University of Southampton Research Repository

ePrints Soton

Copyright © and Moral Rights for this thesis are retained by the author and/or other
copyright owners. A copy can be downloaded for personal non-commercial
research or study, without prior permission or charge. This thesis cannot be
reproduced or quoted extensively from without first obtaining permission in writing
from the copyright holder/s. The content must not be changed in any way or sold
commercially in any format or medium without the formal permission of the
copyright holders.

When referring to this work, full bibliographic details including the author, title,
awarding institution and date of the thesis must be given e.g.

AUTHOR (year of submission) "Full thesis title", University of Southampton, name

of the University School or Department, PhD Thesis, pagination

http://eprints.soton.ac.uk
 UNIVERSITY OF SOUTHAMPTON

Corrosive Sulphur in Large

Transformers: Impact, Quantification
and Detection

by
Pedro Santo Amaro

A thesis submitted for the

degree of Doctor of Philosophy

in the
Faculty of Physical Sciences and Engineering
Department of Electronics and Computer Science

February 2015
 UNIVERSITY OF SOUTHAMPTON

ABSTRACT

FACULTY OF PHYSICAL SCIENCES AND ENGINEERING

DEPARTMENT OF ELECTRONICS AND COMPUTER SCIENCE

Doctor of Philosophy

by Pedro Santo Amaro

Several power apparatus use some type of oil insulation system. Transmission level
transformers, one of the most expensive components in power networks, use mineral oil
as both electrical insulation and cooling. Due to the financial risk posed, if this equip-
ment fails unexpectedly, several quality and condition monitoring techniques are used to
test the insulation system. Nevertheless, in 1990 approximately 50 % of the mineral oils
used in transformers was contaminated with corrosive sulphur species [1], which passed
undetected by the standard corrosion tests at the time, resulting in transformers failures
worldwide. CIGRE estimates that between 2000 and 2009 there were approximately 100
related failures, not considering small distribution transformers too small for any inves-
tigation to be carried out.
The work presented investigates how Cux S faults develop, examines possible techniques
to monitor the corrosion process and the failure scenarios caused by sulphur corrosion.
A sample production method capable of generating 90 mm square Kraft paper samples
with homogeneous Cux S deposition, with a less than 1 % variation, and a reproducibility
of ±3.5% is detailed. Empirical data demonstrate how the generation of Cux S deposits
to DBDS concentration, temperature, ageing atmosphere, paper-copper interfaces and
location of deposits. Based on the empirical data obtained, a copper sulphide deposition
model using multiple parallel Cux S formation mechanisms was developed.
The development of condition monitoring techniques and laboratory-scale validation ex-
periments have been completed successfully, with the use of both field and laboratory
samples. It has been shown that x-ray fluorescence (XRF) is technically capable of
quantifying very accurately sulphur and copper in oil, with a limit of quantization at
4.05 and 1.95 ppm respectively. In addition, XRF measurements demonstrated that
through historical sulphur trends, sulphur depletion rates can be calculated to extrap-
olate if there is an onset sulphur corrosion. Frequency dielectric spectroscopic analysis
of Cux S contaminated Kraft paper demonstrated empirically that Cux S deposits have
resonance frequency, at approximately 1 mHz, and that the magnitude of the resonance
relates to the quantity of Cux S. Furthermore, laboratory samples accurately replicate
field samples containing Cux S deposits and other impurities in terms of dielectric data
analysis.
The effect of sulphur corrosion and possible failure scenarios are investigated. The elec-
trical breakdown experiments demonstrate that the Cux S deposits reduced the electrical
breakdown strength of each Kraft paper layer by 79.5 %. By applying DP life expectancy
models to empirical data it is demonstrated that the corrosion process reduces 25 % of
the transformer life expectancy at normal working conditions. The excess dielectric
loss, generated from a Cux S contaminated insulation paper system, has increased by a
factor of approximately 27.5. The thermal conductivity also increased 75%. It is demon-
strated the generation of a hotspot is very likely, which then generates a snowball effect
of generating further Cux S deposits and further increasing the local temperature.

ii
Contents

Nomenclature v

Acknowledgements x

1 Introduction 1
1.1 Electrical Power System in Britain . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Project Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 TOPICS Research Project . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Power Transformer Design, Insulation Systems and Corrosive Sulphur 9

2.1 Transformers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Winding Conductors . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Dielectrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.3 Transformer Core and Tank . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Corrosive Sulphur . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Sulphur Compounds . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.2 Sulphur Corrosion Process . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.3 Detection and Condition Monitoring . . . . . . . . . . . . . . . . . 32
2.2.4 Mitigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 The Copper Sulphide Deposition Process 38

3.1 Analytical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.1 Energy-Dispersive X-Ray Spectroscopy . . . . . . . . . . . . . . . . 39
3.1.2 X-ray Photoelectron Spectroscopy . . . . . . . . . . . . . . . . . . 40
3.1.3 Thermogavimetric Analysis . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Development of Sample Production Method . . . . . . . . . . . . . . . . . 41
3.3 Modifications to the Sample Production Method . . . . . . . . . . . . . . 57
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 Monitoring the Corrosion Process 62

4.1 Analytical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.1.1 X-ray Fluorescence . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1.2 Frequency Spectroscopy Analysis . . . . . . . . . . . . . . . . . . . 63
4.2 Tracking Copper and Sulphur . . . . . . . . . . . . . . . . . . . . . . . . . 68

iii
 4.2.1 Experiment Overview & Sample Preparation . . . . . . . . . . . . 69
4.2.2 Calibration Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3 Frequency Dielectric Spectroscopy Analysis of Copper Sulphide Deposits
in Paper Insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.3.1 Sample Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3.2 Experimental Arrangement . . . . . . . . . . . . . . . . . . . . . . 84
4.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

5 The Impact of Copper Sulphide Deposition on the Thermal Properties

and Electrical Strength of Kraft Paper 92
5.1 Analytical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.1.1 Breakdown Life Data Analysis . . . . . . . . . . . . . . . . . . . . 93
5.1.2 Degree of Polymerisation . . . . . . . . . . . . . . . . . . . . . . . 95
5.2 Paper Insulation Electrical Breakdown Failure . . . . . . . . . . . . . . . . 96
5.2.1 Experimental Arrangement . . . . . . . . . . . . . . . . . . . . . . 96
5.2.2 Experimental Method . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.2.3 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3 Chemical Degradation of Paper . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.1 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4 Thermal Runaway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4.1 Experimental Arrangement and Sample Preparation . . . . . . . . 109
5.4.2 Dielectric Loss Measurements . . . . . . . . . . . . . . . . . . . . . 112
5.4.3 Thermal conductivity . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

6 Conclusions 117
6.1 Copper Sulphide Deposition Process . . . . . . . . . . . . . . . . . . . . . 118
6.2 Monitoring the Corrosion Process . . . . . . . . . . . . . . . . . . . . . . . 118
6.3 The impact of Copper Sulphide Deposition on the thermal properties and
electrical strength of Kraft Paper . . . . . . . . . . . . . . . . . . . . . . . 119
6.4 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.4.1 Corrosion Process . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.4.2 Monitoring Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 120

References 122

iv
Nomenclature

A Amps [A]
Aaf Arrhenius pre-exponential factor
Acs Cross section in the direction of the hot spot [m2 ]
As Area of sample [m2 ]
AC Alternating Current
Ag Silver
AST M American Society for Testing and Materials
B Magnetic flux density [T]
Bef f Flux density of the applied rms voltage [T]
Bmax Maximum magnetic flux density [T]
BiBZ Bibenzyl
BSI British Standards Institution
BT A Benzotriazole
◦C
Degrees Celsius
C Capacitance [F]
C Coulombs
Coil Concentration of DBDS-Cu complex in oil [ppm]
C0 Saturated DBDS concentration in oil [ppm]
Minimum concentration of DBDS-Cu complex in oil to start being absorbed by
CA
paper [ppm]
Minimum concentration of DBDS in oil to generate Cux S deposits in the paper
CG
[ppm]
CL Capacitance of the contaminated tape layer [F]
Cs Copper concentration at the source [ppm]
CCD Covered Conductor Deposition
CEB Central Electricity Board
CEGB Central Electricity Generating Board
International Council on Large Electric Systems (in French: Conseil International
CIGRE
des Grands Rseaux lectriques)
CO Carbon Monoxide
CO2 Carbon Dioxide

v
CT Current Transformer
Cu Copper
Cux S Copper Sulphide
cdf Cumulative density function
D Magnetic electric displacement field, also known as the electric flux density [A s m−2 ]
d Displacement vector pointing from the negative charge to the positive charge [m]
Minimum concentration of DBDS-Cu complex absorbed by paper to start generating
DG
Cux S deposits [ppm]
Df Diffusion coefficient of copper through insulation paper layers
Dp Concentration of the absorbed DBDS-Cu complex in insulation paper [ppm]
dc Thickness of the transformer core [m]
ds Sample thickness [m]
DBDS Dibenzyl Disulphide
DBS Dibenzyl Sulphide
DBSO Dibenzyl Sulphoxide
DC Direct Current
DDF Dielectric Dissipation Factor
DGA Dissolved Gas Analysis
DHDS Dihexyl Disulphide
DHS Dihexyl Sulphide
DN O Distribution Network Operating
DP Degree of Polymerization
DPt DP at time t
DP0 Initial DP
E Electric field stress [V m−1 ]
El Local electric field [V m−1 ]
Ea Activation energy [J mol−1 ]
EDX Energy-Dispersive X-ray
emf Electromotive force [V]
erf c Complementary error function
ESEM Environmental Scanning Electron Microscope
eV Electron Volt (1.602176565 × 1019 J)
F Farads
Ff Median rank approximation of the cumulative probability of failure
Fw2p Weibull two-parameter cdf
F DS Frequency Dielectric Spectroscopy
f Frequency [Hz]
fw2p Weibull two-parameter pdf
g Grams
GC − AES Gas Chromatography and Atomic Emission Spectroscopy

vi
h Hours
H Magnetic field strength [A m−1 ]
H2 Dihydrogen
HV High Voltage [V]
Hz Hertz
I Current [A]
IEC International Electrotechnical Commission
IF T Inter-Facial Tension
J Current density [A m−1 ]
J Joules
Jc Conduction current density [A m−1 ]
Jce Effective electric conduction current density [A m−1 ]
Jde Effective displacement electric current density [A m−1 ]
Ji Impressed (source) current density [A m−1 ]
Jt Total current density [A m−1 ]
K Kelvin
k Thermal conductivity [W m K−1 ]
kD Equilibrium constant between dissolved and absorbed DBDS-Cu complexes
k10 Initial rate of polymer chains breaking up
k2 Rate of change of k10
ke Material constant
kh Material constant
kQ Equilibrium constant between absorbed DBDS-CU complexes and deposited Cux S
KOH Potassium Hydroxide
l Litres
Ln Layer number
LoD Limit of Detection [ppm]
LoQ Limit of Quantisation [ppm]
LV Low Voltage [V]
m Meters
meanblank Instrumental average response of a blank samples [ppm]
mol Moles
N Newtons
n Number of molecules per unit volume contributing to the polarisation
ns Steinmetz exponent
O2 Dioxygen
OfGEM Office for Gas and Electricity Markets
P Power [W ]
p Dipole moment [C m]
Pd Dielectric power loss [W]

vii
Pmacro Macroscopic polarisation
PD Partial Discharge
pdf Probability density function
pH Power of Hydrogen
ppm Parts per million
PTFE Polytetrafluoroethylene
Q Concentration of Cux S deposits in insulation paper
q Electric charge [C]
R Resistance [Ω]
Rg Universal gas constant 8.3144621 [J mol−1 K−1 ]
Rw (t) Weibull reliability function
RSD Relative Standard Deviation
RSH Mercaptans
rms Root mean square
S Siemens
S Seconds
SEM Scanning Electron Microscope
SF6 Sulphur Hexafluoride
SO System Operator
T Tesla
T Incubation period of DBDS in oil to generate Cux S deposits [s]
Th Temperature rise away from a hotspot [◦ C]
t Time [s]
Heating time required for DBDS-Cu complex to start to be absorbed by the insula-
tA
tion paper
tG Heating time for Cux S to start depositing in the paper insulation
tan δ Loss tangent, also known as the dissipation factor
tan δe Effective loss tangent
TO Transmission Operator
T OP ICS Transformer Oil Passivation and Impact of Corrosive Sulphur
Empirical constant for the incubation period of DBDS in oil to generate Cux S
U
deposits
V Volts
VL Potential difference across a Cux S contaminated paper layer [V]
W Watts
Wb Webber
We Eddy current losses [W]
Wh Hysteresis losses [W]
wt Weight [g]
XM A X-ray Micro Analyser

viii
XP S X-ray Photoelectron Spectroscopy
XRF X-Ray Fluorescence
Z Complex impedance
α Polarisability
β Shape parameter
β1 Shape parameter with respect to E
∆d Thickness of the paper layers [m]
δ Loss angle
0
εr Real relative permittivity
00
εr Imaginary relative permittivity
ε0 Electric permittivity of free space
emf Generated electromotive force [V]
εr Relative permittivity
η Scale parameter [s]
ηE Scale parameter with respect to the E [kVmm−1 ]
λ Failure rate function, also known as the hazard function
ρT Thermal resistivity of oil-paper insulation[K m−1 W−1 ]
ρT C Electrical resistivity of the transformer core [Ω m−1 ]
σe Effective conductivity [S m−1 ]
σs Static conductivity [S m−1 ]
ΦB Magnetic flux through the circuit [Wb]
χ Electric susceptibility
Ω Ohms
ω Angular frequency (2πf )[rad s−1 ]
ω0 Natural frequency [Hz]

ix
 Acknowledgements

I would like to express my sincere gratitude to the whole team involved in this research
project for their valuable contributions throughout this project; Professor Paul L.
Lewin and Dr. James A. Pilgrim, my PhD supervisors, Dr. Gordon Wilson and Paul
Jarman, National Grid project supervisors, Marco Facciotti and Dr. Alex Holt, my
project research colleagues, Professor Richard C. D. Brown, Chemistry advisor, and
finally Dr. Richard J. Heywood and Dr. Hongzhi Ding, Doble Engineering consultants.

I am also grateful to all the members of the Tony Davies High Voltage Laboratory,
who have made me welcome to the group and providing insightful knowledge.
Particular thanks to Brian Rodgers, Mike Smith, Neil Palmer, Ernest Asiedu and
Charlie P. Reed for their technical support.

I would like to thank my parents, Luis Santo Amaro and Ana Gonçalves, my brother,
Miguel Santo Amaro, and Catarina Almeida Santos, for their constant support and
encouragement throughout my academic career.

Most of all, I would like to express my greatest gratitude to the friends made during my
time at Southampton; especially Dr. Richard Chippendale and Dr. Mathew Praeger
for our high-level virtual meetings and Friday gatherings, Dr. Ziyi Huang for our
stimulating discussions, Dr. Ian L Hosier from his endless optimism, and Rui Huang,
Stelios Christou, Dr. Celia Yeung, Dr. Jack A. Hunter, Karim Budabuss Aleksandar
Denkov, etc. you have all made this last six years a truly amazing experience.

x
Chapter 1

Introduction

Electrical power systems were established in the late 19th century due to the invention of
the incandescent light bulb, by Joseph Swan (1878) in Britain and later on improved by
Thomas Edison. DC power systems were predominant during the 1870’s and 1880’s, as
it only consisted of ‘small’ systems. Direct current (DC) systems powered factories and
small city centre areas, but did not reach 95% of residents. In Britain, regulation came
into place to foster this new industry, with The Electric Lighting Act 1882 and 1888
allowing the setting up of supply systems by persons, companies or local authorities.
As the electrical power system grew it became clear that to supply whole cities with a
electric power system would require centralised generation of power. In 1891 the first
alternating current (AC) power system is introduced in Britain with London Electric
Supply Corporation, opening Deptford Power Station. AC and DC systems coexisted
up until the middle of the 20th century.
At the start of the 20th century, there were over 600 separate electricity companies in the
UK, in order to increase availability and efficiency there was a move towards centralising
the electrical power industry and creating regional monopolies. The Electricity (Supply)
Act 1919 created electricity authorities, where each new authority would acquire all the
power stations, transmission and distribution system in their area. This allowed the
new electricity authorities to build larger electrical power stations and interconnect a
number of systems, enabling electrical authorities to transfer power around the country
in response to supply and demand. Further legislation was enacted to help this new
centralised electrical power network. The Electricity Supply Act 1925 created the Cen-
tral Electricity Board (CEB), which established the National Grid transmission system
linking the biggest and most efficient power stations across the country and supplying
power to the existing district operators.
In the 1930’s, a government scheme began to help people connect their houses to the
electrical grid, at the time domestic use of electricity was limited to the wealthy. By
1936, 12000 premises had been connected to the electrical grid. By the end of the Second
World War many key industries were nationalised, including the electrical industry on

1
1st April 1948, which created the Central Electricity Generating Board (CEGB). After
the war, demand grew faster than supply, which saw the CEGB’s mission as providing
an adequate and secure supply (“keeping the lights on”), with little concern for costs.
Eventually efficiency and economies of scale took over, which started the development
of control rooms operating the transmission and distribution networks that continuously
predict demand and supply, and direct power stations to increase or decrease production.
To make sure that production was always achieved at the lowest possible cost a ‘merit
order’ system was created, where all generators in power stations were ranked based on
the unit cost of electricity. During this state ownership period there were considerable
investments made to the network and research centres. It was also during this tenure
that the 400kV, 275kV and 132kV network systems were built.
In 1989 with the new Electricity Act, the industry was privatised and became what is
known as today’s modern electrical power industry. At the start of the second millen-
nium the Utilities Act 2000 introduced a new regulation to facilitate the integration of
renewable energy sources into the power sources industry mix. As the government and
society become more environmentally aware a new path of low carbon emission is being
pursued in the power industry. The integration of renewable energy sources is making
power networks systems revert back to a distributed power generation system.

1.1 Electrical Power System in Britain

The electricity network industry in Britain, as explained previously, is constituted of

several regional natural monopolies. There are currently three transmission operators
(TOs) Scottish Power Transmission Limited for southern Scotland, Scottish Hydro Elec-
tric Transmission plc for northern Scotland and the Scottish islands groups and National
Grid Electricity Transmission plc for England and Wales. The TOs develop, operate
and maintain a high voltage system within their own distinct onshore transmission ar-
eas. There is also a System Operator (SO), National Grid Electricity Transmission plc,
which is responsible for ensuring a stable and secure operation of the whole transmission
system.
The electricity distribution networks are responsible for transporting electricity from
the high voltage transmission grid to industrial, commercial and domestic users. In
Britain there are 14 licensed distribution network operators (DNOs) owned by six dif-
ferent groups, as shown in Figure 1.1 and each is responsible for a regional distribution
services area.

2
 Figure 1.1: Britain Electricity Distribution Operator Regions and Owners [2]

Three-phase distribution and transmission networks in the UK range from 400 V to 132
kV and from 275 kV to 400 kV, respectively [3]. Figure 1.2 demonstrates the centralised
electrical system, from generation to customer in the UK, noting that 132 kV may also
be a transmission voltage in Scotland. The ac electrical network consists of either three-
phases or one-phase configurations depending on where it is in the network. In the ac
electrical network, three-phase systems are the most common system used, as it has
been demonstrated to be the most efficient way of transporting power [4]. Single-phase
electrical systems are normally used at the extremities of the network; generation and
at the customer’s end.

Figure 1.2: England Transmission and Distribution Electricity Systems, modified

from [3]

The Office for Gas and Electricity Markets (OfGEM) monitors the electrical industry
natural monopolies; ensuring customers are getting the best service and price possible.
It also establishes investment policies, which for the electrical transmission industry is

3
RIIO-T1. OfGEM also regulates the electrical transmission and distribution industry
by penalising outages and persistent faults. The higher the voltage level a fault occurs,
the greater impact it normally has in the network, due to a domino effect. Therefore,
transmission operators normally have redundancy built into their network to allow the
possibility of one or more failures without needing to put customers off-line. When
there is an electrical breakdown the utility operator needs to assess the continuity of
supply, environmental impact, material damage and all issues relating to health and
safety, each aspect carries its own financial and social impacts. The ideal electrical
power utility operator would reliably access, manage the risk of their assets and replace
the asset just before breakdown. In addition, the operator would need to manage the
operating conditions of the assets to maximise their lifetime, for example by using on-line
continuous condition monitoring techniques.

1.2 Project Motivation

Refined crude oil is the most common type of insulating fluid system used in the elec-
trical power industry. Transformers and reactors are the equipment that most often
uses oil as an electrical insulation. Insulation structures are critical to any electrical
apparatus; in the case of the transformer it is the insulation that defines the lifespan of
the asset [5, 6, 7], hence it is vital to monitor any degradation of these systems. Several
condition monitoring techniques have been developed in order to detect certain changes
that might cause a fault or show an underlying problem. Through proper condition mon-
itoring techniques, it is possible to extend the equipment life with minimum workload
for the network operator, by providing an opportunity to correct any problems before
they cause any real damage.
Corrosive sulphur related degradation is one problem currently affecting oil insulated
power equipment, such as power transformers. Corrosive sulphur compounds can orig-
inate from the crude oil and from additives to insulating oil. Sulphur corrosion may
degrade the electrical insulation properties of solid insulation media by the formation of
semi-conductive deposits. Equipment issues related to corrosive sulphur have appeared
before, but it was not considered a major technical issue in the power industry. This
was due to the relatively few failures being attributed to corrosive sulphur, meaning
that they were considered as singular events, which lead to a lack of study and detailed
analysis of this issue [8]. The problem was initially dealt in the early 1990’s by making
some improvements to the oil purification techniques and adding a control mechanism
such as corrosive tests as part of quality assurance [9, 10]. Around the 21st century, this
problem resurfaced [6, 11, 12]. It is suspected that higher operating temperatures and
the addition of foreign chemicals to the insulation oil was the cause [13]. These new
types of corrosive sulphur compounds passed undetected by the old standardised tests,
which led to major revision of these tests. After the standard tests were optimised it

4
is now possible to detect the new corrosive compounds [11]. Nevertheless, as corrosive
related faults take a long time to develop to an extent where it shows any detectable
change, by the time corrosive sulphur was detected it had already spread on a global
scale [14]. The magnitude of this problem is much greater today due to the large number
of contaminated and failed units [8]. The lack of research into this topic has left the
industry with very little knowledge about corrosive sulphur related faults. It is known
that corrosive sulphur contamination of the oil insulation system leads to the deposition
of copper sulphide (Cux S) particulates in kraft paper, but the process is still unknown.
The mechanism leading to the breakdown of the equipment remains unclear. No condi-
tion monitoring technique designed for detecting sulphur corrosion is currently available,
making it impossible to detect or evaluate the present damage to the equipment with
non destructive techniques [11].
Transformers are the primary assets in any ac network in the world, and one of the most
expensive, with high voltage (HV) super grid transformers costing millions of pounds.
Therefore having a fleet of transformers contaminated with corrosive sulphur oils repre-
sents a high financial and operational risk. Preventive measures have been recommended
by research bodies, i.e. CIGRE, to avoid proliferation of corrosive oil and to prevent
failure of infected equipment. The main preventive action to stop failures has been to
passivate copper by the addition of triazole based passivators to the oil. Due to the
circumstances at the time, the protective properties of passivators were considered to
outweigh the unknown side effects of adding them [11].
Currently there are two parallel problems; one is the corrosive sulphur compounds and
the other is the unknowns of the passivators; more precisely, what its side effects are
and for how long it can protect the copper from corrosive compounds. If these problems
are left unsolved, it adds a level of uncertainty to the electrical network that would be
unsustainable from the point of view of preserving continuity of supply, which is required
by the electrical distribution and transmission regulator [10].

1.3 TOPICS Research Project

The work in this thesis has been funded by National Grid plc (UK) through the project
Transformer Oil Passivation and Impact of Corrosive Sulphur (TOPICS). The scope of
this project was to review the current state-of-the-art and common practices in oper-
ating electrical power transformers throughout the United Kingdom with a particular
attention towards corrosion phenomena and its effect on dielectric performances. This
was a three-year research project, where the research team consisted of two postgradu-
ate researchers, one from Electrical Engineering (myself) and another from Chemistry
(Marco Facciotti). In addition, there was a Research Fellow (Dr. Alex F. Holt) inves-
tigating silver corrosion. The academic staff involved with the project were Prof Paul

5
L. Lewin, Dr James A. Pilgrim and Prof Richard C. D. Brown. Paul Jarman and Dr.
Gordon Wilson from National Grid together with Dr. Richard J. Heywood from Doble
Engineering provided industrial input for the TOPICS project through quarterly meet-
ings, where results and new ideas were discussed. This multidisciplinary research project
had four main research areas:

1. The Corrosion and Cux S Deposition Process

2. The Passivation of Power Transformers

3. Corrosion of Silver Tap Changers

4. Condition Monitor Techniques for Corrosion

The work presented in this thesis covers the first and last research areas. Nevertheless,
some of the work carried out was undertaken collectively. Marco Facciotti collected the
Scanning Electron Microscope (SEM) / Environmental Scanning Electron Microscope
(ESEM) -Energy-Dispersive X-ray (EDX) spectra. Dr N Sano (NEXUS) at Newcastle
University collected X-ray Photoelectron Spectroscopy (XPS) spectra as part of Project
IDB610(C:S:P:C: - Corrosive Sulphur Paper Contamination). The X-Ray Fluorescence
(XRF) sample preparation and XRF spectra collection was undertaken by these re-
searchers. In addition, field samples were collected with the help of Dr. Hongzhi Ding
from Doble Engineering during the decommissioning of certain National Grid transform-
ers.

1.4 Thesis Contributions

Initial empirical tests were undertaken to validate the relationships between Cux S depo-
sition with temperature, oxygen and DBDS concentration. Further empirical tests were
then undertaken to increment the state-of-the-art by demonstrating that the increase
of either DBDS concentration or temperature has similar effects in increasing the rate
of corrosion, and copper-paper interfaces are ‘preferred’ areas for corrosion and depo-
sition. Furthermore, a procedure for homogeneous Cux S deposition was developed for
large kraft paper samples. From data collected with the XPS, EDX and the empirical
studies a Cux S deposition process model, composing of two parallel mechanisms, has
been proposed.
This thesis proposes a new method for tracking sulphur species based on XRF spec-
troscopy of oil samples. Parallel to the sulphur tracking method, it is also demonstrated
that the XRF test method is capable of accurately measuring the dissolved copper in
oil. Empirical studies also demonstrate that copper dissolution does not increase due to
the presence of DBDS in the mineral oil.
The dielectric spectroscopy of Cux S contaminated paper results demonstrate that it is
6
possible to detect Cux S deposits and it is also possible to quantify how degraded the
dielectric properties are with respect to the Cux S deposits.
Breakdown strength measurements demonstrate that as the Cux S contaminates the kraft
paper layer and its insulation properties decrease substantially. Data analyses of the de-
gree of polymerisation concludes that the impact solely caused by corrosive oil is not
significant enough to be considered the cause of the DP related failures. The power
loss generated from a Cux S contaminated insulation is very significant, which combined
with only a relatively small increase in thermal conductivity makes the generation of a
hotspot very likely, which may lead to a thermal runaway. The following is a publication
list of the work under the TOPICS project.
Learned Journal Articles:

1. Amaro, P S, Facciotti, M, Holt, A F, Brown, R C D, Lewin, P L, Pilgrim, J A,

Wilson, G and Jarman, P N X-Ray Fluorescence as a Condition Monitoring Tool
for Copper and Corrosive Sulphur Species in Insulating Oil. IEEE Transactions
on Dielectrics and Electrical Insulation, In Press.

2. Facciotti, M, Amaro, P S, Holt, A F, Brown, R C D, Lewin, P L, Pilgrim, J

A, Wilson, G and Jarman, P N (2014) Contact-based corrosion mechanism lead-
ing to Cux S deposition on insulating paper used in oil-immersed electrical power
equipment. Corrosion Science, 84, 172-179.

International Conferences Papers:

3. Holt, A F, Facciotti, M, Amaro, P S, Pilgrim, J A, Lewin, P L, Brown, R C

D, Wilson, G and Jarman, P N (2013) An Initial Study Into Silver Corrosion
in Transformers Following Oil Reclamation. In, INSUCON 2013, International
Electrical Insulation Conference, Ottawa, CA, 02 - 05 Jun 2013. 4pp, 469-472.

4. Amaro, P S, Holt, A F, Facciotti, M, Pilgrim, J A, Lewin, P L, Brown, R C D,

Wilson, G and Jarman, P N (2013) Impact of Corrosive Sulfur in Transformer
Insulation Paper. In, IEEE 2013 Electrical Insulation Conference, Ottawa, CA,
02 - 05 Jun 2013. 5pp, 459-463.

5. Holt, A F, Facciotti, M, Amaro, P S, Brown, R C D, Lewin, P L, Pilgrim, J A,

Wilson, G and Jarman, P (2013) Silver Corrosion in Transformers. In, IEEE 2013
Conference on Electrical Insulation and Dielectric Phenomena, Shenzhen, CN, 20
- 23 Oct 2013. , 448-451.

6. Amaro, P S, Facciotti, M, Holt, A F, Pilgrim, J A, Lewin, P L, Brown, R C D,

Wilson, G and Jarman, P (2013) Tracking Copper Sulfide Formation in Corrosive
Transformer Oil. In, IEEE 2013 Conference on Electrical Insulation and Dielectric
Phenomena, Shenzhen, CN, 20 - 23 Oct 2013. 4pp, 144-147.

7
 7. Facciotti, M, Holt, A F, Amaro, P S, Brown, R C D, Lewin, P L, Wilson, G
and Jarman, P N (2013) XPS study on direct detection of passivator Irgamet 39
on copper surfaces aged in insulating mineral oil. In, IEEE 2013 Conference on
Electrical Insulation and Dielectric Phenomena, Shenzhen, CN, 20 - 23 Oct 2013.
4pp, 1097-1100.

1.5 Thesis Structure

Following the Introduction, the second chapter introduces the transformer in the context
of the electrical power system, followed by a detailed review of the winding conductors,
dielectric materials and transformer core & tank. Finally, it introduces corrosive sulphur
in the context of oil insulated power apparatus, followed by a detailed review of the cur-
rent understanding of the sulphur corrosion process, detection and condition monitoring
techniques, and mitigation procedures.
The third chapter presents a sample production method developed in this project. While
developing this method advances in understanding have been made in terms of the rela-
tionship between environmental variables and the effects on the corrosion process. This
chapter concludes with a new theory to explain the Cux S deposition process.
Having developed a sample production method, Chapter 4 investigates multiple tech-
niques used to analyse samples and detect changes in the properties of a transformer’s
dielectric materials. First, an x-ray spectroscopic technique is investigated as corrosion
monitoring tool, by analysis of the transformer oil. Frequency dielectric spectroscopy
(FDS) is also investigated as a condition monitoring tool for transformer insulation pa-
per.
The fifth chapter details the failure events of corrosive sulphur related faults in power
apparatus. A range of techniques from dielectric spectroscopy to degree of polymerisa-
tion is used to investigate potential triggers leading to failure. Two distinct hypotheses
have been considered and tested and one found to be verifiable. Finally, the last chapter
summarises the main achievements of this thesis and offers new questions based on the
insights gained from this work. Future work is also outlined.

1.6 Summary

This chapter has presented a broad overview of the power system development in Britain
and the electrical utility today and its regulator. In addition, this chapter has explained
the project motivation and briefly outlined the research challenges. Finally, this thesis
was situated in the context of the broader interdepartmental research project, TOPICS,
and the thesis contributions and structure were outlined.

8
Chapter 2

Power Transformer Design,

Insulation Systems and Corrosive
Sulphur

Power transformers are one of the oil insulated electrical equipment critically affected
by corrosive sulphur. Hence, before introducing the current knowledge about corrosive
sulphur the transformer design and its insulation systems are presented. The trans-
former design discusses some of the current winding structures and materials used, core
structure and design, tank structure and general transformer losses. In addition, the
dielectric transformer systems are presented individually, starting by the insulation fluid
and then the cellulose based system such as paper and pressboard.
Corrosive Sulphur is discussed by first presenting the sulphur compounds, their origin
and respective corrosive level. This is followed by research into the current particular
sulphur compound affecting the electrical industry. Some of the currently proposed cor-
rosion methods are also demonstrated. Finally the mitigation techniques being applied
by the industry are discussed.

2.1 Transformers

Transformers work by generating a magnetic flux, using an electrical conductor with a

changing current, and then directing it to pass through another conductor, by providing
a low reluctance path, where it induces an electromotive force (emf) [V]. The principle
is known as electro-magnetic induction and was first described by Michael Faraday in
1832 [15], it is known as Faraday’s Law of Induction, i.e.

dΦB
|emf | = | | (2.1)
dt

9
Where emf is the emf generated [V] and ΦB is the magnetic flux through the circuit
[Wb]. Transformers are fundamental in ac power systems; they are able to change volt-
age and current levels of the network, where in the ideal case the change in voltage is
inversely proportional to the change in current so that power is kept constant. This
voltage level change is achieved by manipulating the ratio of turns between the primary
and secondary windings, as first discovered by Nicholas Callan [16]. Although the gen-
eral principals were developed it took another 50 years to build and patent the first
transformer, in 1886, by Karoly Zipenowski, Miska Deri and Otto Blathy from Granz in
Hungary [17].
The modern electrical power systems consist of several electrical networks intercon-
nected together by transformers. Transformers differ depending on the systems to be
interconnected. The biggest differentiation in an electrical power system is between the
transmission and distribution networks, which are defined by the voltage level. There
are several different types of transformers, which correspond to design variations in the
coils, core, insulating medium and the tank, but the normal lifetime of a transformer
operating at rated conditions is between 30 to 40 years [6, 18]. Variations of each com-
ponent within a transformer are in order to increase efficiency, or improve resilience, or
due to space restriction or with the aim of reducing maintenance costs. The following
sub-sections will discuss each component in greater depth.

2.1.1 Winding Conductors

Silver is the second most electrically conductive material and is used in some small parts
of the transformer [6], the high cost of silver prevents it being more widely used. Copper
is the next most electrically conductive material and transformers coils are almost ex-
clusively made out of high-conductivity copper, which is simply highly purified copper
manufactured by using electrolysis to remove impurities. The addition of silver, from
0.1 % to 0.3 % concentration in copper, can be seen to increase the softening tempera-
ture significantly, with negligible effect on the electrical conductivity. As the addition of
silver tends to harden the metal alloy it is therefore mainly applied to large units, where
the higher mechanical strength outweighs the decreased malleability of the windings [6].
As stated earlier, transformer design is influenced by the desire to increase efficiency.
Load losses are generated by joule heating in the windings, which varies as the square
of the load current I [A],
P = I 2R (2.2)

where P is power [W] and R is resistance [Ω]. The size of the windings is the determin-
ing factor of the transformer size. The cross-sectional area of the conductor is inversely
proportional to load loss, which radiates the heat generated and therefore influences the
cooling systems. The size of the windings are interconnected with the size of the trans-
former meaning that reducing the size of the conductors results in savings in materials
and core losses. Although, it also increase the thermal loss of the conductor, which in
10
turn increases the size of oil ducts needed to dissipate the excess heat generated [6].
Therefore, a transformer design engineer needs to decide what to prioritise depending
on the customers’ demands. The temperature of transformers varies depending on the
applied load. Transformer ratings are used to calculate the maximum loading to a trans-
former without decreasing its normal lifetime expectancy. For modern transformers the
rated load can produce a maximum average temperature rise of 65 ◦ C , with respect to
ambient temperature. For older transformers the maximum average temperature rise
is 55 ◦ C , the discrepancy is due to advances in insulating materials [18]. In the UK,
the maximum and minimum temperature since 1995 is 38.5 ◦ C and -27.2 ◦ C respectively
[19]. This means modern transformers are designed to a maximum rated temperature
of 103.5 ◦ C . Nevertheless, transformers operate at temperatures of between 60 to 80
◦C [20].
Copper winding wires are normally in the form of rectangular cross-section wire or
strip complying with EN 13601:2002 [21]. Circular cross-sectional area wires have worse
space-factors and less mechanical strength, therefore its use is limited to transformers
with less than a few kVA, the selection of this circular type wire requires a special type
of winding arrangement [6].
The low voltage(LV) and HV winding have very different designs, because they have
different electrical stresses. The current density of most oil filled transformers is be-
tween 2 and 4 A mm−2 , limited due to thermal losses [6]. Compared to the HV side
the LV side will have more than double the current passing through its winding, which
means the cross-sectional area is much greater. Due to non-uniform magnetic field and
the large cross-section area of the LV conductor, eddy currents are generated. Eddy
currents are swirling electrical currents induced within conductors, which oppose the
changing orthogonal magnetic field plane. To decrease eddy currents the LV conductors
are subdivided into smaller strands individually insulated from each other, as shown in
Figure 2.1. To prevent circulating currents due to different conductor strands experienc-
ing different flux densities, which induces different voltages, the strands are transposed
ensuring that all conductor strands experience identical leakage flux [6].

Figure 2.1: Transposed LV Transformer Conductor [6]

11
The HV winding is normally subdivided into two strands or not subdivided at all since
the current is much smaller. The challenge with the HV winding is that it requires
many more turns than the LV, therefore the HV side is wound around the same place
several times creating discs/sections, as shown in Figure 2.2. The disc can be normally
wound using a continuous disc winding arrangement, where each disc is wound in reverse
compared to its adjacent disc so that the ‘start’ of one disc is adjacent to the ‘finish’
of the other disc. In large transformers, fully or partially interleaved HV windings are
used to improve the voltage distribution in the winding [22]. The electrical field stress
per turn will vary greatly depending on the transformer size and design, but as a rule,
large transformers operate at about 300 V per turn [23]. Transformers are tested at
about twice the working voltage and most manufacturers have generous safety margins
to avoid test failures.

Figure 2.2: Arrangement of a continuous disc HV winding [6]

For efficiency and size purposes, the windings in power transformer are normally placed
around the same core limb. Most commonly, the LV winding is wound between the core
limb and the HV winding, as shown in Figure 2.3 [6]. In a three-phase transformer, there
are three pairs of windings, which are normally separated by 120 electrical degrees, as
shown in Figure 2.4. The three-phases can connect in three possible ways delta (∆),
star (Y ) and zig-zag (Z). It is also possible to shift electrically the phase with respect to
the primary by a certain degree, by rearranging the order of the phases in the secondary
winding, this phase shift is defined by using hours where 1 hour equals 30 ◦ [24]. There
is a specific notation used to describe what winding set-up is used in a transformer, as
shown in Figure 2.5.

12
 Figure 2.3: 400kV Three-Phase Transformer with LV winding wound between the
core limb and the HV winding

Figure 2.4: Three-Phase Voltage, with each phase separated by 120◦

Figure 2.5: Transformer winding configuration notation

Transformers can have other windings besides the primary and secondary windings.
Tertiary delta-connected windings are sometimes used to provide a low impedance path
for third harmonic currents, which are generated due to the non-linearity of the B-H
curve of the core. Another winding section that practically all power transformers have
are tapping windings, which allows for variation of the applied voltage while keeping

13
the output voltage the same. For distribution and small auxiliary transformers there is
a normally tap changer allowing a 5 % variation, with adjustment off-line. On larger
transformers, there may be an on-load tap changer or even an auto-tap changer, which
maintains on-line the output voltage constant. The tap changer variation in larger
transformers is normally in the order of 10 %. The tap changer is normally located in
the HV side because it has a lower current. On larger transformers, the tappings are
accommodated using separate tapping winding, so as not to disturb the electromagnetic
balance of the transformer tertiary winding [6].

2.1.2 Dielectrics

Both liquid and solid dielectric materials are used in transformers. The solid dielectrics
provide both structural support and insulation, while the liquid dielectric acts as both
an insulator and a coolant.

2.1.2.1 Insulation Fluid

There are several liquid dielectrics suited to use within transformers, for instance silicone
liquids, synthetic ester fluids, vegetable oil and mineral oil. Silicone is normally used in
places where fire hazards must be avoided due to its high flash point (above 350 ◦ C ),
high thermal stability and chemical inertness. Synthetic ester fluids have been designed
to meet HV insulation fluid specifications and can be produced mainly from vegetable
oil. They have a very low toxic level and good biodegradable characteristics, the dis-
advantage being that this is reflected in high costs. Normally transformers worldwide
are filled with the mineral oil as it is the most economical feasible and widely available
solution. Also in highly congested places where transformers are placed in closed spaces,
sulphur hexafluoride (SF6 ) gas insulation is used. This research focuses on super grid
transformers where the vast majority of transformers use mineral oil complying with
IEC 60296 [25].
Mineral oil is a product generated by crude oil, through several purification proce-
dures. Crude oil is mainly made up of hydrocarbon molecules, but it contains significant
amounts of other impurities such as sulphur, oxygen, nitrogen, inorganic salts, metals
and aromatic compounds [26]. The crude oil’s impurities vary widely from field to field
[27] creating distinctive electrical and chemical properties in each mineral oil. The main
type of transformer mineral oil used is naphthenic based because of its low temperature
viscosity and pour point properties. As naphthenic crude oils constitute less than 5 % of
the world’s crude, paraffinic crude oils are occasionally used especially in warm countries
where low temperature viscosity properties do not cause a problem [28, 29].
The refining process of crude oil to transformer mineral oil starts by first separating the
different hydrocarbon molecules over a selected temperature range by distillation under
atmospheric pressure. It is then further distilled at reduced pressure, to get a more
14
precise control over the boiling range and remove further undesirable parts of the crude
oil. Afterwards, two procedures can be used to remove sulphur, oxygen, nitrogen com-
pounds and reactive hydrocarbons. The first is solvent refining, which relies upon the
selective solubility of each of the substances being removed using a carefully controlled
solvent system. The second procedure is hydrotreating, which is a pressure, tempera-
ture and hydrogen catalyst driven process that removes nitrogen, sulphur, oxygen and
aromatics by a chemical reaction called hydrogenation [27]. Hydrocracking is another
form of hydrogenation, however removes aromatics to a much higher degree resulting
in a water-white colour fluid, which is corrosive sulphur free. It is this technology that
the paraffinic transformer oils are based upon. The intensity of Hydrotreating and Hy-
drocracking processes is denoted by a term called ‘Severity’. Severity is driven by the
catalyst type, the magnitude of pressure and temperature, the ratio of hydrogen to oil
and the contact time allowed for the reaction, the higher the severity the higher the
purity of the oil [28]. These advances in refining processes have been decreasing the sul-
phur content since 1964, and with the development of severe hydrogenation techniques
it is now possible to reduce sulphur compounds to negligible levels, less than 1 ppm [26],
as shown in Figure2.6.

Figure 2.6: Average sulphur content of mineral insulating oils as a function of the
year of transformer production [1]

Depending on the purity of the mineral oil the severity of both the solvent and hydro-
gen treatment can be altered. The objective of refining is not to remove all impurities,
but instead to remove as little of the impurities as required from the transformer min-
eral oil so that it has the required properties. The reason for using a combination of
modest solvent extraction and mild hydrogen treatment is because an optimum amount
of aromatics is needed to maintain good oxidation resistance and inhibitor response

15
[27]. Additives, such as synthetic oxidation inhibitor and pour point depressants, can
be added to mineral oils to obtain the desired properties. The oil is classified according
to standards [29, 30], which depend on the additives and impurities contents.
Transformer mineral oil degrades mainly due to electrical and thermal stresses under an
oxidizing atmosphere. Chemical degradation is seen when the hydrocarbon compounds
lose their stability and decompose by the oxidisation of the oil, precipitation as sludge
and the increase of the acidity level [27]. The electrical degradation is demonstrated
by an increase of the dissipation factor and a decrease in interfacial tension, dielectric
resistance and breakdown voltage. The rate of degradation will depend on the electrical
and thermal stresses being applied, contaminants introduced such as moisture content
and degradation compounds from the solid cellulose materials, and the initial quality of
the mineral oil [31].
The lifetime of the liquid insulation in a transformer is expected to be the same as the
transformer, which is approximately 50 years [27]. As the insulation liquid is the biggest
bulk of the dielectric system and it is the most accessible, it is used for the main con-
dition monitoring techniques. Dissolved gas analysis (DGA) tests are one of the most
common standard oil tests performed in industry [32]. The information given in this
test is used to assess the transformer mineral oil and indirectly assess the systems such
as kraft paper insulation. Moisture content is also monitored as it influences greatly the
dielectric properties of the solid and liquid insulation systems and its lifetime [33, 34].
In addition, online sensors placed in transformers to detect partial discharges, which is
the initial stage of degradation of electrical properties of the insulation system, ultra-
high frequency, acoustic sensors and optical sensors are a few of the techniques used
[35]. Standard dielectric dissipation factor (DDF) and relative permittivity tests are
also used to check the electrical properties of the oil insulation system [36, 37]. Table
2.1 shows the oil properties for a standard grade uninhibited insulating oil. Mineral
oils increase in viscosity with decreasing temperature. The pour-point of a fluid is the
lowest temperature at which the fluid is capable of any observable flow. Volatility is a
measure of how easily the oil changes composition. Flash point is the temperature at
which a particular organic compound gives off sufficient vapours to ignite in air. Chem-
ical stability is very important in oil in order to retain low electrical conductivity and
good thermal properties. It is mainly affected by three factors: temperature, oxygen
and catalysts. The main characteristics of insulating oil are; low viscosity, low pour
point, high flash point, excellent chemical stability and high electrical strength. Less
important but still desirable properties are a high specific heat capacity, high thermal
conductivity, good impulse strength, low solvent power, low density, good arc quenching
properties, non-toxic and above all low cost and ready availability. No oil meets all
the desired properties, thus when choosing the insulation oil, the engineer has to find a
compromise between certain properties [27].
Due to the high thermal expansion coefficient of transformer oil, it is necessary to pro-
vide a mechanism to accommodate this expansion, in the form of a transformer breath-

16
 TEST LIMITS TYPICAL
PROPERTY UNIT
METHOD MIN MAX DATA
1 - Function
Viscosity, 40 ◦ C mm2 s−1 ISO 3104 12.0 9.6
Viscosity, -30 ◦ C mm2 s−1 ISO 3104 1800 1100
Pour point ◦C ISO 3016 -40 -51
Water content mg kg−1 IEC 60814 30 <20
Breakdown voltage
- Before treatment kV IEC 60156 30 40-60
- After treatment kV 70 >70
Density, 20 ◦ C kg dm−3 ISO 12185 0.895 0.875
DDF at 90 ◦ C IEC 60247 0.005 <0.001
2 - Refining/stability
Appearance IEC 60296 Clear, no sediments complies
Acidity mg KOH g −1 IEC 62021 0.01 <0.01
Interfacial tension mN m−1 EN 14210 40 48
Corrosive sulphur DIN 51353 NC NC
Potentially corrosive
IEC 62535 NC NC
sulphur
Corrosive sulphur ASTM D1275B NC NC
DBDS mg kg−1 IEC 62697-1 ND ND
Antioxidant wt% IEC 60666 ND ND
Metal passivator
mg kg−1 IEC 60666 ND ND
additives
2-Furfural and related
mg kg−1 IEC 61198 0.05 <0.05
compounds content
Aromatic content % IEC 60590 9
3 - Performance
Oxidation stability at
IEC 61125C
120 ◦ C ,164 hours
Total acidity mg KOH g−1 1.2 0.55
Sludge wt% 0.8 0.18
DDF at 90 ◦ C 0.500 0.080
4 - Health, safety and environment (HSE)
Flash point ◦C ISO 2719 135 152

Table 2.1: Standard grade uninhibited insulating mineral oil properties [38], ND - not
detectable, NC - not corrosive

ing system. All transformers need this with the exception of the smallest distribution
transformers. Refrigeration breathers, a specific breathing system which continuously
dehydrates the insulation oil, are present in all 400 kV and 275 kV transformers in the
UK [6].

17
2.1.2.2 Insulation Paper

Paper is among the cheapest and best electrical insulation materials known. Transformer
insulation paper must meet certain physical and chemical standards [39]. It must also
meet the required electrical properties specification such as high dielectric strength, a
dielectric constant that matches to that of oil and low dielectric loss. The dielectric
constants of the paper and the oil should be as similar as possible in order for them to
share the electrical stress more evenly. Kraft paper is made entirely from unbleached
softwood pulp manufactured by the sulphate process; it is unbleached because residual
bleaching agents might be hazardous to the electrical properties of the paper [6]. This
process produces paper with a slightly alkaline pH level, between 7 and 9. Paper or
pressboard strength is primarily determined by the bonding forces between fibres, which
are influenced by the type and extent of the refining process. Softwood cellulose is the
most suitable for electrical insulation due to its fibre length of 1-4 mm, which gives the
highest mechanical strength [6]. Nevertheless small quantities of other harder woods
may be added to create a kind of alloy that offers better properties than the individual
constituents do. Cotton fibres have been used in the past but the higher cost and lower
intrinsic bonding strength makes them unsuitable as an individual material, although
they can be used as a mixture with softwood fibres to produce superior electrical strength
and mechanical properties. IEC 60554 lists the properties of the main paper types
being used currently [39]. There are also specially designed papers to meet the specific
requirements of the transformer industry. Some of these are:

• Crêpe paper

• Highly extensible paper

• Thermally upgraded paper

• Diamond dotted press paper.

Crêpe paper is extensible, making it ideal for hand applied coverings where there are
irregular contours. One main disadvantage of this paper is its tendency to lose elasticity
through time meaning the tightness of the covers might be lost. A more recent devel-
opment is the highly extensible paper, it has smooth surface, stretch and cross machine
tear properties and retains a high tensile strength. The high mechanical strength of
the paper makes it ideal for machine application. Thermally upgraded paper is treated
with stabilisers which can reduce thermal degradation. Diamond dotted press paper is
synthetic resin bonded paper tubes and is normally used in locations having lower elec-
trical stress but where high mechanical strength is important. The reason for their more
selective use is due to the large ratio of resin to paper necessary to obtain the required
mechanical strength, which makes it very difficult to impregnate it with transformer oil
[6].
18
Pressboard in its simplest form is just thick insulation paper made by laying up a num-
ber of layers of paper at the wet stage of the manufacturing process. There are two main
types of pressboard - those built up purely by stacking up paper layers of a thickness of
2-3 mm, and those constructed by adding suitable adhesive between the layers, with a
thickness of 4.5-6 mm [40]. There are three sub-categories techniques for the pressboard
material:

• Calendered pressboard, which has 5 % moisture and density from 1.15-1.30 g cm−3 ,

• Mouldable pressboard, which contains 5 % moisture, density of 0.9 g cm−3 , has

good oil absorption and is malleable to some degree.

• Pre-compressed pressboard, which produces a stronger, more stable, stress-free

material with density of 1.23 g cm−3 . It is capable of retaining its shape and
dimensions throughout the stage of the transformer manufacture and the thermal
cycling in oil under service condition to a far better degree than the other two.

The latter type is the most used for pressboard, but mouldable pressboard is used for
the insulation between the LV and HV windings, due to the irregularity of the windings
separation gap. Laminated pressboard starts at 10 mm thick and goes to 50 mm or
more; this can be any type of the un-laminated pressboard material mentioned above.
Dried wood can also be used in small distribution transformer for core frames. For larger
transformers wood can be economically used for lead supports. It can also provide an
alternative to pressboard for winding end support slabs.
Both paper and pressboard are cellulose based materials, which is a natural polymer
chain with repeating glucose rings. The average number of glucose monomers per chain
is called the degree of polymerisation (DP). Degradation occurs as the polymer chains
break down decreasing the DP, which decreases the tensile strength [20]. Typical cel-
lulose degradation by-products formed are carbon monoxide, carbon dioxide and furan
compounds. The full degradation mechanism of cellulose base systems is complex and is
associated with a variety of causes, many of which are design specific, but temperature,
moisture, air, acidity and electrical stress are recognised to strongly influence cellulose
ageing process [41, 20, 42]. Eventually the cellulose base insulation systems degrade
to such an extent that it loses all its mechanical strength and becomes susceptible to
mechanical damage, which puts the electrical integrity of the equipment at risk. In
addition, as the paper degrades so does the electrical breakdown strength, which causes
partial discharges to occur that in turn, accelerate even further the degradation process
[20].
The lifetime of a transformer is mainly limited by the lifetime of the solid insulation,
more specifically the lifetime of the winding paper insulation. Therefore knowing the
current state of the winding insulating paper is extremely valuable. DP analysis is a
good parameter to evaluate the state of the kraft insulation paper. A new unused paper

19
has a typical DP value of approximately 1300, as it degrade this value drops and when
it reaches 200-150 is considered to be at the end of its life [20]. The problem with DP
analysis is that the current standard technique used to measure the DP is destructive
and requires removing a sample from the transformer (BS EN 60450:2004). Dissolved
gas analysis (DGA) has been investigated to track the degradation of paper by mon-
itoring carbon monoxide and carbon dioxide levels [43]. Research has also focused on
using furan concentration in oil to measure indirectly the DP, suggesting that there is
a logarithmic relationship between 2-furfuraldehyde and the DP [33]. Partial discharge
(PD) is an early consequence of the degradation of insulation and therefore vast amounts
of research have gone into PD detection and analysis systems [44]. As the solid insula-
tion degrades it changes its dielectric properties, a standard technique to measure these
changes is to use tan delta/power factor measurements. Although it is possible to use
dielectric measurements and data analysis techniques to verify how conductive an insu-
lation system is, narrowing down the cause is challenging and the procedure varies from
transformer to transformer [33, 45, 46, 7].

2.1.3 Transformer Core and Tank

The most commonly used iron core shape in the UK is the 3-phase 3-limb core, which
has no return limb because the sum of the three phases always equal zero, Figure 2.7.
The top and bottom yokes must have the same cross-sectional area as the limbs. When
the height dimension is restricted a five limb core-type transformer is used, Figure 2.8.
Single-phase transformers are used in large generation sites for economical reasons and
physical size restrictions.

Figure 2.7: Three limb core-type transformer [6]

20
 Figure 2.8: Five limb core-type transformer [6]

This core needs to provide a low reluctance path for the magnetic flux, which allows
us to redirect the magnetic flux so that it links the primary and secondary windings
in an efficient way. There are some losses due to both changing polarity of the field,
known as hysteresis losses, and eddy currents being generated within the core. Also the
alternating fluxes generate low frequency acoustic losses which can be invasive to the
environment, especially in the case of large transformers. The losses of a transformer
can be determined for a given design. The hysteresis losses, Wh [W] are

ns
Wh = kh f Bmax (2.3)

where kh is a material constant, f is the applied frequency [Hz], Bmax is the maximum
flux density [T] and ns is the Steinmetz exponent that is approximately between 1.6 and
2.5. The eddy current losses, We [W], are

2
ke f 2 d2c Bef f
We = (2.4)
ρT C

where ke is a material constant, dc is the thickness of the transformer core [m], ρT C is

the electrical resistivity of the transformer core [Ω m−1 ] and Bef f is the flux density of
the applied rms voltage [T].
To improve efficiency the core must be designed with materials that have a minimal
hysteresis loop area and an increased resistivity to reduce eddy currents, also by having
stacked laminated sheets design it will further minimise eddy currents. The IEC60404
covers the specification of magnetic materials including core steels [47].
The transformer tank provides containment for all the physical parts of the transformer
and for the dielectric fluid; on larger units it may provide additional structural support.
The tank (Figure 2.9) allows access to bushing connections, winding temperature CTs,
core earthing links, off-circuit tapping links and the rear of tapping selector switches [6].

21
 Figure 2.9: Transformer tank [6]

2.2 Corrosive Sulphur

The international standard definition of corrosion is the “physicochemical interaction

between a metal and its environment which results in changes in the properties of the
metal and which may often lead to impairment of the function of the metal, the environ-
ment, or the technical system of which these form a part” [48]. Noting that “corrosion”
is a process, the rate at which it occurs is dependent on the initial corrosive element,
the metal, and can be influenced by other variables acting as catalysts or inhibitors. In
oil insulated transformers the corrosion process is normally due to sulphur compounds
interacting with the copper surface. As the copper is mainly located in the windings,
which are normally wrapped in kraft paper, the corrosion process creates copper com-
plexes that compromise the winding insulation system.
A surge of corrosive sulphur related faults in oil insulated power apparatus has been
reported since the start of the 21st century [12, 49]. Although it has been investigated
before [9], the few cases were treated as isolated events leading to superficial analysis
and missing the real cause of the problem [8]. The recent set of failures appears to be
different as the volume of failed unit is much larger and the affected units, mainly shunt
reactors and power transformers, are not contained in any specific geographical region
[8]. The failed units were all within established industry standards and practice, and
no manufacturing errors could explain the failures. The common denominator found
was the type of mineral oil used [8], suggesting that some new corrosive compound was
introduced to the insulation mineral oil, which was able to pass undetected through the
standard corrosion test. Insulation oil flows in a thermal convectional pattern through

22
the transformer. This provides a mechanism, which allows contaminants, such as cor-
rosive sulphur compounds, to spread throughout the transformer and cause damage. In
the past decade a considerable amount of corrosion related research was added to the
state-of-the-art, new corrosive sulphur detection tests and updates to corrosion stan-
dards [50, 51, 52], mathematical models for corrosion were suggested [53] and mitigating
techniques were developed[54]. This section will analyse this research in more detail.

2.2.1 Sulphur Compounds

As explained previously, some sulphur compounds come from the crude oil and only a
certain amount of sulphur is removed for the oil so that it passes the required tests. Re-
moving all the sulphur compounds will remove the natural oxidation inhibiting properties
of the oil. Some oils, such as iso-paraffin based oils, using decontamination processes
such as severe hydrogen treatment to remove almost all sulphur, will then require syn-
thetic oxidation inhibitor additives that may be sulphur based [28]. There are tens to
hundreds of different sulphur compounds present in mineral transformer oil [55], but the
problem only resides with the corrosive sulphur species.
The British Standards Institution (BSI) and the International Electrotechnical Com-
mission (IEC) have defined potentially corrosive sulphur as “organo-sulphur compounds
present in transformer oils that may cause Cux S formation” [52]. The IEC 62535 stan-
dard also notes that some of these compounds may initially be stable and only become
corrosive under certain operating conditions. Some sulphur compounds are thermally
activated by hot metal surfaces, such as conductors [50], but Cux S formation is due
to multiple compounds. Hence lowering the temperature of the transformer does not
prevent Cux S deposition completely [1]. The concentration of the compound in the
transformer oil and its reactivity are the main variables affecting the chemical reaction
rate, leading to the creation of Cux S complexes. Types of sulphur normally found are,
in increasing order of corrosivity [50]:

1. Thiophenes

2. Disulphides and polysulphides

3. Thio-ethers

4. Mercaptans

5. Elemental sulphur.

Current refining techniques are to a standard that amounts elemental sulphur and mer-
captans are below detectable levels [56]. Thio-ethers are moderately corrosive com-
pounds, which can be found after the oil refining processes with concentration relative
to the severity [28]. Disulphides and polysulphides are normally stable compounds but
23
some species such as dibenzyl disulphide (DBDS) can become susceptible to heat break-
down and form mercaptans (RSH) which are corrosive in transformers [28, 56]. That
is why some corrosive sulphur compounds have been considered in the past to be good
oxidation inhibitors, leading to significant concentrations of corrosive sulphur, i.e. 100
to 350 ppm of DBDS [1], in the insulation oil after the refining process [12]. Some
other possible sources, which have not yet been confirmed, are gaskets, water-based
glues, copper and Kraft paper [12, 13]. Thiophens are very stable compounds found
after all refining processes, which act as natural oxidation inhibitors and is necessary
for totally uninhibited oil [28]. It has been shown that sulphur content in oil has been
reduced significantly in the 1990’s, (Figure 2.6). Conversely, approximately 50 % of the
power transformers installed since 1990 have tested positive for corrosive sulphur, with
a predisposition to failure [1]. Figure 2.10 demonstrates the increasing trend in the de-
tection of corrosive oils. These two contradicting trends suggest that although sulphur
compounds in crude oil might play a part in corrosive oils, it does not account for the
increasing trend seen from the late 20th century. This suggests that an additional prod-
uct, containing sulphur compounds, has been added after the refining process. A study
was performed on 14 unused inhibited and uninhibited naphthenic mineral oils, using
gas chromatography and atomic emission spectroscopy (GC-AED), which found DBDS
to be predominant in all oils [1]. This was then tested with co-chromatography and
mass spectrometry to match its reactivity in oil with copper, as shown in Figure 2.11,
demonstrating that DBDS may become corrosive to copper, especially under high tem-
peratures. Transformer design changes in the 1980’s increased the copper temperatures,
which may have transformed some sulphur based additives, like DBDS, into corrosive
compounds [13]. It is now recommended that oil purchased from the late 1990’s to the
early 2000’s be tested for corrosive sulphur [57].

Figure 2.10: Percentage of transformer oils found to be corrosive as a function of the

year of transformer production,using D1275A ASTM standard corrosion test method-
ology (blue) and the 48 hour extension version (red), modified from [1]

24
 Figure 2.11: GC-MS graph for a mineral oil sample containing DBDS before and
after ageing at 140 ◦ C in a fan oven for 72 hours, the identified peaks corresponds to
the quantity DBDS in before and after ageing [1]

DBDS has been identified through experimental studies to be the main compound in
corrosive sulphur related faults [50]. When compared with other potentially corrosive
compounds like dihexyl sulphide (DHS), dihexyl disulphide (DHDS), dibenzyl sulphide
(DBS) and dibenzyl sulfoxide (DBSO), DBDS corrodes the copper surface as much as
the other compounds, but the rate at which DBDS forms Cux S deposits is much greater
[58]. At low temperatures, DBDS is stable and acts as an oxidation inhibitor, but at high
temperatures, it becomes corrosive. It has been suggested that DBDS has a triggering
temperature, which is known to be below 150 ◦ C [59]. Due to the lack of standardised
quantitative tests for detecting corrosive sulphur compounds there are some discrepan-
cies in the threshold of DBDS concentration which is considered to render oil corrosive
[9, 60, 61].

2.2.2 Sulphur Corrosion Process

The sulphur corrosion process and its effects on oil insulated power equipment have been
the primary focus of this new surge of investigation into corrosive sulphur in the power
industry. With no available solution, only mitigation strategies, the industry focused
research on what affected the rate of corrosion and what effects corrosion has on both
the conducting and insulating media. Currently there is no agreement on the overall
process by which corrosive sulphur related faults develop, from the moment the corrosive
sulphur compounds contaminate the transformer oil, to the moment a fault occurs due

25
to Cux S deposits. This can be segmented into three parts: the chemical reaction that
produces Cux S particulates, the deposition of Cux S particles in the insulation paper,
and the failure mechanism.
Studies have shown that the deposition rate of Cux S is directly proportional to the
DBDS concentration and the incubation period is inversely proportional to the DBDS
concentration [62]. The incubation period is seen as the time required for DBDS to
generate and deposit Cux S on paper. Suggesting that halving the concentration of
DBDS, the incubation period is doubled effectively doubling the lifetime due to corrosion.
The Cux S deposition is low in oils that contain less than 1 ppm of DBDS [62], but
the corrosion process is still present [63]. The type of mineral oil affects corrosion
process, although in every case DBDS enhances the corrosion of copper. It has been
shown experimentally that the rate and amount of deposition of Cux S on paper varies,
also the saturation of DBDS in the oil will vary by more than a factor of two from
oil to oil [64]. Temperature is directly proportional to the corrosion rate as shown
in numerous field and laboratory experimental studies [64, 1, 65]. Some studies show
that the deposition rate is approximately doubled when the temperature is increased
by 10 ◦ C [13, 62]. This proportionality terminates when the temperature exceeds the
DBDS decomposition temperature, which occurs between 150 ◦ C [63] and 180 ◦ C [61], as
shown in Figure 2.12. It has also been empirically demonstrated that the concentration
of oxygen in the atmosphere has an impact in the deposition process [66]. Under a
nitrogen atmosphere, the Cux S deposition process is not fully developed [53]. Under an
air atmosphere, Cux S is deposited on paper, demonstrating that la ow oxygen content
decreases the deposition rate [67, 53]. Laboratory studies have also shown that when 2,6-
di-ter-butyl-p-cresol (DBPC), which is an industry used synthetic oxidation inhibitor,
is present with DBDS under oxygen atmosphere it significantly increases the rate of
deposits of Cux S on insulation paper [66, 68].

120

100

80
200 °C
DBDS (ppm)

200 °C + Copper
60 180 °C

180 °C + Copper

40 100 °C

100 °C + Copper

20

0
0 5 10 15 20 25 30 35
Time (Days)

Figure 2.12: Thermal degradation and oxidation of mineral oil with DBDS in an
open vessel to air as function of time, at various temperature and where some samples
contain a copper specimen (+ Copper), modified from [61]

26
Due to the lack of understanding of the chemical processes in the complex matrix of
mineral oil, simplified models have been developed to model Cux S deposition. Based
on the variables investigated several models have been suggested. A chemical reaction
has been proposed, which suggests a two-step process; initially the DBDS and some
oil soluble copper complexes are formed. Secondly, the copper and sulphur complexes
are absorbed into the paper insulation, where they then decompose into Cux S, dibenzyl
sulphide (DBS) and bibenzyl (BiBZ) [51]. Figure 2.13 shows the proposed chemical
process diagrammatically.

Figure 2.13: Schematic of proposed Cux S formation mechanism [11]

Following this process a model was proposed, where first there is dissolved DBDS-Cu
complexes in the oil Coil (t) [ppm]. Then DBDS-Cu complexes are absorbed by the
insulation paper Dp (t) [ppm], before finally deposits of Cu2 S particulates build up on
the insulation paper Q(t) [ppm], as shown in Figure 2.14 below [1, 62]. Assuming the
first reaction occurs at the copper surface, the concentration of DBDS-Cu complex in
oil Coil (t) [ppm] is given by

Coil (t) = C0 (1 − e−kC xt ) (2.5)

Where x is the DBDS concentration [ppm], which is proportional to the generation rate
of DBDS-Cu complexes and kC is the proportionality constant. The C0 is the saturated
DBDS concentration in oil [ppm] and t is time [s]. By rearranging the Equation 2.5
we obtain tA [s], the heating time required for the DBDS-Cu complex to start being
absorbed by the insulation paper,
 
 CA 
ln 1 −

C0 
tA = − (2.6)
kC x
27
 Figure 2.14: Cux S deposition model as a function of heating time at a constant
temperature, modified from [62]

It is assumed that absorption of DBDS-Cu complexes starts only when Coil (t) reaches
a minimum concentration of CA [ppm]. The concentration of the absorbed DBDS-Cu
complex in insulation paper Dp (t) [ppm] is given by

Dp (t) = kD [Coil (t) − CA ] (2.7)

where kD is the equilibrium constant between dissolved and absorbed DBDS-Cu com-
plexes. Applying the same principles above to find the heating time for Cux S to start
depositing in the paper insulation tG we find that
 D 
 CA + k G 
ln 1 − C0
D 

tG = − (2.8)
kC x

The concentration of Cux S deposits Q(t) is given by

Q(t) = kq [Dp (t) − DG ] (2.9)

where kq is the equilibrium constant between absorbed DBDS-Cu complexes and de-
posited Cux S, DG is the minimum concentration of DBDS-Cu complex absorbed by pa-
per to start generating Cux S deposits, and CG is the minimum concentration of DBDS in
oil for it to generate Cux S deposits. This model takes C0 , CA , kC and kD to be constant
under constant temperature. Analysing the model it is possible to verify mathematically
that tA is inversely proportional to the DBDS concentration, tG is inversely proportional
to the DBDS concentration, and that the deposition rate is proportional to the DBDS
concentration. This model is limited as it takes temperature as a steady-state variable
and a DBDS concentration must be approximately 10 ppm or above [62]. Increments

28
to the model above have been suggested, such as the behaviour of catalysts such DBPC
[68] and the impact of oxygen [53] in the three stages of the process. Other studies tried
to use polynomial fits to data to model the depletion of DBDS [63]. Some studies used
the Arrhenius equation to model the activation energy (temperature) required to start
the corrosion process [59] and model the rate of corrosion due to different variables [67].
A study, [64], derived an empirical formula,

U
T = + tg (2.10)
x − CG

to calculate T [s], the incubation period of DBDS in oil to generate Cux S deposits, where
U is a constant determined empirically.
Although the two-step process model has wider consensus, it still agreed that this model
does not fully represent the corrosion and deposition process [11]. The deposition pat-
tern of Cux S is also an important aspect of the corrosion process. Copper can travel into
the oil in three possible ways diffusion, where copper-sulphur complexes concentration
is described using Fick’s law; dispersion, where copper-sulphur complexes move with oil
flow and are filtered by the paper; dissolution, copper-sulphur complexes are soluble in
oil and decompose in the paper [69]. Dissolution appears to have the greatest agreement
amongst researchers [62, 68, 11], where it has been suggested that mercaptans [65] and
H2 S [13] react with copper to from oil soluble complexes. It has also been suggested
that Cux S is formed at the copper surface by direct reaction between copper and sulphur
compounds [1, 70] and which is then dispersed into the paper. Scant support has been
given so far for any of these theories.
Forensic analysis of failed transformer units demonstrate that the highest concentration
of Cux S occurs at the interface between the Kraft paper layer and the copper conduc-
tor. Thereafter its concentration decreases with each successive outer tape paper layer,
which suggests a diffusion gradient with an inverse function of the distance from copper
surface [60, 49]. However, some cases have shown high concentration of Cux S on the
outer layer [71, 72]. An elemental surface concentration of copper and sulphur on kraft
paper layers suggests that there are two deposition processes; one process has a high
initial concentration, but a low diffusion rate [73], the second process has a low initial
concentration, but a higher diffusion rate. This study also models the Cux S deposition
using Fick’s law of diffusion for the boundary condition of planar diffusion into a solid
membrane bounded by sources of constant concentration (turn-to-turn copper conductor
at x=0 and x=l) for each process, which can be simplified by two, one for each process,
as shown by  
Ln Ln
xLn = Cs erf c(l − ) + erf c(l + ) (2.11)
2(Df )0.5 2(Df )0.5
where Ln indicates the layer number, erfc the complementary error function, xLn the
copper concentration at the layer Ln [ppm], and Cs is the copper concentration at the
source [ppm] and Df is the diffusion coefficient of copper through insulation paper layers,
which is temperature dependant [73]. Figure 2.15, demonstrates the development of

29
copper contamination in Kraft paper through time in a turn-to-turn conductor scenario,
where each conductor has 15 Kraft paper layers wrapped.

Figure 2.15: Copper concentration as a function of Kraft paper layer position, between
two copper boundaries (0 and 30), at various times in service (1 day to 30 years)
assuming a constant ageing temperature of 25 ◦ C , modified from [73]

2.2.2.1 Effects of Corrosion

The contamination of oil-impregnated insulation paper with Cux S demonstrates signif-

icant changes in the insulating properties, as shown in Table 2.2. The ability of oil to
deposit copper compounds onto the insulation paper has been shown to occur in both
corrosive sulphur contaminated oil and non-contaminated oil, where the deposits vary
from Cux S to copper oxide respectively [74]. Literature says that deposits are made of
a double form of Cux S, CuS and Cu2 S, where there is a concentration of 80.4 % Cu and
19.6 % of S although the accuracy is not stated [50]. The amount of copper particulates
alters the electrical properties of the insulation paper. The tangent of loss angle, tanδ,
changes from 0.003 approximately to greater than 1.0 and the resistivity drops from MΩ
to the tens of kΩ [1, 62, 11].

tan(δ) Resistivity(Ωm−1)
Insulating paper windings, without
0.003 5 × 1012
visible Cux S contamination
Insulating paper windings, with low
0.005 5 × 1010
Cux S contamination
Insulating paper windings, with high
>1 5 × 104
Cux S contamination

Table 2.2: Effects of Cux S Contamination on the Electrical Properties of Oil-

Impregnated Kraft-Paper [50]

30
The increase in tan δ leads to an increase of temperature. The temperature rise away
from a hotspot, Th [◦ C], can be described using

∆d
Th = −Pd ρT (2.12)
Acs

where ∆d [m] is the thickness of the paper layers, Acs [m2 ] is the cross section in the
direction of the hot spot and ρT [K m−1 W−1 ] is the thermal resistivity of oil-paper
insulation [75]. Pd [W] is the dielectric power loss and represents the heat generated at
the hot-spot, which is created by the losses within the Cux S contaminated paper, given
by
Pd = ωCL VL2 tan δ (2.13)

Where CL [F] is the capacitance of the contaminated tape layer, VL [V] is the voltage
across it and tan δ is the dissipation factor of the contaminated layers comprising the
hot-spot [1, 76]. If the tan δ values from the table above are used for uncontaminated
and highly Cux S contaminated paper, then the heat power generated is over 300 times
greater. The temperature increase might not be as significant, because Cux S changes
the thermal resistivity.
In a majority of field cases, a direct relationship was found between copper content in oil
and tan δ [50]. Further studies have shown that suspended copper complexes does not
significantly affect the dielectric properties of oil, but on the other hand soluble copper
complexes significantly affect the dielectric loss [77]. It has been demonstrated that if the
permittivity of kraft paper increases above the permittivity of the oil there is a consid-
erable reduction in the breakdown strength [78]. The addition of DBDS to transformer
oils was to improve oxidation stability [1]. The sulphur corrosion process depletes DBDS
at rate relative to the generation of copper sulphate complexes [63]. Therefore, one of
the effects of sulphur corrosion to oil is the depletion of oxidation inhibitors that lead to
poor oxidation stability. The effects of Cux S deposits or copper-sulphur complexes on
the electrical fields or vice-versa have not been thoroughly investigated [69].
The development of transformer Cux S faults has been explained by discussing the effects
that sulphur corrosion has on the insulation properties of different media in the trans-
former. Although one understands the development, it does not mean that the failure
event is straightforward. Being able to understand what is the failure event can help
better predict imminent failure and lifetime expectancy. There are two main theories
for the failure event; one theory suggests that the heat generated by the Cux S deposits
on paper generates a hotspot, which then increases the rate of corrosion, increasing the
rate of Cux S deposits, further increasing the temperature of the hot spot. This snowball
effect eventually reaches a point where the heat generated can no longer be dissipated
by the oil, generating a thermal runaway [1]. A second theory suggests that the Cux S
deposits on the paper change the dielectric properties of the kraft paper and lower its
breakdown strength. The fault occurs when a bridge of Cux S is formed between two
adjacent turns in a winding or when the breakdown strength is too weak to withstand

31
the turn-to-turn electrical stress levels [79, 80].
According to the Mineral oil IEC 60599 standard fault temperatures below 300 ◦ C are
deemed low thermal faults, meaning it has a larger effected region [29]. This leads to
Cux S deposits being more diffused throughout the winding conductors [60], which mean
that this type of fault is hard to pinpoint in the transformer until it actually happens.
Nevertheless it is agreed that corrosive sulphur related transformer faults are usually of
the turn-to-turn type, commonly located at the top of the winding where the tempera-
ture is usually higher [10, 81].

2.2.3 Detection and Condition Monitoring

Detection of corrosive sulphur species is performed initially by the transformer oil man-
ufacture and then by the transformer operator. Although there are standard corrosion
tests, the results are qualitatively analysed. All standard corrosive tests use a metal
strip (copper or silver) with the specimen immersed in the oil that is to be tested. Some
methods also apply some layers of insulating paper around the metal and expose the oil
to oxygen for greater veracity with the field conditions [1]. The test results are either
pass or fail, based on the discolouration of the copper surface compared to a standard
colour scale Figure 2.16, which make the test subjective [57]. Some of these tests are
shown in Table 2.3.

Figure 2.16: Standard ASTM Copper strip corrosion colour chart

32
 Temperature Time Oxygen
Method Material Quantity
(◦ C) (h) (O2 )
100ml oil
DIN51353 Ag 100 18 Absent
1600mm2 Ag
250ml oil
ASTM D1275A Cu 140 19 Absent
300mm2 Cu
250ml oil
ASTM D1275B Cu 150 48 Absent
300mm2 Cu
15ml oil 540mm2
IEC62535 Cu 150 72 Limited
Cu 540mm2 paper

Table 2.3: Standard Oil Corrosivity Tests

The test conditions in ASTM D1275B are more severe stress ageing conditions than
ASTM D1275A, which results in an increased sensitivity in detecting corrosion species
in oils. Both ASTM D1275A and DIN51353 are recognised not to be very reliable, there-
fore a CIGRE working group developed the IEC 62535 standard, which uses a 30 mm
copper strip immersed in 15 ml of oil within a sealed vial which is kept at a temperature
of 150◦ C for 72 h [52]. Nevertheless, this test also has some issues; firstly, the oxygen
contained in the sample may lead to oxidation. Secondly, as temperature increases the
air above the oil in the sealed vial is pressurised which may lead to some unforeseen
chemical reactions. Finally it is difficult to separate any interactions between the oil
and the copper due to the vial being sealed [81]. These methods can be classified into
two categories, one being the use of oil and plain copper strips and the other being oil
and paper-wrapped copper samples. The latter are designed to evaluate the ability of
oil to form Cux S on the paper’s layers instead of detecting the reactivity of corrosive
species to copper surfaces. These tests are also known as Covered Conductor Deposition
(CCD) [1]. Studies have shown that insulation oil will pass one type of the test but not
the other and vice-versa [57]. It is recommended that both types of tests are performed
but the CCD test is considered more important.
The standard corrosion tests evaluate if the oil contains enough corrosive sulphur species
to render it corrosive under the test condition. It does not detect any specific corro-
sive species. As DBDS was deemed to generate Cux S deposits more easily, quantitative
methods were developed. A technique developed by a Mitsubishi research team using
gas chromatography-mass spectrometry together with a alumina-based solid phase ex-
traction has reportedly detected dibenzyl sulphide (DBS), bidenzyl (BiBZ) and dibenzyl
disulphide (DBDS) to a level of approximately 0.1 ppm [51]. There are other nonconven-
tional techniques used to quantify DBDS and other corrosive sulphur compounds [82].
A standard test method now exists to quantify how much DBDS there is in the oil [83].
There are no non-destructive techniques developed to date that are capable of either di-
rectly measuring or detecting Cux S deposits in the transformer. Destructive techniques
exist where kraft paper specimens are collected and analysed for copper and sulphur
deposits by X-ray micro analyser (XMA) [62]. The use of EDX or other comparable

33
methods in order to calculate quantitatively the sulphur content on the copper test strip
and insulation papers is considered crucial to eliminate false positives [84].
Several research groups have been trying to develop indirect techniques of either de-
tecting copper sulphides, by trying to measure its effects on the dielectric systems or
detecting by products of the corrosion process. Power factor, turns-ratio, insulation re-
sistance and dissolved gas analysis (DGA) tests have presently been unable to provide an
effective method of detecting corrosive sulphur damage in transformers [70, 57]. There
are some initial developments in techniques using some sort of frequency spectroscopic
analysis, i.e. frequency response analysis and dielectric frequency response, where it has
been possible to observe some effects due to Cux S deposits [72, 10, 74]. In addition,
it has been suggested that the first event of a complex and interlinked chemical pro-
cess leading to the deposition of Cux S in insulation paper is the copper dissolution [85].
Therefore, a copper dissolution monitoring technique could lead to an early detection
system.

2.2.4 Mitigation

To date there are no techniques available that can remove Cux S deposits or repair their
associated damage. Therefore, mitigation techniques are used to prevent further dam-
age. Passivation has been recommended by manufacturers and oil refineries since 2005,
in order to prevent a high failure rate in power system assets, such as transformers
and shunt reactors [54]. Passivation is a technical term used to define the formation
of non-permeable surface layers on metal [85]. Triazole-based passivators such as 1,2,3-
benzotriazole (BTA) and Irgamet 39T M (CIBA Speciality, Basel, Switzerland) have been
confirmed to suppress the deposition of Cux S and therefore prevent degradation of the
insulation paper, in both air and nitrogen atmospheres [86]. BTA is a white powder
that needs to be stirred to dissolve properly, with a recommended concentration of 30
ppm, which is the same as the Japanese standard for stream charging. Irgamet 39T M is
a high-viscosity liquid at room temperature, but it is more readily dissolved than BTA.
The recommended concentration by the manufacturer is of 100 ppm [86].
Historically passivation has been studied for over 30 years in Japan for streaming charg-
ing tendency, which is an electrostatic tension build up, as oil moves through the trans-
former, becoming positively charged in opposition to the negatively charged fixed struc-
tures [87]. Australia, Belgium and other countries which previously used BTA as an
antioxidant [54] are immune to the current corrosive sulphur problems in the power in-
dustry [88, 89]. The depletion of the BTA and Irgamet is known to accelerate in the
presence of oxygen and elevated temperatures [58]. The properties of the triazole-based
passivators is directly related to the complex layers and they only stop working when
they are damaged, either by elevated temperatures, thermal deterioration, oxidation or
some other chemical reaction that breaks down the complex layers. The protective layers
do not degrade with the depletion of the passivator. Hydroperoxides in high concentra-

34
tion have been found to decay the protective layers, although this protection seems to
persist with ageing. The protective layer is retained for a long time even with very severe
oil oxidation and without any passivator present in the oil. Furthermore the passivator
is stable to heat alone, but it decays under ultra-violet light [85]. It is expected that
once the complex layers have been damaged they cannot self-heal. This implies that the
duration of these passivators are directly proportional to the durability of these complex
layers on the copper surface [86].
Adding a passivator does not remove the corrosive compounds. That is the reason why
the term mitigation is used, as opposed to remediation, which is not achievable. It is
estimated that if the passivator drops below 25-30 mg kg−1 it will cause the oil to revert
to its original corrosive state [90]. High depletion rates of metal passivator are sometimes
seen in highly loaded power transformers where oxidation has already started [60, 90].
For originally passivated transformers this seems not to be a problem [60]. Although
the oil may revert to its corrosive state defined by BS EN 62535, the copper, which has
been passivated, is still protected against sulphur corrosion [90]. Insulation paper has
been shown to absorb significant quantities of passivator [91]. It is perceived that the
paper can effectively act as a reservoir [85].
Stray gassing arises from the addition of passivator in transformers, resulting in the gen-
eration of abnormal gases, normally termed “passivator stray gassing”. An increase in
the concentration of hydrogen (H2 ), carbon monoxide (CO) and carbon dioxide (CO2 )
is known to occur in the first seven days after adding the passivator to the insulation
oil. It is thought that the reason relates to residual oxygen present in the transformer
oil [54]. A possible issue with passivators is its ability to withstand high temperatures,
because DBDS only becomes corrosive at high temperatures, meaning that if DBDS is
corrosive then passivation will not work, if there is low temperature then the DBDS
is not corrosive making the passivator obsolete [59]. A positive side effect of adding
Irgamet 39 to either inhibited or uninhibited oil is an increase in thermal stability, as
it is shown to produce less sludge and acidity due to oxidation than non-passivated oils
[61]. In addition, passivators reduce both dissolved copper in oil and copper deposition
in paper [74].
An alternative mitigating technique that can be used is to replace the oil with non-
corrosive oil, which effectively reduces the concentration of corrosive sulphur compounds
by diluting the remnant corrosive oil with the new non-corrosive oil. Care must be taken
in making sure that the retro-filled oil has the same properties as the oil to be replaced,
otherwise the oil dielectric properties might change. The percentage of the corrosive
oil, remaining after retro filling is approximately 5-12 % depending on the quantity of
paper and other absorbing materials, as well as the shape of the transformer tank and
the location of the drainage valve [14, 60]. In order to reduce residual corrosive oil to a
minimum it is important to use a small extra quantity of oil to rinse the interior of the
tank and the active parts [14]. It is worth noting that this technique, although effec-
tive, it is the most costly choice [11, 57]. Retro-filling with natural ester fluid has been

35
shown experimentally to decrease the reactive sulphur compounds on a copper strip by
approximately 76 % and to extend the time of discoloration of copper by corrosion by a
multiple of 5 when compared with corrosive oil prior to mixing [84].
Another technique is oil depolarisation, which is an on-site technique that can be used
both on or off-load. It is based in a combination of solid reagents, chemicals and sor-
bents. This technique is capable of reducing DBDS content to a concentration lower than
5 mg kg−1 and removes passivation chemicals. This technique has been demonstrated
to have several other advantages. Filtering the insulation oil in order to diminish the
corrosive sulphur content in the oil, it also reduces the water content in the oil and
increases interfacial tension (IFT). Besides this, the loss angle tan(δ) seems to reduce
with corrosiveness of the oil [60]. Oil reclamation uses sorbents, which are chemical sub-
stances capable of collecting molecules from other substance by sorption, a mix of metal
oxides, mechanical filters and a degasser. The oil reclamation rigs have been widely
adopted as the sorbent can be reactivated several times before it needs to be replaced,
which means that the overall financial cost of this procedure drops significantly. Some
recent investigations of equipment with reclaimed oil have encountered silver corrosion
[92]. The cause as to why the reclamation rigs are transforming non-corrosive oil into
silver corrosive oil is still under investigation.

2.3 Summary

It was demonstrated that corrosive sulphur is not a new problem and that a new surge
in corrosive sulphur related faults started from the beginning of the second millennium.
DBDS was shown to be the main sulphur compound generating Cux S deposits in this
new surge of faults. There still is no agreed sulphur corrosion mechanism in power
transformers, although there are a few suggested models for the chemical reactions,
formation rates of Cux S deposits and deposition patterns in insulation paper. The
effectiveness of standard corrosion detection methods and other condition monitoring
technique, such as DGA, were also discussed. The current IEC 62535 corrosion detection
method and its limitation was analysed together with the currently available mitigation
techniques, i.e. passivation.
The analysis of the transformer design and its individual components demonstrated that
the winding insulation paper is normally what limits the transformer life expectancy and
it is the main system to be compromised by this type of corrosion. Furthermore, the
understanding of the lack of accessibility to the winding paper insulation, due to the
transformer tank and the insulation oil, explained the difficulty the electrical industry has
when developing a condition monitoring technique for Cux S deposits and the degradation
of Kraft paper.
It has now been explained the scope and motivation of the project and its context in
the power industry, and a review of the current understanding about corrosive sulphur

36
in the electrical power industry. This provides a platform to present the results of this
thesis.

37
Chapter 3

The Copper Sulphide Deposition

Process

When investigating a process using an empirical methodology it is important to establish

a standard process, which produces a repeatable and desired outcome. In this project,
the focus of the research was Cux S deposits in insulation Kraft paper. The standard
corrosion test, BS EN 62535, describes a technique that should produce paper samples
with Cux S deposits when the oil is corrosive, specially if the corrosive compound is
DBDS. However, the paper samples from this test are extremely small, and are not suit-
able for breakdown tests, impedance measurements and spectroscopic analyses. For this
research, the desired outcome from a standard sample production method is a uniform
deposition layer of Cux S on insulation Kraft paper with a sample size of approximately
90 mm square.
This chapter describes the development of a standard procedure that yields the desired
outcomes. This is followed by details of deviations from the standard procedure that
were performed in order to understand the relationship between certain variables and
the corrosion process. Qualitative, BS EN 62535, and quantitative, spectroscopic and
thermogravimetric techniques have been used to analyse the nature of the Cux S deposits
and the deposition patterns. The chapter then concludes by outlining a possible expla-
nation of the corrosion process that occurs in oil insulated power apparatus containing
copper.

3.1 Analytical Techniques

This project has used two types of analytical techniques to analyse samples of Kraft
paper. First is a qualitative analysis following the standard test method for detection
of potentially corrosive sulphur in used and unused insulating oil, i.e. BS EN 62535

38
[52]. According to the standard “the result is positive if the copper strip has one of
the following colours: graphite grey, dark brown or black”. The result is also positive
if the Kraft paper has copper sulphide deposits, which “appears metallic, from clearly
shiny to almost lustreless, often with a lead- or tin-like appearance. It can also have
the appearance of silver, brass or bronze. The metal-like surface of sulphide can have
an overlay of blue and/or purple due to interference phenomena. Other discolourations
(i.e. by-products of paper ageing and oil deterioration) shall not to be taken as copper
sulphide formation” [52].

3.1.1 Energy-Dispersive X-Ray Spectroscopy

As suggested in the BS EN 62535 EDX is used as an additional analytical technique.

EDX is an X-ray microanalytical technique that can provide information on the chemical
composition of a sample for elements with atomic number more than 3 [93]. It works
by applying a focussed electron beam on the sample, the electrons penetrate the sample
and interact with the atoms from which it is made. The size of the interaction volume
is directly proportional to the accelerating voltage of the electron beam and inversely
proportional to the mean atomic number (density) of the sample. The interaction of
the electron beam and the sample generates secondary electrons from the topmost part
of the interaction volume; backscattered electrons come from the top half of the in-
teraction volume; and X-rays from within the whole interaction volume in an attempt
to dissipate to excess energy provided by the electron beam. EDX spectroscopy takes
advantage of the latter method of energy dissipation, emission of X-rays. Two types
of X-rays are produced, Bremsstrahlung and characteristic X-rays. The first, also re-
ferred as background X-rays, are produced by the slowing down of the beam electrons
by the electric field surrounding the nuclei of the atoms in the sample. Characteristic
X-rays are produced by the atomic electron transitions between the electron shells of
that atomic element. Each different atomic element emits x-rays at a specific energy,
because of its unique electronic configuration interacting with the beam electrons, hence
the name characteristic X-Rays. Therefore, the energies of the characteristic X-rays
allow the elements making up the sample to be identified, while the intensities of the
Characteristic X-ray peaks allow the concentrations of the elements to be quantified.
The quantification can be either fully standardised quantitative analysis or standardless
quantitative analysis, also known as semi-quantitative analysis. In standardless quanti-
tative analysis the spectra collected are compared with data collected from standards in
the factory which is stored with the system software, normally the spectra for a single
reference material is collected to verify the quantitative system is working. In a fully
standardized quantitative analysis, the spectra from the standards are collected on the
same instrument as the spectra from the sample being analysed [94].
EDX detectors are commonly used with scanning electron microscope (SEM) applica-

39
tions for their flexibility and performance [93]. The instrument used for this experiment
was a JEOL JSM5910 SEM. Copper samples obtained under the same set of conditions
were analysed together using a multiple pin sample holder. A limited number of sam-
ples was allowed on the sample holder therefore the longest series had to be split in to
two subsequent vacuum sessions. The accelerating voltage was set to 20 kV while the
spot size parameter was varied from 35 for imaging to 40 for elemental analysis. The
working distance (WD) was kept constant at 10 mm as required for the EDX analyses
while contrast and brightness were adjusted manually. EDX quantification was made
using the NSS software suite. Although all elements present were identified only C (K
line), S (K line) and Cu (L line) were quantified. Carbon was kept as a reference. Semi-
quantitative analysis was performed; the single element calibration was performed using
a Co standard disc fitted onto the sample holder.
An environmental scanning electron microscope (ESEM) was used for paper samples,
using a Philips XL-30 instrument. Paper samples obtained under the same set of con-
ditions were analysed together using a multiple pin sample holder. No sample coating
was necessary since the environmental mode was always kept on. A pressure limited
aperture (PLA) was not used. Vacuum in the chamber was considered to be estab-
lished after six flushes of water vapour to replace all air in the analysis chamber. The
WD was kept constant at 10.000 mm as required for the EDX analyses while contrast
and brightness were adjusted manually. In case of lower contamination samples the
spot size and the partial pressure of water (pH2O ) was decreased to minimize charging
(pH2O = 93 − 133 : P a). EDX quantification was carried using the same process. The
depth penetration of this technique is estimated to be in the range of 0.5-5 µm. All
graphs show the atomic percentage of three elements considered disregarding all other
impurities detected.

3.1.2 X-ray Photoelectron Spectroscopy

X-ray Photoelectron Spectroscopy (XPS) was also used to provide information about
the surface deposits. This technique uses X-rays to excite the electrons of the sample
atoms and if their binding energy is lower than the X-ray energy, they will be emit-
ted from the parent atom as a photoelectron. Only the photoelectrons at the extreme
outer surface (10-20 nm) can escape the sample surface, making this a surface analysis
technique. Each element produces a characteristic set of XPS peaks at specific binding
energy values that directly identify each element that exists in or on the surface of the
material being analysed. The number of detected electrons in each of the characteristic
peaks is directly related to the amount of element within the XPS sampling volume [95].
XPS was carried out using a Theta Probe with a monochromatic Al K α X-ray source
of 1486.6 eV. Samples were placed in the pre-vacuum chamber overnight and analysed
the following day. A flood gun was used to minimize charging effects. Survey spectra

40
(3 repetitions) from 1350 to 10 eV (binding energy, B.E.) were collected on the inner
side of each paper sample. The pass energy and energy steps were 200 and 1.000 eV
respectively. High-resolution scans (10 repetitions) for quantitation were made on C
1s, S 2p and Cu 2p peaks. In this case, the pass energy and energy steps were 40 and
0.100 eV respectively. Data processing was performed using Casa XPS software suite
and Shirley background was used. The depth penetration of this technique is estimated
to be in the range of 10-20 nm. All graphs show the atomic percentage of the three
elements considered disregarding all other impurities detected.

3.1.3 Thermogavimetric Analysis

Thermogavimetric analysis (TGA) is a method in which physical and chemical proper-

ties of materials are measured by mass changes, as a function of increasing temperatures
at a constant rate or as a function of time with constant temperature. TGA can provide
information about physical phenomena, such as second-order phase transitions, chemical
phenomena (e.g. decomposition), and solid-gas reactions (e.g., oxidation) [96]. In the
case of this work, TGA is used to directly determine the quantity of inorganic products
in the Kraft paper, which should mainly be copper-sulphur products.
The equipment used is Perkin Elmer Pyris 1 TGA, using the low range data collection
mode. The temperature calibrated using a multi-point calibration system using four
Curie points (Alumel(154.2 ◦ C ), Nickel (355.3 ◦ C ), Perkalloy (596 ◦ C ), Iron(780 ◦ C ))
with an increase rate of 10◦ C min−1 . The balance was calibrated using a single-point
calibration with a reference 100 mg mass. The gas flow rate was set to 20 ml min−1 and
platinum samples pans were used for all measurements.

3.2 Development of Sample Production Method

As discussed in the Chapter 2, DBDS is the main corrosive sulphur compound in gen-
erating Cux S deposits on paper. Therefore it is used in this sample production method,
DBDS (Sigma Aldrich, 98 %) was added to the mineral oil to make it corrosive. The
mineral oil used for the experiment was Nynas Nytro Gemini X an inhibited mineral
oil with a maximum total sulphur content of 0.05 %, no detectable DBDS, no declared
metal passivator and classified as not corrosive according to BS EN 62535 [97]. This
oil was chosen because it it has a low impurity level and it is used in industry. Nat-
urally corrosive oil, Nynas 10GBN, was used as a comparison only. The reason being
naturally corrosive oil is of limited availability and it has a higher impurity level with
multiple corrosive sulphur species and other unknown compounds, which introduce more
uncertainty in an already complex corrosion process. Although Nytro Gemini X is less

41
representative of field data, it provides a less complex matrix making it possible to in-
vestigate the effects of single variables to the whole corrosion process more accurately.
The first objective was to produce oil that generates Cux S deposits on paper using the
standard corrosive test procedure, BS EN 62535, as a guide line. The general method
used for the experiments was as follows:

1. Transfer the oils being used into a beaker, quantity depends on number of samples
being produced, a margin between 25 % to 50 % of the volume required should be
added.

2. Add DBDS to the oil, quantity depends on concentration(ppm) desired.

3. Use a stirrer-hot plate set to (60 ◦ C -75 ◦ C) with magnetic stirrer, to dissolve
DBDS in the oil.

4. Remove the extra paper layers from conductor using tweezers or gloves to avoid
contamination of paper samples.

5. Place the copper samples inside the vials and add 15 ml of the prepared oil with
dissolved DBDS.

6. Seal the vial using polytetrafluoroethylene (PTFE)-faced silicone septum and alu-
minium cap, both of which are sulphur free.

7. Place in fan oven at 150◦ C for 72 h.

8. Remove vial from oven, extract conductor, separate paper from copper and de-
grease both by washing twice with cyclohexane.

9. Visually examine for signs of Cux S deposition, using the colour scales as defined
in BS EN 6235.

This approach is slightly different from that of the standard because the Kraft paper
is wound in a half lap configuration, nevertheless this does not affect the test results.
Adding DBDS to the oil (Steps 2-3), ensures the presence of a corrosive sulphur specie.
The reason for using a paper wrapped copper conductor was that this is the same pri-
mary insulation arrangement found in a large proportion of oil paper insulated high
voltage equipment. Table 3.1 summarises the corrosion tests performed.
An experiment with no DBDS was performed to verify that Gemini X is not corro-
sive and also to act as a control sample, Figure 3.1. This initial test also established
qualitatively the appearance of a thermally aged noncorrosive sample of Kraft paper
and therefore served as a comparison for other samples. Repeat tests were performed
because it has been shown that in some cases the deposition does not always occur, as
it is a probabilistic event [98].

42
 BS EN 6235
Experiment Oil DBDS (ppm) Figure
Paper Copper
1 Gemini X 0 Negative Negative 3.1
2 Gemini X 100 Negative Negative 3.2 & 3.3
3 Gemini X 1000 Positive Positive 3.4
4 10GBN 0 Negative Positive 3.5

Table 3.1: Summary of standard corrosion test results for experiment 1 to 4. “Neg-
ative” means the sample does not show signs of corrosion and “Positive” means the
opposite

(a) (b)

Figure 3.1: Experiment 1, Gemini X with copper samples wrapped in 1 paper layer
tested according to BS EN 62535. Figure 3.1(a) shows oil samples after ageing. 3.1(b)
shows the unwrapped paper sample and copper from each respective vial

Nytro GeminiX with 100 ppm of DBDS corrosion test results for both paper and copper
samples are negative, meaning there was no visible tarnish of the copper surface or
metallic appearance on the paper surface. The test involved 8 samples, four 1 paper
layer samples 3.2(b) and four 4 paper layers samples 3.3(b). The results indicate that
with only 100 ppm of DBDS in Gemini X, the BS EN 62535 stress ageing conditions
are not enough to generate any sign of corrosion even though the test oil is known to
contain corrosive sulphur species.

(a) (b)

Figure 3.2: Experiment 2, Gemini X with 100ppm of DBDS and copper sample
wrapped in 1 paper layer tested according to BS EN 62535. Figure 3.2(a) shows oil
samples after ageing. Figure 3.2(b) shows the unwrapped paper sample and copper
from each respective vial

43
 (a) (b)

Figure 3.3: Experiment 2, Gemini X with 100ppm of DBDS and copper sample
wrapped in 4 paper layers tested according to BS EN 62535. Figure 3.3(a) shows oil
samples after ageing. Figure 3.3(b) shows the unwrapped paper sample and copper
from each respective vial

It should be noted that for this research, the purpose of this experiment is to obtain a
homogeneous Cux S deposition across Kraft paper samples and not necessarily find the
minimum DBDS concentration to produce positive results. Therefore, following from
the previous result the concentration of DBDS was increased by tenfold to 1000 ppm,
ensuring the oil is corrosive. Figure 3.4(b) demonstrates the deposits on the insulation
paper, which appear to be, in qualitative terms, a homogeneous Cux S deposition across
a Kraft paper sample. Only half of the insulation paper has deposits because the paper
is wound in a half lap configuration, where half of the paper is overlapped with the next
turn, meaning only half of the paper is in direct contact with the copper.

(a) (b)

Figure 3.4: Experiment 3, Gemini X with 1000ppm of DBDS and copper sample
wrapped in 1 paper layer tested according to BS EN 62535. Figure 3.2(a) shows oil
samples after ageing. Figure 3.2(b) shows the unwrapped paper sample

As expected Figure 3.5(b) shows the naturally corrosive oil aged copper samples have
a greyish metallic surface colour, which according to the standard corrosion test BS
EN 62535 [52] is a positive result. Although looking at the paper samples the result
is negative. Suggesting that the ageing conditions of the standard test BS EN 62535
are insufficient to produce Cux S deposits on Kraft paper for this naturally corrosive oil,
10GBN.

44
 (a) (b)

Figure 3.5: Experiment 4, 10GBN with copper sample wrapped in 1 paper layer
tested according to BS EN 62535. Figure 3.5(a) shows oil samples after ageing. Figure
3.5(b) shows the unwrapped paper sample and copper from each respective vial.

The aged oils from the experiments 1-4, figure 3.1 to 3.4 parts (a), appear to have no
or little effect on the translucency of the oil. The natural corrosive oil has a lighter
appearance in general, as shown in Figure 3.5(a), but also demonstrates to be more
unstable as in one sample vial the oil has a black appearance .
Based on the above preliminary results an initial hypothesis was formulated that by
scaling up the standard corrosive test procedure, BS EN 62535, with Nytro GeminiX oil
containing 1000 ppm of DBDS it should be possible to produce corroded samples with
Cux S deposits on Kraft paper. The variables controlled in this experiment are temper-
ature, DBDS concentration, sample geometry and ageing time. The main objective of
this experiment is to produce viable samples for analytical techniques; therefore, the
variables are modified with the purpose of maximising a uniform deposition of Cux S
whilst keeping the paper ageing to a minimum. The reason for keeping the ageing to a
minimum is to try to reduce any other paper degradation effects. As much as possible
the BS EN 62535 experiment was used as a reference, for example the copper to oil
ratio, 31.5 mm2 ml−1 , was maintained. The original vial was unsuitable due to the size
of the enlarged copper sample, hence, a crystallising dish of 140 mm diameter was used.
The copper used was a grade C106 with a thickness of 0.5 mm with an area of 1800
mm2 . The paper used was a standard electrical insulation Kraft paper with thickness
of 0.07 mm. Just before the experiment started, the copper surface was cleaned using
a polishing wheel and cyclohexane to remove any surface contaminates. The paper was
always handled with gloves to avoid contamination and was fixed to the copper either
using cable ties or nylon thread.
The initial experiment, aged at 150 ◦ C in a fan oven, failed because it generated an ex-
cess amount of fumes due to the increased surface area of oil and the unsealed containers.
A series of trial and error experiments followed, as shown in Table 3.2.

45
 Time Temperature DBDS
Experiment Sample Result
(Days) (◦ C) (ppm)
Failed: Excess amount
5 1 150 1000 flat
of oil fumes generated
Partial Cux S deposits
6 25 110 1000 flat
on paper and conductor
7
Partial Cux S deposits
(Continued 9 130 1000 flat
on paper and conductor
from 6)
Paper completely
8
burned no visible Cux S
(Continued 10 140 1000 flat
deposits and very
from 7)
brittle
No Cux S deposition on
9 3 140 1000 flat
paper
Clear Cux S flakes
10 3 140 2000 flat covering partially the
paper
Cux S deposition across
11 7 140 2000 curved
the whole paper
Uniform Cux S
12 1-8 140 2000 curved deposition is reached at
the 7th day
Cux S deposition occurs
at all concentrations,
100 - 1000/2000 ppm
13 7 140 curved
10000 appears to have the
most uniform Cux S
deposition

Table 3.2: Experimental iterations in developing enlarged Cux S paper samples using
a trial and error methodology

The temperature was lowered and the ageing time increased (Experiments 6-9), but this
still did not produce paper samples with homogeneous Cux S deposition. The results
clearly demonstrate the disadvantages of having a very prolonged ageing time, where in
Experiment 8 with a total of 44 days in the oven the paper samples were completely
carbonised and brittle. Any analysis of these samples would relate more to the paper
degradation than to the Cux S deposits, hence the necessity to find an optimal balance
between temperature, time and DBDS where the Kraft paper samples are not exces-
sively thermally degraded.

46
DBDS concentration was increased to 2000 ppm in experiment 10, which showed an in-
creased Cux S deposition but not uniform. This result suggests that DBDS concentration
is related to the rate of Cux S deposition, therefore increasing the concentration decreases
the ageing time. In experiment 11 the DBDS concentration was kept at 2000 ppm and
the ageing time was increased to 7 days. The surface contact between copper and paper
was also investigated, in experiment 11, by bending copper plates to produce a curved
plate with a roughly constant radius, giving rise to concave and convex surfaces. This
curvature in the copper plate allows the convex side to have a greater contact area with
the Kraft paper, while not restricting oil flow. The curved geometry as show in Figure
3.6 was found to be essential for generating uniform Cux S deposition on the paper and
demonstrates that the surface contact area between copper and paper is important in
the deposition step of the corrosion process.

Figure 3.6: Sample modification from flat to curved copper sheet

With a working procedure, the experiments were then undertaken to pinpoint the ideal
duration of the experiment and DBDS concentration. Experiment 12, as shown in Figure
3.7, investigates the ageing time required to produce uniform Cux S deposition across the
paper samples with a DBDS concentration of 2000 ppm and an ageing temperature of
140 ◦ C . Deposition starts after 48 hours (2 days) it then grows in patches. On the 7th
day, the Cux S deposition appears to have reached homogeneity. On the 8th day, it starts
to grow in patches, possibly growing further layers of Cux S and repeating the observed
cycle.
All observations were initially qualitative following the standard corrosion test, BS
EN 62535. However, a SEM-EDX was used to quantify the amount of copper and
sulphur within the surface of the paper. Figure 3.8 shows quantitatively that the copper
concentration stabilised in the paper between the 6 th and the 7 th day. Carbon, the main
constituent of paper, as expected had a very high initial concentration, which decreased
as the copper-sulphur products were deposited on the paper surface. The stoichiometry
of copper sulphide appears to be more complex than simple Cu2 S, sulphur relative
concentration still appears to have an increasing trend over the eight days of ageing. By
the 8th , the relative ratio of copper to sulphur was approximately 3:1.

47
 Figure 3.7: Experiment 12, enlarged samples aged in corrosive oil for different times,
where green: homogeneous disposition; orange: patchy disposition

Figure 3.8: EDX data for Experiment 12 Kraft paper samples, the atomic percentage
is normalised to the sum of sulphur, carbon and copper

Parallel to the EDX, XPS was used to analyse the same samples. The reason for per-
forming this additional spectroscopic analysis was to decrease the penetration depth of
assessment, which for XPS is 10 - 20 nm as opposed to the 0.5 - 5 µm of the EDX.
Therefore EDX data provides information about the bulk of the paper insulation, while
the XPS data is more related to the surface features. The XPS data, Figure 3.9, show
a different trend compared to the EDX data, the stoichiometry appears to be reversed,
with more sulphur near to the surface of the paper than copper. The quantity of copper
and sulphur is also very different compared to EDX, unlike the behaviour observed in
the paper bulk, neither copper or sulphur relative concentrations appear to increase over

48
the duration of the experiment.

Figure 3.9: XPS quantitation of C, S and Cu from the inner surface of paper samples
aged in oil containing 2000 ppm of DBDS under an atmosphere of air at 140 ◦ C over 8
days

XPS and EDX data have been normalised to sum of the three elements shown. To
obtain an absolute value of copper-sulphur products in the paper sample, TGA was
performed. Figure 3.10 clearly demonstrates that the longer the ageing period the more
copper-sulphur products are found in the samples. In addition, there are some periods
where there is a considerable jump in contamination, e.g. from the 5th to 6th day and
from the 10th to 13th day. It is also interesting to observe the different degradation
stages in air that occur at 200 ◦ C , 300 ◦ C , 350 ◦ C , 500 ◦ C and 700 ◦ C .
The initial mass loss is assumed to be due to moisture (50◦ C - 300◦ C ) and other volatile
impurities in the paper, the major degradation step, mass loss from 40 - 70 %, is con-
sidered to be the paper cellulose decomposing (300◦ C to 350◦ C ). Between 350◦ C and
700◦ C the by-products of the cellulose degradation degrade further into gas products.
Performing TGA measurements in air is expected to increase the mass initially, due to
the formation of oxides, and then, at a higher temperature, to increase the rate of mass
loss through combustion.
To confirm the Kraft paper samples residues from the TGA were in fact copper deposits,
EDX analysis was perform. The residues of a Kraft paper sample, aged for seven days
at 140 ◦ C with 2000 ppm of DBDS under air after thermogavimetric analysis under air
atmosphere shown in Figure 3.11, were analysed using EDX analysis as a representative
sample, shown in Table 3.3. The residues are mainly composed of copper and oxygen,
the sulphur is minimal as copper sulphide decomposes at approximately 500 ◦ C , carbon
is also present due to charring.
49
 100
90 13 Days

10 Days
80
8 Days
70 7 Days

Weight (%) 60 6 Days

5 Days
50
4 Days
40 3 Days

30 2 Days

1 Days
20
10
0
0 100 200 300 400 500 600 700 800 900 1000
Temperature (°C)

Figure 3.10: TGA in air atmosphere at rate of 10 ◦ C /minute data for Kraft paper
samples aged at 140 ◦ C in Nytro GeminiX with 2000 ppm of DBDS under air, from 2
to 13 days

Site Carbon (%) Oxygen (%) Sulphur (%) Copper (%)

1 15 % 27 % 2% 57 %
2 15 % 26 % 3% 56 %
3 10 % 28 % 2% 60 %
Average 13 27 2 58
SD (±) 2.7 1.1 0.5 1.9

Table 3.3: EDX data for three individual sites on the TGA Kraft paper sample
residues

Figure 3.11: ESEM picture of TGA Kraft paper sample residues.

50
Experiment 13, as shown in Figure 3.12, investigates the ideal concentration of DBDS
in Nytro GeminiX, for 7 days at 140 ◦ C . In qualitative terms, BS EN 62535, all Kraft
paper samples demonstrate traces of Cux S deposition although with different levels of
homogeneity. The ideal concentration appears to be between 1000 ppm and 2000 ppm,
below 1000 ppm the Cux S deposition is still not homogeneous and above 2000 ppm it
becomes patchy and starts flaking off, similar to the samples after the 7th day of ageing,
which emphasises the relationship between DBDS concentration and ageing time.

Figure 3.12: Experiment 13, enlarged samples aged in corrosive oil with different
DBDS concentrations, where green: homogeneous disposition; orange: patchy disposi-
tion

Figure 3.13 shows EDX data for Experiment 13 and demonstrates as expected that
Kraft paper samples aged with 0 ppm and 100 ppm of DBDS have similar results, as the
ageing time and temperature is not enough for 100 ppm of DBDS to generate noticeable
effects. DBDS concentrations of 500 ppm and 1000 ppm have similar results suggesting
a possible saturation is reached, but the concentrations from 2000 ppm to 10,000 ppm
demonstrates another incremental deposition step. The copper to sulphur ratio appears
stable from 500 ppm onwards at approximately 2 : 1, which differs from the results
presented in Figure 3.8. EDX is an elemental analyis technique and it is not able to
distinguish sulphur from corrosive sulphur compounds in oil and the Cux S deposits.
Therefore it is only possible to infer that the changing ratio of copper to sulphur, as
shown in Figure 3.8, is because the sample contains a significant amount of copper-
sulphur compounds, which then decompose into Cux S and other by-products.
An increasing trend of sulphur and copper is also visible in the XPS data, as shown in
Figure 3.14, although the rate of increase is much lower than that seen for the EDX data
(Figure 3.13). The top few nanometres of the paper sample show a maximum sulphur
contamination of 25 %, which is very similar to the EDX data of 22 %. Although
when observing the maximum copper contamination of the Kraft paper the XPS and
EDX data differ significantly, with 13 % on the surface (XPS) and 46 % in the bulk
(EDX). This can be explained by considering the two copper-sulphur product formation

51
 Figure 3.13: EDX data for Experiment 13 Kraft paper samples(140 ◦ C for 7 days in
air atmosphere with varying DBDS concentration), the atomic percentage is normalised
to the sum of sulphur, carbon and copper

processes that are occurring, one at the bulk where the concentration can reach high
levels such as 70 % with a copper to sulphur stoichiometry of approximately 2:1. The
other copper-sulphur product formation process occurs at the surface of the paper where
the concentration level is much lower, between 5 - 30 %, and the stoichiometry appears
to be different with more sulphur present than copper.

52
 Figure 3.14: XPS quantitation of C, S and Cu from the inner surface of Experiment
13 Kraft paper samples(140 ◦ C for 7 days in air atmosphere with varying DBDS con-
centration), the atomic percentage is normalised to the sum of sulphur, carbon and
copper

The TGA data shown in Figure 3.15 demonstrates that the higher the DBDS concentra-
tion the higher the copper-sulphur products. Similar to obtained TGA data for ageing
time (Figure 3.10), there are high DBDS concentration levels, e.g. 2000 ppm and 5000
ppm, that produce a significantly higher increment in copper-sulphur products in the
paper compared to the lower concentrations. This data support the hypothesis that a
concentration of DBDS higher than 1000 ppm should be used in sample production in
order to maximise the copper-sulphur products on and in the Kraft paper.

53
 100
10 Days
90 13 Days

80 8 Days

7 Days
70
6 Days
Weight (%) 60 5 Days

50 4 Days

3 Days
40
2 Days
30
1 Days

20
10
0
0 100 200 300 400 500 600 700 800 900 1000
Temperature (°C)

Figure 3.15: TGA in air atmosphere at rate of 10 ◦ C /minute of Kraft paper samples
aged at 140 ◦ C for 7 days in GeminiX with 0 to 5000 ppm of DBDS under air

Further tests could be performed in terms of variations to Experiment 12 (Figure 3.8)

with different DBDS concentrations, and variations to Experiment 13 (Figure 3.13) with
different ageing times. It is predicted that the higher the concentration of DBDS the
less time it will need to react. The reason for not increasing the DBDS concentration
as much as possible to decrease the ageing time is that at higher concentrations of 2000
ppm of DBDS it is not considered a realistic condition for field transformers. In addition,
increasing the concentration of DBDS above 2000 ppm might alter the corrosion process
away from that observed in the field. Although DBDS concentration in a transformer
is typically up to 350 ppm [1], a higher level of DBDS concentration is used because
the ageing time required to generate copper sulphide deposits at lower concentrations is
too time consuming for this research. Furthermore, the degradation of the paper due to
prolonged ageing periods will contribute significantly to loss of degree of polymerisation,
which might mask some of the effects of copper sulphide deposits on the paper.
The standard sample production method requires that samples have a homogeneous
Cux S distribution. To analyse quantitatively the homogeneity three sites were selected
in the same sample, as shown in Figure 3.16 each site is analysed using EDX. Table 3.4
details the variation of carbon, sulphur and copper from one site to the other, which for
copper and sulphur is ± 1 %. For every measurement, the samples were analysed in this
way and the variations are displayed in the EDX figures using error bars for every data
point.

54
 Figure 3.16: SEM picture of paper sample aged for 7 days at 140 ◦ C with 2000 ppm
of DBDS. The rectangular areas are analysed using EDX.

Site Carbon (%) Sulphur (%) Copper (%)

1 35 21 44
2 37 21 42
3 34 22 44
Average 35 21 43
SD (±) 1.59 0.67 0.98

Table 3.4: EDX data for three sites on the same paper sample aged for 7 days at 140
◦
C with 2000 ppm of DBDS.

Now that the homogeneity of deposition across the sample has been analysed, it is
necessary to evaluate the similarity between replicate samples as shown in Figure 3.17.
Each sample was produced at different times, for each experiment the same procedure
was followed. Qualitatively they have very similar depositions with a metallic greenish-
grey surface colour. Quantitatively, using EDX, there is a variation of ± 2 %, as shown
in Figure 3.18.

(a) (b)

Figure 3.17: Aged Kraft paper samples with Cux S uniform deposition.

55
 Figure 3.18: EDX sulphur data for four replicate Kraft paper samples aged for 7 days
at 140 ◦ C with 2000 ppm of DBDS. The atomic percentage is normalised to the sum of
sulphur, carbon and copper.

Using both qualitative and quantitative analysis methods the following standard proce-
dure has been developed to produce 9 mm square homogeneous Cux S deposition Kraft
paper samples:

1. Use a clean crystallising dish of 140 mm diameter and 70 mm height to pour 512
ml (453 g) of Nytro Gemini X mineral oil.

2. Add 2000 ppm (906 mg) of DBDS to the Nytro Gemini X mineral oil.

3. Use a stirrer-hot plate and a clean magnetic bar to mix the oil with the DBDS at
a temperature of 60 ◦ C ± 10◦ C until all DBDS crystals have dissolved, as shown
in Figure 3.19, which takes approximately 20 min.

4. Use a clean 9 mm square copper plate with a curved surface as shown in Figure
3.6. After cleaning the copper with the polishing wheel and cyclohexane, gloves
must be worn when handling the copper plate.

5. Use Kraft paper to cover the copper plate’s outer curved surface and use two
plastic zip ties or nylon thread to maintain surface contact area of the paper and
the copper, as shown in Figure 3.6.

6. Immerse the paper wrapped copper plate in oil.

7. Heat the oven up to 140 ◦ C.

8. Age the paper in the oven for 7 days.

56
 Figure 3.19: Dissolving DBDS in oil

3.3 Modifications to the Sample Production Method

To investigate the Cux S formation further, the standard sample production variables
were modified. The first modification was to investigate the effects of the atmosphere;
the experiment was adjusted to remove all the oxygen and moisture in order to see if the
presence of oxygen has a retardant effect on the formation of Cux S as it is competing with
the sulphur for copper. To do this the whole system was first degassed, paper copper
and oil, as shown in Figure 3.20(a) for a minimum period of 24 hours in full vacuum at a
rotary pump pressure. Afterwards the samples were heated to the standard ageing tem-
perature of 140 ◦ C under a positive pressure, nitrogen atmosphere and the experiment
carried out as normal. The experiment was performed under several different DBDS
concentrations, as shown in Figure 3.21, because a different atmosphere may change
the DBDS concentration required to form Cux S deposits. The results demonstrate two
important characteristics of the corrosion process, one being that the presence of oxygen
has negligible effects on Cux S formation on the surface of the copper, as shown in Figure
3.20(b). The second is that oxygen is detrimental to the deposition of Cux S on paper
Figure 3.20(c). These results contradict the current perception of the corrosion sulphur
process [51], where it is suggested that copper sulphide is formed in the paper instead
of the copper conductor.
EDX and XPS measurement data demonstrate that under a nitrogen atmosphere no
copper-sulphur product formation process is passed into or onto the paper, as shown in
Figure 3.22 and 3.23. It is suggested in literature that the presence of oxygen generates
carboxylic acids, which attack the copper displacing it into the oil and therefore creating
deposits on paper [53]. These results seem to support this idea, but it is hypothesised
that the carboxylic acids displace the Cux S products on the surface of the copper as well
as creating copper-sulphur complexes, which dissolve in the oil. This hypothesis suggests

57
 (a) (b) (c)

Figure 3.20: Copper and paper samples after ageing procedure under nitrogen with
2000 ppm of DBDS at 140 ◦ C for 7 days, (a) Sample before ageing, (b) Aged copper
plate, (c) aged kraft paper

Figure 3.21: Standard enlarged Cux S sample experiment, but aged under nitrogen
atmosphere

that two copper-sulphur product deposition process occur in parallel, with possibly two
different stoichiometries of Cux S.

Figure 3.22: EDX data for standard enlarged Cux S Kraft paper sample, but aged
under a nitrogen atmosphere

58
 Figure 3.23: XPS data for standard enlarged Cux S Kraft paper sample, but aged
under a nitrogen atmosphere

The hypothesis that paper enhances the local concentration of DBDS and therefore
increases overall corrosion process at the copper paper interface is validated from results
shown in Figure 3.24. It can be concluded that the paper increases the rate of corrosion
by comparing the areas where the paper is in contact with the copper as opposed to the
areas where there is no such contact, like the convex and concave sides of the copper
plate respectively, and the middle and the outer part of the inner side of the insulation
paper respectively.

Figure 3.24: Corroded samples using the procedure described in the Experiment
Method

59
To further validate the hypothesis that Kraft paper increases local DBDS concentration,
an experiment was developed. The experiment uses the standard procedure to produce
samples, but the paper covering the copper plate has a hole, as shown in Figure 3.25(a).
Results shown in Figure 3.25(b) and (c) clearly validate the hypothesis; the copper,
which was not in contact with the paper, would pass a standard corrosion test, BS
EN 62535. The copper paper interface seems to create a “preferred” corrosion site,
which results in other sites not being corroded for the given conditions of the standard
procedure. Given a higher concentration or more time the other sites will eventually
become corroded, this is demonstrated in the Figure 3.21 where the concave side of the
copper plate is only corroded when the DBDS concentration is 5000 ppm or higher.

Figure 3.25: Enlarged samples experiment with whole in the Kraft paper, (a) before,
(b) after ageing process, and (c) visual inspection

3.4 Summary

This chapter has detailed the development of a sample production method capable of
generating 90 mm square Kraft paper samples with homogeneous Cux S deposition, with
a 1 % variation (Table 3.4), and a good reproducibility of ±2% (Figure 3.18). In the
process of developing the sample production method, it was found that increase of
either DBDS concentration or temperature has similar effects in increasing the rate of
corrosion. From observation it is clear that the paper enhances the local concentration
of DBDS making paper-copper interfaces ‘preferred’ corrosion sites. By conducting the
experiment under dry nitrogen it can be concluded that oxygen displaces Cux S from the
copper surface to the Kraft paper, and increases the available corrosion sites.
Based on the results obtained, it is possible to develop a hypothesis to explain copper
sulphide deposition using multiple Cux S formation mechanisms, one for reaction at the
copper surface and another one for reactions in the paper bulk. This is summarized in
Figure 3.26.

60
 Figure 3.26: Schematic representation of the parallel mechanisms proposed for: sur-
face contamination (1) and bulk contamination of insulating paper (2) [99]

A complete summary of this chapter has also been published in the Journal of Corrosion
Science[99].

61
Chapter 4

Monitoring the Corrosion Process

Having established an understanding of the development of the corrosion process, it is

important to investigate techniques that are able to detect and possibly monitor the
progress of corrosion. The ultimate condition monitoring tool would be non-destructive,
able to detect and quantify the degradation caused to the insulation media, while main-
taining the asset on-line with minimal expertise required and little financial investment.
The current state-of-the-art is very limited with respect to monitoring Cux S formation
in electrical apparatus. There is a standard for detecting DBDS [83], but as DBDS
will become less likely to be found in the transformers, as has happened with other
classes of compounds in the past [100], it is important to investigate tools with generic
applicability. Therefore, the objective of this research project concentrates on assess-
ing the feasibility of various techniques, it also focuses on developing laboratory scale
experiments to investigate the deterministic nature of the data collected from various
analytical techniques.
This chapter focuses on two techniques: one being an oil based analytical technique
known as x-ray fluorescence (XRF). The second technique involves analysing Kraft pa-
per using frequency dielectric spectroscopy (FDS), where it investigates how the forma-
tion Cux S in Kraft paper changes its dielectric properties. The chapter concludes with
a proposal of the development of the techniques to the point where they can be applied
in the field.

4.1 Analytical Techniques

This chapter presents data from multiple analytical techniques, compares results and
then proposes an approach for monitoring corrosion. SEM-EDX is used again to anal-
yse the elemental composition of copper samples surfaces. In addition, XRF and FDS,
are introduced in this chapter. XRF is used to analyse the elemental composition of oil

62
samples and FDS is used to analyse the polarizability of Cux S deposits.

4.1.1 X-ray Fluorescence

XRF is an elemental technique similar to SEM-EDX, although it bombards the sample

with high-energy X-rays as opposed to electrons. The term fluorescence by itself is the
phenomenon that results in the emission of a radiation (normally at a lower energy) due
to the absorption of another specific energy radiation. XRF refers to the characteristic
emission of secondary radiation by matter that has been excited by a highly energetic
primary radiation. The elemental identification technique is similar to the SEM-EDX,
which is based on the fact that each different atom emits radiations at a specific and
characteristic wavelength, or energy as a consequence of its unique electronic configura-
tion [101].
The instrument used in the experiments was a Philips Magix-Pro Wavelength Disper-
sive X-ray Fluorescence Spectrometer fitted with a 4 kW Rh-anode, end-window X-ray
tube. The count time was set to 40 s on the element peaks and 40 s on the background.
Both copper and sulphur were quantified using Kα transitions (8.047 keV and 2.308
keV respectively). The excitation power was 60 kV and 66 mA for copper and 24 kV
and 100 mA for sulphur. All samples were measured in He atmosphere. Being liquid
samples, disposable 40 mm diameter XRF cells (Fluxana GmbH & Co. KG) were used
once fitted with 4 µm polypropylene film, as shown in Figure 4.1.

Figure 4.1: XRF stainless steel liquid sample holder alongside a disposable plastic
cup (40 mm diameter) with lid and polypropylene film base filled with oil and ready to
be analysed

4.1.2 Frequency Spectroscopy Analysis

Lumped circuit methods where the dielectric sample is modelled as a parallel circuit of
capacitors and resistors, can be applied in broadband dielectric analysis techniques for
frequencies between 10−6 to 107 Hz [102]. Experimentally FDS is a pair of electrodes
between a sample where a signal of frequency(f ) is applied and current is then measured.
63
The variation of the current magnitude and phase, with respect to the input signal, will
depend on several properties of the dielectric material such as bulk resistance, surface
charge and polarisation effects. The measurement of polarisation effects allows differen-
tiation between insulation types. Data analysis is not always straightforward, real and
imaginary permittivity () are normally the variables used to analyse the different po-
larisation effects. To understand the relationship between conductivity, loss angle and
other material properties with permittivity and frequency it is necessary to consider
Maxwell’s equation relating to Ampere’s Law.

∇ × H = J + jωD (4.1)

Where J is the current density [Am−2 ] and D is the magnetic electric displacement
field/electric flux density [Am−2 ]. The current density J is composed of two parts, Ji
the impressed (source) current density [Am−2 ] and the conduction current density Jc
[Am−2 ], such that

∇ × H = Ji + Jc + jωD (4.2)

Insulation materials consist, at an atomic level, of positive and negative charges (±q)
balancing each other on the microscopic level. Macroscopically there may be some
localised bipolar space charge present, but even then an overall charge neutrality still
exists. When an electric field is applied to the dielectric sample, by a voltage across
the electrodes, the charges are displaced and aligned with respect to the field and their
polarity. At an atomic level the centre of gravity of negative and positive charge are
displaced in opposite direction creating a distance d [m] between them, creating dipoles
in neutral species (atom or molecules). In a dielectric sample, under an applied electric
field there will be several dipoles, each one with its specific dipole moment, p [C m],

p = qd, (4.3)

Where p is related to the local electric field El [V m−1 ] by the polarisability (α) [C m2 V−1 ],
where
p = αEl (4.4)

There are several types of polarisation,

• Electronic Polarisation: Movement of electrons in atoms.

• Ionic Polarisation: Charged atoms and molecules.

• Dipolar Polarisation: Permanent dipoles.

• Interfacial Polarisation: Charge accumulation between two different dielectric ma-

terial, like oil impregnated paper.

64
Macroscopic polarisation, Pmacro [C m−2 ], is given by the sum of the each polarised atom
or molecule,
Pmacro = nαEl = χε0 E (4.5)

where n is the number of molecules per unit volume contributing to the polarization, χ
[(m/V)n−1 ]is the electric susceptibility and ε0 is the electric permittivity of free space
(8.854 × 1012 ) [F m−1 ]. The flux density is given by,

D = ε0 E + Pmacro = ε0 E + χε0 E = ε0 E(1 + χ) = εE = εr ε0 E (4.6)

where E is the electric field [V m−1 ] and εr is the relative permittivity, which is defined
as
εr = (1 + χ) (4.7)

In vacuum, permittivity is equal ε0 , εr = 1, and D has zero phase shift with respect to
the applied field E. In a dielectric material the relative permittivity is greater than 1
and varies with frequency due to the fact the response is causal to the applied field. The
relative complex permittivity, ε∗r , takes this causality into effect by having a phase shift,

0 00
ε∗r = εr (ω) − jεr (ω) (4.8)

0 00
Where εr is the real relative permittivity and εr is the imaginary relative permittivity.
Every type of polarisation will have an effect on the complex permittivity as shown
in Figure 4.3 and each type will resonate at a specific frequency known as the natural
frequency, ω0 [rad s−1 ], as shown in Figure 4.2.

Figure 4.2: Spectrum of different types of polarisation with respect to frequency

65
 Complex permittivity
ε' Anomalous
dispersion

able to follow ω Unable to follow ω

ε''

Resonance occurs

ω0 ω

Figure 4.3: Real and imaginary permittivity variation with increasing frequency.

00
Therefore, expanding from Equation 4.2 it is then possible to determine how ε relates
0
to the conductivity of the sample and how ε relates to the ability to store energy of the
sample, i.e.
h 0 00
i
∇ × H = Ji + Jc + jω ε (ω) − jε (ω) ε0 E
 00
 0 0
(4.9)
= Ji + σs + ωε E + jωε = Ji + σe E + jωε E

where σe is the effective conductivity [S m−1 ], the first term σs is the static conductivity
[S m−1 ] derived from conduction current density, where Jc is equal to σs E. The second
00 0
term ωε is the conductivity due to an alternating field. For pure ohmic conduction ε
is independent of frequency, while for non ohmic conduction or polarization it increases
with decreasing frequency. The loss tangent (tan δ) can be calculated by rearranging

66
Equation 4.9 to give
" 00
!#
0
 σe  0 σs ε
∇ × H = Ji + jω 1 − j 0 E = Ji + jωε 1 − j − 0 E
ωε ωε0 ε (4.10)
0
= Ji + jωε (1 − j tan δe ) E

The σe and the effective loss tangent (tan δe ) are normally the values given by manufac-
turers although it is only specified as conductivity and loss tangent. The loss tangent,
σs
tan δe , has two terms, the first static electric loss tangent ( ωε0 ) describes loss due to

collision of electrons with other electrons, which is the predominantly ac loss in conduc-
00
ε
tors. The second term, ε0
, describes how much energy supplied by an external field is
dissipated as motion and heat, which is the predominant ac loss in insulators.
From (4.10) the total current density, Jt , is:

Jt = Ji + Jce + Jde = Ji + Jcd (4.11)

where Jce is the effective electric conduction current [Am−2 ], such that

00
σe E = (σs + ω )E (4.12)

Jde is the effective displacement electric current [Am−2 ], such that

0
jω E (4.13)

and Jcd is the effective displacement and electric conduction current [Am−2 ] such that

Jce + Jde (4.14)

For good dielectrics, the displacement current density is much greater than the conduc-
tion current density, meaning that Jcd is approximately Jce and therefore:

0
∇ × H ≈ Ji + jω E (4.15)

For good conductors, conduction current density is much greater than the displacement
electric current density, meaning that Jcd is approximately Jde and therefore:

∇ × H ≈ Ji + σe E (4.16)

The Solartron 1260 Impedance Analyzer together with Solartron 1296 Dielectric Inter-
face was used to analyse the dielectric behaviour of the samples from 1 mHz to 1 MHz.
The instruments were set up to record impedance magnitude and phase angle. The
integration period was set to a maximum of 75 cycles with a standard deviation of 1
% of the capable measured accuracy. The signal strength was set to an AC level of
6 V. As the samples had different geometries, the impedance analyser was not used to

67
process data and compare results. The complex relative permittivity (εr ) and loss factor
tan(δ) were calculated from the capacitance (C), which was initially derived from the
complex impedance, using the geometric properties of each sample. This experiment
uses the Solartron Materials Research and Test (SMaRT) software, which controls the
FDS apparatus.
The relative complex permittivity is calculated by first measuring the current phasor and
comparing it to the applied voltage phasor, therefore calculating the complex impedance
(Z). Additional calculations performed by the SMaRT software include,

j
Z = R − jX = R − (4.17)
ωC
1
C= (4.18)
ω Img(Z)
C ×d
ε∗r = (4.19)
A × ε0
00
εr = εr sin[arg(z) +π /2 ] (4.20)
0
εr = εr cos[arg(z) +π /2 ] (4.21)

Where ds is the sample thickness [m] and As sample area [m2 ].

4.2 Tracking Copper and Sulphur

This research investigates whether XRF spectroscopy is capable of reliably detecting

changes in the absolute sulphur quantities in the oil as the corrosive sulphur species at-
tack the copper surfaces. The development of a technique that is capable of monitoring
all sulphur species is expected to be a long lasting monitoring technique for corrosion in
transformers.
The CIGRE Transformer WG A2.32 Final Report suggests that a more precise risk as-
sessment for copper sulphide is required, as well as a deeper understanding of factors
influencing Cux S formation and deposition [11]. The CIGRE transformer working group
A2.40, as part of the copper sulphide long-term mitigation and risk assessment studies,
has expressed its interest in the reliability of XRF as a means of indirectly monitoring
the depletion of corrosive sulphur species in oil [103]. EDX and XRF were both used
to investigate the elemental composition of copper surfaces and oil as the DBDS reacts
with copper. XRF and EDX are quantitative and semi-quantitative analytical chemical
techniques, respectively, which are used in other transformer related forensic investiga-
tions [104].

68
4.2.1 Experiment Overview & Sample Preparation

For each set, a sample rack with ten vials with 15 ml of Nynas GeminiX oil and a copper
bar placed inside each vial was prepared, as shown in Figure 4.4. For samples aged in
nitrogen, the sample rack was placed inside the vacuum oven, evacuated and purged
with nitrogen three times. Every 24 h ± 30 min, a vial was removed from the oven and
the remaining copper samples replaced by new ones. The nitrogen purging process was
repeated every time the oven was opened, for the samples aged in nitrogen.

Figure 4.4: Sample rack containing vials with oil and copper bars.

Each experiment set consists of ten vials therefore at the end of the experiment there
are 10 oil vials with an ageing time range of 1 day to 10 days, and 55 copper samples
aged for 24 h (1 day) in the same type of oil, but with 10 different degradation stages.
The size of the copper sample is approximately 30 mm by 7.5 mm by 1.5 mm and the
glass headspace vial has 20 ml capacity with a nominal diameter of 23 mm, as shown
in Figure 4.5, which are the same dimensions as BS EN 62535 the standard corrosion test.

Figure 4.5: Sample preparation apparatus: oil vial (left), copper bar (right), auto-
matic pipette (top), sample bag & tweezers (middle).

69
The oil with DBDS is prepared in bulk just before the experiment, which for each exper-
iment set is approximately 200 ml. The DBDS is dissolved in the oil using a stirrer and
heater keeping the oil at good mixing speed while not producing bubbles and keeping
the oil at 50 ◦ C to 70 ◦ C until all particulates are dissolved.
The copper bar is initially wrapped in kraft paper, which is cut into 30 mm pieces and
then unwrapped just before being placed in oil. If the experiment uses wrapped copper
samples then the last kraft paper layer is kept. Once the oil vials are removed from
the oven and cooled down, each copper sample is extracted and the vial is sealed with
a PTFE-faced silicone septum and aluminium cap and placed under vacuum at room
temperature. After removing the copper sample from oil it is cleaned with cyclohexane
twice to remove the oil, when dried it is sealed in a plastic sample bag and kept under
vacuum.
The SEM-EDX requires no extra coating of the samples therefore the samples are sim-
ply placed in the sample holder inside the chamber. The copper samples are always
handled with gloves and clean tweezers to minimize contamination. The XRF requires
no sample preparation, the oil is simply poured into a specific cup in the machine, the
simplicity in sample preparation is one of the reasons XRF is being investigated as a
possible chemical analysis tool for the power industry.

4.2.2 Calibration Curves

Before analysing the oil samples with the XRF, it is first necessary to obtain the cali-
bration curves for copper and sulphur. The concentration range for copper and sulphur
was from 0 to 70 ppm and from 0 to 5000 ppm respectively, Table 4.1 lists the nominal
values of the calibration points. Both Cu and S certified stock solution were using Base
Oil 20 (20 mm2 sec−1 at 40 ◦ C ), which is further diluted with Base Oil 20 containing a
copper concentration of <0.5 µg g−1 and a sulphur concentration of <1 µg g−1 (SPEX-
CertiPrep, Inc.). The dilution method was performed using weight measurement ratios.
The calibration range for each element, Cu and S, is according to the expected in-service
transformer oil concentrations. Copper standards were prepared by dilution from a cer-
tified reference material (1000 µg g−1 pm1%, SPEXCertiPrep, Inc.) using Base Oil 20
mixed directly into head-space glass vials. Sulphur standards were prepared similarly
using certified reference material (5000 µgg−1 pm1%, 50 µgg−1 pm1% SPEXCertiPrep,
Inc.). Due to the broad sulphur concentration range, to minimize errors in the dilu-
tions, two different certified stock solutions were used. The 5000 ppm stock solution was
used to produce samples down to 100 ppm while the 50 ppm stock solution was used
to prepare all the lower concentrations. The calibration plots for copper and sulphur
are shown in Figure 4.6 and 4.7 respectively. The Origin software package was used to
process the data using the least-squares method to perform the linear fit of the data.

70
 Element Nominal concentration (ppm)
Copper 0, 0.5, 1, 2.5, 5, 10, 20, 30, 40, 50, 70
Sulphur 0, 5, 10, 25, 50 100, 250, 500, 1000, 2500, 5000

Table 4.1: Nominal Copper and Sulphur concentration of standard solutions in Base
Oil

Figure 4.6: Calibration curve for copper in base oil 20. Average values with standard
deviation error bars are plotted, linear fit is performed using least squares method.

Figure 4.7: Calibration curve for sulphur in base oil 20. Average values with standard
deviation error bars are plotted, linear fit is performed using least squares method.

The Limit of Detection (LoD)[ppm] is calculated by measuring the instrumental aver-

age response of a blank sample (meanblank ) and adding three times standard deviation

71
(SDblank shown in Table 4.2), i.e.

LoD = meanblank + 3 · (SDblank ) (4.22)

To calculate the Limit of Quantisation (LoQ)[ppm] the multiplication factor of the stan-
dard deviation is increased so that

LoQ = meanblank + 10 · (SDblank ) (4.23)

Sulphur Copper
Nominal Concentration SD Nominal Concentration SD
0 0.01 0 0.10
5 0.01 1 0.04
10 0.02 2 0.14
25 0.06 3 0.14
50 0.04 5 0.06
100 0.10 10 0.09
250 0.42 20 2.85
500 1.46 30 0.12
1000 0.61 40 0.67
2500 1.20 50 0.33
5000 1.95 70 0.39

Table 4.2: Standard deviation for copper and sulphur calibration points

Theory allows the use of a multiplication factor ranging from 5 to 10 in Equation 4.23,
to minimise errors a more conservative approach was chosen. Table 4.3 lists a summary
of statistical assessment of the linearity and sensitivity of the XRF technique for Cu and
S.

Experiment Parameters and Statistics Values (errors) Standard Errors

Intercept 3.73 0.18
Slope 0.902 0.02
Copper
Number of points 11 -
Adjusted R-square 0.996261 -
Intercept 0.018 0
Slope 0.022 6.67 · 10−5
Sulphur
Number of points 11 -
Adjusted R-square 0.999881 -

Table 4.3: Summary of statistics for the assessment of linearity1 and sensitivity2 of
the method for Cu and S.

72
After both limits were identified as an expression of signal response an interpolation us-
ing the calibration curves allowed their expression in concentration values (ppm). The
results for both elements are summarized in Table 4.4. For mathematical reasons while
determining the LoD and LoQ, the regression line for sulphur was forced to pass through
the mean value of the blank. No relevant variations were observed on other parameters
and it was possible to obtain positive values of concentration while trying to interpolate
an unknown value of signal response in the LoD and LoQ determination.

Element Limit of Detection, LoD (ppm) Limit of Quantitation, LoQ (ppm)

Copper 0.64 1.64
Sulphur 1.23 4.05

Table 4.4: Calculated LoD and LoQ for the method for Cu and S

XRF will detect all the sulphur in the oil, corrosive and noncorrosive species, therefore
variations will occur even from different batches of the same oil as shown in Figure 4.8.
Hence, XRF data was normalised with respect to the first data point of the trend. This
demonstrates one of the disadvantages of this technique in that it needs multiple data
points when analysing sulphur depletion in a transformer mineral oil.

Figure 4.8: Sulphur XRF data from two batches of Nynas Gemini X mineral oil, with
added DBDS concentrations.

4.2.3 Results

Four variables were investigated: concentration of DBDS, ageing atmosphere (air and
nitrogen), ageing temperature (120 ◦ C , 140◦ C and 170 ◦ C ) and insulation Kraft paper.
From these four variables six variable combinations were tested, as shown in Table 4.5.

73
For all experiments two elemental analysis techniques were used, XRF in oil to measure
copper and sulphur concentrations, and EDX to measure sulphur concentration on cop-
per samples. Therefore, there are 3 data sets per sample.

Experiment 1 2 3 4 5 6
Temperature (◦ C ) 140 140 140 140 120 170
Atmosphere Air N2 Air N2 N2 N2
2000 2000 2000 2000 2000 2000
1000 1000 1000 1000 1000 1000
DBDS (ppm)
500 500 500 500 500 500
100 100 100 100 - 100
Copper (◦ C ) Cu Only Cu Only Wrapped Wrapped Cu Only Cu Only

Table 4.5: Experiments performed and variables investigated

The XRF LoQ is well within the sulphur depletion range due to corrosion, as demon-
strated by Figure 4.9 where oil samples from Experiment 1,“Cu Only” aged under oil
with multiple DBDS concentrations in air atmosphere at 140 ◦ C , are analysed for sul-
phur. The absolute quantity of sulphur decreases by either precipitating out of the oil,
as an ageing process, or by reacting with other elements, i.e. copper. It also increases
as copper sulphur species are then dissolved back into the oil as shown in Figure 4.9.
The 0 ppm samples set shows a much flatter trend, demonstrating noncorrosive species
do not deplete as fast corrosive species as the oil ages. Nevertheless, the smaller the
ratio between corrosive and noncorrosive species is, the noisier the sulphur trend is, as
shown by 100 ppm sample set in Figure 4.9. The noisier the trend, the more data points
and time is required to detect the corrosive sulphur species depletion. Nevertheless,
the complex nature of the corrosion process may make the S depletion not in line with
DBDS concentration for a small window of time, as shown in Figure 4.9, 2000 ppm. It
is hypothesised that with time the 2000 ppm sulphur concentration will decrease.
Figure 4.10 shows the EDX results for the same experiment. The oils with DBDS
concentration of 2000 ppm, 1000 ppm and 500 ppm, demonstrate that copper-sulphur
products are being formed on the copper surface, depleting corrosive sulphur species in
the oil. The 100 ppm samples show an initial copper-sulphur product formation in the
copper surface until the 7th day. Although corrosive species appear to be depleted, in
the data set “100 ppm” 7th to 10th day in Figure 4.10, copper-sulphur products dis-
solved in the oil may still precipitate and noncorrosive species may still change slightly
the sulphur concentration in the short term, as shown in the data set “100 ppm” 7th to
10th day in Figure 4.9.

74
 105

100

95

Sulphur Trend (%)

90

85
0

2000ppm
80
1000ppm

500ppm
75
100ppm

70
0 2 4 6 8 10
Time (Days)

Figure 4.9: Experiment 1 (140 ◦ C , air atmosphere and Cu only), XRF data for S in
oil versus days.

30 100ppm

500ppm

1000ppm
25
2000ppm

20
Sulphur (%)

15

10

0 2 4 6 8 10
Time (Days)

Figure 4.10: Experiment 1 (140 ◦ C , air atmosphere and Cu only), EDX data for S
deposition on Cu versus days, every day the Cu samples are replaced by new samples.

Temperature has the greatest influence on sulphur depletion in the oil, by corrosion. Fig-
ure 4.11 demonstrates this by the increase of the negative gradients trends, as tempera-
ture increases. Figure 4.12 shows how the same concentration in DBDS has a completely
different effect on copper samples due to temperature variation. A 50 ◦ C increment has
a ten times higher impact on formation of sulphur compounds on copper surfaces. This
suggests that hotspots in transformers are particularly vulnerable to this sort of dam-
age. The operating temperature in power transformers is approximately around 60 ◦ C ,
which explains why sulphur corrosion related faults take decades to develop.

75
 100

90

Sulphur Trend (%)

80 170°C

140°C

70 120°C

60

50

40
0 2 4 6 8 10
Time (Days)

Figure 4.11: 2000 ppm of DBDS, copper samples aged in N2 atmosphere with varying
temperature, XRF data for sulphur concentration in oil versus days.

35 170°C

140°C
30 120°C

25
Sulphur (%)

20

15

10

-5
0 2 4 6 8 10

Time (Days)

Figure 4.12: 2000ppm of DBDS, copper samples and varying temperatures, EDX
data for S deposition on Cu samples versus days.

Figures 4.13 and 4.14 show the sulphur concentration, in oil and copper surfaces re-
spectively, across multiple experiments demonstrating how variable changes affect the S
depletion process. The first couple of samples/days appear to show a concentration in-
crease followed by a decrease in sulphur concentration. The sample sets aged in air have
greater negative gradient trends, which are expected because oxygen has been shown
to increase the displacement of Cux S from the copper surface [99], which results in in-
creased copper sites for corrosion. In addition, air increases the oxidation of the mineral
oil, increasing the rate of precipitation from the oil, which depletes the sulphur faster.
The copper samples, “Cu Only+N2 ” and “Cu Only+Air”, have a higher decrease rate
compared to wrapped samples, because of the Kraft paper decreases the initial corro-
sion rate. EDX data for Cu and S in paper has been investigated previously in Chapter 3.
76
 110

100

90

Sulphur Trend (%)

80

70

Cu Only+Air
60
Cu Only+N2

Paper+Air
50
Paper+N2

0 2 4 6 8 10
Time (Days)

Figure 4.13: 2000 ppm of DBDS, aged at 140 ◦ C , XRF data for sulphur concentration
concentration in oil versus days.

Cu Only+Air
30
Cu Only+N2

Paper+Air
25
Paper+N2

20
Sulphur (%)

15

10

0 2 4 6 8 10
Time (Days)

Figure 4.14: 2000 ppm of DBDS, aged at 140 ◦ C , EDX data for sulphur deposition
on Cu samples versus days.

Figure 4.15 shows that the concentration of copper in oil has no dependence on DBDS
concentration and there are no clear links between other experimental variables and
copper dissolution. The samples in general have a very low [Cu], close to the 1.95 ppm
LoQ.
The sample set that more closely resembles field samples are the “Wrapped” samples,
copper samples wrapped with Kraft paper samples, and they have the lowest [Cu] in oil
with less than 5 ppm throughout the 10 days. The “Cu Only + N2 ” data trend, copper
samples aged in nitrogen, have the greatest [Cu], which is expected since the copper
surface is more exposed and the nitrogen prevents dissolved copper from oxidising and
precipitating, nevertheless [Cu] is always less than 10 ppm, which is not very significant.
77
 Cu Only + Air + 140°C

Cu Only + N2 + 140°C

Wrapped + N2 + 140°C

Wrapped + Air + 140°C

9
Cu Only + N2 + 120°C

Cu Only + N2 + 170°C

7
Copper (ppm)

0 2 4 6 8 10
Time (Days)

Figure 4.15: 2000 ppm of DBDS with multiple temperatures (120-140 ◦ C )and with
bare copper samples(Cu Only) or Kraft paper wrapped copper samples(wrapped), XRF
data for [Cu] in oil versus days.

Figure 4.16 shows the [Cu] for a variety of different mineral oils, which have been stress
aged as at 140 ◦ C in air with copper. Results demonstrate that dissolved copper com-
plexes stay within the same range, proving that the long term increasing trend of copper
dissolution is not caused by DBDS. “Gemini X”, which is DBDS free, demonstrates the
typical dissolution of copper in noncorrosive oil which can be seen in the field. HyVolt
III with 2000 ppm of DBDS added shows very similar concentration with a difference
of less than 2 ppm throughout the 10 days. The “No Cu Change”, Gemini X with 2000
ppm of DBDS and the copper samples are not replaced every 24 h, shows an initial spike
in [Cu] and then it levels off at concentration similar to the “Gemini X”. The [Cu] in
“10GBN”, which is a naturally corrosive mineral oil, shows a similar data trend initially
to Gemini X, but after the 6th day the concentration drops. The [Cu] fluctuates due
to copper corrosive species, but because they mainly form insoluble products, such as
Cux S, the concentration decreases as the dissolved copper precipitates out of the oil, as
demonstrated by “10GBN”.
The following data investigate the reproducibility of the observed sulphur and copper
trends. Figures 4.17 and 4.18 demonstrate there is a reasonable repeatability of the
data collected. Variations between the two data sets are evident, but the overall trend
is the same. This means that if XRF is used as a diagnostic tool for transformers, then
conclusions cannot be taken on single point measurements. Instead, there must be a
dataset of historical data points in order to detect the underlying trend, which provides
diagnostic information.

78
 Gemini X
9
HyVolt III + DBDS (2000ppm)

No Cu Change

10GBN
8

Copper (ppm)
6

2 4 6 8 10
Time (Days)

Figure 4.16: XRF data for Cu concentration in oil as a function of ageing time, for
various oils (GeminiX, HyVolt III with 2000ppm of DBDS, No Cu Change - GeminiX
with 2000ppm, 10GBN) at a constant ageing temperature of 140 ◦ C , aged in an air
atmosphere

650
1

2
600

550
Sulphur (ppm)

500

450

400

0 2 4 6 8 10
Time (Days)

Figure 4.17: Repeatability measurements for experiment 1, 1000 ppm of DBDS, XRF
data for sulphur concentration in oil versus days.

79
 8
1

2
7

Copper (ppm)
5

0 2 4 6 8 10
Time (Days)

Figure 4.18: Repeatability measurements, 1000ppm of DBDS, XRF data for [Cu] in
oil versus days (140 ◦ C in air, copper only).

The repeatability accross different mineral oils has also been investigated and the XRF
spectroscopic technique has shown to work across a variety of mineral oils as demon-
strated in Figure 4.19. Some mineral oils, such as Ergon HyVolt III mineral oil that
contains less noncorrosive sulphur species, have few sulphur reactions which leads to a
more linear decay which . Figure 4.19 “No Cu Change”, also demonstrates that not
replacing the copper sample every day depletes the sulphur faster, which makes the de-
tection method for copper sulphide more sensitive over a shorter period of time. The
repeatability measurements are also confirmed in HyVolt III oil plus 2000 ppm of DBDS.

80
 110

100

90

80 Gemini X
Sulphur Trend (%)

GeminX+ DBDS (2000ppm)

70 GeminX+ DBDS (2000ppm) II

HyVolt III + DBDS (2000ppm)

60 Constant Cu II

Constant Cu
50

40

30

20
2 4 6 8 10
Time (Days)

Figure 4.19: XRF data for S concentration in oil as a function of ageing time, for
various oils with and without DBDS, at a constant ageing temperature of 140 ◦ C , aged
in an air atmosphere and with and with same/constant Cu specimen

4.2.4 Discussion

It has been shown that XRF is technically capable of quantifying very accurately sulphur
and copper in oil, the LoQ is 4.05 and 1.95 ppm respectively. From such measurements,
historical S trends can be built to extrapolate if Cux S deposits are being produced. The
volume required for this test is approximately 10 ml and the current standard method
for sampling insulating liquid BS EN 60475:2011 could be applied to this test.
Since this technique is based on a sulphur trend analysis, historical data are funda-
mental. This is due to the stochastic nature of the corrosion process and the complex
oil matrix in a transformer, where multiple reactions occur simultaneously, which can
mask the data relating to corrosive sulphur. As the technique evolves and a database
is developed, the historical data requirement might not be as important. The sampling
frequency of oil insulated assets will depend primarily on standard oil corrosion tests,
BS EN 62535, and loading of the asset. If an asset is overloaded and has corrosive oil
then a short sampling frequency is required, for 140 ◦ C every 2 days then for every 10
degrees lower sampling frequency can be half [17], i.e. at 130 ◦ C a sample every 4 days.
It is also important to monitor copper dissolution as it has been proven to have a major
impact on the dielectric properties of insulating oil [9]. XRF can additionally be used in
a similar way to DGA, to provide the information required to track copper, sulphur and
possibly other elements in the oil. Electrical monitoring techniques, such as dielectric
dissipation factor (DDF), reveal symptoms of a chemical degradation that XRF could
be able to clarify its causation, thanks to its elemental specificity.

81
The level of expertise required is relatively low. Once the equipment has been setup to
identify copper and sulphur, which can be done by the manufacturer, it only requires a
three point calibration. In terms of data analysis there is very little complexity, as the
user simply needs to identify linear trends, negative for sulphur and positive for copper.
With the development of this technique it would be relatively simply to produce software
that analyses the data trends and outputs a warning message if suspicious trends are
detected.
The results indicate that XRF spectroscopy would provide valuable information as a
condition monitoring tool for oil filled power equipment. The calibration curves demon-
strated the capability to track minute sulphur concentration changes in the oil allowing
it to detect sulphur concentration decreases, indicating the onset of sulphur depletion
processes, such as corrosion. In addition, it demonstrates that the long term increasing
trend of dissolved copper in oil is not caused by DBDS or naturally corrosive oils, such
as 10GBN. Nevertheless, results show that XRF is capable of quantifying copper very
accurately, down to 2 ppm.
The simple test procedure required to perform and interpret the readings also increases
the feasibility of this technique to be used as a condition monitoring tool, which can be
performed on site using commercially available bench top XRF equipment.
Future work should focus on development of data analysis algorithms to detect the
underlying trend from the minimum number of historical points and to be able to differ-
entiate between different situations that may cause sulphur depletion, such as oxidation
of oil and corrosion.

4.3 Frequency Dielectric Spectroscopy Analysis of Copper

Sulphide Deposits in Paper Insulation

The hypothesis for this experiment is that at a specific frequency spectrum the Cux S
particulates will resonate and the magnitude of that resonance will be related to the
quantity of Cux S deposits on the sample. FDS is already used by industry to moni-
tor other variables [105]. Therefore, if a relationship between Cux S and FDS data can
be found with a reasonable signal to noise ratio it would be very easy to deploy this
technique to field transformers. To use FDS as a Cux S deposition monitoring technique
it needs first to have good repeatability and accuracy results with the ability to differ-
entiate between highly contaminated from low contaminated Cux S Kraft paper. This
section presents a series of laboratory and field sample analyses using FDS, undertaken
as a feasibility study. The samples were analysed in both room humidity levels and dry
conditions.

82
4.3.1 Sample Preparation

Paper samples consist of two types, artificially aged and serviced aged. The artificially
aged samples are generated by the standard sample production experiment, previously
explained in Chapter 3. These samples can be controlled and therefore will be as ‘clean’
as possible from other impurities besides Cux S. The samples are cut into 35 mm diam-
eter circles with varying thickness due to copper sulphide deposits. The serviced aged
samples contain several other impurities, and were used to verify the feasibility of this
technique as a condition monitoring tool for Cux S. The serviced aged samples were
collected from a 400 kV/275 kV transformer, with 17 years of service [106]. The oil was
previously identified to be corrosive and therefore was passivated, but at the time of
decommissioning the oil did not pass corrosive tests. The transformer was decommis-
sioned due to a combination of corrosive oil and high operating temperature leading to
a high probability of heavy Cux S deposition. This was confirmed by visually inspecting
the conductors and insulation paper, which also had a degree of polymerization (DP)
of around 200. Several samples were collected from different sections of the transformer
and using the ASTM D7671 copper strip corrosion standard it was possible to obtain
samples that were slightly tarnished (1b), dark tarnished (3a) and corroded (4b) as
shown in Figure 4.21. The transformer conductor has two copper parts divided by a
nomex layer. The conductor was wrapped with six half lap layers of insulation paper
of 1 cm width. Due to the nature of the sample geometry and degradation, it was not
possible to produce a 35 mm diameter circular sample. Therefore a pigtail sample was
used, Figure 4.20, where two conductors were clamped together to act as electrodes
between the insulation paper, which replicates more closely real transformer conditions
between two coil turns.

Figure 4.20: Pigtail sample configuration

83
 Figure 4.21: Field samples and copper corrosion evaluation using the ASTM D7671
corrosion scale

4.3.2 Experimental Arrangement

Dry conditions were established by drying in vacuum oven, at rotary pump pressure
at room temperature, for a minimum period of 48 h. Due to the rapid absorption of
moisture in kraft paper, shown in Figure 4.22, and the long measurement time required,
the moisture content needs to be kept low during FDS measurements. Therefore, during
measurements the samples were placed in an air tight box containing silica gel and purged
with nitrogen. While dielectric measurements were taken, the hydrometer EasyLog
EL-USB-2 was used to measure temperature and relative humidity, Figure 4.23 shows
the setup. Figure 4.24 demonstrates that the moisture was around 6 % and room
temperature fluctuated between 20 ◦ C ± 4◦ C, the peaks represent the sample changes.

Figure 4.22: Kraft paper’s rate of moisture absorption, red: absolute mass change,
blue: percentage mass change relative to the dry sample

84
 Figure 4.23: Schematic of moisture free dielectric test cell

Figure 4.24: Temperature and moisture during FDS measurement of dry samples

4.3.3 Results

The FDS results of the kraft paper aged at 140 ◦ C for 72 h with several concentrations
of DBDS are shown in Figures 4.25 and 4.26. The permittivity of the samples has in-
creased with respect to the new kraft paper sample and the magnitude of that increase
0
appears to be related to the increase in Cux S deposits. The εr shows the kraft paper’s
ability to store charge, which indicates the polarizability of the medium, which for more
00
corroded samples is higher due to dipolar structures such as Cux S. The εr represents
the loss component of the medium, which correlates to the degree of corrosion of the
samples where the Cux S deposits have a higher conductivity than cellulose paper. The
serviced aged pigtail samples show similar increases in the ε∗r . From Figure 4.27 and 4.28
the FDS results show sample 4 has the highest values for both real and imaginary parts
of permittivity. Sample 4 has a corrosion scale of 4B according to the standard ASTM
D7671 corrosive sulphur table and a heavy Cux S deposition on the inner layer of the

85
insulation papers as in shown Figure 4.21. The results also demonstrate that samples 2
and 7 are very similar. From Figure 4.21, sample 2 and 7 have different corrosion scales,
1B and 3A respectively, but the insulation paper in both samples had no visible traces
of Cux S deposits resulting in similar FDS data.

0
Figure 4.25: Real part of relative permittivity (εr ) for the aged and new samples at
room moisture and temperature

00
Figure 4.26: Imaginary part of relative permittivity (εr ) for the aged and new
samples at room moisture and temperature

86
 0
Figure 4.27: Real part of relative permittivity (εr ) of the serviced aged pigtail samples
at room moisture and temperature

0
Figure 4.28: Imaginary part of relative permittivity (εr ) of the serviced aged pigtail
samples at room moisture and temperature

4.3.3.1 Room Temperature and Low Moisture

Figures 4.29 and 4.30 show the same trends as previous results under room conditions,
but the trends have diminished by a factor of 2-3 orders of magnitude, suggesting that
moisture has a great influence on the FDS results in this frequency range. Cux S deposits
have an effect on the dielectric properties of the paper that appears to be between 0.1
Hz and a lower frequency. Figures 4.31 and 4.32 show the results for the serviced
87
aged pigtail samples and again exhibit similar FDS changes with respect to samples
prepared in laboratory. Figure 4.33 compares all the samples together; the serviced
aged as expected, have the highest loss factors due to the severely aged paper with an
approximate DP of 200 and a working lifespan of 17 years. Looking at the two sets of
samples separately, i.e. artificially and serviced aged, at low frequency the higher the
corrosion the higher the loss factor. The results suggest that if the frequency spectrum
were increased there would be two maxima points, one in the low frequency (10 µHz)
and another at a high frequency (10 MHz).

0
Figure 4.29: Real part of relative permittivity (εr ) of the aged and new samples at
low moisture and room temperature

0
Figure 4.30: Imaginary part of relative permittivity (εr ) of the aged and new samples
at low moisture and room temperature

88
 0
Figure 4.31: Real part of relative permittivity (εr ) of the serviced aged pigtail samples
at low moisture and room temperature

00
Figure 4.32: Imaginary part of relative permittivity (εr ) of the serviced aged pigtail
samples at low moisture and room temperature

4.3.4 Discussion

Dielectric properties are dependent on many factors, such as frequency, temperature

and atomic structure of sample. This technique is already a standardised test for trans-
formers, but it is only at power frequency, which is sometimes insufficient to detect even
strong changes with the dielectric properties of complex systems as the transformer’s
insulation[107]. From Figure 4.27 it is possible to see two peak at around 100 mHz and
1 mHz suggesting these may be relaxation frequencies, suggesting two possible species.
Figure 4.31 shows that the peak at 100 mHz has disappeared suggesting it was due
89
 Figure 4.33: Loss Factor for all samples at low moisture and room temperature

to moisture. This demonstrates that it is fundamental to first detect Cux S resonance

frequency and then analyse how that changes in a more realistic/noisier environment.
The data show that the hypothesis that Cux S has resonance frequency is correct, al-
though the applied signal frequency needs to be reduced. The second hypothesis that
the magnitude of the resonance relates to the quantity of Cux S has also been validated.
The laboratory prepared samples also seem to replicate reasonably well the field sam-
ples in both the detection of Cux S deposits and FDS response to other impurities, e.g.
moisture. The initial technical feasibility study has been performed, where the theoret-
ical principles were shown to be applicable. In the future extended research is required,
where kraft paper is tested in more realistic environmental conditions. Furthermore,
the applied signal to the dielectric material needs to be redesigned as there is no direct
access to Kraft paper in transformer.

4.4 Summary

XRF and FDS have been investigated as two possible condition monitoring techniques
for corrosive sulphur and copper sulphide deposits in paper, respectively. Laboratory
scale trials and field samples have been used to investigate and validate the condition
monitoring techniques presented. Results from both these techniques have contributed
to the state-of-the-art through a learned journal publication in the IEEE Transaction
of Dielectric and Electrical Insulation [108], which provides a complete summary of the
XRF work, and an international conference publication in the Electrical Insulation Con-
ference, which provide a summary of the work on kraft paper analysis with FDS [109].
The results indicate that XRF spectroscopy can provide valuable information as a condi-
tion monitoring tool for oil filled power equipment. The calibration curves demonstrated

90
that XRF is cable of quantifying sulphur and copper down to a level of 4.05 and 1.64
ppm, respectively, as shown in Table 4.4. In addition, it was demonstrated that the
XRF capability to track minute sulphur concentration changes in the oil allows it to
detect sulphur concentration decreases which can indicate the onset of sulphur depletion
processes, such as corrosion. It was suggested that from XRF measurements, historical
S trends can be built to extrapolate if Cux S deposits are being produced. The volume
required for this test is approximately 10 ml and the current standard method for sam-
pling insulating liquid BS EN 60475:2011 could be applied to this test. Finally, results
have demonstrated that the long term increasing trend of dissolved copper in oil is not
caused by DBDS or naturally corrosive oils, such as 10GBN.
The FDS data has validated the hypothesis, that Cux S has relaxation frequency and
that the magnitude of the resonance relates to the quantity of Cux S, as shown in Figure
4.29 and 4.30. Results have also demonstrated that the resonance frequency of Cux S
is approximately around 1 mHz. The laboratory prepared samples also seem to repli-
cate well the field samples in terms of detecting Cux S deposits and its effects to other
impurities, e.g. moisture.

91
Chapter 5

The Impact of Copper Sulphide

Deposition on the Thermal
Properties and Electrical
Strength of Kraft Paper

Copper sulphide related failures of oil-filled plants have become more common around
most parts of the world over the last couple of decades. As discussed in Chapter 2, there
are two main theories for the failure event suggested by the current state-of-the-art,
which are thermal runaway and turn-to-turn disk electrical breakdown. Forensic analy-
sis of failed units can determine the cause of the fault; but it is not possible to accurately
verify the trigger event due to difficulties in determining the failure event has to do with
the similarities in the effects of each proposed mechanism. Electrical breakdown and
hotspot failures in transformers result in the destruction of the solid insulation in the
surrounding area, and very often the ignition of the mineral oil.
There have been some forensic investigations of decommissioned transformer units,
where copper sulphide related degradation was well advanced [106]. These reports pro-
vide some of the structure to the experiments performed in this chapter to investigate
what the failure event might be. There are three main areas investigated, electrical
breakdown of kraft paper insulation, the chemical degradation of the paper and finally
the electrical and thermal conductivity of kraft paper contaminated with copper sul-
phide.

92
5.1 Analytical Techniques

The analytical techniques used in this experiment consist of Weibull data analysis for
the electrical breakdown tests. The chemical degradation of paper is quantified using
the degree of polymerisation. Finally, dielectric spectroscopy and thermal analysis are
used to examine the heat generated due to increased losses and the samples ability to
dissipate excess heat. Some of the techniques that have not been previously introduced
in this thesis are explained in greater detail below.

5.1.1 Breakdown Life Data Analysis

The failure events, i.e. electrical breakdown, of any component are a probabilistic
event. Hence, when it is stated by researchers or manufacturers the electrical break-
down strength of a certain insulation system it is related to a certain statistical analysis
method. The standard for electric strength of insulating materials uses the median
test data to calculate the electrical breakdown strength [110]. In this thesis a Weibull
two-parameter probability density function (pdf), fw2p (t), is used as it has been shown
to provide more information than the standard analysis technique [111]. The Weibull
two-parameter pdf is defined as
 β−1  β
β t − t
fw2p (t) = e η (5.1)
η η

Where β is the shape parameter and η [s] is the scale parameter. By integrating fw2p (t)
from zero to t, where t is a moment in time [s], gives Fw2p (t), the cumulative density
function (cdf);
 β
− ηt
Fw2p (t) = 1 − e (5.2)

The Weibull reliability function, Rw (t), is then easily calculated as one minus the cdf.
R(t) describes the probability of survival up to t, such that
 β
t
−
Rw (t) = e η
(5.3)

The failure rate function, also known as the hazard function λ specifies the instantaneous
rate of failure at time t of a specimen given that it survives to time t, where
 β−1
fw2p (t) β t
λ(t) = = (5.4)
Rw (t) η η

93
To visually access the fit of data to a Weibull distribution a Weibull plot is used, as
shown below;
 β
− ηt
Fw2p (t) = 1 − e
 β
t (5.5)
ln(1 − Fw2p (t)) =
η
ln(− ln(1 − Fw2p (t))) = β ln(t) − β ln(η)

By using these axes it allows the cdf plot to be linearised, as demonstrated in Figure
5.1.

Figure 5.1: Weibull standard plot with shape and scale parameter.

β, also known as the slope parameter, describes the gradient of the regression line in
the cdf plot. In terms of the pdf plot, β represents the spread of data, where the higher
β the narrower the spread is. In terms of failure rate β describes how failure develops
with time in the following ways:

β < 1 indicates that the failure rate decreases over time.

β = 1 indicates that the failure rate is constant over time.

β > 1 indicates that the failure rate increases over time.

In the Weibull plot, Figure 5.1, η represents the point (t) where the sample is most
likely to fail. The cdf probability, F(t), is based on the time constant 1 − 1e , which
is approximately 63.2 %. η. The increase of η shifts the pdf distribution to the right

94
and lowers the height, and the decrease of η shift the pdf distribution to the left and
increases the height.
It is also necessary to explain how Weibull analysis is applied specifically to progressive
E tests, with a constant increase rate of E. The cdf of failure in terms of E is given by
[111]:
 β1
− ηE
P (E) = 1 − e E (5.6)

Where β1 = β + b, b being the new constant that influences the shape parameter in the
Weibull function with respect to E. ηE [kVmm1 ] represents the scale parameter with
respect to the E.
When plotting the breakdown data points in a Weibull plot it is necessary to assign a
cumulative probability of failure, Ff (i, n), to each point. As a computational technique
was used to process the data, the maximum likelihood technique [112] was used as
opposed to the median rank approximation [113], Equation 5.7.

i − 0.3
Ff (i, n) = (5.7)
n + 0.4

where n is the number of data points and i is the order of the data point in the data set.

5.1.2 Degree of Polymerisation

Chemical degradation of kraft insulation paper occurs as the polymer chains break
down decreasing the tensile strength of the paper. A polymer is a molecule made up
of monomer units, which repeat themselves along the chain. The average number of
monomers per chain is called the degree of polymerisation (DP). The degradation of
natural and synthetic polymers is dependent on time and the ageing conditions. As
explained in Chapter 2, the DP of Kraft paper starts approximately from 1300, when
new, and then decreases as it degrades. When the DP reaches approximately 200 the
Kraft Paper is considered to have reached its end of life.
The decrease of DP in the insulation kraft paper does not greatly affect the dielectric
properties [20, 114]. Nevertheless, chemical degradation of Kraft insulation paper may
cause the transformer to fail as the paper becomes brittle and starts breaking away from
the transformer windings and insulation disappears [115]. In addition, local carbonisa-
tion of the paper increases the conductivity causing overheating and possibly conductor
faults.
Considerable work has gone into modelling the degradation of DP with respect to time
and ageing condition [116, 117, 42, 20, 118, 41]. One model of the degradation of DP
through time at a given temperature [119] is given as

1 1 k1 h i
− = 0 1 − e−k2 t (5.8)
DPt DP0 k2

95
where DPt is the DP at time t, DP0 is the initial DP, k10 is the initial rate at which
bonds break and k2 is the rate at which k10 changes. The model constants, k10 and k2 ,
are based on the Arrhenius equation,
−Ea
k = Aaf e Rg T (5.9)

where Ea is the activation energy [J mol−1 ], Rg universal gas constant 8.3144621 [J mol−1 K−1 ]
and Aaf is the pre-exponential factor.

5.2 Paper Insulation Electrical Breakdown Failure

The hypothesis for this experiment is that the higher the concentration of Cux S deposits
in the paper the lower the electrical breakdown level will be, because Cux S has a much
lower resistivity than paper and therefore reduces the dielectric properties of the whole
system. AC breakdown tests have been performed on new paper and aged paper in
multiple layer systems.
Initial results, undertaken to the standard test method ASTM D149 [120], obtained
using ball bearing electrodes with diameter of 6.35 mm (1/4 inch), suggested that the
electrical breakdown level reduced with the addition of paper layers, which is known
to be incorrect. Therefore, the electrode system was redesigned to a mushroom type
electrode with a flat circular head to eliminate any triple junctions between different di-
electric materials, which can cause partial discharge or corona discharge due to a locally
high electric field strength.
The objective of this electrical breakdown experiment is to test the suggested turn-to-
turn breakdown transformer failure mechanism in a Cux S related fault.

5.2.1 Experimental Arrangement

The kraft paper samples contaminated with copper sulphide, were produced using the
method described in Chapter 3. There were 20 electrical breakdown sites per Kraft
paper sample. A stencil made out of Melinex was placed on the sample to mark the
breakdown test area, ensuring all breakdown sites were equidistant from each other to
minimise the possibility of flashover, as shown in Figure 5.2.

96
 Figure 5.2: Stencil for BD Samples and thickness measurement of sample

The design of the test cell is demonstrated in Figure 5.3. The load mass of 250 g on the
top electrode provides a constant pressure to the sample independent of sample thick-
ness. Each electrode has a 314.5 mm2 , 10 mm diameter, surface contact area with the
sample.

Figure 5.3: Breakdown test cell

A dedicated container was filled with the desired mineral oil, Nynas Nytro Gemini X.The
test cell was immersed in the mineral oil as shown in Figure 5.4.

97
 Figure 5.4: Breakdown test cell with sample under oil, the top electrode (red) is the
HV electrode and the bottom electrode (green) is the earth electrode

The applied ac voltage ramp rate to the test samples was 50 V s−1 and the transformer
range was from 0 to 25 kV. The paper layer were always orientated with the Cux S
deposited side facing upwards towards the top electrode, which is the HV electrode.
Layer 1 is the top layer, closer to the HV electrode, and the layer 3 is the bottom layer,
closer to the earth electrode.

5.2.2 Experimental Method

The following method was used:

1. Place stencils on test samples.

2. Measure and record the thickness of each breakdown site, as show in Figure 5.2

3. Place the sample into the container together with the dielectric test cell and dry
it in vacuum at rotary motor pump pressure for a minimum of 12 hours.

4. Select breakdown site in the sample and place it between the electrodes, while
avoiding exposing the sample to air to prevent formation of air bubbles.

5. Start ramping voltage at a constant rate of 50 V s−1 .

6. Record breakdown voltage and breakdown site in the sample.

7. Perform and record 20 electrical breakdowns per sample.

98
The breakdown sites are then analysed using an optically confirm a breakdown, as shown
in Figure 5.5. Only breakdowns inside the 10 mm diameter circles of the electrode are
considered. The data was then subsequently analysed using the MatLab.

Figure 5.5: Ac breakdown site of a single layer of kraft paper insulation.

5.2.3 Results and Analysis

A multiple paper layered sample was designed to more realistically represent paper in-
sulation use in a transformer. The multiple paper layered sample is constituted of three
individual paper layers. There are three types of individual paper layers; Cux S deposits
paper, aged paper without Cux S and new paper. A total of 27 samples were produced,
exploring all the possible combinations, as shown in Table 5.1.
The samples shown in red in the Table 5.1, 3rd , 5th and 7th , were discarded. There are
two reasons to discard any sample, one is due to the high conductivity of the samples,
which triggers the electrical breakdown detection mechanism (current limiter system)
preventing the voltage from ramping across the sample. The other reason is the occur-
rence of competing BD mechanisms, normally resulting widespread data demonstrated
by a low β value, which make the data unrepresentative of the electrical breakdown
mechanism being studied. After visual inspection not all samples have 20 electrical
breakdown data points, the minimum number of electrical breakdown points per sample
is 15.
By looking at the Table 5.2 it is possible to see that an aged sample and a new sample
have statistically the same electrical breakdown strength. The aged sample undergoes
the same stress ageing as the Cux S contaminated samples, but the oil does not contain
any DBDS. The fact that new and aged samples have very similar dielectric breakdown
strengths demonstrates that if the Cux S contaminated sample has any degradation of
the electrical breakdown strength it is solely to do with the corrosion process rather than
paper age. This set of data can be used as the standard electrical breakdown strength
of a new 3 paper layer dielectric system. To calculate the standard electrical breakdown
strength the average was taken from the 8 samples, which gives 52.31 ± 4.44 kVmm−1 .

99
 Sample N◦ Top Middle Bottom β η (kV mm−1 )
1 New New New 20 55
2 New Cux S New 10 36
3 New New Cux S 6 31
4 Cux S New New 22 47
5 Cux S Cux S Cux S
6 Cux S Cux S New 19 27
7 Cux S New Cux S
8 New Cux S Cux S 16 20
9 New New Aged 26 57
10 New Aged New 34 54
11 Aged New New 22 57
12 Aged Aged Aged 27 55
13 Aged Aged Cux S 25 35
14 Aged Cux S Aged 14 36
15 Cux S Aged Aged 22 45
16 Aged Aged New 22 46
17 Aged New Aged 41 47
18 New Aged Aged 33 48
19 Cux S Cux S Aged 20 26
20 Cux S Aged Cux S 37 7
21 Aged Cux S Cux S 9 18
22 Aged New Cux S 31 30
23 Aged Cux S New 22 40
24 New Aged Cux S 21 35
25 Cux S New Aged 23 44
26 Cux S Aged New 22 45
27 New Cux S Aged 25 37

Table 5.1: Sumary of ac Breakdown Data, with shape and scale parameters

Sample N◦ Top Middle Bottom β η (kV mm−1 )

1 New New New 20 55
9 New New Aged 26 57
10 New Aged New 34 54
11 Aged New New 22 57
12 Aged Aged Aged 27 55
16 Aged Aged New 22 46
17 Aged New Aged 41 47
18 New Aged Aged 33 48
Average 52 ± 4

Table 5.2: Summary of AC Breakdown Data for samples with no Cux S.

Table 5.3 demonstrates the electrical breakdown data for samples with one paper layer
contaminated with Cux S, which is subdivided into three datasets representing the lo-

100
cation of the paper layer contaminated with Cux S in the three layer system. By doing
this subdivision it makes the effect of the location of Cux S deposits on the multi-layered
sample clearer. The closer the Cux S deposits are to the earthed electrode the lower the
electrical breakdown strength. Figure 5.6 demonstrates the Weibull plot of the three
different data sets, which makes it possible to see that each data set is actually dis-
tinct. The average electrical breakdown strength for this sample set is 39 ± 5 kV mm−1 ,
which is 25 % lower than the average for samples with no Cux S. If a linear relationship
between sample layers and electrical breakdown strength is assumed, then it seems to
suggest that the paper layer contaminated with Cux S has almost no contribution to the
dielectric strength of the sample.

0.99
0.95
0.9
0.8
0.7
0.6
0.5
Probability of Failure

0.4
0.3
Sample 13 β = 25 α=35
0.2
Sample 26 β = 22 α=45
Sample 27 β = 25 α=37
0.1

0.01
30 40 50
Breakdown E Field (kV mm−1)

Figure 5.6: Weibull plot of samples set with one Cux S contaminated layers with
confidence bounds of 90 %.

The sample set with two layers of paper with contaminated Cux S is smaller than ex-
pected, because the 7th sample has been excluded as previously explained. Table 5.4
demonstrated the valid data for this sample set. The average electrical breakdown
strength for this sample set is 20 ± 8 kV mm−1 , which is 48 % lower than the average
value for samples with one layer contaminated by Cux S. Again, the electrical break-
down strength of the contaminated layer appears to be considerably diminished. The
geographical effects of the Cux S deposits in the samples are greater in this sample set,
as shown in Figure 5.7. When the Cux S deposits are at the extremities of the sample,
the electrical breakdown strength decreases to about 35 % of the average value for this
data set, as shown by the 20th sample. The 7th sample, which also had Cux S deposits at
the extremities of the sample, was too conducting for the AC breakdown testing system.

101
 Sample N◦ Top Middle Bottom β η (kV mm−1 )
25 Cux S New Aged 23 44
26 Cux S Aged New 22 45
15 Cux S Aged Aged 22 45
4 Cux S New New 22 47
2 New Cux S New 10 36
27 New Cux S Aged 25 37
23 Aged Cux S New 22 40
14 Aged Cux S Aged 14 36
13 Aged Aged Cux S 25 35
22 Aged New Cux S 31 30
24 New Aged Cux S 21 35
Average 39 ± 5

Table 5.3: Summary of AC Breakdown Data for samples with one Cux S contaminated
paper layer.

0.99
0.95
0.9
0.8
0.7
0.6
0.5
Sample 6 β = 19 α=27
Probability of Failure

0.4
Sample 8 β = 16 α=20
0.3 Sample 19 β = 20 α=26
0.2 Sample 20 β = 17 α=7
Sample 21 β = 8 α=18

0.1

0.01
10 20 30
Breakdown E Field (kV mm−1)

Figure 5.7: Weibull plot of samples set with two Cux S contaminated layers with
confidence bounds of 90%.

Sample N◦ Top Middle Bottom β η (kV mm−1 )

6 Cux S Cux S New 19 27
19 Cux S Cux S Aged 20 26
8 New Cux S Cux S 16 20
21 Aged Cux S Cux S 9 18
20 Cux S Aged Cux S 18 7
Average 20 ± 8

Table 5.4: Summary of AC Breakdown Data for samples with two Cux S contaminated
paper layer

102
Figure 5.8 shows samples 8, 10 and 23 in a Weibull plot, demonstrating the effect of Cux S
deposits in the paper insulation system, where one contaminated layer (sample 23) has a
25 % decrease in electrical BD strength and a two contaminated layers (sample 8) has a
63 % of the electrical BD strength compared to a three uncontaminated Kraft paper layer
system (sample 10), which has an electrical BD strength of 54 kV/mm. As a kraft paper
layer becomes contaminated with copper sulphide the permittivity of the contaminated
layer (cp ) increase significantly as demonstrated in Chapter 4 in the Frequency Dielectric
Spectroscopy Analysis of Copper Sulphide Deposits in Paper Insulation. Hence, the field
across the contaminated paper layers (Ecp ) reduces significantly and the field across the
uncontaminated paper layers (Eup ) increases significantly, because

up Eup = cp Ecp (5.10)

Where up is the permittivity of the contaminated paper layer. If the extreme case is
considered, where an individual paper layer, in a sample of three paper layers, increase
its permittivity drastically (effectively becoming a conductor) it reduces the electrical
breakdown strength by one third. Therefore analysing the data from Sample 23, one con-
taminated paper layer, which has a decrease of 27 % it indicates that the contaminated
layer has effectively reduced its E by 79 %.

0.99
0.95
0.9
0.8
0.7
0.6
0.5
Probability of Failure

0.4
0.3
Sample 8 β = 16 α = 20
0.2 Sample 10 β = 34 α = 54
Sample 27 β = 25 α = 40
0.1

0.01
20 30 40 50
Breakdown E Field /kV.mm−1

Figure 5.8: Weibull plot of Sample 8(two Cux S contaminated layers) , Sample 10
(no Cux S contaminated layers) and Sample 23 (one Cux S contaminated layer) with
confidence bounds of 90 %.

103
5.2.4 Discussion

The data have clearly demonstrated that the Cux S deposits reduced the electrical break-
down strength of each kraft paper layer by 79 %. Nevertheless, the electrical field
strength in a transformer turn-to-turn disc winding is not significant. The maximum
turn-to-turn E varies from transformer to transformer and it is not normally divulged
by transformer manufacturers, but it is expected to be approximately 200 V mm−1 [23].
Therefore due to the large number of paper layers and relative low stress field applied
across a turn-to-turn the probability of full electrical breakdown of the paper insulation
system due to Cux S deposits is low. Nevertheless, the effects of partial discharge due to
a diminished electrical field strength, which may further degrade the paper insulation
to a point of failure, have not been investigated.

5.3 Chemical Degradation of Paper

The aim of this experiment is to investigate the impact of corrosion and copper sulphide
deposition on the DP of the insulation paper. The samples were prepared using the
procedure in Chapter 3 and then sent to a commercial laboratory. A viscosity based
method was used to measure the DP following the standard BS EN 60450:2004 [121],
where two measurements where made per sample.
Three sample sets were analysed, as shown in Figure 5.9. The green sample set had no
Cux S deposits, as the ageing oil contained no DBDS (0 ppm). The blue sample set had
corrosion signs at different ageing times, from 1 to 13 days, while keeping the DBDS
concentration in the oil at 2000 ppm. The red sample set had various degrees corrosion
signs, as the DBDS concentration varied from 0 to 5000 ppm, while keeping the ageing
time at 7 days.

Figure 5.9: Samples used for DP analysis.

104
5.3.1 Results and Analysis

The sample sets were initially analysed individually to understand how each variable
DBDS and ageing time, affects the polymer chains in the paper. Figure 5.10 demon-
strates that the ageing procedure alone has a great effect on the DP of paper. The paper
has an initial DP of 1098, after 7 days of ageing at 140 ◦ C in mineral oil under air the
DP decreases 54 % to 498. This demonstrates that the stress-ageing test alone is already
degrading the paper polymer chains almost to its end of life.

Figure 5.10: DP versus Ageing time, for samples aged in non-corrosive oil (0 ppm of
DBDS) at 140 ◦ C

Having analysed the DP of paper samples without corrosive effects it was established
the base from which it is possible to analyse the effect of corrosion to the DP. Figure
5.11 plots the DP of the blue sample set, which is ageing at various times (1-13 days) in
oil with 2000 ppm DBDS. This sample set demonstrates a very similar depolymerisation
trend with a slightly higher depolymerisation rate, where the DP at 7 days is 433 that
is approximately 13 % lower than the sample aged without DBDS.
To get a clearer understanding of how DBDS affects the DP of the paper the red
sample set was analysed. Figure 5.12 shows how the DP changes with a varying DBDS
concentration, while keeping the ageing time constant at 7 days. The depolymerisation
trend is again similar to the previous trends although the magnitude is much diminished,
where at maximum concentration of 5000 ppm there is only a decrease of approximately
18 %, which is a DP difference of 88. Using Equation 5.9 to model both data trends the
life expectancy of the Kraft paper insulation for each condition is calculated. Hence,
this model calculates the significance of the corrosive oil in the depolymerisation of the
paper insulation in terms of life expectancy. Using the software package MatLab it is
possible to fit the model to the data and find the k10 and k2 , as shown in Figure 5.13.
The data fitting parameters are displayed in the Table 5.5.

105
 Figure 5.11: DP versus Ageing time, with 2000 ppm of DBDS at 140 ◦ C

Figure 5.12: DP versus DBDS, with a constant ageing time of 7 days at 140 ◦ C

1100 DBDS 0ppm

DBDS 2000ppm
1000 Heywood1
Heywood2
900

800

700
DP

600

500

400

300

200
0 10 20 30 40 50 60 70 80
Time (Days)

Figure 5.13: DP versus Ageing time, for two samples sets. One aged in oil with 2000
ppm of DBDS and the other with 0 ppm of DBDS.
106
 Experiment Parameters and Statistics Values
k10 2.8137 · 10−9
k2 6.88 · 10−7
Number of points 11
DBDS 2000ppm (Heywood1) Adjusted R-square 0.9844
DP 210 Time intercept 47.81
k10 2.136 · 10−9
k2 5.406 · 10−9
DBDS 0ppm (Heywood2) Number of points 5
Adjusted R-square 0.9816
DP 210 Time intercept 78.71

Table 5.5: Summary of Heywood’s models [20] demonstrated in Figure 5.13, bound-
aries applied to both models were, as t = 0 DP = 0 and as t = ∞ DP = 200

The samples used to calculate the DP values cover a relative short time span compared
to the actual life expectancy, which introduces errors when fitting the model to the data.
Heywood DP studies [20], on aged Kraft paper in mineral oil at 140 ◦ C , demonstrate
that Kraft paper tends to a DP value of 200 as time tends to infinity. Hence, this limit
has been introduced when calculating the model constants, k10 and k2 , for both models
Heywood1 and Heywood2 shown in Table 5.5. To compare the two models a DP values
of 210 is assumed as the end-of-life for kraft paper insulation. Heywood2, based on
the sample set aged with 2000 ppm of DBDS, reached the end-of-life at 48 days and
Heywood1, based on the sample set aged with no DBDS, the end-of-life is reached at
79 days. Hence, the sulphur corrosion process has decreased the life expectancy of kraft
paper by 33%, at an operating condition of 140 ◦ C in mineral oil under air atmosphere.
The ageing temperature of 140 ◦ C is significantly higher than field samples tempera-
tures, which is why the expected life is considerably shorter compared to the normal
life expectancy of a transformer. Literature demonstrates it is possible to change the
Heywood model constants so that the DP rate of degradation is with respect to a nor-
mal operating temperature [20]. As explained previously the model constants are based
on the Arrhenius equation. Therefore, having defined all the variables at 140 ◦ C it is
possible to adjust k10 and k2 by changing T from 140 ◦ C (413.15 K) to 60 ◦ C (333.15
K), as shown in Table 5.6. The activation energies (Ea ) were defined from literature
[119, 20] as it has been shown to be the optimum values. The Ea value for k2 Heywood1
model might not be the optimal value as the DBDS introduces a degradation of the
Kraft paper’s DP that has not been previously modelled. The pre-exponential factors
(Aaf ) were calculated based on the initial k10 and k2 values calculated. Figure 5.14
demonstrates the Heywood model with the new constants, which demonstrates that the
sulphur corrosion process reduces the life expectancy of kraft paper insulation by 63
years, which is 25 % less compared to the sample aged with no DBDS.

107
 Heywood 1 Model (Samples Aged with 2000 ppm of DBDS)
k10 Aaf Ea (J mol−1 ) R(J mol−1 K−1 ) T(K)
2.81E-09 413.2
1.03410E+06 115200 8.314
8.95E-13 333.2

k2 Aaf Ea (J mol−1 ) R(J mol−1 K−1 ) T(K)

6.88E-07 413.2
7.62360E+09 126900 8.314
9.66E-11 333.2

Heywood 2 Model (Samples Aged with 0 ppm of DBDS)

k10 Aaf Ea (J mol−1 ) R(J mol−1 K−1 ) T(K)
2.14E-09 413.2
7.85030E+05 115200 8.314
6.79E-13 333.2

k2 Aaf Ea (J mol−1 ) R(J mol−1 K−1 ) T(K)

5.41E-07 413.2
5.99030E+09 126900 8.314
7.59E-11 333.2

Table 5.6: Change of k constants in Heywood model, using the Arrhenius equation.

1100
DBDS 2000ppm
1000 DBDS 0ppm

900

800

700
DP

600

500

400 191.2 Years 255 Years

300

200

100
0 200 400 600 800 1000
Time (Years)

Figure 5.14: Heywood DP kinetics time degradation model at 60◦ C for Kraft paper
insulation under mineral oil

108
5.3.2 Discussion

It has been shown that the corrosion process degrades the chemical cellulose chain bonds,
as demonstrated by Figure 5.12. The impact of the DBDS on the DP of kraft paper
was shown to decrease to approximately a fifth of the DP value for a given thermal
ageing time. Nevertheless, the thermal stress on kraft paper was shown to be much
more detrimental the DP of the kraft paper.
By applying DP life expectancy models to the data it was demonstrate that the corrosion
process reduces 25 % of the transformer life expectancy at normal working conditions.
It is necessary to mention that although the error bars are plotted for the DP measure-
ments,no confidence bounds were plotted. If the errors from both the experimental data
and variables obtain from Heywood model [20] are taken into consideration in the DP
models, both sample sets become statistically the same, considering a 90 % confidence
bound. Therefore, the results have shown that effects of corrosion on the DP of Kraft
paper are not significant enough to be the main failure event, it also emphasised the
detrimental effects that hotspots can have on paper insulation.

5.4 Thermal Runaway

The following experiments test the hypothesis that Cux S deposits increase the dielectric
losses of oil impregnated kraft paper, which generates thermal losses to a point that
exceeds its thermal dissipation properties. To analyse if a Cux S deposits can generate
a thermal hotspot in the transformer windings the tan(δ) and the thermal conductivity
of oil impregnated kraft paper are measured. As explained in Chapter 2 in the Effects
of Corrosion the thermal hotspot temperature rise can be modelled using equation 2.12
and 2.13, repeated here for convenience:

Pd = ωCL VL2 tan δ (5.11)

∆d
Th = −Pd ρT (5.12)
Acs
Where ρT is described as
1
ρT = (5.13)
k
Where k is the thermal conductivity [W m K−1 ].

5.4.1 Experimental Arrangement and Sample Preparation

The Solartron 1260 Impedance Analyzer with the Solartron 1296 Dielectric Interface
was used to measure the dielectric properties at power frequency (50 Hz). A pigtail
sample configuration shown in Figure 5.15 was used for the dielectric measurements.

109
The samples were analysed at room temperature, under dry oil in an evacuated cham-
ber. The vacuum chamber has two electrical coaxial feed-throughs connecting to the
dielectric analyser, as shown in Figure 5.16. The tan(δ) data was measured for new
(unaged) samples and for aged samples with copper sulphide deposits. Multiple tan(δ)
measurements were recorded for each sample. In addition, two similar samples were
measured for each type of sample.

Figure 5.15: Sample diagram for tan delta measurements.

Figure 5.16: Experiment for tan(δ) measurement of oil immersed pigtail kraft paper
wrapped copper conductor samples.

The samples were prepared from approximately 10 cm sections of kraft paper wrapped
transformer conductor. The paper adjacent to the copper is wound gap to gap according
to the standard for potentially corrosive sulphur (BS EN 62535). Each individual section
was aged under the standard sample production conditions, which is 140 ◦ C in a fan
oven, immersed in Nynas Gemini X mineral oil with 2000 ppm of DBDS for a duration
of 7 days. Afterwards, the samples were cleaned with cyclohexane and allowed to dry
at room conditions. Finally, the pigtail configuration was assembled, using zip-ties to
maintain the contact pressure between the two paper wrapped copper sections. Two
different types of aged samples were assembled, one with a single kraft paper layer per
conductor and another with two kraft paper layers per conductor, generating a two kraft
paper layer interface and a four kraft paper layer interface, respectively. Table 5.7 lists
the samples analysed and Figure 5.17 demonstrates the disassembled pigtail samples
after performing the measurements.

110
 Sample Type Paper layer (per conductor)
1 New 2
2 New 1
3 Aged 2
4 Aged 2
5 Aged 1
6 Aged 1

Table 5.7: Sample set for dielectric measurements

Figure 5.17: Pigtail Samples used for dielectric measurements

The oil drying process was performed by placing it under full rotary pump vacuum
pressure for a period over 2 weeks. The moisture content was then measured using a
GRScientific Aquamax Karl Fischer Coulometric Titrator. The recorded measurements
are shown in Table 5.8, where the average moisture in the oil is 15.19 ppm.

Measurement Moisture Content (ppm)

1st 15.93
2nd 15.04
3rd 14.59

Table 5.8: The moisture content for mineral oil vacuum dried for a period of two
weeks

The kraft paper samples were kept under new Nynas Gemini X mineral oil at full rotary
pump vacuum pressure for a period over 2 weeks to remove air bubbles and moisture
from the kraft paper.

111
The samples for the thermal conductivity tests were prepared using the standard sam-
ple production discussed in Chapter 3. The kraft paper samples are impregnated with
Nynas Gemini X mineral oil to perform the thermal conductivity tests. There are two
sets of samples, new kraft paper and new mineral oil, and aged kraft paper with Cux S
deposits and aged mineral oil with 2000 ppm of DBDS. Figure 5.18 demonstrates the
two samples measured. The thermal conductivity was measured across oil impregnated
Kraft paper layers, where the new sample had a stack of 45 new Kraft paper layers and
the aged sample has a stack of 36 aged Kraft paper layers. The difference in layers has to
do with the equipment sample holder arrangement and it was taken into account when
calculating the thermal conductivity.

Figure 5.18: Thermal conductivity test samples, (A)- New Gemini X oil and new
Kraft paper, (B)- Aged Gemini X with 2000 ppm of DBDS at 140 ◦ C for 7 days in a
fan oven and aged Kraft paper with Cux S deposits

5.4.2 Dielectric Loss Measurements

As expected there is a clear increase in the dielectric loss of the insulation paper due to
the Cux S deposits, demonstrated in Table 5.9 and Figure 5.19. The results for the new
pigtail samples (1 & 2), which have one and two paper layers per conductor respectively,
have similar tan(δ) values (±8.2%), 0.00587 and 0.00491 respectively. Sample 1 contrary
to what was expected due to an increased dielectric interface, had a slightly higher tan(δ)
value compared to sample 2. An explanation could be the imperfections in the sample
geometries of the pigtail configuration and the heterogeneous pressure distribution across
the conductor-paper-conductor interface that leads to different sample thickness across
the sample. The measurement error is at most 3.2 %, as indicated by the standard
deviation values in the Table 5.9, demonstrating that the data variation has to do with
the sample reproducibility. Nevertheless, the results for samples 1 and 2 are realistic
and serve as control variables for the aged samples (3-6).

112
The aged samples demonstrate a significant increase of the mean tan(δ) value compared
to the new samples. The tan(δ) of each individual sample varies greatly from 0.068 to
0.148, which compared to the new samples, with an average tan(δ) value of 0.00539, is an
increase of 1162 % to 2651 % respectively. Although the sample variation is considerable,
the results demonstrate that with time the tan(δ) value can reach 0.148 or higher as
shown in other case studies [1].

T an(δ) [10−3 ] Mean S.D.

Sample
1 2 3 4 5 6 7 [10−3 ] [10−3 ]
New 2 Layers 6.12 5.82 5.69 5.87 0.223
New 1 Layer 4.96 4.96 4.90 4.90 4.88 4.93 4.88 4.91 0.031
Aged 2 Layers 65.5 68.5 67.4 71.3 67.1 68.0 2.16
Aged 2 Layers (II) 141.2 146.8 150.6 151.7 148 4.73
Aged 1 Layer 99.7 100.4 100.5 100.1 100.1 100.1 100.1 100.0 0.251
Aged 1 Layer (II) 142.7 142.9 143.4 144.0 143 0.056

Table 5.9: The tan(δ) measurements of the pigtail samples, S.D.- Standard Deviation

Figure 5.19: Dielectric loss of pigtail samples.

The tan(δ) results have demonstrated that the kraft paper with Cux S deposits can gen-
erate a significant higher dielectric loss, up to factor of 27.5 times higher. This excess
loss is then to some extent converted to thermal losses, which the oil impregnated kraft
paper insulation system needs to dissipate to prevent a thermal runaway.

113
5.4.3 Thermal conductivity

The thermal conductivity for the two samples were measure at room temperature and
130 ◦ C as shown in Table 5.10 and 5.11. For each test 15 repetitive measurements were
taken in sets of 5, where between sets the sample was removed and the sensor cleaned in
order to investigate the intrinsic repeatability of the sample demonstrated in the results
by relative standard deviation (RSD).
The hypothesis is that Kraft paper with Cux S deposits has a higher thermal conductiv-
ity than new Kraft paper, as Cux S deposits increase the electrical conductivity that is
normally positively correlated with the thermal conductivity. The increase in thermal
conductivity is expected, although the magnitude of that increase is unknown. Table
5.10 validate such hypothesis, at room temperature the Kraft paper with Cux S deposits
has a thermal conductivity value of 0.166 W m K−1 ± 0.7 W m K−1 , which compared to
the new Kraft paper sample is an increase of 16.9 %.

Kraft Paper New k (WmK−1 ) RSD (%)

Test 1 Average N=5 0.139 0.2
Test 2 Average N=5 0.142 0.4
Test 3 Average N=5 0.145 0.3
Overall Average N= 15 0.142 2

Kraft Paper Cux S k (WmK−1 ) RSD (%)

Test 1 Average N=5 0.167 0.3
Test 2 Average N=5 0.165 0.1
Test 3 Average N=5 0.167 0.1
Overall Average N=15 0.166 0.7

Table 5.10: Thermal conductivity analysis results of samples at room temperature

Increasing the temperature decreases the thermal conductivity, but the rate at which
each sample decrease differs. At 130 ◦ C the new Kraft paper sample decrease signifi-
cantly by 38.7 % to 0.086 WmK−1 , compared to a decrease of 9.6 % by the Kraft paper
sample with Cux S deposits to 0.150 WmK−1 . This resulted in the Kraft paper sample
with Cux S deposits having a thermal conductivity that is 74.4 % greater than the new
Kraft paper sample, as demonstrated in Table 5.11.

114
 Kraft Paper New k (WmK−1 ) RSD (%)
Test 1 Average N=5 0.087 1.2
Test 2 Average N=5 0.086 0.8
Test 3 Average N=5 0.085 1.0
Overall Average N=15 0.086 1.2

Kraft Paper Cux S k (WmK−1 ) RSD (%)

Test 1 Average N=5 0.147 0.5
Test 2 Average N=5 0.152 0.3
Test 3 Average N=5 0.151 0.8
Overall Average N=15 0.150 1.6

Table 5.11: Thermal conductivity analysis results of samples at 130 ◦ C

From Equation 5.11 and 5.12 the tan(δ) is inversely proportional to ρT , where

Th Acs 1
tan(δ) = − 2 (5.14)
ωCL VL ∆d ρT

Hence, the thermal conductivity increase of a factor between 1.17 and 1.74 is not enough
to compensate a dielectric loss increase of a factor between 11.6 and 27.4, which validates
that the presence of copper sulphide generates hotspots, which then start a snowball
effect of further Cux S deposits that increase further the dielectric losses and increasing
further the local temperature.

5.4.4 Discussion

The excess dielectric loss generated from a Cux S contaminated insulation paper system is
very significant, with an increase of approximately 27.5 times as demonstrated by Figure
5.19. The thermal conductivity has demonstrated to also increase and therefore increase
the capability of the oil impregnated paper to dissipate thermal losses. Nevertheless,
the thermal conductivity does not increase enough to compensate for the excess heat
generated and therefore a formation of a hotspot is very likely.
These experiments have demonstrated that a scenario where a hotspot is generated by
Cux S deposits is very likely, which then generates a snowball effect of generating further
Cux S deposits and further increasing the temperature.

115
5.5 Summary

This chapter has analysed three possible failure scenarios and the likelihood of corrosive
oil causing each one. The electrical breakdown strength experiments demonstrated that
the Cux S deposits reduced the electrical breakdown strength of each kraft paper layer
by 79 %. Nevertheless, the large number of paper layers and relative low field stress
applied across a turn-to-turn the probability of a full electrical breakdown of the paper
insulation system due to Cux S deposits is low. It was also demonstrated that the cor-
rosion process degrades the chemical cellulose chain bonds, where DP-life expectancy
models demonstrated that the corrosion process reduces 25 % of the transformer life
expectancy at normal working conditions. Nevertheless, the thermal stress on kraft pa-
per was shown to be much more detrimental to the DP of the kraft paper, indicating
that DP related failures are more likely to be due to thermal degradation than corrosive
sulphur. The power loss generated from a Cux S contaminated insulation paper system
increases the overall power loss by factor of approximately 27.5 as demonstrated by Fig-
ure 5.19. Although thermal conductivity increases, it is not enough to compensate the
excess heat generated. Therefore a formation of a hotspots in Cux S contaminated areas
is suggested to be very likely.
The results collected from all experiments suggest that a scenario where a hotspot is
generated by Cux S deposits is the most likely, which then generates a snowball effect of
generating further Cux S deposits and further increasing the local temperature.

116
Chapter 6

Conclusions

A general introduction of power transformers was expanded by a consideration of the be-

haviour of the insulation systems and their impact on the transformer’s life expectancy.
Furthermore, an overview of transformer design and its accessibility was discussed with
the focus on the development of condition monitoring techniques. A comprehensive re-
view of corrosive sulphur in the context of the power industry was then followed by a
closer look at the corrosive sulphur compounds and their origin, the corrosion process
and the current available detection and mitigation techniques.
Sample production methods for paper with Cux S deposits were investigated and an en-
larged Kraft paper sample with homogeneous Cux S deposition process was developed. It
was shown that DBDS, temperature, copper-paper interfaces and oxygen are positively
correlated with the generation of Cux S deposits. From the empirical studies presented,
a copper sulphide deposition model was proposed.
The development of condition monitoring techniques and laboratory-scale validation ex-
periments have been completed successfully, with the use of both field and artificial
samples. The XRF results indicate that its capability to track minute sulphur concen-
tration changes in the oil allows it to detect sulphur decreases, which can indicate the
onset of sulphur depletion processes, such as corrosion. The FDS data demonstrated the
correlation between the Cux S resonance magnitude with the quantity of Cux S deposits.
Finally a series of possible Cux S related failure events were analysed in terms of the
probability of corrosive oil causing each scenario. The breakdown strength experiments
demonstrated that the Cux S deposits reduced the breakdown strength of each kraft
paper layer by 79 %. The DP experiments demonstrated that the corrosion process
degrades the paper by 25 % of the Kraft paper’s life expectancy at normal working
conditions. Nevertheless, both degradation of breakdown strength and DP did not
individually increase the likelihood of failure. The power loss generated from Cux S con-
taminated insulation results demonstrates an increase of 2650 %, which combined with
only a 75 % increase in thermal conductivity makes it very likely that hotspots, are
created which may lead to a thermal runaway.

117
Each area presented contributes individually to the state-of-the-art, a summary of those
contributions are described in the following sections.

6.1 Copper Sulphide Deposition Process

At the start of this project, October 2011, there were several models for copper sulphide
deposition on paper with no agreement among the researchers. In addition, there was
no available sample production method to generate consistent enlarged Cux S contami-
nated paper samples. Empirical tests analysed in Chapter 3 validated the relationships
between copper sulphide deposition with temperature, oxygen and DBDS concentration.
Therefore advancing state-of-the-art in sample production and in the understanding cop-
per sulphide deposition process.
Chapter 3 has detailed the development of a sample production method capable of gen-
erating 90 mm square Kraft paper samples with homogeneous Cux S deposition, with
less than 1 % variation (Table 3.4), and a good reproducibility of ±3.5% (Figure 3.18).
In the process of developing the sample production method, it was found that increase
of either DBDS concentration or temperature has similar effects in increasing the rate of
corrosion. From observation it is clear that the paper enhances the local concentration
of DBDS making paper-copper interfaces ‘preferred’ corrosion sites. By conducting the
experiment under dry nitrogen it can be concluded that oxygen displaces Cux S from the
copper surface to the Kraft paper, and increases the available corrosion sites.
Based on the results obtained, it is possible to develop a hypothesis to explain copper
sulphide deposition using multiple Cux S formation mechanisms, one for reaction at the
copper surface and another one for reactions in the paper bulk. Figure 3.26 summarizes
diagrammatically these findings.

6.2 Monitoring the Corrosion Process

In terms of monitoring the corrosion in oil insulated power apparatus, the current stan-
dard techniques are very limited. There is currently two standard test methods one for
detection of potentially corrosive sulphur in mineral oil BS EN 62535 and another for
quantitative determination of DBDS BS EN 62697. At the time, of writing the power
industry had still not adopted any other condition monitoring techniques to a significant
scale.
XRF and FDS have been investigated as two possible condition monitoring techniques
for corrosive sulphur and copper sulphide deposits in paper, respectively. The results
indicate that XRF spectroscopy can provide valuable information as a condition mon-
itoring tool for oil filled power equipment. The calibration curves demonstrated that
XRF is capable of quantifying sulphur and copper down to levels of 4.05 and 1.64 ppm,

118
respectively, as shown in Table 4.4. It was demonstrated that from XRF measurements,
historical S trends can be built to extrapolate if Cux S deposits are being produced as
shown in Figure 4.19. A feasibility study of the XRF test method being applied to the
field was discussed. In addition, it was demonstrated that the 10 ml samples, required
to perform the test, could be collected under the current standard method for sampling
insulating liquids BS EN 60475:2011. Finally, results have demonstrated that the long
term increasing trend of dissolved copper in oil is not caused by DBDS or naturally
corrosive oils, such as 10GBN.
The FDS data has validated the hypothesis, that Cux S has a resonance frequency and
that the magnitude of the resonance relates to the quantity of Cux S, , as shown in Figure
4.29 and 4.30. The target frequency spectrum was demonstrated to be around 100 µHz
to 1 mHz, which requires the use of both polarisation and depolarisation currents (PDC)
and FDS techniques to reduce measurement time to a feasible amount. The laboratory
prepared samples demonstrated to closely replicate the field samples in terms of detect-
ing Cux S deposits and its effects to other impurities, e.g. moisture.
A complete summary of the XRF work has also been published in the IEEE Transac-
tion of Dielectric and Electrical Insulation. The work on kraft paper analysis with FDS
has also been summarised in work presented at the Electrical Insulation Conference [109].

6.3 The impact of Copper Sulphide Deposition on the ther-

mal properties and electrical strength of Kraft Paper

There have been many transformer failures due to corrosive sulphur in the mineral oil,
which has led the industry to re-evaluate their asset risk analysis policy for mineral oil
insulated power assets. Understanding how each insulation system degrades, especially
oil impregnated Kraft paper is crucial to precise and accurately pinpoint the current
degradation state and therefore the current risk of the asset failing.
Chapter 5, has analysed three possible failure scenarios and the likelihood of corrosive
oil causing each scenario. The breakdown strength experiments demonstrated that the
Cux S deposits reduced the breakdown strength of each kraft paper layer by 79 %.
Nevertheless, the large number of paper layers and relative low field stress applied across
a turn-to-turn makes the probability of a full electrical breakdown of the paper insulation
system due to Cux S deposits low. Empirical DP studies have demonstrated that the
corrosion process degrades the chemical cellulose chain bonds, where DP-life expectancy
models demonstrated that the corrosion process reduces 25 % of the transformer life
expectancy at normal working conditions. Nevertheless, the thermal stress on kraft
paper was shown to be much more detrimental to the DP of the kraft paper, indicating
that if a DP related failure occurs it is more likely to be due to thermal degradation than
corrosive sulphur. The power loss generated from a Cux S contaminated insulation paper

119
system was demonstrated to increase the overall power loss by factor of approximately
27.5 as demonstrated by Figure 5.19. The thermal conductivity was also demonstrated
(Table 5.10) to increase, but a 75 % increase, at 130 ◦ C , was shown to be insufficient
to compensate for the excess heat generated. Therefore a formation of hotspots in
Cux S contaminated areas is considered to be very probable scenario. The formation of
a hotspot may then generate a snowball effect, generating further Cux S deposits and
further increasing the local temperature, leading to a possible thermal runway.

6.4 Future Work

The work developed in this thesis started from a relatively small volume of previous
research that had mainly been developed recently due to the surges in corrosive sulphur
related failures in the power industry. Therefore the research undertaken in this thesis
set out to establish the initial concepts and relationships with environmental variables.

6.4.1 Corrosion Process

This thesis focused on the present corrosion related faults which centred around the cor-
rosive sulphur specie DBDS. The sample production method stemmed from the DBDS
and its known propensity to create copper sulphide deposits in kraft paper insulation.
Hence, the proposed model for the corrosion process was developed from empirical stud-
ies using DBDS. Investigation of the corrosion process of other corrosive species and
how it relates and differs from the proposed model is a possible area for future work.
Corrosive oil normally contains multiple corrosive species, therefore future investigations
of how multiple corrosive species affect the transformer windings and how the corrosion
process is affected by environmental factors could help improve the current proposed
model. Work presented in this thesis has contributed to a better understanding of this
new wave of corrosive related faults in the 21st century, but as industry learns from
its current mistakes as it has done in the past, future corrosion process may involve
different corrosive species. Having a model that is flexible to cope with these changes
may provide long lasting benefits for the power industry.

6.4.2 Monitoring Techniques

The XRF test method for the detection of corrosion by the periodic evaluation of sul-
phur concentration in the oil has been instigated in a laboratory scale experiment and
presented in this thesis. Furthermore an initial field implementation feasibility analysis
was also discussed. The final evaluation of this technique is field testing. Future work
focusing on building an historical sulphur concentration database for high value oil insu-
lated power assets, such as transformers, could enable the development of data analysis
120
technique to more accurately detect the onset of corrosion.
XRF related work presented in this thesis analysed copper concentration in oil. Al-
though no correlation was found between corrosive sulphur species and copper dilution,
monitoring the copper concentration in oil is important due to its high impact on the
insulation systems.
The dielectric spectroscopy is in a more infancy state where theoretical hypothesis were
proven to work. Future work could possibly focus on the development of field measure-
ment methodologies for transformer.

121
References

[1] F. Scatiggio, V. Tumiatti, R. Maina, M. Tumiatti, M. Pompili, and R. Bartnikas,

“Corrosive sulfur induced failures in oil-filled electrical power transformers and
shunt reactors,” IEEE Trans. Power Del., vol. 24, no. 3, pp. 1240–1248, Jul. 2009.

[2] Energy Network Association, “Electricity distribution map,” Online, Tech.

Rep., Sep. 2014. [Online]. Available: http://www.energynetworks.org/info/faqs/
electricity-distribution-map.html

[3] Parliamentary Office of Science and Technology, “UK Electricity Networks (post-
note 163),” UK Parliament POST Note, Tech. Rep., 2001.

[4] U. Bakshi and V. Bakshi, Electrical Engineering and Control Systems. Technical
Publications Pune, 2008.

[5] S. Yang, M. Dong, G. jun Zhang, and Z. Zhang, “Condition evaluation of oil-paper
insulation based on dielectric spectroscopy,” in Proc. IEEE Conference on Elec-
trical Insulation and Dielectric Phenomena (CEIDP ’09), Virginia Beach, USA,
Oct. 2009, pp. 77–80.

[6] M. Heathcote, J & P Transformer Book, 13th ed. Newnes, 2007.

[7] A. Setayeshmehr, I. Fofana, C. Eichler, A. Akbari, H. Borsi, and E. Gockenbach,

“Dielectric spectroscopic measurements on transformer oil-paper insulation under
controlled laboratory conditions,” IEEE Trans. Dielectr. Electr. Insul., vol. 15,
no. 4, pp. 1100–1111, Aug. 2008.

[8] C. Bengtsson, M. Dahlund, J. Hajek, L. F. Pettersson, K. Gustafsson, R. Leander-

sson, and A. Hjortsberg, “Oil corrosion and conduction Cu2 S deposition in power
transformer windings,” CIGRE, Tech. Rep. A2-111, 2006.

[9] T. Oommen, “Corrosive and non-corrosive sulfur in transformer oils,” in Proc.

IEEE of Electrical Electronics Insulation Conference (EEIC’93), North Carolina,
USA, 1993, pp. 309–311.

[10] P. J. Griffin and L. R. Lewand, “Understanding corrosive sulfur problems in electric

apparatus,” in Proc. 74th Annual International Doble Client Conference, Boston,
USA, 2007.
122
[11] M. Dahlund, I. Atanasova-Höhlein, R. Maina, N. Dominelli, T. Ohnstad, T. Ami-
moto, C. Claiborne, M.-H. Ese, J. Lukic, V. Mezhvynskiy, C. Perrier, P. Smith,
and J. Tanimura, “Copper sulphide in transformer insulation final report,” CI-
GRE, Tech. Rep. WG A2-32, 2009.

[12] S. Ren, Y. Xu, X. Cao, L. Zhong, Q. Yu, and R. Jeanjean, “A research summary
of corrosive sulfur in mineral oils,” in Proc. IEEE 9th International Conference
on the Properties and Applications of Dielectric Materials (ICPADM’09), Harbin,
China, Jul. 2009, pp. 353–356.

[13] L. R. Lewand, “The role of corrosive sulfur in transformers and transformer oil,”
in Proc. 69th Annual International Conference of Doble Clients. Boston, USA:
Doble Engineering Company, USA, 2002.

[14] F. Scatiggio, R. Maina, V. Tumiatti, M. Pompili, and R. Bartnikas, “Long term

stability of insulating mineral oils following their corrosive sulfur removal,” in Proc.
IEEE International Symposium on Electrical Insulation (ISEI’10), San Diego,
USA, Jun. 2010, pp. 1–4.

[15] M. Faraday, “Experimental researches in electricity,” Phil. Trans. R. Soc. Lond.,

vol. 122, pp. 125–162, Jan. 1832.

[16] N. Callan, “LXXXVIII on a new galvanic battery,” Lond. Edinb. Dubl. Phil. Mag.,
vol. 9, no. 3, pp. 472–478, 1836.

[17] K. Zipenowski, M. Deri, and O. Blathy, “Electric Distribution by Alternating

Currents,” 1888.

[18] J. Winders, Power Transformers: Principles and Applications, ser. Power Engi-
neering (Willis). Taylor & Francis, 2002.

[19] Met Office. (2014, Sep.) UK climate - Extremes. [Online]. Available: http://www.
metoffice.gov.uk/public/weather/climate-extremes/#?tab=climateExtremes

[20] R. Heywood, “The degradation models of cellulosic transformer insulation,” Ph.D.

dissertation, University of Surrey, Chemistry Deparment, 1997.

[21] IEC TC 10, Copper and copper alloys (BS EN 13601), BSI Std., 2002.

[22] S. Kulkarni and S. Khaparde, Transformer Engineering: Design, Technology, and

Diagnostics, 2nd ed. Taylor & Francis, 2012.

[23] P. N. Jarman, personal correspondence, 2014.

[24] IEC TC 14, Power transformer - Application Guide (BS IEC 60076-8), BSI Std.,
2011.

[25] IEC TC 10, Fluids for electrotechnical applications Unused mineral insulating oils
for transformers and switchgear (BS EN 60296), BSI Std., 2012.
123
[26] S. Krawiec, “Production of corrosive sulphur free transformer fluids,” in Proc.
IEEE International Symposium on Electrical Insulation (ISEI’08). IEEE, Oct.
2007, pp. 174–177.

[27] R. Bartnikas, Engineering Dielectrics: Volume III Electrical Insulating Liquids.

ASTM International, 1994.

[28] J. M. Rasco, “Petroleum Refining Production of Naphthenic Transformer Oil Con-

trol of Corrosive Sulfur,” in Proc. 74th Annual International Doble Client Confer-
ence, Boston, USA, 2007.

[29] IEC TC 10, Mineral oil-impregnated electrical equipment in service. Guide to the
interpretation of dissolved and free gases analysis (BS EN 60599), Std., 2007.

[30] ASTM D27.01, Mineral Insulating Oil Used in Electrical Apparatus (ASTM
D3487), ASTM International Std., 2009.

[31] M. Meshkatoddini, “A practical method for lifetime estimation of the used min-
eral oils,” in Proc. IEEE 14th International Conference on Dielectric Liquids
(ICDL’02), Graz, Austria, Jul. 2002, pp. 305–308.

[32] IEC TC 10, Mineral oil-impregnated electrical equipment in service - Guide to the
interpretation of dissolved and free gases analysis (BS EN 60599 Draft), BSI Std.,
2013.

[33] T. Saha, “Review of modern diagnostic techniques for assessing insulation condi-
tion in aged transformers,” IEEE Trans. Dielectr. Electr. Insul., vol. 10, no. 5, pp.
903–917, Oct. 2003.

[34] M. Wang, A. J. Vandermaar, and K. Srivastava, “Review of condition assessment

of power transformers in service,” IEEE Electr. Insul. Mag., vol. 18, no. 6, pp.
12–25, Nov. 2002.

[35] B. H. Ward, “A survey of new techniques in insulation monitoring of power trans-

formers,” IEEE Electr. Insul. Mag., vol. 17, no. 3, pp. 16–23, Jun. 2001.

[36] W. S. Zaengl, “Applications of dielectric spectroscopy in time and frequency do-

main for HV power equipment,” IEEE Electr. Insul. Mag., vol. 19, no. 6, pp. 9–22,
2003.

[37] IEC TC 10, Insulating liquids. Measurement of relative permittivity, dielectric dis-
sipation factor (tan δ) and d.c. resistivity (BS EN 60247), BSI Std., 2004.

[38] “Standard Grade Nytro Libra Electrical Insulating Oil,” NYNAS, Tech. Rep., Apr.
2012.

[39] IEC SC 15C, Cellulosic papers for electrical purposes. Part 2: Methods of test (BS
EN 60554-2), BSI Std., 2002.
124
[40] IEC TC 14, Power transformers. Part 3: Insulation levels, dielectric tests and
external clearances in air (BS EN 60076-3), BSI Std., 2001.

[41] A. Kachler and I. Hohlein, “Aging of cellulose at transformer service temperatures.

Part 1: Influence of type of oil and air on the degree of polymerization of press-
board, dissolved gases, and furanic compounds in oil,” IEEE Electr. Insul. Mag.,
vol. 21, no. 2, pp. 15–21, Mar. 2005.

[42] A. Emsley and G. Stevens, “Review of chemical indicators of degradation of cel-

lulosic electrical paper insulation in oil-filled transformers,” IEE Proc. Sci, Meas.
& Tech., vol. 141, no. 5, pp. 324–334, Sep. 1994.

[43] P. J. Baird, H. Herman, G. C. Stevens, and P. N. Jarman, “Non-destructive mea-

surement of the degradation of transformer insulating paper,” IEEE Trans. Di-
electr. Electr. Insul., vol. 13, no. 2, pp. 309–318, Apr. 2006.

[44] S. Okabe, G. Ueta, H. Wada, and H. Okubo, “Partial discharge-induced degra-

dation characteristics of oil-impregnated insulating material used in oil-immersed
power transformers,” IEEE Trans. Dielectr. Electr. Insul., vol. 17, no. 4, pp. 1225–
1232, Aug. 2010.

[45] U. Gafvert, L. Adeen, M. Tapper, P. Ghasemi, and B. Jonsson, “Dielectric spec-

troscopy in time and frequency domain applied to diagnostics of power transform-
ers,” in Proc. IEEE 6th International Conference on Properties and Applications
of Dielectric Materials (ICPADM’00), vol. 2, Xi’an, China, 2000, pp. 825 –830.

[46] T. Saha, R. Middleton, and A. Thomas, “Understanding frequency & time domain
polarisation methods for the insulation condition assessment of power transform-
ers,” in Proc. IEEE Power Energy Society General Meeting (PES ’09), July 2009,
pp. 1–8.

[47] IEC TC 68, Magnetic Materials. Part 8-6: Specifications for individual materials
Soft magnetic metallic materials (BS EN 60404-8-6), BSI Std., 2009.

[48] ISO/TC 156, Corrosion of metals and alloys – Terms and definitions (ISO
8044:1986), ISO Std., 1989.

[49] A. H. n. Rocha, “Faults in 500 kV shunt reactors caused by corrosive sulfur in

the insulating oil,” in Proc. 68th Annual International Doble Client Conference,
Boston, USA, 2001.

[50] F. Scatiggio, V. Tumiatti, R. Maina, M. Tumiatti, M. Pompili, and R. Bartnikas,

“Corrosive sulfur in insulating oils: Its detection and correlated power apparatus
failures,” IEEE Trans. Power Del., vol. 23, no. 1, pp. 508–509, Jan. 2008.

[51] S. Toyama, J. Tanimura, N. Yamada, E. Nagao, and T. Amimoto, “Highly sensitive

detection method of dibenzyl disulfide and the elucidation of the mechanism,”
IEEE Trans. Dielectr. Electr. Insul., vol. 16, no. 2, pp. 509–515, Apr. 2009.
125
[52] IEC TC 10, Insulating liquids, Test method for detection of potentially corrosive
sulphur in used and unused insulating oil (BS EN 62535), BSI Std., 2011.

[53] H. Kawarai, Y. Fujita, J. Tanimura, S. Toyama, N. Yamada, E. Nagao,

N. Hosokawa, and T. Amimoto, “Role of dissolved copper and oxygen on copper
sulfide generation in insulating oil,” IEEE Trans. Dielectr. Electr. Insul., vol. 16,
no. 5, pp. 1430–1435, Oct. 2009.

[54] F. Scatiggio, M. Pompili, and R. Bartnikas, “Oils with presence of corrosive sulfur:
mitigation and collateral effects,” in Proc. IEEE Electrical Insulation Conference
(EIC’09), Montreal, Canada, May 2009, pp. 478–481.

[55] L. Lewand and S. Reed, “Destruction of dibenzyl disulfide in transformer oil,” in

Proc. 75th Annual International Doble Client Conference, Boston, USA, 2008, pp.
1–20.

[56] L. Lewand, “Dibenzyl disulfide in transformer oil,” International Electrical Testing

Association World, vol. 30, no. 3, pp. 1–4, 2008.

[57] J. Smith and P. Sen, “Corrosive sulfur in transformer oil,” in Proc. IEEE Industry
Applications Society Annual Meeting, Houston, USA, Oct. 2010, pp. 1–4.

[58] K. Mizuno, S. Toyama, H. Kawarai, J. Tanimura, Y. Fujita, F. Kato, T. Amimoto,

N. Hosokawa, and E. Nagao, “Identification of compounds leading to copper sulfide
formation on insulating paper in transformers and the degradation of suppressing
effect of 1, 2, 3-benzotriazole and Irgamet 39 in insulating oil,” in Proc. 76th
Annual International Doble Client Conference, Boston, USA, 2009, pp. 1–7.

[59] L. Arvidsson, “Conditions for DBDS induced corrosive sulfur formation,” in Proc.
IEEE International Symposium on Electrical Insulation (ISEI’10), San Diego,
USA, Jun. 2010, pp. 1–4.

[60] R. Maina, V. Tumiatti, M. Pompili, and R. Bartnikas, “Corrosive sulfur effects in

transformer oils and remedial procedures,” IEEE Trans. Dielectr. Electr. Insul.,
vol. 16, no. 6, pp. 1655–1663, Dec. 2009.

[61] M. A. G. Martins and A. R. Gomes, “Experimental study of the role played by

dibenzyl disulfide on insulating oil corrosivity-effect of passivator Irgamet 39,”
IEEE Electr. Insul. Mag., vol. 26, no. 4, pp. 27–32, Jul. 2010.

[62] T. Amimoto, N. Hosokawa, E. Nagao, J. Tanimura, and S. Toyama, “Concen-

tration dependence of corrosive sulfur on copper-sulfide deposition on insulating
paper used for power transformer insulation,” IEEE Trans. Dielectr. Electr. Insul.,
vol. 16, no. 5, pp. 1489–1495, Oct. 2009.

[63] G. A. Oweimreen, A. M. Y. Jaber, A. M. Abulkibash, and N. A. Mehanna, “The

depletion of dibenzyl disulfide from a mineral transformer insulating oil,” IEEE
Trans. Dielectr. Electr. Insul., vol. 19, no. 6, pp. 1962–1970, Dec. 2012.
126
[64] F. Kato, T. Amimoto, E. Nagao, N. Hosokawa, S. Toyama, and J. Tanimura,
“Effect of DBDS concentration and heating duration on copper sulfide formation in
oil-immersed transformer insulation,” IEEE Trans. Dielectr. Electr. Insul., vol. 18,
no. 6, pp. 1869–1876, Dec. 2011.

[65] I. Fofana, A. Beroual, A. T. Ngnui, E. C. D. Lyon, and A. G. D. Collongue,

“Parameters affecting the formation of sulphide deposition on copper conductors,”
in Proc. XVII International Symposium on High Voltage Engineering (ISH’11),
Hannover, Germany, Aug. 2011, pp. 1993–1997.

[66] H. Kawarai, Y. Uehara, K. Mizuno, S. Toyama, E. Nagao, N. Hosokawa, and

T. Amimoto, “Influences of oxygen and 2,6-di-tert-butyl-p-cresol on copper sul-
fide deposition on insulating paper in oil-immersed transformer insulation,” IEEE
Trans. Dielectr. Electr. Insul., vol. 19, no. 6, pp. 1884–1890, Dec. 2012.

[67] S. Ren, L. Zhong, Q. Yu, X. Cao, and S. Li, “Influence of the atmosphere on the
reaction of dibenzyl disulfide with copper in mineral insulation oil,” IEEE Trans.
Dielectr. Electr. Insul., vol. 19, no. 3, pp. 849–854, Jun. 2012.

[68] S. Toyama, K. Mizuno, F. Kato, E. Nagao, T. Amimoto, and N. Hosokawa, “Influ-

ence of inhibitor and oil components on copper sulfide deposition on kraft paper
in oil-immersed insulation,” IEEE Trans. Dielectr. Electr. Insul., vol. 18, no. 6,
pp. 1877–1885, Dec. 2011.

[69] R. Crutcher and K. Warner, “Copper Mobility and Sulfur in Electrical Equip-
ment,” in TechCon - Europe 2008, Liverpool, UK, Nov. 2008.

[70] F. Scatiggio, C. Marchiori, and P. Botelho, “Understanding and management of

sulphur corrosion from insulating oil,” in Proc. 74th Annual International Doble
Client Conference, Boston, USA, 2007.

[71] B. Pahlavanpour, H. Odelius, and K. Sundkvist, “Corrosive sulfur issues and reme-
dial actions,” in Proc. 74th Annual International Doble Client Conference, Boston,
USA, 2007.

[72] C. Claiborne, G. Frimpong, W. Ball, and A. Hjortsberg, “Transformer oils, corro-

sive sulfur and effects on transformers,” in Proc. 74th Annual International Doble
Client Conference, Boston, USA, 2007.

[73] T. B. Whitfield, J. E. Castle, C. Saracco, and M. Ali, “Transport of copper over

paper in high-voltage electrical insulation,” Surf. and Interface Anal., vol. 34,
no. 1, pp. 176–179, 2002.

[74] R. Maina, V. Tumiatti, M. C. Bruzzoniti, R. M. De Carlo, J. Lukic, and

D. Naumovic-Vukovic, “Copper dissolution and deposition tendency of insulating
mineral oils related to dielectric properties of liquid and solid insulation,” in Proc.

127
 IEEE International Conference on Dielectric Liquids (ICDL’11), Trondheim, Nor-
way, Jun. 2011, pp. 1–6.

[75] R. Bartnikas and K. Srivastava, Power and Communication Cables: Theory and
Applications. IEEE Press, 2000.

[76] 60287, Electric cables. Calculation of the current rating. (BS IEC 60287), BSI
Std., 2012.

[77] R. M. De Carlo, M. C. Bruzzoniti, C. Sarzanini, R. Maina, and V. Tumiatti,

“Copper contaminated insulating mineral oils-testing and investigations,” IEEE
Trans. Dielectr. Electr. Insul., vol. 20, no. 2, pp. 557–563, Apr. 2013.

[78] I. L. Hosier, A. S. Vaughan, and R. D. Chippendale, “Permittivity mismatch

and its influence on ramp breakdown performance,” in Proc. IEEE International
Conference on Solid Dielectrics (ICSD’13), Bologna, Italy, Jun. 2013, pp. 644–647.

[79] M. Chen, Y. Cheng, C. Deng, S. Liu, and C. Diao, “Research on the process of
the transformer winding inter-turn discharge caused by corrosive sulphur,” in Proc.
IEEE Conference on Electrical Insulation and Dielectric Phenomena (CEIDP’11),
Cancun, Mexico, Oct. 2011, pp. 476–479.

[80] M. Chen, Y. Yan, Y. Cheng, and C. Diao, “The characteristic of transformer wind-
ing inter-turn discharge under sulphur corrosion and thermal ageing,” in Proc. In-
ternational Conference on Intelligent System Design and Engineering Application
(ISDEA’12), Sanya, China, Jan. 2012, pp. 563–566.

[81] P. M. Mitchinson, P. L. Lewin, and P. Jarman, “A mechanism for the formation

of copper sulphide in oil filled electrical equipment,” in Proc. IEEE International
Symposium on Electrical Insulation (ISEI’10), San Diego, USA, Jun. 2010, pp.
1–4.

[82] V. Tumiatti, C. Roggero, M. Tumiatti, S. Di Carlo, R. Maina, and S. Kapila,

“IEC 62697-2012: State of the art methods for quantification of DBDS and other
corrosive sulfur compounds in unused and used insulating liquids,” IEEE Trans.
Dielectr. Electr. Insul., vol. 19, no. 5, pp. 1633–1641, Oct. 2012.

[83] TC GEL/10, Test methods for quantitative determination of corrosive sulphur

compounds in unused and used insulating liquids. Test method for quantitative
determination of dibenzyldisulfide (DBDS) (BS EN 62697-1), BSI Std., 2013.

[84] K. J. Rapp, A. W. Lemm, L. A. Orozco, and C. P. McShane, “Corrosive sulfur

phenomena mitigation by using natural ester dielectric fluids - field experience
in Latin America,” in Proc. IEEE Transmission and Distribution Conference and
Exposition (TDCE’08), Bogota, Colombia, Aug. 2008, pp. 1–6.

128
[85] P. Wiklund, M. Levin, and B. Pahlavanpour, “Copper dissolution and metal pas-
sivators in insulating oil,” IEEE Electr. Insul. Mag., vol. 23, no. 4, pp. 6–14, Jul.
2007.

[86] T. Amimoto, E. Nagao, J. Tanimura, S. Toyama, and N. Yamada, “Duration and

mechanism for suppressive effect of triazole-based passivators on copper-sulfide
deposition on insulating paper,” IEEE Trans. Dielectr. Electr. Insul., vol. 16, no. 1,
pp. 257–264, Feb. 2009.

[87] M. Higaki, Y. Kako, M. Moriyama, M. Hirano, K. Hiraishi, and K. Kurita, “Static

electrification and partial discharges caused by oil flow in forced oil cooled core type
transformers,” IEEE Trans. Power App. Syst., vol. PAS-98, no. 4, pp. 1259–1267,
Jul. 1979.

[88] T. Amimoto, F. Kato, J. Tanimura, S. Toyama, N. Hosokawa, and E. Nagao,

“Verification of non-corrosive sulphur in transformer oils used in Japan,” in Proc.
76th Annual International Doble Client Conference, Boston, USA, 2009.

[89] T. Kawamura, T. Kobayashi, T. Amimoto, and H. Murakami, “Failure modes of

oil-immersed transformers due to static electrification and copper sulphide gener-
ation, and suppressive effect of BTA,” CIGRE, Tech. Rep. A2-206, 2008.

[90] M. A. Martins, A. Gomes, and B. Pahlavanpour, “Experimental study of a pas-

sivated oil corrosiveness, after depletion of the passivator,” IEEE Electr. Insul.
Mag., vol. 25, no. 6, pp. 23–27, Nov. 2009.

[91] R. Asano, B. R. W. Miethke, and J. McIver, “Siemens experience on passivated

oils,” in Proc. 74th Annual International Doble Client Conference, Boston, MA,
USA, 2007.

[92] M. Dahlund, H. k. Johansson, U. Lager, and G. Wilson, “Understanding the Pres-

ence of Corrosive Sulfur in Previously Non-Corrosive Oils Following Regeneration,”
in Proc. 77th Annual International Doble Client Conference, Boston, MA, USA,
2010.

[93] D. Bell and A. Garratt-Reed, Energy Dispersive X-ray Analysis in the Electron
Microscope, ser. Microscopy handbooks. Taylor & Francis, 2003.

[94] J. Russ, Fundamentals of Energy Dispersive X-ray Analysis, ser. Butterworths

Monographs in Materials. Butterworths, 1984.

[95] P. van der Heide, X-ray Photoelectron Spectroscopy: An introduction to Principles

and Practices. Wiley, 2011.

[96] P. Gabbott, Principles and Applications of Thermal Analysis. John Wiley &
Sons, Inc., 2008.

129
 [97] “High Grade, Nytro Gemini X, High Performance Insulating Oil,” Nynas, Tech.
Rep., Apr. 2012.

[98] S. Okabe, M. Kohtoh, S. Toyama, and T. Amimoto, “Analysis methods of sulfide

and sulfoxide compounds in mineral insulating oil for diagnosis on electrostatic
charging of power transformers,” IEEE Trans. Dielectr. Electr. Insul., vol. 19,
no. 1, pp. 181–187, Feb. 2012.

[99] M. Facciotti, P. S. Amaro, A. F. Holt, R. C. Brown, P. L. Lewin, J. A. Pilgrim,

G. Wilson, and P. N. Jarman, “Contact-based corrosion mechanism leading to
copper sulphide deposition on insulating paper used in oil-immersed electrical
power equipment,” Corros. Sci., vol. 84, pp. 172–179, 2014.

[100] F. M. Clark and E. L. Raab, “Detection of corrosive sulfur compounds in mineral

transformer oil.” ASTM, vol. Preprint, pp. 1–10, 1948.

[101] G. R. Lachance and F. Claisse, Quantitative X-ray fluorescence analysis: theory

and application, Wiley, Ed. Wiley, 1995.

[102] F. Kremer and A. Schönhals, Broadband Dielectric Spectroscopy. Springer Berlin

Heidelberg, 2003.

[103] WG2.40, Paris, Aug. 2012.

[104] A. F. Holt, M. Facciotti, P. Amaro, R. C. D. Brown, P. L. Lewin, J. A. Pilgrim,

G. Wilson, and P. Jarman, “Silver corrosion in transformers,” in Proc. IEEE Con-
ference on Electrical Insulation and Dielectric Phenomena (CEIDP’13), Shenzhen,
China, Oct. 2013, pp. 448–451.

[105] I. Fofana, H. Hemmatjou, F. Meghnefi, M. Farzaneh, A. Setayeshmehr, H. Borsi,

and E. Gockenbach, “On the frequency domain dielectric response of oil-paper
insulation at low temperatures,” IEEE Trans. Dielectr. Electr. Insul., vol. 17,
no. 3, pp. 799–807, Jun. 2010.

[106] H. Ding, “Report 6692/R,” Doble PowerTest Ltd, Tech. Rep. 2, 2012.

[107] W. S. Zaengl, “Dielectric spectroscopy in time and frequency domain for HV power
equipment. I. Theoretical considerations,” IEEE Electr. Insul. Mag., vol. 19, no. 5,
pp. 5–19, 2003.

[108] P. S. Amaro, M. Facciotti, A. F. Holt, J. A. Pilgrim, P. L. Lewin, R. C. Brown,

G. Wilson, and P. Jarman, “X-Ray Fluorescence as a Condition Monitoring Tool
for Copper and Corrosive Sulphur Species in Insulating Oil,” IEEE Trans. Dielectr.
Electr. Insul., pp. 1–9, Sep. 2014, in Press.

[109] P. S. Amaro, A. F. Holt, M. Facciotti, J. A. Pilgrim, P. L. Lewin, R. C. D. Brown,

G. Wilson, and P. N. Jarman, “Impact of corrosive sulfur in transformer insulation

130
 paper,” in Proc. IEEE Electrical Insulation Conference (EIC’13), Ottawa, Canada,
Jun. 2013, pp. 459–463.

[110] IEC TC 112, BSI Standards Publication Electric strength of insulating materials
Test methods. Part 1 : Tests at power frequencies (BS EN 60243-1), BSI Std.,
2013.

[111] L. Dissado and J. Fothergill, Electrical Degradation and Breakdown in Polymers,

9th ed. IEE Materials & Devices, 1992.

[112] W. Nelson, Applied Life Data Analysis, ser. Wiley Series in Probability and Statis-
tics. Hoboken, NJ, USA: John Wiley & Sons, Inc., Feb. 1982.

[113] J. Fothergill, “Estimating the cumulative probability of failure data points to be

plotted on Weibull and other probability paper,” IEEE Trans. Dielectr. Electr.
Insul., vol. 25, no. 3, pp. 489–492, Jun. 1990.

[114] J. Gielniak, A. Graczkowski, and H. Moscicka-Grzesiak, “Does the degree of cellu-

lose polymerization affect the dielectric response?” IEEE Trans. Dielectr. Electr.
Insul., vol. 18, no. 5, pp. 1647–1650, Oct. 2011.

[115] A. M. Emsley, R. J. Heywood, M. Ali, and X. Xiao, “Degradation of cellulosic

insulation in power transformers .4. effects of ageing on the tensile strength of
paper,” IEE P. Sci. Meas. Tech., vol. 147, no. 6, pp. 285–290, Nov 2000.

[116] X. Zou, T. Uesaka, and N. Gurnagul, “Prediction of paper permanence by accel-

erated aging I. Kinetic analysis of the aging process,” Cellulose, vol. 3, no. 1, pp.
243–267, 1996.

[117] X. Zou, T. Uesaka, and N. Gurnagul, “Prediction of paper permanence by acceler-

ated aging II. Comparison of the predictions with natural aging results,” Cellulose,
vol. 3, no. 1, pp. 269–279, 1996.

[118] I. Hohlein and A. Kachler, “Aging of cellulose at transformer service temperatures.

Part 2. Influence of moisture and temperature on degree of polymerization and
formation of furanic compounds in free-breathing systems,” IEEE Electr. Insul.
Mag., vol. 21, no. 5, pp. 20–24, Sep. 2005.

[119] A. M. Emsley, R. J. Heywood, M. Ali, and C. M. Eley, “On the kinetics of degra-
dation of cellulose,” Cellulose, vol. 4, no. 1, pp. 1–5, Mar. 1997.

[120] D09.12, Standard Test Method for Dielectric Breakdown Voltage and Dielectric
Strength of Solid Electrical Insulating Materials at Commercial Power Frequencies
(ASTM D149 - 09), ASTM Std., 2013.

[121] IEC TC 15, Measurement of the average viscometric degree of polymerization of

new and aged cellulosic electrically insulating materials (BS EN 60450), BSI Std.,
2004.

131