See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/314298289

SDMF based interference rejection and PD interpretation for simulated defects

in HV cable diagnostics

Article in IEEE Transactions on Dielectrics and Electrical Insulation · February 2017

DOI: 10.1109/TDEI.2016.006002

CITATIONS READS

6 98

9 authors, including:

Xiaosheng Peng Chengke Zhou

Glasgow Caledonian University Glasgow Caledonian University
16 PUBLICATIONS 48 CITATIONS 211 PUBLICATIONS 4,566 CITATIONS

SEE PROFILE SEE PROFILE

Alistair Reid Donald M. Hepburn

Cardiff University Glasgow Caledonian University
66 PUBLICATIONS 569 CITATIONS 28 PUBLICATIONS 154 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Scalable Non-invasive Radiometric Wireless Sensor Network for Partial Discharge Monitoring in the Future Smart Grid View project

Special Issue "UHF and RF Sensor Technology for Partial Discharge Detection" View project

All content following this page was uploaded by Alistair Reid on 21 February 2019.

The user has requested enhancement of the downloaded file.

 SDMF based Interference Rejection and PD Interpretation
for Simulated Defects in HV Cable Diagnostics
Xiaosheng Peng, Jinyu Wen, Zhaohui Li, Guangyao Yang
State Key Laboratory of Advanced Electromagnetic Engineering and Technology
School of Electrical and Electronic Engineering
Huazhong University of Science and Technology, Wuhan, China, 430074

Chengke Zhou, Alistair Reid, Donald M. Hepburn

School of Engineering and Built Environment
Glasgow Caledonian University
Cowcaddens Road, Glasgow, UK, G4 0BA

Martin D. Judd and W. H. Siew

Department of Electronic & Electrical Engineering
University of Strathclyde
Royal College, 204 George Street, Glasgow, UK, G1 1XW

ABSTRACT
Partial Discharge (PD) in cable systems causes deterioration and failure, identifying
the presence of PD is crucial to Asset Management. In this paper presents methods for
interference signals rejection and for PD interpretation for five types of artificial defect
in 11 kV ethylene-propylene rubber (EPR) cable. Firstly, the physical parameters of the
artificial defects used for PD signal generation are introduced. Thereafter, the sample
stress regime, PD testing and detection systems, including IEC 60270 measurement
system and High Frequency Current Transformer (HFCT), are outlined. Following on,
a novel Synchronous Detection and Multi-information Fusion (SDMF) based signal
identification method is developed, to separate PD and interference signals within raw
data. Finally, a comparative PD analysis of two detection systems is carried out and
several characteristics of insulation related PD signals of EPR cables are reported. The
SDMF based data pre-processing and the comparative PD activity analysis contribute
to improvement of PD pattern recognition and assist in on-line PD monitoring systems.
Index Terms — Partial Discharge, Cable System, Signal Identification, Interference
Rejection, Feature Extraction.

1 INTRODUCTION investigation can take place.

PD testing in a controlled HV laboratory environment is
PARTIAL Discharge (PD) based condition monitoring
invaluable as the PD defect type and PD signals are both
systems have been increasingly adopted for power apparatus in
known quantities. Comparative PD testing between an IEC
recent years [1-4]. For successful PD recognition and
60270 system and commercial HFCT to obtain information on
interpretation in Medium Voltage (MV) and High Voltage
differences in PD characteristics is of great value to allow
(HV) cables there are two essential challenges which need to
understanding of the relationships between these measures.
be overcome.
Secondly, for pattern recognition by automated intelligent
Firstly, most of the research into PD in cables is based on
(AI) software algorithms, a large volume of training data
IEC60270 systems, e.g. [5-8]. The authors contend that the
with known signal types and defect types is demanded. The
characteristics of PD signals from cables that are detected by
effective operation of AI systems requires both a large
commercial HFCT sensors need to be studied in more detail.
volume of experimental data and the ability to
For most on-line PD monitoring systems it is difficult to
automatically pre-process raw data to reject interference
correlate PD signals with corresponding cable defect types as,
signals [8-10]. IEC 60270 recommends a threshold of two
if cable maintenance is not carried out after PD signals are
times that of detected white noise to separate the
detected, then the fault is damaged beyond recognition, and
interference signals from the PD data [11] as the Signal to
cable sections are often disposed of before forensic
Noise Ratio (SNR) of an IEC 60270 system is high. For PD
signal detection by HFCT, even in HV laboratory bedding, aluminum wire armor and a PVC oversheath with
conditions, the SNR is low. PD signal interpretation from reduced flame propagation characteristics [12].
HFCT detection systems is, therefore, challenging if The five defect topologies investigated in this study,
interference signals are not effectively rejected. reported to exist in cables, are simulated as illustrated in Fig. 2
To address the first issue, a comparative experiment and Fig. 3.
which combined the laboratory-based IEC 60270 system
and a typical on-line PD monitoring setup for cable systems
was established. Comparative analysis of PD signal
characteristics of the two systems are carried out and, based 0.4
on this, recommendations for on-line PD monitoring of 0.4
cable systems are made. Defect
Defect
To address the second issue, an automatic, Synchronous type1 type2
Detection and Multi-information Fusion (SDMF) signal
identification method, capable of identifying and separating 2 2
PD and repetitive pulsed interference, is presented. The 3.75 (a) 3.75 (b)
method is shown to be effective for data pre-processing of
raw experimental data detected by HFCT with low SNR,
and which could be applied for PD lab testing of other
power apparatus, e.g. HV transformer, motor, Gas Insulated 0.4 7
Switchgear (GIS), etc. The output of SDMF based signal
Defect Defect
identification, which are PD data from HFCT, could be
type3 type 4
applied to both statistical PD signals interpretation and to
PD criteria identification and pattern recognition for HFCT 2
based on-line PD condition monitoring systems, this aspect (c) (d)
3.75
will be presented in a future paper.
Fig. 2. Defect type and dimension: (a) 1: void in insulation; 0.4mm
diameter, 2mm depth. (b) 2: protrusion on outer conductor; 0.4mm
2 EXPERIMENTAL SETUP AND ARTIFICIAL diameter, 2mm depth. (c) 3: floating protrusion; 0.4mm diameter, 2mm
DEFECTS depth. (d) 4: breach in outer conductor; 7mm*7mm.
Laboratory testing of five types of artificial defects in 11 kV
EPR cable was carried out. The artificial defects will be
discussed in Section 2.1.
The PD data was collected using both a capacitive coupling
method, based on IEC 60270, and a commercially available
HFCT connected to a 1 GHz LeCroy digital oscilloscope.
A high voltage AC supply was connected to the cable under
test, with associated artificial defects, to generate PD signals
for pattern recognition. Additional information on the
experimental set-up will be provided in Section 2.2.
2.1 EPR CABLE AND ARTIFICIAL DEFECTS
The construction and dimensions of the layers of the EPR
cable are shown in Fig. 1.

Fig. 3. Defect type 5: Surface discharge around end termination.

Aluminium core
Inner semiconductor To generate defect type 1, void in cable insulation, a
EPR layer cylindrical void was generated by drilling a hole from the outer
Outer semiconductor
21 23 30.5 32.5 34.5 wall towards the inner conductor, using a precision 0.4 mm
Sheath (Wound Cu tape) diameter printed-circuit-board drill. As shown in Fig. 3(a),
Aluminium armour
copper tape is placed over the hole, in touch with the outer
PVC over sheath
conductor, to seal the defect.
Defects type 2 and 3, replicate the situation where a metallic
Fig. 1. The dimensions and layers of the EPR cable sample. Units: mm protrusion occurs due to a manufacturing defect or external
damage [12]. The presence of these defects leads to a strongly
As indicated in Fig. 1, the single-core 11 kV cable has a inhomogeneous field which will provide a site where PD is
stranded, compacted aluminum core, EPR insulation, likely to occur. Type 2 protrusions are created by inserting a
removable semiconducting screen, copper tape screen, PVC printed-circuit-board drill into the cable: the drill is left in
contact with the outer conductor, the aluminium armour and one of the five artificial defects described in section 2.1, were
the PVC over sheath. Defect type 3 is a floating protrusion, i.e. connected (one at a time) to the high voltage supply for PD
there is no connection between the conductive parts and the testing. As shown in Fig.6, a voltage probe was used to obtain
drill. the 50 Hz phase information and as the trigger for PD data
Defect type 4, simulates a breach in outer conductor, a acquisition. The signal from the HFCT is amplified by a wide-
delamination or interruption in the semiconducting or band amplifier and, as the PD signals from the IEC 60270
conductor layers, which may also occur due to manufacturing measurement impedance Zm are of high amplitude, a wide-
defects or during installation. This type of fault is created by band attenuator was applied to reduce the signal to the input
cutting off a 7mm*7mm area from the semiconductor, sheath, limits of the oscilloscope (±10V). The signals from the HFCT
aluminum armor, PVC over sheath layers of the EPR cable, as and the IEC 60270 measurement impedance Zm are
shown in Figure 3(d). synchronously logged for equitable comparisons. The
PD may also occur due to incomplete or incorrect oscilloscope captured raw PD data using a data acquisition rate
installation of stress cones at a cable’s end terminations. of 100 MS/s. The time base was set to 20 ms, with the period
Provided sufficient gap exists between inner and outer of the 50 Hz power supply frequency.
conductors at the end termination, the likelihood of breakdown One experimental setup is shown in Fig. 6.
is less than that of the internal fault types described above [12].
Defect type 5, shown in Fig. 3, is created by exposing part of
the outer copper sheath and connecting it to earth.
2.2 EXPERIMENTAL SETUP
A commercially available HFCT and an IEC 60270 [11] EPR cable sample Artificial defect
testing system were combined in the laboratory for PD testing.
The commercial HFCT adopted has been applied in PD on-
line monitoring systems for many years, and used successfully
by some authors in previous work [13]. The frequency response
of the HFCT, as shown in Fig. 4, has a passband of 20 kHz to
20 MHz. The connection of the systems to the LeCroy 104Xi
digital oscilloscope is shown in Fig. 5.
Stress cone

Earth strap
HFCT
Fig. 6. Experimental setup for EPR cable sample, showing an example
defect, earth strap and stress cone.

The voltage was increased in 1kV increments from 0 to PD

Inception Voltage (PDIV), then taken beyond to a pre-
determined maximum level. For defect types 1, 4 and 5, the
maximum value was the rated voltage of the cable, 11kV. For
defect types 2 and 3, the maximum voltage was 13 kV and 12
Fig. 4. HFCT frequency response kV respectively. Data was logged for each voltage level. The
test voltages and the number of data sets for each test are
Voltage Stress Artificial shown in Table 1.
defect
probe cone
Due to the high PD count for defect type 5, fewer data sets
Cable sample were required to obtain useful information. To identify the
Earth strap
defect type from the acquired data, it is expected that over 400
HFCT
transient pulses will need to be extracted from the raw data.
AC ~
LeCroy 104Xi, 1GHz +25 Table 1. PD test voltage and number of sets of 20 ms data for each defect.
dB
-30 Wide-band Defect 5 6 7 8 9 10 11 12 13 SUM
Wide-band dB amplifier Type kV kV kV kV kV kV kV kV kV
attenuator
Type 1 50 50 50 50 50 50 50 0 0 350
(b) Type 2 0 50 50 50 50 50 54 53 52 409
Fig. 5. PD detection system combining a commercial HFCT and an Type 3 0 51 50 50 26 50 52 61 0 340
IEC60270 system. Data was measured relative to the 50 Hz AC supply. Ck
represents the IEC coupling capacitor and Zm the measurement impedance. Type 4 0 0 50 50 51 52 51 0 0 254

Type 5 0 0 10 10 10 10 10 0 0 50
EPR cable samples of length 1.5 m to 2 m, each containing
3 SDMF BASED INTERFERENCE REJECTION FOR window. If a pulse has the correct shape and this time
EXPERIMENTAL DATA dependency, then the pulse is considered to be a PD pulse.
3 6
3.1 NOISE AND INTERFERENCE FROM DEFECTS Interference signals
2 4
Example sets of data from defect types 1 to 5 as detected by

PD Magnitude (mV)

PD Magnitude (V)
1 2
the HFCT and IEC 60270 systems are shown in Fig. 7. Fig. 7(a) 0 0
to (e) shows signals from defect types 1 to 5 respectively. Data -1 -2
in green, with axis on the right hand side of the figure, are PD -2 -4
PD signals
signal detected by the IEC 60270 system. Data in red, with
-3 -6
axis on the left hand side of the figure, are from the HFCT. 0 2 4 6 8 10
(a)
12 14 16 18
Time (ms)
20
2 5
Three typical noise and interference signals were detected PD signals

PD Magnitude (mV)

PD Magnitude (V)
during the experiments: white noise, regular interference 0 0
signals and random interference signals.
White noise was detected during PD testing the five -2 -5
Regular interference signals
types of defect. The SNR of data gathered using the HFCT
is lower than that in signals gathered using the IEC testing -4 -10
0 2 4 6 8 10 12 14 16 18 20
system. IEC 60270 recommends setting the threshold value 2
(b) Time (ms)
10
for PD pulse detection at two times the level of white noise

PD Magnitude (mV)

PD Magnitude (V)
but, as shown in Fig. 7 (a), this is a challenge for PD
detection using HFCT. When PD signals have a similar 0 0

magnitude to the noisy raw data then significant loss of

detected PD will occur unless appropriate interference -2
-10

Regular interference signals

rejection techniques are applied. PD signals

-20
In addition to the white noise detected in all data sets, other 0 2 4 6 8 10
(c)
12 14 16 18
Time (ms)
20

4 20
pulsative noise signals were detected. During testing of defect Time(ms)
3 15
types 2 and type 3 regular pulsative interference signals were
PD Magnitude (mV)

2 10

PD Magnitude (V)
detected, examples of strong regular pulses are indicated with 1 5

arrows in Fig. 7 (b) and Fig. 7 (c). Additionally, random 0 0

-1 -5
pulsative interference signals were detected during some
-2 -10
PD signals
tests. As some of the regular interference pulses have similar -3 -15
magnitude to PD, if the threshold for PD extraction is set at a -4
0 2 4 6 8 10 12 14 16 18
-20
20
level appropriate to the white noise then both PD and regular 200
(d) Time (ms)
500
interference signals will be extracted from raw data. 150
100 250
PD Magnitude (mV)

For signal interpretation and pattern recognition of PD

PD Magnitude (V)
50
detected by HFCT, it is therefore essential to correctly 0 0
extract PD signals from raw data with white noise, regular -50

interference signal and random interference signal. In the -100 -250

-150 PD signals
following section the development of an automatic
-200 -500
algorithm, capable of processing large volumes of data to 0 2 4 6 8 10 12 14 16 18 20
(e) Time (ms)
separate PD from other signals and to organise the data for Signal detected by HFCT Signal detected by IEC 60270 system
recognition systems is outlined. (channel 2 of the oscilloscope) (channel 3 of the oscilloscope)

3.2 SDMF BASED INTERFERENCE SIGNAL Fig. 7. Examples of (a) Type 1, (b) Type 2, (c) Type 3, (d) Type 4, (e)
Type 5 PD data.
REJECTION
Examples of transient PD pulses contained within the 2
4
signals captured and shown in Fig. 7 (b) are presented in Fig. 8.
Transient signals from HFCT and IEC 60270 systems are
PD Magnitude (mV)

1 2
PD Magnitude (V)

clearly different, as seen in Fig. 8. PD pulse shapes from the

0
HFCT have different characteristics compared with those 0
detected of IEC 60270 (IEC pulses), e.g. the oscillation tail of -2
HFCT transients persist for longer than that of IEC pulses. IEC
-1
pulses always arrive at the oscilloscope about 100 ns after the 5.109 5.11 5.111 5.112
Time(ms)
5.113 5.114 5.115

HFCT pulses, due to the different responses and cable transfer Time
window
Signal detected by HFCT
(channel 2 of the oscilloscope)
Signal detected by IEC 60270 system
(channel 3 of the oscilloscope)
times. If a fixed time window of 500 ns is placed around an
HFCT pulse then the IEC pulse is contained in the time Fig. 8. PD Transient pulses from HFCT (lower trace) and IEC 60270
(upper trace).
 A novel Synchronous Detection and Multi-information interference signals identified by the method shown in Fig.
Fusion (SDMF) based signal identification method has been 9. Fig. 10 (a) shows the PRPD pattern in terms of maximum
developed for automatic PD and interference signal magnitude of transient pulses; Fig. 10 (b) shows the PRPD
identification from the raw data captured in the experiments pattern in terms of Equivalent Bandwidth [14] of transient
discussed above. A flowchart of SDMF based PD signal pulses; Fig. 10 (c) shows the PRPD pattern in terms of
identification process is shown in Fig. 9. skewness of transient pulses. The visualisation of PRPD of
different parameters proves the effectiveness of the SDMF
Synchronously detected PD raw based interference signals rejection.
Data from Channel 2 and Channel 3

Channel 2: Channel 3:
Low SNR Data from High SNR Data from IEC
HFCT 60270 system

Threshold
Selection
Threshold
Selection
Transient Pulses
(a) Time (ms)
Extraction

Transient Pulses PRPD Based

Extraction Validation

Data Fusion PD Signals of

Channel 3

Location
PD From Yes Information of (b) Time (ms)
Channel 3 in 500ns PDs
time windows?

No

Interference PD Signals of
Signals of Channel 2
Channel 2

Fig. 9. Flow chart of SDMF based PD signal identification

(c) Time (ms)
In Fig. 9, data acquisition for channel 2 and channel 3 are PD of Defect Type 1 Interference Signals

fully synchronised with the AC signal collected in Channel Fig. 10. Identified PD and interference signals of defect type one captured
1. The raw data from an HFCT and from an IEC detection by HFCT, (a) PRPD of maximum voltage, (b) PRPD of equivalent
system (channels 2 and 3 respectively) are logged and bandwidth, (c) PRPD of skewness.
processed in pairs. SDMF based PD identification of five defect types are
Data from the IEC 60270 system has a high SNR and carried out: results are shown in Fig.11. IEC system results
few interference signals. This data is processed first, are shown in Fig. 11 (a), HFCT are shown in Fig. 11 (b).
according to threshold selection, transient pulse extraction
and PRPD based PD signal validation. Using the PD signal
location information, e.g. 16.556ms, the information is
applied to PD signal identification for the HFCT data.
Data from the HFCT channel has a low SNR: data
extracted, both PD and interference signals, are fused with
the location information from the IEC channel. The data
fusion judgement is carried out based on the 500 ns window
discussed previously. If the signals do not show good
correlation the pulses are judged to be interference signals.
SDMF based PD and interference signals detected by
HFCT from defect type 1 are shown in Fig. 10.
Data in green are PD signals, data in black are
 Value 1
10
15
8
6 Value 2 10
4 Value 3
Value 4 5
2
Value 5
(a) (f)
7
2.5
5
(a) Time (ms)
1.5 3

0.5 1
(b) (g)
7 20

PD Magnitude (mV)

PD Magnitude (V)
15
5
10
3
5
1
(c) (h)
(b) Time (ms) 150
20
PD of Defect Type 1 PD of Defect Type 2 PD of Defect Type 3
15 100
PD of Defect Type 4 PD of Defect Type 5
10
50
Fig. 11. PD signals from defect types 1 to 5, (a) PD signals from the IEC 5
60270 system, (b) PD signals from the HFCT. 0
(d) (i)

100 6
4 COMPARATIVE ANALYSIS OF PARTIAL 80
4
DISCHARGE DATA DETECTED BY HFCT AND 60
40 2
IEC 60270 SYSTEM 20
0 0
Based on the PD data extracted by SDMF methods, 5 6 7 8 9 10 11 12 13
(e)
5 6 7 8 9 10 11 12 13
(j)
statistical analysis is carried out to compare the PD activity Test Voltage (kV) Test Voltage (kV)
behaviour of the five types of defects. Maximum Value Median Value

4.1 VARIATION IN PD MAGNITUDE WITH APPLIED Fig. 12. Violin plot of PD magnitude from defects (a) Type 1 HFCT, (b)
Type 2 HFCT, (c) Type 3 HFCT, (d) Type 4 HFCT, (e) Type 5 HFCT: (f)
VOLTAGE Type 1 IEC, (g) Type 2 IEC, (h) Type 3 IEC, (i) Type 4 IEC, (j) Type 5
Violin plots are a convenient way of graphically depicting IEC.
groups of numerical data, they combine a boxplot and a
(doubled) kernel density plot [15]. For each violin plot, five key From Fig. 12 the following conclusions may be drawn:
data parameters are indicated, i.e. maximum, minimum and Firstly, as the magnitudes of HFCT PD signals are in mV
median values and first and third quartiles. Violin plots of and those of IEC PD signals are in V, the PD signal level from
maximum PD magnitude of transient pulses in terms of the IEC 60270 system is always much greater than that from
applied voltage are shown in Fig. 12. In Fig. 12 (a), value 1 is the HFCT sensor. It is, however, essential that research on
the maximum value; value 2 is the third quartile; value 3 is the signal processing and pattern recognition is based on
median value; value 4 is the first quartile and value 5 is the commercial HFCTs if effective PD on-line monitoring system
minimum value. Violin plots also shows the statistical kernel is to be carried out.
density of the samples, in terms of a double distribution curves. Secondly, as PD signals from the commercial HFCT are of
Fig. 12 (a) contains 7 violin plots, which represent the small magnitude, e.g. the median value of PD signals for
statistical results of PD magnitude of defect type 1 changing defect types 1, 2 and 3 is less than 5 mV (Fig. 12(a) – (c)), it is
with the applied voltage, from 5 kV to 11 kV. a challenge to extract PD signals in data from noisy industrial
Fig. 12 (a) to (e) show violin plots of PD magnitude for five environments, as the white noise of most industrial detection
artificial defect types using statistical analysis of the HFCT systems is more than 5mV.
data. Fig. 12 (f) to (j) are the PD magnitude violin plots of the Thirdly, although the median PD magnitudes increase with
statistical analysis of the IEC data for the same faults. increasing voltage for most faults, for type 2 the median PD
magnitude decreases with voltages over 9 kV (Fig. 12(b) and
Fig. 12(g)).
 4.2 VARIATION IN PD PULSE NUMBER WITH A rapid decrease of PD magnitude (Fig. 12 (b)) and a rapid
APPLIED VOLTAGE increase of pulse numbers (Fig. 13 (b)) might be an effective
Violin plots of PD numbers in 20 ms with applied voltage identifier of serous insulation degradation, which should be
are shown in Fig. 12. As the number of PD pulses detected by cautioned during on-line PD monitoring of EPR cables.
HFCT and IEC 60270 systems is the same only one set of
figures is presented.
For defect types 1, 3 and 4 the number of PD pulses in each
set of data is low: in some cases there is only one PD pulse in
20 ms. For types 1 and 3 the variation in PD numbers with
increasing voltage is inconsistent: for type 4 the number shows
an increasing number of PD with increasing voltage. For these
defects the trends for the PD number show strong similarity to
the PD magnitude trends, Fig. 12.
(a)
For defect type 2, there is a rapid increase in PD number
when the voltage is increased to 11 kV and above. If compared
with Fig. 12 (b), there is a rapid decrease in the median value
of the PD magnitude at 11kV and above.
For defect type 5, there is a rapid increase in PD number as
the voltage is increased throughout the test. This trend is
similarly shown in PD magnitude, Fig. 12 (e) and (j).
5 6
(b)
4 5
Fig. 14. Flashover of defect type 2 Damage to (a) EPR cable, (b) needle.
4
3
3
2 2 4.4 VARIATION IN PD CHARACTERISTICS WITH
1 1 SIZE OF VOID DEFECT
(a) (d)
To investigate the variation of PD characteristics when the
PD Pulse Number in 20 ms

20
15 size of a void in a cable increases, e.g. with insulation
15
10
degradation, the diameter of a void, shown as a defect type 1 in
10
Fig. 2, is changed from 1 mm to 3 mm in steps of 0.5 mm. The
5 5 depth of the voids is maintained at 2 mm, as defined in the
(b) 5 6 7 8 9 10 11 12 13
characteristics of the defects shown in Table 2. The test voltage
4 (e) for each defect type was varied, in 1 kV steps, up to the rated
Test Voltage (kV) voltage of the cable, 11kV. The values of voltages for which
3
Maximum Value results are reported are listed in Table 2.
2
Median Value Table 2. Characteristics of defect types 6 to 10 and test voltages
1
Diameter of Depth of
5 6 7 8 9 10 11 12 13 Defect Type Testing voltage
void (mm) void (mm)
(c)
Test Voltage (kV) Type 6 1 2 8kV, 9kV, 10 kV, 11kV

Fig. 13. Violin plot of PD Number in 20 ms for defects (a) Type 1, (b) Type 7 1.5 2 10 kV, 11kV
Type 2, (c) Type 3, (d) Type 4, (e) Type 5.
Type 8 2 2 8kV, 9kV, 10 kV, 11kV

Type 9 2.5 2 8kV, 9kV, 10 kV, 11kV

4.3 CASE STUDY: FLASHOVER DURING DEFECT
TYPE 2 EXPERIMENT Type 10 3 2 9kV, 10 kV, 11kV

During a test on defect type 2, after the voltage was

increased to 13 kV, a flashover incident occurred. This caused Violin plots of maximum PD magnitude for different
damage to both the tip of the protrusion and the EPR insulation diameters of void are shown in Fig. 15.
of the cable, as shown in Fig. 14. Analysis of data presented in From Fig. 15, the following observations can be made:
sections 4.1 and 4.2 indicates increasing number of PD, Fig. 13 Generally speaking, although the relationship is not linear, PD
(b), and decrease of PD magnitude, Fig. 12 (b). It is possible magnitude increases with size of void.
that, before serious insulation degradation, the fault is in stage Data analysis, not presented, has shown that there is no
one, which shows a higher PD magnitude and a lower pulse change in the number of PD pulses as the size of the void
number; after the serious insulation degradation, the fault is in changes. For defect types 6 to 10 the number of PD pulses
stage two, with lower PD magnitude and higher pulse numbers. stays at 1 to 5 per 20 ms.
 types 1 to 4 are fewer than 6 in 20 ms; the median
values of PD pulse number for defect types 1 to 4 are
500 1, 2 and 3. Sometimes there is only one negative
PD Magnitude (V)

0.006 400
pulse in 20 ms (Fig. 13 (a) (c) (d)).
0.004 300
• From analysis of PD characteristics prior to the
200
0.002
100
flashover incident, indicator of serious insulation
0 0
degradation of EPR cables might be the rapid
1 1.5 2 2.5 3 1 1.5 2 2.5 3 decrease of PD magnitude (Fig. 12 (b)) and the rapid
(a)
Maximum Value
Diameter of
void (mm)
(b) Diameter of
void (mm)
increase of PD pulse numbers (Fig. 13 (b)).
Median Value
• Analysis of PD characteristics when the size of a
Fig. 15. Violin plot of PD magnitude for a range of void diameters: (a)
HFCT, (b) IEC 60270. void is increased, to simulate insulation degradation,
shows that PD magnitude increases with size of void
but the relationship between PD magnitude and size of
From the analysis in sections 4.1, 4.2 and 4.4 of the paper void is not linear, and there is no change in the number
conclusions may be drawn, i.e. for PD based condition of PD pulses when the size of void is increased.
monitoring of a cable system, the symptoms of serious Based on the work carried out in this paper, 5672 sets of
insulation degradation might be: PD signals and 9687 sets of interference signals detected by
Rapid increase of PD magnitude (Fig. 15) due to changing HFCT have been automatously extracted from raw data.
dimension of the void; 2000 sets of PD signals with known defect type and 2000
Rapid decrease of PD magnitude (Fig. 12 (b)) due to interference signal have been randomly selected for further
changing local stress with defect type; pattern recognition investigation. The results of the study
will be published in a future paper.
Rapid increase in PD number (Fig. 13 (b)) due to variation
in local stress.
ACKNOWLEDGEMENT
5 DISCUSSION AND CONCLUSIONS The authors would like to thank National Natural Science
Foundation of China for funding the research under grants
This paper has presented information gathered from 51541705. The authors also would like to thank EPSRC
experimental work on 5 defects introduced to 11kV single- (UK) for funding the research, under grants EP/G028397/1
phase EPR cable. The defects replicate faults which are and EP/G029210/1.
found in in-service cables. The range of electrical stress
parameters applied to the cable faults is provided. A method
REFERENCES
of comparing IEC and HFCT data (Synchronous Detection
[1] A. J. Hunter, P. L. Lewin, L. Hao, C. Walton, M. Michel,
and Multi-information Fusion) allows signal interpretation
“Autonomous classification of PD sources within three-phase 11 kV
and pre-processing of PD data from the defects. Based on PILC cables”, IEEE Transactions on Dielectrics and Electrical
the comparative analysis of data collected from an IEC Insulation, Vol. 20, No. 6, pp. 2117-2124, 2013.
62070 system and a commercially available HFCT, [2] X. Peng, C. Zhou, D. M. Hepburn, M. D. Judd, W. H. Siew,
“Application of K-Means Method to Pattern Recognition in On-line
characteristics of PD signals are reported. Cable Partial Discharge Monitoring”, IEEE Transactions on
Conclusions are summarized as follows: Dielectrics and Electrical Insulation, Vol. 20, No. 3,pp. 754-761,
2013.
• The proposed SDMF based PD signal identification [3] L. Hao ， P. L. Lewin, “Partial Discharge Source Discrimination
is effective for automatous PD and interference using a Support Vector Machine”, IEEE Transactions on Dielectrics
signal identification and to processing of large and Electrical Insulation, Vol. 17, No.1, pp. 189 -197. 2010.
volumes of experimental PD data if two kinds of [4] A. Rizzi，F. M. F. Mascioli，F. Baldini, C. Mazzetti, R. Bartnikas,
defection system are fully synchronised. “Genetic Optimization of a PD Diagnostic System for Cable
Accessories”, IEEE Transactions on Power Delivery, Vol. 24, No.3,
• PD from EPR cable detected by a commercial HFCT pp. 1728 -1738, 2009.
sensor was of relatively low magnitude, i.e. median [5] H. Ma, J. C. Chan, T. K. Saha, C. Ekanayake, “Pattern Recognition
value of PD magnitude from defect types 1 to 3 was Techniques and Their Applications for Automatic Classification of
Artificial Partial Discharge Sources”, IEEE Transactions on
below 5 mV (Fig. 12 (a) – (c)).
Dielectrics and Electrical Insulation, Vol. 20, No. 2, pp. 468-478,
• The magnitude of PD signals from an IEC 60270 2013.
system was observed to be several hundred times [6] A. J. Mcgrail, E. Gulski, E. R. S. Groot, D. Allan, D. Birtwhistle, T.
greater than those from a commercial HFCT sensor R. Blackburn, “Data Mining Techniques to Assess the Condition of
High Voltage Electrical Plant”, CIGRE Working Group 15.11, 2002.
(Fig. 12). As IEC systems are challenging to install [7] C. Zhou, M. Michel, D. M. Hepburn, G. Zhang, “Rough Set Theory
in practical situations, alternative methods of signal for Data Mining in an On-line Cable Condition Monitoring System”,
processing must be developed and deployed in order CIRED 2007, Vienna, Austria.
to make full use of HFCT technology. [8] J. Tang, F. Liu, X. Zhang, X. Liang, Q. Fan, “Partial discharge
recognition based on SF6 decomposition products and support vector
• Statistical analysis shows that for insulation related machine”, IET Science Measurement & Technology, Vol.6, No.4,
PD, pulse number in a 20 ms period is low. Despite pp. 198-2041, 2012.
insulation breakdown, PD pulse number in defect
[9] L. Li, J. Tang, Y. Liu, “Partial Discharge Recognition in Gas Alistair J. Reid (M’11-SM’15) graduated from
Insulated Switchgear Based on Multi-information Fusion”, IEEE the University of Strathclyde, UK, in 2004 with a
Transactions on Dielectrics and Electrical Insulation, Vol. 22, No. 2, B.Eng. (Hons) degree in Electrical and
pp. 1080-1087, 2015. Mechanical Engineering and received the Ph.D.
[10] S. M. Gargari, P.A.A.F Wouters, P. van der Wielen, E. F. Steennis, degree in 2007 for research on partial discharge
“Partial discharge parameters to evaluate the insulation condition of monitoring. Subsequently, he has worked as a
on-line located defects in medium voltage cable networks”, IEEE Post-Docroral Research at the University of
Transactions on Dielectrics and Electrical Insulation, Vol. 18, No. 3, Strathclyde, a Research Fellow at Glasgow
pp. 868-877, 2011. Caledonian University and a Visiting Research
[11] IEC International Standard 60270, “High Voltage Test Techniques - Associate at California Institute of Technology.
Partial Discharge Measurements”, International Electrotechnical He is presently a Lecturer in the School of Engineering and Built
Commission (IEC), Geneva, Switzerland, 3rd edition, 2000. Environment at Glasgow Caledonian University. His research interests
[12] A. J. Reid, X. Peng, X. Hu, M. D. Judd, W. H. Siew, C. Zhou, D. M. include diagnostic monitoring techniques and systems, partial discharge,
Hepburn, “Comparison Of Partial Discharge Characteristics From sensor design, embedded systems and energy harvesting. Alistair is a
Insulation Defects In 11kV EPR Cable”, Proc. 17th International Senior Member of the IEEE and a committee member of the Universities
Symposium on High Voltage Engineering, Hannover, Germany, High Voltage Network (UHVnet) in the UK.
2011.
[13] C. Zhou, M. Michel, D.M. Hepburn, X. Song, “On-line partial Martin D. Judd (M’02-SM’04) is the Technical
discharge monitoring in medium voltage underground cables”, IET Director of High Frequency Diagnostics Ltd. He
Science Measurement and Technology, Vol.3, Iss.5, pp. 354–363, graduated from the University of Hull in 1985
2009. with a first class (Hons) degree in Electronic
[14] G. C. Montanari, A. Cavallini, “A New Approach to Partial Engineering, after which he gained 8 years of
Discharge Testing of HV Cable Systems”, IEEE Electrical Insulation industrial experience, first with Marconi
Magazine, Vol. 22, No. 1, pp. 14-23, 2006. Electronic Devices and then with EEV Ltd.
[15] M. Correll, M. Gleicher, “Error Bars Considered Harmful Exploring Martin received his PhD from the University of
Alternate Encodings for Mean and Error”, IEEE Transactions on Strathclyde in 1996 for research into the
Visualization and Computer Graphics, Vol. 20, No. 12, pp. 2142- excitation of UHF signals by partial discharges in
2151, 2014. gas insulated switchgear. He has worked
extensively on UHF partial discharge location techniques for power
Xiaosheng Peng (M’11) received the B.Sc. and transformers and was latterly Professor of High Voltage Technologies at
M.Sc. degrees from Huazhong University of the University of Strathclyde, where he managed the High Voltage
Science and Technology, China in 2006 and 2009, Research Laboratory. In 2014 he founded High Frequency Diagnostics, a
respectively, and the Ph.D. in electrical engineering contracting and consultancy business that works in partnership with
at Glasgow Caledonian University in 2012 funded companies and universities to maximize the impact of R&D outputs by
by EPSRC. He has worked as a Post-Doctoral utilizing them in new technologies and applications.
Researcher in Glasgow Caledonian University
funded by EDF Energy. He is currently a lecturer in W. H. Siew (M’95) is a Reader in the Department
School of Electrical and Electronic Engineering of of Electronic & Electrical Engineering, University
Huazhong University of Science and Technology. of Strathclyde, Glasgow, Scotland. He is a triple
His research interests are partial discharge signal processing and condition alumnus of the University of Strathclyde with
monitoring of power plant. He is a member of IET. degrees of B.Sc. (Hons) in electronic & electrical
engineering; Ph.D. in electronic & electrical
Chengke Zhou (M’06) received the B.Sc. and engineering; and Master of Business
M.Sc. degrees in electrical engineering from Administration. His areas of research interest
Huazhong University of Science and Technology, include large systems electromagnetic
China in 1983 and 1986, respectively, and the Ph.D. compatibility; cable diagnostics; lightning
degree from the University of Manchester U.K., in protection; and wireless sensing systems. He is Convener of the CIGRE
1994. Since then, he worked in Glasgow WG C4.208 and a member of the Technical Advisory Panel for the IET
Caledonian University, U.K., as a Lecturer, Senior Professional Network on Electromagnetics. He is also a member of IEEE
Lecturer and in Heriot-Watt University as a Reader TC7. He is a Chartered Engineer and an MIEE.
until 2007 when he went back to Glasgow
Caledonian University as a Professor. He has
published more than 100 papers in the area of partial discharge based
condition monitoring of MV/HV plant and power system analysis. He is
member of IET.

Donald M. Hepburn received his B.A. (Hons)

from the Open University in 1987 and the Ph.D.
degree from Glasgow Caledonian University
(GCU) in 1994. He has many years of industrial
research experience and has been involved in
research into high voltage insulation systems at
GCU for over 20 years. His research interests
cover monitoring of chemical changes to
insulation materials, application of electrical,
acoustic and RF monitoring equipment to HV
components and application of advanced digital signal processing to
information from the monitoring techniques. He is a Senior Lecturer at
GCU and is involved in industrial and academic research projects.

View publication stats