C C I I R R E E D D 18

lr
lrlerral|ora| Corlererce or E|eclr|c|ly 0|slr|oul|or Tur|r, ê-9 Jure 2005

ClRE02005

3ess|or No 1

LIMITATIONS IN THE APPLICATION OF ON-LINE AND OFF-LINE PD MEASUREMENT
SYSTEMS

Ian COTTON, Vincent O’DONNELL, Nick CHRISTOFIDES
University of Manchester, United Kingdom
ian.cotton@manchester.ac.uk


INTRODUCTION

The first underground cables to be installed in the UK were
paper-insulated cables and their installation dates back as
early as the 1920’s. Although the manufacturing
technology was primitive compared to that available today,
there are still 50-60 year old cables that remain in working
condition. In the modern power system, these cables are
both an asset and a liability. Failure can lead to
interruptions in supply and customer minutes lost, indices
used by the regulator to reward distribution network
operators. Companies are therefore under pressure to
somehow improve the performance of cable assets.
However, the economic costs relating to the replacement of
all cable assets at the end of their life are prohibitive and
strategies to determine the relative health of cable assets
are becoming increasingly important.

While a number of cable failures are caused by third-party
damage and cannot be predicted, a variety of partial
discharge testing techniques can be employed on a power
cable in an attempt to characterise its ‘health’. Partial
discharges (PD) are small electrical breakdowns within an
insulation system that partially bridge the space between
two regions at different potentials. PD activity can
eventually degrade the insulation of a cable to such an
extent where it can no longer withstand the electric stress
across it. At a basic level, cables with a high PD activity
can be classified as being more at risk of failure in
comparison to cables in which no PD activity can be
detected.

Both on-line and off-line partial discharge measurement
systems can be used to help characterise the health of a
power cable, in particular ones constructed with paper
insulation. However, the way in which signals propagate
through these cables, the influence of joints, the
termination techniques and ambient noise levels at the
substations all influence the levels of partial discharge
signal that can be measured.

This paper discusses a number of practical problems that
will be faced in the implementation of partial discharge
measurement systems on cable networks. The basic means
by which pulses propagate through a cable system are
explained along with the effect of joints and variable
impedance terminations. The level of detectable partial
discharge is shown to strongly correlate with the location
of the defect site in relation to the substation and the
measurable noise level. The impact of the use of different
types of transducers to measure the partial discharge
signals is also discussed.

OFF-LINE / ON-LINE TECHNIQUES

Methods of PD testing can be split into two main types,
off-line and on-line. Off-line testing has traditionally been
used to provide information about partial discharge
location along a cable using time domain reflectometry
(TDR). Off-line testing can be carried out using 50Hz
generators although these have to supply considerable
capacitive current to the cable and are therefore extremely
large. Very-low frequency (VLF) allows the cable to be
energised at 0.1Hz therefore preventing any large
capacitive current from being drawn and minimising the
size of the test set. While off-line testing is a valuable
resource, it is costly in terms of manpower, results in a loss
of power system security while a circuit is taken off-line
and can only take a snapshot of the cable health.

In contrast, on-line detection can monitor the health of a
cable for prolonged periods and can be achieved by fitting
current transformers or other transducers around earth
straps connected to the sheath of a cable. There is no
system security risk associated with on-line testing but it
can be susceptible to noise and location is more difficult.

Whatever the method of testing, the way in which PD
pulses propagate through a power cable play an important
role in the accuracy of the measurements. Attenuation or
distortion of pulses as they propagate along the cable will
alter the signal received at the measuring equipment.
Amongst other issues, this can lead to the apparent
presence of high levels of partial discharge close to the
measurement site and lower levels further away while in
reality the original magnitudes may be reversed. Further
problems result from the consideration of the influence of
joints, ring-main units and cable tee-sections.

PROPAGATION OF HIGH FREQUENCY SIGNALS
THROUGH A CABLE SYSTEM

Basic Principles

A PD occurring between phase and earth in a power cable
results in a flow of current on both the phase conductor and
the power cable sheath. The maximum duration of the PD
current resulting from the breakdown of small voids is at
most a few nanoseconds [1]. This type of pulse contains
significant high frequency components. When the current
pulse is formed, it splits and travels in two opposite
C C I I R R E E D D 18
lr
lrlerral|ora| Corlererce or E|eclr|c|ly 0|slr|oul|or Tur|r, ê-9 Jure 2005

ClRE02005

3ess|or No 1
directions towards each end of the cable (used as an
advantage in off-line TDR techniques). An equal and
opposite PD current will be found on any location of the
power cable sheath and conductor at any moment in time.
The magnitude and shape of the current at any location
varies due to the behaviour of the cable as a low pass filter.

A simplistic high frequency model of the cable illustrates
why this behaviour takes place. Both the conductor and the
sheath consist of resistance and inductance. At high
frequencies, the series impedance of this path is high. The
insulation between the conductor and the sheath can be
modelled as capacitance and resistance and therefore has
low impedance at high frequencies. The high frequency
components of the discharge do not propagate great
distances down the power cable but flow between the
conductor and the sheath close to the discharge site.

Using the Telegrapher’s equations, a relationship between
the current and the voltage entering and leaving a cable
section of length ‘d’ can be given by equation 1 [2] where
¸
c
is the propagation constant and Z
c
is the characteristic
impedance. Through the use of Fourier transforms, the
voltage and current at any point along a cable system
owing to the input of a specific pulse can therefore be
calculated.

|
|
.
|

\
|
|
|
.
|

\
|
=
|
|
.
|

\
|
) (
) (
) cosh( ) sinh(
) sinh( ) cosh(
) 0 (
) 0 (
d !
d v
d Z d
d Z d
!
v
c
c
c c c
c c c
c
c
¸ ¸
¸ ¸
(1)

Attenuation Of PD Pulses Along A Uniform Cable

A fast partial discharge pulse is rich in both high and low
frequency content when it is represented in the frequency
domain. However, the high frequency content cannot
propagate through the cable for reasons already described
and is increasingly attenuated as a function of cable length.

Both the magnitude and the shape of the pulses therefore
change as they propagate with a finite velocity along a
cable.

Figure 1 shows the results of a simulation in terms of the
peak voltage produced by a current transducer placed at a
substation when a 20000pC PD pulse originates at a PD
site in a cable (i.e. 10000pC propagates to the substation
and 10000pC proceeds in the opposite direction down a
semi-infinite cable length). As is expected, the magnitude
of the voltage measured at the CT decreases as the distance
the PD must propagate to the substation increases. The
most important consequence of this decrease in partial
discharge magnitude is the ultimate reduction in the output
of the CT to below the noise level present within the
measuring environment.

The figure clearly shows the need to correct measured
partial discharge pulse levels as a function of cable length.
A large PD a long distance away could produce a smaller
signal at the substation in comparison to a small PD a short
distance away. If partial discharge maps such as those
produced in VLF testing are to be interpreted accurately,
correct modification of the plots is important. The result is
significant for both on-line and off-line testing.
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Cable Length
P
e
a
k

V
o
l
t
a
g
e

O
u
t
p
u
t

f
r
o
m

C
T

[
V
]

Figure 1 – Measured signal versus distance

With any form of measurement, but particularly for on-line
techniques, continuous ‘white’ noise-levels of at least a few
millivolts can be expected within a substation environment
although transient power electronic noise present on cable
sheaths can be magnitudes higher than this level. The
consequence of this noise is to limit the level of PD that
can be detected at the substation. Figure 2 shows the
minimum PD levels required to cause triggering of a
detection circuit at a substation when the noise level is
50mV or 75mV. It can be seen that for a cable length of
2km and a noise level of 50mV, a 6000pC pulse would be
the minimum required for detection.
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
Cable Length
M
i
n
i
m
u
m

d
e
t
e
c
t
a
b
l
e

P
D

l
e
v
e
l

[
p
C
]
minimum PD level needed for 50mV noise
minimum PD level needed for 75mV noise

Figure 2 - Minimum detectable PD level[V2]
Attenuation Of Pulses Through A Mixed Cable
Network

A common and necessary practice in distribution circuits is
to join cables of different size and type. Different sizes are
found where repairs have been carried out or where cable
sizes are simply tapered along a radial length containing
tees. With most utility companies now preferring XLPE to
C C I I R R E E D D 18
lr
lrlerral|ora| Corlererce or E|eclr|c|ly 0|slr|oul|or Tur|r, ê-9 Jure 2005

ClRE02005

3ess|or No 1
paper cables, it is also very common to find paper and
XLPE cables connected together.

With the surge impedance of cables varying according to
size and insulation type, discontinuities in PD propagation
due to impedance mismatch are expected. Travelling wave
theory can be used to calculate the transmission and
reflection coefficients that indicate the relative proportions
of the pulses transmitted or reflected at an interface made
up of two different size cables. The change in pulse
magnitude due to a cable interface and the attenuation
along the cable length must be known to be able to find the
original magnitude of a PD pulse.

Equation (2) calculates the current transmission coefficient
for a 240 to 95mm
2
cable interface. In terms of pulse
current, it shows that only about 80% of the pulse
magnitude entering the interface will be transmitted to the
remaining 95mm
2
. For an interface made up of 95mm
2
to
240mm
2
cable, 120% of the pulse would be transmitted to
the 95mm
2
cable as shown in (3).

0.778
3 3 21
21 * 2
* 2
2 2
2
95 240
240
95 240
=
+
=
+
=
÷
mm mm
mm
Z Z
Z
| (2)
222 . 1
33 1 2
33 * 2
* 2
2 2
2
95 240
95
240 95
=
+
=
+
=
÷
mm mm
mm
Z Z
Z
|
(3)
Attenuation Of Pulses Due To A Teed Connection
(Cable Or Transformer)

When a partial discharge pulse propagates past a teed
connection, the pulse will change in magnitude and
propagate down both paths. Two specific situations can be
considered, one where the tee consists of a cable of
significant length and one where the tee consists of a
transformer connected by a short cable (the case found in a
typical UK urban system using ring main units).

Taking the simple situation of one cable splitting into two,
for a current of 1pu magnitude flowing into the junction, a
current of 0.67pu will flow into each of the two cables. The
significant attenuation of the pulse therefore further
restricts the distance that a specific sensor can ‘see’ down a
cable circuit when the noise level is accounted for. If the
teed-cable is short in length (and therefore has a low travel
time), there is also a possibility that a reflection from the
end of the cable will cause distortion in the magnitude and
shape of the pulse attempting to be measured.

Where a cable circuit is connected through a ring main unit
and there is a short piece of cable terminated with a
transformer, there is not likely to be a significant effect
since for most frequencies the transformer will appear as an
open-circuit. The impedance of an 11kV/415V transformer
measured phase-to-earth is shown in Figure 3. The
impedance does not reduce below that of typical cables (20
to 30O) until a frequency of some 300kHz. For a ring main
unit consisting of a long cable and a transformer, the cable
alone will dominate the effect on the partial discharge
pulse.

1
10
100
1000
10000
1.0E+02 1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07
Frequency [Hz]
M
a
g
n
i
t
u
d
e


Figure 3 – Transformer open circuit impedance

IMPLICATIONS FOR ON-LINE AND OFF-LINE
MEASURING SYSTEMS

Off-Line Systems

In the typical off-line measurement of partial discharge
using the time domain reflectometry technique, it is usual
to plot a graph of pulse magnitude versus distance. The
information presented above implies the following
limitations in this technique:

- An error in the absolute magnitude of the partial
discharges will result from attenuation of the pulses.
Where multiple partial discharge sites are present in the
cable circuit, it would not be possible to judge relative
severities unless a suitable correction was made.

- When long cable circuits are tested, small partial
discharge pulses may not be detected. Testing from
both ends would help eliminate this problem. With an
off-line system this implies the circuit will be
disconnected from the power system for a longer period
meaning more risk to the system security and higher
man-hours of labour.

- For long cables, the use of time domain reflectometry
would be difficult even if partial discharges can be
measured. This is because of the increased attenuation
of the reflected pulse that has, as a minimum, got to
travel the full length of the cable circuit and sometimes
almost double the length before it is detected.

On-Line Systems

For on-line systems utilising current transducers placed
onto the earthed sheath of cables, the ambient noise level is
likely to be significantly higher than the off-line case.
There is also more likely to be an impact of power
electronic / switching noise, transient in nature but high in
magnitude, on the measurement system. The limitations
C C I I R R E E D D 18
lr
lrlerral|ora| Corlererce or E|eclr|c|ly 0|slr|oul|or Tur|r, ê-9 Jure 2005

ClRE02005

3ess|or No 1
are, however, similar to those identified for the off-line
system:

- An error in the absolute magnitude of the partial
discharges will result from attenuation of the pulses.
Where multiple partial discharge sites are present in the
cable circuit, it would not be possible to judge relative
severities unless a suitable correction was made (as
before).

- When long cable circuits are tested, small partial
discharge pulses may not be detected. Testing from
both ends would help eliminate this problem. With an
on-line system, the only real impact of this is on the
initial installation and regular maintenance costs of the
measurement system. It is unlikely that such a system
would be applicable to XLPE cables as typical
detection levels would not be low enough.

- If localisation of the partial discharge pulses is being
attempted by examining the shape of the pulse, shape
distortion due to reflections or noise will be
problematic.

POSSIBLE SOLUTIONS

Matched Filtering

Detection of PD and location of the source under noisy on-
line conditions has previously been shown to be improved
by matched filtering [3]. A single filter is considered
matched to a particular waveform if its impulse response is
that waveform reversed in time. A bank of filters matched
to different waveforms can be used to optimally detect
which specific waveforms are present in a noisy signal.
Using models of PD pulse propagation through different
lengths of distribution cables [4], a matched filter bank can
be constructed with different waveforms for different PD
source locations covering the full length of a cable. Some
sample model pulses are shown in Figure 4. Clearly
received propagated PD become more alike the further
along the cable they have travelled and in the presence of
noise, the filter bank will select a variety of possible
locations with the trend indicating the most persistent PD
source.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
x 10
−6
−0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
S
c
a
l
e
d

f
o
r

U
n
i
t

E
n
e
r
g
y
Time (s)
100m
200m
500m
1000m
2000m

Figure 4 - Modelled propagated PDs
A typical noisy signal segment is shown in Figure 5.


0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
−0.02
−0.015
−0.01
−0.005
0
0.005
0.01
0.015
0.02
N
o
i
s
y

S
i
g
n
a
l

(
V
)
Time (s)

Figure 5 - Typical noisy PD signal capture (40ms).

A typical scatter plot combining matched filter detected PD
pulse-shapes from successive measurements is shown in
Figure 6. This shows a range of pulse shapes and relative
amplitudes are present on the circuit.
0 500 1000 1500 2000
0
0.01
0.02
0.03
0.04
0.05
0.06
ONLINE LOCATION: 907 m
M
a
t
c
h
e
d

F
i
l
t
e
r

O
u
t
p
u
t

P
e
a
k

L
e
v
e
l
Location (m)

Figure 6 - On-line localisation by matched filter

A typical detected pulse is shown in Figure 7 along with
the best matching propagated model PD waveform scaled
to show best, the similarity.

500 1000 1500 2000
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
8.74 8.745 8.75 8.755 8.76 8.765 8.77 8.775 8.78 8.785
x 10
−4
−2
0
2
4
6
x 10
−3
S
i
g
n
a
l

(
V
)
Time (s)
actual signal samples
900 m PD model

Figure 7 - Typical detected and matching model pulses

The matched filter output can be conveniently shaped as a
sequence of detected PD pulses. The basic elements of such
a sequence are pulse locations (i.e., indexes for the best
matching pulse shapes, e.g., 250m), pulse arrival times
C C I I R R E E D D 18
lr
lrlerral|ora| Corlererce or E|eclr|c|ly 0|slr|oul|or Tur|r, ê-9 Jure 2005

ClRE02005

3ess|or No 1
(i.e., arrival samples relative to start of longshot, e.g.,
sample 35,343), pulse ‘magnitudes’ (i.e., with some
calibration, the estimate of the pulse magnitude scalings).

Other useful information could be bundled into the
matched filter output. For example, actual noisy samples of
each detected pulse (for further analysis, noise rejection
etc.) or other measurements derived from these (e.g., time
and frequency standard deviations [5]).

Matched filter output PD sequences can be used for
investigations into separating multi-source PD, classifying
incident and reflected pulses (with TDR), fault location by
pulse shape, fault type vs PD phase (of power frequency),
relative distribution of positive and negative pulses, pulse
magnitudes, correlation to cable temperature, pulse
repetition rates, eliminating PD-like disturbances and so
on, tracking the various features over time. In this way, the
matched filter output can support the basic efforts to relate
PD measurements to assessment of condition of cable and
accessories.

CONCLUSIONS

This paper has described how on-line and off-line methods
used to detect partial discharge signals in paper cables can
be affected by the circuit characteristics and the presence of
electrical noise. The effect of attenuation down a length of
cable, the impact of transformers and of teed cable sections
has been described.

It is important to correct partial discharge activity measured
at a substation to account for attenuation and to therefore
give a truer picture of the real problems the cable may
have. The use of matched filters to detect partial discharge
signals within noisy environments has also been described.

REFERENCES

[1] E. Kuffel, W.S. Zaengl, J. Kuffel, High Voltage
Engineering: Fundamentals, 2
nd
edition, Newnes, 2000

[2] J. Veen, P.C.J.M. van der Wielen and E.F. Steennis,
“Propagation Characteristics of three-phase Belted Paper
Cable for On-line PD Detection.”, Conference Record of
the 2002 IEEE lntemational Symposium on Electrical
Insulation, Boston, MA USA, April 7-10,2002

[3] J. Veen and P. C. J. M. van der Wiellen, "The
application of matched filters to PD detection and
localization," Electrical Insulation Magazine, IEEE, vol.
19, pp. 20-26, 2003.

[4] N. Christofides, I. Cotton, C. Walton, and M. Michel,
"Partial discharge propagation in belted paper power
cables," presented at International Symposium On High
Voltage Engineering, Delft, 2004.

[5] A. Cavallini, G.C. Montanari, A. Contin, and F. Puletti,
“A New Approach to the Diagnosis of Solid Insulation
Systems Based on PD Signal Inference”, IEEE Electrical
Insulation Magazine, vol. 19, no. 2, pp. 23-30, 2003.