Fast fault detection for power distribution systems

Magnus Öhrström

Stockholm 2003

Licentiate Thesis
Royal Institute of Technology
Department of Electrical Engineering
Akademisk avhandling som med tillstånd av Kungl Tekniska Högskolan
framlägges till offentlig granskning för avläggande av teknisk licentiatex-
amen tisdagen den 29 april 2003 kl 10.00 i sal H1, Teknikringen 33, Kungl
Tekniska Högskolan, Stockholm.

TRITA-ETS-2003-06
ISSN 1650-674x
ISRN KTH/EEK/R 0302-SE

c Magnus Öhrström, April 2003

Universitetsservice US AB, Stockholm 2003

Abstract

T HE main topic of this licentiate thesis is fast fault detection. The thesis
summaries the work performed in the project “Fast fault detection for
distribution systems”.
In the first chapters of the thesis the term “fast” is used in a general
manner. The term is later defined based upon considerations and conclusions
made in the first chapters and then related to a specific time.
To be able to understand and appreciate why fast fault detection is
necessary, power system faults and their consequences are briefly discussed.
The consequences of a fault are dependent of a number of different factors,
one of the factors being the duration of the fault.
The importance of the speed of the fault detection depends on the type of
equipment used to clear the fault. A circuit breaker which interrupt currents
only when they pass through a natural zero crossing might be less dependent
on the speed of the fault detection than a fault current limiter which limits
the fault current before it has reached its first prospective current peak.
In order to be able to detect a fault in a power system, the power sys-
tem must be observed, i.e., measurements of relevant quantities must be
performed so that the fault detection equipment can obtain information of
the state of the system. The fault detection equipment and some general
methods of fault detection are briefly described.
Some algorithms and their possible adaptation to fast fault detection are
described. A common principle of many algorithms are that they assume
that either a signal or the power system object can be described by a model.
Sampled data values are then fitted to the model so that an estimate of
relevant parameters needed for fault detection is obtained. An algorithm
which do not fit samples to a model but use instantaneous current values
for fault detection is also described and evaluated.
Since the exact state of a power system never is known due to variations
in power production and load, a model of the power system or of the signal
can never be perfect, i.e., the estimated parameter can never be truly correct.
Furthermore, errors from the data acquisition system contribute to the total
error of the estimated parameter.
Two case studies are used to study the performance of the (modified)
algorithms. For those studies it has been shown that the algorithms can
detect a fault within approximately 1 ms after fault inception and that
one of the algorithms can discriminate between a fault and two types of
common power system transients (capacitor and transformer energization).

iii
iv Abstract

The second case study introduced a system with two sources which required
a directional algorithm to discriminate between faults inside or outside the
protection zone.
It is concluded that under certain assumptions it is possible to detect
power system faults within approximately 1 ms and that it is possible to
discriminate a power system fault from power system transient that regularly
occurs within power systems but which not are faults.
TRITA-ETS-2003-06 • ISSN 1650-674x • ISRN KTH/EEK/R 0302-SE
Acknowledgements

This thesis concludes the work I have performed at the Division of Electric
Power Systems, KTH during my three and a half years there.
First, I would like to thank Leif Lundin (ABB) and Ove Albertsson
(ABB) that made it possible to start the project. Thanks goes also to
Weikko Alsterfalk (ABB) and Hans Gleimar (ABB) for letting me continue
with the project in spite of bad times for ABB Medium Voltage Products.
I would also like to thank my supervisors at KTH. When I started with
the project professor Göran Andersson was head of the department and my
supervisor. Thanks also to professor Lennart Söder who became head of the
department and my supervisor after professor Andersson moved to ETZ in
Zürich, Switzerland.
There was a small reference group attached to the project and I would
like to thank the members Henrik Breder (ABB), Stig Holst (ABB) and pro-
fessor emeritus Åke Ekström for valuable comments and suggestions. Pro-
fessor emeritus Åke Ekström was also running a project, “New techniques
for electricity supply of industrial systems including e.g. local generators.”
at the Competence center in electric power engineering at KTH from 1997
until 2002, where my work was discussed.
Thanks goes also to my colleagues at ABB and KTH for interesting
discussions and good company.
A special thanks goes to Jonas Persson (KTH) and Henrik Breder (ABB)
for proof-reading parts of this thesis.
Many thanks goes to Margareta Surjadi for all help with practical details.
The financial support from the Elektra-program (Elforsk), Vetenskapsrådet,
and ABB is gratefully acknowledged.
Finally, I would like to thank my family for their support. This thesis
would not have been possible without them.

v
vi
Contents

1 Introduction 3
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Main contributions of the thesis . . . . . . . . . . . . . . . . 5
1.5 List of publications . . . . . . . . . . . . . . . . . . . . . . . 5

2 Faults and their consequences 7

2.1 Faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Shunt faults . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Series faults . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 General . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Faults involving an arc . . . . . . . . . . . . . . . . . 11
2.2.3 Consequence steps . . . . . . . . . . . . . . . . . . . 12
2.2.4 Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.5 System aspects . . . . . . . . . . . . . . . . . . . . . 13

3 Fault clearing 15
3.1 Fuses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Circuit-breakers . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Current limiting . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.1 Series reactor . . . . . . . . . . . . . . . . . . . . . . 16
3.3.2 Fuses . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3.3 A switch-fuse combination current limiter . . . . . . 17
3.3.4 Solid-state fault current limiters . . . . . . . . . . . 18
3.3.5 Superconducting fault current limiters . . . . . . . . 19
3.3.6 Current diverter . . . . . . . . . . . . . . . . . . . . 19

vii
viii Contents

3.4 Motivation for fast fault detection . . . . . . . . . . . . . . 20

4 Fault protection systems 21

4.1 Fault clearing systems . . . . . . . . . . . . . . . . . . . . . 21
4.2 Relay protection system . . . . . . . . . . . . . . . . . . . . 21
4.2.1 Transducers . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 Wiring . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.3 Trip coil . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.4 Relays . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Basic protection principles . . . . . . . . . . . . . . . . . . . 27
4.3.1 Magnitude relays . . . . . . . . . . . . . . . . . . . . 27
4.3.2 Directional relays . . . . . . . . . . . . . . . . . . . . 28
4.3.3 Impedance relays . . . . . . . . . . . . . . . . . . . . 28
4.3.4 Differential relays . . . . . . . . . . . . . . . . . . . . 28
4.3.5 Pilot relaying . . . . . . . . . . . . . . . . . . . . . . 28
4.4 Speed of protection . . . . . . . . . . . . . . . . . . . . . . . 29

5 Algorithms 31
5.1 Waveform algorithms . . . . . . . . . . . . . . . . . . . . . . 31
5.1.1 Two samples . . . . . . . . . . . . . . . . . . . . . . 32
5.1.2 Fourier methods . . . . . . . . . . . . . . . . . . . . 33
5.1.3 LSQ-methods . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Model algorithms . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 UHS-relaying . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.4 Instantaneous current algorithm . . . . . . . . . . . . . . . 38
5.4.1 Low-pass filter . . . . . . . . . . . . . . . . . . . . . 39
5.4.2 Current differential . . . . . . . . . . . . . . . . . . . 40
5.4.3 Extension to provide directional properties . . . . . 40
5.5 Arc detector . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

6 Sources of errors 43
6.1 Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2 Power system transients . . . . . . . . . . . . . . . . . . . . 43
6.2.1 Capacitor energization . . . . . . . . . . . . . . . . . 44
6.2.2 Transformer energization . . . . . . . . . . . . . . . 44
6.2.3 Faults . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.2.4 Identification of differences . . . . . . . . . . . . . . 46
6.3 Data acquisition equipment . . . . . . . . . . . . . . . . . . 48
6.3.1 Transducers . . . . . . . . . . . . . . . . . . . . . . . 49
Contents ix

6.3.2 A/D converter . . . . . . . . . . . . . . . . . . . . . 51

6.3.3 Processor word length . . . . . . . . . . . . . . . . . 52
6.3.4 Processor speed . . . . . . . . . . . . . . . . . . . . . 53
6.4 Algorithm dependent errors . . . . . . . . . . . . . . . . . . 53

7 Case Studies 59
7.1 IEC case study . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.1.1 Common considerations . . . . . . . . . . . . . . . . 61
7.1.2 Fault detection with the LSQ-method . . . . . . . . 62
7.1.3 Fault detection with the differential equation method 64
7.1.4 Fault detection based on instantaneous current values 67
7.1.5 Results of the IEC case study . . . . . . . . . . . . . 68
7.2 SSAB case study . . . . . . . . . . . . . . . . . . . . . . . . 69
7.2.1 The electrical power system at SSAB . . . . . . . . . 70
7.2.2 Fault detection with the LSQ-method . . . . . . . . 71
7.2.3 Fault detection with a differential equation method . 72
7.2.4 Fault detection based on instantaneous current values 72
7.2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.3 Shortcomings of the simplified power systems . . . . . . . . 73

8 Conclusions and future work 75

8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

A Analysis of voltage dips measured at SSAB Oxelösund 79

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
A.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
A.2.1 Characterization of voltage dips . . . . . . . . . . . . 80
A.2.2 Comparison at different voltage levels . . . . . . . . 82
A.2.3 Cause and consequences . . . . . . . . . . . . . . . . 83
A.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.3.1 Characterization of voltage dips . . . . . . . . . . . . 83
A.3.2 Comparison at different voltage levels . . . . . . . . 83
A.3.3 Causes and consequences . . . . . . . . . . . . . . . 88
A.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
A.5 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
x Contents

B Experiments performed as computer based calculations 93

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.1.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.1.2 Theory dependence . . . . . . . . . . . . . . . . . . . 94
B.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
B.2.1 Parameters and variables . . . . . . . . . . . . . . . 95
B.2.2 Observations and measurements . . . . . . . . . . . 96
B.2.3 Effects of observation . . . . . . . . . . . . . . . . . 97
B.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
B.3.1 Interpretation problems . . . . . . . . . . . . . . . . 97
B.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
B.4.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . 99
List of Figures

2.1 Shunt faults . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Typical fault phase currents, each including a dc-component 10

5.1 The power system used in the differential equation algorithm. 36

6.1 The effect of the dc-component on the estimated current (ex-

pressed in kA) . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 The current associated with capacitor energization . . . . . 47
6.3 The current associated with transformer energization . . . . 48
6.4 The current associated with a power system fault . . . . . . 48
6.5 Estimated magnitude for a sensitive algorithm . . . . . . . 54
6.6 Estimated magnitude for a not so sensitive algorithm . . . . 55
6.7 The derivative at two different sampling frequencies . . . . 57

7.1 IEC case study single line diagram . . . . . . . . . . . . . . 60

7.2 A block diagram describing the implementation of algorithms
in EMTDC . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.3 The estimated current and the instantaneous current . . . . 63
7.4 The estimated resistance during a fault. . . . . . . . . . . . 65
7.5 SSAB case study single line diagram . . . . . . . . . . . . . 70

A.1 Characterization with the RMS-method . . . . . . . . . . . 81

A.2 Minimum remaining RMS voltage . . . . . . . . . . . . . . 86
A.3 Minimum characteristic voltage . . . . . . . . . . . . . . . . 88
A.4 Propagation of voltage dips . . . . . . . . . . . . . . . . . . 90

xi
xii
List of Tables

7.1 Maximum fault detection time . . . . . . . . . . . . . . . . 63

7.2 Maximum fault detection time . . . . . . . . . . . . . . . . 66

A.1 Minimum remaining voltages and duration of the dips. . . . 84

A.2 Minimum characteristic voltages and duration of the dips. . 85
A.3 Dip magnitude at different voltage levels . . . . . . . . . . . 87
A.4 Transformer connections . . . . . . . . . . . . . . . . . . . . 88
A.5 Cause and consequences . . . . . . . . . . . . . . . . . . . . 89

xiii
xiv
I dedicate this thesis to my dear family:
Hillevi, Jennifer and Cecilia
Chapter 1

Introduction

1.1 Background

T O prevent people and property from damage or injury, electrical faults

in a power system must be cleared fast. In the early days of electrical
power systems the fault clearing was administered by the maintenance staff,
who visually detected the fault and manually operated a switch to clear the
fault. As fault currents became larger and the operating requirements of the
electric power system became more stringent, the need for automatic fault
clearance became a necessity.
A typical fault clearing system consists of a circuit breaker and a re-
lay protection system. The relay protection system consists of transducers,
wiring, relay, auxiliary power supply, and the operating coil of the circuit
breaker1 .
In the early days of automatic fault clearing, a fault was detected by
electromechanical relays2 . The measured quantity, such as for example a
voltage or a current, was transformed to a mechanical force which operated
the relay when a preset threshold was exceeded. Following the advent of
electronics such as transistors and operational amplifiers, solid-state relays
were developed. The characteristic of such relays were implemented by
circuit design. Today, new relays are normally numerical relays. They are
1
It might seem odd that the operating coil is included in the relay protection system,
but since the operating coils is connected in series with the wiring of the relay protection
system, it is closely integrated with the relay protection system.
2
In fact, electromechanical relays are still in common use and can be obtained by a few
manufacturers. However, numerical relays are taking over more and more.

3
4 Chapter 1. Introduction

built around a microprocessor in which the relay characteristic is digitally

implemented. The analogue measurements are converted to digital signals
for evaluation within the microprocessor. The recent development of fast
microprocessors has led to the possibility to implement highly sophisticated
relay characteristics within the microprocessor.
The trend in protection relay seems to go towards so-called relay termi-
nals which for example can contain all protection relay functions needed to
protect a power transformer. This is opposite to a couple of years ago when
it was necessary to have one relay for differential protection, one relay for
earth-fault protection and so on.
The other main part of the fault clearing system is the circuit breaker.
The operating times of circuit breakers have gradually been reduced, but
since all circuit breakers are dependent on a current zero-crossing to inter-
rupt the current, they can never protect the power system from the first
peak of the short-circuit current. Fault current limiters have been proposed
and evaluated for almost 30 years by now. Recent research has proposed a
number of installations of fault current limiters based on solid-state breakers
or superconducting properties. Another approach to limit the fault current
is to install a series reactor.
Since it is easier to close a current path than to open it (provided that the
switch is dimensioned for the mechanical forces that will stress the switch
during the closing), the possibility to commutate a fault-current to earth at
the source with an earthing-switch has been proposed. The possibilities of
today to supervise and control a power system seem to be sufficient to allow
such a solution. The required apparatus and the control system exist but a
field installation is required to prove the design.
The detection of faults is an essential part of the installation irrespective
of whether a current-limiter or an earthing-switch is used. Allowing for a
mechanical operating time of a few milliseconds, faults must have been de-
tected within one millisecond or so to allow the power system to be protected
from the first peak of the fault-current.

1.2 Objectives
The objective of this licentiate thesis is to present results from the project
“Fast fault detection in power distribution systems” performed at the Royal
Institute of Technology in Stockholm, Sweden since the spring of 2000.
Within the project algorithms suitable for fast fault detection have been
1.5. List of publications 5

investigated and their performance evaluated. It turned out that the ex-
pression “fast” is not defined in the context of fault detection. A study of
expressions used for speed requirements of protection equipment has been
performed. The process of measuring and conversion of measurands has
been investigated and the need of processing capacity has been discussed.

1.3 Outline
The thesis begins with a chapter (chapter 2) on electrical faults and their
consequences in a power system. Methods and apparatus for mitigation and
clearing of faults are described in chapter 3. The fault protection system
is briefly outlined in chapter 4 together with common general principles for
fault detection. Next, a few algorithms used for relaying purposes are de-
scribed in chapter 5. Some common sources of errors and how they effect
the fault detection are discussed in chapter 6. The application of the algo-
rithms to two case studies are presented in chapter 7. Finally, conclusions
and ideas for future work are discussed in chapter 8.
To put the work within this thesis in perspective, two appendices have
been included. Appendix A contains a study on voltage dips, possibly caused
by short-circuit faults and appendix B contains a study on simulations of
power systems.

1.4 Main contributions of the thesis

The main contributions of this thesis are:
• a structured survey on the speed of fault detection and on the benefits
of fast fault detection. In particular, “fast” fault detection is defined.
• an evaluation of possible algorithms appropriate for fast fault detec-
tion.
• requirements on equipment and algorithms used for fast fault detec-
tion.
• a case study of the application of fast fault detection in one typical
grid and in one grid at a Swedish steel plant.

1.5 List of publications

The work during the project has been described in the following publications:
6 Chapter 1. Introduction

Conference papers
• Öhrström, M. and Söder, L., “Fast fault detection for power distribu-
tion systems” [1], presented at the 7th IASTED conference, Marina
del Rey, California, USA, 12–15 May, 2002
• Watson, J. and Öhrström, M., “Current transformers, couplers &
coils: A century of Overcurrent Measurement for Power System Pro-
tection” [2], presented at the 37th International Universities Power En-
gineering Conference, Staffordshire University, 9–11 September 2002
• Öhrström, M., Söder, L. and Breder, H., “Fast Fault Detection for
Peak Current Limitation Based on Few Samples” [3], accepted for the
CIRED2003 conference in Barcelona, Spain, 12–15 May, 2003
• Öhrström, M. and Söder, L., “A Comparison of Two Methods used
for Voltage Dip Characterization” [4], accepted for the IEEE Power
Tech conference in Bologna, Italy, June, 2003

Technical reports
• Öhrström, M., “Characterization of voltage dips recorded at a Swedish
industrial plant during 1999” [5], internal KTH-report
• Öhrström, M., “Analysis of voltage dips” [6], internal KTH-report
Chapter 2

Faults and their

consequences

T HE detection and clearing of electrical faults in power systems is the

main topic of this licentiate thesis. To appreciate why a fast and
reliable fault clearing is important this chapter contains an overview of the
consequences caused by electrical faults and relates the consequences to the
duration of the fault.

2.1 Faults
The consequences (in most cases damage or potential hazard to humans and
property) caused by electrical faults in power systems strongly depend on the
magnitude of the fault current, which in turn depends on the type of fault,
the location of the fault, the system earthing, the source impedance, and
the impedance of the fault. The duration of the fault is also of considerable
importance when estimating the consequences of a fault.
One way to characterize the types of faults is to describe them as shunt-
or series faults. Shunt faults are faults when one or more of the phases are
short-circuited (possibly to earth). Shunt faults are in general more severe
than series faults, which could be described as an interruption in one or
more of the phases.
The following definition of a short-circuit is taken from an IEEE stan-
dard [7]:
“An abnormal connection (including an arc) of relatively low
impedance, whether made accidentally or intentionally, between

7
8 Chapter 2. Faults and their consequences

two points of different potential. Note: The term fault or short-

circuit fault is used to describe a short-circuit.”

2.1.1 Shunt faults

An IEC-standard [8] has the following definition of a shunt fault1 :
A fault that is characterized by the flow of current between two
or more phases or between phase(s) and earth at the frequency
of the associated power system.
It is important to realize that fault currents are mainly of power frequency
character but that they also can contain high-frequency components. Such
high-frequency components can consist of the discharge current of a capaci-
tor bank or stray capacitances in cables and bushings. As will be discussed
later many fault detection algorithms use only the power frequency compo-
nent of the current to detect a fault.
The different types of shunt faults are illustrated in figure 2.1. Fault

Figure 2.1. Shunt faults

currents due to shunt faults depend on the system impedance as seen from
the fault location, and of the fault impedance. In general, fault currents
1
The I nternational E lectrotechnical C ommittee (IEC) publish standards that are used
by manufacturers and customer for standardization purposes.
2.2. Consequences 9

are much larger than load currents. However, the magnitude of single-phase
fault currents is largely dependent on the system grounding2 and can be
large in magnitude (of the same order as three-phase fault currents in solidly
grounded systems) or small (a few Amperes in high-impedance grounded
systems).

2.1.2 Series faults

An IEC-standard [8] has the following definition of a series fault:
“A fault for which the impedances of each of the three phases
are not equal, usually caused by the interruption of one or two
phases.”
Series faults give rise to fault currents proportional to the load currents.
Series faults can be due to a broken conductor, a fuse operation in one or
two phases, or a circuit breaker malfunction in one or several phases.

2.2 Consequences
The consequences of a fault can be divided into one part caused by the
initiation of the fault (e.g. insulation breakdown) and one part which is
dependent on the duration of the fault. The initial consequences cannot be
reduced by faster fault detection whereas the part dependent on the duration
of the fault can.

2.2.1 General
Mechanical forces
For parallell conductors in a single- or a three-phase system, the maximal
force imposed on one of the conductors can be calculated by using equa-
tion 2.1 [9].
2
Fmax = k · kr · · i2s kp/m (2.1)
d
where Fmax is the largest force (in kp3 /m) imposed on the conductors, is
is the peak current expressed in kA, d is the distance between neighboring
2
In Sweden, the transmission system (voltages more than 130 kV) is solidly earthed, but
the distribution system (at least for 10 – 40 kV) is either non-earthed or high impedance
earthed. The impedance can be a resistance, an inductance, or a combination thereof.
3
1 kp (kilopond) is equal to 9.82 kN.
10 Chapter 2. Faults and their consequences

conductors expressed in cm, and k and kr are constants. Since k, kr , and

d are design parameters, it can be concluded that for a given power system
component, the maximal force imposed upon it is proportional to the square
of the peak-current. Thus, when a short-circuit current is carried by the
phase conductors, there will be a mechanical force upon them and that force
will grow rapidly with increasing short-circuit currents since it depends on
the square of the current.
In case of a short-circuit, the fault current will contain a decaying dc-
component that depends on the instant of fault initiation. Thus, the fault
current is largest immediately after the fault initiation and the largest me-
chanical force imposed on the power system components is caused by the
largest peak current. An example of typical fault phase currents is illustrated
in figure 2.2.
100

80

60

40

20

−20

−40

−60

−80
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Figure 2.2. Typical fault phase currents, each including a dc-component

Power system equipment is designed to manage a current peak of a cer-

tain magnitude. The organisation IEC4 have published a proposed standard
value of the maximal allowed peak current expressed in terms of a peak fac-
tor that relates the RMS5 -value of the short-circuit current to the peak
current (equation 2.2)6 .
Ipeak = 2.5 · Irms (2.2)
4
International Electrotechnical Commission
5
Root M ean S quare
6
The figure 2.5 is based upon measurements
√ and calculations in a multitude of actual
power systems. Another figure 2 · 2 ≈ 2.83 is often used as an estimate of the peak
2.2. Consequences 11

Thermal stress
Due to resistance in the conductors of a power system, heat losses according
to equation 2.3 are produced when a current flows through them.

P = R · I2 (2.3)

In equation 2.3, P denotes the heat losses for a conductor caused by the
current I, when passing through the resistance R. I is the RMS-value of the
current.
Since the losses depend on the square of the current, fault currents give
rise to major heating. Depending on the size and material of the conductor,
the heating will eventually lead to meltdown and destruction of the con-
ductor. Power system equipment is designed to manage fault currents for
a certain amount of time. The organisation IEC have published standard
values of allowed short-circuit currents and how long they are allowed to
persist. If the fault current is lower than the rated value, the time before
meltdown occurs is increased. An estimation of the allowed period of time
can be obtained by calculating the heat loss developed in the conductor when
the rated short-circuit current I1 is passed through the conductor during the
rated time t1 . If the actual current is I2 , an estimation of the allowed time
t2 can be calculated by using equation 2.4.

I12 · t1 = I22 · t2 (2.4)

Power system apparatus are normally designed to carry fault currents during
a rather long time (one or three seconds are common rated values according
to the IEC). If the thermal consequences where the only consequences there
would be plenty of time to detect and clear faults.

2.2.2 Faults involving an arc

If the fault occurs outdoors (or at least in a non-enclosed environment) the
fault current will be limited due to the resistance of the arc which can be
estimated by an empirical formulae published by Warrington [10]. Still, an
electric arc burning outdoors will eventually cause damage to the power line
and the fault current must be cleared.
factor. It is derived by assuming a dc-component of 100 % and that there is no damping
in the system so that the fault current barely approaches zero. Close to large generators,
the dc-component can be even larger and if the damping is poor, the fault current might
not pass through zero for several periods.
12 Chapter 2. Faults and their consequences

If the fault occurs in an enclosed environment there is no empirical for-

mulae for the arc resistance and the fault current limitation due to the arc
is negligible. When an arc burns in an enclosed environment, such as for
example a switchgear cubicle, the damage will be severe. First, the power
dissipation will cause heating of the air inside the cubicle. The pressure
inside the cubicle will increase and a pressure wave will travel through the
switchgear. Approximately 10 ms [11] after the initiation of the arc the
pressure wave will be fully developed. Modern switchgear cubicles are in
general equipped with pressure relief valves that open to prevent some of
the damage caused by the rise in pressure. If the arc is not interrupted
the temperature will continue to rise. After approximately 100 ms [11] the
conductors will begin to melt and the switchgear will need extensive repair
or most likely a replacement. A fault current of 40 kA through an arc will
vaporize approximately 4 kg of material7 if burning for 1 s [11].

2.2.3 Consequence steps

Once a fault has been initiated it will cause some initial consequences which
are difficult (impossible) to do something about. What one can do is to
protect the system from further consequences. The concept of consequence
steps starts with the initial consequences. Subsequent consequences are
time-dependent. The faster the fault can be cleared, the less are the con-
sequences. Based on the previous discussion some steps can be identified
such as: reduction of the pressure caused by an open arc, reduction of the
mechanical stresses by limiting the first current peak, and limitation of the
thermal consequences by minimizing the fault time.

2.2.4 Cost
The concept of consequence steps could also be related to the cost caused
by the fault. The initial consequence could for example lead to insulation
breakdown in a cable so that the cable needs to be repaired. If the fault
current is allowed to persist, the cable will be more and more damaged
so that it might need to be replaced. Eventually, the fault current will
damage not only the cable, but also other equipment in the faulted current
path, such as circuit breakers, instrument transformers, transformers, or
even generators. Replacing or repairing part of a cable might not be very
7
If the conductor is made out of copper with an area of 10x100 mm2 , 4 kg of copper
corresponds to a meltdown of approximately 10 cm of the conductor.
2.2. Consequences 13

expensive, but replacing a power transformer most certainly is. Another

aspect of cost is that the larger the damage is the longer time the repair
will take before the cable can be put into service again. Depending on the
layout of the power system more or less customers might be without power
during the repair.

2.2.5 System aspects

A short-circuit normally gives rise to high fault currents in the faulted cur-
rent path and consequences associated therewith. At the same time, the
voltage will drop, giving rise to a voltage dip that can be noted throughout
large parts of the power system and disturb sensitive processes. A study of
voltage dips measured at an industrial plant is included in appendix A. For
that particular study, the critical duration of a voltage dip depended on the
remaining voltage during the dip8 , but many processes managed a voltage
dip if the duration was less than 100 ms. However, if the duration of the
dip were even shorter, its consequences would be further reduced. Transient
stability considerations is generally not considered for distribution systems.
However, the increasing amount of distributed generation, might lead to
such considerations even for distribution systems.

8
The magnitude of the dip is an expression that sometimes is used. But if the mag-
nitude of a dip is say 30% it is unclear if the remaining voltage is 30 % or if the missing
voltage is 30%.
14
Chapter 3

Fault clearing

I N the early days of power systems (late 1800s), an electrical fault could
be detected visually by the operator and then manually removed. Today,
electrical faults are automatically cleared by fuses and circuit breakers, or
limited by fault-current limiters. This chapter gives an overview of methods
and apparatus used for fault clearing.

3.1 Fuses
Fuses have been used in power systems since the late 19th century [12]. A
fuse is a device which can carry load currents, but when the current rises
above a certain threshold for a certain time, the conducting path will break
down and the fuse will clear the current. Fuses are common in low voltage
systems and in some extent in medium voltage systems but not so common
in high voltage systems. Depending on the design (there are fuses that are
non-current limiting and there are fuses that are current limiting), a fuse can
limit the first current peak of the short-circuit current. The fuse is thus an
excellent device for fault clearing but they have a number of disadvantages
which limit their use:
• After a fuse has cleared a current it has to be replaced.
• When load currents flow through a fuse they give rise to losses propor-
tional to the square of the current (as previously discussed in chapter 2
equation 2.3). For large load currents the losses can be unacceptably
high.

15
16 Chapter 3. Fault clearing

• There are power systems where it can be difficult to coordinate fuses

with the remaining protection system so that selectivity1 is obtained.

3.2 Circuit-breakers
Power circuit-breakers are used at all voltage levels in a power system. Most
common today in medium voltage systems are either SF6 circuit-breakers2
or vacuum circuit-breakers. The basic principle of current interruption of
an alternating current is based on the natural zero-crossings of the current
twice per period of the power frequency. Once the contacts of the circuit-
breaker is opened, the current will continue to flow through an arc until a
natural zero-crossing is reached. At that moment the arc is extinguished
and if the contact separation is large enough and the arcing channel has
cooled sufficiently, the current will be interrupted. Otherwise, the current
will flow in the circuit until next natural zero crossing before an interruption
can take place.
Since the interruption principle is based on the natural zero-crossings
of the short-circuit current it is impossible to protect the system from the
mechanical forces caused by the first current peak by using a circuit-breaker.
There is no way the first current peak can be avoided with such circuit-
breakers. Even if the operating time of the contacts is made infinitely small,
the circuit-breaker would still need a zero-crossing to interrupt the current.

3.3 Current limiting

The expression “current limiting” is used for apparatus which can protect
the power system from the mechanical forces associated with the first current
peak of the fault current by either interrupt or limit the fault current before
the prospective current peak is reached.

3.3.1 Series reactor

A traditional method of limiting short-circuit currents has been to install a
so-called series reactor in the main circuit. However, the reactor does not
1
Selectivity is a term used for the coordinated operation of a protection system so that
only as small part of the power system as possible is disconnected due to the clearance of
a fault.
2
Due to environmental considerations there is a discussion whether the use of SF6
should be prohibited by regulation or not.
3.3. Current limiting 17

only limit the short-circuit current but the installed short-circuit power is
also reduced. Since the reactor is a series circuit element it also contributes
with additional losses when load currents pass through it. The series reactor
is always connected to the circuit3 and needs no external signal or control to
limit the fault current. Actually, the fault current is not limited very much,
just as much as needed for the existing protection system to cope with it.

3.3.2 Fuses
There are fuses that are of so-called current limiting type i.e. they can limit
the fault current before the first current peak of the prospective fault current.
When the fault current exceeds a certain threshold, the fuse rapidly melts
and the fault current flows through an arc within the fuse until the next
current zero where it is extinguished. The impedance of the arc limits the
fault current during the arcing time. Even though the current limitation can
be significant, the limited current still flows in the circuit until it is finally
cleared. As for the series reactor the fuse does not need any external signal
or control to operate but is “triggered” by the fault current itself.

3.3.3 A switch-fuse combination current limiter

In general, the designer of a power system wants to have as high avail-
able short-circuit power as possible to allow smooth operation of the power
system. A higher short-circuit power allows the connection of larger loads
without voltage reductions or special considerations when connecting the
load4 . However, a higher short-circuit power has the side-effect that in case
of a fault, higher short-circuit currents will flow in the power system. When
the power system is supplied from several sources a solution utilizing fuses
have been used for several years by now. The system is sectionalized through
a switch-fuse combination, which in case of a short-circuit operates the fuse,
hence separating the two sources from each other. After the fuse has oper-
ated, the protection system of each individual source detects and clears the
fault, leaving healthy parts of the power system in operation.
The switch-fuse combination consists of a conductor in parallell with a
current-limiting fuse. When a short-circuit fault is detected, the conductor
3
In some installations, the reactor can be removed from the circuit by a bypass switch.
4
A synchronous machine can for example be started directly which is the most severe
way to start the machine or by using thyristors to gradually increase the power to the
machine which is a nicer way to start the machine.
18 Chapter 3. Fault clearing

is forced open by a small explosive charge forcing the current to commutate

to the fuse. Due to the high current, the fuse melts down and limits the
current before the first peak of fault current, hence providing a protection
from the mechanical force associated with the first current peak.
The benefit with using a switch-fuse combination is that connections
otherwise not allowed due to possibly high short-circuit currents can be
used. The disadvantage of using a switch-fuse combination is that after it
has operated, parts of it must be manually replaced before it can be taken
into service again.

3.3.4 Solid-state fault current limiters

The advent of high power solid-state devices such as thyristors, GTOs, and
IGBTs has provided a means of building a solid-state breaker which is not
dependent on a natural zero crossing to interrupt the current. Once a short-
circuit is detected, a signal is sent to the control-gear of the solid-state
breaker, which turns off the current in the main circuit immediately.
Inductances are not only explicitly as components in a power system
but also natural as currents give rise to magnetic flux. Magnetic energy is
stored in the inductances according to W = Li2c , where ic is the current
immediately before interruption. When the current is interrupted close to
the peak of the current, the energy is at its maximum and manifests as a
transient over-voltage across the breaker which can be of such magnitude
as to disturb or damage equipment in the system. When ic is high the
energy must be handled by for example a varistor across the breaker. The
advantages of a solid-state breaker is that they can break the current in
practice instantaneously. The disadvantages with a solid-state breaker are
however:
• the cost5 .
• depending on the components used, the components can not extin-
guish large currents6 .
• the losses are quite high.
By adding a switch in parallel with the solid-state equipment that commu-
tates the current to the solid-state breaker just before switching, some of
the disadvantages might be resolved. In that way it is possible to use less
5
The cost of solid-state equipment has been an issue for discussions for several years.
The trend seems to be towards better performance and towards lower cost.
6
A thyristor is dependent on the zero-crossing of the current for its interruption.
3.3. Current limiting 19

expensive solid-state components7 thus lowering the cost (as long as the
switch is cheaper than the solid-state components). The control system of
the apparatus can be constructed so that the switch never operates if the
current is larger than the solid-state components can handle. Then a new
disadvantage is that the component might not be able to limit the current
before the first current peak (if the switch is slow). The losses can also be
reduced with a switch in parallel that carries the load current. It is eas-
ier to construct a mechanical switch with low impedance than a solid-state
component with low impedance. It must be observed that to commutate
a current can be a difficult task and that the operating time of the switch
adds to the total fault clearing time.

3.3.5 Superconducting fault current limiters

One type of superconducting fault current limiter is based on a transformer
that has a secondary winding that is made as a superconductor. In nor-
mal service operation the superconducting state persists but in case of a
short-circuit, the superconducting state is disturbed causing the secondary
winding to be normal-conducting. The inductance of the secondary winding
is then changed drastically and the current hence limited. A disadvantage
with superconducting current limiters are that they are quite expensive.
One advantage of superconducting current limiters is that no equipment to
detect a short-circuit current is needed, since the short-circuit current by
itself disturbs the superconducting state.

3.3.6 Current diverter

Since interrupting a current is more difficult than closing a switch in a
faulted circuit another solution that provides functionality similar to current
limiting apparatus is to divert the fault current from the power system at the
source [13]. One or a number of fast switches diverts the current to earth
once a short-circuit is detected. The current is then interrupted by the
ordinary protection system. In case of a meshed grid with several sources8 ,
a switch is placed at each source, responsible for diverting that sources part
of the fault current. The effect of using current diverters are that the current
7
The components might be degraded because they are not in the main circuit all the
time but only when a switching takes place.
8
The word source is used to denote either a local generator or a transformer that is
connected to a transmission grid or a local distribution grid.
20 Chapter 3. Fault clearing

at the fault point becomes virtually zero, but instead a temporary voltage
dip down to zero is imposed on the system for the time it takes for the
ordinary protection to interrupt the fault, open the current diverters, and
close the current path again. But then again, without the current diverter
this type of power system connection would be difficult to realize. Other
types of current interruption equipment also give rise to a voltage dip when
clearing the fault current. A current limiter might get away with a less
severe voltage dip but the cost is considerable higher than that of a bypass
switch.

3.4 Motivation for fast fault detection

Except for the fuse and the superconducting current limiter all of the com-
ponents mentioned in this chapter are in need of fast and reliable fault
detection to operate as intended and to protect the power system. To be
able to limit the fault current before the first current peak, the fault has to
be detected in at least five ms after fault initiaton9 (if it is assumed that
the apparatus used to clear the fault has an operating time that is zero). In
practice, the operating time of a switch that is used for current diversion is
approximately 2–3 ms. If a safety margin of 1–2 ms is enough to make sure
that the fault current is limited before the first current peak, approximately
1 ms is left for the fault detection.

9
corresponding to a quarter of a period of the power frequency.
Chapter 4

Fault protection systems

4.1 Fault clearing systems

A fault clearing system consists of a relay protection system and a circuit-
breaker. In case of a fault, the task of the circuit-breaker is to clear the fault
and the task of the relay protection system is to detect the fault. The circuit-
breaker has already been discussed in the previous chapter 3. It is important
to understand that the time to clear a fault is dependent on both the time
required to detect the fault and the time needed for the circuit-breaker to
clear the fault.

4.2 Relay protection system

The relay protection system can be further divided into transducers, auxil-
iary power, trip-circuits, and relays. The relay performs the actual detection
but cannot work without proper inputs and outputs.

4.2.1 Transducers
To be able to detect a fault, the state of the power system must be observed
and analyzed. The most common signals used for fault detection are cur-
rents, and voltages. Furthermore, light can be used to detect an open arc
fault inside an enclosed switchgear and is not further treated here. A pres-
sure gauge can be used in a transformer to detect pressure waves generated
by a fault.

21
22 Chapter 4. Fault protection systems

Currents and voltages cannot be measured directly, i.e. it is not possible

to connect an amperemeter or a voltmeter at high voltage, but some mea-
suring equipment that reduces the voltage must be used. Some common
groups of measuring equipment are described below.

Instrument transformers
Instrument transformers has been in use since the late 19th century. They
are commonly abbreviated as CT for current transformers, VT for voltage
transformers (or PT for potential transformers), and CVT for capacitive
voltage transformers. Instrument transformers consists in principle of two
windings wound round an iron core. The connection of a CT and VT is
different since the CT is connected in series with the main current path and
the VT is connected as a shunt-device to earth (if phase-to-earth voltages
are measured) or between phases (if phase-to-phase voltages are measured).
During the years CTs and VTs have been extremely reliable components
(with respect to their accuracy of reproducing the primary voltage or current
at the secondary terminals) and their behavior is with a few exceptions well
known to power system engineers. Things to consider when using VTs, CTs
or CVTs as inputs for fast fault detection are:
• Saturation of the core of the CT, which is caused either by a high
ac-component or a dc-component. A saturated CT supply little or no
secondary current (at least when the primary current is of power fre-
quency character) depending on the secondary burden and the amount
of saturation. However, it takes a while for a CT to saturate so for fast
fault detection CT saturation might not be a problem. If fast fault
detection is supposed to be used in systems where fast reclosing is
used, CT saturation could be a consideration if the fault still persists
when the reclosing is made and there is remanent flux in the core.
Saturation of the core of a VT is not considered as a problem for
fast fault detection since the voltage normally is reduced during a
fault thus providing larger margins to the knee-point on the BH-curve
where saturation commences.1
• The bandwidth of an instrument transformer is a measure on how the
instrument transformer will behave at various frequencies. An instru-
ment transformer for use in power systems is normally designed for
1
A saturation phenomenon related to VTs is ferroresonance which is a resonance be-
tween the non-linear inductance of the VT and the capacitances of the network in which
the VT is connected.
4.2. Relay protection system 23

use at 50 or 60 Hz. For fast fault detection, it might be of interest to

study transients at higher frequencies than 50 or 60 Hz. The band-
width will then tell us how the instrument transformer is likely to
react for those transients. If the power frequency component is used
for the fault detection the bandwidth of instrument transformers is of
course sufficient since they are designed for such frequencies.
• Impulse and step responses are tools used to estimate the bandwidth
of an instrument transformer. An impulse contains by definition all
frequencies. A comparison of the frequency content on the secondary
side of an instrument transformer will reveal the frequency response
of the instrument transformer. The step response gives additional
information regarding possible delay and oscillations introduced by
the instrument transformer.
• Accuracy of the instrument transformer. The error of an instrument
transformer can be expressed as either a magnitude error and a phase
error or as a composite error that is a mean value over a period of
power frequency. The composite error takes into account both the
magnitude and the phase error. The phase error of an instrument
transformer is a measure of the delay of the instrument transformer.
The phase error of typical voltage and current transformers are small
and no significant time delays are introduced. In IEC-standards, pro-
posed standard values of the magnitude and phase errors of instru-
ment transformers are published. The maximal phase error that is
allowed is in the range of 3 degrees, which corresponds to a delay of
20 × 3/360 ms ≈ 0.17 ms in a power system with fundamental fre-
quency of 50 Hz, and approximately 0.14 ms in a power system with
a power frequency of 60 Hz.
It is important to consider instrument transformers as an integral part of
the power system since the connected burden effects its performance.

Non-conventional transducers

Non-conventional transducers is a bad expression and is used by power sys-

tem engineers to describe other transducers than instrument transformers.
Well known examples of non-conventional transducers are:
• Voltage dividers that are mainly capacitive, resistive or combinations
thereof. A capacitive voltage divider is used as the front-end in a
capacitive voltage transformer.
24 Chapter 4. Fault protection systems

• Rogowski coils, which are wound round a non-magnetic structure

(i.e. the Rogowski coil is an air core transformer) and are sensitive
to the flux change caused by the power system current. The output
is proportional to the time-derivative of the current and is a small
voltage signal (millivolts). The output is integrated and processed
in a microprocessor to give the current in the conductor. The Ro-
gowski coil is not a new concept but was investigated and published
by Rogowski as early as in the 1920s.
• Low power current transducers which in fact is a current transformer
with the secondary winding connected to a fixed resistance. The sec-
ondary output is taken as the voltage drop across the resistance and
processed in a microprocessor to give the current.
• Optical sensors are sensitive to either the magnetic (Faraday-effect)
field or the electric field (Pockel-effect).

Of the above mentioned transducers the voltage divider, the Rogowski coil,
and the low power current transducer have found some use for switchgear
manufacturers. The optical sensors are typically on the development stage
and awaits further investigation.

Location of transducers

The location of transducers in the power system can be a difficult issue.

There is a fine balance between the room and the cost as opposed to the
requirement of the protection system. Conventional instrument transform-
ers with their large amount of transformer iron in the core are expensive
apparatus in the power system. Non-conventional transducers are much
cheaper if only the single component is studied. However, to interpret the
signals some kind of merging unit and processing capability is needed. How-
ever, non-conventional transducers are often possible to integrate in other
power system equipment such as bushings, circuit breakers, insulators, or
cables. This way a larger number of measuring points are economically fea-
sible as compared to when conventional instrument transformers are being
used. The nonconventional transducers can also be performed as combined
sensors that give both voltage and current measurement where conventional
instrument transformers would have given only one of the quantities or both
CTs and VTS would have had to be used.
4.2. Relay protection system 25

4.2.2 Wiring

In a distribution system, the transducers are typically located close to the

relay so that the amount of wiring is limited. In a large transmission sub-
station the distance from the transducer to the relay can easily be 50 m or
more.

4.2.3 Trip coil

Traditionally, circuit breakers are operated by releasing energy stored in a

large spring. The latch which holds the spring is released by energizing a coil
that provides a force that acts on the latch thus releasing it. The opening
coil needs a rather large power to operate the latch. The power is supplied by
a circuit to which the contact of the relay is connected. Thus the relay must
have contacts that are capable of closing and opening currents of say 2–3
A. Such contacts contribute to the operating time of the relay and is added
to the detection time. The additional operating time must be considered
when comparing the operating times of different relays. The contacts of a
relay can typically have an operating time of 5 ms, which means that relays
are not suitable for fast fault detection. However, the operating time can be
reduced by using for example a field effect transistor to close the contacts
of the relay, a method that has been used for an application where it is
important with a well defined closing time.

4.2.4 Relays

Electromechanical relays

Electromechanical relays were the first relays to be used in power systems for
protection purposes. The inputs to an electromechanical relay are currents
and voltages from CTs and VTs in the power system.

Solid-state relays

Solid-state relays also take their input signals from the power system with
the aid of CTs and VTs. The relay characteristic in a solid-state relay is
achieved by circuit design based on operational amplifiers.
26 Chapter 4. Fault protection systems

Digital and numerical relays

The advent of the microprocessor led to the investigation of using computers

for the protection of power systems. Since the cost of components were high
it was thought that one computer would be used for all protection tasks in a
substation. A numerical relay2 consists of a signal conditioning subsystem,
a conversion subsystem, and a digital processing subsystem [14].
The signal conditioning subsystem contains transducers, and analogue
filtering. The conversion system contain the A/D converter and sample-and-
hold circuits. Analogue to Digital (A/D) converters are essential in modern
protective relay equipments to be able to handle the data digitally. The A/D
conversion can be treated in many different ways. Often it is important to
have simultaneously taken samples of say phase voltages and phase currents.
One way of achieving that is to put A/D converters on every input. However,
the A/D converter is an expensive component so it would be advantageous
to use less A/D converters. A solution is to sample-and-hold which means
that the samples are taken simultaneously on all input channels, hold by a
circuit consisting of a capacitor and then selected for digital conversion by
a so-called multiplexer. The samples is then sequentially converted but due
to the hold-process they are in fact simultaneous. The A/D converter can
be one of typically three of the following types which are discussed in more
detail by Johns and Salman [14] or Demler [15]:
• sequential counter. A counter is started from zero and the output of
the counter is compared to the analogue input after a D/A3 conversion.
If the input is larger than the counter, the counter is incremented and
the comparison is performed again. The process will continue until
the input is larger than the value of the counter which then is taken
as the digital representation of the analogue input. The principle is
easy, but a disadvantage is that the conversion takes longer time for
large numbers since the counter starts from zero for each sample.
• single or dual slope. The input of the A/D converter is integrated for
a fixed time T1 and stored in a register. At that time, a fixed reference
voltage is connected to the input of the A/D converter and integrated.
The time T2 required for the value in the register to become zero is a
measure of the input of the A/D converter.

2
A microprocessor relay where the relay characteristic can be programmed into the
microprocessor.
3
Digital to Analogue
4.3. Basic protection principles 27

• parallel comparators. This type of A/D converter consists of equally

many comparators as the resolution of the A/D converter, i.e. a 8-
bit A/D converter consists of 28 = 256 comparators. The input of
the A/D converter is applied to the comparators. The output of the
comparators are either zero if the input is less than the value of that
comparator or one if the input is larger than the value of that com-
parator. The number of zeros and ones is then converted to a 8-bit
digital number that is the digital representation of the input.
The three types of A/D converters are quite similar and all of them will
suffice for fast fault detection purposes. For measuring equipment intended
for very high frequencies, the choice of A/D converter is of more concern
and it is probable that the last described type will be used, i.e. parallel
comparators.

PC platform based relays

An early paper on digital protection predicted that all protective task within
a substation would be handled by a single computer due to cost reasons.
Instead the trend has been towards single protective relays for each feeder of
the substation (and also one single protective relay for different protection
tasks in that feeder). The advent of bay terminals made it possible to
gather different protective tasks in a feeder into one single unit. Today, thirty
years later, there exists system where all protective tasks of a substation is
performed by one single industrial pc. One such system is described in (x)
where its primary function is the control of an HVDC converter station,
but all protective tasks in the substation was possible to implement in the
industrial pc. The feeders are equipped with I/O circuit boards which gather
current and voltage signals from the feeders and transfer them to the pc.
It is possible to implement simple protective task in the I/O circuit board
which contains a processor.

4.3 Basic protection principles

4.3.1 Magnitude relays
The magnitude of for example a current is compared to a predetermined
threshold level. Whenever the magnitude is above the threshold it is de-
termined that a fault is detected. The magnitude is not the instantaneous
28 Chapter 4. Fault protection systems

value of the current but instead the RMS-value or the peak value of the
current.

4.3.2 Directional relays

The magnitude relay is extended so that a phase angle between the current
and a so-called polarizing quantity is estimated. Then it is possible to
determine whether the fault is in the forward or in the reverse direction
as seen from the relay depending on the value of the estimated phase angle
compared to the expected phase angle.

4.3.3 Impedance relays

The impedance relay measures a voltage and a current and makes an esti-
mate of the impedance of the protected object. The impedance relay can
be made directional and newer relays can have very complicated tripzones
(i.e. the part of the power system that the relay is designed to protect).
Distance relays are another name for an impedance relay but can also be
equipped with a magnitude relay to further improve its ability to provide
protection for the power system object that it is designed to protect.

4.3.4 Differential relays

The principle of differential relaying is always a unit protection, i.e. a differ-
ential relay protects only the object it is supposed to protect. It cannot be
used for backup protection. The principle is simple; the current measured
into the zone of protection must be equal the current measured out of the
zone of protection, otherwise a fault has occurred within the zone of protec-
tion. Even though the principle is simple many things must be considered
when implemented in a power system such as rated burden and turn ratio.
A single relay requires inputs from at least two different sets of CTs thus the
communication of those signals within the substation must be considered.
Differential protection is a common protection principle for transformers,
motors, and generators.

4.3.5 Pilot relaying

Pilot relaying is not really a relaying principle. It can consist of any of the
basic types mentioned above, the difference is that a communication link is
added so that two relays that protect a transmission line can communicate
4.4. Speed of protection 29

with each other even when placed at separate ends of a transmission line.
If the concept of pilot relaying is interpreted as the ability of two or more
relays to communicate with each other the increased use of numerical relays
for power system protection will provide an excellent platform for all relays
to be pilot relays.

4.4 Speed of protection

Fault detection has traditionally been performed by protection relays taking
a certain amount of time to detect a fault. The time needed for the fault
clearing is thus dependent on the fault detection time and the apparatus
operating time and in case of an ac-breaker it is also dependent on the
time until next natural current zero crossing. However, the speed of the
protection equipment is not well-defined in standards. When discussing the
speed of fault detection a number of key expressions are commonly used
such as:
• High-speed relay, defined by IEEE standard [7] as: “A relay that
operates in less than a specified time. Note: The specified time in
present practice is fifty milliseconds (three cycles on a 60 Hz basis)”
• Fast fault detection, commonly used for fault detection within a period
of power frequency but sometimes used for fault detection as fast as
40 microseconds [16].
• Very fast fault detection, commonly used for fault detection within
half a period of power frequency but sometimes as fast as a few mil-
liseconds [17].
• Ultra high speed relaying, commonly used for fault detection within a
quarter of a cycle. Often implemented to detect the travelling waves
caused by faults in EHV or UHV overhead lines [18].
For the remainder of this thesis fast fault detection is defined as:

Definition 4.1 Fast fault detection (FFD) is defined as “fault detection

for systems that require fault removal or limitation before the first current
peak after the initiation of the short-circuit.”
30
Chapter 5

Algorithms

Algorithms for fault detection in numerical protection relays have been of

academic and industrial interest since around 1970. Early algorithms were
constrained by the available computer performance of that time and com-
monly implemented in a low level machine language1 in order to speed up
the algorithm. The performance of computers of today makes it possible to
implement algorithms in a high-level language such as for example c or c++
thus making it possible to use the same software in different microproces-
sors. This chapter contains a brief description of common algorithms used
in numerical protection relays. The source of the original material is Johns
and Salman [14], and Phadke and Thorp [19]. The additional discussion on
the suitability to use the algorithms for fast fault detection is contributed
by the author (except when explicitly stated).

5.1 Waveform algorithms

This class of algorithms has a common property, namely the assumption
that the post-fault voltage and current can be described by a sinusoidal
signal s(t) as for example in equation 5.1.

s(t) = Smagn · sin(ωt + ϕ) (5.1)

Sampled data values of current and/or voltage are then fitted to the sinu-
soidal waveform using one of a number of available methods. The result is
1
Often unique to a family of microprocessors and not easily transferred to other mi-
croprocessors.

31
32 Chapter 5. Algorithms

estimated values of Smagn , ω = 2πf , and ϕ. The frequency f of the sinu-

soidal signal is often assumed to coincide with the nominal power frequency2
of the power system so that only the magnitude and the phase needs to be
estimated. By comparing the estimated magnitude with the magnitude dur-
ing normal operation, a fault can be detected (the principle of a so-called
magnitude relay as described in chapter 4).
The short-circuit current due to a fault often contains a decaying dc-
component with a magnitude depending on the fault inception angle (as
illustrated in figure 2.2). Equation 5.1 does not take into account that
dc-component. Whenever the fault current contains a dc-component, the
estimation of the fault current will therefore contain an error. The wave-
form algorithm that was selected for the case studies (the LSQ-algorithm of
section 5.1.3) does not take into account the dc-component. However, it was
demonstrated in chapter 6 that the error introduced by the dc-component
in the fault current was negligible (for this algorithm) for the first current
peak of the fault current.

5.1.1 Two samples

Description

Assume that two consecutive current samples i0 and i1 at the time instants
t0 and t1 respectively are available and that the angular frequency ω in equa-
tion 5.1 corresponds to the nominal power frequency of the power system.
Substituting into equation 5.1 gives two equations for solving the unknown
parameters (Imagn , and ϕ).

i0 = i(t0 ) = Imagn sin(ω0 t0 + ϕ) (5.2)

i1 = i(t1 ) = Imagn sin(ω0 t1 + ϕ) (5.3)

Imagn , and ϕ can now be solved from equations 5.2 and 5.3.
If it is first observed that the time-derivative of equation 5.1 is:

i (t) = ωImagn cos(ωt + ϕ) (5.4)

2
A reasonable assumption because the fault current caused by a shunt-fault is of power
frequency character. Furthermore, the subject of this thesis being FFD a frequency de-
viation will not cause any significant error within the detection time (approximately 1
ms).
5.1. Waveform algorithms 33

and then that:

i2 (t)
i2 (t) + 2
= Imagn sin2 (ωt + ϕ) + Imagn
2
cos2 (ωt + ϕ)
ω2 (5.5)
2
= Imagn (sin2 (ωt + ϕ) + cos2 (ωt + ϕ)) = Imagn
2

Then the magnitude Imagn can be estimated from one current sample and
one current derivative sample. The current derivative however is not always
readily available. Two consecutive current samples can be used to estimate
the derivative 5.6.
i (t) = (i(t1 ) − i(t0 ))/∆t (5.6)
where ∆t is the time between the two samples (∆t = t1 −t0 ). The magnitude
can then be estimated from equation 5.5 by substitution of equation 5.6
which gives equation 5.7.

2 1
Imagn (t1 ) = i21 + (i1 − i0 )2 (5.7)
ω 2 ∆t

Suitability for fast fault detection

Since only two samples are needed to estimate the magnitude, the algorithm
has the potential to be fast. However, the calculation of the derivative
can produce poor estimates if the samples are noisy or of poor quality.
The derivative is calculated by taking the differential between two current
samples and then divide the differential with the time difference between
the two samples 5.8.
(i(t) − i(t − ∆t))/∆t (5.8)
If the time difference ∆t is small the differential will be divided with a small
number, hence magnifying possible errors in the differential. An example of
this is given in chapter 6, where the algorithm is applied to an actual sampled
voltage signal that is sampled at a high frequency but with poor quality of
the samples. The algorithm may be interesting for FFD, but not further
studied in this thesis. Another waveform algorithm (see section 5.1.3) was
selected for further investigations.

5.1.2 Fourier methods

Description
The discrete Fourier transform (DFT) can be used to estimate the magnitude
of a sampled signal. The assumption that the signal can be described by
34 Chapter 5. Algorithms

equation 5.1 still holds, i.e. , the signal is assumed to be sinusoidal and of a
known frequency which coincides with the nominal power frequency of the
power system. The literature describes full-wave and half-wave versions of
DFT based algorithms. The full-wave algorithm uses sampled data from a
full period of the nominal power frequency, whereas the half-wave algorithm
uses sampled data from a half period of the nominal power frequency.
If the measurement noise is assumed to have zero mean and to be un-
correlated between samples, the DFT can be shown to give the optimum fit
of the sampled data to the sinusoidal waveform. Algorithms based on DFT
methods are used in many modern numerical relays due to its excellent prop-
erties (optimum fit, noise reduction, and harmonic filtering) and simplicity.
The fast Fourier transform (FFT) is an algorithm which calculates the DFT
of a discrete signal. The FFT can be applied when the signal is of a length
N (N = number of samples) so that 2n = N for some integer value of n.

Suitability for fast fault detection

Algorithms based on the discrete Fourier method are not sensitive to an

increase in the sampling rate (as the derivative was) as will be shown in
chapter 6, where the algorithm (calculation of the DFT and identification
of the power frequency component) is applied to actual sampled data. How-
ever, since sampled data from at least half a period is used, the speed of
protection is not increased by raising the sampling frequency. The esti-
mate of the magnitude will still only be available after at least half a period
(corresponding to 10 ms when the power frequency is 50 Hz).
Therefore, fault detection within 1 ms from fault inception is not possible
with the described method.

5.1.3 LSQ-methods
Description

The abbreviation LSQ is a short notation for Least Squares. Phadke and
Thorp [19] describes an algorithm for estimation of the magnitude and the
phase of a voltage or a current signal based on three consecutive samples;
each sample giving one equation for solving the magnitude and the phase.
Since only two unknowns are estimated and three equations are available,
the system of equations is over determined and the algorithm fits the samples
by a least square error method to a sinusoidal of nominal power frequency
5.2. Model algorithms 35

(see equation 5.9, which can be derived from 5.1 by using a trigonometric
identity sin(x + y) = sin(x)cos(y) + cos(x)sin(y)).

i(t) = Ic cos(2πf t) + Is sin(2πf t) (5.9)

If i−1 , i0 and i1 denote three consecutive current samples, the estimated

magnitude of the current (at time t0 ) is given by:


|I| = Ic2 + Is2 (5.10)

where
i1 cos θ + i0 + i−1 cos θ
Ic = , (5.11)
1 + 2 cos2 θ
and
i1 − i−1
Is = (5.12)
2 sin θ
and θ is the power frequency angle between two consecutive samples θ =
2πf0 · ∆t, where f0 is the power frequency of the power system and ∆t is
the time difference between two consecutive samples.

Suitability for fast fault detection

This algorithm is also potentially fast since only three consecutive samples
are needed for an estimate of the magnitude. However, a similar reasoning
as when taking the derivative of a signal is applicable (see section 5.1.1).
The division by 2 sin θ in equation 5.12 (a small number for high sampling
frequencies (∆t → 0)) is numerically sensitive to errors in the sampled signal.
This is further illustrated in chapter 6, where the algorithm is applied to
actual sampled data of poor quality.
The algorithm will be further investigated in the case studies in chapter 7.

5.2 Model algorithms

Description
A common property for model algorithms are that the protected power
system object is modelled by differential equations. A transmission line
can for example be modelled by an RL-link or a pi-link. Sampled values
of voltage and current measured in the power system are then fitted to
the differential equations. The result from the algorithm is estimates of
36 Chapter 5. Algorithms

the impedance of the protected power system objects. A fault is detected

by comparing the estimated impedance to the nominal impedance of the
protected object, i.e. a ratio relay as described in chapter 4.
Phadke and Thorpe [19] describe an algorithm for estimation of apparent
impedance to a fault based on three consecutive samples. The algorithm
fits the samples by solving a differential equation that models the protected
object as a resistance in series with an inductance (see figure 5.1).

Source
i Z=R+jωL

u Fault

Figure 5.1. The power system used in the differential equation algorithm.

The relationship between i and u can be written:

di
u=L + Ri (5.13)
dt
The impedance R + jωL can now be estimated by fitting current and
voltage samples to equation 5.13. Let (ik , ik+1 , ik+2 ), and (uk , uk+1 , uk+2 )
denote the sampled values of current and voltage, the estimated impedance
of the protected object are then given by equation 5.14. If the estimated
impedance is high there is no fault and R + jωL is an estimation of the
impedance of the load. If the estimated impedance is low there is a fault
and R + jωL is an estimation of the impedance of the fault.
 
(uk+1 + uk ) (ik+2 − ik+1 ) − (uk+2 + uk+1 ) (ik+1 − ik )
R=
(ik+1 + ik ) (ik+2 − ik+1 ) − (ik+2 + ik+1 ) (ik+1 − ik )
  (5.14)
∆t (ik+1 + ik ) (uk+2 + uk+1 ) − (ik+2 + ik+1 ) (uk+1 + uk )
L=
2 (ik+1 + ik ) (ik+2 − ik+1 ) − (ik+2 + ik+1 ) (ik+1 − ik )

The dc-component in the fault current does not introduce errors with
this algorithm because it is accounted for in the model. A decaying dc-
R
component C1 e− L t is a part of the solution to equation 5.13.
5.3. UHS-relaying 37

Suitability for fast fault detection

Even this algorithm has the potential of being fast. The denominator, how-
ever, will become small for high sampling frequencies introducing similar
numerical difficulties as previously discussed. This algorithm will be further
investigated in chapter 7.

5.3 UHS-relaying

Description

The abbreviation UHS is a short notation for U ltra H igh S peed. As men-
tioned in section 4.4, the expression UHS-relaying is used for fault detection
within a quarter of a period of the power frequency. Common algorithms
used for UHS-relaying are so-called travelling wave algorithms. Travelling
wave algorithms measure the travelling waves generated by the initiation
of the fault. One algorithm uses measurements from both ends of a trans-
mission line to determine whether the fault is on the transmission line or
not [18].

Suitability for fast fault detection

The travelling waves travel along the transmission line with a speed close
to that of light. For distribution systems, which in general deliver power to
a rather small area, travelling wave algorithms have the potential of being
fast. Travelling wave algorithms are mostly used for transmission systems
were the damping of the transients is low (R X) and the transients
are easier to detect. In distribution systems however, the damping is often
higher and the travelling waves not easily detected. A further consideration
regarding travelling waves is that a distribution system is smaller (but can
have more feeders) than a transmission system. Distribution systems are
often (at least in Sweden) a mixture of overhead lines and cables with lots
of junctions. When a travelling wave reaches a junction, one part of it is
reflected and one part of it is transmitted but now even more attenuated
making it harder to detect.
38 Chapter 5. Algorithms

5.4 Instantaneous current algorithm

Description

A straightforward method to detect a fault is to use the instantaneous value

of the sampled current and determine that a fault has occurred when a
certain level has been exceeded. However, any error in the data acquisition
system or a power system transient could give a single current value higher
than the trigger level, thus causing a false fault detection. To make the
fault detection less sensitive to random instantaneous current values above
the trigger level, a technique similar to the fault type classifier described by
Phadke and Thorp [19] can be used. Phadke and Thorp uses a fault type
classifier algorithm to determine which phase(s) that possibly is faulted and
performs further calculations on that phase(s) to reduce the computational
burden of the microprocessor. No trip signal is based solely on the fault
type classification scheme. In this thesis the fault type classifier algorithm
will be used to detect a fault and a trip signal will be issued if the threshold
is exceeded. Furthermore, a higher sampling frequency (as compared to
Phadke and Thorp [19]) is used in this thesis.
The algorithm can be described by the following steps:

(i) Set the counter to zero

(ii) Take samples of the three phase-currents. The phase-currents are
filtered before the samples are taken. The filter is a low-pass filter
that will filter out high-frequency components such as for example
capacitor energization currents from the measured currents. The
low-pass filter will be described in section 5.4.1.
(iii) If the absolute value of the current is larger than the trigger level
increase the counter by one. Otherwise decrease the counter by
one (unless the counter is zero).
(iv) If the counter is equal to five, calculate the differential of the
current and if the differential is above a threshold, issue a trip
signal. The current differential is used to discriminate a fault
from a transformer energization as described in section 5.4.2.
(v) Repeat from (ii)

The original fault type classifier is extended with a low-pass filter and a
current differential criteria.
5.4. Instantaneous current algorithm 39

Suitability for fast fault detection

A sampling frequency high enough to allow fault detection within 1 ms will
be selected. If no delays whatsoever are added to the sampled data values,
the condition of a fault detection within 1 ms would correspond to a sampling
frequency of fs = 5 · 1e1−3 Hz ≈ 5 kHz. In practice, a higher sampling
frequency will be used, since the sampled data values will be somewhat
delayed. This algorithm will be further investigated in the case studies
(chapter 7).

5.4.1 Low-pass filter

Capacitor energization can produce high inrush currents with high deriva-
tives. The inrush current however is not of nominal power frequency but of
higher frequencies determined by the inductance and capacitance of the ac-
tual circuit. One method to discriminate a capacitor inrush current from a
fault current is to use a low-pass filter that can be applied before the instan-
taneous samples are taken in the algorithm. The anti-aliasing filter in the
data acquisition system is a low-pass filter that makes sure no frequencies
above half the sampling frequency is included in the signal to avoid erroneous
result when reconstructing the signal (half the sampling frequency is called
the Nyqvist frequency) 3 . However, since it is desired to use a high sample
rate so that the time for five consecutive samples to fall above the trigger
level is small, the anti-aliasing filter determined by the sampling frequency
will let through much of the capacitor energization transient. To filter the
capacitor energization transient the cut-off frequency of the anti-aliasing fil-
ter will be modified (lowered). It is possible to use such a filter since the
fault current given by a shunt fault is of power frequency. Furthermore,
a low-pass filter introduces a phase delay that also must be taken into ac-
count when studying the fault detection time. The phase delay grows larger
as the cut-off frequency is decreased. Once a cut-off frequency is determined
it must be verified that the phase delay is not too large so that the fault
detection takes more than 1 ms. If the phase delay is less than 1 ms it is
possible to achieve FFD within 1 ms by increasing the sampling frequency
sufficiently.
A low-pass filter is not ideal4 , i.e. frequency components lower than the
cut-off frequency will be attenuated as well as frequency components higher
3
A classical result from signal processing theory. See for example Proakis et. al. [20].
4
Ideal as in mathematically perfect.
40 Chapter 5. Algorithms

than the cut-off frequency will be passed through the filter. The actual
properties of low-pass filters has not been investigated in this thesis. In the
chapter on case studies (chapter 7), existing models of low-pass filters have
been used and the cut-off frequencies has been selected as if the low-pass
filters were ideal.

5.4.2 Current differential

Power system transients such as for example capacitor energization or trans-
former energization could give rise to current values above the trigger level
even though no fault has occurred. Sometimes it is possible to increase the
trigger level to make sure that capacitor energization is not detected as a
fault. If it is not possible to raise the trigger level, the derivative (through-
out this thesis, the derivative is always assumed to be taken with respect
to time if not otherwise stated) of the current can be used as a conditional
fault detector so that the trip signal is only given at the same time as the
derivative is above a certain level. As previously discussed, there might
be numerical difficulties to calculate the derivative since the division with
a small time step will magnify possible errors in the sampled signal. The
differential current in-between two time steps (i(t) − i(t − ∆t)) will be used
instead of the derivative ((i(t) − i(t − ∆t))/∆t) to avoid some of the nu-
merical difficulties. At least, potential errors in the measured signal are not
amplified. A study of the differential in case of a fault will be compared
to the differential when energizing a capacitor and a transformer to deter-
mine whether the differential can be used to discriminate between the three
events.

5.4.3 Extension of the method to provide directional prop-

erties
The algorithm based on instantaneous current samples described above is
not directional. If the power system contains two sources (as for example a
local generator), high currents may flow in the system for faults outside the
protection zone that shall not be detected as a fault. In [1] the application of
an algorithm (described in section 5.2) that uses three consecutive samples
to form an estimate of the impedance of the protected object is described.
Such an algorithm has directional properties but is sensitive to errors in the
sampled signals if the sampling rate is selected too high. For that particular
study, the estimated impedance was reasonable at a sampling rate of 4 kHz.
5.5. Arc detector 41

At that sampling rate it is possible to calculate an estimate of the impedance

in less than 1 ms. If the two methods are executed in parallel, a fast direc-
tional algorithm can be obtained. The extension of the algorithm (which
will be executed in parallel to the instantaneous current value method) can
be described as:
(i) Sample the current and voltage at 4 kHz.
(ii) Estimate the impedance of the protected object using equation 5.14.
(iii) If a fault is detected inside the protection zone and if the instan-
taneous current value method has produced a trip signal, forward
the trip signal as an output of the algorithm.

5.5 Arc detector

In case an arcing fault occurs inside an enclosed switchgear, the damage
can be extensive as previously discussed in chapter 2. Thus, it is of great
value to be able to disconnect such faults as quickly as possible independent
of the method used to mitigate the fault. A typical magnitude relay can
detect a fault within 20 ms after the fault inception. It is relatively easy to
reduce that time if the detection is based on the light from the arc. Such a
device can detect arcing faults within 1 ms after fault inception. However,
whenever the light is not easily available or ambiguous, fault detection based
on other principles such as those described in this chapter are required.
42
Chapter 6

Sources of errors

6.1 Reliability
Reliability of a relay protection system can be described from two aspects:
dependability and security. The dependability of a relay protection system
is the ability to detect and disconnect all faults within the protected zone.
The security of a relay protection system is the ability to reject all power
system events and transients that are not faults so that healthy parts of
the power system are not unnecessarily disconnected . The two aspects are
contradictive in the way that the more dependable the system is, the more
sensitive it is and the risk for false detection is then increased.

6.2 Power system transients

A power system normally is in a quasi-static state (quasi because it is im-
possible to achieve a perfectly static state since the load and the generation
change and because most power systems are AC-systems, i.e. the voltage
and current changes from sample to sample). Whenever an electrical path
is changed in the power system, a transient state will occur until the power
system has settled in a new quasi-static state. The change of the electrical
path can be due to for example circuit breaker operation and power sys-
tem faults. A few frequently occurring transients which might give rise to
currents similar to those of a fault are briefly discussed here.

43
44 Chapter 6. Sources of errors

6.2.1 Capacitor energization

When energizing a capacitor there is a transient in form of an inrush cur-
rent. The magnitude and frequency of the current transient depends on the
impedance of the actual circuit and the instant when the switching takes
place. If the capacitor is energized when the voltage across the switch is
zero there is a small transient; if the capacitor is energized when the volt-
age across the switch is near its peak the transient will be larger. If the
capacitor is used solely for reactive power generation, the transient will be
larger than if it is equipped with an inductor, thus forming a filter. The
current transient will cause a transient in the voltage as well. The magni-
tude of the voltage transient depends on the strength of the system and of
the magnitude of the current transient.
When a charged capacitor is located electrically close to an uncharged
capacitor the current transient (measured between the capacitors) when
charging the capacitor will be very large since only the impedance between
the capacitors limit the inrush current. That impedance is small if the
capacitors for example are connected to the same busbar with only a few
meters of cables and/or conductors in-between.
A remedy against capacitor switching transients has already been touch-
ed upon, namely to make sure that the capacitor is energized at zero voltage,
thus limiting the transient. Such solutions, known as synchronized switch-
ing, exist and have been proven in service for many years. Protection sys-
tems based on a common hardware platform can, as previously discussed
in section 4.2.4, perform more tasks than just protection. If the capacitor
switch (and as will be shown later also transformer switch) can be synchro-
nized, the transients will be mitigated thus making it less likely that the
fault detection algorithm is disturbed. Low-pass filtering also reduces the
probability that the inrush current will influence the fault detection since
the inrush current is of a high frequency character (see figure 6.2).

6.2.2 Transformer energization

The energization of an unloaded power transformer can sometimes lead to
current transients which can be large in magnitude and highly distorted.
Since the resistive load is low (the transformer is unloaded), the damping
is also low and such transients can last for many seconds before they are
attenuated. The distortion has a typical second harmonic content [19] that
can be used to discriminate between transformer inrush currents and power
6.2. Power system transients 45

system faults. However, estimating the level of the second harmonic current
takes too long time and for fast fault detection other means of discrimina-
tion must therefore be investigated. The magnitude of the inrush current
transient depends on the switching instant and on the remanence of the
transformer core [19]. If the direction of the inrush current is so that the
flux in the core coincides with the direction of the remanence flux, the core
might become saturated, hence providing large currents for its magnetiza-
tion.

A remedy against transformer inrush current has already been touched

upon, namely to make sure that the transformer is energized when the volt-
age is at its peak (another example of synchronized switching). The inrush
current which is of an inductive nature will lag the voltage by a quarter of a
period of power frequency, hence starting from a low value. The probability
that the core will be driven into saturation is lowered and the inrush currents
will be smaller. If the direction of the remanent flux can be estimated or
measured, the synchronized switching can be made even better if the switch
is controlled so that it closes its contacts in a way so that the remanence
flux decreases with the first current half-period.

When no synchronized switching is used, unfavorable switching instants

will occur from time to time so the worst case scenario is not an unusual
event. Low-pass filtering will not help to discriminate between a fault and
transformer energization as it could for capacitor energization. Since the
transformer inrush-current has a high content of the second harmonic, a low-
pass filter designed to filter out the second harmonic will contribute with a
phase error, hence introducing a delay. A second order Butterworth filter for
example will contribute with a time delay of approximately 3.5 ms when the
cut-off frequency is selected to 75 Hz1 . It will be demonstrated in the case
study that it is possible to discriminate between a transformer inrush current
and a fault by evaluating the current derivative. The current derivative is
larger for a power system fault than for a transformer energization.

1
The time delay has been estimated with EMTDC simulations by using a Butterworth
filter from a library with common power system components available within EMTDC.
46 Chapter 6. Sources of errors

6.2.3 Faults
Power system electrical faults are also of transient character. Faults that
are inside the zone of protection 2 are not considered as errors because they
are meant to be detected. Faults outside the zone of protection can disturb
the algorithm if the fault currents are close to, or larger than, the trigger
level. Even though a fault is detected it might be so that the protection
system of a neighboring system should have cleared that fault.
Another potential source of error with respect to fault currents is the
decaying dc-component (as illustrated in figure 2.2) that is appearing in
the fault current immediately after fault inception. For example the LSQ-
algorithm described in chapter 5, three consecutive samples are used to
estimate the magnitude of the fault current under the assumption that they
can be fitted to a sinusoidal waveform. The dc-component however is not
a sinusoidal waveform and will lead to an error of the estimated current.
Figure 6.1 contains the estimated current and the instantaneous current for
a fault where the fault current contains maximal dc-component. The figure
is taken from the IEC case study (chapter 7) and the sampling frequency is
4 kHz. It can be concluded by analyzing figure 6.1 that the dc-component
does not contribute with a large error and for the purpose of fault detection
within 1 ms from the fault inception the error is negligible.

6.2.4 Identification of differences

The differences between capacitor energization, transformer energization,
and a fault current is summarized in this section. The currents associated
with each of the three power system transients are plotted in figures 6.2, 6.3,
and 6.4. The load current before the transients were applied was 630 ARM S .
The current associated with the energization of a capacitor is plotted in fig-
ure 6.2. The current is higher than the load current, and contains lots of
high frequency components that depends on the inductance and capacitance
of the actual circuit. The current associated with the energization of a trans-
former is plotted in figure 6.3. The current is higher than the load current,
and contains a dc-component and a second harmonic current. Finally, a
fault current is plotted in figure 6.4. The fault current is (in this case) much
higher than the load current and is of power frequency character. For this
2
The zone of protection is a part of the power system that a protective device is designed
to protect. All faults inside the zone of protection will be detected by the relay, but not
faults outside the zone.
6.2. Power system transients 47

100

80

60

40

20
Current (kA)

0
ia

−20

−40

−60 ia:estimated

−80

−100
0.3 0.302 0.304 0.306 0.308 0.31 0.312 0.314 0.316 0.318 0.32
Time (s)

Figure 6.1. The effect of the dc-component on the estimated current (ex-
pressed in kA)

1
Current (kA)

−1

−2

−3
0.25 0.26 0.27 0.28 0.29 0.3 0.31 0.32 0.33 0.34 0.35
Time (s)

Figure 6.2. The current associated with capacitor energization

particular case it is possible to discriminate between a fault and the other

two power system transient by selecting the current level for which a fault
is detected (trigger level) sufficiently high. For other power systems the gap
between the load current and the fault current might be smaller, thus mak-
ing it difficult to find a trigger level with which it is possible to discriminate
between the three events. In sections 5.4.1, and 5.4.2 it is described an
48 Chapter 6. Sources of errors

1
Currents (kA)

−1

−2

−3
0.25 0.26 0.27 0.28 0.29 0.3 0.31 0.32 0.33 0.34 0.35
Time (s)

Figure 6.3. The current associated with transformer energization

100

80

60

40

20
Current (kA)

−20

−40

−60

−80

−100
0.25 0.26 0.27 0.28 0.29 0.3 0.31 0.32 0.33 0.34 0.35
Time (s)

Figure 6.4. The current associated with a power system fault

alternative way to discriminate between the three power system transients.

6.3 Data acquisition equipment

Before power system data can be used in the microprocessor for fault de-
tection, it must be measured in the power system and then transferred to a
format that the microprocessor can handle. The microprocessor works with
6.3. Data acquisition equipment 49

digital signals whereas the output of the measurement equipment often is an

analogue signal. The analogue signal must be converted to a digital signal
before it is further processed. Communication channels are also needed to
transfer both analogue and digital signals between transducers, protection
equipment and possibly other equipment within the substation.

6.3.1 Transducers
The word transducer is in this thesis used in the sense that it is a power
system component used to create an image of the signal under study which
can be used by the protection equipment (after processing).

Voltage transformers
Voltage transformers (sometimes referred to as potential transformers) has
been subject to standardization by for example IEC and IEEE. The accuracy
of a voltage transformer is well-defined by the standards. The IEC proposes
two accuracy classes for voltage transformers used for protective purposes3 ,
namely class 5P and 10P. A voltage transformer that is rated according to
class 5P has an error of less than 5% at rated voltage and rated burden.
A voltage transformer that is rated according to class 10P has an error of
less than 10% at rated voltage and rated burden. The voltage transformer
is rated for a given frequency (most often 50 or 60 Hz), but the standard
does not specify how a voltage transformer behaves at frequencies other
than the rated one. For transient measuring the step-response and the
bandwidth of a voltage transformer is crucial. For voltage transformers used
for fast fault detection according to principles previously mentioned, the high
frequency properties are not that important since the fault detection mainly
is based on investigating the power frequency current or voltage. According
to Lehtonen [21], the magnitude response of a magnetic voltage transformer
is flat up to at least 1000 Hz. For certain burdens the phase error might
be an issue but not for the power frequency component. In the case studies
(chapter 7) it was assumed that the magnitude response of both CTs and
VTs is equal to one for all frequencies so the possible attenuation of high
frequencies inherent to a VT is not accounted for.
Saturation of the iron core is normally not considered an issue for voltage
transformers since the voltage usually drops in case of a fault, thus increasing
3
There are other accuracy classes for voltage transformers used e.g. for instruments or
revenue metering.
50 Chapter 6. Sources of errors

the marginal to saturation. However, a single line to earth fault in a power

system with high impedance system earthing will produce a voltage rise
in the two healthy phases. Voltage transformers used in such systems are
constructed to cope with the voltage rise (at least for a limited time). It is
instead more crucial that the accuracy is maintained even for low voltages if
the voltage samples are used in fault detection algorithms such as described
in section 5.2.
Voltage transformers used in distribution systems (typically up to say
66 kV [12]) are typically pure magnetically coupled iron core transformers.
Due to cost reasons it would be disadvantageous to build such transformers
for higher voltages. At these voltages C apacitive V oltage T ransformers
(CVTs) are used instead. They consist of a capacitive voltage divider which
has an output in the kV-range. A magnetically coupled core type voltage
transformer is then connected in series with the capacitive divider, providing
an output of reasonable magnitude for use with secondary equipment such
as protection equipment. An inductor is also connected in series with the
capacitor voltage divider and the magnetic voltage transformer [22]. The
inductor is selected so that the LC-circuit thus formed is tuned to the power
frequency of the power system. As for the magnetic voltage transformer,
the behavior of the CVT at frequencies other than the power frequency is
not specified by the standards. Capacitive voltage transformers produce a
voltage transient when the voltage is suddenly changed such as when a fault
occurs. Generally, the voltage transient must be considered when applying
algorithms that use the voltage but is not further considered in this thesis.

Current transformers
Current transformers also have been subject to standardization. The accu-
racy of a current transformer is well defined by standards and IEC allows
a multitude of accuracy classes for current transformers for protective pur-
poses. A current transformer rated for protective purposes might for exam-
ple be designated 5P20, where the P stands for P rotection, the number 5
stands for the maximum allowed composite error of the current transformer
expressed in percent, and the figure 20 stands for the overcurrent factor
for which the maximum error is allowed (and for all currents lower than
that given by the overcurrent factor). The overcurrent (Ioc ) is related to
the rated current of the current transformer by multiplying the overcurrent
factor with the rated current Ioc = 20 · In , where In is the rated current of
the current transformer.
6.3. Data acquisition equipment 51

When a fault occurs in the power system the probability that the current
transformer will become saturated is quite high. Especially in power sys-
tems where fault current limiters are used where the fault currents are likely
to be large. A fault current has previously been demonstrated to contain
a decaying dc-component as well as a steady state ac-component (fig 2.2).
A current transformer can be saturated due to the dc-current and due to
the ac-current. The result of the saturation however is the same; the out-
put of the current transformer will be small or even zero depending on the
connected burden. It is not sure whether saturation is a concern when ap-
plying fast fault detection since it depends on the choice of transformer, the
burden of the connected data acquisition equipment, and the fault current.
The fault is probably detected and cleared before the current transformer
becomes saturated. The use of correctly rated current transformer how-
ever is crucial. If two power systems are connected with a sectionalizing
fault current limiter, consideration must be made to make sure the current
transformers are correctly rated to avoid possible saturation.
The step-response and the bandwidth of a current transformer is also
crucial when fast fault detection algorithms are applied. However, the meth-
ods described previously (chapter 5) in this thesis use primarily the power
frequency component of the fault current for the fault detection. When
applying fault detection based on other principles such as for example a
travelling wave, the high frequency properties of the current transformer
must be considered. High frequency properties of a current transformer is
better than the corresponding property for a voltage transformer. Lehto-
nen [21] has shown that the magnitude response for a particular current
transformer was flat up to almost 10000 Hz.

6.3.2 A/D converter

The properties of an A/D (Analogue/Digital) converter including anti-alias-
ing filter, sample (and hold) circuits, and quantization circuits also con-
tributes to the error of the measured data.
When the sampling rate has been selected the characteristics of the anti-
alias filter can be determined. Due to the Nyqvist criteria the anti-aliasing
filter must be a low-pass filter with a cut-off frequency corresponding to half
the sampling rate. For high sampling frequencies the cut-off frequency of
the anti-aliasing filter is also high, thus letting through much of the content
within the signal including high-frequency noise which can be interpreted
as an error by the fault detection algorithms. The phase error introduced
52 Chapter 6. Sources of errors

by the anti-aliasing filters however tends to zero when the cut-off frequency
tends to infinity. For a method discussed earlier 5.4, it was proposed that to
be able to discriminate a capacitor switching from a fault, a low-pass filter
could be used. The cut-off frequency of such a filter should be set much lower
than a regular anti-aliasing filter (i.e. half the sampling frequency), hence
introducing a phase error upon the signal. The phase error constitutes a
delay in the sampled signal and the delay depends on the design of the filter
and in particular on the choice of the cut-off frequency.
For cost saving reasons, one A/D converter is often used together with an
analogue multiplexer so that the samples of different signals are not truly si-
multaneously but instead sequentially sampled. The speed of the conversion
determines the maximal allowed sampling rate and the number of signals
that can be multiplexed into one A/D converter. If better performance is
required, one A/D converter per channel can be used.
The accuracy of the A/D converter is often expressed in bits. An 8-bit
A/D converter provides 28 = 256 possible digital levels for the signal. The
analogue signal will be rounded to fit into one of these 256 digital levels.
If the signal is ±10 kV (peak value), each digital level will correspond to
20000/256 ≈ 78 V. If a 12-bit A/D converter is considered, then it provides
212 = 4096 possible digital levels for the signal. The analogue signal will be
rounded to fit into one of these 4096 digital levels. If the signal is ±10 kV
(peak value), each digital level will correspond to 20000/4096 ≈ 4.88 V.
As will be shown later in the section on algorithm dependent errors, an 8-
bit A/D converter can limit the sampling frequency for a certain algorithm
though it provides quite a lot of noise in terms of round-off errors, which
the algorithm is sensitive to.

6.3.3 Processor word length

The word length of the processor also contributes to the error. However, mi-
croprocessors used in modern protection relays typically have a word length
of 16 bits. If the accuracy of the A/D converter is 8-bits or 12-bits, the word
length will not be a limiting factor. If complicated algorithms are used then
there might be a small error due to the word length but probably not of
any importance. For an algorithm that uses only the instantaneous current
values, the word length is not a limiting factor because no calculations are
performed, only a comparison with the trigger level.
6.4. Algorithm dependent errors 53

6.3.4 Processor speed

The processor speed determines how fast an algorithm can be executed. It

is essential that the algorithm can be executed in between two consecutive
samples, otherwise the sampling rate has been selected too high for that
particular algorithm. Choosing a low-level program language such as as-
sembler instead of a high-level program language such as Fortran or c++,
may increase the execution speed but is seldom necessary for standard fault
algorithms used in existing relay protection of today. Industrial PCs used
in so-called common hardware platforms are probably fast enough to allow
sampling rates much higher than those proposed in this thesis. An exam-
ple of such a system is manufactured under the designation MACH-II [23].
Microprocessors used for calculating time instants for triggering pulses in
power electronics applications such as current controlled thyristor convert-
ers should be suitable for fault detection as well. The hardware platform
used for controlling large HVDC converters (manufactured as MACH-II)
surely manages sampling frequencies at those rates proposed in this thesis
especially since the algorithm is very simple.

6.4 Algorithm dependent errors

Algorithms contribute to the error in the estimated signal (which could lead
to a false fault detection) in various ways. It is important to realize that the
actual signal is not known in advance. An algorithm that assumes that the
signal is a perfect sinusoidal signal will generate an error due to the harmonic
content and other noise in the power system signal. Another type of “error”
an algorithm can contribute to is to amplify errors in the sampled data if
not carefully designed. The word “error” is put inside quotes because the
algorithm is correct under the assumptions made. However, the algorithm
can amplify potential errors in the sampled signal. Numerically unstable
might be a better word to use than “error”.
An algorithm that is used to estimate the magnitude of a voltage or
a current signal (studied in more detail in section 5.1.3) was found to be
dependent on the sampling rate. The algorithm fits sampled data values
to a sinusoidal signal and the magnitude is given by equation 6.1. The
54 Chapter 6. Sources of errors

magnitude of the current at sampling value k is denoted by |I|.


|I| = IC2 + IS2 , where
[i1 cos θ + i0 + i−1 cos θ]
IC = , and (6.1)
1 + 2 cos2 θ
[i1 − i−1 ]
IS =
2 sin θ
The sampling rate determines the value of ∆t and if selected too high, ∆t
will be a small number, thus any error contributed by the measurement
and by the subtraction of two almost equal numbers will be amplified when
divided by a small number. This is illustrated when applying the algorithm
from equation 6.1 on measured data sampled at 20 kHz and data from the
same measurement downsampled to 1 kHz. As can be seen in figure (6.5),
the algorithm gives a much smoother estimation of the magnitude when
sampled at 1 kHz as compared to when sampled at 20 kHz.
16000 16000

14000 14000

12000 12000

10000 10000

8000 8000

6000 6000

4000 4000

2000 2000

0 0
−1 0 1 2 3 −1 0 1 2 3

Figure 6.5. Estimated magnitude when applying a sensitive algorithm to

measurements with different sampling rates. The left-hand graph contains
an estimation of a voltage downsampled to 1 kHz and the right-hand graph
contains an estimation of the same voltage sampled at 20 kHz. The timescale
of the x-axis is seconds.

The algorithm based on the discrete Fourier transform (see section 5.1.2)
is not so sensitive to the sampling rate. This algorithm uses data samples
from a whole period and the result is given in figure (6.6) for two different
sampling rates (1 kHz and 20 kHz respectively.) However, since data samples
6.4. Algorithm dependent errors 55

10000 10000

9000 9000

8000 8000

7000 7000

6000 6000

5000 5000

4000 4000

3000 3000

2000 2000

1000 1000

0 0
−1 0 1 2 3 −1 0 1 2 3

Figure 6.6. Estimated magnitude when applying a not so sensitive algo-

rithm to measurements with different sampling rates. The left-hand figure
contains a plot of the estimated magnitude when the sampling frequency was
1 kHz and the right-hand figure contains a plot of the estimated magnitude
when the sampling frequency was 20 kHz.

from a whole period is used not much is gained by raising the sampling
frequency. The estimate of the current magnitude is not available until
after about 20 ms (corresponding to one period of power frequency).
The algorithm based on three consecutive samples of voltage and current
(see section 5.2) shows to be equally sensitive to errors in the data as the
algorithm described by equation 6.1. This time the algorithm assumes that
the protected power system can be described by a differential equation e.g. a
RL-link describing a transmission line (or/and a load). If three consecutive
samples of voltage and current are available, the apparent impedance of the
transmission line can be estimated by using equation 6.2 that was given in
section 5.2. If the apparent impedance of the transmission line deviates from
the expected, it is likely that a fault is present.
 
(vk+1 + vk ) (ik+2 − ik+1 ) − (vk+2 + vk+1 ) (ik+1 − ik )
R=
(ik+1 + ik ) (ik+2 − ik+1 ) − (ik+2 + ik+1 ) (ik+1 − ik )
  (6.2)
∆t (ik+1 + ik ) (vk+2 + vk+1 ) − (ik+2 + ik+1 ) (vk+1 + vk )
L=
2 (ik+1 + ik ) (ik+2 − ik+1 ) − (ik+2 + ik+1 ) (ik+1 − ik )
Simplification of the common denominator of equation 6.2 gives that it can
be written as:  
−2 i2k+1 − ik ik+2 (6.3)
56 Chapter 6. Sources of errors

The expression given by equation 6.3 tends to zero as the sampling rate
tends to infinity. A small denominator tends to amplify errors in the sampled
signals, which was demonstrated for the LSQ-method earlier in this section.
Taking the numerical approximation of a derivative of a sampled signal
is subject to numerical difficulties. The derivative of a signal f (t) is defined
as:
f (t) − f (t − h)
f  (t) = lim (6.4)
h→0 h
When the signal f (t) is not known analytically an estimate of the derivative
can be formed by taking the differential between two samples and dividing
with the time difference between the samples.

f (t) − f (t − ∆t)
f  (t) ≈ (6.5)
∆t
The estimate will be poor if ∆t is selected too large and the signal varies
rapidly. On the other hand, the estimate can also be poor if ∆t is selected
too small, possible errors in the sampled signal will be magnified when their
differential is divided by a small number ∆t. This is illustrated in figure 6.7,
where the derivative has been calculated at a sampling frequency of 20 kHz
and after that the signal has been down sampled to 1 kHz. The derivative
calculated at 20 kHz sampling frequency is extremely noisy and difficult to
analyze, whereas the derivative calculated at 1 kHz sampling frequency is
quite smooth.
6.4. Algorithm dependent errors 57

6 6
x 10 x 10
5 5

4 4

3 3

2 2

1 1

0 0

−1 −1

−2 −2

−3 −3

−4 −4

−5 −5
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02

Figure 6.7. The derivative of a signal sampled at two different sampling

frequencies
58
Chapter 7

Case Studies

Two case studies have been performed in order to investigate the perfor-
mance of algorithms for fault detection and current limiting concepts on
a model of a real power system. The modelling and simulation has been
performed with EMTDC1 — a computer program for studying transient
electrical problems [24]. The case studies were performed using two power
systems, the first one based on standard values of rated system voltage,
rated nominal load current, and rated short-circuit current as given by the
IEC (denoted as the IEC case study), and the second case study was based
on power system data from an electrical power distribution system at SSAB
Oxelösund, Sweden (denoted as the SSAB case study). Three different types
of fault detection algorithms have been studied; one waveform algorithm (see
section 5.1), one model algorithm (see section 5.2), and one instantaneous
current algorithm (see section 5.4).

7.1 IEC case study

The international standard IEC 60694 [25] publish preferred ratings of swi-
tchgears and controlgears. For this case study the following ratings have
been selected and power system impedances have been calculated based
upon them:
• system voltage uh = 12 kV
• nominal load (phase) current in = 630 A (measured at the load ZL in
figure 7.1)
1
Electro-Magnetic Transients in DC systems

59
60 Chapter 7. Case Studies

• short-circuit phase current in case of a solid three-phase fault

ik = 40 kA
The time-constant2 of the power system (when no load is connected to the
power system) was selected to 45 ms, corresponding to a power factor of
cos ϕ = 0.0705. The power factor of the load was selected to cos ϕ = 0.8,
a typical value for common load types. Figure 7.1 contains the single line

Source Zs B1 B2 B3 ZL

Fault
Measuring point C T

Figure 7.1. IEC case study single line diagram

diagram of the power system used in the IEC case study. The source is
modelled as an infinite source, that is with no limits on active or reac-
tive power production. The source impedance is ZS = 0.0122 + j · 0.173
Ω, which corresponds to a short-circuit current of 40 kARMS . The load
impedance was ZL = 8.79 + j · 6.44 Ω, which corresponds to a load cur-
rent of 630 ARMS . Three circuit-breakers, B1 , B2 , and B3 respectively, were
used to apply three power system transients. The first circuit-breaker was
used to simulate faults imposed on the power system. Mostly three-phase
faults were studied, however a few two-phase faults were also studied to
investigate if they also could be detected using algorithms from this thesis.
Single-phase faults are only briefly considered within this thesis, since dis-
tribution systems are commonly high impedance earthed3 so that the fault
current of a single-phase fault is in the order of a few Amperes. The second
circuit-breaker was used to simulate capacitor energization, and the third
2
The time constant of a power system is a measure of the time it takes for transients
to develop or to attenuate.
3
At least in Sweden except for the 400 V distribution system that is solidly earthed.
7.1. IEC case study 61

circuit-breaker was used to simulate transformer energization. The capac-

itor and transformer data was taken from data sheets of an actual power
system, namely the same power system as that one used in the SSAB case
study below. The capacitor was rated at 4.08 MVAr at 12 kV (10.5 kV in
the SSAB case study) corresponding to a capacitance of 90.19 µF, and the
transformer was rated at 10.2 MVA at 12 kV (10.5 kV in the SSAB case
study). The capacitor and the transformer was implemented in EMTDC,
using basic building blocks from a library with common power system com-
ponents available within EMTDC. The transformer model is not just a ratio
changer but saturation is also included. Since not all parameters of the
actual transformer were known, standard values from the EMTDC power
system component library were used. Finally, a user defined component was
used to implement three algorithms that were investigated with respect to
speed of fault detection. The user defined component can be described by
the block diagram of figure 7.2.

Power system Transducers Filter A/D Algorithm

Figure 7.2. A block diagram describing the implementation of algorithms

in EMTDC

7.1.1 Common considerations

The trigger level of the fault detection, i.e. a fault is detected when the cur-
rent exceeds the trigger level, was selected to 3 times the RMS-value of the
nominal load current (a typical setting of an overcurrent relay is within 2.5
– 6 times the nominal load current according to an experienced relay pro-
tection engineer). A simple data acquisition system consisting of a low-pass
filter and a simplified A/D converter4 was added to the case study to be able
to study some of the requirements and limitations of such a system. The
measuring transformers however were implemented as ideal ratio-changers.
For two of the fault detection algorithms that were implemented (the LSQ-
method and the differential equation method), the low-pass filter was de-
signed with a cut-off frequency equal to half the sampling frequency to avoid
4
The Analogue to Digital converter model did not include quantization, i.e. no round
off errors were introduced.
62 Chapter 7. Case Studies

aliasing. For the third method (the instantaneous current method) the low-
pass filter was selected in a different manner that is described in more details
later in this chapter. The input data to the algorithms (i.e. sampled values
of voltage and current) was measured close to the source (to the right of
ZS in figure 7.1). Since the scope of this thesis is fast fault detection for
power systems where current limitation or interruption before the first peak
is required, the algorithms studied are intended to detect all faults within
the power system5 . In a power system with a traditional relay protection
system, each feeder (the source, the capacitor, the transformer, and the load
in figure 7.1) would have had its own protection equipment6 , but in a power
system where current limiting is used it is likely that most faults have to
be limited rapidly by a current diverter (see chapter 3) or a current limiter
(see chapter 3), which are typically located at the source or at a sectional-
izing position. There is however an advantage to perform measurements in
several feeders, namely that the location of a fault might be easier to find.

7.1.2 Fault detection with the LSQ-method

A LSQ-method proposed by Phadke and Thorp [19] was previously described
in section 5.1.3. The method uses three consecutive current samples and fit
them to a sinusoidal current of unknown magnitude and phase by a LSQ-
method. The result is an estimate of the magnitude (and phase) of the
current. Whenever the estimated current exceeds the predetermined trigger
level a fault is likely to have occurred and a trip-signal is issued7 .
First, three-phase faults were imposed on the power system. The fault
inception angle, i.e. the phase angle of the system voltage when the fault
occurs, was swept over a time interval corresponding to one period of fun-
damental power frequency. A total of 20 simulations were performed and
for each simulation the fault detection time was observed. The sampling
frequency was also allowed to vary and the 20 simulations were repeated for
four different sampling frequencies, namely 1, 2, 4, and 8 kHz respectively.
5
At least all faults that will give rise to fault currents that will be potentially high and
have to be limited.
6
Even though fast fault detection is implemented, the existing protection equipment in
each feeder is still important. It will protect the feeder for all faults that do not require fast
fault detection, such as for example interturn faults in the stator of a rotating machine.
7
The trip-signal is implemented in EMTDC as a binary signal which can take two
values, the binary signal is zero if no fault is detected and one if a fault is detected.
The binary signal can be used to control e.g. a circuit-breaker or a fault current limiting
apparatus, but that is not included in this case study.
7.1. IEC case study 63

The result is summarized in table 7.1 from which it can be concluded that to
achieve a fault detection time within 1 ms a sampling frequency of at least 2
kHz is required, but to allow for a margin, 4 kHz was selected as an appro-
priate sampling frequency. The estimated current magnitude at a sampling

Sampling frequency Maximum detection time

1 kHz 3 ms
2 kHz 1 ms
4 kHz 0.5 ms
8 kHz 0.38 ms
Table 7.1. The maximum fault detection time for different sampling fre-
quencies.

frequency of 4 kHz is shown in figure 7.3 together with the instantaneous

current samples. When the sampling frequency was increased to 16 kHz,
it was apparent that the estimated current contained bad estimates similar
(but not as bad) as illustrated in figure 6.5.

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8

−1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Figure 7.3. The estimated current and the instantaneous current values.
In EMTDC the source voltage is ramped up to its nominal value at the start
of the simulation and consequently the current is also ramped to its nominal
value since the load is a pure impedance.
64 Chapter 7. Case Studies

Transformer and capacitor energization were also simulated. The capac-

itor energization gave rise to large currents that was falsely interpreted as a
fault. An attempt to lower the cutoff frequency of the low-pass filter was not
successful. Even though the cutoff frequency was lowered to 300 Hz, the in-
rush current due to the capacitor energization was still interpreted as a fault.
The low-pass filter was replaced with a bandpass filter that was given such
a characteristic as to let through the power frequency component (±10%).
The simulation was repeated but this time the capacitor energization was
not interpreted as a fault. A further simulation was performed to study the
speed of the fault detection with the new filter. The fault detection time
was prolonged to approximately 1.25 ms.
A few two-phase faults were also simulated. With nothing else changed,
the fault detection time was now approximately 3.5 ms for some fault in-
ception angles. When the sampling frequency was raised to 8000 Hz, it was
possible to lower the fault detection time to 3.38 ms, and when the sam-
pling frequency was raised to 16000 Hz, the fault detection time was lowered
to 3.32 ms. No further attempt was made to raise the sampling frequency
since the LSQ-algorithm produced bad estimates (similar to figure 6.5) for
sampling frequencies above 16000 Hz.
Finally, single phase faults were simulated to verify that such faults were
not detected by the algorithm. For some fault inception angles, a fault was
detected in that phase were the single-phase fault was applied. However,
such situations can be avoided by the requirement that a fault must be
detected in at least two phases. For this particular study, the reason that
a single phase fault gave rise to fault currents large enough to be detected
as a fault is probably that the modelling of the system grounding resistance
is inaccurate. The grounding resistance has no or little effect on three and
two phase faults so the inaccurate choice of grounding resistance does not
effect the results obtained in this case study concerning three and two phase
faults.

7.1.3 Fault detection with the differential equation method

A differential equation method (described in section 5.2) was implemented
as a user defined component in EMTDC to investigate if it was a possible
algorithm to use for fast fault detection. The trigger level was treated in a
slightly different manner for the differential equation method than for the
LSQ-method since a low value of estimated resistance and inductance is
a sign of a fault in the power system, whereas a higher value of estimated
7.1. IEC case study 65

resistance and inductance is a sign of normal operating conditions. Trying to

find a trigger level that makes it possible to compare the different methods
to each other, the trigger level was selected to one third of the nominal
resistance and inductance of the protected object8 . A first simulation was
made to determine the nominal resistance and inductance of the protected
power system i.e. the source and the load of figure 7.1. It was concluded
that the nominal impedances of the protected power system were: Rnom =
8.791 Ω, and Lnom = 0.0205 H corresponding to a reactance of Xnom =
6.44 Ω. I.e. the estimated impedances are equal to the load impedance if
no fault is applied. Hence, the trigger level was set to Rnom /3 and Lnom /3.
An example of an estimated resistance is given in figure 7.4, where the
resistance and the trigger level is plotted. For that particular simulation,
the sampling frequency was 1 kHz and the fault was applied at t = 0.32
s. The fault detection in this particular phase was 5 ms. Once the trigger

10
Trigger level
Estimated resistance
9

7
Resistance (ohm)

0
0.32 0.321 0.322 0.323 0.324 0.325 0.326 0.327 0.328 0.329 0.33
Time (s)

Figure 7.4. The estimated resistance during a fault.

level was selected, a number of simulations were performed (20) and the
fault inception angle was moved 1 ms in between each simulation. The fault
8
Instead of 3 times the nominal current as for the LSQ-method and the instantaneous
current values method.
66 Chapter 7. Case Studies

detectors were placed in each of the three phases of the power system and
the sampling frequency was 1 kHz as earlier mentioned. The 20 simulations
were then repeated but now with sampling frequencies of 2, 4, and 8 kHz
respectively. The result of the 4 · 20 simulations is summarized in table 7.2,
where the largest detection time for each sampling frequency is given. Since
the point of wave of the fault initiation is a random value, fault initiation
will eventually take place in a time so that the maximum fault detection
time is obtained. The maximum detection time must always be shorter than
approximately 1 ms if fault clearing before the first prospective current peak
is the target. From the maximum fault detection times given in table 7.2 it

Sampling frequency Maximum detection time

1 kHz 3 ms
2 kHz 1 ms
4 kHz 0.75 ms
8 kHz 0.38 ms
Table 7.2. The maximum fault detection time for different sampling fre-
quencies.

can be concluded that for this particular system, fast fault detection can be
obtained with the differential equation method with a sampling frequency
of at least 2 kHz. To provide a larger margin it would be appropriate to
select the sampling frequency to at least 4 kHz.
Once the sampling frequency had been determined to 4 kHz, further
simulations were performed. The energization of a power transformer and a
capacitor were simulated for a number of simulations (20) were the point of
wave of the switching instant was varied over a period of power frequency.
The trigger level was the same as before and the purpose of the simulations
was to find out how the differential equation algorithm responded to com-
mon power system transients that are not faults. The energization of the
capacitor proved to constitute a problem since it was falsely detected as a
fault. An attempt was made to lower the cutoff frequency of the low-pass
filter since capacitor energization produces a high frequency (with respect
to the power frequency of the power system) inrush current. However, it
turned out that it was not possible to filter out all of the high frequency con-
tent of the inrush current. Even though the cutoff frequency was selected to
100 Hz, a fault was still detected when the capacitor was energized. When
the low-pass filter was replaced by a bandpass filter that was designed to
7.1. IEC case study 67

let through only the power frequency component (±10%), the capacitor en-
ergization was not detected as a fault. A bandpass filter however produces
a phase error which will constitute as a time delay so that the fault detec-
tion is delayed. At 4 kHz sampling frequency the fault detection time was
now approximately 1.8 ms (as compared to 0.75 ms when using the conven-
tional low-pass filter as determined by the Nyqvist criterion). An attempt
was made to raise the sampling frequency and it turned out that when the
sampling frequency was set to 8 kHz the time to detect a fault was still ap-
proximately 1.5 ms. An attempt to further increase the sampling frequency
to 16 kHz did not turn out well since the algorithm did not respond well at
that frequency and produced bad estimates (similar to figure 6.5) that were
not reliable. Thus, the differential equation method (at least this version)
does not seem appropriate for fast fault detection in power systems where
capacitors are present. It is likely that other transients such as for example
cable energization, asynchronous motor starts, synchronous motor starts,
and incipient cable faults will disturb the algorithm.
Transformer energization was also simulated, both with the conventional
low-pass filter (according to the Nyqvist criterion) and with the bandpass
filter that was described above. With the conventional low-pass filter faults
were falsely detected when the transformer was energized. The cutoff fre-
quency of the low-pass filter was lowered in order to find out if it was possible
to find a cutoff frequency for which the algorithm did not interpret trans-
former energizations as faults. Similar to the capacitor energization it was
not possible to find a cutoff frequency so that the transformer energization
was not interpreted as a fault. If the low-pass filter is replaced by a band-
pass filter, the algorithm will probably discriminate faults from transformer
energization, but it is not necessary to perform that simulation because the
fault detection time will be more than 1 ms and not possible to improve by
raising the sampling frequency as was discussed above.

7.1.4 Fault detection based on instantaneous current values

The trigger level was selected to 3 times the nominal load current. The fault
type classifier (without the current differential and the modified low-pass fil-
ter) that was described in section 5.4 was implemented in EMTDC and three
power system events were simulated. The three-phase fault was correctly
detected and when the sampling frequency was selected to 10 kHz, the fault
detection time was never more than 0.5 ms for 20 different fault inception
angles corresponding to a complete power frequency period. However, both
68 Chapter 7. Case Studies

the transformer energization and the capacitor energization also produced a

trip signal. Investigation of the sampled data signals revealed that it would
be quite easy to raise the trigger level to avoid false trip signals for this
particular case and still be able to detect faults within 1 ms. However, for
systems were it is not possible to raise the trigger level other solutions must
be investigated.
First the differential of the current (∆i = in − in−1 ) was investigated.
The differential between the current value when the trip signal was issued
and the current value one sample earlier was compared for the fault current,
the capacitor energization and the transformer energization respectively. It
turned out that it was possible to discriminate between a fault and a trans-
former energization by using the current differential, but not between a fault
and a capacitor energization. The current differential for a fault current was
more than seven times the current differential for a transformer inrush cur-
rent.
Second, the characteristics of the low-pass filter was modified so that
the cut-off frequency was lowered below the cut-off frequency determined
by the Nyqvist criterion. A second order Butterworth filter with a cut-off
frequency of 300 Hz was found to be suitable for this particular study. Now
it was possible to discriminate between a fault and a capacitor energization.
With the current differential criteria and the modified low-pass filter the
fault detection time was now approximately 1.0 ms and the simulated fault
was the only power system event for which a trip signal was issued. The
capacitor energizing was discriminated from a fault by low-pass filtering of
the signal and the transformer energization was discriminated from a fault
by using the current differential. The phase delay of the filter was estimated
to be approximately 0.75 ms (included in the detection time of 1 ms) by
investigating the sampled signals.

7.1.5 Results of the IEC case study

It was found that the method of using instantaneous current values (without
the modified low-pass filter and without the current differential criterion
described in section 5.4) and combine those instantaneous current samples
in an appropriate manner was suitable for fast fault detection for the IEC
case study. With a sampling frequency of 10 kHz the fault detection time
was less than 1 ms.
It was also found that the method could discriminate between power
system faults and capacitor energization but then the method would have
7.2. SSAB case study 69

to be extended so that the instantaneous current samples are subject to

low-pass filtering before processed.
A further extension of the method was to study the current differential
between two samples in order to be able to discriminate between a fault
and a transformer energization. It was shown that transformer energization
could be discriminated from a fault by using the current differential.
A method based on a differential equation that fitted sampled voltage
and current signals to a power system model in order to produce an estimate
of the apparent impedance of the protected object was shown to be not
particularly well suited for fast fault detection (at least not for this particular
power system). Three phase faults were correctly detected by the method
but capacitor and transformer energization imposed problems that could
not be solved without exceeding a fault detection time of 1 ms.
A method based on a least-square fitting algorithm to produce an esti-
mate of the peak value of the measured current was also proven to detect
faults within 1 ms with a sampling frequency of 4 kHz. However, with the
traditional filter settings given by the Nyqvist criterion, capacitor energiza-
tion also produced estimated currents that were interpreted as faults. When
substituting the low-pass filter with a bandpass filter it was possible to dis-
criminate between a three-phase fault and a capacitor energization. Further
simulations proved that the energization of a transformer also was possible
to discriminate from a fault by using the bandpass filter. Two-phase faults
were also possible to detect but then the fault detection time was more than
1 ms, even after the sampling frequency was raised to 16 kHz.

7.2 SSAB case study

SSAB9 is a Swedish steel company with business at several locations in
Sweden as well as abroad.
The steel plant at SSAB Oxelösund, Sweden, is the largest Nordic manu-
facturer of heavy plate [26]. SSAB Oxelösund has participated in the project
“New techniques for electricity supply of industrial systems including e.g.
local generators.”10 and has provided the power system data used in this
case study.
9
Svenskt Stål AktieBolag
10
The project was performed at the Competence center in electric power engineering at
KTH from 1997 until 2002.
70 Chapter 7. Case Studies

7.2.1 The electrical power system at SSAB

A reconnection of the local generator at SSAB has been investigated by
Wikström et.al. [27]. Such a reconnection would be beneficial since it would
lower the power losses in the system and since the short-circuit power at
10.5 kV would be higher, also allowing for a more efficient operation of the
power system. However, the reconnection would not be possible without
major strengthening of the 10.5 kV power system since the short-circuit
withstand capability would be exceeded. An alternative to strengthening of
the system could be to use so-called fault current limiters (see chapter 3).
If the fault current limiters shall be able to limit the current before the first
peak of the prospective fault current, fault detection within 1 ms combined
with an apparatus fault clearing time of not more than 2 ms is required.
Currently, the steel plant at SSAB is supplied by three power lines at
the voltage level 130 kV. The short-circuit power at the 130 kV terminals
at SSAB is approximately 2250 MVA. In addition SSAB has a local gener-
ator supplied with process gas. The local generator is rated 81.25 MVA at
10.5 kV, but is usually operated at 30-40 MVA11 . The single-line diagram
in figure 7.5 gives a brief summary of (a simplified part of) the electrical
power system at SSAB after the reconnection of the local generator.

Generator ZG

Source ZS B1 B2 B3 ZL

Fault 0.4 kV system

Measuring point C T

Figure 7.5. SSAB case study single line diagram

11
According to Pasi Hyvönen at SSAB
7.2. SSAB case study 71

The 130-kV grid and a 55 MVA transformer (130 kV/10.5 kV) were im-
plemented as an infinite source in series with an impedance (source and ZS
in figure 7.5). The local generator was implemented in EMTDC using a
basic building block from a library with common power system components
available within EMTDC (generator and ZG in figure 7.5). Most of the pa-
rameters of the local generator were known but for a few parameters that
were not known, standard values values from the EMTDC power system
component library were used. The load is implemented as an impedance
(ZL = 5.84 + j · 2.83 in figure 7.5). Three circuit-breakers, B1 , B2 and B3
respectively, were used to apply three power system transient. The first
circuit-breaker was used to simulate faults imposed on the power system.
Mainly three-phase faults were studied, but a few two-phase faults were
studied to investigate whether such were possible to detect within 1 ms us-
ing algorithms from this thesis. Single-phase faults were not studied since
the system is high impedance earthed so that fault currents caused by single-
phase faults are in the order of a few Amperes. The second circuit-breaker
was used to simulate capacitor energization and the third circuit-breaker was
used to simulate transformer energization. The capacitor and transformer
data was taken from data sheets of the actual power system at SSAB. The
capacitor was rated at 4.08 MVAr at 10.5 kV corresponding to a capaci-
tance of 117.8 µF. The capacitor and the transformer was implemented in
EMTDC using basic building blocks from a library with common power sys-
tem components available within EMTDC. The transformer model is not
just a ratio changer but saturation is also included. Since all parameters of
the actual transformer not were known, standard values from the EMTDC
power system component library were used. Finally, an user defined com-
ponent was used to implement three algorithms that were investigated with
respect to speed of fault detection.
To determine an appropriate trigger level, three-phase faults were first
simulated at the 0.4 kV level for different fault inception angles. The maxi-
mal current measured in the 10.5 kV system was 1.75 kAp , which is approx-
imately 1.5 times the peak current of the nominal load current. Thus it can
be concluded that the trigger level can be set to 3 times the nominal current
for this system.

7.2.2 Fault detection with the LSQ-method

A number of simulations have been performed to estimate the current magni-
tude when a fault occurs, using the equations of section 5.1.3. The sampling
72 Chapter 7. Case Studies

rate, fault inception angle and the detection level, i.e. when the magnitude
of the current exceeds a set value corresponding to a fault in the system,
have been varied throughout the calculations and their effect on the fault
detection time evaluated. When setting the detection level to 3 times the
pre-fault current in the load, it is possible to reach a detection time of just
about 3 ms with a sampling rate of 1 kHz, a detection time of just about
1.5 ms with a sampling rate of 2 kHz and a detection time of 0.75 ms with
a sampling rate of 4 kHz. Hence, for this case study fast fault detection
can be achieved based on monitoring the current magnitude. However, the
energization of a capacitor or a transformer were also (falsely) detected as
faults.

7.2.3 Fault detection with a differential equation method

To be able to distinguish a fault in the 10.5 kV system from a fault in the
135 kV system, some kind of directional criterion is required. Otherwise, a
fault in the 135 kV system would operate the current limiter or diverter even
though the 10.5 kV system can withstand such a fault. The second algo-
rithm proposed (equations from section 5.2) can provide such a directional
criterion. For faults within the 10.5 kV system the apparent resistance and
inductance will always be positive whereas a fault in the 135 kV system will
yield negative apparent resistances and inductances. Faults in the 0.4 kV
system correspond to much higher apparent resistance and inductance than
the other two fault locations. Hence, it is possible to distinguish a fault in
the 10.5 kV system from a fault in the 0.4 kV and the 135 kV systems by
measuring the voltage and current at a single location. However, also for
this algorithm, capacitor and transformer energization were falsely detected
as faults.

7.2.4 Fault detection based on instantaneous current values

The algorithm using instantaneous current values (see section 5.4) have also
been applied to the SSAB case study.
The same power system events as for the IEC case study were simulated.
For faults at the 0.4 kV level no trip signal was issued but for faults at the 130
kV level a trip signal was issued. The calculation of apparent impedance was
now added to the simulation and the trip signals of the two algorithms were
connected in series so that a trip signal was issued only if both algorithms
detected a fault. The instantaneous current method detects fault both in
7.3. Shortcomings of the simplified power systems 73

the 130 kV and in the 10.5 kV system whereas the impedance algorithm
can distinguish between faults at different voltage levels (or at least if the
fault is upstream or downstream as seen from the measuring transformers).
The simulations showed that it was possible to select trigger levels so that
fault detection was possible within 1 ms and so that faults at the 130 kV
level did not produce a trip signal. Furthermore, capacitor and transformer
energizations were simulated. None of those energization produced a false
trip.

7.2.5 Results
The application of the LSQ-method and the differential equation method
gave as result that it is possible to use them for fast fault detection — at least
in the simplified system. However, since both capacitor and transformer
energization were falsely detected as faults, the methods are severly limited
The instantaneous current method however, was able to detect a fault
within 1 ms from fault inception and to discriminate a fault from capacitor
and transformer energization.

7.3 Shortcomings of the simplified power systems

Both of the power systems used for the case studies are quite oversimplified.
No consideration is given to transient phenomena that is not related to
power system faults except for transformer and capacitor energization. It is
of course possible to model more possible sources of errors in the simplified
model of SSAB but impractical to consider all sources of errors. The IEC
case study is purely theoretical i.e. there is no measurements made that
could be used to verify that the model is implemented correctly and that
the results obtained are reliable. The SSAB case study is built upon a model
developed by Wikström [28] and to some extent verified against measured
power system data.
A simulation can never substitute actual measurements and experiments
in a real power system, only complement it. In practice there are very few
power system engineers that would be allowed to perform actual testing of
power system protection equipment in a real power system. Since faults in a
power system has great consequences it is not advisable to create real faults
in the power system just to test the fault protection equipment. Simulations
are a better way to test the power system protection equipment. It is im-
portant to realize that a simulation is (even if quite complex events can be
74 Chapter 7. Case Studies

simulated) a simplified model of the real power system. Some thoughts on

making calculations instead of performing experiments in the power system
is available in appendix B.
Chapter 8

Conclusions and future work

8.1 Conclusions
Lets for a moment consider the list of the main contributions of this thesis
as given in chapter 1. The list is repeated here for convenience.
• a structured survey on the speed of fault detection and on the benefits
of fast fault detection. In particular, “fast” fault detection is defined.
• an evaluation of possible algorithms appropriate for fast fault detec-
tion.
• requirements on equipment and algorithms used for fast fault detec-
tion.
• a case study of the application of fast fault detection in one typical
grid and in one grid at a Swedish steel plant.
Fast fault detection was defined in chapter 4 after finding out that the ex-
isting definitions of speed of protection were quite crude. When the rate of
rise of the fault current and the operating time of the fault clearing equip-
ment was taken into account, fast fault detection was defined to mean fault
detection within 1 ms measured from the inception of the fault.
The evaluation of algorithms suitable for fast fault detection led to the
selection of three algorithms that would be further investigated, namely
the LSQ-method, the differential equation method, and the instantaneous
current values method.
The requirements on equipment and algorithms pointed out items to
consider when dimensioning a system for implementation of fast fault de-
tection. Many of the requirements were related to high frequency sampling
of analogue signals. The case study showed that a sampling frequency of

75
76 Chapter 8. Conclusions and future work

10 kHz was suitable for the instantaneous current method. A sampling

frequency of 10 kHz is not a particularly high sampling frequency so that
existing microprocessor-based platforms currently used for power systems
applications should be able to serve as a platform even for fast fault detec-
tion.
The case study was performed using two power systems and by using
the simulation software EMTDC. The three algorithms mentioned above
were implemented in EMTDC and several simulations were performed. It
was concluded that the algorithm using instantaneous current values was
suitable for fast fault detection for the systems under study. The differen-
tial equation algorithm was also suitable for detecting faults but it turned
out that it was difficult to discriminate between faults and common power
system transients that are not faults but still produce higher currents than
the nominal load current. The LSQ algorithm gave somewhat better results
than the differential equation algorithm and it turned out to correctly de-
tect three phase fault while still discriminating between faults and common
power system transients. When two-phase faults were simulated the LSQ
algorithm detected them but not within 1 ms for a few fault inception an-
gles. The instantaneous current values algorithm did manage to detect two
phase faults within 1 ms, thus determined to be the most suitable for fast
fault detection out of the three algorithms under study.

8.2 Future work

Some ideas for future work are summarized in this section. The algorithm
based on instantaneous current values will be further investigated with an
extended set of power system transients. Modifications in order to make the
algorithm more robust will be proposed and evaluated. Since a fault current
contains a large power frequency component it could for example be possible
to use a PLL (Phase Locked Loop) and build criteria for fault detection on
deviations from the locked state. Methods (including PLL) that also, in
addition to detect faults, could be used to control apparatus used for fault
current limitation or solid-state circuit breakers will be investigated. Fault
current limiters built from controllable power electronics need a reference to
generate trigger signals at appropriate time instants. Some configurations of
fault current limiters are dependent on operation close to a natural current
zero crossing which the above mentioned methods (including PLL) might
8.2. Future work 77

be able to predict. A proposed continuation of the project will concentrate

on:
1. A case study on how current diverters, fault current limiters, solid-
state breakers, and conventional circuit breaker can be combined in a
power system. Different combinations will be studied and possible ad-
vantages and disadvantages of those combinations will be discussed.
An important issue to be examined is whether protection and con-
trol algorithms can be used by all apparatus in a combination, thus
providing coordination benefits.
2. Signal processing methods to enhance the understanding and behav-
ior of methods which extract the power frequency component (in-
cluding PLL) in a short time.
3. Evaluate criteria for fault based on the use of methods that extract
the power frequency component. The criteria will be evaluated with
the aid of EMTDC or SIMPOW simulations of power systems, prefer-
ably correlated with actual measurements. An important work will
be to study deviations in the methods and how sensitive apparatus
depending on current or voltage zero are to those deviations. In case
of large harmonic content in the signals (e.g. capacitor bank switch-
ing) extra voltage or current zero crossings can occur. What are
the consequences in terms of transients in the power system if the
methods falsely predicts one of the extra zero crossings instead of the
power frequency component zero crossing?
4. Alternative criteria for fault detection such as:
(a) Injection of unique frequency components into the power sys-
tem. The unique frequency should be selected as to not inter-
fere with any resonance frequencies in the system and higher
than the power frequency of the power system. In case of a
fault involving zero-sequence components it might be possible
to make a fast fault detection by analyzing zero-sequence com-
ponents. Since the injected frequency is unique, precise filters
can be used to find that frequency and since the frequency is
selected higher than the power frequency, it will rise faster.
(b) Use methods from transmission systems to detect faults by
observing the so-called travelling waves caused by the initiation
of the fault.
78
Appendix A

Analysis of voltage dips

measured at SSAB
Oxelösund

A.1 Introduction

T HE content of this appendix is based upon work carried out within

the project “New techniques for electricity supply of industrial sys-
tems including e.g. local generators” performed at the Competence center
in electric power engineering at KTH1 since 1997. The project team con-
sists of members from KTH, ABB, Birka Energi, Vattenfall, and SSAB2
Oxelösund (denoted as SSAB for the remainder of this chapter).
Power quality was early identified as an important topic within the
project. Voltage dips were identified as the main cause of electrical dis-
turbances within industries. SSAB had for many years been aware of the
problem, which even led to the building of a new overhead line (three 130
kV lines altogether) in order to strengthen the grid at SSAB. The num-
ber of voltage dips at SSAB is well documented by years of measurements
in the local 230 V grid [29]. The measurements were performed as single-
phase measurements using the electrical outlets. I.e. the voltages that were
measured was phase-voltages.
1
Kungl Tekniska Högskolan, Stockholm, Sweden
2
S venskt S tål Aktiebolag

79
80 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

Since the measurements at SSAB were performed in the 230 V grid,

i.e. single-phase measurements, the project team decided to make measure-
ments in the 10 kV grid in order to collect data from a three phase system.
The measurement equipment was installed and monitored by ABB Corpo-
rate Research from May 1999 until March 2000. During that time nearly 40
voltage dips were recorded, most of them during the summer.
In this report two methods for characterization of voltage dips are de-
scribed. Furthermore, voltage dips measured in three different voltage levels
are correlated and compared. Finally, causes and consequences of some of
the recorded voltage dips are investigated.

A.2 Methods
A.2.1 Characterization of voltage dips
This report describes two methods of voltage dip characterization and pre-
sent the result of the methods applied to measurements. The measurements
were made at SSAB in Oxelösund by ABB Corporate Research during 1999.
Altogether 112 recordings are available spanning from 1999-06-13 until 2000-
03-16. However, out of the 112 recordings only 38 have captured voltage
dips. Due to a broken tap-changer, the power system was operated at a
higher voltage than normal, thereby causing unwanted trigging of the mea-
surement equipment.
The equipment used for the measurements was:
• 3 pieces of Fluke 42 for RMS voltage measurement
• a SMR transient recorder
• and 3 pieces voltage probes (x10)
Furthermore, a modem was installed so that the measurement equipment
could be monitored and the measurements downloaded for off-line evalua-
tion. The measuring equipment was installed in a 10.5 kV switchgear, using
three existing voltage transformers3 . The measurement equipment was trig-
gered when the RMS-voltage was outside the interval 5600 V < Urms <
6500 V. Each measurement contains four channels, time and three phase-
voltages. Each channel contains 65536 samples. The sampling frequency
was f = 20 kHz, corresponding to a recorded time of 3.2768s for each mea-
surement. Thus, the resolution in time is good, but the resolution in voltage
magnitude is poorer since the resolution of the A/D converter only was 8
3
I mina anteckningar p KTH
A.2. Methods 81

bits peak-to-peak i.e. each quantization level corresponds to a voltage of 100

V4 .

Characterization by the RMS-method

A straightforward way to characterize a single-phase voltage dip is to calcu-
late the minimum value of the RMS-voltage during the dip (see figure A.1).
For each time instant, the RMS-voltage is calculated over a time window
corresponding to one period of power frequency. If the number of samples
per period is denoted by N , the RMS-voltage at sample point n is calculated
by using equation (A.1), which is valid from sample n = N :

 i=n
1

Vrms (n) = vi2 (A.1)

N
i=n−N +1

The duration of the dip is also of importance. In this report the duration is
defined as the time during which the RMS-voltage is below 90% of the pre-
fault value (see figure A.1). In the case of a three-phase system, a voltage
1.05

1 Dip duration

0.95

0.9

0.85
Value used for characterization

0.8
0 5 10 15 20 25

Figure A.1. Characterization with the RMS-method

dip can be characterized by the minimum RMS-voltage during the dip. In

case of a symmetrical voltage dip, the minimum RMS-voltage is the same
4
The quantization level depends on the voltage range and on the bit resolution as V /28 .
For these measurements V was selected so that the quantization level was 100 V.
82 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

in all phases. For unsymmetrical voltage dips, the phase with the deepest
dip is selected for the characterization.

Characterization by characteristic voltage

The material in this subsection was derived by Zhang [30] and summarized
here for convenience.
A three-phase voltage dip can be characterized by three parameters:
1. The dip-type is either A, C or D. The dip is of type A if equally
deep in all three phases. The dip is of type C if deeper in two phases
than in the third one. The dip is of type D if deeper in one phase
than in the other two.
2. The characteristic voltage is a phasor which quantifies the severity of
a three-phase voltage dip. It is defined as the subtraction of positive-
sequence voltage and negative-sequence voltage of type5 Ca .
3. The Positive and Negative factor (PN-factor) is an additional phasor
to quantify a three-phase unbalanced dip where the system’s positive-
and negative- sequence impedances are not equal. If the PN-factor is
close to unity, the dip can be quantified by the characteristic voltage
alone.
Furthermore, the duration of the voltage dip is given. I have used the same
duration as for the RMS-characterization for easy comparison of the two
methods.

A.2.2 Comparison of voltage dips recorded at different volt-

age levels
Whereas the mechanical, and thermal consequences of a short-circuit fault
often is limited to small areas of the power distribution system, the voltage
dip associated with the fault can often be noticed in large parts of surround-
ing power systems. For example, a fault in Hallsberg (120 km from SSAB
as the crow flies), would give a voltage dip down to 77% at the 130 kV
terminals at SSAB in Oxelösund [29].
Measurements at 130 kV and 0.4 kV voltage levels are available from the
same time period as the measurements at 10.5 kV. The measurements at
130 kV were performed by Vattenfall in a power system electrically close to
SSAB. The measurements at 0.4 kV were performed in the electrical socket
5
Ca denotes a dip of type C, with phase a being the deviant. The same notation is
used for Cb , Cc , Da , Db and Dc .
A.3. Results 83

in an office at SSAB. There have been no synchronization of the clocks of the

three measurement devices, but recordings differing not more than a couple
of minutes have been defined as simultaneous, at least if the magnitude of
the recorded dips are in the same range and no other events have occurred
in the same period.

A.2.3 Cause and consequences

In order to cope with voltage dips it is important to know how they are
caused and what consequences they have. SSAB has provided event data
from the plant related to the times when voltage dips have occurred. Fur-
thermore Vattenfall, the supplier of power to SSAB, has provided event
data from the 130 kV grid related to the times when the voltage dips oc-
curred. For each dip, the RMS-voltage, duration, cause, and consequences
are tabulated.

A.3 Results
A.3.1 Characterization of voltage dips
Characterization by applying the RMS-method to the recordings resulted in
table A.1 which is copied from the report [5] and also the plot in figure A.2.
Only measurements where the RMS-voltage during the dip dropped below
0.90 p.u. are presented. Furthermore, the characterization by characteristic
voltage resulted in table A.2 taken from the report [5] and also the plot in
figure A.3. Only recordings were the voltage dropped below 0.90 p.u. are
presented. The dip-type is either A, C or D. The dip is of type A if it is
equally deep in all three phases. The dip is of type C if it is deeper in two
phases than in the third one. The dip is of type D if it is deeper in one
phase than in the other two. Furthermore, Ca denotes a dip of type C, with
phase a being the deviant. The same notation is used for Cb , Cc , Da , Db
and Dc .

A.3.2 Comparison of voltage dips recorded at different volt-

age levels
Table A.3 contains dip magnitudes of voltage dips measured simultaneously
at three different voltage levels. The measurements on 130 kV and 10.5 kV
were made in three phases, whereas the measurement on 0.4 kV were made
84 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

Date and number Minimum remaining voltage (pu.) Duration (ms)

Phase A Phase B Phase C
1999-06-13-01 0.90 0.95 0.81 120
1999-06-30-02 0.85 0.86 0.86 130
1999-06-30-03 0.76 0.77 0.77 110
1999-07-07-04 0.89 0.90 0.90 190
1999-07-11-05 0.80 0.87 1.16 230
1999-07-14-06 0.84 0.85 0.84 620
1999-07-14-08 0.83 0.84 0.84 1240
1999-07-14-10 0.75 0.89 0.76 100
1999-07-14-11 0.75 0.77 0.90 90
1999-07-14-12 0.87 0.88 0.88 90
1999-07-14-13 0.85 0.94 0.86 170
1999-07-14-14 0.86 0.87 0.87 60
1999-07-14-15 0.82 0.82 0.82 350
1999-07-14-16 0.84 0.85 0.85 900
1999-07-14-17 0.93 0.66 0.85 270
1999-07-17-18 0.95 0.86 0.94 80
1999-07-20-19 0.82 0.83 0.83 100
1999-07-20-20 0.67 0.67 0.67 90
1999-07-20-21 0.65 0.83 0.93 140
1999-07-20-22 0.56 0.57 0.67 100
1999-07-22-23 0.86 0.93 0.92 100
1999-07-27-24 0.99 0.87 0.95 390
1999-07-29-26 0.96 0.81 0.93 60
1999-07-30-27 0.95 0.87 0.96 80
1999-07-31-28 0.96 0.83 0.94 70
1999-08-07-29 1.09 0.84 0.90 160
1999-08-09-30 0.68 0.52 0.91 170
1999-08-18-31 0.91 0.96 0.84 70
1999-11-22-67 0.81 0.88 0.97 110
1999-11-29-69 0.83 0.82 1.12 150
1999-11-29-70 0.83 0.83 1.13 150
1999-11-29-71 0.83 0.83 1.13 140
1999-11-29-72 0.78 0.78 0.94 160
Table A.1. Minimum remaining voltages and duration of the dips.
A.3. Results 85

Date Diptype Characteristic PN-factor Duration

and number voltage (ms)
1999-06-13-01 Dc 0.81 0.95 120
1999-06-30-02 A 0.86 0.86 130
1999-06-30-03 A 0.77 0.77 110
1999-07-07-04 A 0.89 0.89 190
1999-07-11-05 Da 0.83 0.93 230
1999-07-14-06 A 0.84 0.84 620
1999-07-14-08 A 0.83 0.83 1240
1999-07-14-10 Cb 0.71 0.86 100
1999-07-14-11 Cc 0.72 0.86 90
1999-07-14-12 A 0.87 0.87 90
1999-07-14-13 Cb 0.84 0.89 170
1999-07-14-14 A 0.87 0.86 60
1999-07-14-15 A 0.81 0.82 350
1999-07-14-16 A 0.85 0.85 900
1999-07-14-17 Db 0.69 0.88 270
1999-07-17-18 Db 0.85 0.96 80
1999-07-20-19 A 0.83 0.82 100
1999-07-20-20 A 0.67 0.67 90
1999-07-20-21 Da 0.67 0.82 140
1999-07-20-22 A 0.65 0.56 100
1999-07-22-23 A 0.94 0.86 100
1999-07-27-24 Db 0.92 0.95 390
1999-07-29-26 Db 0.81 0.96 60
1999-07-30-27 Db 0.87 0.96 80
1999-07-31-28 Db 0.83 0.96 70
1999-08-07-29 Db 0.87 0.95 160
1999-08-09-30 Cc 0.55 0.79 170
1999-08-18-31 Dc 0.84 0.92 70
1999-11-22-67 Da 0.86 0.94 110
1999-11-29-69 Cc 0.86 0.87 150
1999-11-29-69-2 Cc 0.87 0.88 140
1999-11-29-70 Cc 0.87 0.93 150
1999-11-29-71 Cc 0.88 0.83 140
1999-11-29-72 Cc 0.71 0.88 160
Table A.2. Minimum characteristic voltages and duration of the dips.
86 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

0.9

0.8

0.7
Remaining voltage (pu)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 200 400 600 800 1000 1200 1400
Duration (ms)

Figure A.2. Minimum remaining RMS voltage as function of the duration

in one phase since the measuring equipment was installed in an electrical

socket. Some of the voltage dips are deeper at 130 kV than at the lower
voltage levels, probably because synchronous machines connected at 10.5
kV strengthens the 10.5 kV grid during the dips.
Voltage dips can transform when propagating through transformers6 .
According to Zhang [30], and taking into account the transformer connec-
tions at SSAB (table (A.4)), a dip propagating from 130 kV to 10 kV will
not transform. However, a dip propagating from 10 kV to 0.4 kV will trans-
form so that a two-phase dip turns into a single-phase dip (or the other way
around). The theory is verified by calculations with EMTDC. The result
from a dip originating in the 130 kV grid looks exactly the same when mea-
sured in the 130 kV grid or in the 10 kV grid respectively (see figure (A.4)).

6
Depending on the transformer connections (Y-Y, Y-∆, ∆-Y, ∆-∆), a single-phase dip
can transform into a two-phase dip (or the other way around). The details are described
by Zhang [30] and not further considered in this thesis.
A.3. Results 87

Date and number Vattenfall SSAB SSAB

(130 kV) (10 kV) (0.4 kV)
in p.u. in p.u. in p.u.
1999-06-13-01 0.64 0.81 0.84
1999-06-30-02 0.83 0.85 0.84
1999-06-30-03 0.78 0.76 0.75
1999-07-07-04 - 0.89 -
1999-07-11-05 - 0.80 -
1999-07-14-06 0.82 0.84 0.84
1999-07-14-07 0.89 0.91 -
1999-07-14-08 0.82 0.83 0.84
1999-07-14-09 - 0.91 0.92
1999-07-14-10 0.71 0.75 0.72
1999-07-14-11 0.70 0.75 0.83
1999-07-14-12 0.85 0.87 0.87
1999-07-14-13 0.82 0.85 0.83
1999-07-14-14 0.84 0.86 0.86
1999-07-14-15 0.79 0.82 0.82
1999-07-14-16 0.82 0.84 0.84
1999-07-14-17 0.56 0.66 -
1999-07-17-18 - 0.86 -
1999-07-20-19 0.80 0.82 0.82
1999-07-20-20 0.79 0.67 0.67
1999-07-20-21 0.44 0.65 0.74
1999-07-20-22 0.16 0.56 0.60
1999-07-22-23 0.87 0.86 0.88
1999-07-27-24 - 0.87 -
1999-07-29-26 0.78 0.81 -
1999-07-30-27 0.83 0.87 -
1999-07-31-28 0.78 0.83 -
1999-08-07-29 - 0.84 -
1999-08-09-30 - 0.52 0.78
1999-08-18-31 0.79 0.84 -
1999-11-22-67 0.28 0.81 -
1999-11-29-69 - 0.82 -
1999-11-29-69-2 - 0.83 -
1999-11-29-70 - 0.83 -
1999-11-29-71 - 0.83 -
1999-11-29-72 - 0.78 -
Table A.3. Dip magnitude at different voltage levels
88 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

0.9

0.8

0.7
Remaining voltage (pu)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 200 400 600 800 1000 1200 1400
Duration (ms)

Figure A.3. Minimum characteristic voltage as function of the duration

Transformer High-voltage Low-voltage

130/10 Star, solidly earthed Star, Impedance earthed
10/0.4 (kv/kv) Delta Star, solidly earthed
Table A.4. Transformer connections

A.3.3 Causes and consequences

According to [29] non of the recorded dips gave a disturbance of the blast
furnace whereas almost all dips during production time gave a disturbance
of the plate making plant. The probable causes of the voltage dips are
tabulated in table (A.5).

A.4 Discussion
The large amount of voltage dips recorded and their consequences, indicates
that they are a major problem for the industry. However, one must bear in
A.4. Discussion 89

Date and number Probable Cause

1999-06-13-01 Lightning — 130 kV
1999-06-30-02 Lightning — 130 kV
1999-06-30-03 Lightning — 130 kV
1999-07-07-04 Unknown
1999-07-11-05 Unknown
1999-07-14-06 Lightning — 130 kV
1999-07-14-07 Unknown
1999-07-14-08 Lightning — 130 kV
1999-07-14-09 Lightning — 130 kV
1999-07-14-10 Lightning — 130 kV
1999-07-14-11 Lightning — 130 kV
1999-07-14-12 Lightning — 40 kV
1999-07-14-13 Lightning — 130 kV
1999-07-14-14 Lightning — 130 kV
1999-07-14-15 Vattenfall, 130 kV, Other
1999-07-14-16 Lightning — 130 kV
1999-07-14-17 Lightning — 130 kV
1999-07-17-18 SVK, 400 KV, Unknown
1999-07-20-19 Lightning — 130 kV
1999-07-20-20 Unknown
1999-07-20-21 Lightning — 130 kV
1999-07-20-22 Lightning — 130 kV
1999-07-22-23 SVK, 220 kV, Lightning
1999-07-27-24 Unknown
1999-07-29-26 Vattenfall, 130 kV, Other nature
1999-07-30-27 SVK, 400 kV, Unknown
1999-07-31-28 Vattenfall, 130 kV, Other nature
1999-08-07-29 Unknown
1999-08-09-30 Lightning — 130 kV
1999-08-18-31 SVK, 400 kV, Unknown
1999-11-22-67 Lightning — 130 kV
1999-11-29-69 Unknown
1999-11-29-69-2 Unknown
1999-11-29-70 Unknown
1999-11-29-71 Unknown
1999-11-29-72 Unknown
Table A.5. Cause and consequences
90 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

Voltages

130 kV
120

80

40
kV

-40

-80

-120

10 kV
10

2
kV

-2

-6

-10
0.48 0.5 0.52 0.54 0.56 0.58 0.6 0.62
Time (s)

Figure A.4. Propagation of voltage dips

mind that many of the voltage dips occurred close in time to each other. In
those cases when a voltage dip caused the plant to trip, the plant would most
probably not have started again when the following dip occurred. Hence,
the only voltage dip having consequences was the one which tripped the
A.5. Future work 91

plant. Secondly, many of the voltage dips occurred during July, a month
when the plant usually is shut down for maintenance, thereby causing no
consequences.
The duration of the majority of the measured voltage dips was less than
200 ms, but a typical high-voltage circuit-breaker interrupts a short-circuit
current within 40 ms [31]. Hence, by examining the relay protection it might
be possible to reduce the duration of most voltage dips to less than 100 ms.
Further possible solutions to the problem with voltage dips are:
• Strengthen the grid by increasing the short-circuit power.
• Disconnect the plant from the grid during bad weather, relying on a
local generator.
• Installation of a Dynamic V oltage Regulator (DVR).
• Improve the lightning protection of the overhead lines by for example
top lines or surge arrestors.
The first solution mentioned will lead to an increase in the short-circuit
power. In an existing grid, however, it might not be possible to add that
power without rebuilding or substituting large parts of the equipment in
the grid. The problem is that an increased short-circuit power leads to
an increased short-circuit current in case of an electrical fault in the grid.
Two solutions that allows the benefit of increasing the short-circuit power
without the disadvantage of increasing the short-circuit current is described
below:
• The installation of a fault current limiter at one or more properly se-
lected places in the power system will in case of a short-circuit limit
the fault current to a level that the components of the system will
manage. There are fault current limiters available to the market to-
day [32] and research towards new types of fault current limiters are
performed at several universities [27] [33].
• A fault diverter system [34] placed at every source feeding the power
system will divert the fault current contribution from each source to
ground, thereby preventing superimposed fault currents in the power
system. Dynamic simulations of a fault diverter system has been
performed and the results seem promising [28].

A.5 Future work

Future possible connections at SSAB, which will give a more flexible grid will
be studied. Then again, the problem with a stronger grid will be apparent.
92 Appendix A. Analysis of voltage dips measured at SSAB Oxelösund

The methods discussed to control the short-circuit currents need further

investigation and the focus will be on fault detection and fault handling. The
methods described in this report will not be used for fault detection due to
a delay time of at least one period of power frequency when performing the
RMS-calculation. Conventional protection relays using Fourier algorithms
are also out of the question due to the delay time introduced by the Fourier-
calculation. Methods that will be studied are:
• detection of the wavefront caused by the fault using for example wave-
let analysis
• current differential algorithms based on sample-to-sample comparison
• algorithms based on current level and current derivative
Dynamic simulations, preferably performed with EMTDC, will be used to:
• test algorithms for fault detection and fault handling.
• find how fast fault current limitation that actually is needed.
Appendix B

Experiments performed as
computer based calculations

B.1 Introduction
Before an experiment is started it is important to sort out what will be
observed. Furthermore, the experiment shall be planned so that the obser-
vation really can take place [35]. A theoretical model of the experiment (if
such exists) can be of great help when planning the experiment.

B.1.1 Basics
My experiments are carried out with the aid of a calculation program im-
plemented in a computer. My research focuses on fast and automatic fault
disconnection in electrical distribution systems and is a project carried out
at the department of electrical power systems at KTH. The projects per-
formed at the Electric Power Systems group at the department of Electrical
Engineering span issues from modelling of power system equipment to mod-
elling of electrical markets. For most of the projects performed in the group
calculation programs are frequently used . What is calculated and in how
fine details varies between projects. Common types of calculations are:
• load flow calculations, which are carried out to calculate power flows
within electrical power systems.
• system planning calculations, which are carried out to estimate the
operation of an electrical power system in different time scales such
as yearly, monthly and weekly.

93
94 Appendix B. Experiments performed as computer based calculations

• short-circuit calculations, which are used as a basis for determination

of protection settings and testing of protection systems.
The above mentioned types of calculations have in common that they are
possible to perform by hand given that the power system under study is
small. The number of components and state of operations grows fast as
the power system grows, which means that calculation by hand becomes
impractical (or even impossible). The level of detail differs between the three
types of calculations mentioned above. In a system planning calculation it is
possible to lump several loads together into one equivalent component. But
for short-circuit calculations to study protection setting for different loads
in a power system more detailed models are needed.
The remainder of this article will focus on short-circuit calculations per-
formed in purpose to study methods for fast and automatic disconnection
of faults.
One alternative method to calculate short-circuit currents by hand has
already been mentioned for small power systems. A third alternative is
to build a scaled model of the power system under study in an electrical
laboratorium. Experiments performed with a scaled model of the power
system would probably give a good understanding of the power system since
physical components are used.
The benefit of using computer calculations is that it is relatively easy to
start using the program and get results.

B.1.2 Theory dependence

Methods for short-circuit calculations is either based on the assumption
that the quantities under study always is periodic so that RMS-values1 of
voltages, currents and circuit impedances can be used, or based on the
solving of the differential equations describing the system. The first method
is used when the so-called steady state2 values of voltages and currents are of
interest. The other method is used when the instantaneous values of voltages
and currents are of interest. In case of an electrical fault in a power system
1
RMS is an abbreviation of Root M ean S quare, which is a mean value for periodic sig-
nals. For sinusoidal signals the RMS-value is calculated as the peak value of the sinusoidal
signal divided by the square root of two.
2
A power system is said to be in a steady state when the RMS-values of all signals
and parameters do not change. Even though the power system is in steady state the
instantaneous signals do change since power systems mainly contains sinusoidally varying
signals.
B.2. Methods 95

the steady state is disturbed and during a transient state transferred to a

new steady state corresponding to the current circuit parameters. Transients
are possible to study with calculation programs based on the second method
mentioned above.
When choosing a calculation program it is not only needed to consider
whether the program calculates RMS-values or instantaneous values but
also to consider how the program solves the differential equations, and if the
program uses a fixed time step or a time step that is allowed to change during
the course of the calculation. A fixed time step is determined before the
calculation starts and remains the same throughout the calculation whereas
a time step that is allowed to change is made small when the solution changes
rapidly and made larger when the solution changes slowly.
In the market lots of calculation programs are available for short-circuit
studies. One of the programs we use at my department is EMTDC [24].
The modelling of the power system is made through a graphical interface
and the differential equations are hidden in the calculation program. The
differential equations are solved by the trapezoidal method described in a
paper by Dommel [36] in the early seventies.
To evaluate whether the calculations are performed correctly it is advis-
able to start with a small power system which easily can be calculated by
hand for comparison. Furthermore so-called benchmark systems have been
published which can be used to calibrate the models [37]. Further verification
is possible by comparing the calculated results with actual measurements in
a real power system, modelled in the calculation program.

B.2 Methods
The next step in designing the experiment is to determine which parameters
that will be included in the experiment and how they will be treated during
the experiment. It must also be possible to make observations (measure-
ments) in the calculation program. Measurements in a real power system
always influence the observation, but in a calculation program there is usu-
ally ideal observations.

B.2.1 Parameters and variables

For all types of calculations of power systems it can be difficult to separate
between parameters and variables. For one calculation the system voltage
of a node might be considered as constant and consequently treated as a
96 Appendix B. Experiments performed as computer based calculations

parameter. For another type of calculation the system voltage of the same
node might depend on the load and consequently vary throughout the cal-
culation. When performing short-circuit calculations one often studies what
happens when a switching or a sudden change of circuit topology is intro-
duced in the power system. Then the system voltage of a node is considered
as a parameter until the time of the switching and thereafter treated as a
variable. Thus, the system voltage is treated as both a parameter and a
variable for that particular calculation.
The independent variable in short-circuit calculations are most often the
time. Currents, voltages and other variables are calculated as functions of
the time. Sometimes the time step can be thought of as a parameter (fixed
time step) or as a variable (variable time step). Stochastic variables can be
used to simulate switchings or changes in the power system such as location
of the fault, the type of the fault, and the duration of the fault.

B.2.2 Observations and measurements

To make observations in a calculation program is easy. As a rule there are no

limitations to which quantities one is allowed to measure. Furthermore there
are seldom limitations on how many measurements that are allowed. In a
real power system there are limitations with respect to space and cost reasons
on how many measurement transducers that can be used. Furthermore, the
more measurements one makes the more difficult it is to analyze them. After
a computer calculation the calculated values can be saved and analyzed
afterwards but in a real power system the analyze must take place in real
time.
Most calculation programs offers the possibility of using ideal voltage
and current transducers. An ideal transducer does not contribute any error
but are simplified and hence not possible to use for all types of calculations.
One can choose to make more detailed models of measuring transducers and
calibrate those models with actual measurements but often it is enough to
know the range of the fault current. When calculating short-circuit currents
with the purpose of studying fast and automatic disconnection of faults the
measurement error must be accounted for and implemented in the calcula-
tion. The actual measuring fault (in case of sinusoidal signals) is composed
by a magnitude error and a phase error. Phase errors introduce a time delay,
which can be of importance when estimating how fast fault disconnection is
possible.
B.3. Results 97

B.2.3 Effects of observation

All measurements in a real power system affects the accuracy of the ob-
servation, but in a calculation program the observation sometimes can be
ideal. Compared to a real power system observational effects depending on
the calculation program is also an issue to consider. Since the calculation
program solves the differential equations at discrete time steps a small error
is introduced in each time step and adds up in the final solution [38]. Such
faults can be minimized by careful selection of solving method and by suit-
able choice of fault tolerance. Depending on how the discrete time step is
treated by the calculation program there might be an additional error de-
pending on the properties of the system under study. For each power system
one or more time constants can be calculated. These time constants reflects
the response of the power system to a disturbance. If the discrete time step
is larger that the time constants of the power system it is not possible to
observe all phenomenons in the calculation. Calculation programs using a
variable time step automatically adapts to the time constants of the power
system but for calculation programs using a fixed time step the choice of
time step is a science in itself. If time constants are possible to calculate or
estimate they can be used as a guide. If one is unsure if the correct time
step is chosen one can repeat the calculation with half the time step and
then compare the results. If the results differ significantly the time step is
not chosen correctly but should be even smaller. On the other hand, if a
too small time step is chosen, the calculations will take unreasonably long
time depending on the size of the power system and how the parameters are
varied.

B.3 Results
B.3.1 Interpretation problems
When comparing the results from a calculation with measurements made in
the real power system one should thoroughly study which parameters that
have been used and how they have been selected. Loads can for example
be modelled as constant power loads, constant current loads, and constant
impedance loads, or a combination of those3 [39]. Loads can vary with time.
3
A constant power load draws a specified power from the source independent of the
voltage, a constant current load draws a specified current from the source independent of
the voltage, and a constant impedance load draws a current depending on the voltage.
98 Appendix B. Experiments performed as computer based calculations

In most cases one never sees the real power system but instead works with
drawings and product data sheets. A visit to the real power system can
explain differences in the results because it often happens that the reality
and the drawings do not agree. The actual operating conditions can not be
seen in a drawing. Does one really compare the calculation with the real
power system under the same conditions?

When comparing results obtained with different calculation programs

one should study how the calculation programs work, how they solve the
differential equations, how fault tolerance is treated, and how the time step
is treated. If both calculation programs use a fixed time step it is important
to check that the time step is equal for both programs. Modelling of compo-
nents can differ between different calculation programs and also depends on
the level of detail wanted. A more realistic model might give more accurate
results but in turn take longer to calculate. Calculation programs using a
graphical interface often hides the differential equations that describe the
component and it can be difficult to compare between program for example
how a power transformer is modelled. Further differences might be explained
by examining how switchings are treated by the programs. In calculation
programs based on a fixed time step one can imagine a switching event tak-
ing place between two discrete time steps. Then you can have differences
in the results depending on how that is treated by the programs. The cal-
culation program EMTDC allows switching in between discrete time steps
because the program uses an advanced algorithm for interpolation between
time steps so that the switching is correctly treated [40]. One more reason
for differences between calculation programs can be that different parame-
ters have been used which can be easy to do depending on if the parameters
are given as for example inductances or impedances, or in Ohm or per unit4 .
Overhead lines or cables can be modelled in a number of ways as for exam-
ple pi-link, t-link or as a distributed model with or without transposing
the phases [40]. Some models make use of the length of the line or cable
and its geometrical shape to estimate its parameters, but in other models
impedance values might be given.

4
A short definition of per unit is given by IEEE standards as “The reference unit, estab-
lished as a calculating convenience, for expressing all power system electrical parameters
on a common reference base.” [7]
B.4. Discussion 99

B.4 Discussion
B.4.1 Summary
In a calculation program one always makes a model of reality. The model
can be made more or less detailed depending on the purpose of the study
and which level of detail that one is interested in. One must find a balance
between the level of detail, the clearness, and the amount of time needed to
perform the calculation. In my opinion one of the reasons that calculation
programs are widely spread is that one can quite fast get results that one
can work with. Another reason is that the simulated results is in good agree-
ment with actual measurements. The models are quite accurate. Thanks
to the rapid improvement of computer performance a power system can be
calculated quite fast. The calculation programs available today often have
the possibility to perform multiple calculations varying the parameters in
between. The calculation program EMTDC gives the possibility to vary a
parameter linearly, nonlinearly, or stochastically within a given intervall.
Another reason for using a calculation program is that it is easy to share
models between users and relatively easy to find another user having similar
problems or possibilities.
An alternative to computer calculations is to build a scaled model of the
real system, equip the model with a control system and a data acquisition
system. Such a model would probably give a better understanding of the
power system because it contains physical components whose behavior can
be studied and explained. The drawbacks of building a scaled model is that
it takes time to get started and obtain results, and that it would cost more
than performing computer calculations.
To perform calculations based only on data obtained from drawings and
product data sheets is not recommended. A visit to the real power system
is strongly recommended. Partly to verify that the right data is used as pre-
viously discussed but also to obtain the understanding that a power system
is not merely lines on a drawing but has a rather large physical extension.
Since results are easily obtained with a calculation program there is a
risk to put to much trust in the results. One should have a theoretical
knowledge of electricity and power systems to understand the result and to
eliminate obviously false results.
100
List of References

[1] M. Öhrström and L. Söder. Fast fault detection for power distribution
systems. In Proceedings of the IASTED conference Power and Energy
Systems. ACTA Press, 2002.
[2] J. Watson and M. Öhrström. Current transformers, couplers & coils:
A century of overcurrent measurement for power system protection. In
37th International Universities Power Engineering Conference (UPEC
2002), volume 1, pages 277–281, 2002.
[3] M. Öhrström, L. Söder, and H. Breder. Fast fault detection for peak
current limitation based on few samples. Accepted for publication at
CIRED2003, 2003.
[4] M. Öhrström and L. Söder. A comparision of two methods used for
voltage dip characterization. Accepted for publication at IEEE Power
Tech, Bologna, 2003.
[5] M. Öhrström. Characterization of voltage dips recorded at a swedish
industrial plant during 1999. Technical Report A-EES-0011, Electric
Power Systems, Royal Institute of Technology, Stockholm, Sweden,
2000.
[6] M. Öhrström. Analysis of voltage dips. Technical report, KTH, 2001.
[7] IEEE Standard Dictionary of Electrical and Electronic Terms. The
Institute of Electrical and Electronics Engineers, Inc, fourth edition,
1988.
[8] IEC 60050(448): International Electrotechnical Vocabulary, Chapter
448: Power system protection. Bureau Central de la Commission Elec-
trotechnique International, 3, rue de Varembé Genéve, Suisse, second
edition, 1995.

101
102 List of References

[9] S. Rusck. ASEAs Handbok nr 1: Kortslutningsstrmmar, deras berkning

och verkningar. 1968.

[10] A. R. Van C. Warrington. Protective Relays: Their Theory and Prac-

tice, volume Vol. I. Chapman & Hall Ltd., London and John Wiley &
Sons, New York, second edition, 1968.

[11] L. Gauffin. Design of personally safe 1 – 72.5 kv switchgear rooms,

physical characteristics and design aspects. Technical report, ABB Dis-
tribution, 721 71 Västerås, 1988.

[12] C. Christopoulos and A. Wright. Electrical Power System Protection.

Kluwer Academic Publishers, Dordrecht/Boston/London, 2nd edition,
1999.

[13] Schneider Electric. ARC TERMINATORtm arc extinguishing

systems for use in medium voltage switchgear. Available
http://www.squared.com/us/products/switchgr.nsf, December 2002.

[14] A.T. Johns and S.K. Salman. Digital Protection For Power Systems.
Peter Peregrinus Ltd., on behalf of the Institution of Electrical Engi-
neers, London, United Kingdom, 1995.

[15] M.J. Demler. High-speed analog-to-digital conversion. Academic Press,

Inc., San Diego, California 92101, USA, 1991.

[16] T. Ueda, M. Morita, H. Arita, Y. Kida, Y. Kurosawa, and T. Yam-

agiwa. Solid-state current limiter for power distribution system. IEEE
Transactions on Power Delivery, 1993.

[17] A. Poeltl and K. Frohlich. Two new methods for very fast fault type
detection by means of parameter fitting and artificial neural networks.
IEEE Transactions on Power Delivery, 1999.

[18] M. Chamia and S. Liberman. Ultra high speed relay for ehv/uhv trans-
mission lines-development, design and application. IEEE Transactions
on Power Apparatus and Systems, PAS-97, 1978.

[19] A.G. Phadke and J.S. Thorp. Computer Relaying for Power Systems.
Research Studies Press Ltd., Somerset, England, 1988.
List of References 103

[20] J.G. Proakis and D.G. Manolakis. Digital Signal Processing, Princi-
ples, Algorithms, and applications. Prentice-Hall International, Inc.,
Simon & Schuster/A Viacom Company, Upper Saddle River, New Jer-
sey 07458, third edition, 1996.

[21] M. Lehtonen and T. Hakola. Neutral Earthing and Power System Pro-
tection. ABB Transmit Oy, Relays and Network Control, P.O.Box 699,
FIN-65101 Vaasa, Finland, 1996.

[22] J.M. Gers and J.H. Holmes. Protection of electricity distribution net-
works. Power and Energy series. The institution of Electrical Engineers
(IEE), 1998.

[23] L. Carlsson. “classical” hvdc: still continuing to evolve. Modern Power

Systems, June 2002.

[24] Available http://www.pscad.com/main/pscad/pscad v3/index.html,

November 9 2002. Copyright 2000-2001 Manitoba HVDC Research
Centre Inc.

[25] Common specifications for high-voltage switchgear and controlgear stan-

dards. International Electrotechnical Commission (IEC), 2002.

[26] Available http://www.oxelosund.ssab.se/encompany/index.htm. Copy-

right 1999 SSAB Oxelsund AB.

[27] A. Wikström et al. Power quality improvement by use of a solid state

current limiting device. In Proceedings. Ninth International Conference
on Harmonics and Quality of Power, volume 3, pages 921–925, 2000.

[28] A. Wikström. Initial results from simulations with current limiter and
arc eliminator. Technical Report A-EES-0001, Electric Power Systems,
Royal Institute of Technology, Stockholm, Sweden, 2000.

[29] P. Hyvönen. Voltage dips recorded at the electrical power system for
ssab oxelösund ab. Technical report, SSAB Oxelösund, 1999.

[30] L. Zhang. Three-phase Unbalance of Voltage Dips. PhD thesis,

Chalmers University of Technology, 1999.

[31] ABB Switchgear AB, SE 771 80 Ludvika. HPL SF6 Circuit-breakers,

2000. SESWG/B 2314en.
104 List of References

[32] ABB Calor Emag Mittelspannung GmbH, Oberhausener Strasse 33,

D-40472 Ratingen. Is -limiter, 1999. DECMS 2216 99 E.

[33] C.S. Chang et al. Voltage quality enhancement with power electronics
base devices. In IEEE Power Engineering Society Winter Meeting,
2000.

[34] H. Breder and L. Liljestrand. Fault diverter system. Technical report,

ABB, 2000.

[35] S-O. Hansson. Konsten att vara vetenskaplig. Filosofienheten, KTH,

2002.

[36] H.W. Dommel and W.S. Meyer. Computation of electromagnetic tran-

sients. Proceedings of the IEEE, 62:983, 1974.

[37] http://www.ee.washington.edu/research/pstca/. Power systems test

case archive - uwee. URL-link.

[38] G. Strang. Introduction to applied mathematics, pages 562-586. Welles-

ley Cambridge Press, Wellesley, Massachusetts, 1986.

[39] G. Andersson. Dynamic Phenomena in Electric Power Systems. Com-

pendium, Electrical Power Systems, KTH, 2001.

[40] Manitoba HVDC Research Centre Inc. PSCAD/EMTDC Manual.