Final SB Thesis 01apr09 SQ

Asset Management and the Role of Power
Quality on Electrical Treeing in Epoxy Resin
A thesis submitted to The University of Manchester

for the degree of Doctor of Philosophy
in the Faculty of Engineering and Physical Sciences
2009
Sanjay Bahadoorsingh
School of Electrical and Electronic Engineering
Table of Contents
List of tables ...................................................................................................................... 6
List of figures..................................................................................................................... 7
List of abbreviations ....................................................................................................... 11
List of abbreviations ....................................................................................................... 11
Abstract ............................................................................................................................ 12
Declaration ....................................................................................................................... 13
Copyright ......................................................................................................................... 13
Acknowledgements......................................................................................................... 14
Dedication ........................................................................................................................ 14
1. Introduction................................................................................................................ 15
1.1. Supergen V - AMPerES ....................................................................................... 15
1.2. Background .......................................................................................................... 15
1.3. An interpretation of insulation ageing................................................................... 16
1.4. Space charge ....................................................................................................... 19
1.4.1. Electroluminescence ..................................................................................... 19
1.5. Partial discharges................................................................................................. 20
1.6. Water trees........................................................................................................... 21
1.6.1. Transition from water trees to electrical trees................................................ 22
1.7. Electrical trees...................................................................................................... 23
1.7.1. Electrical tree types ....................................................................................... 24
1.7.2. Electrical tree initiation................................................................................... 25
1.8. Power quality and electrical ageing...................................................................... 26
1.8.1. The role of harmonics.................................................................................... 27
1.8.2. Modelling electrical stress with harmonic content ......................................... 30
1.8.3. Impact of harmonics on electrical ageing ...................................................... 32
1.9. Low voltage ageing .............................................................................................. 39
1.9.1. Influence of ageing factors ............................................................................ 39
1.9.2. Discussion ..................................................................................................... 42
1.10. Literature review findings ..................................................................................... 44
1.11. Asset management overview ............................................................................... 44
1.12. Asset management approaches .......................................................................... 46
1.12.1. Condition monitoring...................................................................................... 46
1.12.2. Reliability centered maintenance................................................................... 46
1.12.3. Complimentary roles of condition monitoring and reliability centered
maintenance .................................................................................................. 47
1.13. State of the art asset management ...................................................................... 48
1.13.1. General industry ............................................................................................ 48
1.13.2. Rail industry................................................................................................... 49
1.13.3. Aerospace industry........................................................................................ 51
1.13.4. Power industry............................................................................................... 52
3
1.13.5. Discussion ..................................................................................................... 54

1.14. The future of asset management - PAS 55 .......................................................... 56
1.14.1. Key principles of PAS 55 ............................................................................... 56
1.14.2. PAS 55 in the energy sector .......................................................................... 57
1.15. Review of asset management .............................................................................. 58
1.16. Aims and objectives ............................................................................................. 60
1.17. Thesis structure.................................................................................................... 60
2. A multifactor framework linking insulation ageing to asset management .......... 61
2.1.1. Overview........................................................................................................ 61
2.1.2. Asset management........................................................................................ 61
2.1.3. Material state ................................................................................................. 62
2.1.4. Stress factors................................................................................................. 64
2.1.5. Ageing mechanisms ...................................................................................... 66
2.1.6. Measurands ................................................................................................... 67
2.1.7. Multifactor framework .................................................................................... 69
2.1.8. Application ..................................................................................................... 72
2.1.9. Discussion ..................................................................................................... 73
3. Development of experimental method and equipment .......................................... 75
3.1. Test equipment..................................................................................................... 78
3.1.1. Waveform generation .................................................................................... 78
3.1.2. Amplification .................................................................................................. 79
3.1.3. Image capture................................................................................................ 80
3.1.4. Electroluminescence capture......................................................................... 80
3.1.5. Hardware assembly ....................................................................................... 81
3.2. Partial discharge instrumentation ......................................................................... 84
3.2.1. Design............................................................................................................ 85
3.2.2. Simulation ...................................................................................................... 86
3.2.3. Software development ................................................................................... 89
3.3. Limitations ............................................................................................................ 95
3.3.1. Sampling rate and sample size...................................................................... 95
3.3.2. Disturbances.................................................................................................. 96
3.3.3. Improvements for disturbance mitigation....................................................... 97
3.4. Disturbance reduction .......................................................................................... 98
3.4.1. Implementation of balanced circuit detection................................................. 98
3.4.2. Disturbance analysis.................................................................................... 100
3.5. System integration.............................................................................................. 102
3.6. Sample preparation ............................................................................................ 103
3.7. Assessment of partial discharge sources........................................................... 105
3.8. Experimental plan............................................................................................... 108
3.8.1. Influence of power quality on electrical tree growth and breakdown times.. 109
3.8.2. Influence of power quality on electrical treeing partial discharge patterns .. 110
4
4. Experimental results and analysis......................................................................... 112

4.1. Influence of needle lubricant coating.................................................................. 112
4.1.1. Partial discharge patterns............................................................................ 113
4.1.2. Electrical tree growth ................................................................................... 115
4.1.3. Initiation and breakdown times .................................................................... 119
4.1.4. Gaseous activity .......................................................................................... 120
4.1.5. Summary ..................................................................................................... 124
4.2. Sequence of electrical tree growth..................................................................... 125
4.3. Influence of power quality on electrical tree growth and breakdown time .......... 128
4.3.1. Electrical tree growth ................................................................................... 128
4.3.2. Deductions................................................................................................... 137
4.3.3. Breakdown time........................................................................................... 137
4.3.4. Deductions................................................................................................... 146
4.4. Influence of power quality on partial discharge patterns .................................... 148
4.4.1. Partial discharge modelling ......................................................................... 151
4.4.2. The statistical evaluation of the partial discharge patterns.......................... 161
4.4.3. Visual correlation with partial discharge pattern and statistical indices ....... 167
4.4.4. Deductions................................................................................................... 174
5. Summaries and outcomes...................................................................................... 176
5.1. Asset management ............................................................................................ 176
5.2. Test facility ......................................................................................................... 177
5.3. Lubricant coating................................................................................................ 178
5.4. Direction of electrical tree growth....................................................................... 179
5.5. Power quality and electrical tree growth ............................................................ 179
5.6. Power quality and breakdown times .................................................................. 180
5.7. Power quality and partial discharge patterns ..................................................... 181
6. Major contributions ................................................................................................. 183
6.1. Achievements..................................................................................................... 183
6.2. Conclusions........................................................................................................ 183
6.3. Further work ....................................................................................................... 184
References ..................................................................................................................... 185
Appendix A Weibull plots of breakdown times ....................................................... 194
Appendix B Partial discharge data and tree images............................................... 196
Appendix C List of publications................................................................................ 239
40,000 words
List of tables
Table 1-1: Electrical fields required for different electrical trees in polyethylene [46]. ..................... 24
Table 1-2: Categories and typical characteristics of disturbances in the power network [50].......... 26
Table 1-3: Voltage distortion limits [49]. ........................................................................................... 29
Table 1-4: Typical harmonic current relative to fundamental from common sources....................... 30
Table 1-5: Percentage of residual insulation life during peak periods [59]....................................... 34
Table 1-6: Percentage of residual insulation life during the entire day [59]. .................................... 34
Table 1-7: Quantitative analysis illustrating increased partial discharge activity as a
consequence of ageing with 50 Hz compared to 50 Hz + 10 % 11th harmonic [72]....... 36
Table 1-8: Summary of the impact of electrical stress factors on electrical ageing mechanisms. ... 38
Table 3-1: Properties of the seven test waveforms. ......................................................................... 76
Table 3-2: AWG vertical sensitivity. .................................................................................................. 79
Table 3-3: AWG horizontal resolution............................................................................................... 79
Table 3-4: Power amplifier parameters. ........................................................................................... 79
Table 3-5: Summary of problems, solutions & tradeoffs. ................................................................. 84
Table 3-6: Summary of problems, solutions & tradeoffs. ................................................................. 89
Table 4-1: Sample details to investigate lubricant effects. .............................................................113
Table 4-2: Interpretation of the width/length ratio. ..........................................................................116
Table 4-3: Densities of compounds. ...............................................................................................123
Table 4-4: Conclusions on the influence of the needle lubricant coating. ......................................124
Table 4-5: Breakdown results of 42 test samples...........................................................................138
Table 4-6: Graphically determined and values from the breakdown time data........................142
Table 4-7: Comparison of the 5th and 7th harmonic influence on breakdown time values...........145
Table 4-8: Influence of 7th harmonic magnitude on breakdown time values. ..............................146
Table 4-9: Variation in Weibull and values for breakdown times to corresponding Ks and
THD indices for each composite waveform. .................................................................147
Table 4-10: Description of samples tested. ....................................................................................148
Table 4-11: Upper quartile values for wave 7 is the lowest, while waves 12 and 11, exceed
wave 1 whose THD is significantly higher. ...................................................................167
Table 5-1: Variation in breakdown time and values to corresponding KS and THD (same as
Table 5-9)......................................................................................................................180
List of figures
Figure 1-1: Factors which influence insulation ageing......................................................................17
Figure 1-2: Dry ageing of polymeric insulation [8]. ...........................................................................18
Figure 1-3: Wet ageing of polymeric insulation [8]............................................................................22
Figure 1-4: Schematic representation of typical electrical tree growth [3]. .......................................23
Figure 1-5: Possible routes to electrical tree initiation [47]. ..............................................................25
Figure 1-6: Harmonic orders of 50 Hz fundamental (top) polluting the fundamental, influencing
the shape of the resultant (below)...................................................................................27
Figure 1-7: Links between power quality disturbances and electrical stress factors [65].................32
Figure 1-8: Influence of harmonic content on phased resolved partial discharge plots. A) Pure
test voltage B) 17 % - 3rd harmonic C) 11 % - 5th harmonic [73].....................................35
Figure 1-9: Partial discharge phase-resolved plots after 720 hours of 50 Hz ageing (left) and 50
Hz + 10 % 11th harmonic ageing (right) [72]. ..................................................................36
Figure 1-10: Simulation model (left) and experimental partial discharge activity (right)
incorporating 11 % of the 11th harmonic, showing good correlation to phase location
of discharge activity [73]..................................................................................................37
Figure 1-11: Typical overvoltage propagated from MV to LV network via a transformer [90]. .........41
Figure 1-12: Induced overvoltage on the line from indirect lightning i.e. strike to ground in the
vicinity of the line [95]......................................................................................................42
Figure 1-13: Typical overvoltages at varied locations from direct lightning strike to the line [90].....42
Figure 1-14: Asset managers management processes - the big picture [100]. The resource,
cost and work control loops are feedback loops which influence the control loop to
improve management of the physical assets..................................................................45
Figure 1-15: Asset management balance of costs, risks and performance [101].............................46
Figure 1-16: Pyramid of railway infrastructure condition monitoring highlighting the three major
contributors: maintenance policies, technologies and infrastructure [109]. ....................50
Figure 1-17: Classification of maintenance strategies [115]. ............................................................53
Figure 1-18: Scope of PAS 55 [101]. ................................................................................................57
Figure 1-19: Considerations for asset managers in a dynamically changing environment. .............59
Figure 2-1: Interaction of stress factors influencing the mechanisms of failure in context to the
asset managers decisions forming the asset management layer of the framework. .....62
Figure 2-2: Insulation failure flowchart. .............................................................................................63
Figure 2-3: Electrical stress factors...................................................................................................65
Figure 2-4: Ageing mechanisms are dynamic and may change in time as the material and local
stresses change [131]. ....................................................................................................67
Figure 2-5: Flow of information to the asset manager. .....................................................................69
Figure 2-6: Multifactor framework of insulation life. ..........................................................................71
Figure 2-7: Future development of the framework with defined asset management strategies
tailored to the companys business plan. ........................................................................74
Figure 3-1: Wave 1 THD=40 % Ks=1.56. ..........................................................................................77
Figure 3-2: Wave 7 THD=0 % Ks=1.00. ............................................................................................77
Figure 3-3: Wave 8 THD=5 % Ks=1.03. ............................................................................................77
Figure 3-4: Wave 9 THD 5 % Ks=1.06. .............................................................................................77
Figure 3-5: Wave 11 THD=17.8 % Ks=1.60. .....................................................................................77
Figure 3-6: Wave 12 THD=7.85 % Ks=1.60. .....................................................................................77
Figure 3-7: Wave 13 THD=5 % Ks=1.27. ..........................................................................................78
Figure 3-8: Slew rate variation with capacitive load for the amplifier at its 20 mA limit. ...................80
Figure 3-9: Overview of test equipment and their interfaces............................................................ 81

Figure 3-10: Test facility fully assembled. ........................................................................................ 82
Figure 3-11: Amplifier: 30 kV 20 mA................................................................................................. 82
Figure 3-12: High voltage testing area (top) with close up view of the camera and sample
under test (below) ........................................................................................................... 82
Figure 3-13: Schematic of high voltage test facility. ......................................................................... 83
Figure 3-14: Overview of partial discharge measuring system. ....................................................... 84
Figure 3-15: Straight detection circuit............................................................................................... 85
Figure 3-16: Measurement instrument - amplifier filter circuit. ......................................................... 86
Figure 3-17: Simulation circuit in MicroCap...................................................................................... 86
Figure 3-18: AC response................................................................................................................. 87
Figure 3-19: Transient response....................................................................................................... 87
Figure 3-20: Input 50 pC calibrating pulse........................................................................................ 88
Figure 3-21: Output 100 pC/V discharge pulse. ............................................................................... 88
Figure 3-22: Flowchart of the modules essential to the partial discharge measuring system.......... 89
Figure 3-23: Flowchart of modules 2 and 3 of the partial discharge instrumentation....................... 90
Figure 3-24: All input data point captured......................................................................................... 91
Figure 3-25: Input data points above noise threshold within 20 s window ..................................... 92
Figure 3-26: Partial discharge detected at sampling rate of 10 MSps.............................................. 93
Figure 3-27: Flowchart of process to produce PRPD plot. ............................................................... 94
Figure 3-28: Comparison of commercially available LDS 6 (left) and the in-house test facility
(right) showing good correlation of partial discharge activity from an electrical tree...... 95
Figure 3-29: Sampling rate of 5 MSps.............................................................................................. 96
Figure 3-30: Noise (magnified time scale on right plot) from energised high voltage amplifier at
output = 0 V, 100 pC/V. .................................................................................................. 97
Figure 3-31: Balanced detection circuit. ........................................................................................... 98
Figure 3-32: Implemented balance circuit integrated with the amplifier filter stages........................ 99
Figure 3-33: Noise from high voltage amplifier energised at output = 0 V, 100 pC/V from
straight circuit detection (left) and balanced circuit detection (right). ............................. 99
Figure 3-34: Power spectrum of FFT of sampled noise data. ........................................................100
Figure 3-35: Offset of input data points nullified at 100 pC/V (left) and 50 pC/V (right). ................101
Figure 3-36: Flowchart of hardware and software integration. .......................................................102
Figure 3-37: Sample screen shots of integrated software..............................................................103
Figure 3-38: Schematic of epoxy resin sample. .............................................................................104
Figure 3-39: Produced epoxy resin samples. .................................................................................104
Figure 3-40: Sample production rig. ...............................................................................................105
Figure 3-41: Physical setup of the sample in preparation for testing. ............................................105
Figure 3-42: Partial discharge patterns in ethylene-acrylic acid copolymer point-plane geometry
samples (gap = 10 mm, tip radius = 3 m), from artificial channel diameter = 40 m,
length = 2 mm (plot A) and length = 1 mm (plot B) both at 4 kV, while length = 2 mm
at 4.5 kV (plot C) [155, 160]..........................................................................................106
Figure 3-43: Partial discharge patterns in ethylene-acrylic acid copolymer samples at 12 kV for
point-plane geometry (gap = 12 mm, tip radius = 3 m), from electrical tree growth,
after 1 min (plot A), 35 min (plot B), 2 h (plot C), 6 h (plot D), 6 h 45 min (plot E) and
6 h 55 min (plot F) [157]................................................................................................106
Figure 3-44: Influence of aquadag and lubricant coating on discharge activity. ............................107
Figure 3-45: General plan for each sample under test. ..................................................................109
Figure 3-46: Flowchart of the experimental process to investigate the influence of power quality
on partial discharge due to electrical treeing. ...............................................................110
Figure 4-1: Hypodermic needles soaked for 12 days (upper) resulted in greater lubricant
retention conveyed by the glossy needle surface compared to needles soaked for 3
days (lower)...................................................................................................................112
Figure 4-2: Partial discharge activity from electrical trees of length 30 m at 50 Hz sinusoidal
reference with lubricant coating on needles (top) and without lubricant coating on
needles (below).............................................................................................................114
Figure 4-3: Illustration of typical tree growth A) with lubricant and B) without lubricant. ................116
Figure 4-4: Tree growth images for T444-07-Y with lubricant coating (left) and T213-07-N
without lubricant coating (right). ....................................................................................116
Figure 4-5: Reduced electrical tree length and width measurements with lubricant coating
compared to measurements without lubricant coating. ................................................117
Figure 4-6: 3D plots showing reduced width/length ratios for samples with lubricant coating
relative to samples without lubricant coating. ...............................................................118
Figure 4-7: Scatter of initiation and breakdown times with and without lubricant coating. .............120
Figure 4-8: Illustration of gaseous activity.......................................................................................121
Figure 4-9: Gas percentage vs cycles in the electrical tree channels of polyethylene. 10 s
pause between the full cycles 50 Hz, 30 kV (gap = r mm, electrode tip radius = 5
m) [164]. ......................................................................................................................122
Figure 4-10: Simple example with dimensions of tree channel. .....................................................123
Figure 4-11: Sequence of electrical tree growth for sample T345-09-N.........................................126
Figure 4-12: Plot of electrical tree length vs time of all samples. Inset the cluster of 2 mm tree
length () and scatter of breakdown () points. T325-09-N exhibits significant
growth relative to all samples........................................................................................131
Figure 4-13: Normalized plot of electrical tree length vs time of all samples. Insulation gap of
length = 2 mm used as reference. Lengths 2 mm registered due to branches
growing upward beyond the needle tip e.g.T325-09-N.................................................132
Figure 4-14: Plot of width/length ratio vs time of all samples. Inset the cluster of 2 mm tree
length () and scatter of breakdown () points. ............................................................133
Figure 4-15: 3D plot of width/length ratio for all samples highlighting scatter of markers. ..........134
Figure 4-16: 3D plot of width/length ratio as a function of THD for all samples..............................135
Figure 4-17: 3D plot of width/length ratio as a function of Ks for all samples. ................................136
Figure 4-18: Breakdown time vs THD illustrating the mean and standard deviation......................139
Figure 4-19: As THD increased at constant peak voltage, the variation in breakdown trends did
not reveal a deterministic relationship with THD. Lines are not for best fit or trend
purposes but to assist the reader identify result groups ...............................................140
Figure 4-20: Breakdown time vs Ks illustrating the mean and standard deviation..........................140
Figure 4-21: As Ks increased at constant peak voltage, the variation in breakdown revealed a
potential region at Ks=1.27 for maximum breakdown times. Lines are not for best fit
or trend purposes but to assist the reader identify result groups..................................141
Figure 4-22: Weibull plots with and values for the total population of tested samples,
subsets of Ks=1.60, THD=5 % and undistorted waveform where Ks=1.0 & THD=0
%. ..................................................................................................................................143
Figure 4-23: The probability density and cumulative distribution function plots are similar
shapes except wave 8 and wave 1 which correspond to minimum and maximum
values respectively containing the highest values. Inset at T = 4000 s wave 9 is
most influential. .............................................................................................................144
Figure 4-24: Phase-resolved partial discharge plots for sample K115. ..........................................150
Figure 4-25: Time domain representation of derivatives and electrical treeing partial discharge
activity captured from wave 7 (plot 6) and wave 13 (plot 8) from K106 for one
acquisition (80 ms). .......................................................................................................151
Figure 4-26: Changes of voltage in partial discharge (PD) source at pure and harmonic test
voltages in a solid dielectric with a) PD source (void); t=thickness of void b)
Equivalent circuit diagram (abc), where Ca=capacitance of solid dielectric,
Cb=capacitance of solid dielectric in series with void and Cc=capacitance of void c)
PD mechanisms at pure sinusoidal test voltage d) Effect of harmonics in test
voltage on void voltage Uc and PD [73]. .......................................................................153
Figure 4-27: Wave 13 discharge pattern compared to the V and dV/dt plots from four tests. .......154
Figure 4-28: Plots of the cosh and sinh hyperbolic functions illustrating potential to model
dead zones of partial discharge activity. .....................................................................155
Figure 4-29: Normalisation of the waveform to prevent operation in the asymptotic region of the
hyperbolic functions cosh and sinh with amplitude = 1 (left) and amplitude = 10
(right).............................................................................................................................156
Figure 4-30: Improved modelling of partial discharge patterns with cosh(V+dV/dt). The dotted
circles highlight improved dead zone recognition. ......................................................157
Figure 4-31: K106 partial discharge patterns due to the composite waveforms with comparison
to the cosh(V+dV/dt) model. The dead zones highlighted by the dotted circles do
not fully correlate with the recorded partial discharge activity. .....................................158
Figure 4-32: Improved dead zone recognition highlighted by the dotted ellipses for K101 for
waves 9 and 8...............................................................................................................159
Figure 4-33: Typical tree growth curve with suggested operating region for further experiments
investigating electrical tree partial discharge modelling. ..............................................160
Figure 4-34: Comparison of cosh(V+dV/dt) to partial discharge activity for a triangle waveform. .160
Figure 4-35: Electrical treeing partial discharge activity due to a triangle wave illustrating the
magnitude of discharge was related to not only the instantaneous voltage [159]........161
Figure 4-36: Weibull plots of negative and positive discharges for sample K115. .........................163
Figure 4-37: Linear best fit plots of charge magnitude and values showing no dependence
on THD and Ks. .............................................................................................................164
Figure 4-38: Box and whisker plots of determined Weibull values for all waveforms,
combined charge polarities as well as positive and negative charge polarities. ..........165
combined charge polarities as well as positive and negative charge polarities. ..........166
Figure 4-40: Electrical tree growth images for K115. Each plot = 2 mins, 14.4 kV peak. ..............168
Figure 4-41: Electrical tree growth images for K104. Each plot = 4 mins, 10.8 kV peak. ..............169
Figure 4-42: Test K104 Weibull analysis of negative charges for plots 1-14 showing the
variation during plots 6, 7 and 8 as a result of the sudden tree growth, accompanied
by increased magnitude partial discharge activity. .......................................................169
Figure 4-43: Graph showing the sudden change between plots 6-8 for characteristic and
values of test K104, indicating a change in the insulation state. ..................................170
Figure 4-44: Progression of partial discharge pattern with and values for successive tests
K112 and K113 on sample T273. Plots show change in discharge patterns
suggesting change of dominant ageing mechanism and change in state of
insulation.......................................................................................................................171
Figure 4-45: Visual images showing electrical tree growth for tests K112 and K113. ...................171
Figure 4-46: Weibull plots of wave 1 discharge activity for K112 and K113 illustrating the
variability due to minute changes in electrical tree growth preventing consistent
plots. .............................................................................................................................172
Figure 4-47: Example of spread of Weibull plots. Waves 13 and 12 have similar scatter while
the other waveforms clustered together. Relative to wave 7 (fundamental) each
composite waveform has a different scatter as a result of different partial discharge
patterns. ........................................................................................................................173
10
List of abbreviations
AWG
Arbitrary Waveform Generator
BD
Breakdown
BSI
British Standards Institute
CBM
Condition Based Monitoring
CCD
Charge Coupled Device
CM
Corrective Maintenance
CML
Customer Minutes Lost
EL
Electroluminescence
EPR
Ethylene Propylene Rubber
ET
Electrical Tree Growth
FFT
Fast Fourier Transform
HV
High Voltage
IAM
Institute of Asset Management
IEC
International Electrotechnical Commission
IEEE
Institute of Electrical & Electronic Engineers
IMD-UMS
Integrated Mechanical Diagnostic Health & Usage System
ISO
International Organization for Standardization
LV
Low Voltage
MCM
Motor Condition Monitoring
MFCP
Maintenance Finite Capacity Planning
MI
Measurement Instrument
MV
Medium Voltage
NI
National Instruments
ODR
Operator Driven Reliability
OFGEM
Office of Gas and Electricity Markets
OHSAS
Occupational Health & Safety Advisory Services
PAS
Publicly Available Specification
PCI
Peripheral Components Interconnect
PD
Partial Discharge
PET
Polyethylene Terephthalate
PP
Polypropylene
PRPD
Phase-resolved Partial Discharge
PTFE
Polytetrafluoroethylene
PVC
Polyvinylchloride
RAMSYS
Rail Asset Management Systems
RCM
Reliability Centered Maintenance
TBM
Time Based Maintenance
TDD
Total Demand Distortion
THD
Total Harmonic Distortion
TIV
Tree Inception Voltage
TTL
Transistor Transistor Logic
WT
Water Tree Growth
XLPE
Cross-linked Polyethylene
11
Abstract
Power network operators in developed countries are faced with the challenge of
effectively managing network performance with an ageing asset population. A significant
proportion of equipment is already operating well beyond design life, testifying to the
success of the many insulation systems employed. Increased production of renewable
and distributed energy has resulted in changes of load flows on the network, while
demand-side management schemes cause variation in load demands. A steady rise in the
number of power electronic devices results in reduced power quality from disturbances
including harmonics. Consequently, there is a gradual change in the working environment.
Hence at the plant level, insulation systems will age differently influencing electrical
ageing mechanisms such as partial discharges and electrical treeing.
This research encompasses the plant level, where diagnostic data is interpreted to
determine asset management decisions, at the system level. A novel structured
framework has been developed linking the physics and chemistry of insulation
degradation as well as the management of network power quality, to plant reliability and
asset management. The development of a test facility for electrical treeing investigations,
using composite waveforms uniquely consisting of six harmonic components has been
described. The conducted experimental studies sought to qualitatively and quantitatively
identify any distinguishing features of partial discharges and electrical tree growth
characteristics, as a consequence of harmonic content impacting power quality. In power
network and laboratory research the power quality dynamically varies, although this is
often not monitored. In this research, the total harmonic distortion (THD) and waveshape
(Ks) indices were varied to a maximum of 40 % and 1.6 respectively. Electrical trees were
developed in point-plane geometry using 2 m tip radius hypodermic needles and a 2 mm
gap in epoxy resin (LY/HY 5052) samples at a constant voltage of 14.4 kV peak.
The results illustrated firstly, a return growth of the electrical tree from the ground
electrode towards the needle tip after the original (downward) growth of the electrical tree
(from the needle tip to the ground electrode) traversed the insulation gap. Secondly, no
changes were detected in electrical tree growth characteristics due to variation of
harmonic content in the excitation voltage. Thirdly, composite waveforms with increased
magnitude of the 7th harmonic resulted in reduced failure times and low values of the
Weibull shape parameter describing the increased scatter of these times. Penultimately,
the composite waveforms influenced the partial discharge pattern produced, leading to
possible misinterpretation of the dominant ageing mechanism. If this change in partial
discharge activity is a result of an unmonitored change in power quality, overestimation of
the insulations ageing state will occur resulting in inappropriate asset management
decisions taken. Finally, modelling partial discharge activity due to electrical treeing with
dV
the function cosh V +

provided a good fit for identifying locations of partial discharge
dt
peaks on the phase-resolved partial discharge plots and also identified periods of low
discharge activity.
It is concluded that at constant peak voltage, harmonic content influences electrical
ageing mechanisms and further investigation of the role of the 7th harmonic is required.
12
Declaration
No portion of the work referred to in the thesis has been submitted in support of an
application for another degree or qualification of this or any other university or institute of
learning.
Copyright
(1) Copyright in text of this thesis rests with the Author. Copies (by any process) either in
full, or of extracts, may be made only in accordance with instructions given by the author
and lodged in the John Rylands University Library of Manchester. Details may be obtained
from the Librarian. This page must form part of any such copies made. Further copies (by
any process) of copies made in accordance with such instructions may not be made
without the permission (in writing) of the Author.
(2) The ownership of any intellectual property rights which may be described in this thesis
is vested in The University of Manchester, subject to any prior agreement to the contrary,
and may not be made available for use by third parties without the written permission of
the University, which will prescribe the terms and conditions of any such agreement.
(3) Further information on the conditions under which disclosures and exploitation may
take place is available from the Head of the School of Electrical and Electronic
Engineering.
13
Acknowledgements
I would like to thank and praise the Almighty for His continued guidance and support. With
Him all things are possible and thy will be done.
I can only attempt to express my deepest gratitude to my supervisor, Dr. Simon Rowland.
His guidance, support and encouragement during my PhD study have been a beacon in
the darkest hours. Thank you Simon.
A special thank you to the EPSRC Supergen consortium for their financial support.
I would also like to extend my gratitude to the other academics in the department for their
invaluable guidance and contributions in technical discussions.
A special thank you to Bobby, Anabel, Anish and Nicole who have supported me on my
quest for excellence. Most importantly, thank you for your prayers.
Many thanks to all my colleagues at The University of Manchester who have assisted me
in various ways.
Dedication
to my parents
14
Chapter 1. Introduction
1. Introduction
1.1.
Supergen V - AMPerES
Supergen V is a collaborative research partnership amongst six leading UK universities,

nine industrial partners of the electrical energy sector and the Engineering and Physical
Sciences Research Council (EPSRC). The theme of Supergen V is Asset Management
and Performance of Energy Systems - AMPerES. The broad aims of Supergen V include:
To deliver intelligent diagnostic tools for plant; enabling optimum usage,

maintenance and replacement.
To provide platform technologies for integrated network planning and asset

management.
To investigate alternative plant and reduce environmental impact of networks.
To develop models and recommendations for improved network operation and

management in terms of economic performance and new generation
connections.
1.2.
Background
Maintaining a reliable energy supply at minimal cost is a requirement of any power

system. The challenge of this task is increasing in this era of ageing plant with a global
drive to increase the integration of renewable and distributed generation. Hence gradual
evolution of the network configuration to incorporate new generation and demand trends
result in each individual item of plant experiencing a change in its working environment
and conditions. These changes include:
New forms of generation and changes in the load characteristics.

Consequently this will change network flows, fault currents, thermal loading of
plant equipment and voltage profiles.
Increased usage of power electronic devices which alter the natural waveform
introducing high frequency sinusoids and pulse trains.
Modification in scheduled maintenance procedures.
Consequently, there is a gradual change in the working environment experienced by

insulation systems. As networks undergo this evolutionary process, the insulation systems
will age differently. As an example of network evolution, we might consider a part of the
network which was previously subjected to a steady low level of loading. The network
15
equipment might be old, but not considered as aged since it has not been highly thermally
stressed. Additionally, the plant may be in a location where reliability was not critical and
so maintenance may not have been a priority. However, if that location is now part of a
wind farm connection link, there is a need for high reliability to facilitate transmission unto
the network. Therefore the plant may be highly loaded at given intervals. Consequently,
we might expect more extreme and more regular thermal excursions than previously
experienced. Similarly, connection of non-linear loads lead to increased network harmonic
content and reduced power quality, not experienced previously. Harmonics can potentially
result in significant changes of the time-domain features of the power frequency waveform
i.e. from a pure sinusoidal to a non-sinusoidal (distorted) waveform. This will increase
electrical and thermal ageing of the insulation. Intrinsic contaminants, imperfections,
protrusions and voids remain in these insulation systems and will continue to play a major
role influencing ageing and failure mechanisms. Thus to improve the interpretation of
captured diagnostic data, increased understanding of dielectric ageing under such nonpower frequency conditions is required.
This rapid metamorphosis of the power system network illustrates that there are some key
issues relating to asset management which must be fully understood to efficiently manage
the networks ageing assets. There is therefore a need to link performance of individual
items of plant to system performance addressing the diagnostic needs of the network
operators, equipment suppliers and service companies.
1.3.
An interpretation of insulation ageing
Champion et al. [1] defined ageing as the reflection of the chemical and physical changes,
in electrical materials or electrical systems resulting from stresses with the passage of
time. However, ageing is much more complex and Figure 1-1 offers a more detailed
perspective. Figure 1-1 is by no means exhaustive but provides a good platform to
appreciate and improve comprehension of the multifactor nature of insulation ageing.
16
INSULATION
BREAKDOWN
accelerate with
electric field
accelerate with
electric field
PHYSICAL
CHEMICAL
THERMAL
ELECTRICAL
MECHANICAL
Time
Environment
Usage
Ultraviolet
Humidity
Ionizing
Radiation
Oxidation
Gases
Chemicals
Joule Heating
Dielectric Heating
Eddy Currents
Temp Cycling
Temp Gradient
Voltage AC, DC
Impulses
Polarity
Frequency
Current
Tensile Stress
Compressive
Stress
Vibration
Bending
Torsion
Figure 1-1: Factors which influence insulation ageing.
Physical ageing is affected by all the other ageing types either directly or indirectly.
Physical and chemical ageing exist without an electric field but application of an electric
field accelerates the degradation process. Additionally, both are influenced by
environmental factors but Bonten et al. [2] identified that for polymers, physical ageing
mechanisms were reversible (as long as physical rupture has not occurred) in contrast to
chemical ageing mechanisms which irreversibly modify the polymer structure. Physical
ageing affected the molecular arrangement of the polymer structure and its intermolecular forces [2] as a result of the inability of polymer chain bonds to return to their
equilibrium position after thermal or mechanical stress [3]. Chemical ageing is primarily
due to oxidation and molecular bond breakage through events (some outlined in Figure
1-1) liberating electrons and ions. Fundamentally, chemical ageing may either enhance
the electric field or cause a reduction in the breakdown strength [3]. Thermal ageing, often
a by-product of electrical stress leads to physical ageing changing the insulations
microscopic structure influencing its chemical stability [3, 4]. Mechanical ageing is
influenced by mechanical stress which may have resulted during the manufacturing and
transportation phases of the insulation system and even whilst in operation from
electrodynamics and thermal forces [5]. Mechanical ageing significantly influences
physical ageing and may accelerate electrical ageing. Compressive stress results in
breakage of bonds which generate defects in the insulation whereas tensile stress results
in crack initiation and growth allowing molecular chains to rotate, translate, unfold and
disentangle [3]. Thus mechanical stresses aid crack propagation allowing space charge
17
deposition. Consequently the crack may lengthen leading to mechanical failure and or
partial discharge activity leading to electrical tree formation and eventually breakdown.
The main degradation mechanisms of electrical ageing in solid polymeric insulation are:
Space charge accumulation
Partial discharge (PD)
Water tree growth (WT)
Electrical tree growth (ET)
Partial discharges and water trees may be predecessors for electrical trees as seen in
Figure 1-2. Partial discharge activity is one of the most prominent indicators of defects and
on-going degradation processes in electrical insulation systems, thus is it the primary
online and offline diagnostic tool employed [6, 7]. A significant degree of research has
already been conducted to identify factors affecting the degradation mechanisms listed
above in many insulation systems and environmental conditions [8]. One example in
Figure 1-2 illustrates dry ageing in polymeric insulation and highlights intrinsic and
extrinsic ageing. Intrinsic ageing is defined as irreversible changes of the fundamental
material properties in an insulation system caused by ageing factors. Conversely, extrinsic
ageing is a source of irreversible changes of insulation properties stemming from the
ageing factors acting on imperfections in the insulation system [5].
Figure 1-2: Dry ageing of polymeric insulation [8].
18
1.4.
Space charge
Space charge is the net difference between positive and negative charge (electrons,
protons, ions) present in a dielectric. The presence of space charge, enhances or reduces
the local electric field [9-11], influencing partial discharge activity, electrical tree growth
and thus eventual failure of the insulation system. Polymeric insulation systems contain
micro-voids produced during the manufacturing process [12]. The differences in the
permittivity of the air and the polymer will enhance the field in the void but an initiating
electron is required to start partial discharge activity. Space charge is injected from the
surface of the electrode into the insulation [13] and the charges gain energy from the
applied field and lose it through collisions with the polymer [8]. Hence, the initiating
electron is transported through the insulation by the conduction process and is trapped as
space charge at the polymer-void interface [12]. Once the critical electric field at the
interface is exceeded, these charges are injected into the void, accelerated by the electric
field and ionize gas molecules giving rise to hot-electron avalanches. These collisions
cause damage to the lattice and accumulate on the opposite end of the void-polymer
interface depositing electrons and positive ions on the cavity walls. This repetitive process
leads to the formation of voids and growing pits in the polymer leading to electrical treeing
[12]. Significant research has confirmed that space charge injection has been observed at
field magnitudes in the range of one-fifth to one-third the magnitudes required for
breakdown in homogenous dielectrics and one-tenth the magnitude for inhomogeneous
dielectrics [14]. This confirms that the space charge can influence electrical ageing at
comparable rated voltages. In polyethylene the critical field for space charge injection is
100 kV/mm [14, 15] while for tree initiation it is 500 - 700 kV/mm. In epoxy resins the
critical field space charge injection 300 kV/mm [14]. Indirect evidence for space charge
is quantified by the intensity of electroluminescence activity [8].
1.4.1. Electroluminescence
The presence of space charge has also been associated with the occurrence of
electroluminescence (EL). EL occurs prior to partial discharge inception in polymeric
insulation at high voltage and there is no measurable degradation of the polymer below
EL inception voltages [16, 17]. EL represents one of the few measurable quantities that
accompanies the electrical tree initiation process and electrical ageing [10, 18, 19].
Dissado et al. [20] described a model involving bipolar injection, trapping and
recombination of mobile and trapped charges [8, 21]. The explanation highlighted that on
one half cycle of a waveform, mobile injected charge recombines with trapped charge of
19
opposite polarity, thereby reducing their concentration and producing a pulse of EL. The
remaining space charge is trapped resulting in an accumulation of space charge of the
same polarity as the injecting electrode (homocharge) reducing the local electric field. In
the following half cycle the same processes occur again leading to EL and a polarity
reversal of the space charge [20]. No field threshold is necessary for recombination to
occur, only bipolar injection and trapped charges are required.
Ignoring space charge effects, the applied fields necessary for electroluminescence in
epoxy resins are in the range 200 - 800 kV/mm [22]. EL pulses were identified by 2 ns rise
and fall times with 10 ns pulse widths. As the voltage increased there was a noticeable
shift of activity toward shorter wavelengths [22]. EL activity was captured over the entire
visible light spectrum with the maximum activity occurring at 500 nm [22, 23]. Evidence
suggested EL emission occurred within the ultraviolet spectrum beyond 300 nm albeit self
absorption in the material occurred [23], resulting in a total bandwidth of 375 - 725 nm for
epoxy [22] and 300 - 600 nm for cross-linked polyethylene (XLPE) [24]. The energy of EL
photons can be responsible for breaking chemical bonds [20] and inducing chemical
damage of the dielectric [10]. The ultraviolet radiation can cause photo dissociation,
photochemical reactions and charge transfers which create free radicals, promote bond
scissions and it is thought creates a micro cavity in which partial discharges can occur and
lead to electrical treeing [18].
1.5.
Partial discharges
A partial discharge can be defined as localized electrical discharge that only partially
bridges the insulation between conductors. It may or may not occur adjacent to a
conductor [25]. Partial discharges can be categorized as a symptom and a mechanism
associated with insulation degradation [7]. Partial discharge activity can occur at operating
voltages in electrical trees, voids, cuts, cracks and at fillers or contaminants with poor
adhesion to the polymer and delaminating sites at interfaces of the insulation [26]. Partial
discharges are, in general, a consequence of local electrical stress concentrations in the
insulation due to voids, contaminants, protrusions and defects on the surface of the
insulation. Voids may form as a consequence of electrostrictive forces due to the applied
field and by electrochemical effects such as water treeing [3]. Partial discharges can be
described by pulses with rise-times as short as 1 ns [27] and are often accompanied by
emissions of sound, light and heat as well as chemical reactions [25]. The magnitude of
partial discharges is proportional to the size of the degradation site. The frequency of
discharges is an indication of the number of degraded sites in the insulation system. Other
20
parameters such as the phase relation and applied voltage magnitude compliment such
information to provide more accurate estimates about the nature and extent of partial
discharge activity taking place in the insulation system [28].
Partial discharge patterns served as unique signatures to identify sources of defects and
ageing states of insulation systems, with the aid of artificial intelligent techniques [7, 2937]. Hence, provided adequate data parameters describing the partial discharge patterns
are available, identification of an existing defect such as a void or an electrical tree is
possible [26, 38]. Partial discharge activity in voids will either increase the conductivity of
the void walls, extinguishing discharge activity, or erode the walls forming pits eventually
leading to the inception of electrical trees [26].
1.6.
Water trees
A water tree is a propagating dendritic pattern of water-filled voids which over time
increases in length [39]. Water trees consist of strings of hydrophilic micro voids (which
were originally hydrophobic before a chemical change such as oxidation occurred), of the
order of 1 m diameter filled with water [40]. Water trees have been present in a variety of
polymer based insulation systems and have been a major mechanism of in-service cable
failures over an extended period of time. These water trees can be classified as:
Vented trees have a stem joining them to the surface of the insulation and are
therefore in direct contact with a reservoir of aqueous electrolyte [3].
Bow-tie trees which originate from a contaminant, boundary surface or water

filled void within the insulation where there is limited access to an aqueous
reservoir [3].
Water trees occur at much lower fields than those required for electrical trees. Fothergill et
al. [40] highlighted that the water tree inception rate was more dependant on the electric
field than the applied voltage. The conditions for water tree manifestation must be
conducive and include factors such as pH, type and concentration of the electrolyte [40].
Densley et al. [8] reviewed polymeric insulation in wet environmental conditions as
depicted in Figure 1-3. Under wet conditions, the degree of moisture influencing the
insulation system is quite significant and the voids are likely to be filled if not partially filled.
These voids suppress discharges but become initiation sites for bow-tie trees [8].
21
Figure 1-3: Wet ageing of polymeric insulation [8].
While water trees represent one form of polymeric insulation degradation, water trees can
cross the insulation without causing insulation failure but can also initiate an electrical tree
[3].
1.6.1. Transition from water trees to electrical trees

Boggs et al. proposed a mechanism for the conversion of water trees to electrical trees
under impulse conditions by the electro-thermo-mechanical phenomena [41]. Boggs et al.
postulated that the impulse voltage induced a transient current causing the water in the
tree channel to boil, creating a void which supports partial discharge activity and
eventually electrical tree initiation [41]. This study revealed that water trees did not cause
failure of in-service XLPE cables, instead electrical trees were initiated from water trees as
a result of lightning surges. This explained the frequent failure of cables after torrential
rain and lightning [41]. Densley et al. contributed by highlighting that significant oxidation
may occur in water trees at high temperatures leading to an increase in water absorption,
higher conductivity and eventual thermal runaway [8]. The main cause for this transition
seems to be temperature, and if the degree of thermal exposure is reduced the transition
time to an electrical tree will be lengthened. This argument is quite consistent with the
mechanism Boggs et al. [41] discussed resulting in the time frame for such a transition
having a large spread.
22
1.7.
Electrical trees
An electrical tree is a path of damage incurred by polymeric insulation as a consequence

of electrical stress and resembles the shadow of a tree [42]. Electrical trees are easily
initiated at sites of defects and once initiated can develop over a period of time to bridge
the insulation and eventually cause failure [43]. Electrical trees can also initiate from
eroded surfaces in a void, water trees and also stress enhancements without voids [26].
Electrical trees are composed of micrometre (m) diameter and length hollow channels
[3].
Figure 1-4 outlines three distinct stages of electrical tree growth. The inception stage is
characterized by a finite inception time, the propagation stage exhibits a decelerating
growth rate (similar to water trees) which then accelerates leading into the runaway stage
before breakdown. During the propagation stage, extension of the tree channels occur
with the main channels expanding into diameters > 10 m, discharges 100 pC
accompanied by increased acoustic and light emissions [3].
Figure 1-4: Schematic representation of typical electrical tree growth [3].
Auckland et al. [44] explained that tree growth was controlled by the number of discharges
and the residual charges in existing tubules (fine channels). The residual charges in their
respective tubules prevented further discharges in that tubule forcing new tubules to be
formed [44]. The light intensity emitted by partial discharges is 100 times more intense
than that from EL and by monitoring the light radiation from the sample, the transition from
tree initiation to tree growth is readily identified [22]. Increased temperature decreases the
23
inception time of an electrical tree and increases the tree growth rate. In the runaway
stage the leading channels are typical channels of the inception stage i.e. very thin, < 3
m and magnitude discharges < 5 pC [3].
1.7.1. Electrical tree types

Dissado et al. classified electrical trees into 3 types [3]:
A branch tree which has multiple branched structures with channel diameters
of the order of tens of microns ( 30 m) in the main channel (trunk) to one
micron ( 1 m) in the channel tips (branches) [3].
A bushy tree where the tubules are densely packed [3].
A bush-branch tree which is primarily a bush tree with one or more branches
projecting [3].
Jiang et al. [45] explained that branch tree channels were semiconducting and discharges
occur near the tips of the branches, distorting the electric field and reducing the likelihood
of branch formation along the tree channel. Conversely bushy tree channels cannot be
conducting since one channel would effectively short the field required to produce
discharge in other channels [45]. Jiang et al. continued to suggest that the channels are
probably full of surface charge resulting in a wildly distorted field pattern within the bush
tree thus giving rise to the random directions of the tree channel [45]. An increase in the
applied field results in a transition from branched tree to bushy tree to bush-branched tree
as illustrated in Table 1-1 [46]. This was confirmed by Jiang et al. [45] and Guastavino et
al. [38] who provided evidence confirming that branch like tree growth resulted in faster
breakdown than bushy type tree growth.
Table 1-1: Electrical fields required for different electrical trees in polyethylene [46].
Tree Type
Branch
< 540
Bushy
540 - 600
Bush-Branch
24
Field Magnitude (kV/mm)
> 600
1.7.2. Electrical tree initiation

The incubation period as seen in Figure 1-5 is defined as the time required for tree
initiation from the time the voltage is applied [47]. This can be defined as the time for a
tree length of 10 m to be formed [3]. Dissado et al. [3] defined the onset of tree inception
prior to a visible tree channel when 0.04 - 0.30 pC discharges are detected. In addition to
the space charge mechanisms which initiate an electrical tree, Noto et al. [46] also
attributed dielectric heating as a source of void generation and Maxwell stress as initiating
cracking around the needle tip. The resultant dielectric field would be reduced resulting in
partial discharges. Shimizu et al. [13] outlined two possibilities for tree initiation; long term
leading to gas filled cavities or short term leading to local field enhancements. Long term
tree initiation times (>> 1 s) under low voltage results in void formation. This is the
consequence of a cumulative degradation process which generally has been observed for
AC stresses, although similar processes can operate under repetitive pulsed voltages.
Tree initiation within a short time (< 1 s) under high voltages occurs under impulse, DC
and AC voltage applications. Tree initiation occurs when the local field exceeds the
breakdown field leading to localized electron avalanches and local breakdown [13].
APPLIED VOLTAGE
Extrinsic
Process
Charge Injection &

Extraction
Electrostrictive
Force
Void Discharge
Material Ageing due

to Other Processes
Pit Formation
Decrease in
Breakdown Voltage
Incubation Period
Intrinsic
Process
Electric Field Induced

Ageing
Small Gap
Formation
Field Distortion
Charge Accumulation
Joule Heating
Oxidation
Intensified Electric
Field
Weak Channel
Formation
Decrease in
Breakdown Voltage
Partial
Discharge
TREE INITIATION
Figure 1-5: Possible routes to electrical tree initiation [47].
25
1.8.
Power quality and electrical ageing
To a power engineer, power quality concerns powering and grounding of equipment to

ensure successful operation [48-50]. Transient, short or long term, as well as steady state
disturbances impact the networks power quality. Table 1-2 provides a condensed table of
such disturbances extracted from the IEEE 1159 standard, Recommended Practice for
Monitoring Electric Power Quality [50].
Table 1-2: Categories and typical characteristics of disturbances in the power network [50].
Categories
Transients
Short duration
Long duration
Spectral content
Duration
Impulsive
0.5 ns 0.1 ms rise
ns ms
Oscillatory
kHz MHz
ms s
0 8 pu
Interruption
0.5 cycles 1 min
< 0.1 pu
Sag
0.5 cycles 1 min
0.1 0.9 pu
Swell
0.5 cycles 1 min
0.1 1.8 pu
Interruption
> 1 min
0.0 pu
Undervoltage
> 1 min
0.8 0.9 pu
Overvoltage
> 1 min
1.1 1.2 pu
steady state
0.5 2.0 %
steady state
0.0 0.1 %
Voltage imbalance
DC offset
Waveform distortion
th
Harmonics
1 100 H
steady state
0.0 20 %
Interharmonics
0 6 kHz
steady state
0.0 2.0 %
Notching
Noise
Voltage fluctuation
Power frequency
Voltage
steady state
broadband
steady state
0.0 1.0 %
< 25 Hz
intermittent
0.1 7.0 %
< 10 s
Short duration sags and swells, as well as long duration overvoltages and undervoltages,
influence the rms, amplitude and rise-time voltage attributes. The rise-time denotes a rate
of change of voltage which the insulation experiences, e.g. under fault conditions. Swells
and overvoltages can represent increased electrothermal stressing depending on the
peak, rms and duration of such disturbances. An insulation component at a particular site
in the network can be electrically vulnerable due to weak system impedance and so suffer
from a poor quality of supply e.g. as a result of harmonics. Voltage waveform distortion
levels from harmonic phenomena and transient disturbances observed on the
transmission and distribution networks are an important problem. These non-power
26
frequency disturbances have resulted in new working environments for ageing insulation
systems.
1.8.1. The role of harmonics

The harmonic of a wave is an integer multiple of the fundamental frequency. The
fundamental is the 1st harmonic. In power engineering the fundamental frequency is 50 or
60 Hz depending on the network, realising harmonics of 100 or 120 Hz, 150 or 180
Hzetc, shown in Figure 1-6.
Harmonic orders
Magnitude/Per Unit
Fundamental
2nd Harmonic
3rd Harmonic
4th Harmonic
0.5
0
-0.5
-1
0
10
Time/ms
12
14
16
18
20
Harmonic influence on resultant
Magnitude/Per Unit
Fundamental
Resultant
10% 3rd Harmonic
20% 5th Harmonic
30% 7th Harmonic
0.5
0
-0.5
-1
0
10
Time/ms
12
14
16
18
20
Figure 1-6: Harmonic orders of 50 Hz fundamental (top) polluting the fundamental,

influencing the shape of the resultant (below).
On power networks there may be an unacknowledged harmonic contribution. This may

also be true for reported experimental laboratory work. It is therefore important to know
the thresholds at which power quality affects insulation ageing. Comprehending the effect
of the abnormalities in any resultant waveform experienced by the insulation is crucial to
defining which time-domain waveform features and characteristics are most influential to
insulation degradation. Only then can the crucial harmonic combinations be identified.
The sources of harmonics are non-linear loads, consuming non-sinusoidal currents, from
a sinusoidal voltage on the distribution networks. These non-sinusoidal currents are
27
harmonic currents yielding harmonic voltages across the network impedances. This
distortion of the power frequency waveform on the grid, commonly known as harmonic
pollution mainly results from the use of power electronic devices such as adjustable speed
drives, switching power supplies, inverters and other high speed switching devices [48].
Other sources of harmonics include static power converters, electric arc furnaces,
electrical equipment with magnetic cores (transformers and motors) and static var
compensators. Other disturbances originate from utility switching, fault clearing and
lightning. This distortion of the sinusoidal waveform, often leads to malfunctioning of
sensitive electronic devices, unexpected tripping of relays due to high harmonic currents,
overheating and accelerated electrothermal ageing of cables, motors and transformers,
reducing the functional life of electrical components [48, 51, 52].
Harmonics, described as a frequency-domain representation of time-domain occurrences
[53] may increase the peak, rms and rate of change of an electric field within a dielectric,
increasing dielectric losses whilst creating a temperature rise within the dielectric [48, 54].
High frequency harmonics may lead to increased Joule heating and mitigation is achieved
by derating transformers and cables [48, 49, 55, 56]. However, thermal stressing of the
dielectric still occurs. Thermal runaway of the insulation due to high harmonic currents
may be mitigated by engaging network protection. The transients introduced with
protection switching further pollute the network. For example SF6 and vacuum switchgear
produce hundreds of surges during switching operations with 300 ns rise-times and rates
of 200 - 3000 surges per second [57].
The 3rd order harmonics and the multiples of this frequency commonly known as the
triplen harmonics pose a serious threat to thermal overloading neutral cables [52]. Such
harmonic components contain a zero sequence and the vector addition of each harmonic
phase current results in a magnitude three times the fundamental in the neutral cable.
Major sources of such components include electronic ballasts, switch mode power
supplies and personal computers, all connected at low voltages [48]. The triplen
harmonics are usually filtered from propagating to higher voltage levels with the delta star
transformer [55]. Thus on the distribution networks the 3rd, 5th and 7th harmonic orders
dominate, whereas the 5th and 7th would be the most influential to power quality and
potentially to insulation system failure at increased voltage levels of transmission.
The harmonic components produced by pulse converters are governed by expression (11), where n is the integer number (n = 1, 2, 3) and p = the number of rectifiers in the
circuit.
28
n p 1
(1-1)
Thus the characteristic 5th, 7th and higher order harmonics consistent with expression (1-1)
are produced by 6-pulse motor drives [55]. The 5th and 11th harmonic components have a
negative sequence since the motor is being driven in reverse and might induce
overheating or cause over-current protection devices to operate. The 5th harmonic
dominates [55, 58]. The advent of the more expensive 12-pulse motor drives eliminated
the 5th and 7th order harmonics but introduced the 11th and 13th harmonic components
[59]. The 23rd and 25th harmonic components are generated within HVDC transmission
links [58].
The most common measure of harmonic content on the utility side is an index expressed
as a percentage, known as total harmonic distortion (THD) in equation (1-2).
2
V
THD (%) = h 100
h = 2 V1
N
(1-2)
Where h represents the harmonic order while Vh and V1 are the rms voltages of the hth
harmonic order and fundamental respectively.
Table 1-3 specifies the voltage distortion limits outlined in the IEEE 519 standard,
Recommended Practices and Requirements for Harmonic Control in Electrical Power
Systems [49]. At local industrial sites, the maximum THD index for voltage distortion
recorded can double these limits, as reported in [60, 61].
Table 1-3: Voltage distortion limits [49].
Bus voltage
Individual voltage distortion (%)
Total harmonic distortion (%)
69 kV
3.0
5.0
69 kV < V < 161 kV
1.5
2.5
> 161 kV
1.0
1.5
The current distortion limits define the maximum harmonic currents the end-user loads
may inject into the network. The current distortion is dependent on the bus voltage level
and the ratio of short-circuit current to maximum fundamental load current at the point of
common coupling, collectively known as the total demand distortion (TDD) [49]. In his
Master of Science (MSc) lecture notes on Quality of Supply, Professor J.V. Milanovi
provided a summary of real measurements, highlighting the magnitude ratio of the
29
harmonic currents produced by common sources relative to the fundamental, expressed
as a percentage shown in Table 1-4.
Table 1-4: Typical harmonic current relative to fundamental from common sources.
Contributing Source
Harmonic Order (%)

3rd
5th
7th
11th
13th
23rd 25th
Pulse Width Modulation Drive
3.9
82.8
77.5
46.3
41.2
1.5
2.5
DC Drive
1.2
33.6
1.6
8.7
1.2
2.8
1.2
Switch Mode Power Supply
65.7
37.7
12.7
5.3
2.5
0.8
0.4
Variable Speed Drive
3.9
39.7
18.9
6.8
3.8
1.8
1.7
Electronic Ballast
19.9
7.4
3.2
1.8
0.8
0.1
--
Voltage distortion can be further magnified by parallel resonances of harmonic currents in

networks employing power factor compensation capacitors [55, 62]. Harmonic filters are
employed to reduce the THD to an acceptable level at the point of common coupling, in
accordance with the IEEE 519 standard [49]. Estimates of filters cost are in the range
(USD) $ 75 - $ 250 /kVA [48].
1.8.2. Modelling electrical stress with harmonic content

Montanari et al. [63, 64] formulated three parameters to model distorted, non-sinusoidal
voltage waveforms as a consequence of harmonic content. These were:
waveshape parameter - Ks and Kf
peak parameter - Kp
rms parameter - Krms
The waveshape parameters are defined in equations (1-3) and (1-4).
Ks =
1
0
Kf =
30
h
2
h =1
h
h =1
2
h
2
h
(1-3)
(1-4)
Where h =
Vh
V1
and h is the harmonic component number. In equation (1-3) the ratio of
1
is unity, since 0 is the angular frequency of the 50 Hz waveform and 1 is the
0
angular frequency of the fundamental waveform, also 50 Hz. Hence Ks = Kf will be used to
represent this parameter. Equations (1-3) and (1-4) are proportional to the rms derivative
of the waveform and thus related to its steepness [64]. Hence, the value of this index is
proportional to the distortion of the waveform.
The peak parameter is defined in equation (1-5).
Kp =
Vp
V1p
(1-5)
Kp is ratio of the peak voltage of the resultant waveform Vp to the peak fundamental V1p.
The rms parameter is defined in equation (1-6).
K rms =
Vrms
V1rms
(1-6)
Krms is ratio of the rms voltage of the resultant waveform Vrms, to the rms fundamental
V1rms.
Unity values for all three parameters represent a non-distorted sinusoidal, power
frequency waveform. Figure 1-7 is an illustration of disturbances (outer circle) and their
links to the characteristic descriptions of electrical stress factors including the parameters
outlined above (inner circle) [65].
31
Figure 1-7: Links between power quality disturbances and electrical stress factors [65].
1.8.3. Impact of harmonics on electrical ageing

Life tests conducted on polyethylene terephthalate (PET) and XLPE employing the 3rd, 5th
and 7th harmonic orders separately, confirmed that the resultant composite waveform may
increase the voltage peak and thus the likelihood of partial discharge inception [66, 67].
Therefore the phase and magnitude attributes of the harmonic orders influenced the peak
parameter (Kp) causing significant scatter of failure times. Hence, the voltage peak was
cited as the principal factor for insulation life reduction, even with constant rms voltage
[54, 62, 64, 66-69] and in the absence of partial discharge activity [64]. Montanari et al.
[64] investigated the ageing effect of non-sinusoidal voltages due to harmonic content on
XLPE and polypropylene (PP) in the absence of partial discharge activity due to intrinsic
ageing. The mathematical analysis conducted in [64] suggested the prevailing
degradation mechanism under distorted voltage was similar under sinusoidal voltages in
the absence of partial discharges. Resulting life model equations confirmed the voltage
peak parameter Kp was most detrimental to insulation life for both dielectrics. Furthermore,
the voltage waveshape parameter Ks was marginally more influential than the rms
parameter Krms especially when partial discharge activity occurred [64]. However further
investigations later confirmed the outright dominance of Ks over Krms [59].
Research previously conducted focused on the influence of Kp on insulation life, ignoring
the impact of parameters Ks and Krms [54, 69]. Despite this, life estimations were
determined for cables (XLPE, ethylene propylene rubber (EPR)), motors, transformers
32
and PP capacitors under unrealistic severe harmonic conditions of a test power system
producing significant reduction of insulation life especially for cables [54]. Mazzanti et al.
[59] identified this shortcoming and included parameters Ks and Krms. Mazzanti et al. [59]
analyzed real network data from Romes subway which featured two 12-pulse converters
and a 20 kV XLPE feeder and employed a reduced electrothermal life model, defined in
equation (1-7). This model assumed negligible thermal ageing, since the thermal rating of
the insulation was significantly greater than the levels of thermal stress produced [59].
n
nr
LNS = LS K p p K s ns K rms
(1-7)
LNS and LS represent the mean insulation life under the distorted (non-sinusoidal) and
undistorted (sinusoidal) conditions respectively. The exponents np, nr, ns, are model
parameters derived from insulation specific accelerated laboratory life tests for Kp, Krms
and Ks respectively [54, 69]. LS is defined as a function composed from the inverse power
model of electrical ageing and the Arrhenius model for thermal ageing [54, 70, 71]. For
completeness LS is illustrated in equation (1-8).
n b
LS = L0 (E / E0 ) [ 0 ] e B
(1-8)
Where E is the rms electric field, E0 is a reference field, n0 is voltage endurance coefficient
defined in [68] (coefficient magnitude proportional to voltage endurance), L0 is the lifetime
at the nominal sinusoidal voltage and reference temperature, is conventional thermal
stress (indicative of temperature change), b is a parameter regulating stress synergism
and B is constant proportional to the activation energy of predominant thermal
degradation (magnitude of B proportional to thermal endurance).
The results obtained from this reduced electrothermal life model (under a distorted
regime) in equation (1-7), emphasized that parameters Ks and Kp impact significantly on
insulation life reduction, with Kp dominating especially during periods of significant loading
[59]. Ks = 1.6 was the worst case, while Krms never exceeded 1.08. Mazzanti et al. [59]
then argued that during peak periods where the voltage waveform is characterized by
sudden voltage rise, there is a chance of faster activation of field-assisted ageing
processes.
Table 1-5 and Table 1-6 show the overestimates of insulation life using the distorted
model in equation (1-7), incorporating all three parameters (KP, Ks, Krms) in comparison to
Kp only. These results were expressed as a percentage of residual insulation life, relative
33
to the estimates produced using the undistorted model in equation (1-8). Three case
studies during peak periods and over the entire day (peak and non-peak periods) were
analysed [59].
Table 1-5: Percentage of residual
Table 1-6: Percentage of residual
insulation life during peak periods [59].
insulation life during the entire day [59].
Insulation life (%)

Case
Insulation life (%)

Case
using models with

KP, Ks, Krms
KP only
26.6
81.1
24.4
26.6
using models with

KP, Ks, Krms
KP only
48.1
72.9
71.1
46.4
70.9
81.5
48.0
72.9
Mazzanti et al. [59] reported minimal increase of the rms voltage confirming the influence
of Krms to be very low and independent for correlation with ageing, as a consequence of
distorted waveforms. Therefore, these results in Table 1-5 and Table 1-6 suggest the
influence of the waveshape parameter, Ks on insulation life is too significant to be ignored.
Ks is an indication of the waveform steepness and can be described as a measure of per

unit time change in the electrical field of the dielectric (relative to a sinusoidal). Therefore
time-domain features including high rates of change of voltage and oscillating ripples
should have an impact on the movement of space charge, changing conditions for
electrical ageing processes.
The research in [62, 64, 66, 67] investigated increased harmonic orders with inversely
proportional harmonic magnitudes and large phase increments. To determine the critical
conditions for electrical ageing where the power quality has been compromised,
investigations (simulation and experimental) require smaller increments of phase and
magnitude of harmonic components interacting with the fundamental. Only then can the
critical harmonic levels for electrical ageing be identified.
The phase-resolved partial discharge patterns are greatly influenced by the presence of
harmonic content relative to discharge activity from a purely sinusoidal excitation voltage
[66, 72-74], as seen in Figure 1-8.
34
Figure 1-8: Influence of harmonic content on phased resolved partial discharge plots. A)
Pure test voltage B) 17 % - 3rd harmonic C) 11 % - 5th harmonic [73].
The following effects are observed in Figure 1-8:
Change in maximum discharge magnitude (Plot B < Plot A < Plot C).
Change in the phase location of discharge activity.
Change in the number density of the discharge activity.
Existence of regions with no discharge activity, known as dead zones (Plot B).
Asymmetrical discharge activity for positive and negative cycles (Plot B).
These changes in the partial discharge patterns are deceptive and can lead to erroneous
conclusions about the insulations integrity. Applying analogue and digital filters proved
futile to suppress harmonic frequencies [72, 73]. Hence the influence of harmonics must
be acknowledged.
An investigation [72] was conducted using point-plane geometry (gap = 10 mm, tip radius
= 100 m) epoxy resin samples aged under the influence of a pure 50 Hz waveform and a
composite waveform including 10 % of the 11th harmonic (THD = 10 %) for 720 hours. The
test voltage magnitude of the 50 Hz fundamental was twice the partial discharge inception
voltage. The condensed quantitative partial analysis is provided in Table 1-7. N is the
number of discharges per second (N = N+ + N-) and Qmax is the maximum magnitude (pC)
of partial discharge captured. Both parameters are referenced to the polarity of the
excitation voltage.
In Table 1-7 the increased percentage difference between fundamental and composite
waveform ageing of parameters, confirmed ageing with harmonic content increased partial
discharge activity, both in number and magnitude. Hence this manifests as a change in
the number density of the partial discharge activity patterns captured in Figure 1-9.
Additionally, Table 1-7 reveals a significant difference between the values for N before
35
undergoing fundamental and composite waveform ageing, 1949 s-1 and 860 s-1
respectively. Although an explanation has not been provided in [72], this difference is
insufficient to negate the validity of the conclusions.
Table 1-7: Quantitative analysis illustrating increased partial discharge activity as a
consequence of ageing with 50 Hz compared to 50 Hz + 10 % 11th harmonic [72].
Parameter
N
N
Ageing: 50 Hz
Before
After
1949 s-1
338 s-1
-1
185 s
-1
153 s
-1
1183 s
776 s
-1
Difference
Ageing: 50 Hz + 10% 11th H
Difference
Before
After
-83 %
860 s-1
819 s-1
-84 %
728 s
-1
416 s
-1
-43 %
-80 %
132 s
-1
403 s
-1
205 %
-5 %
74 pC
75 pC
1%
102 pC
150 pC
47 %
Qmax-
53 pC
74 pC
40 %
45 pC
150 pC
233 %
Therefore
harmonic
Qmax
disturbances
infiltrating
diagnostic
systems
can
lead
to
misinterpretation of partial discharge data by the highly complex artificial intelligent

techniques employed, especially if they have been designed on the assumption of a
sinusoidal excitation voltage. Such techniques typically utilize quantitative yet descriptive
partial discharge data parameters including factors of skewness, kurtosis and inception
phase [7, 29-33, 36, 75].
Figure 1-9: Partial discharge phase-resolved plots after 720 hours of 50 Hz ageing (left) and
50 Hz + 10 % 11th harmonic ageing (right) [72].
Florkowski et al. developed a model to identify the phase locations of partial discharge
activity [73]. This basic model is based on the elementary principle that a threshold field
must be exceeded for a partial discharge to occur. The instantaneous voltage magnitude
of the composite waveform (fundamental + harmonic) was evaluated to determine if the
36
inception voltage was exceeded. The model correlated to the phase location of discharge
activity with real data but did not provide any measure of the relative discharge magnitude
expected, as shown in Figure 1-10. The partial discharge activity was captured at 15 kV
from a generator stator bar.
Figure 1-10: Simulation model (left) and experimental partial discharge activity (right)
incorporating 11 % of the 11th harmonic, showing good correlation to phase location of
discharge activity [73].
Literature describing electrical tree growth under harmonic conditions is scarce. However,
investigations into the influence of frequency on electrical tree growth are not uncommon.
Electrical treeing tests conducted using point-plane geometry samples (gap = 1 mm, tip
radius = 5 m) of XLPE [67] did not readily facilitate the capture of the visible aspect of
tree growth or any measurement of tree length within the dielectric, unlike the tests
conducted in transparent unfilled epoxy resin samples of similar configuration [1, 42, 57].
Despite this visual limitation, electroluminescence images captured confirmed at constant
rms voltage, harmonic content increasing the peak voltage beyond the determined tree
inception voltage (9 kV) resulted in bush trees. A reduction of the peak voltage below the
tree inception voltage increased tree inception time, resulting in growth of branch trees
[67]. These tests employed the 3rd, 5th and 7th harmonics.
Auckland et al. reported that as frequency increased from the power frequency, the tree
growth changed from a branch tree into a bushy tree [44]. These tests were conducted
using point-plane geometry samples (gap = 3 mm, tip radius = 25 m) of epoxy and
polyester resins over the frequency range of 50 Hz 1 kHz. Similar observations were
made by Zheng et al. [43] upon investigating the influence of electrical frequency and
mechanical stress on the initiation and growth of electrical trees. 5 mm thick 66 kV XLPE
cable slices were tested over the frequency range 50 Hz 2 kHz. The results indicated
37
that in the absence of mechanical stress, at low frequencies the electrical tree displayed
thinner branches while as the frequency increased the branches grew thicker [43].
The impact of frequency and the parameters Kp, Ks, and Krms on electrical ageing are
concisely outlined in Table 1-8 .
Table 1-8: Summary of the impact of electrical stress factors on electrical ageing
mechanisms.
Factors
Frequency
Partial
Discharge
Water Tree
Electrical
Tree
As frequency
increased PD
magnitude
decreased [7678]
Minimal PD
magnitude is not
frequency
dependent [74,
76]
As frequency
increased
inception rate and
growth rate
increased [40]
At high frequency
only, growth rate
faster and tree
length longer in
comparison to
power frequency
superposed with
high frequency
[79]
As frequency
increased tree
growth changed
from branch to
bush tree [44]
An increase in
DC bias voltage
for composite
waveforms
resulted in
predominantly
bush type trees
observed [80]
When tensile
mechanical
stress applied
with increasing
frequency bush
type trees
observed [81]
When tensile
mechanical
stress applied
with decreasing
frequency
branch and
bush type trees
observed [81]
Voltages
around TIV
resulted in
branch trees
while higher
voltages
resulted in bush
trees [38, 67]
PD at higher
frequencies
initiated close to
minimal inception
voltage required
[76]
Peak, RMS
Waveshape
PD inception
starts earlier
once phase and
amplitude of
composite
waveform sum to
higher value than
fundamental
magnitude and
greater than TIV
[66, 67]
Influences the
PD pattern [7274]
Other Facts
At LV no significant
effect on mean BD
values [4]
An increase in DC bias
voltage for composite
waveforms resulted in a
decrease in failure time
[80]
High frequencies results

in maximum losses and
surface charge
accumulation due to
relaxation phenomena
[82]
effect on mean BD
values [4]
Increase of peak and
rms values increase
dielectric losses and
temperature [54]
Increase of peak
reduces insulation life
[59, 62, 64, 66, 67]
effect on mean BD
values [4]
PD = Partial Discharge, LV = Low Voltage, BD = Breakdown, TIV = Tree Inception Voltage
38
1.9.
Low voltage ageing
The low voltage (LV) supply is defined as the supply of electricity whose upper limit does
not exceed 1 kV rms [83]. Research into low voltage insulation often highlights the work
done on motor insulation and cables. However, the focus here is oriented around the
latter, reviewing the limited information available about failure modes of low voltage AC
insulation whilst highlighting any correlation to failure mechanisms of high voltage (HV)
insulation.
1.9.1. Influence of ageing factors

Electric cables for low voltage 450/750 V systems can be subjected to thermal stress from
high current caused by short-circuits and mechanical stresses due to the laying conditions
[84]. Cables packed in junction boxes are subjected to both thermal and mechanical
stresses. The thermal stress originates from the current flow in the conductors. Nominal
current produces a limited temperature increase, while short-circuits could yield a
substantial temperature rise, increasing the possibility of electrical breakdown, especially
with the synergistic combination of mechanical stress. Significant thermal ageing does not
indicate that failure is imminent. However, for cellulose paper and paper/oil it causes
embrittlement, a reduction in mechanical strength of paper, an increase in dielectric loss
and evolution of gases. Only extreme conditions such as excessively high temperatures or
transient overvoltages lead to limited functionality [26].
Laboratory tests conducted on non-metallic sheath cable insulation investigated the
effects of chemical, thermal and electrical characteristics on low voltage insulation < 1 kV
[85]. The results suggested chemical breakdown of polyvinylchloride (PVC) insulation
occurred at temperatures > 260 C, whilst being 90 % overloaded continuously at 120 V
for 3 hours. In-service cables usually have load currents drawn intermittently rather than
continuously and as such will experience lower loading cause less heat generation.
Hence, chemical and electrical breakdown do not occur at temperatures below auto
ignition temperatures of most common structural materials [85].
High intensity ultraviolet irradiation has also been cited as a contributing environmental
variable in low voltage ageing [86]. Research on 1.1 kV XLPE and PVC cable investigated
the effect of ultraviolet irradiation and mechanical stress indicated that XLPE cables
degrade significantly within one year under direct application of ultraviolet irradiation. The
presence of mechanical stress on the insulation accelerated the effect of ageing. PVC
39
insulation withstood a significant degree of ultraviolet irradiation resulting in minor
degradation [87]. Mechanical and electrical properties of polymeric materials are
weakened by oxidation accelerated by high temperature, or by ultraviolet irradiation even
at room temperature. Oxidation induces changes which cause the material to become
hardened or viscous, initially at the exposed surface. The entire process is irreversible.
The degree of oxidation is dependent on the type of polymer, density of cross-linkage and
other factors dependent on the material and conditions [5].
The environmental conditions during installation must be considered. All the risks of
damage to the cable during this installation process which involve transportation and
storage must be minimized as this can result in costly maintenance and repair activities in
the early life of the cable prior to inevitable failure. Moisture has been cited [88] as a key
accelerant in the degradation of the low voltage cables as it causes a reduction in
electrical strength and increased conductivity [89]. Hence this must be mitigated by
capping the ends of the cable, securing joints and having dry storage facilities. However,
paper impregnated cables can exist in a wet environment but must be kept under load
creating a thermal gradient which repels moisture ingress [89]. This suggested that ageing
was influenced by the oil impregnated in the paper rather than the moisture, causing
significant polarization affecting the papers electrical properties [89]. In urban areas,
installations frequently exist in confined environments with inadequate ventilation for the
cables, often resulting in the cables operating at temperatures beyond their rating. Hence
under overload and emergency conditions, the current is increased while heat dissipation
is poor, therefore thermal ageing is dominant.
The major sources of faults have been identified as damage during construction and
installation of equipment. This often results in lacerations to the cables causing a rupture
requiring joining. If the damage is minimal and the supply to customers is not disrupted,
there may still be damage to the oversheath of the cable from abrasions. Furthermore, the
cable may be subjected to mechanical forces including tensile and compressive forces,
causing cracks. The joining process along with these cracks and abrasions result in
vulnerable points for failure due to thermal runaway (hot spots) and moisture ingression
responsible for corrosion. This leads to chemical reactions causing oxidation,
depolymerisation and an increase in thermal conductivity accompanied by a reduction in
mechanical strength of the cable [89].
On low voltage distribution networks, faults occurring upstream of a transformer produce
temporary overvoltages, generally not exceeding 1.5 kV rms, while transient overvoltages
40
do not normally exceed 6 kV peak [83]. Figure 1-11 is an illustration of a typical
overvoltage, however the initial magnitude exceeds 6 kV [90].
Figure 1-11: Typical overvoltage propagated from MV to LV network via a transformer [90].
During faults there maybe significant overvoltages producing high electrical fields. These
non-power frequency events possess the potential to promote immediate breakdown or
initiate a gradual thermal ageing process. Unlike high voltage cables, low voltage cables
are not designed with an insulation structure aimed at making them free from partial
discharges under moderate overvoltages [91]. Short duration stresses are characteristic of
faults, transients and overvoltages which occur in the system. High fault currents create
an increase in thermal losses resulting in increased operating temperatures and
temperature gradients, which can exceed the insulation thermal properties. This may lead
to oxidation, melting, depolymerisation and eventually breakdown. Oxidation assists in
splitting polymer chains, causing brittleness and hardening of insulation from a depletion
of plasticizer [92], while thermal stresses (with or without oxygen) can contribute to
breakage of polymer chains [92]. This supports the deduction that most significant ageing
factors are chemical in nature [93].
The probability of lightning on low voltage networks is lower relative to the medium and
high voltage networks, as low voltage cables are usually limited in lengths and physically
shielded by houses and trees. Low voltage insulation is not designed to withstand
lightning induced overvoltages which can short-circuit phase to phase, and phase to
neutral conductors [94]. Lightning overvoltages cannot be prevented on overhead lines.
Underground cables will lower the rate of occurrence of overvoltages from lightning but
will not cause a reduction in the magnitude of the overvoltage. Direct lightning strikes to
the phase or neutral conductors generate insulation flashover regardless of the point of
impact. Lightning strikes to the structure and the line produce similar overvoltages.
Indirect lightning induces overvoltages higher than the insulation limits and this magnitude
41
is dependant on the location of neutral conductors and the suspended earth wires [94].
The overvoltage at the low voltage networks is characterised by an oscillating wave of
diminishing amplitude as shown Figure 1-13.
Figure 1-12: Induced overvoltage on the
Figure 1-13: Typical overvoltages at
line from indirect lightning i.e. strike to
varied locations from direct lightning
ground in the vicinity of the line [95].
strike to the line [90].
The indirect and direct lightning result in distinctly different types of overvoltages provided
in Figure 1-12 and Figure 1-13 usually characterised by parameters including amplitude,
steepness of impulse and energy [94]. Hence lightning overvoltages will increase the
electrothermal ageing of the insulation system.
1.9.2. Discussion
Exposed cables are often damage by the natural environment. Abrasions due to overhead
branches, exposure to ultraviolet radiation and windy conditions reduce the mechanical
properties of the insulation and increase the probability of faults and failures. Electrical
properties are also negatively affected but this is usually less significant than the
mechanical properties. Unsurprisingly, deterioration in mechanical strength has been cited
as the principle cause and moisture the principle accelerant in the low voltage ageing from
the literature reviewed for cables. Therefore the insulation thickness provides sufficient
resilience against the effect of electrical stress whose impact is low relative to the other
stress factors. Low voltage cable insulation is designed to withstand a recommended limit
of mechanical strength providing physical protection of the conductor. The thickness of the
insulation cannot be evaluated in practice; the partial discharge test is the only reliable
method to verify the insulations integrity [96]. Diagnostic partial discharge testing is
typically employed on medium and high voltage insulation systems but can be applied to
low voltage insulation systems with some modifications [96]. The electric fields created in
low voltage cable insulation are typically not sufficient to initiate partial discharges within
42
the dielectric. Consequently the formation of water and electrical trees are very
uncommon at this voltage level. Despite this, contaminants, voids, protrusions and defects
will still act as points of electrical field enhancement.
The ageing mechanisms for high and medium voltage (MV) cables maybe similar but the
stress level which initiates the ageing will be certainly different [26]. Stress factors
contributing to failure at both high voltage and low voltage include chemical effects
(oxidation, chain scission, cross-linking etc) and thermal effects. However the electrical
processes are not similar since mechanisms such as electrical treeing, tracking and
dielectropheresis do not normally occur at low voltage. Thus chemical, physical and
thermal models are transferable but electrical degradation models are not. The main low
voltage electrical stress factors have been identified as the waveshape and the rise time
from non-power frequency events. These factors are responsible for the increased energy
transfer to the dielectric facilitating higher than nominal electrothermal stress. Hence,
while electrical requirements are considerably reduced at low voltage, the other stress
factors still dominate and have an increased impact due to the operating conditions and
working environment.
In high and medium voltage networks the key focus is on asset management and quality
of supply. When considering the availability or customer minutes lost (CML) the low
voltage network cannot be neglected as outages here affect the overall network
performance, efficiency and operational costs [97]. Results published in [97], using
London as an example, illustrated that 90 % of faults occurred on the low voltage network,
and proper low voltage network management resulted in significant improvements to the
overall network performance. Furthermore, the data in [97] also confirmed that the
principal causes for failure included damage to low voltage cables by construction and
installation of other cables and pipes in the already limited underground spaces, moisture
ingression into cable sheaths and increased thermal strain on cables in winter and
summer peaks causing overload.
Thus insulation systems on the high and medium voltage networks are very different from
those employed on low voltage networks. The materials used in insulation systems at high
and medium voltages are better defined and so more easily modelled, in addition to being
significantly more expensive. Significant progress into the ageing of such insulation
systems has yielded improved diagnostic tools for network operators. The high cost of the
plant items for such levels of insulation justifies the vast amount of research already and
currently being conducted. Unfortunately, economic realities have meant that diagnostic
43
work on ageing of low voltage plant has not focussed on failure prediction. This is largely
because failures are generally the result of anomalous events such as incorrect
installation, damage, product abuse or poor manufacturing quality, rather than gradual
ageing. Further investigation of the ageing of low voltage assets is required to determine
the financial implications of faults on the low voltage network and to determine if advanced
condition monitoring techniques are feasible to improve network performance. More data
is required for pre-fault and pre-failure events to serve as diagnostic measures rather than
post-fault and post-failure analyses.
1.10.
Literature review findings
The following points summarize the findings from the literature review on insulation
ageing:
Experimental investigations of harmonics on electrical ageing of insulation have

not incorporated more than two harmonic orders in any composite waveform nor
maintained constant peak voltage.
Experimental investigations of harmonics on electrical treeing have not assessed

quantitatively or qualitatively electrical tree growth as a function of the waveshape
parameter Ks.
Significant harmonic content in the excitation voltage during diagnostic testing can
yield modulated phase-resolved partial discharge plots. If unacknowledged, this is
the precursor to erroneous conclusions about the insulations integrity.
Deterioration in mechanical strength has been cited as the principle cause and
moisture the principle accelerant in the ageing of low voltage insulation.
Initial moisture ingress is primarily due to third party damage compromising the
moisture barrier.
At high and low voltage, chemical, physical and thermal models are transferable
but electrical degradation models are not.
1.11.
Asset management overview
Asset management originated in the financial industry building on the principal trade off
between risk and return [98]. However these assets are the physical infrastructure
essential to facilitate business operations. Thus infrastructure asset management is
significantly more complex due to the non-financial aspects of performance and risk
44
surrounding maintenance and replacement for these assets which are members of highly
interconnected power systems characterized by a non-liquid market [98].
An excerpt from the Institute of Asset Management (IAM) related that, The management
of physical assets (their selection, maintenance, inspection and renewal) plays a key role
in determining the operational performance and profitability of industries that operate
assets as part of their core business. Asset management is the art and science of making
the right decisions and optimising these processes [99]. This definition does not outline
an explicit policy to be employed within any specific industry or business but recognizes
that every industry and each business is unique.
Figure 1-14 shows the typical processes involved in the realm of asset management
illustrating the feedback loops due to resources, cost and work scheduling outlining the
multidimensional nature of asset management.
Figure 1-14: Asset managers management processes - the big picture [100]. The resource,
cost and work control loops are feedback loops which influence the control loop to improve
management of the physical assets.
Asset management strives to effectively balance cost, performance and risk with long
term corporate objectives [98, 101] as shown in Figure 1-15, involving a broad spectrum
of any businesss operation from board level to ground level. Hence effective asset
management is the amalgamation of aligned technical and management decisions.
45
Figure 1-15: Asset management balance of costs, risks and performance [101].
1.12.
Asset management approaches
The risks for each industry and business are different hence different strategic
methodologies are often adopted. The following two techniques are not asset
management strategies but approaches to fulfilling asset management practices:
Condition monitoring
Reliability centered maintenance
1.12.1.
Condition monitoring
The term condition monitoring is often used interchangeably with condition based
monitoring, predictive maintenance and on-condition monitoring. Condition monitoring
describes the process of employing techniques which include vibration monitoring,
thermography, tribology (the effects of friction on moving machine parts through the
lubricating medium), ferrography (separates particulate contamination by using a
magnetic field) and ultra-sonic flaw detection for detecting eminent equipment failures in
various industries. Condition monitoring techniques were developed to predict the onset of
failure for components exhibiting random failure characteristics [102].
1.12.2.
Reliability centered maintenance
Reliability centered maintenance was developed in the USA for the airline industry in the
late 1950s and has since been applied to many industries in an attempt to improve
management of assets. This methodology aims to identify components whose failure and
degradation induce loss or degradation in functions performed by the most important
processes or cause appreciable maintenance expenditures [103]. Blischke et al.
46
highlighted that the reliability centered maintenance methodology involves a systematic
and logical consideration of the following [103]:
System, subsystem or component functions.
Failure modes of each function.
Importance associated with the function and its failure.
Prioritizing process that identifies preventative maintenance tasks that cost

effectively reduce failure occurrence.
Reliability centered maintenance uses a function-oriented approach to focus maintenance

efforts only on functionally critical components. Thus reliability centered maintenance
reduced the overall chance of failure by shifting the operating cycle of the equipment into
the low failure-rate region and reducing maintenance-related failures with the reduced
frequency of intrusive maintenance tasks [103]. Reliability centered maintenance
methodology lends itself fully to the application of the correct maintenance procedure
being conducted at the correct time and by the correct personnel [102].
1.12.3.
Complimentary roles of condition monitoring and
reliability centered maintenance

The following points illustrate the key difference between these two asset management
approaches.
Condition monitoring is one of the many function preservation strategies evaluated

during reliability centered maintenance analysis. Other function preservation
strategies include scheduled maintenance design changes, training improvements,
operational changes, on-time changes and run-to-failure [102].
Condition monitoring technologies are often treated as the final assessment tools
rather than a function preservation strategy while reliability centered maintenance
analysis process provides a well-documented and structured program for
evaluating efficient and effective use of condition monitoring technologies [102].
Reliability centered maintenance will not solve all process operation and
maintenance expenditure problems but will address problems arising from
inadequate, incorrect, ineffective or redundant maintenance tasks [103].
Condition monitoring technologies are excellent tools for identifying potential

failure. However these tools need to be applied in the right circumstances, and
47
only when necessary. The reliability centered maintenance process is designed to
identify those circumstances and to determine when such tools are required [102].
1.13.
State of the art asset management
A brief review is provided to compare and contrast progress made within industries for
managing their physical assets. Moreover, the technologies available and employed are
briefly reviewed as well. Comparison of the asset management strategies in rail networks,
aerospace and the power industry are discussed, since these industries are significantly
physically asset based.
1.13.1.
General industry
According to the periodical, Maintenance and Asset Management there are new ways of
rethinking, thus reinventing reliability centered maintenance strategies including a host of
well known and some not so well known tools such as lean thinking, system thinking,
theory of constraints, failure modes and effects analysis, root cause analysis and tool
productive maintenance [104]. These revolutionary views were born as improved
maintenance reliability strategies were being developed to improve the level of condition
monitoring and asset management. Recently published techniques included:
Operator driven reliability (ODR) - A culture driven team-base approach using the
expert human operators senses of touch, hearing, smell and sight with the
necessary tools as part of the maintenance strategy [105].
Maintenance finite capacity planning (MFCP) - A management tool enabling

efficient scheduling and dispatching of sufficient work to fully utilize available
human resource capacity. This can be classified as human resource management
but applied to industry equipment for increased productivity [106].
Motor condition monitoring (MCM) - A model based fault detection and diagnostic
techniques successfully used in space and aviation applications. This condition
monitoring system is immune to sporadic environment behaviour and background
activities. The model consists of differential equations which learn the behaviour
characteristics of the equipment (motor). This facilitates the ability to determine
normal or abnormal operation of the equipment under any operating conditions as
significant diagnostic data can be captured [107].
48
1.13.2.
Rail industry
Rail infrastructure maintenance contributes to a significant percentage of the total

infrastructure costs and any reduction in maintenance costs significantly improve the
efficiency of managing infrastructure costs [108]. Throughout the years, European
infrastructure managers have collected a significant amount of unstructured data about
the infrastructure. This often led to incompatibility amongst many programs and databases
realizing incomplete asset management and maintenance strategies [108].
In the railway sector there is a need to move from the find and fix regime to the predict
and prevent regime. In the UK this move has been embraced by Network Rail and
involved transition from a time based maintenance (TBM) regime to an asset performance
based regime [108]. Asset management systems are dependent on the quality of data
available from condition monitoring systems and for the rail industry must [109]:
Reduce penalties (from the regulator).
Detect irregularities that could endanger safety and reliability of railway traffic.
Allow condition forecasting of assets to have maintenance and renewal planning

whilst maximising maintenance time.
However there are often tradeoffs and hurdles to overcome for implementing the latest
condition based maintenance techniques in a cost effective manner.
Figure 1-16 illustrates the modern concept of integrated railway infrastructure condition
monitoring highlighting the three major contributors (maintenance policies, technologies
and infrastructure). Figure 1-16 also draws attention to data captured from diagnostic
systems which acquire track, overhead line, vehicle dynamics measurements as well as
data from video inspection and other monitoring technologies linked to maintenance
policies [110].
49
Figure 1-16: Pyramid of railway infrastructure condition monitoring highlighting the three
major contributors: maintenance policies, technologies and infrastructure [109].
Software developed by the company Mer Mec, entitled RAMSYS (Rail Asset Management
Systems) incorporated deterioration models providing simulated behaviours of network
components [109]. The behaviour of the railway network components can only be
accurately forecast under operating conditions (e.g. traffic variation) with complete
comprehension of the deterioration models and factors producing damage and or faults
(e.g. railway traffic with certain characteristics) [109]. In RAMSYS, the data from various
diagnostic systems are utilized simultaneously, together with the complete history to
capture and define the behaviour of every asset. This history is used to generate a
forecast of behaviour with the use of sophisticated deterioration models. The optimal
combination of activities proposed for execution were based on the forecast behaviour
and comparison to the required quality and incurred costs [109]. RAMSYS was
successfully implemented on the Dutch railway system, revealing that malfunctions
related to infrastructure were reduced by 70 % and yielding an equivalent reliability
increase of the entire network by 40 % [110]. Major benefits of this system included [109]:
Earlier and timely identification and correction of critical defects.
Mitigation of risks of critical defect occurrence.
Transition from corrective to condition based and predictive-preventative

maintenance.
50
1.13.3.
Aerospace industry
Jet engines used in organisations such as the navy, army, air force and NASA use health
monitoring and life support systems. Jet engine health monitoring involves real time
analysis of components such as blades, disks, bearings and seals. However the extreme
conditions in which these components are required to operate, e.g. high temperatures and
high turbulence, make the application and positioning of sensors very challenging [111]. In
this industry, reliable failure prediction has been identified as a combination of physicsbased deterioration models and state estimation tools. Structural health monitoring is
implemented to monitor vibratory stresses that threaten engine safety due to high cycle
fatigue and other failure behaviours including low cycle fatigue, multi-axial fatigue, thermal
management, fracture mechanics, crack growth and stress corrosion [111]. Real time
monitoring using prognostic models of the jet engines can be achieved by installing time
of arrival and vibration sensors on engines themselves to determine stress levels or
embed sensors within the components of the engine. These embedded sensors included
accelerometers, strain gauges and thermocouples [111].
The future goal of jet engine health monitoring includes the ability to monitor conditions
and states of the engine components located in inaccessible areas and operating under
extreme conditions utilizing compact, smart, wireless, self diagnostic, self calibrating, self
powered and secure sensors. The future of the engine health monitoring systems will
house more advanced fault diagnostic and prognostic models complimented by improved
sensors and software systems [111]. An example is a small, cheap and lightweight sensor
for aircraft wing inspection capable of processing diagnostic data on-board and
transmitting wirelessly to a ground station [112].
Aircraft maintenance systems such as the Goodrich Integrated Mechanical Diagnostic
Health and Usage System (IMD-HUMS) [113] is one example of a very sophisticated and
versatile system. After flights, this system captures vibration data of components and then
applies a myriad of algorithms to generate statistical component condition indicators which
are then mapped to health indicators. The data captured from the system must
incorporate significant filtering techniques to prevent interference of noise from nonsynchronous gears, shafts and bearings, changes in air speed and torque. Functions of
this system included [113]:
Real time logging of events during flight which include structural and operational
usage.
51
Interface for pilot, crew and other maintenance personnel to input problems which
assist in troubleshooting.
Monitoring to detect specific problems related to aircraft operation which can

trigger maintenance actions.
Tracking the status of gears, bearing and shafts allowing early detection of
evolving flaws prior to becoming faults or failure events.
1.13.4.
Power industry
In power networks of the industrialised world a significant proportion of equipment is

operating well beyond its design life. In urban areas, cables which are literally one
hundred years old still carry power, while transformers operate well beyond the 40 years
originally envisaged. These functional, ageing assets are a testament to the success of
the many insulation systems employed [114]. Asset management for the electrical power
industry has been described as the evaluation of the life cycle costs of equipment and
system versus quality of supply according to requirements and regulations resulting in four
key challenges for asset managers [115]:
Alignment of strategy and operations with stakeholders values and objectives.
Balancing of reliability, safety and financial considerations.
To benefit from performance-based financial incentives set by the regulatory body

in a competitive environment.
Living with the output-based regulatory penalty regime.
The generalized nature of these challenges suggests that they can be applied directly and
indirectly to any industry. A crucial aspect of asset management is determining the
condition of the equipment achieved through effective data collection [98]. The best form
of condition estimation and data collection occurs whilst the equipment is in-service but
the enormous efforts and financial implications deem the offline data collection more
attractive. To maximise the use of the acquired data, physical models dependent on the
operating stresses, ageing mechanisms and maintenance specific aspects must be linked
[115]. The weakness of models has been strongly linked to the construction details of
various equipment [115], albeit any variation in equipment construction is usually
designed to increase efficiency (space, thermal, fuel, etc). Variations of the equipment
design influence the severity of operational stresses acting on the equipment. The
synergistic effect of these operational stresses (mechanical, electrical, chemical, thermal,
physical and environmental) prove challenging to quantify, leading to difficulty in
52
developing accurate deterministic or probabilistic deterioration models. The challenge
perhaps lies in ensuring relevant, reliable and accurate data acquisition to assist in model
developments.
Despite this hurdle at the component layer, many models are being implemented in
commercially available simulation software packages e.g. Powersim, Anylogic, iThink, SD-
Library [115] Runenergy [116]. Runenergy, a web based application, employs many
combinations of neutral networks and fuzzy logic principles for diagnostic checks and fault
identification. Notably, this application was deemed applicable to a host of mechanical
equipment including valves, heat exchangers, compressors, filters, generators, boilers in
addition to electrical equipment [116]. These artificial intelligent techniques have been
successfully employed for electrical equipment diagnostics using fuzzy logic for partial
discharge identification characteristics [7, 29-35, 117, 118]. A feature common to all of the
aforementioned packages is the ability to assist asset managers in maintenance
strategies. However, this depends on the focus of the maintenance strategy as different
approaches lead to varying maintenance costs and asset availability [115] exhibited in
Figure 1-17.
Figure 1-17: Classification of maintenance strategies [115].
Historically, time based maintenance strategies have been employed at high and medium
voltage networks. Variation in time interval durations and acceptable availability of assets
can result in high maintenance costs. However, in the low voltage network corrective
53
maintenance was employed and the component operated until failure i.e. no preventative
maintenance. This may not have been the most economic maintenance option as the
reliability of the network suffered. Since condition based monitoring activities are
concerned with the estimation of equipment condition leading to high availability with
moderate maintenance costs, this is presently the preferred maintenance approach
according to German researchers, even at low voltages [115]. On the medium voltage
grids where reliability is the primary concern, condition based monitoring strategies are
being employed as well [115]. This allows reduction of customer minutes lost (CML)
indices. On the high and extra high voltage transmission networks, reliability centered
maintenance and condition based monitoring strategies are used to maintain the
networks integrity and alleviate bottlenecks [115]. Reliability centered maintenance
addresses the risk factor of the component failing, which is important for the transmission
networks to remain operational within their contingency plans. It is evident that the
importance of the asset to the fulfilment of the businesss core competency is critical to
the asset management strategy employed.
These strategies seek to minimise life cycle costs which are functions of maintenance and
replacement schemes as a consequence of an ageing asset base. However, there are
also costs incurred from occurrences of damage, disturbance events on the components
in the network which include cost for fault clearance, repairs, replacement of damaged
equipment and penalties or compensation. Hence, improved understanding of any
components condition leads to improved deterioration models resulting in an increase in
accuracy of reliability data which can reduce stochastic costs [115].
1.13.5.
Discussion
This review of the state of the art of asset management strategies employed across the
rail, aerospace and power industries provides practical insight. Conditioning monitoring of
mechanical equipment has always been at the forefront of industrial applications,
especially using vibration signatures. The limited literature reviewed indicates that the
deterioration models employed in both aerospace and rail industries are substantially
more developed than the models employed in the power industry, albeit these are
different types of equipment operating in different environments. This finding engages the
thought that perhaps mechanical degradation is better understood hence the integrated
approaches to condition monitoring for mechanical components incorporating the latest
technologies and software packages. However, in the power industry, mechanical
properties of electrical insulation are synergistically influenced by other stress factors
54
(electrical, thermal, chemical, environmental and physical) and so do not solely dictate the
failure mode.
Similar component forecasting in the rail industry based on traffic conditions and utilizing
data captured from different diagnostic systems should be employed in the power
industry. The traffic conditions are analogous to variation in load flows, for which data is
accessible and can be forecasted. The integrated use of load flow data and physical
measurable properties in diagnostic systems illustrates that crucial state estimations have
been identified and can be improved. Different and new measurable properties will require
diverse measuring systems impacting on costs and, in some extremes, data compatibility
issues. Effective data capture is essential to providing a data-driven and auditable asset
management system [98].
Captured and processed data can be incorporated into the deterioration models for
electrical insulation systems. However, these models will not yield accurate life forecasting
unless the physics of the deterioration process are fully understood. The aerospace
industry adopted models based on the physics of the failure processes. Such an approach
allows increased comprehension of the synergistic effects of the other stress factors. The
stress factors for electrical insulation often vary as a result of geographic location,
construction and design of the component. There are concerns that equipment
construction design affects the accuracy of deterioration models. However, improved
comprehension of insulation systems ageing and the working environment will negate the
uncertainties in modelling dynamics on the construction of the component.
Condition monitoring development in the power industry is lagging other industries.
Emphasis must be on development of sensors and improved prognostic sensing
techniques. Significant partial discharge technologies have been developed and are
currently improving. Fundamental research must continue to evaluate new and pertinent
physical measurable properties as well as links to the physics and chemistry of ageing
mechanisms influenced by the host of stress factors [114, 119, 120]. Monitoring the
operational environment (circumstance monitoring) where and when possible, will
undoubtedly improve the accuracy of deterioration models. Improved conditioning
monitoring will only enhance asset management strategies yielding maximum potential
benefit from both engineering and financial perspectives. There is the need for production
of a complete structural blueprint linking the component level to the system level which
allows the asset manager to quantify assets more reliably and save money on life cycle
costs.
55
1.14.
The future of asset management - PAS 55
Globally, many specifications exist addressing portions of asset management but no

defined standard exists for management of physical assets [101]. The Publicly Available
Specification, PAS 55 created by the Institute of Asset Management and endorsed by the
British Standards Institute (BSI) provides an avenue to efficiently deal with this
shortcoming. The definition of asset management according to PAS 55 is Systematic and
coordinated activities and practices through which an organization optimally manages its
physical assets and their associated performance, risks and expenditures over their
lifecycles for the purpose of achieving its organizational strategic plan. Jay [101]
simplified this definition to mean the set of disciplines, methods, procedures and tools to
optimize the whole life business impact of costs, performance and risk exposures of the
companys physical assets.
In order to draft a standard such as PAS 55, which optimized management of physical
infrastructure assets, industrial collaboration was essential. PAS 55 was specifically
designed to be compatible with other management systems such as ISO 9001 (standard
for quality management systems), ISO 14001 (standard for environmental management
systems) and OHSAS 18001 (standard for occupational health and safety management
systems) and does not require a stand alone management system [121]. The end result
was a 21-piece framework of requirements necessary to display asset management
competence [122]. This document was first issued in April 2004, a couple years after the
idea was first floated in February 2002 [101].
1.14.1.
Key principles of PAS 55
PAS 55 is independent of industry sector asset type and asset ownership structure. It
provides a framework for what to do and why. It also ties asset management into the
strategic business plan of the company [101]. It is a very general document and therefore
does not prescribe how to achieve good practice but rather the best appropriate practice
available. Therefore it does not mandate specific changes to the current strategy of asset
management employed. The key principles of PAS 55 can be summarized as [101]:
Aligned - ensuring practices are consistent with business plans and allows
learning, understanding and co-ordination of actions relating to asset plans.
Whole life managed - facilitating proper planning, costing and optimization of these
physical assets.
56
Inclusive - all levels and sectors of personnel are actively involved in contributing
to the best appropriate practice.
Data supported - meaningful data acquisition is important and the specification

does not mandate highly technological data acquisition and retrieval processes but
rather enforces the availability of the right data to the right people to improve
decision support.
Continuous improvement - a source of constant benchmarking for each company

and so providing the company with a goal to continually improve asset
management practices.
Pragmatic - providing holistic situations for ageing assets.
Optimized - ensuring that the optimization of inputs and outputs, short and long
term, capital expenditure, operational expenditure, other costs, risks and
performance are recognized.
Figure 1-18: Scope of PAS 55 [101].
1.14.2.
PAS 55 in the energy sector
The National Grid Transmission Company has undergone the PAS 55 certification
process. Below are some of the benefits relative to the UK electricity transmission
division:
The audit findings yielded more benefits by identifying asset management

improvement areas in asset management proactive environments [123].
57
Potentially, it can be part of regulatory process which affects the Office of Gas and
Electricity Markets (OFGEM) view on asset risk management and thus influence
the price review. The asset risk management was introduced by OFGEM to gain
assurance that systematic and coordinated practices were being employed [123].
OFGEM was satisfied with PAS 55 as it stressed continual improvement and alignment
with their asset risk management survey, ensuring that network operators have [124]:
A clear strategy
Understanding of the risk in their business
Understanding of systems and processes
Audits and reviews
The option of certification for PAS 55 requires least effort by OFGEM and produced the
most gain (since neither an audit nor an aligned asset risk management survey from
OFGEM is required because audits are carried out by an independent auditor).
Additionally, the responsibility of network operators being PAS 55 certified falls unto
themselves. Thus PAS 55 certification represents a hallmark of good asset management
practice recognized by the industry.
1.15.
Review of asset management
The condition of assets leads to performance while conditioning monitoring represents a

component level diagnosis only. A characteristic of any ageing asset is a reduction in
reliability. Hence the need for maintenance, providing remedial action to economically
sustain reliability whilst prolonging useful life and enabling planned asset replacement.
Thus the practice of asset management is a growing profession within power utilities
[119]. Asset management is an integral part of a business plan and involves human
resources looking at the bigger picture from the system level.
At the system level, blackouts are a major catastrophic failure in large interconnected
power systems and prediction of such an event is difficult. One of the many potential
sources of this problem is the increased failure in ageing equipment [125] at the
component level. Even restoration procedures after blackouts result in the presence of
sustained power frequency overvoltages, switching transients and harmonic resonance
[126]. These events momentarily accelerate the ageing rate of the insulation system, thus
increasing the functional age of in-service equipment. Li et al. [127] defined the functional
age as dependant on the deterioration status associated with usage history, operating and
58
environmental conditions. An increase in the functional age increases the likelihood of
failure and increases associated maintenance costs.
An understanding of insulation ageing facilitates the application of diagnostic techniques
for reliability assessment. Reliability indices are static and represent a snapshot of the
current condition of that component. Any changes to the components working
environment will impact on its probability of failure, its risk of failure and overall system
reliability. The drive to increase production of renewable and distributed energy has
resulted in variations of load flows on the network, while demand-side management
schemes cause deviations in load demands. A steady increase in the number of power
electronic devices results in fluctuations of power quality due to impulse transients and
harmonic content. As a result, there is a gradual change in the working environment
experienced by insulation systems [128-130] and as networks undergo this evolutionary
process, the insulation systems will age differently. Consequently, future failure modes
may deviate from the historical norm and traditional heuristics may no longer facilitate the
forecasting of behaviour of in-service insulation systems. There has been no direct
correlation with probability of failure and risk of failure. Questioning the suitability of any
asset management strategy to address such instances does not manifest a solution. The
challenge of an effective asset management system lies with the collection of data which
can help mitigate failure and increase economic productivity [131]. If historically such data
has not been captured there is little that can be done besides employing heuristic
techniques.
Figure 1-19: Considerations for asset managers in a dynamically changing environment.
To conclude, understanding what asset management is and its paramount role in the
power industry facilitated an integrated system level perspective. Effective modeling of
insulation ageing mechanisms at the component level cascading into the system level,
integrated into reliability models with financial considerations will undoubtedly be a huge
benefit to any asset management system.
59
1.16.
Aims and objectives
In this research project entitled Asset Management and the Role of Power Quality on
Electrical Treeing in Epoxy Resin, the aims and objectives were:
To review the impact of harmonics on electrical ageing.
To review ageing of low voltage insulation.
To develop a framework linking knowledge of insulation ageing to asset

management.
To develop an experimental facility to investigate electrical treeing.
To investigate the influence of harmonics on electrical treeing partial discharge

activity and electrical tree growth.
1.17.
Thesis structure
Chapter 2 describes the development of the novel multifactor framework linking

knowledge of insulation ageing to asset management. This vertically integrated five layer
framework facilitates modelling stress factors of the working environment. More
importantly, the framework establishes links between the physics and chemistry of
insulation degradation, linking plant reliability to asset management.
Chapter 3 reports the experimental approach employed in this research. An extensive
description of the in-house test experimental facility (hardware and software) is provided
discussing its merits and limitations. The sample preparation process and experimental
plans have been outlined providing a platform for interpretation of results.
Chapter 4 contains the analysis of experimental results. The influence of the lubricant
coating on hypodermic needles from which electrical trees were grown is first examined,
followed by a documented review of the electrical tree growth process observed. The
influence of the power quality on electrical tree growth as well as a statistical analysis of
the breakdown times and partial discharge activity are evaluated and discussed. This
chapter closes by modelling the electrical treeing partial discharge patterns with the
dV
.
function cosh V +
dt
Chapter 5 concisely reviews the contributions of this thesis, identifying the key outcomes.
Chapter 6 identifies the achievements, conclusions and avenues for future work.
60
Chapter 2. A multifactor framework linking insulation ageing to asset management
2. A multifactor framework linking insulation ageing to

asset management
2.1.1. Overview
There have been numerous ageing flow diagrams and conceptual models capturing the
effect of single stress factors on the path of failure for any insulation system, including
those in [5], and more specific combined effects of stress factors under wet and dry
conditions for polymeric insulation [8]. These flow diagrams have provided the foundation
for the development of a multifactor framework leading the transformation from a heuristic
approach towards a scientific approach of insulation life prediction and ultimately asset
management. These previous models have tended to be bottom-up approaches, led by
physicists and chemists. The model introduced here addresses the needs of asset
managers, coherently linking long term business models with the degradation models of
ageing assets.
The framework allows for stochastic modelling of the working environment, including
stress factors such as electrical, physical, mechanical, environmental, chemical and
thermal. More importantly, the framework establishes links between the physics and
chemistry of insulation degradation, linking plant reliability to asset management. Such a
tool can complement and improve asset management strategies and provide a platform
for future dielectric research. The framework developed is composed of five integrated
layers [53, 65, 114, 119, 131]:
(i)
Asset management
(ii)
Material state
(iii)
Stress factors
(iv)
Ageing mechanisms
(v)
Measurands
2.1.2. Asset management

Densley [26] previously emphasized that to make justifiable decisions concerning ageing
infrastructure, network operators must be knowledgeable of the operating conditions
influencing the ageing factors, ageing mechanisms, rate of ageing, failure mechanisms
and the criteria necessary to determine the corrective actions. Research has produced
many insulation- and process-specific models which have limited applicability to real
insulation systems. Uncertainties tend to lead to conservative decisions, leading to higher
61

asset maintenance or replacement costs than really necessary [53]. The asset
management layer is the first and uppermost of this framework.
K (t )
fi
( j )
G( g j )
G (K, t )
Gend ( K , t )
end (K , t )
(t )
K (t )
Figure 2-1: Interaction of stress factors influencing the mechanisms of failure in context to
the asset managers decisions forming the asset management layer of the framework.
Figure 2-1 depicts the network environment modelled by a time-dependant vector K(t).
This environment ages the material through a range of processes and results in a material
state described by another vector G(t). The asset manager can measure the environment
by circumstance monitoring through variables such as temperature and voltage.
Measurements of the plant itself can be acquired through condition monitoring, producing
a set of measurands, (t). The role of the physicist, material scientist and chemist is then
to link the measurements to the real state of the materials and through models of ageing
and reliability determine a limit on the condition of the plant against which the asset
manager can make judgements. As a result, the asset manager must make tactical
decisions concerning maintenance levels and loading profiles along with strategic
decisions including 40-year investment plans.
2.1.3. Material state

Through consideration of a materials state, asset managers are no longer confined to
considerations of time to failure or likelihood of failure of the insulation. Asset managers
may now consider the likelihood of the insulation reaching a specific measurable condition
deemed critical. Ultimately, the task is to seek the time or stress levels required to take the
62

system to the final failure path identified in Figure 2-2. The decisions explicitly identified in
Figure 2-1 should be informed by a number of knowledge- or data-streams. The most
important is knowledge of equipment functionality and failure.
Consider a single component such as a cable or a transformer operating in a network.
This component is subjected to multifactor ageing. With time, these stress factors may
degrade the insulation by a dominant process until, as a result of changes in the material,
a different mechanism prevails. If this latter mechanism leads rapidly to failure, the critical
point effectively determining the useful life of the insulation is that time at which the new
mechanism is initiated, and is termed here the end point. Figure 2-2 illustrates this
sequence used to produce the second layer of the framework, lying beneath the asset
management layer.
Figure 2-2: Insulation failure flowchart.
Identifying this end point is key for asset managers [7]. At the component level the
concern is whether the component functions correctly facilitating a forecast of future
reliability, while at the system level it is critical to understand the implication of this
components failure. Thus forecasting the end point enables effective asset replacement
63

programmes. Hence comprehension of the ageing process and corresponding changes in
the materials facilitate active management of existing assets.
Examples of numerically defined end points might include; oxidative state, void size or
electrical tree length. In terms of the impact of dielectric reliability, we consider a set of
acceptable limits of physical parameters which describe the state of each insulation
component. For an oil-impregnated-paper insulated cable the moisture content might be a
key parameter. Here we generate a vector G(t) to represent the physical parameters as a
function of time. Given that G(t) represents the state of the material at a microscopic level
we may consider an envelope of acceptable properties and a range of properties which
require some form of action (asset maintenance or replacement). Equation (2-1)
summarises an example of G(t) and its constituents.
g1 (t ) void size
g 2 (t ) oxidative state
g 3 (t ) mositure content
g (t ) morphology
G(t ) = Actual Material State = 4 =

dissolved gases
g 5 (t )
g 6 (t ) tree length
......... ..........................
g (t )
(2-1)
Most large scale insulation systems will normally require a probability density function gj(t)
to describe spatial or temporal variations throughout the insulation [132]. An example of a
system which already works on this basis is sampling oil from a transformer for dissolved
gas analysis. Limits of acceptable gas levels are set and responses pre-determined in
maintenance procedures [133].
2.1.4. Stress factors

The second core knowledge stream feeding into the asset managers decision of Figure
2-1 is the existing and future working environment. This environment controls the stresses
to which equipment (and in particular its dielectric) is subjected. The asset manager may
deem this uncontrollable. However, in some cases it is possible to load plant differently to
alter the system reliability. This does not necessarily mean reducing the load, instead
maintaining a more steady loading level, or reducing transient impulse stresses
(mechanical or electrical). The working environment is described by a another timedependent vector K(t), known as the stress vector. The vector consists of many factors
and similar to the material state vector, may be distributed spatially and change in time, as
64

illustrated in the Equation (2-2). These stress factors describe the working environment,
i.e. circumstance monitoring.
K1 (t ) Mechanical
K 2 (t ) Physical
K
(
t
)
Electrical
K (t ) = Stress Factors = 3 =
K 4 (t ) Thermal
K (t ) Environmental
5 Chemical
K 6 (t )
(2-2)
Each of these major stress factors can be further broken down into sub-factors which can
enhance, as well as compete with, each other. It should be emphasized that for this
framework, the relationship between macroscopic stresses created by the external
network (such as applied voltages and temperatures) must be used to determine the local
microscopic stresses which influence ageing. Frequently models assume steady-state
external stresses. However, key to the study of reliability of real systems is the impact of
short-term events such as switching surges or lightning strikes. While these may be
considered separately, they must be included as an essential part of any integrated
model. Specifics of the electrical stress factors as a consequence of power quality
disturbances have been illustrated in Figure 1-7.
The electrical stress factors outlined in Figure 1-7 provided the origin of the sub-factors
illustrated in Figure 2-3 where the crosses denote key characteristic properties of the
stress types provided. This defined the electrical working environment of the insulation
component. Similar analysis can be done to decouple the other stress factors.
Figure 2-3: Electrical stress factors.
Quite often there exists some uncertainty about the future working environment. As a
consequence of changing generation and loading patterns on many networks, this is a
particularly dynamic part of the framework. Thus knowledge of a particular networks
65

operating environment is critical. In the framework, the stress factors represent the third
layer of the model.
2.1.5. Ageing mechanisms

Failure of a given component can be regarded as a competition between a number of
ageing mechanisms represented as a probability distribution function fi(K(t)), also a
function of time and of the stress factors K(t). Changes in the likelihood of failure occur
because either the working environment (stress factors) changes or the material changes.
These changes may be described as degradation, ageing or failure mechanisms. They all
contribute to end of life. In reality a continuous change in a material will change the local
stress factors. Additionally, it is important to consider the physics and chemistry of the
processes at a local level. The modelling of physical processes allows definition of the
limits of the acceptable material state at a microscopic level (see Equation (2-1)). These
limits might then be associated with gradually degrading material such as oxidative state,
degree of polymerization or might be seen as the change required in a material necessary
for a change in ageing mechanisms. Two examples of such transitions include the growth
of electron mean free path eventually allowing partial discharges, and the growth of a
water-tree enabling transformation to an electrical tree.
Dominant mechanisms under electrical stress factors include space charge trapping,
partial discharges, water treeing and electrical treeing. These ageing mechanisms can
complement or compete with each other with one mechanism dictating the failure mode
whilst another dictates the time to failure. For example, the key rate-determining ageing
process in an XLPE cable might be water tree growth. However, water trees in
themselves do not cause failure, rather they create conditions for electrical trees to
initiate, which can then lead to failure. Thus these mechanisms may have a chronological
sequence to their dominance. This is not always the case, as other stress factors can
influence mechanism dominance in the working environment; one such example is when
mechanical compressive stress leads to hot spots from thermal gradients causing
insulation embrittlement and eventually compromising the mechanical strength of the
cable, resulting in failure.
66
K (t )
K (t )
G (t )
G (t )
Figure 2-4: Ageing mechanisms are dynamic and may change in time as the material and
local stresses change [131].
Figure 2-4 illustrates the cyclical nature of the ageing processes changing the local
stresses (e.g. via space charge accumulation) and material properties. The loop shown
will change radically if the main ageing process is altered. Ultimately, reduced duration of
a stable state indicates rapid mechanism changes leading to failure. The response of a
dielectric to short-term events as identified in Figure 1-7 must also be considered. In
particular, the role of impulse voltages and resultant transient mechanical and electric
stresses on initiation of ageing mechanisms, such as electrical tree growth is important.
For an asset manager, identifying the appropriate protection levels for a device against
transient and infrequent events is a key economic decision.
2.1.6. Measurands
In practical insulation systems, knowledge of the state of a material is limited. For inservice plant equipment, direct measurement of a material is difficult. There is clearly less
restriction in laboratory experiments, particularly on model samples, where destructive
testing and model stresses allow detailed investigation. The measurements available can
be used to derive an understanding of the materials condition. Physical data capture is
needed to provide some estimate of the degradation due to these mechanisms and thus
the ageing state of the insulation.
Measurands are measurements of physical quantifiable properties, acquired either inservice or under laboratory conditions, which assist in describing the aged state of the
67

dielectric, i.e. condition monitoring. Thus measurands are essential to the asset manager.
Consider the electrical stress factors such as magnitude and peak which influence
electrical ageing mechanisms. Partial discharges occur and are a distinguishable
characteristic diagnostic at higher voltage levels which can be used to characterize void
formation and other defect development [34, 134]. Thus partial discharge detection is one
of the preferred methods of condition monitoring at high voltage. Additionally, this
illustrates that mechanisms are also dependent on the operating voltage. Different
mechanisms yield different measurands, some of which can be field based, some not.
Techniques are well developed for the assessment of transformer oils sampled in service
[135]. Dissolved gas analysis [133, 136, 137] as well as furans analysis [138] determine
the state of the oil and the solid insulation of the transformer windings respectively. The
end result in both scenarios is a series of measurements of insulation properties which
lead to conclusions about the state of the insulation system and transformer lifetime
estimates.
Consider electrical tree detection. Whilst the cable insulation is in service, detection is only
possible through partial discharge detection and pattern recognition. Even this method is
not fully accurate but a good example is provided in [117, 118]. However, under laboratory
conditions, electrical tree growth can be detected utilizing electroluminescence detection
including photon counting and spectral frequency analysis [22, 23, 139, 140]. While such
tasks are bounded by physical limitations for in-service equipment, there is a need for new
measurands and diagnostic tools to overcome these restrictions [65].
The measurands which allow an estimation window of material condition represent the
final layer in the framework. The measurands used, n allow an estimation of the actual
physical state of the material. We call this estimate the state estimation matrix, (t) as
denoted in Equation (2-3). The content of the state estimation matrix depends upon the
condition monitoring undertaken at the time and the interpretation available. Examples of
potential measurands for cables include:
1 (t ) dielectric loss
2 (t ) void size
3 (t ) tree length
(t ) mositure content
(t ) = State Estimation Matrix = 4 =

thermal capacity
5 (t )
partial discharge magnitude

6 (t )
......... ..........................
(t )
n
68
(2-3)

The associated probability distributions denoting the physical conditions due to the
dominant ageing mechanism written as gj(K(t),) representing the jth insulation property.
These distributions depend upon the insulation ageing stress factors and the history of the
component. Each distribution gj reflects the physical condition of the material and leads to
an ageing state matrix G(t) in Equation (2-1). However, the asset manager can only
assemble a matrix from measurands which we term the ageing state estimation matrix (t)
in Equation (2-3) providing a window onto the evolution of the material from the input of
real measurands n. This is now normally termed condition monitoring. Ideally the
measurands, n, give direct information on the real physical condition G(t) and the
processes fi, but identifying these links remains one of the major challenges for
researchers in the area. Hence continued work on ageing models is critical to improved
asset management and ultimately network reliability. Figure 2-1 exhibited that the endpoint is better defined as the ultimate critical state Gend(t). This state is a function of the
real-time working environment and so should be written as a function of working stresses
Gend(K,t). This must be interpreted into critical values of measurands yielding an

equivalent end(K,t), a multidimensional critical set of measurands which also depend on
the working environment. By forecasting (K,t) and identifying the likelihood of reaching
any value of end(K,t), a likelihood of failure can be determined in time as a function of
stress K(t).
2.1.7. Multifactor framework

The multifactor framework provides a structured approach to linking the top level of asset
management, through knowledge of the material state, factors or stresses which age the
dielectric and the ageing mechanisms. This is depicted in Figure 2-5, highlighting the
movement of information to the asset manager.
Figure 2-5: Flow of information to the asset manager.
69

The multifactor framework described in Figure 2-6 consists of five vertically integrated
layers described as:
(i)
asset management decision making
(ii)
the actual material or equipment state, G(t)
(iii)
the working environment, K(t)
(iv)
ageing mechanisms, fi(t)
(v)
measurands giving an estimation material state, (t)
Figure 2-6 identifies some of the details of ageing factors, ageing mechanisms, and
measurands often considered by the scientist, engineer and ultimately asset manager.
The stochastic nature of the failure of insulation must be taken into consideration. Once a
property is at a critical limit (end-point) it is not always certain that failure will occur
immediately. Thus, probabilistic representations of the working environment and material
states are essential. The link between asset management and insulation ageing allows
focus on specific stress factors which control insulation ageing in a given environment.
Through knowledge of the physics and chemistry of ageing, a limit on the condition of the
plant may be set in terms of measurands of the condition monitoring, end(K,t), which can
trigger actions such as maintenance or asset replacement [53]. Again, this formed the top
asset management decision-making layer of Figure 2-6, providing an integral link to the
measurands layer via the current state estimation matrix (t). These values of end depend
upon acceptable risk to the asset manager, and the acceptable parameters depend upon
the working environment K(t). This approach has also been advocated by Montanari [141].
As an example, a higher state of oxidation of oil may be acceptable in a transformer in a
lower loading situation. In reality, experts might determine probabilistic limits within which
the state estimator must lie. This last stage enables formation of probability density
functions of plant life expectancy h(t,K) equivalent to the life data acquired in-service or
under laboratory conditions. This information will undoubtedly complement current asset
management strategies, decreasing and identifying uncertainties in the decision-making
process.
70
(t )
(t )
K (t )
Gend (t , K )
K (t )
fi (t , K )
K1 (t , M )
K 2 (t , P)
K 3 (t , E )
j j _ critical
G (t )
K 4 (t , T )
K 5 (t , T , C )
K 6 (t , T , C )
P1 (t )
T1 (t )
P2 (t )
T2 (t )
P3 (t )
T3 (t )
P4 (t )
T4 (t )
M1 (t )
C1 (t )
E1 (t )
M 2 (t )
C 2 (t )
E2 (t )
M 3 (t )
C3 (t )
E3 (t )
M 4 (t )
C 4 (t )
E4 (t )
M 5 (t )
C5 (t )
E5 (t )
M 6 (t )
h(t , K )
E6 (t )
M 7 (t )
E7 (t )
M8 (t )
E8 (t )
f1 (t , K1 , K3 )
f 2 (t , K 2 , K3 )
g j (t , K , j )
g j (t , K , j )
f 4 (t , K 5 , K 6 )
f 3 (t , K3 , K 4 , K5 , K 6 )
g j (t , K , j )
g j (t , K , j )
G (t ) = actual material state

(t ) = state estimation matrix
h(t , K ) = pdf of plant item failure
j = insulation property measurand

f i (t , K ) = pdf of competing ageing mechanisms
g j (t , K , j ) = pdf of competing ageing sub-mechanisms
K (t ) =
K (t ) = stress factors reflecting the working environment

i
Figure 2-6: Multifactor framework of insulation life.
71
2.1.8. Application
This framework is versatile but complex. In practice, it is important to identify nonapplicable factors and mechanisms so they can be omitted for a particular component
operating under given conditions. The first instance of application entails discussing
existing knowledge considering the synergies which occur between mechanisms and
environments, resulting in simplification of the framework and identifying areas of
ignorance.
Consequently, this framework provides the wherewithal for the production of life
forecasting models, often radically simplified, but with a list of uncertainties. For example,
the impact of increased thermal cycling or enhanced frequency of switching surges on real
equipment, through analysis of stresses, ageing mechanisms and their impact on material
condition. Given that this framework can potentially produce a probability density function
for a single plant component, the inclusion of multiple components can ultimately produce
an integrated model for system reliability, enabling holistic network asset management.
This may then allow an asset manager to act accordingly, to mitigate the changing stress
or manage the plant differently.
The multifactor framework has been used to demonstrate the differences between high
and low voltage ageing [53]. The major stress factors contributing to failure at both high
and low voltage include chemical effects and thermal effects (oxidation, chain scission,
cross-linking etc). In low voltage cables, these chemical and thermal stress factors provide
the initiating mechanisms of failure, resulting in property changes to the insulation causing
a reduction in the mechanical and electrical strengths [26, 89, 92], leading to eventual
insulation failure. Furthermore as discussed in section 1.9, a high proportion of low
voltage cable failures have been attributed to physical damage of the moisture barrier
sheath of the cable, through third party damage leading to moisture ingression and
consequent chemical reactions [97]. In contrast, in high voltage cables the early stages of
ageing are dependent on the local electric field conditions. Hence similarity exists in the
ageing mechanisms in the final stages of failure although they have very different initiating
processes [26]. However the electrical processes are not similar since mechanisms such
as electrical treeing, tracking and dielectropheresis do not occur at low voltage. Thus
chemical, physical and thermal models are transferable but electrical degradation models
are not [142]. Insulation systems and diagnostic tools at high voltages are well defined
and being continuously improved. The challenge, however, is at low voltages where
condition monitoring is rarely employed and data is mainly collected after the occurrence
72

of a destructive event. As mentioned in sections 1.9.2 and 1.13.4, this is primarily due to
the financial infeasibility of online low voltage diagnostics.
2.1.9. Discussion
A consequence of dynamically changing networks is the inability to accurately predict life
expectancy of equipment in service. The ever changing working environment dictates
variations in the stress factors which influence the ageing mechanisms, any of which
might degrade the insulation. However, critical limits may be determined by scientists
performing laboratory life estimation experiments where ageing leads to rapid failure.
These critical limits should be set by asset managers who have both system and
component perspectives with an understanding that in-service equipment are exposed to
numerous stress factors. An improved knowledge base of the multifactor ageing of
insulation systems is required to comprehend the competing failure factors and
mechanisms while accurately forecasting insulation life. Hence understanding and
identifying changes to dominant ageing processes during equipments life are critical to
improved asset management. The multifactor framework is proposed as an avenue to
achieve this.
The multifactor framework developed here provides a template for asset managers, plant
managers and those involved in the area of network performance and material reliability to
assess the condition of any insulation system. In order to realise the maximum potential of
this framework, mathematical modelling must be integrated using all existing models, tools
and techniques for assessing insulation systems at the level of ageing processes. This is
no simple task as the scope is extremely wide and deep, spanning all insulation systems
and all modes of failure under a myriad of environmental operating conditions. The
availability of data is undoubtedly vital to the successful completion of this vast puzzle.
Further and future development of the framework from the top level would involve a
tailored integration of the defined asset management documentation with the companys
strategic business plan yielding a completely optimized framework, illustrated in Figure
2-7.
73
(t )
(t )
K (t )
Gend (t , K )
Figure 2-7: Future development of the framework with defined asset management strategies
tailored to the companys business plan.
74
Chapter 3. Development of experimental method and equipment
3. Development of experimental method and equipment

To investigate the role of power quality on electrical treeing, an experimental plan and inhouse test facility were developed. This chapter describes the approach adopted
providing insight into the practical and physical constraints encountered, providing a
platform to understand and interpret the results in the successive chapter.
The experimental plan aims to investigate the following:
1. Influence of power quality on electrical tree growth and breakdown times.
2. Influence of power quality on partial discharge patterns due to electrical treeing.
The power quality is the changing variable in this research activity and can be denoted
using two indices already defined (section 1.8), governed by the following equations:
Total Harmonic Distortion (THD)

2
V
THD (%) = h 100
h = 2 V1
N
(3-1)
Waveshape Factor (Ks)
V
Ks = h h
h =1
V1
N
(3-2)
Where h is the harmonic order present in the composite waveform, Vh the rms voltage of
the hth harmonic and V1 the rms voltage of the fundamental. Analysis of composite
waveforms with varied harmonic content resulted in the selection of six waveforms in
addition to the reference fundamental for testing. These seven waveforms are described
in Table 3-1 and illustrated in the succeeding figures.
The criteria for waveform selection were based on the index values for Ks and THD in
addition to providing a range of waveform rate of change (observed in the derivative
plots). In accordance with literature real case study values of Ks 1.60 [59] and IEEE 519
standard for THD 5% [49] have been included.
75

Table 3-1: Properties of the seven test waveforms.
Composition + 50 Hz
Wave
Harmonic Order
% Magnitude per
Harmonic
Phase
Ks
THD %
40.0
1.56
40.0
0.00
1.00
0.00
5.00
1.03
5.00
5.00
1.06
5.00
11
17.8
1.60
17.8
12
5,7,11,13,23,25
3.20
1.60
7.85
13
5,7,11,13,23,25
2.00
1.27
5.00
Wave 1 represents an extreme waveform with exorbitantly high THD = 40 % due to the 3rd
harmonic. Wave 7 represents the fundamental 50 Hz signal but will be referred to as one
of the composite waveforms (wave 7). Waves 8 and 9 are defined by the THD = 5 %
recommended limit employing the 5th and 7th harmonic orders respectively. Wave 11
exhibits upper defined limits of Ks = 1.6 for the 7th harmonic only, while waves 12 and 13
illustrate limits of THD = 5 % and Ks = 1.6 respectively, both utilizing the 5th, 7th, 11th, 13th,
23rd and 25th harmonic orders. Waves 12 and 13 employ six harmonic orders;
experimental work incorporating more than two harmonic orders [73] has not been
reported previously.
Figure 3-1 to Figure 3-7 also show the time derivative of each waveform. An increase in
the frequency attribute of these composite waveforms (due to increased harmonic content
present) resulted in increasingly complex derivatives. This variation of derivatives
increased justification for further investigation of the impact of distorted waveforms on
threshold field assisted ageing processes [59], in this case electrical treeing and its partial
discharge activity. To experimentally reproduce these seven waveforms in the output
range of 20 kV output 30 kV, calculations revealed equipment capable of
reproducing slew rates 100 V/s were required.
76

Wave 1 THD=40.0% K =1.56
Wave 9 THD=5.00% K =1.06
Waveform
Derivative
3
2
1
0
-1
-2
-3
2
1
0
-1
-2
-3
-4
0
Waveform
Derivative
Magnitude/Per Unit
Magnitude/Per Unit
-4
50
100
150
200
250
300
350
50
100
Phase/deg
350
Waveform
Derivative
3
2
1
0
-1
-2
Waveform
Derivative
Magnitude/Per Unit
Magnitude/Per Unit
300
Wave 11 THD=17.8% K =1.60
-3
2
1
0
-1
-2
-3
-4
-4
50
100
150
200
250
300
350
50
100
Phase/deg
150
200
250
300
350
Phase/deg
Figure 3-2: Wave 7 THD=0 % Ks=1.00.
Figure 3-5: Wave 11 THD=17.8 % Ks=1.60.
Wave 8 THD=5.00% K =1.03
Wave 12 THD=7.85% K =1.60
Waveform
Derivative
3
2
1
0
-1
-2
-3
Waveform
Derivative
Magnitude/Per Unit
Magnitude/Per Unit
250
Figure 3-4: Wave 9 THD 5 % Ks=1.06.
Wave 7 THD=0.00% K =1.00
2
1
0
-1
-2
-3
-4
0
200
Phase/deg
150
-4
50
100
150
200
250
300
Phase/deg
350
50
100
150
200
250
300
350
Phase/deg
Figure 3-6: Wave 12 THD=7.85 % Ks=1.60.
77

Wave 13 THD=5.00% K =1.27
s
Waveform
Derivative
Magnitude/Per Unit
3
2
1
0
-1
-2
-3
-4
0
50
100
150
200
250
300
350
Phase/deg
3.1.
Test equipment
Developing any test facility requires an equipment specification for desired operational
performance within the realms of practical and financial limitations. Hence the test
equipment must satisfy the following objectives:
Voltage source capable of producing programmable non-power frequency

waveforms, providing increased isolation from external disturbances.
Amplifier capable of reproducing waveforms with 100 V/s slew rates with an
output in the range 20 kV output 30 kV.
Instrumentation to monitor, capture and record partial discharge activity.
Visual observation system capable of distinctly monitoring and capturing electrical

tree growth in epoxy resin samples.
Instrumentation to monitor, capture and record electroluminescence activity.
The hardware must be physically housed in a mobile test facility with software
integration providing functionality from a personal computer.
Adhere to all safety policies ensuring correct and safe operation of the equipment.
3.1.1. Waveform generation

An arbitrary waveform generator (AWG) was selected as the most appropriate source of
generating waveforms digitally with minimal external disturbances polluting the desired
waveform [73]. It was therefore critical that waveforms contain controlled desired
imperfections only. A peripheral component interconnect (PCI) based arbitrary waveform
generator was specified; the National Instruments NI PCI 5412 - 100 MSps (Million
78

Samples per second) 14 bit (16,384 levels) AWG (8 Mb). The vertical sensitivity and
horizontal resolution were sufficient for the intended application outlined in Table 3-2 and
Table 3-3.
Table 3-3: AWG horizontal resolution.
Table 3-2: AWG vertical sensitivity.

12
60,000
No of Samples
732
3.662
50 Hz cycles @ 0.1 s/sample
Output Voltage (Vpp)

Volts/level
4,194,176
20
3.1.2. Amplification
The commercially available TREK 30/20A high voltage power amplifier with its
parameters outlined in Table 3-4 was selected.
Table 3-4: Power amplifier parameters.
Output Current
0 to 20 mA DC or Peak AC
Output Voltage
0 to 30 kV DC or Peak AC
Internal Capacitance
50 pF
Slew Rate
> 500 V/s
The operational limits in Figure 3-8 of the amplifier driving a capacitive load with maximum
20 mA, were developed using equation (3-3).
(C
Amp
+ CDUT ) =
I
dV
(3-3)
dt
CAmp is the internal capacitance of the amplifier (50 pF), CDUT is the capacitance of the
device under test and dV
dt
is the rate of change of the excitation voltage.
79
Slew Rate vs Total Capacitance at 20mA
Slew Rate V/us
500
400
300
200
100
0
0
100
200
300
400
500
CDUT pF
Figure 3-8: Slew rate variation with capacitive load for the amplifier at its 20 mA limit.
The capacitance of the device under test (estimated < 5 pF) would realize acceptable slew
rates in the order of 100s V/s. Hence this amplifier satisfied the operational
requirements.
3.1.3. Image capture

The ability to accurately capture images of electrical tree growth has evaded many
researchers. Research in this area employed technologically advanced spectroscopy
equipment including cameras fitted with highly sensitive charge-coupled device (CCD)
image sensors and an image capture interface known as a frame grabber [38, 67, 143,
144]. For this research, a high speed camera capturing multiple frames per second (fps)
was not required. IEEE 1394 interface cameras were identified since the LabVIEW
platform can provide functionality equivalent to the frame grabber. The AVT MarlinF145C2 camera was selected. The colour camera boasted a maximum resolution of 1392 x
1040 pixels and a maximum capture rate of 13 fps. A telecentric C-mount lens was
employed to provide image capture at a distance 0.2 m from the object to the camera
and with a depth of field 5 mm.
3.1.4. Electroluminescence capture

Techniques for monitoring electroluminescence activity for correlating insulation ageing
have been investigated in [16, 22-24, 57, 139, 145]. The intensity of such emissions have
been monitored using photomultipliers and high speed counters. The main requirement is
a dark experimental environment to acquire such measurements. The spectrum of these
emissions have bandwidth of 375 nm to 725 nm for epoxy resin [22] encompassing both
the visible and ultraviolet ranges. The USB4000-UV-VIS from Ocean Optics accepts the
80

emission transmitted through single-strand optical fibre equipped with an ultraviolet/visible
collimating lens. The light is then dispersed via a fixed filter and grating tuned to 200 nm to
850 nm across the linear CCD array detector responsive from 200 nm to 1100 nm. This
was employed to capture intensity and frequency of the electroluminescence activity. The
variable exposure time range of 10 s to 60 s with 12 s between successive exposures
ensured relatively real time outputs to the LabVIEW platform using the USB interface.
3.1.5. Hardware assembly

Figure 3-9 provides the overview of the test equipment specified illustrating the interfacing
links within the computer and external environment. The challenges of developing this
facility include adhering to all safety policies ensuring correct and safe operation of the
equipment and achieving accurate and efficient software control.
Figure 3-9: Overview of test equipment and their interfaces.
The hardware assembly included not only the installation of the high voltage amplifier and
physical positioning of equipment but included the design and implementation of interlock
systems and triggering circuitry for the high voltage amplifier. Figure 3-10 and Figure 3-11
below provide images after the hardware was fully assembled. Figure 3-13 provides a
wiring schematic of the entire system.
81
Figure 3-10: Test facility fully assembled.
Figure 3-11: Amplifier: 30 kV 20 mA.
fire/heat sensor
4 M
earth rod
potential divider
bushing connected to output

of high voltage amplifier
partial discharge
balance circuit
variable intensity light source
dummy sample
sample under test
AVT MarlinF-145C2
camera
light source extensions
Figure 3-12: High voltage testing area (top) with close up view of the camera and sample
under test (below)
Figure 3-13: Schematic of high voltage test facility.
83

Table 3-5 highlights the problems encountered and the respective solutions employed
during the hardware assembly process.
Table 3-5: Summary of problems, solutions & tradeoffs.
Problem
Solution
Tradeoffs
Impedance mismatch from

high voltage remote signal to
unit causing loading.
Implemented a voltage buffer

circuit to increase
impedance.
Output from NI PCI 5412 is

6 V. Input to high voltage
unit is 10 V to get maximum
output of 30 kV.
Implemented a small signal

non-inverting amplification
circuit to increase output to
the high voltage unit.
Red high voltage supply

energised lamp once ON
switch is closed with no high
voltage output being enabled
on amplifier. Green lamp
should be illuminated.
5 V DC supply was powered

from ON switch rather than
set latching switch enabling
power to NAND gate
preventing floating inputs
which trigger the current
transistor to activate the red
lamp relay.
Not all opamps tried worked

correctly in circuit. The output
low TTL input was not
reproduced to trigger the unit.
Many opamps tried until one
provided suitable results.
Opamp selected must have a
bandwidth to facilitate
accurate reproduction of
signals. Circuit incorporated
clipping to limit the input at
the high voltage signal input.
5 V DC supply is available
once the ON switch was
closed. This implied that the
circuit was not completely
isolated as mains voltage
was input to the 5 V DC
power supply.
3.2.
Partial discharge instrumentation
This section is dedicated to the design and implementation of the partial discharge
measuring system and its output, in the form of a phase-resolved partial discharge
(PRPD) plot. Figure 3-14 provides an illustration of the measuring system.
Figure 3-14: Overview of partial discharge measuring system.
84
3.2.1. Design
Partial discharge detection involves converting a current pulse into a voltage signal using
a measuring impedance, as most instruments are conveniently designed to measure
voltage [146, 147]. In accordance with the IEC 60270 standard, High Voltage Test
Techniques Partial Discharge Measurements [25], a straight detection wide-band circuit

was originally implemented as shown in Figure 3-15.
Figure 3-15: Straight detection circuit.
Where U~ is the high voltage supply, Ca is capacitance of the test object, Ck the coupling
capacitor, Zm is the measuring impedance and MI the measurement instrument. The wideband attribute employs the frequency response (defined below) of the measurement
impedance to ensure the test voltage frequency and its associated harmonics do not
affect the measurement instrument [25]. Recommended values of this frequency response
provided in the IEC 60270 standard [25] for the lower and upper limit frequencies, flower
and fupper respectively, yielding the frequency band f are stated below.
30 kHz flower 400 kHz
fupper 500 kHz

100 kHz f 400 kHz
An RLC measurement impedance was utilised. The resonant frequency, fresonant of this
parallel RLC circuit is governed by equation (3.4), outlined below.
fresonant =
1
2 LC
(3.4)
Where L is the inductance (H) and C is the capacitance (F). The measured values of the L
and C components from the Robinson partial discharge detector measurement impedance
were 19.3 mH and 95 pF respectively, yielding fresonant = 117.5 kHz. The measured R value
85

was 1.6 k. This component affects the time constant of the circuit but not the discharge
pulse height. If R is small this results in a low time constant producing a sharp impulse
which may not be successfully reproduced by the amplifier circuit [148].
The circuit following the measurement impedance incorporated a non-inverting signal
amplification circuit of variable gain ( 100) with bandpass filter of frequency response of
100 kHz to 400 kHz, illustrated below. This constituted the measurement instrument.
Figure 3-16: Measurement instrument - amplifier filter circuit.
3.2.2. Simulation
During the design stage, simulation exercises were conducted using MicroCap software.
2nd and 3rd order Chebyshev filters were evaluated in these simulation exercises using the
measured values of the RLC measurement impedance. The capacitance of the test object
lay in the range 1 pF C 4 pF (with minimal connecting leads). Thus a value of 100 pF
was used to incorporate other sources of capacitance in the circuit including the internal
50 pF of the high voltage amplifier. The Chebyshev filter was selected instead of the
Butterworth filter because of the higher roll off at the cut off frequencies, increasing the
bandpass property of the filter. While the Butterworth filter has a constant gain in its output
response, the Chebyshev produces characteristic ripples at the cut off frequencies. In
order to minimize these ripples, 0.01 dB ripple design parameters were used [149]. The
schematic with real component values has been provided in Figure 3-17.
Figure 3-17: Simulation circuit in MicroCap.
86

The simulated results from the sinusoidal magnitude and transient response for the circuit
above are provided in Figure 3-18 and Figure 3-19 respectively.
AC Frequency Response
Transient Response
0.08
-10
0.06
-30
Magnitude/V
Magnitude/dB
-20
-40
-50
0.04
0.02
0
-60
-0.02
-70
-80 4
10
10
10
10
-0.04
0
Frequency/Hz
Figure 3-18: AC response.
10
Time/us
15
20
Figure 3-19: Transient response.
The AC response illustrated that bandwidth attributes could be improved. Improvements

include:
a) Constant gain in the frequency range desired
b) Steeper roll off at boundary frequencies
c) Less attenuation.
The simulated transient response to an injection of 50 pC illustrated that this response
settled comfortably after 12 s. These results were acceptable as the transient response
conformed to the recommended pulse resolution time (Tr) range of 5 s Tr 20 s [25]
and the shape of the bandwidth response was satisfactory within the limitations of the
transient response. However the significant level of attenuation would require
compensation for improved frequency response in a practical environment.
Issues a and b mentioned above can be improved using a 3rd order filter instead of the 2nd
order. However the resulting transient characteristics of the 3rd order filter were not
acceptable; the oscillatory response was insufficiently damped, thus unable to produce a
distinct pulse resolution time (Tr) in the range of 5 s Tr 20 s [25]. This favoured the
2nd order Chebyshev filter. A high gain amplification stage before the filter stage as
illustrated in Figure 3-16 would provide significant compensation for the attenuation issue
c. Moreover, utilizing an increased bipolar voltage supply (eg. 12 V instead of 5 V) will

increase amplification of the pulse height facilitating easier detection and discrimination
against noise. Hence 12 V was selected as the supply voltage for this circuit. The
downside of this choice was the limited availability of high speed operational amplifiers
87

with high gain-bandwidth product 1 GHz. This order of gain-bandwidth product is
essential to accurately reproduce amplified microsecond discharges within the limitations
of experimental work. This is expressed in equation (3.5).
1
gain = gain-bandwidth product
discharge duration
(3.5)
1
s
100 = 1 108
The LT1226 low noise and high speed operational amplifier satisfied the two major
requirements of supply voltage and gain bandwidth providing 12 V and 1 GHz
respectively.
The complete circuit provided a transient response in Figure 3-21 with an output gain =
100 pC/V to the injection of 50 pC, 125 kHz (in Figure 3-20) from the pulse calibrator. This
measured response correlated highly with the simulated response, settling comfortably
after 12 s confirming sufficient gain-bandwidth product from the LT1226 [148]. Thus
establishing this instrumentation system conformed to the IEC 60270 standard [25].
Figure 3-20: Input 50 pC calibrating pulse.
88
Figure 3-21: Output 100 pC/V discharge pulse.

Table 3-6 highlights the obstacles encountered during this initial instrumentation exercise
with the adopted remedies.
Table 3-6: Summary of problems, solutions & tradeoffs.
Problem
Solution
Excessive noise
propagating through the
passive filter with simple
back to back RC filters.
Implemented Chebyshev filter

with 0.01 dB ripple
characteristics to increase the
filter effectiveness.
Bandpass filter
attenuation for a 3rd order
Chebyshev greater than 6
dB. Output signal
attenuating significantly.
3rd order Chebyshev
provided a pulse transient
response with excessive
oscillation characterising a
lightly damped response.
Used 12 V DC instead of 5 V
DC to increase output voltage
and compensate for the
attenuation due to the filter.
Changed the filter to 2nd order
Chebshev. This improved the
transient response significantly.
Tradeoffs
Introduction of inductance
into the filter reducing
selection of parts options and
ripple effect into the
response.
Locating a high bandwidth
opamp with 12 V DC supply
rails was challenging and for
future use replacement parts
could be a problem.
The bandwidth will not be as
flat top as a 3rd order circuit
and roll off at cut off
frequencies not as steep.
The attenuation was
comparable.
3.2.3. Software development
Figure 3-22: Flowchart of the modules essential to the partial discharge measuring system.
89

The software defining the partial discharge measuring system represents a fraction of the
entire designed software in LabVIEW responsible for the overall functionality of the test
kit. A concise software flowchart is provided in Figure 3-22 with the five modules of the
partial discharge measuring system identified. The partial discharge recognition and
analysis process is explained in Figure 3-23. This was executed in MATLAB using the
input data from LabVIEW with the results exported back to the LabVIEW platform for
output display.
Figure 3-23: Flowchart of modules 2 and 3 of the partial discharge instrumentation.
The partial discharge extraction process entailed filtering the captured input data points
followed by a polarity pulse detection process. This was based on the observation that the
polarity of the pulse is determined by the first rising edge. In this data acquisition system
90

such a detection process was necessary as the output waveform of the amplifier and filter
circuit can produce maximum and minimum peaks approximately equal in magnitude as
depicted in Figure 3-21. Therefore this would provide increased confidence in the output
results.
To understand the process, consider a sampling rate of 107 samples per second. This
corresponds to a sample collected every 1
107
seconds = 0.1 s. Figure 3-24 shows the
points collected for a 50 s interval. At this sampling rate it is evident that a positive pulse
is present. However, it is noticeable that the maximum absolute value of the negative
pulse exceeds the maximum value of the positive pulse by 12.5 %. It is this feature which
has led to this sample being used as an illustration.
200
Input Data Points 10 MSps

160.8
Discharge Amplitude/pC
150
100
50
0
-50
-100
-150
-180.9
-200
2.293
2.294
2.295
2.296
2.297
Sample Points
2.298
5
x 10
Figure 3-24: All input data point captured.
The input data to the digitizer contains points from the combined influence of background
and system noise. This noise threshold was predetermined at a value of 30 pC.
91
200
Input Data Points Above Set Noise Threhold
150
Amplitude < Noise

Amplitude > Noise
Noise Threshold
160.8
100
20s
50
0
-50
-100
-150
-180.9
-200
2.293
2.294
2.295
2.296
2.297
Sample Points
2.298
5
x 10
Figure 3-25: Input data points above noise threshold within 20 s window
Figure 3-25 illustrates the filtering of data points for amplitudes noise threshold. Hence
the data points exceeding the noise threshold (blue points) were extracted. Proceeding in
chronological fashion the first (blue) data point was identified to establish the starting point
of a window of 20 s (default setting but can be varied). As determined before, each point
was collected every 0.1 s, hence 200 points represent a window of 20 s. In this window,
the maximum and minimum of the filtered (blue) points were determined with their
corresponding sample points.
The absolute values of the maximum and minimum points were compared and should the
point possessing the greater magnitude occur chronologically earlier, that point identified
the pulse polarity and the magnitude, otherwise the other point did.
92
200
150
Partial Discharge Detected at 10 MSps

Amplitude < Noise
Amplitude > Noise
Noise Threshold
Partial Discharge
160.8
100
C=cluster of points
50
0
-50
-100
-150
-180.9
-200
2.293
2.294
2.295
2.296
Sample Points
2.297
2.298
5
x 10
Figure 3-26: Partial discharge detected at sampling rate of 10 MSps.
Figure 3-26 shows this, as an absolute maximum point of 180.9 occurs at a later time than
the 160.8 point. Hence in accordance with the first rising edge describing the pulse
polarity, this pulse was defined as a positive pulse of amplitude 160.8 pC occurring at its
given sample point in time. It is worth mentioning that the cluster of points, C cannot be
counted as a partial discharge pulse because they are contained in the 20 s window
despite their magnitudes exceeding the noise threshold. Notably from the visual
representation, these points represented the highly damped oscillatory tail of the original
partial discharge pulse and not a separate pulse.
This process is repeated starting at the next data point above the noise threshold for all
the input data points acquired. The locations and the magnitude of the partial discharges
are stored and prepared for output. The phase-resolved partial discharge plot is an
effective and well accepted method of displaying partial discharge activity [146]. This
representation provides comparison with the reference waveform on the same plot.
Equation (3.6) provides the governing equation to determine the phase () of any data
point captured using its sample point.
sample point
360 + initial phase ,360
= Modulo
sampling rate
frequency
(3.6)
93

This equation was incorporated into the flowchart of Figure 3-27 and was implemented in
LabVIEW to produce the modified PD data points with attributes of phase location and
discharge magnitude.
.
Figure 3-27: Flowchart of process to produce PRPD plot.
These partial discharge points were superimposed onto a reference waveform plot and
continuously updated to produce an output display of Figure 3-27. The x-axis represents
the phase locations in degrees whilst the y-axis is formatted in pC for partial discharge
activity.
Figure 3-28 compares short duration partial discharge phase-resolved plots from an
electrical tree captured using both the commercial equipment and the in-house produced
test facility. The density of the partial discharge pattern and the relative discharge
magnitudes correlate well providing confidence in the measurements, albeit the images
acquired lack clarity.
94
Figure 3-28: Comparison of commercially available LDS 6 (left) and the in-house test facility
(right) showing good correlation of partial discharge activity from an electrical tree.
3.3.
Limitations
Acceptance of any developed system is incomplete without identifying the limitations of

the system. In accordance with section 4.5.3 of the IEC 60270 standard [25] sources of
error should be taken into account when evaluating measurements.
3.3.1. Sampling rate and sample size

The balance between sampling accuracy and sampling quantity within the limits of
processing power yielded a sampling rate of 107 samples per second and a sampling size
of 4 x 105 samples. This corresponds to a sample collected every 1
107
seconds = 0.1 s
for a period of 40 ms (two cycles at 50 Hz). Doubling the sample size to 16 x 105 samples
significantly increased the digitizers response time often resulting in system errors and
run time failures. Excessively lowering the sampling rate allowed an increased number of
50 Hz cycles of data to be collected but the maximum partial discharge pulse height
captured may be erroneously defined as the true discharge pulse height.
Figure 3-29 demonstrates that at a reduced sampling rate of 5 MSps the pulse height has
still been identified as illustrated in Figure 3-26, albeit the latter illustrated a different
partial discharge pulse. Therefore each acquisition now permits 80 ms (four 50 Hz cycles)
of continuous data to be captured and processed approximately every 2-3 s depending on
the status of the central processing unit of the personal computer controlling the digitizer.
95
200
Partial Discharge Detected at 5 MSps
150
Amplitude < Noise

Amplitude > Noise
Noise Threshold
Partial Discharge
157
100
50
0
-50
-100
-119
-150
-200
7.35
7.355
7.36
7.365
7.37
7.375
7.38
Sample Points
7.385
4
x 10
Figure 3-29: Sampling rate of 5 MSps.
3.3.2. Disturbances
Disturbances affect measurements and the IEC 60270 standard [25] recommends that
background noise be less than 50 % of the partial discharge magnitude under
investigation. Constant magnitude disturbances can originate from numerous sources
including the high voltage amplifier when energized at zero voltage, external high voltage
tests in the vicinity, internal noise from the measuring system, switching operations in
adjacent circuits, radio transmissions, etc. [25].
This constant magnitude disturbance was quantified in the range 25 pC noise 30 pC
for this current setup. Data captured suggested that any discharge pulse with magnitude
less than this noise value proved extremely difficult if not impossible to distinguish as a
partial discharge. Noskov et al. [143] cited 25 pC as the noise threshold employed. In the
intended application, electrical treeing in epoxy resin under laboratory conditions partial
discharge activity is expected to be 100s pC [143], classifying this disturbance noise
level as acceptable. However, investigations to reduce the source of this noise highlighted
that the internal circuitry of the measuring system in Figure 3-13 was a potential source of
disturbances. The power supply of the amplifier filter circuits were provided by DC
converter modules connected directly to ~230 V mains supply. These low voltage supplies
with solid state switching elements produce disturbances in addition to harmonic distortion
[25]. A different ~230 V mains power supply revealed that the prevailing disturbance was
96

independent of the power supply outlet confirming the source was indeed the high voltage
amplifier. This prevailing disturbance was captured in Figure 3-30.
Figure 3-30: Noise (magnified time scale on right plot) from energised high voltage amplifier
at output = 0 V, 100 pC/V.
3.3.3. Improvements for disturbance mitigation

The current level of disturbance at 25 pC noise 30 pC can be employed to provide
adequate and accurate results. However the following improvements were evaluated to
provide further reduction to disturbances:
Migrate to the balanced circuit detection instead of a straight detection circuit

method. While the complexity of the balanced circuit is greater than its counterpart
the straight detection circuit, the balance circuit provides increased immunity
suppressing external originating events from the device under test [148].
Introduce a filter at the coupling point of the test facility to the mains supply to
suppress external disturbances from penetrating into the measuring system [150].
97
3.4.
Disturbance reduction
3.4.1. Implementation of balanced circuit detection

The balanced circuit detection option was explored in accordance with the IEC 60270
standard [25]. The schematic of this system is provided in Figure 3-31.
Figure 3-31: Balanced detection circuit.
Where U~ is the high voltage supply, Ca is the test object capacitance, Ca1 is the dummy
test object or coupling capacitor of capacitance equivalent to Ca, Zm is the measuring
impedance and MI the measurement instrument. In order to achieve the optimum
performance of such a system the capacitance of Ca and Ca1 should be identical to provide
the measurement instrument (MI) with a mathematical difference of the input signals
thereby producing an output reflecting the discharge activity local to the loop created
between Ca and. Ca1. A parallel coupling capacitor is not required for test objects of small
capacitance since the ratio of these capacitances influence the suppression of external
disturbances [148]. The measuring impedances, Zm employed here are parallel RLC
circuits satisfying equation (3.4) with the resonant frequency fresonant = 200 kHz leading to
component values of C = 100 pF, R = 1.8 k and L = 6.63 mH. Many balanced circuits
can be developed using measuring impedances with variable capacitance and resistance
providing compensation for stray capacitance and variation of the time constant to
reproduce finely-tuned partial discharge pulses [148]. A schematic of the implemented
balanced circuit is provided in Figure 3-32 illustrating the output to the amplifier and filter
circuits.
98
Figure 3-32: Implemented balance circuit integrated with the amplifier filter stages.
After calibration the circuit noise was captured once again with the high voltage amplifier
energised with the output at zero volts. Figure 3-32 displays the results indicating a
significant reduction in the noise level, to 10 pC noise 11 pC.
Figure 3-33: Noise from high voltage amplifier energised at output = 0 V, 100 pC/V from
straight circuit detection (left) and balanced circuit detection (right).
99
3.4.2. Disturbance analysis

A Fast Fourier Transform (FFT) was employed to determine the frequency components
from noise captured in the right plot of Figure 3-33. This was performed using the
developed LabVIEW interface. The FFT magnitude spectrum in Figure 3-34 revealed the
presence of low frequencies 700 Hz f 10 kHz and high frequencies f 0.6 MHz whose
amplitudes have not been sufficiently attenuated by the 2nd order Chebyshev filter,
manifesting as disturbance sources.
Figure 3-34: Power spectrum of FFT of sampled noise data.
The results of this FFT are limited to the frequency range determined by the sampling rate
of the input data, 5 MSps. Therefore high frequency disturbances 2.5 MHz will not be
detected. An offset in the input sampled data was observed. Investigations for the source
of this offset exposed an uncontrollable offset of - 0.033 V in the NI PCI-5112 digitizer
card. This was consistent with the results from the FFT produced. The offset was nullified
by manually setting this interface card with an offset of + 0.033 V producing sampled
results in Figure 3-35.
Figure 3-35 identifies 2 noise bands, NB1 and NB2 captured over 80 ms at a sample rate
of 5 MSps with the high voltage amplifier energised. NB1 and NB2 can be visibly
categorized as noise preventing the identification of partial discharge pulses.
100

Input Pulses at 5 MSps 100 pC/V
Pulses
Upper Noise Threshold
Lower Noise Threshold
15
20
10
NB2
5
0
NB1
-5
-10
-15
-20
0
Input Pulses at 5 MSps 50 pC/V
Pulses
Upper Noise Threshold
Lower Noise Threshold
15
20
10
5
NB2
NB1
-5
-10
-15
2
Sample Points
4
x 10
-20
0
2
Sample Points
4
x 10
Figure 3-35: Offset of input data points nullified at 100 pC/V (left) and 50 pC/V (right).
The high density of data points collected in NB1 indicated this band of disturbance
contained extremely high frequencies in excess of the sampling rate. NB2 contained
frequencies within the bandwidth of the measuring impedance of 100 kHz f 400 kHz.
The magnitude of NB2 10.5 pC dictates the minimum detectable level of discharge
activity which can be evaluated by this measuring system. Hence this undefined
disturbance in the sampled data is characterized by a phase relationship independent of
the reference voltage, leading to a host of points at that magnitude at numerous phase
locations. This would compromise discharge patterns and thus be erroneously accepted
as discharge activity from the test object.
Hence the magnitude of NB2 10.5 pC governed the minimum detectable level of
discharge activity with amplification gain of 100 pC/V. The minimum detectable level of
discharge activity was further reduced to 5.5 pC by increasing the amplification gain in
Figure 3-16 to 50 pC/V. The trade off is the reduction in pC range by half, as the power
supply swing remains constant resulting in saturation of the phase-resolved plots.
However to prevent this, variable gain was implemented for the amplification, facilitating
manual gain control when required.
101
3.5.
System integration
Figure 3-36: Flowchart of hardware and software integration.
Figure 3-36 illustrates the integration of the hardware and software to produce the desired
functionality for the entire test facility implemented on the LabVIEW platform. During the
acceptance testing, some screen shots of the developed software suite were captured as
shown in Figure 3-37.
102
Waveform generation
Image capture
Discharge data
Electroluminescence
capture
Figure 3-37: Sample screen shots of integrated software.
3.6.
Sample preparation
Epoxy resins have been successfully employed as a dielectric material in high voltage
electrical insulation systems. Thus many researchers have conducted investigations on
various types of epoxy resins [19, 42, 72, 151, 152]. The epoxy resin system selected for
this research was employed for previous studies conducted here at The University of
Manchester [42, 153]. This insulation system consists of two parts; the epoxy resin
(Araldite LY 5052) and an amine hardener (HY 5052), both produced by Huntsman.
Point-plane geometry samples were produced with the epoxy resin encased in a hollow
acrylic square tube of dimensions outlined in Figure 3-38 and Figure 3-39 relative to a one
pound coin. The hypodermic needle [154] (tip radius = 3 m, length = 51 mm and
thickness = 1.1 mm) was set with a plane separation gap of 2 0.5 mm [42]. During
manufacture, each hypodermic needle is coated (25 mg/cm2) with a transparent lubricant
as defined in the BSI 7864 standard, Sterile Hypodermic Needles for Single Use [154].
This lubricant coating reduces frictional forces permitting easy movement of the needle
under human skin.
103
Figure 3-38: Schematic of epoxy resin
Figure 3-39: Produced epoxy resin
sample.
samples.
The following steps outline the actual sample preparation process:

1. The hypodermic needles were soaked in Silstrip for 72 hours to remove the
lubricant coating (The commercially available chemical Silstrip derives its name
from its silicone stripping attributes). The needles were removed from the Silstrip
bath, carefully cleaned removing the lubricant substrate (whilst preventing damage
to the tip), rinsed with a fine, high pressure stream of water, blotted and dried with
fibre-free tissue then left to stand allowing further natural drying.
2. The LY/HY 5052 hard cast-type epoxy resin was mixed using a weight ratio of
100:38 (resin:hardener). This mixture which has a pot life of 2 hours was
vacuumed (typically 45 - 60 mins) to remove embedded bubbles.
3. These cubes were taped with thread seal (polytetrafluoroethylene (PTFE) tape) to
prevent permanent bonding of the sample to the mould rig in Figure 3-40 which
has a capacity of six per batch. The hypodermic needles were set to maintain the
desired gap length of 2 0.5 mm.
4. The mixture was delicately poured into the acrylic cubes of Figure 3-38 and
allowed to cure for 24 hours at room temperature followed by a post cure at 100 C
for 4 hours before gradually cooling to room temperature [42].
5. To minimise the variation of material ageing [1] on electrical tree growth and
provide comparison to similar studies [42], only samples left to stand for 10 3
days were tested.
104
Finished sample
Cube based wrapped in with

PTFE tape
Figure 3-40: Sample production rig.
The samples were labelled in the following format:

Batch identifier
Lubricant coating (Y/N)
T131-07-N
Unique sample identifier (1-6)
Waveform sample tested

with, see Table 3-1
3.7.
Assessment of partial discharge sources
Verification that partial discharge activity captured originated from the sample under test
was paramount for interpreting produced phase-resolved partial discharge plots. A
maximum test voltage of 14.4 kV peak proved insufficient to generate external sources of
partial discharges e.g. corona from crocodile clips. Therefore at this voltage, discharge
activity captured from the sample must originate internally, due to treeing activity and/or
externally due to imperfect boundaries in the physical setup of the test sample e.g. air
pockets between the sample base and the brass ground electrode.
Interface with potential air

pockets between brass
grounded electrode and base
of epoxy resin sample
Figure 3-41: Physical setup of the sample in preparation for testing.
105

Partial discharge patterns of artificial channels, voids and electrical tree growth were
reviewed [155-160] and reproduced in Figure 3-42 and Figure 3-43.
Figure 3-42: Partial discharge patterns in ethylene-acrylic acid copolymer point-plane

geometry samples (gap = 10 mm, tip radius = 3 m), from artificial channel diameter = 40
m, length = 2 mm (plot A) and length = 1 mm (plot B) both at 4 kV, while length = 2 mm at
4.5 kV (plot C) [155, 160].
When charges can propagate across short channels or voids, the wing-turtle-like pattern
also known as the tree-void-like pattern of Figure 3-42 plot A develops into the turtle-like
pattern of (Figure 3-42) plot B. This is described by the flat top shaped partial discharge
pattern [155, 160]. Increased voltage also produced a similar flat top feature observed in
(Figure 3-42) plot C.
Figure 3-43: Partial discharge patterns in ethylene-acrylic acid copolymer samples at 12 kV

for point-plane geometry (gap = 12 mm, tip radius = 3 m), from electrical tree growth, after
1 min (plot A), 35 min (plot B), 2 h (plot C), 6 h (plot D), 6 h 45 min (plot E) and 6 h 55 min
(plot F) [157].
Figure 3-43 shows the progressively changing partial discharge patterns produced from a
growing electrical tree [157]. This confirmed increasing tree length changed the partial
discharge from the turtle-like pattern to the wing-like pattern as seen in Figure 3-43 plot
106

A and plot B as well as plot D and plot E respectively. This is consistent with the findings
of [155, 160] captured in Figure 3-42.
To investigate the origins of partial discharge activity from the test setup of Figure 3-41,
graphite based conductive paint, commercially available as Aquadag, was employed. Four
virgin samples were used in this investigation and
stressed at 14.4 kV peak for 15
minutes. The partial discharge data was captured continuously for this duration but logged
in 5 minute segments. Aquadag was applied to the base of two test samples. Figure 3-44
illustrates the captured partial discharge activity for the following samples:
A. No Aquadag on base with full lubricant coating on needle.
B. Aquadag on base with full lubricant coating on needle.
C. Aquadag on base without lubricant coating on needle.
D. No Aquadag on base no lubricant coating on needle.
No Aquadag + Full Lubricant Coating
-500
0
500
100
200
Phase/deg
Aquadag + No Lubricant Coating
100
200
Phase/deg
300
Internal Discharge Activity
-500
0
300
-500
0
Aquadag + Full Lubricant Coating
500
500
500
100
200
Phase/deg
300
No Aquadag + No Lubricant Coating
-500
0
100
200
Phase/deg
300
External Discharge Activity
Figure 3-44: Influence of aquadag and lubricant coating on discharge activity.
In plots A and B as well as marginally in plot D of Figure 3-44, discharge activity appeared
as an isolated cluster whose pattern is representative of partial discharges traversing a
channel or void, creating the characteristic flat top shape seen in Figure 3-42. This was
interpreted as possibly the consequence of air pockets between the sample base and the
brass ground electrode i.e. external discharge activity. However, without the lubricant
107

coating on the needles, there existed a reduction in the density of (Figure 3-44) plots C
and D relative to plots A and B, indicating decreased discharge activity. This suggested
that the lubricant coating facilitated cavity formation along the needle and epoxy resin
interface contributing to the discharge activity appearing as external discharges. Therefore
external discharge type activity must be interpreted as a combination of the lubricant
coating on the needle and air pockets between the sample base and the brass ground
electrode.
The internal discharge activity patterns are characterised as a tree-void-like patterns of
Figure 3-43 especially plots B and C, where the partial discharge magnitude typically
increases with increasing phase of the excitation voltage, until the end of the tree channel
is traversed. The tree-void-like pattern is similar to the wing-like pattern [159] and can
be acquired independent of the presence of a void [156]. All plots of Figure 3-44 illustrated
this pattern characteristic of growing electrical trees, with plot C having low discharge
amplitudes. Thus, the presence of the Aquadag drastically reduced external discharge
activity especially when the lubricant coating was removed.
The arrangement of plot D (without Aquadag) will be employed in the experimental work.
To minimise the influence of external discharge sources the following mitigating steps
were taken:
Adhere to the hypodermic needle preparation outlined in section 3.6.
Ensure the base of the completed sample is free of rough boundaries by carefully
filing the entire base smooth.
If the filing required will affect the insulation gap length, discard sample and assess
the sample production rig of Figure 3-40, recalibrating if necessary.
The physical setup of the test sample on the grounded brass electrode should be
carefully set to minimise if not eliminate any potentially uneven boundaries which
cause air pockets at this surface.
3.8.
Experimental plan
The spectrometer has not been incorporated into these investigations. The acrylic tubes
housing the prepared epoxy resin samples blocked the transmission of the wavelengths in
the ultraviolet region. This is an undesirable result and any limited electroluminescence
measurements recorded using this apparatus would not provide a holistic view of the
electrical treeing process. Hence this limitation prevented the inclusion of the
spectrometer in the experimental plan. Future plans to integrate this spectrometer involve
108

changing the type of acrylic cube or careful removal of the acrylic without disrupting or
damaging the cured sample. Additionally, limiting or ideally eliminating external light
sources must be considered to increase the accuracy of the data capture as the naked
eye can only detect partial discharge activity in the range 500 pC - 1000 pC [148] in the
visible spectrum, which is beyond the magnitudes expected for these experiments.
3.8.1. Influence of power quality on electrical tree growth and

breakdown times
The design of the experimental plan involved a total of 42 samples to be tested, six
samples tested under the influence of each (of the seven) composite waveforms. Figure
3-45 shows a general plan for each sample under test.
Stage 1
Stage 2
Wave 1
Initiation
Growth
Wave 7
50 Hz
Wave 8
Composite
waveform
Stage 3
Ramp
Wave 11
Wave 12
50 Hz
Wave 13
Wave 9
18.0 kV
peak
14.4 kV
peak
10.6 kV
peak to
24.8 kV
peak
<30 mins
150 mins
10 mins
30
180
190
Time
(mins)
Figure 3-45: General plan for each sample under test.
The first stage describes the electrical tree initiation period and will last no longer than 30
mins. If an electrical tree of length 10 m did not initiate within this period under the
influence of 50 Hz at 18.0 kV peak, that sample was discarded and a new sample setup
for testing. Upon electrical tree initiation to the desired length, the composite waveform
was activated without supply interruption using the software suite developed. Thus
entering the second stage where the composite waveform was maintained at 14.4 kV
peak for a maximum of 2.5 hours as the excitation waveform. During this stage, partial
discharge readings were recorded continuously in 15 minute batches and visual images of
tree growth were captured every 2 minutes. Should the sample suffer breakdown during
this stage, this time was recorded. If the sample did not fail after 2.5 hours, a ramp
process was invoked in the final stage increasing the voltage at rate of 1 kV rms/min from
109

10.6 - 24.8 kV peak. Should the sample breakdown during this ten minute interval the
breakdown voltage was recorded. The voltage range employed here is consistent with
previous investigations on epoxy resins using 14.1 kV peak test voltage in [42, 44, 153,
161] with [42] utilizing a range of 14.1 25.5 kV peak.
This exercise sought to comprehensively analyse electrical tree growth measurements
and breakdown times for all 42 samples tested in search of a relationship, should it exist,
as a function of the composite waveforms due to THD and Ks.
3.8.2. Influence of power quality on electrical treeing partial

discharge patterns
This second series of experiments concentrated on partial discharge activity arising from
epoxy resin samples containing electrical trees. Three broadly defined maturity stages of
electrical tree growth were employed as starting points (infant stage tree formed,
moderate spread midway of gap and extensive spread gap traversed), outlined below in
Figure 3-46.
Figure 3-46: Flowchart of the experimental process to investigate the influence of power
quality on partial discharge due to electrical treeing.
110

These samples were subjected to the seven defined composite waveforms at constant
peak voltage of 10.8 kV peak or 14.4 kV peak. The duration of each waveform did not
exceed five minutes. The most important aspect was ensuring that sufficient partial
discharge activity was captured in this period. The entire sequence of seven composite
waveforms was immediately repeated and the partial discharge activity captured for
analysis. The sequence was repeated to determine if the collected partial discharge data
varied from the first cycle to provide any indication of a change in the dominating ageing
mechanism. Hence the time for each test did not exceed (7 x 5) x 2 = 70 mins. A total of
11 tests were conducted.
The initialization of every sequential composite waveform occurred without de-energizing.
This ensured the effects due to residual trapped space charge influencing the acquired
partial discharge patterns were minimal. At the beginning of every stage (composite
waveform) images of the electrical tree growth were captured (where possible) to provide
a visual correlation to the partial discharge pattern from each composite waveform.
This exercise sought to evaluate and model partial discharge patterns due to electrical
treeing as a consequence of distorted waveforms.
111
Chapter 4. Experimental results and analysis
4. Experimental results and analysis

4.1.
Influence of needle lubricant coating
The literature citing the sample production technique adopted include [42, 153, 161],
utilizing hypodermic needles with a 2 0.5 mm gap spacing in a point-plane geometry
configuration. The post-cure temperature was room temperature in [153, 161] but 100 C
in [42]. As mentioned before, each hypodermic needle is coated (25 mg/cm2) with a
transparent lubricant as defined in the BSI 7864 standard, Sterile Hypodermic Needles for
Single Use [154]. However, only [153] detailed this lubricant removal process which
involved soaking the hypodermic needles in Silstrip, for 72 hours. Notably there were no
further guidelines for soaking the needle specifying the surface area of the needle to be
immersed (i.e. tip, or section of shank which would be bonded to the epoxy resin or the
entire needle). Therefore, all needles were wholly soaked in the Silstrip bath for these
experiments.
A visual observation from needles soaked for 12 days revealed a non-uniform stripping
process, despite total coverage of the tip, bore and shank surface areas. Figure 4-1
illustrated this observation where needles soaked for 12 days felt smoother and had a
glossy sheen in comparison to those needles soaked for 3 days. Hence soaking the
needles for longer duration did not enhance lubricant removal.
Soaked 12 days
Soaked 3 days
Figure 4-1: Hypodermic needles soaked for 12 days (upper) resulted in greater lubricant
retention conveyed by the glossy needle surface compared to needles soaked for 3 days
(lower).
As a consequence of this non-uniform conditioning process, visual observation of

conditioned hypodermic needles was conducted before sample production to ensure
experimental reproducibility.
112

The impact of the lubricant coating on the electrical treeing process was investigated
before the planned experiments outlined in section 3.8 were conducted. The findings from
this initial investigation are outlined into the following sections:
Partial discharge patterns
Electrical tree growth
Initiation and breakdown times
Gaseous activity
Samples were created and tested under the conditions outlined in Table 4-1. The sample
name provides information (section 3.6) such as the unique sample identifier (e.g. T131),
the waveform sampled was tested with (e.g. 07) and indicates whether the lubricant
coating was present (e.g. N).
Table 4-1: Sample details to investigate lubricant effects.
Sample
Wave Lubricant Waveshape Ks THD (%) Peak (kV)
T131-07-N
No
1.00
0.00
8.4
T132-02-N
No
2.54
56.6
8.4
T136-06-N
No
3.58
56.6
8.4
T161-06-Y
Yes
3.58
56.6
8.4
T163-10-Y
10
Yes
1.04
5.00
8.4
T165-06-Y
Yes
3.58
56.6
8.4
T166-01-Y
Yes
1.56
40.0
8.4
T212-07-Y
Yes
1.00
0.00
8.4
T213-07-N
No
1.00
0.00
8.4
T214-09-N
No
1.06
5.00
8.4
T215-09-Y
Yes
1.06
5.00
8.4
T355-07-N
No
1.00
0.00
14.4
T383-07-N
No
1.00
0.00
14.4
T444-07-Y
Yes
1.00
0.00
14.4
T446-07-Y
Yes
1.00
0.00
14.4
4.1.1. Partial discharge patterns

During the initiation stage and early growth stage (with lubricant coating on the needle),
there was increased partial discharge activity in the phase regions of 140 - 160 and
320 - 340 . However when the lubricant coating was removed, minimal partial discharge
activity (both in amplitude and number of occurrences) was captured. If discharge activity
113

occurred, the magnitudes were less than the minimum detectable level of 5.5 pC.
Samples tested at 8.4 kV peak (see Table 4-1) were stressed with the 50 Hz sinusoid until
tree inception 10 m [1], after which the composite waveforms were initiated. At the
increased voltage of 14.4 kV peak, all samples were stressed with the 50 Hz sinusoidal
only. However, electrical tree inception and growth occurred very quickly, thus tree length
10 m [1] proved difficult to measure. Therefore the partial discharge activity from
samples in Table 4-1 plotted in Figure 4-2 was produced from excitation of the 50 Hz
sinusoidal only, during electrical tree initiation and the early growth stage where the tree
lengths 10 m (8.4 kV peak) and 30 m (14.4 kV peak).
rabbit-ear phase pattern
Figure 4-2: Partial discharge activity from electrical trees of length 30 m at 50 Hz

sinusoidal reference with lubricant coating on needles (top) and without lubricant coating
on needles (below).
In Figure 4-2, there is a distinct contrast in the partial discharge pattern acquired when the
needle lubricant coating is present or not as outlined by the observed rabbit-ear phase
pattern [134, 158, 162, 163]. Figure 4-2 also highlights a variation of peak magnitudes of
the partial discharge activity. The maximum discharge activity occurred with sample T44607-Y, tested at 14.4 kV peak. During acquisition, there is always the possibility that
discharge activity will not be captured due to the cyclic time to process data until the next
capture (section 3.5). Furthermore, close monitoring of these experiments revealed the
114

occurrence of the rabbit-ear pattern exists only for a limited duration during initiation and
early growth of the electrical tree (length 30 m).
This rabbit-ear pattern has been reported as a function of discharges in a void [158]. The
size of the void has been documented to control both the ability to sustain discharge
activity and the angle of incline of the rabbit-ear pattern [134]. Investigations into the
source of the rabbit-ear pattern in polyethylene and polyimide concluded the synergistic
effect of water and oxygen was responsible for such a pattern [162, 163]. These studies
detailed that during the partial discharge process, oxygen atoms and water molecules are
released. These electronegative particles attach to electrons resulting in an increased
time lag for discharge activity, generating the rabbit-ear pattern with increased magnitude
discharges [162, 163]. These findings suggest that the lubricant coating between the
hypodermic needle and the epoxy resin suffers degradation at the onset of voltage
application leading to favourable conditions for production of the rabbit-ear pattern. This
is consistent with the observation that the rabbit-ear pattern occurs for a limited duration
during initiation and early growth of the electrical tree, i.e. until the lubricant coating has
been sufficiently degraded to inhibit those favourable conditions for such partial discharge
activity. Alternatively this can also be interpreted that perhaps the partial discharge activity
becomes physically separate from the lubricant, occurring at the tree tip and not the
needle
tip.
To
confirm
this
however,
requires
further
investigation
utilizing
electroluminescence measurements.
4.1.2. Electrical tree growth

During the growth phase of the electrical tree, its visual physical characteristics varied
significantly as illustrated in Figure 4-3. The presence of the lubricant coating on the
hypodermic needle resulted in an initial thick growth with little spread relative to the
dimensions of the test sample. Visually, it can be classified as a bushy tree. Upon
increasing voltage or after the tree has grown to an undefined mature stage, fine branches
develop realizing a change; now to a bush-branch tree. Conversely, without the lubricant
coating on the needle the electrical tree grew with a significant spread of extremely fine
tubules, creating a branch tree.
115
Figure 4-3: Illustration of typical tree growth A) with lubricant and B) without lubricant.
Figure 4-4: Tree growth images for T444-07-Y with lubricant coating (left) and T213-07-N
without lubricant coating (right).
To capture this change in tree growth, plots of the ratio of absolute tree width (W) to
absolute tree length (L) were captured from images of progressive tree growth similar to
those depicted in Figure 4-4 with Table 4-2 describing the physical interpretation of the
ratio.
Table 4-2: Interpretation of the width/length ratio.
Width/Length Ratio
Interpretation
<1
Tree is growing toward ground electrode with less spread
Tree has equal spread on both axes
>1
Tree has increased spread along the horizontal axis

No tree further growth.
constant
OR
Tree has spread with uniform rate on both axes.
The width/length ratio increases as the electrical tree growth spreads. A width/length ratio
of constant value can suggest that growth is stagnant or that the growth is occurring
consistently along both dimensions of measurements. Since the gap is fixed at
116

approximately 2.0 mm, when the tree has traversed this gap changes in the width/length
ratio can be attributed to changes along the horizontal axis if the tree continues to spread.
Figure 4-5 and Figure 4-6 show the measurements captured, identifying samples where
the lubricant coating was removed from the surface of the hypodermic needles with
dashed traces (---). Alternatively, the dashed-dotted (--) traces identify samples where the
lubricant coating was present. In Figure 4-5, separate dimensions (absolute and
normalized) of width and length measurements of electrical tree growth are provided in
two-dimensional plots. The measurement process incorporated manual magnification of
captured digital images producing a pixel to mm correlation based on fixed dimensions of
the needle thickness. This was a subjective process and can be easily misinterpreted
hence the dimension-less axes of Figure 4-5 were employed to provide relative not
absolute measurements. Three-dimensional plots are shown in Figure 4-6 with the
Measure of Width/units
Measure of Length/units
combined measurements of Figure 4-5 to produce the corresponding width/length plots.

Absolute Length vs Time
100
80
Normalized Length vs Time

Max
60
40
20
0
300
200
Time/min
400
Absolute Width vs Time
200
Time/min
400
Normalized Width vs Time

Max
200
T444-07-Y
T446-07-Y
T161-06-Y
T163-10-Y
T165-06-Y
T166-01-Y
T212-07-Y
T215-09-Y
T131-07-N
T132-02-N
T136-06-N
T213-07-N
T214-09-N
T355-07-N
T383-07-N
100
0
200
Time/min
400
200
Time/min
400
Figure 4-5: Reduced electrical tree length and width measurements with lubricant coating
compared to measurements without lubricant coating.
In the samples without the lubricant coating showed significantly increased growth along
both length and width absolute dimensions in comparison to the samples with the
lubricant coating. Additionally, samples without the lubricant generally took more time to
saturate plots of normalized length and width suggesting longer periods of growth in
117

comparison to samples with lubricant coating. Samples T444-07-Y and T446-07-Y
illustrated increased growth in contrast to other samples where the lubricant was present.
However these two samples (T444-07-Y and T446-07-Y) were stressed at 14.4 kV peak,
where the partial discharge magnitudes were highest as seen in Figure 4-2. The samples
without lubricant (T355-07-Y and T383-07-Y) stressed at this increased voltage level were
characterized by large magnitudes of length and width measurements than those samples
without lubricant, confirming that the lubricant coating inhibits electrical tree growth,
especially at lower voltages.
Plot of Width/Length Ratio vs Time
= breakdown
= touch ground electrode (L=2mm)
Width/Length
6
4
2
Samples
400
T444-07-Y
T446-07-Y
T161-06-Y
T163-10-Y
T165-06-Y
T166-01-Y
T212-07-Y
T215-09-Y
T131-07-N
T132-02-N
T136-06-N
T213-07-N
T214-09-N
T355-07-N
T383-07-N
300
200
100
Time/min
Figure 4-6: 3D plots showing reduced width/length ratios for samples with lubricant coating
relative to samples without lubricant coating.
In Figure 4-6 where applicable, the symbols and represent instances when the
electrical tree traversed the insulation gap, touching the ground electrode or suffering
breakdown respectively. Notably, the samples tested at 14.4 kV peak (T355-07-N, T38307-N, T444-07-Y and T446-07-Y) have shorter dimensions on the time axis. At this
increased voltage level, the electrical trees initiate and grow faster resulting in accelerated
propagation towards the ground electrode and ultimately a decreased time to breakdown,
relative to the samples tested at the lower voltage. The plots for samples without lubricant
demonstrate increased tree width contributing to the increased width/length indices as
time progresses. In T213-07-N there was a significant delay until tree inception occurs but
the limited growth due to time constraints suggests that the behaviour is otherwise similar
118

to the trees produced from other needles whose lubricant coating was removed. In T35507-N and T383-07-N, the plots indicate that failure () has occurred. At the instant of the
electrical tree bridging the insulation gap (), failure occurred. Thus these events occurred
within quick succession of each other and the overlay of both markers on the Figure 4-6 is
not distinguishable. Hence the failure () markers have taken precedence.
These growth plots have been collected from electrical trees which have been grown
using a variety of stressing waveforms. While all trees were initiated with 50 Hz, some
trees grew into mature trees under the influence of composite waveforms (see Table 4-1).
However a sub-population of T212-07-Y, T213-07-N, T355-07-N, T383-07-N, T444-07-N
and T446-07-Y were subjected to similar conditions to each other. Consequently the
growth characteristics appear to be controlled by the presence of the lubricant and not the
composite waveforms. The main conclusion from Figure 4-6 is the inhibition of electrical
tree growth spread and change in the tree structure when the lubricant coating (on the
needle) is present.
4.1.3. Initiation and breakdown times

Initiation times were only acquired for the samples tested at 8.4 kV peak. These samples
did not breakdown during the testing period (4.0 hrs) at this voltage level. For samples
tested at 14.4 kV peak, the initiation times were not acquired and the samples suffered
breakdown within the testing period (2.5 hours). Different testing periods were used as
compensation for the variation of the employed voltage levels in an attempt to facilitate
sufficient time for breakdown. Figure 4-7 presents the limited data recorded.
119

Initiation and Breakdown Times
16000
14000
12000
Time/s
10000
8000
6000
4000
2000
0
8.4 kV peak tree initiation without lubricant coating
8.4 kV peak tree initiation with lubricant coating
14.4 kV peak tree breakdown without lubricant coating
14.4 kV peak tree breakdown with lubricant coating
Figure 4-7: Scatter of initiation and breakdown times with and without lubricant coating.
The scatter of these points does not provide proof of any dependence of the lubricant
coating to these times. A significantly increased sample size is required to reveal any
correlations, should they exist. Preliminary investigations suggest that the lubricant
coating does affect the initiation and breakdown times. Visual observations from
monitoring and conducting these experiments have therefore provided the following
hypothesis to be challenged: Tree initiation is quicker in the presence of the lubricant
coating on hypodermic needles but the tree grows slower. Results are consistent with the
latter fact increasing failure times with the lubricant coating.
4.1.4. Gaseous activity

The movement of gas bubbles was observed from the needle tip up the bore and shank
on application of voltage ( 8.4 kV peak). This is illustrated in Figure 4-8. The visual
observation of this gaseous activity only occurs when the lubricant coating is present on
the hypodermic needle. This gaseous activity occurs simultaneously with the rabbit ears
partial discharge pattern. The visible gaseous activity did not appear as a constant
process throughout the entire tree growth process, only during electrical tree inception
and early growth (tree length m). This does not eliminate the prospect that such activity
may still be occurring but is visually undetectable at later stages of growth.
120
Gas bubbles moving up the needle
Figure 4-8: Illustration of gaseous activity.
Investigations into the role and the chemical composition of gas evolved during electrical
tree growth have been conducted in [164] on epoxy resin and polyethylene. These
investigations concluded that as the voltage and electrical tree growth increased, the
pressure of the gas in the tree channels increased and the amount of hydrogen liberated
was the same for both dielectrics but oxygen was liberated in epoxy resin only. The
experimental setup used here and described in Figure 3-9 was not designed for any
gaseous measurements, so similar conclusions could not be drawn. Figure 4-9 illustrates
the gas composition of an electrical tree in polyethylene as the electrical tree grew
highlighting that hydrogen existed in high concentration through out the growth of the
electrical tree. During electrical tree growth the decomposition of methane increased the
concentration of hydrogen and other hydrocarbon gases. This research [164] did detect a
similar trend with oxygen and oxide by-products formed in epoxy resin during electrical
tree growth.
121
Figure 4-9: Gas percentage vs cycles in the electrical tree channels of polyethylene. 10 s
pause between the full cycles 50 Hz, 30 kV (gap = r mm, electrode tip radius = 5 m) [164].
As mentioned in section 4.1.1, the production of the rabbit-ear partial discharge pattern
has been attributed to the presence of oxygen and moisture. Assuming the oxygen
decomposition in epoxy resin is similar to the ethane in Figure 4-9, this can potentially
account for the diminishing presence of the rabbit-ear pattern as time elapsed and
electrical tree growth continued.
A preliminary yet basic calculation to determine the volume of gas displaced as a result of
the growth of an electrical tree branch was performed using equation (4-1).
Volumemethane =
Density epoxy Volumeepoxy

Density methane
(4-1)
With the following assumptions:
Total mass of epoxy resin vaporized as a result of growth of electrical tree branch
is converted into 100 % of one hydrocarbon gas e.g. methane.
122
Density of epoxy resin is a function of the individual components only based on

volumes and is unaltered as it changes from liquid to solid states.
Table 4-3: Densities of compounds.
Chemical
Density (kg/m3)
Ratio by Volume
LY 5052
1170
100
HY 5052
940
47
LY/HY 5052 Epoxy Mixture
1096
--
Methane, CH4 (15C,1.01 bar)
0.667
--
Thus producing the following relationship at room temperature and pressure:
Volumemethane = 1640 Volumeepoxy
(4-2)
Assuming a simple cylindrical electrical tree branch of dimensions in Figure 4-10.
Figure 4-10: Simple example with dimensions of tree channel.
This resulted in Volumeepoxy = 7.9 x 10-15 m3 and Volumemethane = 1.3 x 10-11 m3, which
corresponds to 1.3 x 10-2 mm3. Assuming the volume described a spherical gas bubble
radius 10 m yields a cross-sectional area 1 mm2. This magnitude can be
representative of the gaseous activity observed. However, this suggests that perhaps the
gas bubbles may not be spherical but perhaps pancake-like along the needle surface, yet
hot enough to yield increased volume and or pressure in the tree channel, governed by
the ideal gas law in equation (4-3).
PV = nRT
(4-3)
Where P is the pressure (Pa), V is the volume (m3), n is the number of moles, R is the
universal gas constant and T is the temperature (K).
Hence this analysis questioned the source of this gaseous activity i.e. the sample
production, actual curing process, even possible voids near needle tip accelerating the
degradation (autoxidation) [13]. Oxidation has been cited as one of the dominant
processes in the tree initiation stages developing the void for the gaseous activity to occur
123

[165]. The rate of gaseous activity is therefore related to the size of the void and ultimately
governed by the amount of gas present. Since the layer of lubricant on the hypodermic
needle is as an intermediate layer between the epoxy resin and the material of the needle.
The observations made here suggest this layer prevents the epoxy resin from effectively
bonding to the needle, thereby creating voids and subsequently a weakest link path for
this trapped gas and for displaced gas due to the electrical tree growth to escape. This
facilitates reduced pressure at the needle tip and subsequent tree channels formed. This
reduces mechanical stress, decreasing cracking in the structure and this may be the
cause of different growth characteristics with and without the lubricant coating.
Additionally, reduced pressure would yield increased partial discharge activity consistent
with Figure 4-2. Therefore, the interface with the electron source (needle) is always
important and may be critical in all experiments. This maybe thought of as analogous to a
vented water tree. Hence, the fundamental conclusion revealed that further investigations
are required into the influence of the lubricant coating of hypodermic needles on the
gaseous activity.
4.1.5. Summary
Table 4-4 highlights the results unveiled from this investigation on the lubricant coating of
the hypodermic needles. These results have been unreported in the relevant literature [42,
153, 161] and make the comparison of similar tests extremely difficult, again questioning
the sample production technique. Moreover, these findings emphasize the meticulousness
required when preparing samples to ensure the integrity of results is not compromised.
Table 4-4: Conclusions on the influence of the needle lubricant coating.
Characteristics
Partial discharge pattern during
initiation and early stages of
growth
Initiation times
Breakdown times
With Lubricant Coating

The partial discharge pattern
acquired similar to the rabbitear pattern.
Inhibition of electrical tree
growth resulting in limited
spread.
Reduced
Possibly longer
Without Lubricant
The partial discharge activity
captured is minimal, both in
amplitude and pulse count.
increased with significant
spread.
Longer
Possibly shorter
Therefore, to ascertain a high degree of experimental reproducibility and results, the

samples produced for further experimental investigations employed the following
mitigation measures:
124
Hypodermic needles were soaked for a minimum of 5 days in Silstrip.
Visual inspection after 5 days to discard needles with glossy appearance indicating
significant retention of the lubricant coating.
4.2.
Sequence of electrical tree growth
This section primarily aims to record observations during the electrical tree growth which
can serve as a platform for future work in this area. Figure 4-11 provides a condensed
example of the major sequences of events from sample T345-09-N.
The phase-resolved partial discharge plots exhibit the discharge activity captured over a
five minute interval, with the electrical tree images and drawings representing snapshots
during those respective intervals. The discharge activity, captured from the sample can
originate internally, due to treeing activity and/or externally due to imperfect boundaries in
the physical setup of the test sample e.g. air pockets between the sample base and the
brass ground electrode.
Traditionally, electrical tree growth has been depicted by the three stage process outlined
in Figure 1-4. The observations recorded below will describe electrical growth in terms of
four distinct phases: initiation, non-conductive tree growth, conductive return tree growth,
thermal runaway.
Initially, the visual two dimensional images and drawings of the electrical trees attempt to
clearly illustrate extremely fine branches ( 1 m) growing in a three dimensional space
from the point of increased electrical stress (the needle tip). Very faint filaments of light
were observed at the needle tip penetrating and eroding into the epoxy resin. This
filament activity may have even been electroluminescence occurring in the visible
spectrum, albeit electroluminescence occurs faster than partial discharge pulses [22].
However the time delay for the real-time camera image acquisition meant if this filament
activity was of sufficiently high repetition rate, it could appear as a sustained event,
relatively lengthy in duration ( < 2 s). This intermittent light emission occurred prior to
electrical tree initiation, appearing as a newly formed tubule then disappearing, until
definite tree growth occurred where the tubule formed was substantially thicker than the
apparent filament activity. Thus, an electrical tree was initiated and began to grow towards
the ground electrode.
125
T + 20 mins
126
300
Figure 4-11: Sequence of electrical tree growth for sample T345-09-N.

200
Phase/deg
300
The plots above display the partial discharge activity captured over a five
minute interval, with drawings and electrical tree images representing
snapshots during these periods.
The branch tree continues growing

downward and spreading. It is about to
traverse the insulation gap.
T + 70 mins
100
200
Phase/deg
300
T + 115 mins
100
200
Phase/deg
The branch tree bridges the insulation gap.

Tree growth reverses in an upward
direction. Thick growth now observed
upward through existing tubules from the
grounded electrode.
300
Significant increase in the

magnitude of discharge activity.
Discharges > 300 pC
-400
0
-300
-200
-100
100
200
300
Phase Resolved Partial Discharge Plot 21 T345
400
T + 140 mins
Partial breakdown of tubules growing

upward through the thick branches
temporarily extinguish discharge activity.
Existing finer tubules growing downward
from the needle tip also collapse.
200
Phase/deg
300
Sporadic scatter of discharge

activity during and immediately
post breakdown.
Discharges > 400 pC.
100
The thick upward growth continues from

other locations of the grounded electrode
increasing visibility. Tubules near the
needle tip collapse meeting the upward
growing branches. Breakdown.
Breakdown arc
Sustained discharge magnitude and

increased repetition rate until breakdown.
Prior to breakdown increased frequency of
fluctuating discharge activity observed.
-400
0
-300
-200
-100
100
200
300
400
Thick branch developing (upward

growth)
Further increase of discharge activity

mainly on positive slope. Repetition rate
fluctuates and extinguishes momentarily.
Positive rate negative rate.
Tree touches grounded plane and continues spreading
Overall increase of charge magnitude

and repetition rate particularly on
negative slope.
Unsymmetrical trend continues but

with increased activity.
-400
0
-300
-200
-100
100
200
300
400
Epoxy Resin
Increased repetition rate particularly on

negative slope.
Discharge magnitudes unsymmetrical but

spread across the zero crossing.
100
Very fine tubules spreading
The tree has matured propagating

towards the grounded plane as a branch
tree.
Needle
Very low repetition rate.
Discharges occur almost uniformly on

either slope after the zero crossing.
Discharges < 100 pC.
-400
0
200
Phase/deg
-200
-100
100
200
300
-400
0
100
T + 50 mins
400
-300
-300
-200
-100
100
200
300
400

Notably, the return growth of the electrical tree from the ground electrode occurs after the
original tree traversed the insulation gap. A similar observation was documented by
Dissado and Fothergill [3] and Auckland et al. [44]. This return growth manifests itself as a
thick channel growing through existing fine tubules from the ground electrode. As this
return growth continues, the widening of fine tubules from the needle tip in a downward
direction also slowly occurs. During this time, the entire tree structure may not increase its
spread but develop thicker channels increasing its visibility. It was reported that the
increase in length and number of thick channels resulted from the hot ionised gas plasma
in the channels seeking the electrically shortest path between electrodes to create and
sustain the breakdown arc [3].
Generally in an electrical tree, negative partial discharges occur with higher repetition rate
but with decreased magnitude than its positive counterparts, even in polyethylene [158,
159]. However from the onset of the return growth (instant the tree touched the ground
electrode), the repetition rate increases significantly relative to the period of original
(downward) growth (before the tree touched the ground electrode). This observation was
attributed to the discharges eroding the ground electrode, increasing the ionisation in the
tubules which resulted in widening of the fine tubules and charge deposition on the walls
increasing conductivity for the breakdown arc [3]. Since breakdown did not occur when
the original tree growth traversed the insulation gap, this suggested the original growth is
non-conducting while return growth is conducting. Further investigations incorporating
electroluminescence activity can be used to confirm this. On the phase-resolved partial
discharge plots, the increased repetition rate of partial discharges is accompanied by
increased phase spread of these (partial discharge) occurrences. As the thick channels
increase, the repetition rate of discharge activity is momentarily extinguished when a tree
channel suffers partial breakdown, implying (full) breakdown of the insulation gap is
imminent. The discharge activity is restored gradually with further thick conductive
channels developing from nearby fine tubules to create favourable conditions for the
breakdown arc. Immediately prior to the luminous breakdown event, the frequency of this
fluctuating discharge activity increases, further increasing the phase spread. However the
overall magnitude of the discharge activity is significantly reduced. This process continues
for an interval of 10 s until breakdown occurs as a consequence of Joule heating and
thermal runaway from the breakdown arc.
127
4.3.
Influence of power quality on electrical tree growth and
breakdown time
4.3.1. Electrical tree growth
In this investigation, as previously outlined in section 3.8.1, the images have been used to
develop width/length plots. These width/length measurements have been correlated to
produce subsequent three-dimensional plots. During this exercise the following limitations
were experienced:
The setup of the light source at the beginning of each test did not always yield the
clearest visual image as the tree grew. In some instances the visibility of extremely
fine tree branches during the original (downward) growth was hindered by a
surplus, deficit or reflection of light in that particular region. An optical remedy
would be to have an external control (outside of the high voltage test area)
reducing such occurrences, increasing visibility of the tree and decreasing time for
setup and data processing.
Using one camera for acquiring images of a three-dimensional electrical tree

visually limits the accuracy in determining the absolute length of branches and the
general direction of the furthest branches. This limitation has been manifested in
the interpretation of the width/length plots and in one extreme case, resulted in the
omission of one sample (T421-09-N) from the width/length plots.
Determining the actual width/length index can be fairly subjective and prone to
human error as it is a manual process (briefly described in section 4.1.2). Software
to assist in index calculations was developed but proved extremely error prone due
to the inconsistency in optics, which was dependent on the variable nature of the
tree growth.
Most electrical trees spread to cover the entire width of the image. This resulted in
the saturation of the width/length curves. This did not imply the tree spread totally
along the horizontal axis of the entire sample. Instead it suggested that external
control of the camera during the experiment is necessary to facilitate improved
focusing without disturbing the experimentation.
Initiation times were not captured for the reason mentioned in section 4.1.1. At the
increased voltage of 18.0 kV peak for stage one, electrical tree inception occurred
extremely quickly ( seconds). Thus engaging the composite waveform precisely
at 10 m proved impossible. From the moment the tree was initiated and became
visible the composite waveforms were instantaneously initiated as the stressing
128

waveform, maintaining a constant peak of 14.4 kV. This meant that the electrical
tree length would be no greater than 30 m at the instant when the composite
waveform was introduced. The creation of the tree was not the focus of this
exercise but merely the starting point for the investigation on the electrical tree
growth.
Figure 4-12 to Figure 4-17 illustrate plots of the electrical tree growth measurements for
the samples tested. The original (downward) growth of the electrical tree reached
maximum when the electrical tree traversed the insulation gap, touching the ground
electrode and has been indicated by . The locations of these markers along the time axis
are fairly consistent within the limitations of experimental error suggesting that the original
(downward) growth rate of the electrical tree appears to be independent of the composite
waveforms.
Figure 4-12 illustrates measurements acquired of absolute tree length. There was a wide
variation in the values of length at which the plots saturated. This variation was attributed
to branches growing and spreading unconventionally beyond the needle tip and deviations
in the insulation gap length between the needle and the plane compounded by the optics
shortcoming which influenced image capture and measurements. Figure 4-13 provides a
normalized plot of tree length referencing the 2 mm insulation gap, further illustrating the
variation in tree lengths captured. As explained in section 4.1.2 the measurement along
the horizontal dimension is important to provide improved interpretation of electrical tree
growth, thus utilizing the width/length ratio.
Figure 4-14 provides a plot of width/length ratio illustrating the close cluster of points
where the electrical trees traversed the insulation gap touching the ground electrode. This
again suggested the original (downward) growth rate of electrical trees toward the ground
electrode (considering both vertical and horizontal dimensions of electrical tree growth)
appears to be reproducibly independent of power quality variation.
Three-dimensional plots were produced to increase readability of the electrical tree growth
curves. In Figure 4-15 beyond the markers, steep gradients prior to flattening indicate a
significant increase in the spread of the electrical tree. This is consistent with observations
reported in section 4.2. During the return growth of the electrical tree there are instances
where the width/length ratio decreased; producing a dip in the curve, identifying periods
when some branches grew upward beyond the needle tip.
129

Figure 4-16 shows the plot as a function of THD while Figure 4-17 shows the plot as a
function of Ks. These figures are quite compact and it is difficult to determine any
correlation of THD and Ks with the change in width/length ratios as a consequence of the
composite waveforms.
130
0
0
20
40
60
80
100
120
140
20
40
60
= touch ground electrode (L= 2mm)
= breakdown
80
Time/min
100
Electrical Tree Length vs Time
0
0
20
40
60
80
100
120
140
120
50
100
140
150
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
Figure 4-12: Plot of electrical tree length vs time of all samples. Inset the cluster of 2 mm
tree length () and scatter of breakdown () points. T325-09-N exhibits significant growth
relative to all samples.
131
Measure of Length/units
132
upward beyond the needle tip e.g.T325-09-N.
0
0
20
40
60
= breakdown
80
Time/min
100
0
0
Normalized Electrical Tree Length vs Time
120
50
100
140
150
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
Figure 4-13: Normalized plot of electrical tree length vs time of all samples. Insulation gap of
length = 2 mm used as reference. Lengths 2 mm registered due to branches growing
Length/mm
0
0
20
40
60
= breakdown
80
Time/min
100
Width/Length Ratio vs Time
0
0
120
50
100
140
150
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
Figure 4-14: Plot of width/length ratio vs time of all samples. Inset the cluster of 2 mm tree
length () and scatter of breakdown () points.
133
Width/Length Ratio
134
Samples
20
40
60
80
Time/min
100
= breakdown
3D Plot of Width/Length Ratio vs Time
120
140
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
Figure 4-15: 3D plot of width/length ratio for all samples highlighting scatter of markers.
Width/Length Ratio
0
40
Width/Length Ratio
35
30
25
THD %
20
15
10
5
0
20
40
60
Time/min
80
100
= breakdown
120
3D Plot of Width/Length Ratio vs Time as a function of Total Harmonic Distortion (THD)
140
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
Figure 4-16: 3D plot of width/length ratio as a function of THD for all samples.
135
136
Width/Length Ratio
0
1.8
Figure 4-17: 3D plot of width/length ratio as a function of Ks for all samples.
1.6
1.5
1.4
Waveshape Factor (K )
1.7
1.3
1.2
1.1
= breakdown
20
40
80
100
Time/min
60
3D Plot of Width/Length Ratio vs Time as a function of Waveshape Factor (K )
120
140
160
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
4.3.2. Deductions
The following deductions summarize the analysis performed in section 4.3.1:
An electrical tree can grow towards the ground electrode and after, towards the
needle tip through the existing tubules. The rate of the original (downward) growth
appeared to be independent of the composite waveform applied at constant peak
voltage. The rate of original (downward) growth toward the ground electrode was
lower than the subsequent return growth toward the needle tip.
A change in the growth characteristics of the electrical tree in epoxy resin due to
variation in harmonic content has not been detected. To determine a change in
structure of the tree, should such a relationship exist, requires a revised approach
improving the precision and accuracy of the measuring system.
These plots cannot confirm that acceleration of field assisted ageing processes
occur due to composite waveforms. Tree growth measurements may not be the
best measurand to support this claim. One alternative measurand which can be
studied in future work is electroluminescence activity.
The voltage of 14.4 kV peak was consistent with previous studies conducted [42,
153, 161] but may be too high for the waveshape factor to have a dominating
influence on the tree growth characteristics. This voltage level was also justified as
it allowed significant growth and breakdown activity in the 2.5 hour duration,
facilitating numerous experiments. Future work should employ lower voltages for
increased durations for the growth of the electrical trees subjected to the
composite waveforms.
4.3.3. Breakdown time

The breakdown results for all 42 samples have been provided in Table 4-5, with the
following summary:
31 samples (74 %) suffered breakdown in stage 2 within 2.5 hours at 14.4 kV

peak.
3 samples (7 %) suffered breakdown in stage 3 within the 10 minute voltage ramp.
8 samples (19 %) did not breakdown during the entire test.
137

Table 4-5: Breakdown results of 42 test samples.
No.
Sample
THD
(%)
Ks
Vrms
(kV)
Vpkpk
(kV)
Stage 2
Breakdown
Time (s)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
T323-09-N
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T335-12-N
T341-08-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T366-01-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T385-12-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T396-07-N
T411-11-N
T412-09-N
T414-08-N
T415-01-N
T421-09-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T431-07-N
T433-08-N
5.0
5.0
5.0
5.0
0.0
17.8
7.9
5.0
7.9
0.0
5.0
17.8
40.0
40.0
0.0
5.0
5.0
40.0
17.8
7.9
5.0
5.0
5.0
40.0
0.0
7.9
17.8
5.0
5.0
7.9
0.0
17.8
5.0
5.0
40.0
5.0
5.0
7.9
17.8
40.0
0.0
5.0
1.06
1.27
1.06
1.27
1.00
1.60
1.60
1.03
1.60
1.00
1.06
1.60
1.56
1.56
1.00
1.27
1.03
1.56
1.60
1.60
1.03
1.27
1.06
1.56
1.00
1.60
1.60
1.03
1.27
1.60
1.00
1.60
1.06
1.03
1.56
1.06
1.27
1.60
1.60
1.56
1.00
1.03
10.40
10.19
10.40
10.19
10.18
10.14
10.21
9.71
10.21
10.18
10.40
10.14
11.04
11.04
10.18
10.19
9.71
11.04
10.14
10.21
9.71
10.19
10.40
11.04
10.18
10.21
10.14
9.71
10.19
10.21
10.18
10.14
10.40
9.71
11.04
10.40
10.19
10.21
10.14
11.04
10.18
9.71
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
28.8
DNB
5591
7291
7021
5564
6450
DNB
DNB
7366
6180
7375
8600
6909
7070
6888
7518
8810
DNB
7758
6780
7109
7321
6390
7710
7815
DNB
7187
8405
8945
8278
DNB
DNB
8261
DNB
DNB
4274
5625
4609
3469
3590
DNB
DNB
Stage 3
Breakdown
Voltage
(kV)S
DNB
-----15.33
DNB
---------DNB
-------DNB
----DNB
13.83
-DNB
DNB
-----12.67
DNB
DNB = Did Not Breakdown
Figure 4-18 shows a scatter plot of the 31 breakdown points as a function of THD,
highlighting the linear best fit of mean and standard deviation times. Figure 4-18 suggests
as THD increased, breakdown times are increasingly scattered and have reduced mean
values. Distorted environments have yielded a reduction in insulation life [49, 54, 62, 64,
138

66-69]. This is governed by the spread about the mean value and not a shifted mean
value, especially in a small sample size.
Breakdown Time vs THD
10000
9000
8000
Breakdown Time (s)
7000
mean
6000
5000
4000
3000
standard deviation
2000
1000
0
-5
10
15
20
25
Total Harmonic Distortion (%)
30
35
40
45
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T412-09-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T421-09-N
= DNB
Figure 4-18: Breakdown time vs THD illustrating the mean and standard deviation.
At THD = 0 %, (wave 7) the waveform is undistorted and the breakdown times are
characterized by the lowest standard deviation of 968 s, confirming the potential of
distorted waveforms to produce more weak points (greater spread) relative to undistorted
waveforms. The shortest breakdown time recorded (3469 s) was achieved from excitation
employing the 7th harmonic (wave 11), consistent with research conducted by Bozzo et al.
[66].
Each batch of samples was limited to a maximum of six samples due to production
constraints outlined in section 3.6. Batches with at least three tested samples were closely
examined. As THD increased batches T33x, T34x and T38x revealed no trend in the
breakdown times, while batches T39x and T42x showed a reduction in the breakdown
times. Those batches illustrating no trends contained at least one sample tested using the
undistorted wave 7 (THD = 0 %) while batches showing a reduction in breakdown time did
not have any samples tested using wave 7. Thus as THD increased at constant peak
voltage, the variation in breakdown trends did not reveal a deterministic relationship with
THD. This is shown in Figure 4-19.
139

Breakdown Time vs THD for Selected Sample Batches
Breakdown Time (s)
9000
T39x
8000
T34
7000
T38x
6000
T33x
5000
4000
T42x
3000
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
THD (%)
Figure 4-19: As THD increased at constant peak voltage, the variation in breakdown trends
did not reveal a deterministic relationship with THD. Lines are not for best fit or trend
purposes but to assist the reader identify result groups
Breakdown Time vs Ks
10000
9000
8000
mean
Breakdown Time (s)
7000
6000
5000
4000
3000
standard deviation
2000
1000
0
0.9
1.1
1.2
1.3
1.4
Waveshape Factor (K )
s
1.5
1.6
1.7
T324-13-N
T325-09-N
T331-13-N
T332-07-N
T333-11-N
T342-12-N
T343-07-N
T344-09-N
T345-11-N
T346-01-N
T354-01-N
T355-07-N
T363-13-N
T365-08-N
T372-11-N
T373-12-N
T374-08-N
T376-13-N
T381-09-N
T382-01-N
T383-07-N
T391-11-N
T392-08-N
T393-13-N
T394-12-N
T412-09-N
T423-13-N
T424-12-N
T425-11-N
T426-01-N
T421-09-N
= DNB
Figure 4-20: Breakdown time vs Ks illustrating the mean and standard deviation.
Figure 4-20 shows a scatter plot of the 31 breakdown points as a function of the
waveshape parameter Ks, highlighting the linear best fit of mean and standard deviation
times. As in Figure 4-18 and Figure 4-20 suggests that as Ks increases, breakdown times
140

are increasingly scattered with reduced mean values. The standard deviation was lowest
at Ks = 1.00 and Ks = 1.03 at 968 s and 889 s respectively.
In Figure 4-21 samples of the same batches analyzed in Figure 4-19, revealed a region
exists where the waveshape may have decreased influence, potentially resulting in a local
maxima of breakdown time at Ks = 1.27. For Ks = 1.27, samples were subjected to wave
13 (containing the 5th, 7th, 11th, 13th, 23rd and 25th harmonic orders) of THD = 5.0 %.
However due to limited tested samples from the same batch it was difficult to confirm this
observation. Further experimental investigations with more samples per batch are
required to verify the reproducibility of this occurrence. Hence visual analysis of plots
Figure 4-18 through to Figure 4-21 yielded inconclusive findings, warranting a statistical
approach with more data.
Breakdow n Tim e vs Ks for Selected Sam ple Batches
Breakdown Time (s)
9000
8000
7000
T39x
T38x
T34x
T33x
6000
5000
T42x
4000
3000
1.00
1.10
1.20
1.30
1.40
1.50
1.60
Ks
Figure 4-21: As Ks increased at constant peak voltage, the variation in breakdown revealed a
potential region at Ks=1.27 for maximum breakdown times. Lines are not for best fit or trend
purposes but to assist the reader identify result groups.
After the second stage (9000 s) elapsed some samples did not suffer breakdown,
therefore Table 4-5 represented right-censored data. The Weibull distribution has been
applied here to statistically model this failure data using the two parameter cumulative
distribution function in equation (4-4).
F (t ) = 1 e
(4-4)
Where is scale parameter and is shape parameter. The scale parameter represents
the time for the failure probability of 0.632, which is analogous to the mean of the normal
141

distribution. The shape parameter is an estimate of the range of breakdown times [166].
The linear natural logarithmic relationship for parameter estimation via the graphical
method is governed by equation (4-5):
ln(t ) ln = ln ln (1 F (t ) )
(4-5)
To accommodate samples in the data set which did not breakdown, the following mean
ranking estimator was employed for the plot of ln vs ln[-ln(1-F(t))].
F ( i, n ) =
i 0.3
n + 0.4
(4-6)
The processed breakdown data of Table 4-5 yielded Table 4-6 with the determined scale
and shape parameters. All Weibull plots have been archived in Appendix A while
selected plots have been provided in Figure 4-22.
Table 4-6: Graphically determined and values from the breakdown time data.
THD
Composite
Harmonic
Sample
(%)
Waves
Order
Size
ALL
ALL
ALL
3,5,7,11,13,23,25
1.60
ALL
11,12
ALL
5.00
1.00
Ks
Failed
42
31
8466
4.3
5,7,11,13,23,25
12
8724
3.2
8,9,13
5,7,11,13,23,25
18
14
8446
5.0
0.00
--
7820
5.8
1.60
7.85
12
5,7,11,13,23,25
8955
3.4
1.60
17.8
11
8643
2.5
1.03
5.00
9770
6.9
1.06
5.00
8111
3.6
1.27
5.00
13
5,7,11,13,23,25
7565
5.5
1.56
40.0
9494
2.3
142

Ks =1.60 & THD=All
Total population
0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
0
-2
-4
-6
=8467
=4.3
8.2 8.4 8.6 8.8
ln(t/sec)
-2
-4
-6
9.2
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
9.2
0
-2
-6
8.2 8.4 8.6 8.8

ln(t/sec)
Ks =1.0 & THD=0.0%
Ks =All & THD=5.0%
-4
=8725
=3.2
=8446
=5.0
8.2 8.4 8.6 8.8
ln(t/sec)
9.2
-2
-4
-6
=7820
=5.8
8.2 8.4 8.6 8.8
ln(t/sec)
9.2
Figure 4-22: Weibull plots with and values for the total population of tested samples,
subsets of Ks=1.60, THD=5 % and undistorted waveform where Ks=1.0 & THD=0 %.
Figure 4-22 and Table 4-6 illustrate the scale and shape parameters acquired for the
total population with a sample size of 42 samples as well as subsets of the 42 samples
tested. The sample sizes used for subsets Ks = 1.6 and THD = 5.0 % were 12 and 18
respectively which were sufficient for credible results. A sample size less than five can
lead to erroneous estimation of these parameters [166]. Hence with marginal sample
sizes of six for the remaining subsets, there exists some uncertainty in the results of the
parameters estimated for these subsets highlighted in Table 4-6. An increase of the
sample size is the best avenue to address this uncertainty. Alternatively, application of
techniques to determine unbiased estimates of these parameters can be employed [167,
168]. Both methods can be undertaken as parallel activities for future work.
The graphically calculated scale , and shape , parameters have been plotted in Figure
4-23 illustrating the modelled density and cumulative distribution curves. The maximum
experiment time = 9000 s (2.5 hrs), hence right-censored data has been used in
producing these curves, therefore extrapolation occurs beyond 9000 s. In Weibull
probability density plots, when 2.6 < < 3.7 the coefficient of skewness approaches zero
i.e. no distribution tail and this can be approximated to the normal probability density
function [169]. High values of imply a low degree of data variability [170].
143

Probability Density Function
-4
x 10
3
Ks = All THD= 5.00%
Total population
Ks= 1.60 THD= All
1
0
0
2000
4000
0.6
0.4
8000
10000
Time (s)
12000
14000
16000
0.2
Ks= All THD= All
0.15
Ks=1.60 THD= All
Wave 11,12
Ks= All THD=5.00% Wave 8,9,13
0.1
Ks=1.00 THD=0.00% Wave 7
0.05
0
0
2000
Ks=1.60 THD=7.85% Wave 12
4000
Ks=1.60 THD=17.8% Wave 11
0.2
0
0
18000
Cumulative Distribution Function
1
0.8
6000
= 8466 =4.3
= 8724 =3.2 Wave 11,12
= 8446 =5.0 Wave 8,9,13
= 7820 =5.8 Wave 7
= 8955 =3.4 Wave 12
= 8643 =2.5 Wave 11
= 9770 =6.9 Wave 8
= 8111 =3.6 Wave 9
= 7565 =5.5 Wave 13
= 9494 =2.3 Wave 1
Ks=1.03 THD=5.00% Wave 8

Ks=1.06 THD=5.00% Wave 9
2000
4000
6000
8000 10000
Time (s)
12000
14000
16000
18000
Ks=1.27 THD=5.00% Wave 13

Ks=1.56 THD=40.0% Wave 1
Figure 4-23: The probability density and cumulative distribution function plots are similar
shapes except wave 8 and wave 1 which correspond to minimum and maximum values
respectively containing the highest values. Inset at T = 4000 s wave 9 is most influential.
Figure 4-23 shows the mean estimates of the probability density functions closely
clustered, providing very limited information in comparison to the width distributions which
are described by the shape parameter . The Weibull shape parameter estimated for all
42 test samples yielded ALL = 4.3. A reduction of 25.5 % resulted with 11,12 = 3.2, for 3 %
reduction in when the subset of waves 11 and 12 were analyzed collectively (for Ks =
1.60 and THD = 17.8 % and 7.85 % respectively). When waves 8, 9 and 13 were
analyzed collectively for THD = 5 % and Ks = 1.03, 1.06 and 1.27 respectively, there was
an increase of 14 % in the shape parameter resulting in 8,9,13 = 5.0. Notably the
comparison of the respective scale parameter values only differed by 3.2 %. It is difficult
to compare and contrast the degree of distortion the THD and waveshape factor Ks
indices represent, in addition to the physical interpretation each shape parameter conveys
since the composite waveforms are different. Although these indices are mathematically
linked, it would therefore be very useful for future studies to conduct experimental work
with normalised values for a linear comparison of THD and Ks, thus ultimately acquire a
weighting factor for each index.
144

The subset of waves 11 and 12 were analyzed separately with THD = 17.8 % and 7.85 %
respectively, for Ks = 1.60. Separately 11 = 2.5, and 12 = 3.4, hence the shape parameter
increased as THD decreased with only 3 % variation in . When the THD values were
significantly higher in wave 1 at 40.0 % and wave 11 at 17.8 % respectively 1 = 2.3 and
11 = 2.5, albeit Ks differed for these waves by only 2.5 %. These observations confirm
that electrical treeing in a significantly distorted environment where the THD is beyond the
recommended level outlined in [49] will consequently reduce operational life. Conversely,
electrical treeing in an undistorted environment of the fundamental power frequency only
(wave 7) will yield relatively increased operational life, where 7 = 5.8. This observation is
consistent with previous studies in distorted operational environments for electrical treeing
[67] and other mechanisms of insulation ageing [51, 54, 62-64, 66].
The undistorted waveform (wave 7) yielded 7 = 5.8 and 7 = 7820, while wave 8 yielded
8 = 6.9 and 8 = 9770. Wave 8 resulted in a lower voltage of 9.71 kV rms in comparison
to wave 7 with 10.18 kV rms, albeit both waves had the same peak voltage of 14.4 kV.
The statistically weak analysis (only three of six samples failed) suggested that tolerable
THD and Ks indices can describe distorted waveforms yielding increased shape
parameters (relative to undistorted waveforms), provided the distortion decreased the rms
index. Thus suggesting, the amalgamated effect of some harmonic combinations can
prolong insulation life. This was characteristic of the isolated 5th harmonic.
A discrepancy was observed with the collective results from waves 8, 9 and 13 compared
with the isolated results from these waves as seen in Table 4-7.
Table 4-7: Comparison of the 5th and 7th harmonic influence on breakdown time values.
Wave
Harmonic
Order
Ks
Voltage
Magnitude of
Shape
Samples
rms (kV)
Harmonic (%)
Parameter
Failed
---
1.00
10.00
---
5.8
1.03
9.71
5.00
6.9
1.06
10.40
5.00
3.6
13
5,7,11,13,23,25
1.27
10.19
2.00
5.5
Collectively 8,9,13 = 5.0 with THD constant at 5 %. Individually, 8 = 6.9, 9 = 3.6 and 13 =
5.5 where Ks8 Ks9 Ks13. A proportional trend for KS with (as previously observed for
THD with ) might have been expected to yield 8 9 13, instead the results attained
were 9 13 8. The uncertainty surrounding 9 = 3.6 (7th harmonic) requires further
investigation and the following observations must be considered:
145
Waves 8 and 9 have identical THD levels of 5 % but with Ks 3 % resulting in 8

= 6.9 and 9 = 3.6, i.e. 48 % as illustrated in Table 4-7. The rms value of
wave 8 was the lowest compared to wave 7 (undistorted), wave 9 and wave 13
which were all at the same peak voltage. Wave 8 (5th harmonic) appeared least
detrimental having the lowest sample failure rate (50 %) and the highest value
(8 = 6.9). Additionally, the rms value of wave 9 (7th harmonic) was the highest and
resulted in a significantly reduced value (9 = 3.6). This was reflected in the
probability density function of Figure 4-23. Wave 13 characterised by THD = 5 %
contained both 5th and 7th harmonic components and resulted in 13 within the
range 9 13 8. There is therefore the need to further investigate the impact of
the 7th harmonic as it appears more influential than the 5th harmonic for breakdown
due to electrical treeing at constant peak voltage. The shortest breakdown times
(including electrical treeing) at constant rms voltage were also functions of the 7th
harmonic [66, 67].
An increase in the 7th harmonic magnitude resulted in a reduction of the shape
parameter. This was independent of the other harmonics present as illustrated in

Table 4-8.
Table 4-8: Influence of 7th harmonic magnitude on breakdown time values.
Wave Harmonic Order Magnitude of 7th Harmonic (%)
Shape Parameter
13
5,7,11,13,23,25
2.00
5.5
5.00
3.6
12
5,7,11,13,23,25
13.2
3.4
11
17.8
2.5
During the tests using waves 9 and 11, five of the six samples tested failed, within
the 2.5 hour duration of stage two giving similar levels of confidence in the results
acquired.
4.3.4. Deductions
The following deductions summarize the analysis performed in section 4.3.3:
Visual interpretation of the breakdown data collected is quite vague and

inconclusive. However the spread of data points in Figure 4-21 suggest further
experimental investigations are required in the vicinity of Ks = 1.27.
146
At approximately constant Ks, increasing THD reduced Weibull shape parameters

of breakdown times indicating increased failure rates for less than 10 % variation
in the respective Weibull scale parameters. This cannot be confirmed for constant
THD and increasing Ks. This is summarized in Table 4-9.
Table 4-9: Variation in Weibull and values for breakdown times to corresponding Ks and
THD indices for each composite waveform.
Ks
% THD Wave
1.60
7.85
12
8955
3.4
1.60
17.8
11
8643
2.5
1.56
40.0
9494
2.3
1.03
5.00
9770
6.9
1.06
5.00
8111
3.6
1.27
5.00
13
7565
5.5
Confirmation that electrical treeing in a distorted environment exceeding THD

standards [49] will ultimately reduce the operational life of the insulation.
At constant peak voltages, the synergistic effect of the resultant rms and the
waveshape was a function of the composite waveforms frequency components
and their magnitude. Thus distorted waveforms can reduce or increase breakdown
times relative to the undistorted waveform.
The 7th harmonic produced low Weibull shape parameters, 3.6 for breakdown
times due to electrical treeing at constant peak voltage, suggesting it (7th
harmonic) is more detrimental to insulation life than the 5th harmonic, = 6.9.
147
4.4.
Influence of power quality on partial discharge
patterns
In this investigation, (outlined in section 3.8.2) the partial discharge data was analyzed
visually using phase-resolved partial discharge plots and statistically using Weibull plots of
partial discharge magnitudes. Processed partial discharge data with images of electrical
tree growth (where available) have been archived in Appendix B. Examples have been
provided in Figure 4-24, Figure 4-36 and Figure 4-41 from test K115 and discussed in
later sections.
Table 4-10 provides a description of the initial state of the electrical tree prior to stressing
the sample with the seven defined composite waveforms (section 3.7). The use of
constant peak voltages of 10.8 kV peak and 14.4 kV peak were consistent with the
previous experimental work conducted throughout this research and other investigations
[42, 44, 153, 161]. The selected voltage level was determined to ensure a sufficient level
of discharge activity for each test for the respective time per wave allocation.
Table 4-10: Description of samples tested.
Electrical Tree Growth Description
Time per wave
Vpeak
(At beginning of experiment)
(mins)
(kV)
Extensive spread, touching ground electrode.
10.8
T291
Moderate spread, midway to ground electrode.
14.4
K104
T312
One branch very infant stage of tree.
10.8
K106
T314
14.4
K108
T316
One branch very infant stage of tree.
14.4
K109
T311
Bush like structure very infant stage of tree.
14.4
K111
T252
14.4
K112
T273
10.8
K113
T273
14.4
K114
T245
14.4
K115
T245
14.4
Test
Sample
K101
T264
K102
During this exercise the following limitations were experienced:
As the electrical tree grew, its ageing state changed. The dominant ageing
mechanism is expected to influence the nature of the partial discharge activity.
Hence the reduction in time per wave for tests K113, K114 and K115 in an attempt
148

to capture partial discharge activity without a significant change in the electrical
trees ageing state.
Two samples were used twice during testing. Sample T273 for tests K112 and
K113 and sample T245 for tests K114 and K115. These samples were tested
twice to identify any detectable change in the ageing state from the partial
discharge patterns acquired (as mentioned in the previous point).
The starting point for each test would never realize identical conditions due to the
processes that define electrical tree growth.
The setup of the light source at the beginning of each test did not always yield the
best visual image as the tree grew. This has been addressed in section 4.3.1.
However, in this exercise the image acquisition was not the primary source of
data.
Figure 4-24 shows the discharge activity plotted for one sample with the excitation
waveforms described in section 3.7. Distinct features exist for discharge activity from each
waveform. This is consistent with previous work investigating partial discharge patterns as
a consequence of harmonic interference [72, 73] and superposed frequencies [74].
149

Plot 1 K115
-500
50
100
300
50
100
150
200
250
300
50
100
150
200
100
250
300
50
100
250
300
350
150
200
250
300
350
250
300
350
250
300
350
250
300
350
250
300
350
Phase/deg
Plot 6 K115
50
100
Phase/deg
Plot 7 K115
500
200
-500
350
150
Phase/deg
Plot 4 K115
500
Charge/pC
0
50
-500
350
-500
500
Phase/deg
Plot 5 K115
500
-500
350
Charge/pC
Charge/pC
250
-500
Charge/pC
200
Phase/deg
Plot 3 K115
500
Charge/pC
150
Plot 2 K115
500
Charge/pC
Charge/pC
500
150
200
Phase/deg
-500
50
100
150
200
250
300
350
Phase/deg
Plot 8 K115
-500
50
100
300
50
100
150
200
250
300
50
100
150
200
250
300
350
Phase/deg
Plot 14 K115
500
100
150
200
Phase/deg
Plot 11 K115
50
100
150
200
Phase/deg
Plot 13 K115
500
Charge/pC
0
50
-500
350
-500
500
Phase/deg
Plot 12 K115
500
-500
350
Charge/pC
Charge/pC
250
-500
Charge/pC
200
Phase/deg
Plot 10 K115
500
Charge/pC
150
Plot 9 K115
500
Charge/pC
Charge/pC
500
-500
50
100
150
200
Phase/deg
-500
50
100
150
200
250
300
350
Phase/deg
Figure 4-24: Phase-resolved partial discharge plots for sample K115.
150
4.4.1. Partial discharge modelling

Harmonic representation is frequency based. Insulation subjected to electrical stress may
experience rapid voltage changes, sustained voltages as well as progressively changing
voltages. Hence insulation experiences time-domain stress variation and physical models
must be time-domain oriented.
Wave 13 differed from the undistorted wave 7 as it contained the 5th, 7th, 11th, 13th, 23rd
and 25th harmonic orders. Figure 4-25 highlighted the impact of these frequency
components on the time domain characteristics of the resultant waveform. The production
of this rippled wave 13 would appear passive, especially if the power quality is
unmonitored. However, considering wave 7 and wave 13 have the same peak magnitude;
it is a key feature that the magnitudes of the derivative waveforms differ by at least a
factor of two as observed in Figure 4-25.
Wave 7 partial discharges acquired
100
0
-100
20
Wave 7 THD=0.0% Ks=1.00

Waveform
Derivative
0
-2
0
10
200
Charge/pC
Per Unit
20
PRPD Plot
0
-100
-200
0
60
100
Time/ms
Waveform
Derivative
0
-2
180
360
10
80
200
PRPD Plot
100
0
-100
-200
0
20
Phase/deg
Time/ms
Wave 13 partial discharges acquired
200
Charge/pC
40
Time/ms
Wave 13 THD=5.0% Ks=1.27
Charge/pC
-200
0
Per Unit
Charge/pC
200
180
360
Phase/deg
100
0
-100
-200
0
20
40
60
80
Time/ms
Figure 4-25: Time domain representation of derivatives and electrical treeing partial
discharge activity captured from wave 7 (plot 6) and wave 13 (plot 8) from K106 for one
acquisition (80 ms).
151

Hence any insulation subjected to wave 13 defined by limits of THD 5.0 % and Ks 1.27,
would experience a significant increase in rate of change of electrical stressing as a
consequence of these ripples relative to the undistorted wave 7.
Figure 4-25 also revealed that wave 13 produced marginally greater partial discharge
magnitudes and increased density of partial discharges relative to that of wave 7. The
partial discharge patterns for each acquisition were not consistently reproduced over the
four cycles. This was characteristic of partial discharges along a channel [171] in electrical
tree branches. During the 80 ms acquisition, the total number of partial discharges
acquired as a consequence of wave 13 was 167 whilst for wave 7, 110. The acquisitions
of discharge data for both plots were not successive but occurred comfortably within 12
minutes. During this time the electrical tree continued its growth and spread towards the
ground electrode. The onset of the return growth had not yet occurred from the images in
Appendix B. This was confirmed by the relatively comparable magnitude of partial
discharges in Figure 4-25. Hence, this was sufficient evidence to rule out significant
changes in the dominant ageing mechanism. The partial discharge activity from one
acquisition did not provide enough visual evidence to detect the influence of the variation
between wave 13 and wave 7. Accumulated data displayed in Figure 4-24 provided the
visual confirmation that the changes in the waveform produced changes in the partial
discharge patterns.
Understanding why these changes in the shape of the waveform produce varied partial
discharge pattern is critical to modelling partial discharge activity. Consider the simple
void model [73] illustrated in Figure 4-26.
Explanations in the literature have looked solely at the effect of the instantaneous value of
the voltage as the dominant condition for a partial discharge whilst acknowledging the role
of the derivative in producing the partial discharge pattern [72, 158, 159], as observed in
Figure 4-26 [73]. This confirmed the synergistic relationship between both parameters i.e.
V and dV dt . However, the role of the derivative has not been fully explained nor modelled
in conjunction with the instantaneous voltage value, albeit the electrical field is the
causative agent.
152
Figure 4-26: Changes of voltage in partial discharge (PD) source at pure and harmonic
test voltages in a solid dielectric with a) PD source (void); t=thickness of void b) Equivalent
circuit diagram (abc), where Ca=capacitance of solid dielectric, Cb=capacitance of solid
dielectric in series with void and Cc=capacitance of void c) PD mechanisms at pure
sinusoidal test voltage d) Effect of harmonics in test voltage on void voltage Uc and PD [73].
Increased dV dt has the effect of sharply increasing the voltage across a void, thereby
providing a sudden burst of energy on discharge at the molecular level. This undoubtedly
will increase the chance of initiating a free electron provided the threshold electric field is
exceeded. The electric field is a function of the instantaneous voltage. Upon that condition
being satisfied, this increased electrical field strength will accelerate electrons to bombard
molecules in the gas of the void generating more collisions which in turn liberate more
electrons, creating the avalanche effect for a partial discharge to occur. Hence those
ripples in the composite waveforms are the prescription for substantially increased dV dt
magnitudes which can induce overstressing to complement the instantaneous voltage and
influence the partial discharge magnitude and occurrence, i.e. the partial discharge
pattern.
153

As mentioned in section 3.7, the literature reported the discharge magnitude in electrical
treeing was related to the instantaneous value of the applied voltage while the frequency
of discharge activity is dependent on the time derivative of the waveform [159]. However
the waveforms (power frequency sinusoidal and triangle) employed in [159] contained
extreme derivatives. The derivative of the power frequency sinusoidal was low relative to
the derivative of the triangle wave which modelled the fastest rate of change; a square
wave. Hence those conclusions were based on from an unusually extreme data set and
caution must be administered when applying such conclusions as the derivatives of the
composite waveforms employed in this research had a wide range of values.
Test K101
Test K106
partial discharge
V
dV/dt
1
Per Unit
Per Unit
1
0
-1
-1
-2
-2
partial discharge
V
dV/dt
50
100
150
200
Phase/deg
250
300
350
50
100
Test K111
350
1
Per Unit
Per Unit
300
partial discharge
V
dV/dt
1
0
-1
-1
-2
-2
250
Test K115
partial discharge
V
dV/dt
150
200
Phase/deg
50
100
150
200
Phase/deg
250
300
350
50
100
150
200
Phase/deg
250
300
350
Figure 4-27: Wave 13 discharge pattern compared to the V and dV/dt plots from four tests.
Consider wave 13 illustrated in Figure 4-27 whose composition consists of many harmonic
orders. These plots represent the captured discharge data reproduced in Appendix B.
While the magnitudes of the discharges vary, the discharge pattern occurred on the rising
edge of the dV dt slope and not the peaks. There also exist dead zones where no
discharge activity occurred but the dV dt slope magnitudes were comparable to phase
locations where discharge activity did occur. A heuristic approach suggested that since
dead zones occurred in the 2nd (V > 0 and dV dt < 0) and 4th (V < 0 and dV dt > 0)
154

quadrants, a threshold based on polarity and magnitude due to the synergistic effect of V
and dV dt influenced the occurrence of the discharge activity.
Modelling such trends would be best reflected using functions with quadratic and cubic
properties. This led to the application of the trigonometric hyperbolic functions, cosh and
sinh. The graphs for these functions in Figure 4-28 show that at increased values these
functions increase asymptotically thereby amplifying increased magnitudes of V and
dV
dt
.
Hyperbolic trigonometric functions
20
10
dead zone
0
-10
-20
y=cosh(x)
y=sinh(x)
-4
-2
0
x
Figure 4-28: Plots of the cosh and sinh hyperbolic functions illustrating potential to model
dead zones of partial discharge activity.
This provided the platform to apply equation (4-7).
cosh V + dV
dt
(4-7)
This equation can be expanded to reveal equation (4-8) illustrating the relationship
between the cosh and sinh functions.
cosh V + dV
dt
) = cosh(V )cosh (dV dt ) + sinh(V ) sinh (dV dt )
(4-8)
155
The evaluation of cosh(V )cosh dV
evaluation of sinh(V ) sinh dV

dV
dt
dt
dt
will always yield a positive result, while the
will yield a result dependent on the polarity of V and
. The sum of these two parts will determine if the final result lies in the dead zone.
Since cosh > 1, the dead zone lies above this magnitude.
However due to the non-linearity of the hyperbolic functions it was imperative to
investigate the range of operation which suited the intended application of partial
discharge modelling. This has been illsutrated in Figure 4-29.
Wave 13 Amplitude= 1
Rate of change
Amplitude
0
-1
0
50
100
150
200
Phase/deg
250
300
dV/dt
-2
50
100
150
200
Phase/deg
250
300
350
0
-10
0
350
Wave 13 Amplitude= 10
10
Rate of change
Amplitude
50
2
50
100
150
200
Phase/deg
250
300
350
250
300
350
dV/dt
0
-20
0
50
100
150
200
Phase/deg
300
350
cosh(V+dV/dt)
50
x 10
10
250
11
2
1.5
1
0.5
150
200
Phase/deg
20
x 10
cosh(V+dV/dt)
100
100
150
200
Phase/deg
250
300
350
10
2
0
-10
-3
0
cosh(x)
sinh(x)
-2
-1
0
x
cosh(x)
sinh(x)
-2
3
-20
-10
0
x
10
20
Figure 4-29: Normalisation of the waveform to prevent operation in the asymptotic region of
the hyperbolic functions cosh and sinh with amplitude = 1 (left) and amplitude = 10 (right).
In order to capitalise on the possible dead zone recognition of the hyperbolic functions,
the voltage waveform was normalised to prevent erroneous results with increase dead
zones as a consequence of operating in the asymptotic region. This allowed modification
of Figure 4-27 to include plots of equation (4-7) to produce Figure 4-30.
156

Test K101
partial discharge
V
dV/dt
cosh(V+dV/dt)
4
2
-2
-2
50
100
150
200
Phase/deg
250
300
partial discharge
V
dV/dt
cosh(V+dV/dt)
Per Unit
4
Per Unit
Test K106
350
50
100
Test K111
-2
-2
50
100
150
200
Phase/deg
350
300
250
300
350
partial discharge
V
dV/dt
cosh(V+dV/dt)
Per Unit
Per Unit
250
Test K115
partial discharge
V
dV/dt
cosh(V+dV/dt)
150
200
Phase/deg
50
100
150
200
Phase/deg
250
300
350
Figure 4-30: Improved modelling of partial discharge patterns with cosh(V+dV/dt). The
dotted circles highlight improved dead zone recognition.
This modification using the cosh function for modelling of partial discharge patterns with
wave 13 improved reproduction of the phase-resolved plot significantly. Figure 4-30
illustrated the effectiveness of modelling the dead zones particularly in tests K101 and
K106, where the discharge activity was greater in magnitude and possibly repetition rate
than tests K111 and K115. The next challenge was applying this model to the other
composite waveforms to determine its effectiveness at modelling the partial discharge
patterns due to power quality variation. An example has been provided in Figure 4-31.
157

Wave 12 - Plot 9 K106

20
Per Unit
Per Unit
6
2
0
15
10
5
0
-2
0
50
100
150
200
Phase/deg
250
300
350
50

Per Unit
Per Unit
150
200
Phase/deg
250
300
350
300
350
300
350
10
5
0
2
1
0
-1
50
100
150
200
Phase/deg
250
300
350
50
100
150
200
Phase/deg
250

2
Per Unit
2
Per Unit
100
1
0
-1
0
50
100
150
200
Phase/deg
250
300
350
1
0
-1
50
100
150
200
Phase/deg
250

partial discharge
V
dV/dt
cosh(V+dV/dt)
Per Unit
4
2
0
-2
50
100
150
200
Phase/deg
250
300
350
Figure 4-31: K106 partial discharge patterns due to the composite waveforms with
comparison to the cosh(V+dV/dt) model. The dead zones highlighted by the dotted circles
do not fully correlate with the recorded partial discharge activity.
The cosh plots from waves 13, 12 and 11 resulted in high magnitudes hindering
comparison with the discharge activity correlating in Figure 4-31. All the resulting cosh
plots have been archived in Appendix B for further reference. The dead zone in wave 11
(increased magnitude of 7th harmonic only) did not correlate to the cosh model. There was
better agreement in waves 9 and 8, but significant discharge activity towards the end of
the predicted dead zones. In other data sets for waves 9 and 8, better correlation to the
perceived dead zones was observed as illustrated in Figure 4-32.
158

20
Per Unit
Per Unit
6
2
0
15
10
5
0
-2
0
50
100
150
200
Phase/deg
250
300
350
50

Per Unit
Per Unit
150
200
Phase/deg
250
300
350
300
350
300
350
10
5
0
2
1
0
-1
50
100
150
200
Phase/deg
250
300
350
50
100
150
200
Phase/deg
250

2
Per Unit
2
Per Unit
100
1
0
1
0
-1
0
50
100
150
200
Phase/deg
250
300
350
50
100
150
200
Phase/deg
250

partial discharge
V
dV/dt
cosh(V+dV/dt)
Per Unit
4
2
0
-2
50
100
150
200
Phase/deg
250
300
350
Figure 4-32: Improved dead zone recognition highlighted by the dotted ellipses for K101
for waves 9 and 8.
The cosh model illustrated an improved fit for the dead zones for all waveforms except
wave 11 in Figure 4-32. Referring to Table 4-10 the sample condition for test K101 was
characterised by extensive spreading of the electrical tree touching the ground plane.
Unfortunately a series of images were not captured for comparison of this sample.
However test K106 sample was described as a moderate spread electrical tree, therefore
not as mature as the electrical tree in test K101. Nonetheless in Figure 4-30, results from
both tests K101 and K106 illustrated a high degree of correlation with the dead zone
dV
for wave 13. This suggests that the harmonic

modelling from the function cosh V +
dt
composition governing the shape of the wave may produce discharge patterns with
features which cannot be predicted by this cosh function. Additionally, the ageing state of
the insulation and the dominant ageing mechanism may have influenced the nature of
discharge activity for tests K101 and K106. Therefore further investigation for the cosh
models applicability should involve electrical tree partial discharge acquisition operating
within the region of growth closest to linear, highlighted in Figure 4-33. However, due to
the nature of electrical treeing, this region can usually be identified only after entering the
runaway stage. Hence this is not easily achieved but the recommended approach would
159

involve a reduction of the applied voltage magnitude to reduce the growth without a
significant sacrifice in the conditions for increased discharge activity.
Figure 4-33: Typical tree growth curve with suggested operating region for further
experiments investigating electrical tree partial discharge modelling.
The partial discharge activity from a triangle power frequency waveform was subjected to
the cosh function model. The triangle waveform was selected because it contained no
visual features of a sinusoid and was not employed as a test waveform in this research.
Additionally, the triangle waveform is used in electrical treeing research [159] and is easily
reproduced for future verification. Figure 4-34 illustrated good correlation to the dead
zone and the general shape of the partial discharge pattern.
Triangle Wave
2.5
2
1.5
Per Unit
1
0.5
0
-0.5
parital discharge
V
dV/dt
cosh(V+dV/dt)
-1
-1.5
0
50
100
150
200
Phase/deg
250
300
350
Figure 4-34: Comparison of cosh(V+dV/dt) to partial discharge activity for a triangle

waveform.
160

This confirmed that the source of partial discharge activity was due to electrical treeing as
compared in Figure 4-35 and the cosh model can be applied to different waveforms.
Figure 4-35: Electrical treeing partial discharge activity due to a triangle wave illustrating the
magnitude of discharge was related to not only the instantaneous voltage [159].
In conclusion, modelling partial discharge activity due to electrical treeing with the function
dV
appears to be potentially fruitful. Further investigations and analysis are

cosh V +
dt
required to determine why the 7th harmonic in isolation (wave 11) produced results
significantly different from the other waveforms. If successful, this approach could yield
diagnostic interpretation of recorded partial discharge activity to provide a deterministic
measure of the distortion of the power frequency (circumstance monitoring) and detect
any change in the insulations ageing state (condition monitoring).
4.4.2. The statistical evaluation of the partial discharge patterns

The shape parameter of the Weibull function of partial discharge (charge) magnitude
can provide information on the origin of partial discharges whilst serving as a diagnostic
ageing data marker [7]. This correlation was discovered when only relatively limited
bandwidth measuring instruments were economically and technically viable. This was
based on knowledge that specific defects produce characteristic partial discharge charge
magnitudes [172-177]. In [172-175, 178, 179] the Weibull analysis has been applied to the
charge magnitude distribution yielding values as diagnostic sources of discharges and
values providing a measure of discharge activity. Hence the data was modelled using the
two parameter cumulative distribution Weibull function in equation (4-9).
F (q ) = 1 e
(4-9)
161

Where is scale parameter and is shape parameter, yielding the linear relationship:
log(q ) log = log ln (1 F (q ) )
(4-10)
Due to the practical limitation in minimum partial discharge detection the data is leftcensored. However, this typically has negligible impact as the partial discharge detection
system threshold is quite low (5.5 pC) relative to the average charge magnitudes acquired
during experimentation ( 400 pC) from the object. Hence a mean rank estimator defined
by the IEC 56 standard [180] in equation (4-11) was employed here and utilized in
previous studies [177, 181].
F ( i, n ) =
i 0.5
n + 0.25
(4-11)
Figure 4-36 provides an illustration of this statistical output. In accordance with artificial
intelligence techniques employed for partial discharge classification and recognition [29,
31, 32] charge polarities were processed separately.
From the literature, for discharge activity due to voids in epoxy, lies in the range 1.4 <
void < 2.3, but can be larger, especially in periods prior to breakdown [177]. However,
for discharge activity due to electrical treeing in XLPE and EVA (ethylene-vinylacetate
copolymer), lies in the range 0.7 < tree < 1.5 [176]. The assumption is made here that
these ranges were material independent and thus applicable to the epoxy resin
investigated in this research.
162
log[-[ln(1-F(Charge)]
10
10
10
10
Weibull Comparison of Negative Charges K115 [Plots 1-7]
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
177
198
131
103
103
93
90
-2
1.7
1.8
1.4
1.3
1.7
1.8
1.3
-4
10
10
10
log(Charge/pC)
10
10
10
10
Weibull Comparison of Positive Charges K115 [Plots 1-7]
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
40
123
113
86
95
133
106
-2
1.0
1.7
1.4
1.3
1.5
1.7
1.5
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
83
97
110
83
66
83
94
-2
1.3
1.2
1.3
1.6
1.3
1.7
1.5
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
139
147
121
142
139
137
89
-2
1.8
1.9
1.7
1.7
1.8
1.7
1.4
-4
10
10
10
log(Charge/pC)
Figure 4-36: Weibull plots of negative and positive discharges for sample K115.
163

Applying linear best fit plots for charge magnitude and values as a function of Ks and
THD yielded the plots in Figure 4-37. These mathematically averaged plots show
insufficient evidence of deterministic relationships prompting analysis of the spread of the
and values.
Linear best fit of vs Ks
Linear best fit of vs Ks
400
6
All
All
+
-
200
100
0
+
-
300
1.2
1.4
Waveshape (Ks )
1.6
Linear best fit of vs THD
400
1.2
1.4
Waveshape (Ks )
Linear best fit of vs THD
All
All
+
-
200
100
0
+
-
300
1.6
10
20
THD (%)
30
40
10
20
THD (%)
30
40
Figure 4-37: Linear best fit plots of charge magnitude and values showing no
dependence on THD and Ks.
Figure 4-38 provides box and whisker plots with extreme outliers for the calculated
values using the two parameter Weibull function for all 11 tests outlined in Table 4-10. The
horizontal lower and upper lines of the blue boxes represent the first and third quartiles
whilst the red line inside the box denotes the median value of the calculated values. The
whiskers extending from each box encompasses calculated values in the range of three
times the interquartile range, on either side of the first and third quartiles. Thus any points
outside this range are denoted by the red asterisk (*) and classified as extreme outliers.
The positive and negative half cycle discharge activity values of Figure 4-38 lie in the
range 0.7 < < 2.3, suggesting all discharge activity originated from both voids and
electrical trees. The partial discharge repetition rate can be used to further investigate this,
since partial discharges in electrical trees (particularly bush trees) have faster repetition
rates than in voids where charge resulting from discharges can be trapped on cavity walls
[34, 35]. Overall, the interquartile ranges of values did not illustrate any distinguishing
164

features between the composite waveforms and the fundamental waveform (wave 7).
However, wave 12 for the positive half cycle produced the lowest interquartile range with
two outlier points. At present it is unclear how this can interpreted but must be noted.
All Values
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56

THD 5.0 7.85 17.8 5.0
5.0 0.0 40.0
5.0 0.0 40.0
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56

5.0 0.0 40.0
13 12 11 9 8 7 1
Waveforms
Negative Charges values

THD 5.0 7.85 17.8 5.0
Charge Magnitude Values
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56

THD 5.0 7.85 17.8 5.0
Positive Charges values
13 12 11 9 8 7 1
Waveforms
13 12 11 9 8 7 1
Waveforms
combined charge polarities as well as positive and negative charge polarities.
The overall scatter of the values lay within the characteristic tree and void ranges. This
appeared consistent as the ranges of overlapped for electrical trees and voids [34, 35,
174, 177]. Hence this technique cannot uniquely identify a source in the presence of
overlapping discharge activity because the ranges of source identification also overlap.
Multiple sources of discharge activity e.g. air pockets and or trapped gas in the tree
channel as well as an electrical tree may exist. Hence the five parameter Weibull function
in equation (4-12) can be employed to model discharge activity from two dominant
sources.
q 1
q 2
F (q ) = p 1 e 1 + (1 p ) 1 e 2
(4-12)
165

A quality of fit test can determine which (two or five parameter Weibull) is better suited
[177]. This undoubtedly provides an avenue for further statistical analysis and future work.
As the electrical tree grew, was expected to increase as the discharge magnitude due to
electrical treeing increased [176]. Figure 4-39 illustrates the box and whisker plots with
extreme outliers for the 63.2 percentile of discharge activity, the scale factor . The
median values do not suggest any distinguishing features amongst the seven composite
waveforms tested. As the distortion (Ks and THD) increased, the upper quartiles increased
but non-linearly. The upper quartile of the undistorted waveform (wave 7) was the lowest
value (for both positive and negative discharges). Waves 11 and 12 were characterised by
a comparable value of Ks to wave 1, whose THD index doubled waves 11 and 12. Despite
this, the upper quartile for waves 12 and 11 were higher than the upper quartile of wave 1
(for both positive and negative half cycle discharges) in Figure 4-39.
All Values
400
Positive Charges values
400
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56
KS 1.27 1.60 1.60 1.06 1.03 1.00 1.56
THD 5.0 7.85 17.8 5.0
THD 5.0 7.85 17.8 5.0
5.0 0.0 40.0
5.0 0.0 40.0
350
350
300
300
300
250
200
150
350
250
200
150
200
150
100
100
50
50
50
13 12 11 9 8 7 1
Waveforms
13 12 11 9 8 7 1
Waveforms
5.0 0.0 40.0
250
100
Negative Charges values
THD 5.0 7.85 17.8 5.0
400
13 12 11 9 8 7 1
Waveforms
combined charge polarities as well as positive and negative charge polarities.
These have been quantified in Table 4-11 and can be interpreted as confirmation that
partial discharge activity and thus insulation ageing mechanism is influenced by the shape
of the composite waveform which cannot be effectively conveyed by Ks and THD indices.
166

Table 4-11: Upper quartile values for wave 7 is the lowest, while waves 12 and 11, exceed
wave 1 whose THD is significantly higher.
Upper Quartile Values pC

Wave 12
Wave 11
Wave 7
Wave 1
Ks = 1.60
Ks = 1.60
Ks = 1.00
Ks = 1.56
THD = 7.85 %
THD = 17.8 %
THD = 0 %
THD = 40 %
All
113
96
85
92
Positive
119
90
67
76
Negative
101
114
87
96
The positive discharge values 12 of Figure 4-39 revealed the highest interquartile range
coincided with the lowest interquartile range for the corresponding positive 12 previously
noted in Figure 4-38. High values of 12 and narrow ranges of 12 suggest consistently
shaped distributions of varying magnitude. Wave 12 had the narrowest range of and
contained the 5th, 7th, 11th, 13th, 23rd and 25th orders implying that the magnitudes of V and
dV
were adequate to reproduce distinguishable and values from patterns of positive
dt
discharges.
4.4.3. Visual correlation with partial discharge pattern and

statistical indices
This section combines the statistical analysis of partial discharge magnitude with the
captured electrical tree growth images to improve the interpretation of the ageing
mechanisms which impact on the partial discharge activity measurements. Figure 4-40
illustrates an example of the collection of electrical tree growth images. All captured
images (where possible) for the given tests outlined in Table 4-10 have been archived in
Appendix B. Figure 4-40 compliments the Weibull plots in Figure 4-36, providing a visual
indication of the electrical tree growth to associate with the characteristic and values.
There existed a range of 1.0 < < 1.8 (Figure 4-36) which reflect electrical tree growth
(Figure 4-40) for test K115 as these values lay within the characteristic tree and void
ranges. All the Weibull plots of Figure 4-36 were similar in shape, indicating a high degree
of reproducibility for partial discharge activity due to electrical treeing albeit a variation
existed in the power quality of the composite waveforms. The curvature of plots beyond
200 pC suggested another ageing mechanism was dominant and responsible for the
167

increased magnitude discharges, justifying further work using the five parameter Weibull
function model discharge activity from two dominant sources.[177].
Figure 4-40: Electrical tree growth images for K115. Each plot = 2 mins, 14.4 kV peak.
Two examples incorporating power quality variation have been identified for analysis. Test
K104 illustrated a growing electrical tree whilst tests K112 and K113 were utilized to show
the change in the insulation state prior to breakdown.
The first example compared Figure 4-41 with Figure 4-42 to simultaneously illustrate
changes in electrical tree growth with the corresponding Weibull plots of charge
magnitude. In plots 6-8 of Figure 4-42 the shape of the Weibull plots for the negative
charges varied significantly from plots 1-5 and plots 9-14. Plots 6-8 produced knee points
suggesting the presence of two ageing mechanisms and were also isolated from plots 1-5
and plots 9-14 which exhibited relatively increased linearity and were bunched together.
This was confirmed by the abrupt decrease in characteristic and values as shown in
Figure 4-43 during plots 6-8.
168
Figure 4-41: Electrical tree growth images for K104. Each plot = 4 mins, 10.8 kV peak.
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
10
-2
10
----2.4
2.8
2.5
2.7
1.1
1.5
-4
10
10
10
log(Charge/pC)
10
10
----113
127
140
124
87
61
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
10
-2
10
115
254
238
250
299
345
312
1.9
3.0
2.7
2.9
3.2
3.2
2.0
-4
10
10
10
log(Charge/pC)
10
Figure 4-42: Test K104 Weibull analysis of negative charges for plots 1-14 showing the
variation during plots 6, 7 and 8 as a result of the sudden tree growth, accompanied by
increased magnitude partial discharge activity.
169
400
3.5
350
300
250
2.5
2
200
1.5
150
100
50
0.5
1
Shape factor beta
Scale factor alpha/pC
Graph of alpha and beta variation for plots of K104
9 10 11 12 13 14
Plot
alpha
beta
Figure 4-43: Graph showing the sudden change between plots 6-8 for characteristic and
values of test K104, indicating a change in the insulation state.
Comparison of the respective images for plots 6-8 in Figure 4-41 show a sudden growth of
the electrical tree. This growth occurred during wave 7 (plot 6) and continued into the next
waveform, wave 1 (plot 7) before the sequence repeated again commencing with wave 13
(plot 8). It was evident that increased growth can be quantified by the variation in the
shape of these Weibull plots and thus the minima region identified in Figure 4-43. The
subsequent Weibull plots (plots 9 -14) were characterised by increased magnitude partial
discharges reflected by the increased values in Figure 4-43 as a consequence of
electrical tree growth. The subsequent values marginally increased but were
comparable to the values of plots 1-5 (prior to the sudden burst of electrical tree growth)
indicating one and perhaps the same ageing mechanism. The positive charge plots for
test K104 in Appendix B were clustered and did not provide distinct observations as the
negative charge plots shown here.
While this statistical, graphical technique usually revealed marginal scatter in Weibull plots
as a consequence of changes in electrical treeing partial discharge activity arising from
power quality variation, it facilitated identification of significant change in the state of the
electrical tree. However, determining whether the change in the dominant ageing
mechanism as shown in Figure 4-42 was triggered as a consequence of power quality
variation or material ageing or both proved inconclusive under these test conditions using
this analysis. This led to the following example which addressed changing partial
170

discharge activity under the influence of one composite waveform (wave 1) on a well
developed electrical tree, minimising state variability within the tree.
Plot 7 K112
Plot 14 K112
500
=150
=3.9
0
-500
Time =T1
0
100
200
Phase/deg
=41
=1.6
300
100
200
Phase/deg
100
200
Phase/deg
=48
=1.9
=41
=1.4
300
Charge/pC
Charge/pC
300
500
Time =T2
Time =T1+28mins
=37
=2.2
Plot 14 K113
Plot 7 K113
-500
-500
500
=65
=1.4
=76
=1.7
A
Charge/pC
Charge/pC
500
-500
Time =T2+7mins
0
100
200
Phase/deg
=69
=1.2
300
Figure 4-44: Progression of partial discharge pattern with and values for successive
tests K112 and K113 on sample T273. Plots show change in discharge patterns suggesting
change of dominant ageing mechanism and change in state of insulation.
Figure 4-45: Visual images showing electrical tree growth for tests K112 and K113.
Tests K112 and K113 were executed consecutively on sample T273. Figure 4-44
illustrates a progressive change in the discharge patterns. In test K112, moving from plot
A to plot B (28 minutes later) reveals discharge activity at 125 10 and 305 10 ,
which was not present before, indicating a definite change in the dominant ageing
mechanism. This was supported by the increased value of suggesting increased voidlike discharge activity. The images in Figure 4-45 illustrate an increase in the number of
branches from the thick return growth. Moving from plot B to plot C (Figure 4-44) was not
171

an instantaneous switch. A few minutes elapsed without application of electrical stress
followed by six minutes (one minute per composite waveform) energised at an increased
voltage level from 10.8 kV peak to 14.4 kV peak using the other composite waveforms.
This reduction in time per wave was employed to reduce any significant change in the
state of the insulation as a consequence of the increased electrical stress from the
composite waveforms. On application of wave 1 the partial discharge pattern captured
varied from the previous similar acquisition. Partial discharge activity at 125 10 and
305 10 was no longer present. The transition from plot C to plot D also produced
changes in the partial discharge activity and in the characteristic and values. In plot D
(relative to plot C), the discharge magnitude increased while there was less variation in
the range of the values implying the dominant ageing mechanism became stronger. The
images in Figure 4-45 captured the increased return growth of the electrical tree showing
its distance to the needle tip decreasing. There was also visual evidence of increased
branches to support the change in partial discharge activity captured in plot D. Breakdown
occurred momentarily after. These visual images illustrated the progressive growth of the
electrical tree and did not reflect growth as a consequence of wave 1 only.
Therefore evidence from tests K112 and K113 confirmed that any change in the dominant
ageing mechanism resulted in electrical tree growth producing different partial discharge
patterns even under the same applied waveforms. This finding confirms the difficulty of
modelling partial discharge patterns from an electrical tree where the state of the defect
(electrical tree) changes rapidly, shown in Figure 4-46.
Weibull plots for wave 1 of K112 & K113
10
K112 Positive Wave 1/Plot 7

K112 Negative Wave 1/Plot 7
10
10
-1
10
-2
10
-3
10
150
3.9
41
1.6
76
1.7
37
2.2
65
1.4
41
1.4
48
69
1.0
1.2
-4
10
10
10
log(Charge/pC)
10
Figure 4-46: Weibull plots of wave 1 discharge activity for K112 and K113 illustrating the
variability due to minute changes in electrical tree growth preventing consistent plots.
172

In Figure 4-46 only the negative charges of plot 7 illustrated consistency and this is
reflected in relatively similar partial discharge patterns of Figure 4-44 for plot 7. The
spread of Weibull plots acquired for wave 1 implied the overlap of other waveforms will
prevent distinguishing absolute effects due to the power quality variation.
Notably, the scatter of points for waveforms of similar nature and time-domain features
were quite close provided discharge polarities were discriminated. In many cases waves
13 and 12 illustrated extremely close (if not overlapping) points e.g. tests K101 and K106,
while waves 7, 8 and 9 e.g. tests K113 and K115, were also very close but never quite
close as waves 13 and 12. This is illustrated in Figure 4-47. Waves 13 and 12 contained
identical harmonic orders but of differing magnitude yielding similar partial discharge
patterns, implying that the resultant shape of each composite wave can impact on the
Weibull plots i.e. the shift in Weibull plots from the partial discharge activity of each
composite waveform varied as the power quality varied.
Weibull Comparison of Positive Discharges K113 [Plots 1-7]
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
-1
10
log(Charge/pC)
Figure 4-47: Example of spread of Weibull plots. Waves 13 and 12 have similar scatter while
the other waveforms clustered together. Relative to wave 7 (fundamental) each composite
waveform has a different scatter as a result of different partial discharge patterns.
Therefore visually, Weibull plots relate limited information relative to the phase-resolved
partial discharge patterns about variation in power quality. The investigation and analysis
conducted here reveals that caution must be exercised when visually interpreting Weibull
plots, because the influence of the composite waveform shape can be misinterpreted as a
change in the dominant ageing mechanism and vice versa.
173
4.4.4. Deductions
The following deductions summarize the work done in this section:
Wave 13 ( THD = 5.0 % and Ks = 1.27) was characterized by a derivative

magnitude of at least twice that of the undistorted wave 7 albeit both had the same
peak magnitude in Figure 4-25. Wave 13 produced marginally greater partial
discharge magnitudes and increased density of partial discharges relative to wave
7. During one 80 ms acquisition, the total number of partial discharges acquired
from wave 13 was 52 % greater (167 discharges whilst for wave 7, 110
discharges).
Modelling partial discharge activity due to electrical treeing with the function
dV
requires further investigation. It provides a good fit for identifying

cosh V +
dt
locations of peaks in discharge magnitude on the phase-resolved plots and can

also identify periods of low discharge activity.
Increased magnitude of the 7th harmonic on its own produced anomalous results
dV
and did not fit the function cosh V +

well.
dt
The statistical approach using partial discharge magnitudes confirms that the
values can be used as a diagnostic technique to estimate the source of partial
discharges independent of the power quality present. However, this technique
cannot uniquely identify a source in the presence of overlapping discharge activity
because the ranges of source identification overlap. The literature suggested
ranges of 1.4 < void < 2.3 [177] and 0.7 < tree < 1.5 [176], with increased
magnitudes prior to breakdown. The experimental results acquired in this research
lay in the region 0.6 < < 3.9.
As Ks and THD increased, the upper quartiles for increased non-linearly. The
lowest value of the upper quartile (of the scale parameter ) originated from the
undistorted waveform (wave 7) for both positive (66 pC) and negative (86 pC) half
cycle discharges relative to the maximum of positive (118 pC) and negative (110
pC) for all waveforms tested. For Ks 1.60, the upper quartile for wave 12 (THD =
7.85 %) and wave 11 (THD = 17.8 %) were higher than the upper quartile of wave
1 (THD = 40.0 %) for both positive and negative half cycle discharges in Table
4-11. This confirmed that partial discharge activity and thus insulation ageing
174
mechanisms are influenced by the shape (V and
dV
) of the composite waveform,
dt
which is ineffectively conveyed by the Ks and THD indices.
Samples with short electrical trees characterised by minimal tree spread were
more likely to produce Weibull plots with knee points indicative of multiple ageing
mechanisms, responsible for the variation in partial discharge activity as shown in
Figure 4-43. This is consistent with the change in electrical tree growth
characteristics. Conversely, samples with extensive electrical tree spread were
more likely to produce linear Weibull plots indicative of one dominant ageing
mechanism present. This is consistent with minimal growth of mature electrical
trees prior to the onset of breakdown, which have traversed the insulation gap
resulting in the thick return growth. This observation was independent of the
composite waveforms employed as a source of electrical stress.
The statistical approach eliminated the visual interpretation of partial discharge

patterns acquired from the composite waveforms. However, marginal shifts in
successive Weibull plots of the partial discharge activity were reported with varied
harmonic content of the composite waveforms. Therefore, the Weibull plots convey
very limited information about power quality compared to the phase-resolved
partial discharge patterns.
Intrinsic changes in the dominant ageing mechanism can occur within a short time
frame (~s), influencing the partial discharge activity recorded. If this change (in
partial discharge activity) is comparable to unmonitored change in power quality of
the excitation voltage, overestimation of the insulations ageing state will occur,
resulting in untimely asset management decisions being taken.
175
Chapter 5. Summaries and outcomes
5. Summaries and outcomes

5.1.
Asset management
Summary: Asset management strategies in the rail and aerospace industries are
substantially more developed than those employed in the power industry. An asset and
industry independent strategy developed to close this gap is PAS 55, providing a
framework to achieve best practice asset management. To improve the art of asset
management, there is the need for a complete structural blueprint linking the component
level to the system level allowing the asset manager to quantify assets more reliably and
save money on life cycle costs. Thus, a vertically integrated five layer framework in Figure
2-6 linking the knowledge of insulation ageing to asset management has been developed.
This framework facilitates stochastic modelling of the working environment, including
stress factors such as electrical, physical, mechanical, environmental, chemical and
thermal. More importantly, the framework establishes links between the physics and
chemistry of insulation degradation, linking plant reliability to asset management. Such a
tool can complement and improve asset management strategies providing a platform for
future dielectric research. Consequently this framework provides the wherewithal for the
production of life forecasting models, usually simplified, but with defined uncertainties.
Given that this framework can potentially statistically model insulation life for a single plant
component, ultimately it can produce an integrated model for system reliability, enabling
holistic network asset management. This may then allow an asset manager to act
accordingly, to mitigate the impact of changing stress or manage the plant differently.
Outcome: A novel framework linking the knowledge of insulation ageing to asset
management has been developed.
176
5.2.
Test facility
Summary: The in-house test facility shown in Figure 3-10 was successfully designed,
built and operated. It facilitated electrical treeing investigations using composite

waveforms, uniquely employing six harmonic components. This test facility provides a
platform for future dielectric research. It featured:
100 MSps arbitrary waveform generator capable of producing programmable nonpower frequency waveforms whilst providing isolation from external unknown
disturbances.
30 kV, 20 mA high voltage amplifier capable of driving waveforms with slew rates
> 350 V/s at full load.
Automated image capture using an IEEE 1394 colour camera with a maximum
resolution of 1392 x 1040 pixels and a maximum capture rate of 13 frames per
second. The camera was equipped with a telecentric C-mount lens to provide
sufficient optical magnification of the electrical trees in the prepared samples.
Real time partial discharge capture and processing with a minimum detection level
of 5.5 pC.
Automated electroluminescence data capture using a USB spectrometer covering

the range of 200 nm to 850 nm thus encompassing the ultraviolet and visible
spectrums.
The test facility adhered to all safety policies ensuring correct and safe equipment
operation. This mobile test facility was fully controlled from the locally housed
personal computer using the developed software suite shown in Figure 3-37.
Outcome: A test facility was designed and built facilitating electrical treeing investigations
using composite waveforms uniquely employing six harmonic components.
177
5.3.
Lubricant coating
Summary: The investigation of the lubricant coating on the hypodermic needles revealed
the standard stripping process of the lubricant coating was non-uniform influencing the
experimental results. This raised concerns about the sample production technique of
previous studies and resulted in adopted procedures to overcome this problem. The study
of electrical tree growth here led to the use of the width/length ratio. Table 4-4
summarized the influence of the lubricant coating on the hypodermic needle on the
electrical treeing process which included:
The acquisition of the rabbit-ear partial discharge pattern.
The inhibition of electrical tree growth spread resulting in a reduced width/length

ratio.
Preliminary investigations suggest that the lubricant coating present decreased the
electrical tree initiation and breakdown times. Further experiments are required for
confirmation. The observation of mobile bubbles on the needle surface was seen only
when the lubricant coating was present. This gaseous activity occurred simultaneously
with the rabbit-ear partial discharge patterns during early growth of the electrical tree.
These observations suggested this intermediate layer of lubricant coating between the
epoxy resin and the material prevented the epoxy resin from effectively bonding to the
needle. This created a weak layer that allowed trapped gas and displaced gas due to the
electrical tree growth to escape. Escaped gas caused reduction of pressure at the needle
tip and this may be the cause of different growth characteristics with and without the
lubricant. However, further investigations are required into the influence of the lubricant
coating on the gaseous activity.
Outcome: The lubricating coating on hypodermic needles contributes to the formation of
the rabbit-ear partial discharge pattern and inhibit initial electrical tree growth spread,
characterized by a reduce width/length ratio.
178
5.4.
Direction of electrical tree growth
Summary: An electrical tree can grow in two directions. Firstly, the original (downward)
growth of the electrical tree from the needle tip to the ground electrode, where the
insulation gap is traversed. Secondly, the return growth from the ground electrode towards
the needle tip of the electrical tree. This return growth manifested as thick channels
growing through existing widespread fine tubules from the ground electrode. During this
phase the tree does not necessarily increase in spread but thicker channels developed,
increasing the visibility of the entire tree structure. However from the onset of the return
growth the partial discharge repetition rate and magnitude increase significantly relative to
the downward growth (before the tree traversed the insulation gap). Since breakdown did
not occur when the downward growth traversed the insulation gap, this provided the basis
to infer that the downward growth is non-conducting while return growth is conducting.
Outcome: The return electrical tree growth from the ground electrode towards the needle
tip, growing through existing tubules was captured.
5.5.
Power quality and electrical tree growth
Summary: No change in the growth characteristics of an electrical tree in epoxy resin due
to harmonic content variation has been detected. The rate of the original (downward)
growth appeared to be independent of the composite waveform applied at constant peak
voltage. Figure 4-12 to Figure 4-17 did not provide confirmation that acceleration of field
assisted ageing processes occur due to composite waveforms. Electrical tree growth
measurements may not be the best measurand to support this claim. One alternative
measurand which can be studied in future work is the capture of electroluminescence
data. While 14.4 kV peak was consistent with previous studies conducted [42, 153, 161] it
maybe too high for the (waveshape factor) KS to have a dominating influence on the tree
growth characteristics. Future work should employ lower voltages for increased durations
of electrical tree growth subjected to composite waveforms.
Outcome: No changes were detected in electrical tree growth characteristics in epoxy
resin due to variation in power quality with constant peak voltage of 14.4 kV.
179
5.6.
Power quality and breakdown times
Summary: As THD and KS increased, breakdown times were increasingly scattered with
reduced mean values, suggesting increased vulnerability to reduced insulation life as

harmonic content in the working environment also increased. The Weibull statistical
analysis of breakdown times (at approximately constant Ks) revealed increased THD
resulted in reduced shape parameters , for less than 10 % variation in scale parameters
. Such a trend was not confirmed for constant THD and increasing KS as observed in
Table 4-9 and reproduced below in Table 5-1.
Table 5-1: Variation in breakdown time and values to corresponding KS and THD (same
as Table 5-9).
Ks
THD (%) Wave
1.60
7.85
12
8955
3.4
1.60
17.8
11
8643
2.5
1.56
40.0
9494
2.3
1.03
5.00
9770
6.9
1.06
5.00
8111
3.6
1.27
5.00
13
7565
5.5
A correlation of the 7th harmonic on its own to decreased failure times resulting in low
Weibull shape parameters describing the increased scatter of these times has been
revealed at constant peak voltage of 14.4 kV.
Outcome: At a constant peak voltage of 14.4 kV, increased magnitude of the 7th harmonic
resulted in reduced failure times and low values of Weibull shape parameters describing
the increased scatter of these times.
180
5.7.
Power quality and partial discharge patterns
Summary: A composite waveform (wave 13) defined by THD = 5.0 % and KS = 1.27 was
characterized by a time derivative magnitude of at least twice that of the undistorted

fundamental (wave 7) albeit both waveforms had the same peak magnitude as seen in
Figure 4-25. Wave 13 produced marginally greater partial discharge magnitudes and
increased density of partial discharges relative to that of wave 7. During one 80 ms
acquisition, the total number of partial discharges acquired from wave 13 was 52 %
greater than wave 7. Intrinsic changes in the dominant ageing mechanism of electrical
treeing influence the partial discharge pattern recorded. A relatively small change in THD
can result in significant changes to measured partial discharge patterns. If such
measurements are incorrectly interpreted as changes to the state of the insulation rather
than the stressing voltage, overestimation of the insulations aged state will result in
untimely asset management decisions. Modelling partial discharge activity due to
dV
provided a good fit for identifying partial

electrical treeing with the function cosh V +
dt
discharge peaks on the phase-resolved plots and can also identify periods of low
discharge activity. However, waveforms with increased magnitude of the 7th harmonic on
its own (wave 11) produced anomalous results and did not fit the function well, as seen in
Figure 4-31.
The Weibull statistical analysis of partial discharge magnitude confirmed that the shape
parameter, can be used as a diagnostic technique to estimate the source of partial
discharges independent of the power quality present. However, this technique cannot
uniquely identify a source in the presence of overlapping discharge activity because the
ranges of source overlap. The literature suggested ranges of 1.4 < void < 2.3 [177] and
0.7 < tree < 1.5 [176] with increased magnitudes prior to breakdown. The experimental
results acquired in this research lay in the region 0.6 < < 3.9. As KS and THD increased
the upper quartiles of the scale parameter , increased non-linearly for both positive and
negative half cycle discharges in Table 4-11. The lowest value of the upper quartile (of the
scale parameter ) originated from the undistorted waveform (wave 7) confirming that
partial discharge activity was influenced by the shape (functions of peak V and derivative
dV
) of the composite waveforms which is ineffectively conveyed by the KS and THD
dt
indices.
181

While the statistical approach eliminated the visual interpretation of partial discharge
patterns, shifts in the Weibull plots of partial discharge magnitude varied with the
harmonic content of the composite waveforms. Therefore the actual Weibull plots (not
and parameters) conveyed very limited information compared to the phase-resolved
partial discharge patterns about power quality. Electrical trees with minimal spread
possessed increased likelihood of producing Weibull plots with knee points indicative of
multiple ageing mechanisms, consistent with electrical tree growth. Conversely, electrical
trees with extensive spread possessed higher inclination to produce linear Weibull plots
indicative of one dominant ageing mechanism present, consistent with minimal growth of
mature electrical trees prior to the onset of breakdown. This observation was independent
of the composite waveforms employed as a source of electrical stress.
Outcome: Modelling partial discharge activity due to electrical treeing with the function
dV
provided a good fit for identifying locations of partial discharge peaks on

cosh V +
dt
the phase-resolved partial discharge plots and also identified periods of low discharge
activity.
Power quality can influence diagnostic partial discharge patterns and if changes in power
quality are unmonitored, this can produce an overestimation of the insulations ageing
state resulting in untimely asset management decisions taken.
182
Chapter 6. Major contributions
6. Major contributions
6.1.
Achievements
This research project yielded the following achievements:
A novel framework linking the knowledge of insulation ageing to asset

management has been developed.
A test facility was designed and built facilitating electrical treeing investigations
using composite waveforms uniquely employing six harmonic components.
The return electrical tree growth from the ground electrode towards the needle tip,
growing through existing tubules was captured.
6.2.
Conclusions
The following conclusions were drawn from the research:
The lubricating coating on hypodermic needles contributes to the formation of the

rabbit-ear partial discharge pattern and inhibit initial electrical tree growth spread,
characterized by a reduce width/length ratio.
No changes were detected in electrical tree growth characteristics in epoxy resin

due to variation in power quality with constant peak voltage of 14.4 kV.
At a constant peak voltage of 14.4 kV, increased magnitude of the 7th harmonic
resulted in reduced failure times and low values of Weibull shape parameters
describing the increased scatter of these times.
Modelling partial discharge activity due to electrical treeing with the function
dV
provided a good fit for identifying locations of partial discharge

cosh V +
dt
peaks on the phase-resolved partial discharge plots and also identified periods of
low discharge activity.
Power quality can influence diagnostic partial discharge patterns and if changes in
power quality are unmonitored, this can produce an overestimation of the
insulations ageing state resulting in untimely asset management decisions taken.
183
Chapter 6. Major contributions
6.3.
Further work
The following are areas to pursue for further work:
Integration of relevant data and insulation specific life models of different plant
items into the framework to develop probabilistic models of insulation life
identifying further areas of research.
Integration of defined asset management documentation with a companys

strategic business plan to yield a completely tailored and optimized company
specific framework.
Review the sample preparation process to facilitate use of the spectrometer during
testing since the current acrylic cubes employed blocked the transmission of the
wavelengths in the ultraviolet region. This would include either the use of a
different type of acrylic cube or careful removal of the acrylic without disrupting or
damaging the cured sample.
Improvements in the optics of the image capture system and providing control
outside the high voltage testing area.
Experiments to determine the influence of the lubricant coating on the gaseous

activity reported as well as investigate tree initiation and breakdown times.
The peak voltage employed here may have been too high for the waveshape
factor KS to have a dominating influence on the electrical tree growth
characteristics. Future work should employ lower voltages for increased durations,
to investigate tree growth using the composite waveforms defined in this research.
This work and analysis should utilize captured electroluminescence data.
Experiments with normalized values for a linear comparison of THD and KS, to
compare and contrast the degree of distortion each index produces and ultimately
acquire a weighting factor representation.
Further investigations to confirm the 7th harmonic is more influential than the 5th
harmonic for breakdown due to electrical treeing at constant peak voltage.
Further investigations of the 7th harmonic on electrical treeing to determine the

thresholds which influence the correlation to failure times and modelling partial
dV
discharge activity using the function cosh V +

dt
Apply goodness of fit tests and the five parameter Weibull function to statistically
reanalyze the partial discharge data.
184
References
References
[1]
J. V. Champion and S. J. Dodd, "The effect of voltage and material age on the electrical
tree growth and breakdown characteristics of epoxy resins," Journal of Physics D: Applied
Physics, vol. 28, pp. 398-407, 1995.
[2]
C. Bonten and R. Berlich, Aging and Chemical Resistance: Hanser Gardner 2002.
[3]
L. A. Dissado and J. C. Fothergill, Electrical Degradation and Breakdown in Polymers.

London, U.K.: P. Peregrinus on behalf of the Institution of Electrical Engineers, 1992.
[4]
J. P. Bellomo, T. Lebey, J. M. Oraison, and F. Peltier, "Study of the ageing of different

dielectrics under specific electrical constraints," in IEEE International Conference on
Properties and Applications of Dielectric Materials, vol. 2. Brisbane, Australia, 1994, pp.
796-799.
[5]
BSI, "60505 Evaluation and qualification of electrical insulation systems ", 2004.
[6]
T. Ishida and M. Nagao, "Time variation of partial discharge current waveforms with
degradation of insulators," in International Symposium on Electrical Insulating Materials,
vol. 2. Kitakyushu, Japan, 2005, pp. 494-497.
[7]
G. C. Montanari, "Insulation diagnosis of high voltage apparatus by partial discharge

investigation," in IEEE International Conference on Properties and Applications of
Dielectric Materials. Bali, Indonesia, 2006, pp. 1-11.
[8]
J. Densley, "Ageing and diagnostics in extruded insulations for power cables," in IEEE
International Conference on Conduction and Breakdown in Solid Dielectrics. Leicester, UK,
1995, pp. 1-15.
[9]
T. Hibma and H. R. Zeller, "Direct measurement of space-charge injection from a needle

electrode into a dielectric," Journal of Applied Physics, vol. 59, pp. 1614-1620, 1986.
[10]
T. Lebey and C. Laurent, "Charge injection and electroluminescence as a prelude to

dielectric breakdown," Journal of Applied Physics, vol. 68, pp. 275-282, 1990.
[11]
S. S. Bamji, A. T. Bulinski, and R. J. Densley, "Threshold voltage of luminescence and

electrical tree inception in low-density polyethylene," Journal of Applied Physics, vol. 63,
pp. 5841-5845, 1988.
[12]
G. Mazzanti and G. C. Montanari, "Electrical aging and life models: the role of space
charge," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, pp. 876- 890,
2005.
[13]
N. Shimizu and C. Laurent, "Electrical tree initiation," IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 5, pp. 651-659, 1998.
[14]
T. H. Baumann, T. Hibma, P. Pfluger, and H. R. Zeller, "Experimental evidence for charge

injection. A pre-breakdown mechanisms in polymers," in IEEE International Conference on
Conduction and Solid Dielectrics. Erlangen, Germany, 1986, pp. 131-136.
[15]
J. Andrianjohaninarivo, M. R. Wertheimer, and A. Yelon, "Nucleation of electrical trees in

polyethylene," IEEE Transactions on Electrical Insulation, vol. 22, pp. 709-714, 1987.
[16]
T. Mito, M. Watanbe, Y. Muramoto, and N. Shimizu, "Electroluminescence properties under

long time voltage application in XLPE," in IEEE Conference on Electrical Insulation and
Dielectric Phenomena. Kansas City, USA, 2006, pp. 716-719.
[17]
S. S. Bamji, M. Kaufhold, and A. T. Bulinski, "Electroluminescence technique to evaluate

the effect of impulse tests on high voltage cables," IEEE Transactions on Dielectrics and
[18]
S. S. Bamji, A. T. Bulinski, and R. J. Densley, "Evidence of near-ultraviolet emission during

electrical-tree initiation in polyethylene," Journal of Applied Physics, vol. 61, pp. 694-699,
1987.
[19]
G. C. Stone, R. G. Van Heeswijk, and R. Bartnikas, "Investigation of the effect of repetitive

voltage surges on epoxy insulation," IEEE Transactions on Energy Conversion, vol. 4, pp.
754-760, 1992.
[20]
L. A. Dissado, "Understanding electrical trees in solids from experiment to theory," IEEE

Transactions on Dielectrics and Electrical Insulation, vol. 9, pp. 483-497, 2002.
185
References
[21]
J. M. Alison, J. V. Champion, S. J. Dodd, and G. S. Stevens, "Dynamic bipolar charge

recombination model for electroluminescence in polymer base insulation during electrical
tree initiation," Journal of Physics D: Applied Physics, vol. 28, pp. 1693-1701, 1995.
[22]
G. C. Stone, R. G. Van Heeswijk, and R. Bartnikas, "Electroluminescence in epoxy

insulation," IEEE Transactions on Electrical Insulation, vol. 27, pp. 221-232, 1992.
[23]
R. Cosson, C. Paracchini, and G. Schianchi, "Electroluminescence in an epoxy resin,"

Journal of the Electrochemical Society, vol. 125, pp. 581-583, 1978.
[24]
K. Tohyama, S. S. Bamji, and A. T. Bulinski, "Spectra of electroluminescence in XLPE due

to impulse voltage," in IEEE Conference on Electrical Insulation and Dielectric Phenomena.
Atlanta, USA, 1998, pp. 601-604 vol. 2.
[25]
BSI, "IEC 60270: High voltage test techniques partial discharge measurements," 2001.
[26]
J. Densley, "Ageing mechanisms and diagnostics for power cables - an overview," IEEE
Electrical Insulation Magazine, vol. 17, pp. 14-22, 2001.
[27]
R. Bartnikas, "Partial discharges. Their mechanism, detection and measurement," IEEE

[28]
I. J. Kemp, "Partial discharge plant-monitoring technology: present and future

developments," IEE Proceedings - Science, Measurement and Technology, vol. 142, pp. 410, 1995.
[29]
E. Gulski and F. H. Kreuger, "Determination of discharge sources by analysis of discharge

quantities as a function of time," in IEEE International Symposium on Electrical Insulation.
Baltimore, USA, 1992, pp. 397-400.
[30]
E. Gulski and A. Krivda, "Neural networks as a tool for recognition of partial discharges,"
IEEE Transactions on Electrical Insulation, vol. 28, pp. 984-1001, 1993.
[31]
A. Krivda, "Recognition of discharge patterns during ageing," in IEEE Conference on

Electrical Insulation and Dielectric Phenomena. Virginia Beach, USA, 1995, pp. 339-342.
[32]
F. H. Kreuger, E. Gulski, and A. Krivda, "Classification of partial discharges," IEEE

Transactions on Electrical Insulation, vol. 28, pp. 917-931, 1993.
[33]
A. Krivda, "Automated recognition of partial discharges," IEEE Transactions on Dielectrics

and Electrical Insulation, vol. 2, pp. 796-821, 1995.
[34]
A. Cavallini, M. Conti, G. C. Montanari, and A. Contin, "Transition from PD to electrical tree

in solid insulation inferred through enhanced detection tools," in IEEE International
Conference on Solid Dielectrics. Toulouse, France, 2004, pp. 632-635.
[35]
A. Cavallini, M. Conti, G. C. Montanari, C. Arlotti, and A. Contin, "PD inference for the early
detection of electrical treeing in insulation systems," IEEE Transactions on Dielectrics and
[36]
S. M. Strachan, S. Rudd, S. D. J. McArthur, M. D. Judd, S. Meijer, and E. Gulski,

"Knowledge-based diagnosis of partial discharges in power transformers," IEEE
[37]
S. M. Strachan, S. D. J. McArthur, M. D. Judd, and J. R. A. McDonald, "Incremental

knowledge-based partial discharge diagnosis in oil-filled power transformers," in
International Conference on Intelligent Systems Application to Power Systems, 2005, pp.
181-186.
[38]
F. Guastavino and B. Cerutti, "Tree growth monitoring by means of digital partial discharge
measurements," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 10, pp. 6572, 2003.
[39]
S. A. Boggs, "Mechanisms for degradation of TR-XLPE impulse strength during service

aging," IEEE Transactions on Power Delivery, vol. 17, pp. 308-312, 2002.
[40]
J. C. Fothergill, A. Eccles, A. Houlgreave, and L. A. Dissado, "Water tree inception and its
dependence upon electric field, voltage and frequency," IEE Proceedings - Science,
Measurement and Technology, vol. 140, pp. 397-403, 1993.
[41]
S. Boggs, J. Densley, and J. Kuang, "Surge-induced temperature rise in water-containing

defects in XLPE: mechanism for conversion of water trees to electrical trees," in IEEE
186
References
Conference on Electrical Insulation and Dielectric Phenomena 1996. San Francisco, USA,
1996, pp. 311-314.
[42]
H.-Z. Ding and B. R. Varlow, "Electrical treeing studies on the Araldite LY/HY 5052 epoxy
resin over a wide range of stressing voltage," in IEEE Conference on Electrical Insulation
and Dielectric Phenomena. Boulder, USA, 2004, pp. 306-309.
[43]
X. Zheng, G. Chen, A. E. Davies, S. J. Sutton, and S. G. Swingler, "The influence of

survival mechanical stress and voltage frequency on electrical tree in XLPE," in IEEE
Conference on Electrical Insulation and Dielectric Phenomena. Cancun, Mexico, 2002, pp.
955-958.
[44]
D. W. Auckland, J. M. Cooper, and B. R. Varlow, "Factors affecting electrical tree testing,"

IEE Proceedings - Science, Measurement and Technology, vol. 139, pp. 9-13, 1992.
[45]
G. Jiang and S. Boggs, "Mechanism for transition from branch to bush electrical tree," in
IEEE Conference on Electrical Insulation and Dielectric Phenomena. Virginia Beach, USA,
1995, pp. 57-60.
[46]
F. Noto and N. Yoshimura, "Voltage and frequency dependence of tree growth in

polyethylene," in IEEE Conference on Electrical Insulation and Dielectric Phenomena
1975. . Gaithersburg, USA, 1975, pp. 207-17.
[47]
T. Tanaka, "Tree initiation mechanisms," in IEEE International Conference on Properties

and Applications of Dielectric Materials. Tokyo, Japan, 1991, pp. 18-24.
[48]
A. Kusko and M. T. Thompson, Power quality in electrical systems. New York: McGrawHill, 2007.
[49]
IEEE, "Std 519-1992, IEEE recommended practices and requirements for harmonic control
in electrical power systems," 1992.
[50]
IEEE, "Std 1159-1995, IEEE recommended practice for monitoring electric power quality,"
1995.
[51]
A. Cavallini, D. Fabiani, G. Mazzanti, and G. C. Montanari, "Models for degradation of selfhealing capacitors operating under voltage distortion and temperature," in International
Conference on Properties and Applications of Dielectric Materials Xi'an, China, 2000, pp.
108-111.
[52]
B. W. Kennedy, Power quality primer. New York: McGraw Hill, 2000.
[53]
S. Bahadoorsingh and S. M. Rowland, "Development of a multifactor framework linking

insulation ageing and power network environments," in IEEE Conference on Electrical
Insulation and Dielectric Phenomena. Vancouver, Canada, 2007, pp. 388-391.
[54]
P. Caramia, G. Carpinelli, P. Verde, G. Mazzanti, A. Cavallini, and G. C. Montanari, "An

approach to life estimation of electrical plant components in the presence of harmonic
distortion," in International Conference on Harmonics and Quality of Power. Orlando, USA,
2000, pp. 887-892.
[55]
W. E. Reid, "Power quality issues-standards and guidelines," IEEE Transactions on

Industry Applications, vol. 32, pp. 625-632, 1996.
[56]
UL1562, "Standard for Transformer, Distribution, Dry-Type, Over 600 Volts," 1992.
[57]
G. C. Stone, R. G. Van Heeswijk, and R. Bartnikas, "Electrical aging and

electroluminescence in epoxy under repetitive voltage surges," IEEE Transactions on
[58]
J. Arrillaga, N. R. Watson, and S. Chen, Power system quality assessment: John Wiley &
Sons, 2000.
[59]
G. Mazzanti, G. Passarelli, A. Russo, and P. Verde, "The effects of voltage waveform

factors on cable life estimation using measured distorted voltages," in IEEE Power
Engineering Society General Meeting. Quebec, Canada, 2006, pp. 1-8.
[60]
J. V. Milanovic and M. Negnevitsky, "Power quality problems and solutions: current

understanding," in International Conference on Harmonics and Quality of Power, vol. 1,
1998, pp. 30-35.
[61]
M. Negnevitsky, J. V. Milanovic, T. Nguyen, and P. Rayner, "Power quality collaborative

project in Tasmania," in Power Quality, 1998, pp. 1-6.
187
References
[62]
D. Fabiani and G. C. Montanari, "The effect of voltage distortion on ageing acceleration of

insulation systems under partial discharge activity," IEEE Electrical Insulation Magazine,
vol. 17, pp. 24 - 33 2001.
[63]
G. C. Montanari and D. Fabiani, "Searching for the factors which affect self-healing
capacitor degradation under non-sinusoidal voltage," IEEE Transactions on Dielectrics and
[64]
G. C. Montanari and D. Fabiani, "The effect of non-sinusoidal voltage on intrinsic aging of

cable and capacitor insulating materials," IEEE Transactions on Dielectrics and Electrical
Insulation, vol. 6, pp. 798 - 802, 1999.
[65]
S. Bahadoorsingh and S. M. Rowland, "The role of power quality in electrical treeing of

epoxy resin," in IEEE Conference on Electrical Insulation and Dielectric Phenomena.
Vancouver, Canada, 2007, pp. 221-224.
[66]
R. Bozzo, C. Gemme, F. Guastavino, and G. C. Montanari, "Investigation of aging rate in

polymer films subjected to surface discharges under distorted voltage," in IEEE
Conference on Electrical Insulation and Dielectric Phenomena Minneapolis, USA, 1997,
pp. 435-438.
[67]
F. Guastavino, L. Centurioni, G. Coletti, A. Dardano, and E. Torello, "An experimental

study about the treeing phenomena in XLPE subjected to distorted voltages," in IEEE
Conference on Electrical Insulation and Dielectric Phenomena. Albuquerque, USA, 2003,
pp. 600-603.
[68]
A. Cavallini, D. Fabiani, G. Mazzanti, G. C. Montanari, and A. Contin, "Voltage endurance

of electrical components supplied by distorted voltage waveforms," in IEEE International
Symposium on Electrical Insulation. Anaheim, California, 2000, pp. 73-76.
[69]
A. Cavallini, G. Mazzanti, G. C. Montanari, and D. Fabiani, "The effect of power system

harmonics on cable endurance and reliability," in IEEE IAS. Rome, Italy, 2000, pp. 240247.
[70]
G. C. Montanari and L. Simoni, "Aging phenomenology and modeling," IEEE Transactions

on Dielectrics and Electrical Insulation, vol. 28, pp. 755-776, 1993.
[71]
G. C. Montanari, G. Mazzanti, and L. Simoni, "Progress in electrothermal life modeling of

electrical insulation during the last decades," IEEE Transactions on Dielectrics and
[72]
B. Florkowska, M. Florkowski, and P. Zydron, "The role of harmonic components on partial

discharge mechanism and degradation processes in epoxy resin insulation," in IEEE
International Conference on Solid Dielectrics Winchester, UK, 2007, pp. 560-563.
[73]
M. Florkowski and B. Florkowska, "Distortion of partial-discharge images caused by highvoltage harmonics," IEE Proceedings - Generation, Transmission and Distribution, vol.
153, pp. 171-180, 2006.
[74]
T. Kurihara, S. Tsuru, K. Imasaka, J. Suehiro, and M. Hara, "PD characteristics in an airfilled void at room temperature under superimposed sinusoidal voltages," IEEE
[75]
A. Cavallini, M. Conti, A. Contin, and G. C. Montanari, "Advanced PD inference in on-field

measurements. II. Identification of defects in solid insulation systems," IEEE Transactions
on Dielectrics and Electrical Insulation, vol. 10, pp. 528-538, 2003.
[76]
R. Bodega, A. Cavallini, P. H. F. Morshuis, and F. J. Wester, "The effect of voltage

frequency on partial discharge activity," in IEEE International Symposium on Electrical
Insulation., 2002, pp. 685-689.
[77]
A. Cavallini and G. C. Montanari, "Effect of supply voltage frequency on testing of

insulation system," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 13, pp.
111-121, 2006.
[78]
W. Yin, "Failure mechanism of winding insulations in inverter-fed motors," IEEE Electrical

Insulation Magazine, vol. 19, pp. 18-23, 1997.
[79]
H. Suzuki, Y. Ohki, Y. Nakamichi, and K. Ajiki, "Water tree characteristics in low-density

polyethylene under power-frequency voltages with high-frequency components," IEEE
188
References
Conference on Electrical Insulation and Dielectric Phenomena, vol. San Francisco, USA,
pp. 742-745, 1996.
[80]
R. D. Sarathi, S.; Venkataseshaiah, C.; Yoshimura, N.;, "Investigations of growth of

electrical trees in XLPE cable insulation under different voltage profiles," Conference on
Electrical Insulation and Dielectric Phenomena, 2003. Annual Report. , pp. 666-669, 2003.
[81]
B. R. Varlow and D. W. Auckland, "Mechanical aspects of electrical treeing in solid

insulation," Electrical Insulation Magazine, IEEE, vol. 12, pp. 21-26, 1996.
[82]
J. P. Bellomo, P. Castelan, and T. Lebey, "The effect of pulsed voltages on dielectric

material properties," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 6, pp.
20-26, 1999.
[83]
BSI, "BS EN 50160:2000 Voltage characteristics of electricity supplied by public distribution

systems ", 2000.
[84]
F. Guastavino, L. Centurioni, E. Torello, and A. Zaopo, "A new test to characterise low
voltage cables subjected to thermal and mechanical stresses," in IEEE Conference on
Electrical Insulation and Dielectric Phenomena. Cancun, Mexico, 2002, pp. 873-877.
[85]
W. A. Robert, M. James, K. Arun, and E. H. James, "Thermally induced failure of lowvoltage electrical nonmetallic-sheathed cable insulation," Fire Technology, vol. 35, pp. 263275, 1999.
[86]
D. Birtwhistle, M. Celina, G. George, and K. Millers, "Investigations of degradation of XLPE

insulation in low voltage aerial bundle cable systems," in International Conference on
Dielectric Materials, Measurements and Applications. Manchester, United Kingdom, 1992,
pp. 322-325.
[87]
R. Arora and V. Tripathi, "Effect of ultra violet irradiation on low voltage cable insulation," in
IEEE International Symposium on Electrical Insulation. Indianapolis, USA, 2004, pp. 358361.
[88]
E. A. Reeves, Handbook of electrical installation practice: Blackwell Scientific, 1990.
[89]
S. Cherukupalli, V. Buchholz, M. Colwell, J. P. Crine, and R. J. Keefe, "Condition

assessment of distribution PILC cables from electrical, chemical, and dielectric
measurements," IEEE Electrical Insulation Magazine, vol. 20, pp. 6-12, 2004.
[90]
C. Mirra, A. Porrino, A. Ardito, and C. A. Nucci, "Lightning overvoltages in low voltage

networks," in International Conference and Exhibition on Electricity Distribution. Part 1:
Contributions, vol. 1, 1997, pp. 19/1-19/6 vol.2.
[91]
J. P. Steiner and F. D. Martzloff, "Partial discharges in low-voltage cables," in IEEE

International Symposium on Electrical Insulation. Toronto, Canada, 1990, pp. 149 - 152
[92]
T. W. Dakin, "Electrical insulation deterioration treated as a chemical rate phenomenon,"

AIEE Transactions, vol. 67, pp. 113-122, 1948.
[93]
E. L. Brancato, "A pathway to multifactor aging," IEEE Transactions on Electrical

Insulation, vol. 28, pp. 820-825, 1993.
[94]
A. Porrino, I. Alexandri, M. Clement, G. S. Finlay, T. Henriksen, J. Hoeffelman, M. Ishii, P.

Martinez Cid, C. A. Nucci, F. Popolansky, F. Rachidi, G. P. T. Roelofs, and R. Rudolph,
"Protection of MV and LV networks against lightning. Part I: Basic information," in CIRED
International Conference and Exhibition on Electricity Distribution., vol. 2. Birmingham, UK,
1997, pp. 211-217.
[95]
P. P. Barker, T. A. Short, A. R. Eybert-Berard, and J. P. Berlandis, "Induced voltage

measurements on an experimental distribution line during nearby rocket triggered lightning
flashes," Power Delivery, IEEE Transactions on, vol. 11, pp. 980-995, 1996.
[96]
W. Pfeiffer, "Dielectric testing of solid insulation with respect to insulation coordination of

low-voltage equipment," Electrical Insulation Magazine, IEEE, vol. 17, pp. 34-47, 2001.
[97]
C. M. Walton, "Incipient fault detection and management of underground LV networks," in

CIRED International Conference and Exhibition on Electricity Distribution. Amsterdam,
Netherlands, 2001, pp. 179.
[98]
R. E. Brown and B. G. Humphrey, "Asset management for transmission and distribution,"

Power and Energy Magazine, IEEE, vol. 3, pp. 39-45, 2005.
189
References
[99]
IAM, "Institute of Asset Management."
[100]
J. Woodhouse, "Asset Management," in Asset Management Processes and Tools: The

Woodhouse Partnership Limited, 2001, pp. 1-8.
[101]
P. Jay, "Overview of PAS 55 - the business benefits," in The IEE Seminar on Developing &
Applying Best Practice Asset Management. The IEE, Savoy Place, London, 2005.
[102]
R. Overman and R. Collard, "The complimentary roles of reliability-centered maintenance

and condition monitoring," in International Maintenance Conference, 2003.
[103]
W. R. Blischke and D. N. P. Murthy, Case studies in reliability and maintenance. Hoboken,

NJ: John Wiley, 2003.
[104]
R. Blayden, "Rethinking reliability centred maintenance - An innovative approach to

developing operations and maintenance strategy," Maintenance and Asset Management,
vol. 21, pp. 5-11, 2006.
[105]
S. Courtney, "Improved maintenance reliability condition monitoring - the link to machine

condition performance," Maintenance and Asset Management, vol. 21, pp. 5-13, 2006.
[106]
T. Westerkamp, "21st Century maintenance management using finite capacity planning,"

Maintenance and Asset Management, vol. 21, pp. 15-22, 2006.
[107]
C. van.der.Walt, A. Duyer, A. Bates, and G. Walker, "MCM: an inexpensive condition

monitoring technology, simple to use and model based.," Maintenance and Asset
Management, vol. 21, pp. 13-22, 2006.
[108]
D. C. Shaw, "Intelligent infrastructure asset condition monitoring," presented at The IET

International Conference on Railway Condition Monitoring, 2006.
[109]
S. Jovanovic, "Integrated Railway Infrastructure Condition Monitoring for Optimal Asset

Management," in IET International Conference on Railway Condition Monitoring, 2006, pp.
13-19.
[110]
EngineerLive, "Rail asset management system aims to cut delays by 40 per cent," in Rail
International, August 2004.
[111]
M. M. Austin, "Systems health monitoring - from ground to air - the aerospace challenges,"
AIP conference proceedings, vol. 894, pp. 1477-84, 2007.
[112]
X. Zhao, "A wireless ultrasonic guided wave structural health monitoring system for aircraft
wing inspection," American Institute of Physics, vol. 894, pp. 1548-55, 2007.
[113]
E. Bechhoefer and E. Mayhew, "Mechanical diagnostics system engineering in IMD

HUMS," in IEEE Aerospace Conference, 2006, pp. 1-8.
[114]
S. Bahadoorsingh and S. M. Rowland, "The relationship between insulation ageing and

power network performance," in IEEE International Conference on Solid Dielectrics.
Winchester, UK, 2007, pp. 180-183.
[115]
A. J. Gaul, J. Schneider, C. Neumann, J. Hografer, W. Wellssow, M. Schwan, and A.

Schnettler, "Asset management techniques," International Journal of Electrical Power &
Energy Systems, vol. 28, pp. 643-54, 2006.
[116]
EngineerLive, "Maintenance vital to profitability," in Power Engineer Magazine 2006.
[117]
F. Puletti, M. Olivieri, A. Cavallini, and G. C. Montanari, "Risk management of HV

polymeric cables based on partial discharge assessment," in IEEE PES Transmission and
Distribution Conference. Dallas, USA, 2006, pp. 626-633.
[118]
G. C. Montanari, A. Cavallini, and F. Puletti, "A new approach to partial discharge testing of
HV cable systems," IEEE Electrical Insulation Magazine, vol. 22, pp. 14-23, 2006.
[119]
S. Bahadoorsingh and S. M. Rowland, "A framework linking knowledge of insulation ageing

to asset management " IEEE Electrical Insulation Magazine, vol. 24, pp. 38-46, 2008.
[120]
G. C. Montanari, "Envisaging links between fundamental research in electrical insulation

and electrical asset management," IEEE Electrical Insulation Magazine, vol. 24, pp. 7-21,
2008.
[121]
R. Davies, "PAS 55 certification and beyond," in The IEE Seminar on Developing &
190
References
[122]
D. Gooda, "Asset management competencies," in The IEE Seminar on Developing &

[123]
U. Bryan, "PAS 55 implementation and assessment - A utility's experiences," in The IEE

Seminar on Developing & Applying Best Practice Asset Management. The IEE, Savoy
Place, London, 2005.
[124]
S. Argent, "Asset risk management and PAS 55 - the way forward," in The IEE Seminar on
Developing & Applying Best Practice Asset Management. The IEE, Savoy Place, London,
2005.
[125]
P. Pourbeik, P. S. Kundur, and C. W. Taylor, "The anatomy of a power grid blackout," IEEE
Power & Energy Magazine, vol. 4, pp. 22-29, 2006.
[126]
M. M. Adibi and L. H. Fink, "Overcoming restoration challenges associated with major

power system disturbances - Restoration from cascading failures," IEEE Power & Energy
Magazine, vol. 4, pp. 68-77, 2006.
[127]
W. Li, E. Vaahedi, and P. Choudhury, "Power system equipment aging," IEEE Power &
Energy Magazine, vol. 4, pp. 52-58, 2006.
[128]
L. L. Lai, W. L. Chan, C. T. Tse, and A. T. P. So, "Real-time frequency and harmonic

evaluation using artificial neural networks," IEEE Transactions on Power Delivery, vol. 14,
pp. 52-59, 1999.
[129]
F. C. De La Rosa, Harmonics and power systems. Boca Raton, Fla. ; London: CRC/Taylor
& Francis, 2006.
[130]
J. Schlabbach, D. Blume, and T. Stephanblome, Voltage quality in electrical power

systems. London: Institution of Electrical Engineers, 2001.
[131]

to power network asset management " in IEEE International Symposium on Electrical
Insulation, vol. 24. Vancouver, Canada, 2008, pp. 38-46.
[132]
M. Wang, S. M. Rowland, and P. E. Clements, "Moisture ingress into low voltage oilimpregnated paper insulated distribution cables," Science, Measurement & Technology,
IET, vol. 1, pp. 276-283, 2007.
[133]
I. Khan, Z. Wang, I. Cotton, and S. Northcote, "Dissolved gas analysis of alternative fluids
for power transformers," IEEE Electrical Insulation Magazine, vol. 23, pp. 5-14, 2007.
[134]
K. Nakao, T. Kondo, Y. Suzuoki, and T. Mizutani, "phi;-q-n patterns and current shapes of
partial discharges in void," in International Symposium on Electrical Insulating Materials.
Toyohashi, Japan, 1998, pp. 665-668.
[135]
IEC60567, "Guide for the sampling of gases and oil from oil-filled electrical equipment and
for the analysis of free and dissolved gases," 1992.
[136]
IEEE, "Std C57.104-1991, IEEE guide for the interpretation of gases generated in oilimmersed transformers ", 1991.
[137]
IEC60599, "Mineral oil-impregnated electrical equipment in service Guide to the

interpretation of dissolved and free gases analysis," 1999.
[138]
ASTM, "Method D 5837 Standard Test Method for Furanic Compounds in Electrical
Insulating Liquids by High-Performance Liquid Chromatography (HPLC)," 1999.
[139]
R. Ishii, S. Iemura, M. Kubo, N. Nagura, and N. Shimizu, "Electroluminescence for XLPE

cable diagnosis," in International Conference on Properties and Applications of Dielectric
Materials Nagoya, Japan, 2003, pp. 199-202.
[140]
M. Kaufhold, S. S. Bamji, and A. T. Bulinski, "Electroluminescence in XLPE under impulse

voltage stress," in IEEE Conference on Electrical Insulation and Dielectric Phenomena.
San Francisco, USA, 1996, pp. 846-849.
[141]
G. C. Montanari, "Diagnostics of electrical systems: expectations and answers," in IEEE

International Conference on Solid Dielectrics. Winchester, UK, 2007, pp. 530-535.
[142]
M. Wang, S. M. Rowland, and N. G. Van Luijk, "The influence of moisture on low voltage
oil-and-paper insulated distribution cables," in IEEE Conference on Electrical Insulation
and Dielectric Phenomena Nashville, USA, 2005, pp. 112-115.
191
References
[143]
M. D. Noskov, A. S. Malinovski, M. Sack, and A. J. Schwab, "Electrical treeing in epoxy

resin: experimental investigation and simulation," in IEEE Conference on Electrical
Insulation and Dielectric Phenomena. Victoria, Canada, 2000, pp. 313-316.
[144]
M. D. Noskov, M. Sack, A. S. Malinovski, and A. J. Schwab, "Measurement and simulation

of electrical tree growth and partial discharge activity in epoxy resin " Journal of Physics D:
Applied Physics, vol. 34, pp. 1389-1398, 2001.
[145]
S. S. Bamji and A. T. Bulinski, "Electroluminescence - an optical technique to determine

the early stages of polymer degradation under high electric stresses," in Conference on
Precision Electromagnetic Measurements. Ottawa, Canada, 2002, pp. 106-107.
[146]
G. C. Stone, "Partial discharge diagnostics and electrical equipment insulation condition

assessment," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, pp. 891904, 2005.
[147]
P. Osvath, E. Carminati, and A. Gandelli, "A contribution on the traceability of partial

discharge measurements," IEEE Transactions on Electrical Insulation, vol. 27, pp. 130134, 1992.
[148]
F. H. Kreuger, Partial discharge detection in high-voltage equipment: ButterworthHeinemann, 1989.
[149]
C. Bowick, RF circuit design. Indianapolis, Ind.: Sams, 1982.
[150]
S. A. Boggs, "Partial discharge. II. Detection sensitivity," IEEE Electrical Insulation

Magazine, vol. 6, pp. 35-42, 1990.
[151]
M. Olyphant, "Breakdown by treeing in epoxy resins," IEEE Transactions on Power

Apparatus and Systems, vol. 82, pp. 1106-1112, 1963.
[152]
J. V. Champion and S. J. Dodd, "Systematic and reproducible partial discharge patterns

during electrical tree growth in an epoxy resin " Journal of Physics D: Applied Physics, vol.
29, pp. 862-868, 1996.
[153]
H. Z. Ding and B. R. Varlow, "The effect of material age on electrical tree growth in
mechanically prestressed fibre-reinforced composite insulations," in IEEE International
Conference on Properties and Applications of Dielectric Materials, 2003, pp. 940-943.
[154]
BSI, "ISO 7864:1996 Sterile hypodermic needles for single use."
[155]
K. Wu, Y. Suzuoki, T. Mizutani, and H. Xie, "Model for partial discharges associated with
treeing breakdown: I. PDs in tree channels," Journal of Physics D: Applied Physics, vol. 33,
pp. 1197-1201, 2000.
[156]
treeing breakdown: II. tree growth affected by PDs," Journal of Physics D: Applied Physics,
vol. 33, pp. 1202-1208, 2000.
[157]
treeing breakdown: III. PD extinction and re-growth of tree," Journal of Physics D: Applied
Physics, vol. 33, pp. 1209-1218, 2000.
[158]
Suwarno, H. Ichikawa, Y. Suzuoki, T. Mizutani, and K. Uchida, "Partial discharge patterns

of electrical treeing in polyethylene," in IEEE International Conference on Properties and
Applications of Dielectric Materials, vol. 1. Brisbane, Australia, 1994, pp. 379-382
[159]
H. Suwarno, "A comparison between void and electrical treeing discharges in

polyethylene," in IEEE International Conference on Properties and Applications of
Dielectric Materials. Xi'an, China, 2000, pp. 493-496.
[160]
K. Wu, Y. Suzuoki, T. Mizutani, and X. Hengkun, "A novel physical model for partial
discharge in narrow channels," IEEE Transactions on Dielectrics and Electrical Insulation,
vol. 6, pp. 181-190, 1999.
[161]
H. Z. Ding and B. R. Varlow, "Thermodynamic model for electrical tree propagation kinetics
in combined electrical and mechanical stresses," IEEE Transactions on Dielectrics and
[162]
H. Ichikawa, K. Ito, Y. Suzuoki, T. Mizutani, and K. Uchida, "Partial discharge patterns and
degradation diagnosis in polyethylene and polyimide," in IEEE International Conference on
Properties and Applications of Dielectric Materials. Brisbane, Australia, 1994, pp. 630-633.
192
References
[163]
H. Ichikawa, K. Itoh, Y. Suzuoki, T. Mizutani, and K. Uchida, "Partial discharge patterns

and their temporal change in polyethylene and polyimide.," in Conference on Electrical
Insulation and Dielectric Phenomena, 1994, pp. 143-148.
[164]
G. Loffelmacher, "About the physical-chemical procedures of discharge channels in

polyethylene and epoxy resin in an inhomogeneous alternating field.," Gottfried Wilhelm
Leibniz Universitt Hannover, 1976 (In German).
[165]
T. Tanaka, "Charge transfer and tree initiation in polyethylene subjected to AC voltage

stress," IEEE Transactions on Electrical Insulation, vol. 27, pp. 424-431, 1992.
[166]
"IEEE Std 930-2004 Guide for the Statistical Analysis of Electrical Breakdown Data " IEEE
Standards.
[167]
J. S. White, "The Moments of Log-Weibull Order Statistics," Technometrics, vol. 11, pp.
373-386, 1969.
[168]
G. C. Montanari, G. Mazzanti, M. Cacciari, and J. C. Fothergill, "Optimum estimators for

the Weibull distribution of censored data: singly-censored tests," IEEE Transactions on
Dielectrics and Electrical Insulation, vol. 4, pp. 462-469, 1997.
[169]
"Reliability Engineering Resources for the Reliability Professional."
[170]
ReliaSoft, "A high value of beta is not necessarily cause for concern," in Reliability Edge,
vol. 2, 2001.
[171]
H. Kaneiwa, Y. Suzuoki, and T. Mizutani, "Partial discharge characteristics and tree

inception in artificial simulated tree channels," IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 7, pp. 843 - 848 2000.
[172]
M. Cacciari, A. Contin, and G. C. Montanari, "Use of a mixed-Weibull distribution for the

identification of PD phenomena," IEEE Transactions on Dielectrics and Electrical
Insulation, vol. 2, pp. 1166-1179, 1995.
[173]
M. Cacciari, A. Contin, G. Mazzanti, and G. C. Montanari, "Identification and separation of

two concurrent partial discharge phenomena," IEEE Conference on Electrical Insulation
and Dielectric Phenomena, vol. 2, pp. 476-479, 1996.
[174]
A. Contin, G. Contessotto, G. C. Montanari, and M. Cacciari, "Comparing different

stochastic models for the identification and separation of concurrent partial discharge
phenomena," in IEE Conference on Dielectric Materials, Measurements and Applications,
2000, pp. 374-379.
[175]
A. Contin, G. C. Montanari, and C. Ferraro, "PD source recognition by Weibull processing

of pulse height distributions," IEEE Transactions Dielectrics and Electrical Insulation, vol. 7,
pp. 48-58, 2000.
[176]
R. Bozzo, C. Gemme, F. Guastavino, M. Cacciari, A. Contin, and G. C. Montanari, "Aging

diagnosis of insulation systems by PD measurements. Extraction of partial discharge
features in electrical treeing," IEEE Transactions on Dielectrics and Electrical Insulation,
vol. 5, pp. 118-124, 1998.
[177]
A. Contin, E. Gulski, M. Cacciari, and G. C. Montanari, "Inference of PD in electrical

insulation by charge-height probability distribution. Diagnosis of insulation system
degradation," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 5, pp. 110117, 1998.
[178]
A. Contin, G. C. Montanari, M. Conti, and M. Cacciari, "An invariant diagnostic marker for
the identification of partial discharge sources in electrical apparatus," in IEEE International
Conference on Solid Dielectrics. Eindhoven, Netherlands, 2001, pp. 287-290.
[179]
R. Bozzo, A. Contin, C. Gemme, F. Guastavino, and G. C. Montanari, "An analysis of the

probability distributions of charge in electrical trees generated by needle tests," in IEEE
Conference on Electrical Insulation and Dielectric Phenomena. San Franciso, USA., 1996,
pp. 770-774.
[180]
IEC56, "Procedures for Goodness-of-fit Tests, Confidence Intervals and Lower Confidence
Limits for Weibull Distributed Data."
[181]
M. Cacciari and G. C. Montanari, "Discussion, with reply, on 'Estimating the cumulative

probability of failure data points to be plotted on Weibull and other probability paper'," IEEE
Transactions on Electrical Insulation, vol. 26, pp. 1224-1229, 1991.
193
Appendix A
Appendix A Weibull plots of breakdown times

y = 4.3x - 38.9
0.0
-1.0
-2.0
-3.0
-4.0
y = 3.2x - 29.0
1.0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
1.0
-5.0
0.0
-1.0
-2.0
-3.0
-4.0
8.0
8.5
9.0
9.5
8.1
8.4
ln(t/sec)
8.9
9.1
Figure A-1: Weibull plot for total
Figure A-2: Weibull plot for THD=All and
population
Ks=1.6
y = 5.0x - 44.9
1.0
0.0
-1.0
-2.0
-3.0
y = 3.4x - 30.5
0.0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
8.6
ln(t/sec)
-4.0
-0.5
-1.0
-1.5
-2.0
-2.5
8.3
8.5
8.8
9.0
9.3
8.4
8.6
ln(t/sec)
8.8
9.0
9.2
ln(t/sec)
Figure A-3: Weibull plot for THD=5.0 %
and Ks=All
and Ks=1.6
y = 2.5x - 22.9
0.0
-1.0
-2.0
-3.0
y = 6.9x - 63.2
0.0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
1.0
-0.5
-1.0
-1.5
-2.0
-2.5
8.1
8.4
8.6
8.9
9.1
ln(t/sec)
8.9
8.9
9.0
9.0
9.1
9.1
ln(t/sec)
and Ks=1.6
and Ks=1.03
194
Appendix A
y = 3.6x - 32.2
0.0
-0.5
-1.0
-1.5
-2.0
y = 5.5x - 49.3
2.0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
0.5
1.0
0.0
-1.0
-2.0
-3.0
-2.5
8.2
8.4
8.6
8.8
9.0
8.6
9.2
8.7
8.8
8.9
9.0
9.1
ln(t/sec)
ln(t/sec)
and Ks=1.06
and Ks=1.27
y = 2.3x - 20.8
-0.5
-1.0
-1.5
-2.0
y = 5.8x - 52.0
0.0
ln(-ln(1-F(t)))
ln(-ln(1-F(t)))
0.0
-0.5
-1.0
-1.5
-2.0
-2.5
-2.5
8.0
8.2
8.4
8.6
8.8
9.0
ln(t/sec)
8.6
8.7
8.8
8.9
9.0
ln(t/sec)
and Ks=1.56
and Ks=1.00
195
Appendix B
Appendix B Partial discharge data and tree images
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
70
72
52
55
52
51
62
-2
1.3
1.5
1.7
1.8
2.2
2.2
2.1
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
26
26
24
27
28
30
47
-2
1.1
1.1
1.2
1.0
0.9
1.0
1.1
-4
10
10
10
log(Charge/pC)
Figure B-1: K101 (1st run) phase resolve plots (top) and Weibull plots (bottom).
196
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
60
64
52
49
49
43
39
-2
1.9
2.1
2.1
2.2
2.4
2.9
2.1
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
28
29
88
79
75
66
57
-2
1.0
1.1
2.0
2.0
2.2
2.0
1.7
-4
10
10
10
log(Charge/pC)
Figure B-2: K101 (2nd run) phase resolve plots (top) and Weibull plots (bottom).
197
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
21
--------25
23
21
30
-2
2.2
--------2.1
1.9
1.5
1.6
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
15
--------13
7
7
2
-2
1.6
--------1.5
1.1
1.4
0.5
-4
10
10
10
log(Charge/pC)
198
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
27
26
25
22
19
18
26
-2
1.8
1.7
1.9
1.9
1.9
1.8
1.6
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
6
5
5
5
4
5
5
-2
1.5
1.6
1.6
2.6
1.4
2.7
1.7
-4
10
10
10
log(Charge/pC)
199
Appendix B
Figure B-5: K102 Tree growth images.
200
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
----113
127
140
124
87
61
-2
----2.4
2.8
2.5
2.7
1.1
1.5
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
----39
39
40
40
37
37
-2
----1.8
2.2
2.0
2.2
1.6
1.8
-4
10
10
10
log(Charge/pC)
201
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
115
254
238
250
299
345
312
-2
1.9
3.0
2.7
2.9
3.2
3.2
2.0
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
47
63
38
34
18
22
46
-2
2.2
1.8
1.4
1.6
10.6
1.0
1.3
-4
10
10
10
log(Charge/pC)
202
Appendix B
203
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
68
85
83
70
74
82
96
-2
2.3
2.8
2.3
1.7
1.9
2.1
2.3
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
45
53
50
51
59
67
54
-2
2.3
2.2
1.9
2.1
2.1
2.4
2.0
-4
10
10
10
log(Charge/pC)
204
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
92
88
82
65
62
53
68
-2
2.1
2.0
2.2
2.0
2.0
2.0
2.1
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
44
46
50
49
49
54
70
-2
1.8
1.8
1.9
1.9
1.9
1.8
1.9
-4
10
10
10
log(Charge/pC)
205
Appendix B
Figure B-11 K106 Tree growth images
206
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
61
67
67
66
68
73
112
-2
1.9
1.9
2.0
1.9
1.9
1.9
2.1
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
33
40
41
40
43
49
75
-2
1.7
1.8
1.8
2.0
1.9
1.8
1.7
-4
10
10
10
log(Charge/pC)
207
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
95
94
96
88
82
78
85
-2
2.1
2.1
1.8
1.9
2.0
1.9
1.7
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
72
68
69
62
59
57
48
-2
1.6
1.7
1.8
1.7
1.7
1.7
1.6
-4
10
10
10
log(Charge/pC)
208
Appendix B
209
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
247
258
178
283
271
236
238
-2
1.1
1.4
1.1
1.2
1.9
1.4
1.6
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
135
147
136
132
144
143
165
-2
2.0
2.1
1.6
1.9
2.7
2.2
2.1
-4
10
10
10
log(Charge/pC)
210
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
228
218
214
203
195
197
219
-2
2.3
2.1
2.7
3.0
2.9
3.0
3.1
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
157
165
171
161
158
157
162
-2
2.7
2.9
3.3
3.2
3.2
3.0
2.5
-4
10
10
10
log(Charge/pC)
Figure B-16 K109: (2nd run) phase resolve plots (top) and Weibull plots (bottom).
211
Appendix B
Plots 8-14 unavailable

212
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
33
31
23
15
13
15
15
-2
1.8
1.7
1.6
2.5
1.6
1.8
2.3
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
12
13
10
12
12
14
14
-2
0.6
2.0
0.9
2.4
1.3
1.6
1.8
-4
10
10
10
log(Charge/pC)
213
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
15
14
17
19
36
31
40
-2
1.5
1.4
1.1
1.1
1.4
1.6
1.6
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
10
10
11
14
35
24
34
-2
1.2
1.5
0.7
0.8
1.3
1.1
1.2
-4
10
10
10
log(Charge/pC)
214
Appendix B
215
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
44
42
43
38
36
33
41
-2
2.3
2.7
1.6
1.8
2.1
1.7
1.6
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
47
59
54
107
142
127
150
-2
1.7
1.8
1.8
2.1
5.1
2.1
3.9
-4
10
10
10
log(Charge/pC)
216
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
55
94
41
24
25
31
37
-2
1.5
2.3
2.5
1.3
1.2
1.3
2.2
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
163
231
21
97
68
64
76
-2
2.8
3.1
1.4
2.1
2.0
1.7
1.7
-4
10
10
10
log(Charge/pC)
217
Appendix B
218
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
38
52
39
50
32
40
41
-2
1.9
2.1
1.5
1.8
1.4
2.2
1.4
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
108
118
72
77
82
66
65
-2
1.8
1.8
1.4
1.3
1.4
1.3
1.4
-4
10
10
10
log(Charge/pC)
219
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
51
51
52
39
49
49
69
-2
2.1
1.9
2.1
1.3
1.8
1.3
1.2
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
87
67
22
26
19
26
48
-2
1.6
1.3
0.9
0.9
0.8
0.8
1.0
-4
10
10
10
log(Charge/pC)
220
Appendix B
221
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
59
66
51
66
52
63
66
-2
1.6
1.9
1.7
1.9
2.3
1.7
2.0
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
92
101
95
58
43
33
27
-2
1.7
1.8
1.7
1.4
1.3
1.2
1.1
-4
10
10
10
log(Charge/pC)
222
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
58
69
63
107
124
117
127
-2
2.0
2.0
1.4
2.0
2.1
2.0
2.0
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
22
24
21
22
18
20
19
-2
2.2
1.9
0.8
1.3
1.5
1.8
0.6
-4
10
10
10
log(Charge/pC)
223
Appendix B
224
Appendix B
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
177
198
131
103
103
93
90
-2
1.7
1.8
1.4
1.3
1.7
1.8
1.3
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 1
Wave 12/Plot 2
Wave 11/Plot 3
Wave 9/Plot 4
Wave 8/Plot 5
Wave 7/Plot 6
Wave 1/Plot 7
40
123
113
86
95
133
106
-2
1.0
1.7
1.4
1.3
1.5
1.7
1.5
-4
10
10
10
log(Charge/pC)
Figure B-30 K115: (1st run) phase resolve plots (top) and Weibull plots (bottom).
225
Appendix B
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
83
97
110
83
66
83
94
-2
1.3
1.2
1.3
1.6
1.3
1.7
1.5
-4
10
10
10
log(Charge/pC)
10
10
10
10
Wave 13/Plot 8
Wave 12/Plot 9
Wave 11/Plot 10
Wave 9/Plot 11
Wave 8/Plot 12
Wave 7/Plot 13
Wave 1/Plot 14
139
147
121
142
139
137
89
-2
1.8
1.9
1.7
1.7
1.8
1.7
1.4
-4
10
10
10
log(Charge/pC)
Figure B-31 K115: (2nd run) phase resolve plots (top) and Weibull plots (bottom).
226
Appendix B
227
Appendix B
Figure B-33: K101 Cosh model plots with the partial discharge activity captured.
228
Appendix B
229
Appendix B
230
Appendix B
231
Appendix B
Figure B-37: K108 Cosh model plots with the partial discharge activity captured:
232
Appendix B
233
Appendix B
234
Appendix B
235
Appendix B
236
Appendix B
Figure B-42 K114: Cosh model plots with the partial discharge activity captured.
237
Appendix B
238
Appendix C
Appendix C List of publications

[1]

to asset management " IEEE Electrical Insulation Magazine, vol. 24, pp. 38-46, 2008.
[2]

to power network asset management " IEEE International Symposium on Electrical
Insulation, Vancouver, Canada, pp. 38-46, 2008.
[3]
S. Bahadoorsingh and S. M. Rowland, "The role of power quality in electrical treeing of

epoxy resin," IEEE Conference on Electrical Insulation and Dielectric Phenomena,
Vancouver, Canada, pp. 221-224, 2007.
[4]
S. Bahadoorsingh and S. M. Rowland, "Development of a multifactor framework linking

insulation ageing and power network environments," IEEE Conference on Electrical
Insulation and Dielectric Phenomena, Vancouver, Canada, pp. 388-391, 2007.
[5]
S. Bahadoorsingh and S. M. Rowland, "The relationship between insulation ageing and

power network performance," IEEE International Conference on Solid Dielectrics,
Winchester, UK, pp. 180-183, 2007.
239

Final SB Thesis 01apr09 SQ

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Final SB Thesis 01apr09 SQ

Transféré par

Droits d'auteur :

Formats disponibles

Asset Management and the Role of Power

Quality on Electrical Treeing in Epoxy Resin

A thesis submitted to The University of Manchester

School of Electrical and Electronic Engineering

1.13.5. Discussion ..................................................................................................... 54

4. Experimental results and analysis......................................................................... 112

Figure 3-9: Overview of test equipment and their interfaces............................................................ 81

Arbitrary Waveform Generator

British Standards Institute

Condition Based Monitoring

Charge Coupled Device

Customer Minutes Lost

Ethylene Propylene Rubber

Electrical Tree Growth

Fast Fourier Transform

Institute of Asset Management

International Electrotechnical Commission

Institute of Electrical & Electronic Engineers

Integrated Mechanical Diagnostic Health & Usage System

International Organization for Standardization

Motor Condition Monitoring

Maintenance Finite Capacity Planning

Operator Driven Reliability

Office of Gas and Electricity Markets

Occupational Health & Safety Advisory Services

Publicly Available Specification

Peripheral Components Interconnect

Phase-resolved Partial Discharge

Rail Asset Management Systems

Reliability Centered Maintenance

Time Based Maintenance

Total Demand Distortion

Total Harmonic Distortion

Tree Inception Voltage

Transistor Transistor Logic

Water Tree Growth

the function cosh V +

Supergen V is a collaborative research partnership amongst six leading UK universities,

To deliver intelligent diagnostic tools for plant; enabling optimum usage,

To provide platform technologies for integrated network planning and asset

To investigate alternative plant and reduce environmental impact of networks.

To develop models and recommendations for improved network operation and

Maintaining a reliable energy supply at minimal cost is a requirement of any power

New forms of generation and changes in the load characteristics.

Modification in scheduled maintenance procedures.

Consequently, there is a gradual change in the working environment experienced by

An interpretation of insulation ageing

Figure 1-1: Factors which influence insulation ageing.

Space charge accumulation

Partial discharge (PD)

Water tree growth (WT)

Electrical tree growth (ET)

Figure 1-2: Dry ageing of polymeric insulation [8].

Bow-tie trees which originate from a contaminant, boundary surface or water

Figure 1-3: Wet ageing of polymeric insulation [8].

1.6.1. Transition from water trees to electrical trees

An electrical tree is a path of damage incurred by polymeric insulation as a consequence

Figure 1-4: Schematic representation of typical electrical tree growth [3].

1.7.1. Electrical tree types

A bushy tree where the tubules are densely packed [3].

Field Magnitude (kV/mm)

1.7.2. Electrical tree initiation

Charge Injection &