April, 1965, pp. 77 1-785

37.A. Basak and A.A.A. Qader, Fundamental and Harmonic Flux Behaviour In a JOOkVA
Distribution Tranfor,ner Core, IEEE Trans. on Mag., Vol. Mag -19, No.5, September
1983, pp. 2100-2102.
 107

38. Stanley H. Charap, Magnetic Hysteresis Model, IEEE Trans. on Mag., vol. Mag -10,
No. 4, December 1974, PP. 1091-1096.

3 9.T. Nakata, Numerical Analysis of Flux and Loss Distribution in Electrical Machinery,
(Invited Paper), IEEE Trans. on Magnetics, Vol. Mag 20, No. 5, September, 1984.
-

40.R. J. Larry and L. R. Turner, Survey of Eddy Current Programs, IEEE Trans. on
Magnetics, Vol. Mag -19, No. 6, November 1983, .
2477
-
2474
pp.
41.G.W. Swift, Power Transformer Core Behavior Under Transient Conditions, IEEE
Trans. Power App. Syst,vol. PAS-90, No.5, September/October, 1971, pp. 2206-2210.

42.H. W. Dommel, A. Yan, R. J. Marcano and A. B. Milliani, Case Studies For Elec
tromagnetic Transients, UBC Department of Electrical Engineering Internal Report,
Vancouver, 1983.

43 .Microtran Power System Analysis Corporation, Transient Analysis Program Reference

Manual, Vancouver, Canada, 1991.

44.Westinghouse Electric Coorporation, Electric Utility Engineering Reference Book, Vol.

3, Distribution Systems, Chapter 6, Pittsburgh, Pensylvania, 1959.

45.S. Prusty and M. V.S. Rao, A Direct Piecewise Linearized Approach to Convert rms
Saturation Characteristic to Instantaneous Saturation Curve, IEEE Transactions on
Magnetics, Vol. Mag-16, No.1, January 1980, pp. 156-160.

46.W. L. A. Neves and H. W. Dommel, On Modelling Iron Core Nonlinearities, IEEE

Transactions on Power Systems, Vol. 8, No. 2, May 1993, pp. 4 17-425.

47.W. L. A. Neves, H. W. Dommel, Saturation Curves of Delta-Connected Transformers

From Measurements, to appear in IEEE Transactions on Power Delivery. Paper 94 SM
459-8 PWRD presented at IEEE PES Summer Meeting, July 24-28, 1994, San
Francisco, CA.

48.W. L. A. Neves, H. W. Dommel and Wilsun Xu, Practical Distribution Transformer

Modelsfor Harmonic Studies, to appear in IEEE Transactions on Power Delivery. Paper 94
SM 406-9 PWRD presented at IEEE PES Summer Meeting, July 24-28, 1994, San
Francisco, CA.

49.A. Otter and W. Neves, Loss Measurement Programs at TRIUMF, Proceedings of The
Kaon PDS Magnet Design Workshop, TRIUMF, Vancouver, October 3-5, 1988, Pp.
132-135.
 108

50.M. B. Stout, Basic Electrical Measurements, Section 16-8, New York, Prentice Hall
Inc., 1950.

5 1.S.N. Talukdar, J.K. Dickson, R.C. Dugan, M.J. Sprinzen, C.J: Lenda, On Modeling
Transformer and Reactor Saturation Characteristics for Digital and Analog Studies,
IEEE Trans. on PAS, vol. PAS-94, 1975, pp. 612-621.

52.C. G. A. Koreman, Determination of the Magnetizing Characteristic of Three-Phase

Transformers in Field Tests, IEEE Transactions on Power Delivery, Vol. 4, No. 3, July
1989, pp. 1779-1785.

53. Leuven EMTP Center (LEC), ATP Rule Book, Section XIX-G, Revision July, 1987.

54. CIGRE Working Group 36.05, Harmonic, Characteristic Parameters, Methods of

Study, Estimates ofExisting Values in the Network, Electra No. 77, pp. 3 5-54.

55.C57 Collection, Distribution, Power and Regulating Transformers, IEEE Press, 1991

56.Canadian Electrical Association Project RP-070-D-165, Power Line Carrier on

Distribution Systems, Ontario Hydro’s Research.

57.Wilsun Xu, Multiphase Harmonic Load Flow Program (MHLF) Reference Manual.
Powertech Labs, Inc., February, 1991

58.Allan Greenwood, Electrical Transients in Power Systems, John Wiley & Sons Inc., New
York, 1991.

59.ANSIIIEEE C3 7.011-1979, IEEE Application Guide for Transient Recovery Voltage for
AC-High Voltage Circuit Breakers Rated on a Symmetrical Current Basis.

60.CIGRE Working Group 02 of Study Committee 13, Interruption of Small Inductive

Currents: Chapter 5- Part I, Electra No. 133, pp. 79-96.

61.Ernst A. Guillemin,, Synthesis of Passive Networks, Chapter 8, John Wiley & Sons Inc.,
New York, 1957.

62.Athanasios Papoulis, The Fourier Integral and Its Applications, McGraw Hill Book
Company Inc.,1962, pp. 195-196.

63 .Jiri Vlacb, Computerized Approximation and Synthesis of Linear Networks, Chapter IV,
John Wiley & Sons Inc., New York, 1969.

64.B. D. Cullity, Introduction to Magnetic Materials, Addison Wesley Publishing Company,

1972.
 109

65.J. J. Becker, Magnetization Changes and Losses in Conducting Ferromagnetic

Materials, Journal of Applied Physics, Vol. 34, No. 04 (Part 2), April, 1963, pp. 1327-
1332.

66.David F. Tuttle, Electric Networks Analysis ant Synthesis, Chapter Xl, McGraw Hill Inc.
1965.

67.United States Steel, Electrical Steel Sheets Engineering Manual, 4

t h
Edition, 525
William Penn Place, Pittsburgh 30, PA.

68.William H. Press, Saul A. Telkoisky, William T. Vetterling, Brian P. Flannery, Numerical

Recipes, The Art of Scient/Ic Computing, 2 Edition, Section 5.13, Cambridge
University Press, New York,1992.

69.L. Stuehm, B. A. Mork and D. Mairs, Five Legged Core Transformer Equivalent
Circuit, IEEE Transactions on Power Delivery, Vol. PWRD-4, July 1989, pp. 1786-
1793.

70. A. C. Franklin and D. P. Franklin, The J & P Transformer Book: A Practical Technology
of the Power Transformer, 11
th
Edition, Butterworths, 1983.

71.A. E. Emanuel, Nonsinusoidal Waveforms in Power Systems .Basic Definitions, IEEE

Tutorial Course 90EH0327-7-PWR.
 AppendixA /

Orthogonality Between ‘r and I

Consider the circuit of Figure 3.1 a. The voltage across the transformer terminals and
its correspondent flux linkage can be written in the form

v(O)=VksinO, (A.1)

and

A(O)=—AkcosO, (A.2)

respectively.

Let us use Fourier analysis to represent the current i(O) through the nonlinear
resistance and the current i,(O) through the nonlinear inductance. Due to the odd symmetry
of the v—i,. and A—i
1 curves, Ir(O) and i,() will have only odd harmonic components in the

form

jr(0)thi sinO-t-a
sin3O+••• +asinpO+•.
3 (A.3)

and

cosO+b
1
i,(9)=b
c
3 os39+ +bcospO+”., (A.4)

w&te p is odd.

110
 Appendix A. Orthogonality Between Ir and 11111

The total current 4(9) is then:

i,(9) 1r(O)+Ii(O), (A.5)

i.e.,

where y = arctan(b I as).

Evaluating the rrns values of 4(0), i,(O and i (O, we have

Ir, = ja + a + + +• •, (A. 7)

(A.8)

and
+a+b+••-, (A.9)

respectively.

From equations (A.7), (A. 8) and (A.9), it can be seen that

i =r. +I. (A. 10)

 Appendix B

Measurement of the Initial Magnetization Curve

The initial magnetization curve is a plot of the locus of the DC symmetrical

hysteresis loop tips for different peak values of magnetization. Figure B. 1 is the circuit used
to measure it.

> To Analyzer

V Epstein Frame

Figure B. 1: Measurement of the initial magnetization curve.

The magnetizing winding of the Epstein frame (primary winding) is connected to a

DC power supply through a reversing switch S, ammeter and a decade resistance box R.
The secondary winding is connected to a digital waveform analyzer where the voltage

112
 Appendix B. Measurement of the Initial Magnetization Curve 113

waveform is obtained and numerically integrated to give the flux linkage across the

secondary winding.

The Epstein frame is demagnetized before any measurement is taken. This is

accomplished by driving the core into saturation using alternating current at power

frequency and gradually reducing the core excitation to zero.

After demagnetization, R is set to provide a low current, and S is reversed several

times to assure the sample is in a definite hysteresis cycle (AA’ and A’A trajectories of

Figure B.2). Then, the first reading takes place. The voltage across the secondary winding of

the Epstein frame is integrated and the flux difference between AA’ is obtained. This value

is divided by two and segment OA is plotted.

C,

Figure B.2: Hysteresis ioop locus.

After the first reading, R is changed to give a slightly greater value of the current in

the primary winding and the process is repeated up to the desired limit.
 Appendix C

Computation of Triplen Harmonic Components

Although only the computation of triplen harmonics is needed, it is appropriate to

show the derivation of all odd harmonic components of the current that may be produced by
saturation curves during standard tests. The equations below are developed for nonlinear
inductances. For nonlinear resistances, one just needs to replace A. by and L by R in all

equations.

Consider the piecewise nonlinear inductance of Figure 4.2c. For a sinusoidal flux
A(O) = Ak sin( 0), the current can be written in a Fourier series form containing odd

harmonics only’. So,

3
(
1
i
s in30-I-..+bsi
0)=1sin0+b np0+•., (C.1)

where p is odd, and

b =-fi(O)sinp0d0. (C.2)

For segment k =2,

b
=±[J 0
2
A

sin —A
1
+ ‘‘2
}inocio].

114
 Appendix C. Computation of Triplen Harmonic Components 115

li
1 1 1
Substituting F =—=— and F
2 into the equation above, the fundamental
1
L “

current (p = ) and its harmonics (p> ) are:

11
b +-[(F 1
A
2
=F )(s(0
—F
)
+
2
c
) os0
2
])

and
b =--[(Fj )
)(g(p,
—F
)
1
+
2 -cosp0
A
J. 0

With
s(0i)=.{0i _sin20i)

and
sin[(p 1)0] sin[(p + 1)8]
g(p, 0)
—

= —1) —
2(p +1)

Finally, for any segment k, one can write

= — r )(s0
, )k + A., cos
1
+

and

= 1
(r — )2 +-1cosP0ii)
11
g(p,0

where

s(0i)={0i ..Jsin2Oi)

and

sin[(p 1)0k] sin[(p + 1)0k]

g(p, 0,)
—

= —1) —
2(p +1)
 AppendixD

Rational Approximation of the Real Part of Y(s)

D.1 Fitting Procedure

Similar to equation (5.8), the frequency-dependent resistance could be expressed
as a rational function in -s
2

(_s2+a?)(_s2+a). (-s+a)
R (s) R
—

(D. 1)
)(—s + b).
1
b
2 b)’
-

(_s2 + (s2 +

and with s=jw, R(co) is written in the form:

c0
1
(c
2 1).. 2
+ 1)(c
w+
2 (co + 1)
R(a)— KC (D.2)
2 + 1)(d
(dm w +1).. .(dw
2 2 + 1)
—

Where K, C. = d = 1/12 are positive real numbers, with i = 1,2,

1/a and 1 , n The right hand
side of (D. 1)is fitted to the known function R(o). The goal here is to compute K, c. and

. The number of terms n will depend on the frequency range of interest. R(co) is fitted
1
d

interval by interval as shown in Figure D. 1 starting with the interval between o and Oi and

ending with the interval between o and co.

116
 Appendix D. Rational Approximation of the Real Part of Y(s) 117

R(co)

(01 max CO

Figure D. 1: Frequency dependent resistance.

The algorithm proceeds as follows:

1 For interval 1, write R(w) as

-

KC
+1)Kd0
(cIO.
+
2
-Kc(d
R(w)_ +1) (do,
2 2 +1)

or

2
KdC0 - 2+K
dR(co)w 0 = R(w) (D.3)

2 - Estimate .
Take m frequency sample points between o
1 and w . Build the

, Kd and d
overdetermined system of equations (D.3) with only the parameters K
0 1

unknown. A weighted linear least square fitting routine[68] is used to find these parameters
0 is also computed). If the maximum error in the fitting is less than a predifined
(c = Kd /K

value, proceed to next step, If not, oi is reduced by small steps & i and the system of

equations (D.3) is reevaluated.

3-For interval 2 in Figure D.1, write R(a) as

(c 2 + 1)(c Q)2 + 1)
—
 Appendix D. Rational Approximation of the Real Part of Y(s) 118

with K, c
1 and d. already known from the previous step. Estimate w
2 and find c
2 and d
2
following the procedure of step 2, with the ni frequency samples now taken between o

ando.

For the remaining intervals in Figure D. 1, c. and d

1 are found in the same fashion and the
constants 1?,,, a and b of (D. 1) are determined.

Experience has shown that a first estimate of w, 5o (with Ao),=0. 1ci) and
subsequent estimates of co+, 10 o (with Aw
1 = 0.1 wi for i 2) are usually very close to
the final values for an error of 3% and 100 frequency samples used as input parameters to

the fitting routine.

D.2 G(o)) Obtainedfrom Lamination Data

Steel manufactures may supply core loss vs. frequency curves for the most

commom lamination grades subjected to constant flux amplitudes’. These data are generally

obtained on Epstein samples using a sine wave voltage source following the test procedure

of ASTM Standard Method A-34. With this information one can obtain either the frequency-
dependent resistance R(co) or conductance G(w) as described next.

In the sinusoidal steady state the average power P

1 is:

P,=P(co). (D.4)

1 To avoid confusion of trade names, the American Iron and Steel Institute has assigned AISI type numbers
to electrical steel. These consist of the letter “M” (for magnetic material) followed by a number which, when
the designations were originally made, was about to ten times the core loss in watt/lb at 15 kilogauss (1.5 T)
and 60 Hz for 29 gage sheet (0.0 140 in). Core losses have since been reduced but the type numbers remain
[59].
 Appendix D. Rational Approximation of the Real Part of Y(s) 119

R(o) is obtained by:

2
V Aco
=
2
R(w)= i (D.5)

where Vm and Am are the amplitude of the applied voltage and flux, respectively. Since the

flux is kept constant, R(w) can be written in the form:

R(w)=Kp(w). (D.6)

It is convenient to normalize R(w) assuming that the resistance measured at the transformer

rated frequency o is R(o)=1. So, the constant K in the equation above is

P(o.)
K=, (D.7)
r

and the normalized resistance RN(e) is

2
P(O)r)
RN(ftI)= (D.8)
2
w P(w)

So, if the transformer lamination loss vs. frequency curve is known, an estimate of the

frequency-dependent core loss resistance can be made by scaling the normalized resistance

RN(w) to the resistance at rated frequency R(o). Then,

R(co)=RN(co)• R(o.ir), (D.9)

or for the conductance G(o)

G(o.I)=R).
 Appendix D. Rational Approximation of the Real Part of Y(s) 120

D.3 G(w) Obtainedfrom Standard Tests

The frequency-dependent core resistance could also be estimated from transformer

standard tests. The plot of loss/cycle can be crudelly approximated by a straight line, as
/
shown in Figure D.2.

Loss/cyc:

fr f

Figure D.2 Transformer core loss curve at rated flux.

Power transformer manufacturers usually supply two or three values of no-load

loss at different frequencies, with the flux kept at the rated level. From Figure D.2, the loss

per cycle coud be written as

1o.s j47h+ J
f
wheref is the frequency in which the loss is measured, Jr is the rated frequency, Wh is the
hysteresis loss per cycle and Weddy is eddy loss per cycle at the rated frequency.

Taking (D. 11) and writing the loss for the angular frequency w2itf,
 Appendix D. Rational Approximation of the Real Part of Y(s) 121

1
(D.12)

From (D.6) and (D. 12), the resistance R(o) is /

2
w
m
R(o,)= (D.13)
2Ploss =
—[wh +w_

Normalizing R(co), as done for (D.7) and (D.8),

Ko
RN(w)=
(1)’ D14
ratio+—
COr

where ratio=Ww’Weddy and

ratio + 1

The resistance and conductance are then:

R(o)=RN(w)•R(cor), (D.15)

and G(CO)=R(’).

If the transformer manufacturer does not supply the core loss vs. frequency curve, a typical

ratio Wl/Wed+ is commonly taken as unity [70]. In practice, the straight line of Figure D.2

could be drawn using loss measurement data gathered from routine tests. For instance, it

could be obtained from the loss measurement taken at rated voltage and rated frequency
and, as recommended by ANSI C57. 12.90 (Induced Overvoltage Withstand Test), from the
loss measurement taken at twice the rated frequency and double voltage amplitude.