You are on page 1of 6

2006 IEEE PES Transmission and Distribution Conference and Exposition Latin America, Venezuela

Partial Discharge Analysis and Asset


Management: Experiences on Monitoring of
Power Apparatus
A. Cavallini, Member, IEEE, G.C. Montanari, Fellow, IEEE, and F. Puletti

Index Terms--Asset management, Monitoring, On-line tests,


Partial discharges, Reliability, Risk assessment.

I. INTRODUCTION

WO fundamental drivers must be taken into account when


dealing with electrical asset management: the need of cost
reduction and that of availability (and reliability)
maximization. Actually, asset management would greatly
benefit, from both an economical and a technical point of
view, as well as from the opportunity to plan maintenance
actions according to the real state of the asset. Under these
premises, Condition Based Maintenance (CBM) practices
seem to be the most promising and performing tool to support
asset management decisions. However, in order to carry out
proper CBM actions, effective diagnostic techniques aimed at
providing reliable information for the assessment of ageing
and performance (conditions) of the electrical asset must be
employed.
As a matter of fact, failure risk estimation for a certain
electrical apparatus is, in general, a complex task. For
example polymeric insulation cable systems have been
observed to experience sudden failures even if their overall
A. Cavallini is affiliated with the Department of Electrical Engineering,
University of Bologna, Italy (e-mail: andrea.cavallini@mail.ing.unibo.it).
G. C. Montanari is affiliated with the Department of Electrical
Engineering,
University
of
Bologna,
Italy
(e-mail:
giancarlo.montanari@mail.ing.unibo.it).
F. Puletti is affiliated with TechImp Systems, Bologna, Italy
(fpuletti@techimp.com)

1-4244-0288-3/06/$20.00 2006 IEEE

condition had been considered satisfactory (i.e. weak bulk


ageing or no ageing at all). This is mainly to be traced back to
the presence of weak points (defects) able to trigger localized
degradation phenomena. Therefore, the possibility to verify
the presence of localized defects is fundamental for electrical
insulation systems availability and reliability improvement.
Partial Discharge (PD) testing and analysis have proven to be
the most effective way to evidence the presence of local
degradation mechanisms. Nowadays, PD testing is
successfully employed (a) during quality control of electrical
systems to check for manufacturing problems, (b) during
commissioning tests to verify installation quality, (c) during
equipment life as a condition based maintenance tool to derive
information about the state of an electrical asset. In this paper
an innovative and improved approach to partial discharge
detection and analysis is presented, and an example of its
application to the continuous monitoring of an electrical asset,
i.e., a high voltage cable system is reported and commented.
II. PD INVESTIGATION APPROACHES: PROS AND CONS
PD measurements have historically been employed in the
laboratory for quality control of electrical apparatus, with the
aim of highlighting the presence of defects due to
manufacturing processes. Lately, thanks to improved detection
and analysis capabilities of PD diagnostic instruments and
procedures, PD have started to be widely applied on site, with
the purpose of both detecting installation problems
(commissioning tests) and assessing the condition of electrical
systems by inferring the presence of on-going local
degradation mechanisms (diagnostic tests). As far as on site
PD investigation is concerned, an evaluation must be carried
out about the possible different testing approaches in order to
highlight their effectiveness and applicability as well as their
cost/performance ratio.
As regards on-site PD investigations, these can be carried
out by on line or off line procedures. Off line tests are carried
out by disconnecting the asset from the grid and energizing it
through a mobile test set. This offers several advantages but
has also some drawbacks. If the capacitance of the asset is
large (typically, large cable systems, generators), it is very
difficult to carry out tests at industrial frequency. Therefore,
alternative energization methods have been introduced, such

as Damped AC Voltages (DAS), variable frequency resonant


test set, Very Low Frequency (VLF) generators [1]. Testing
with frequencies that differ too much from the industrial
frequency may however provide results that do not reproduce
what happens in real conditions, since the physics of PD is
strongly influenced by supply frequency [2]. In addition, off
line test circuits may provide inherently an electric field
profile inside the insulation system that is remarkably
different from that experienced in service. For instance,
supplying a generator winding with open star connections
during an off line tests creates an electrical field distribution
which is very different from the stress present during machine
operation. Moreover, the overall condition of the system
(temperature and humidity profiles, vibrations, clearances,
etc.) during off line tests may be very different from those
experienced in service conditions. As a consequence, there is
a risk that off line testing activates phenomena that do not
occur during service (e.g., PD from cavities near the star point
in rotating machines). On the contrary, PD sources that are
active during service may not be activated during off line
tests.
On the other hand, off line investigation proves usually to
be very effective as far as measurement sensitivity is
concerned. In particular, by stressing and testing one phase a
time, cross talk phenomena can be eliminated. By using test
ring electrodes and other suitable devices, corona and
interference from the grid can be minimized. Moreover,
trough off line testing procedures is it possible to measure
inception and extinction voltage of PD phenomena and
evaluate the PD behaviour at various voltage levels, thus
improving defect characterization. Therefore, it can be
speculated that off line tests pose less difficulties for noise
rejection and PD phenomena identification.
On line tests are characterized by advantages and
disadvantages that are, in some sense, complementary to those
relevant to offline testing. In particular, tests are carried out at
operating conditions; therefore, all and solely the criticalities
that may affect the asset during operations are detected and
analyzed. However, noise and pulsed interference from
neighbouring HV apparatus may affect PD readings,
complicating interpretation and diagnosis. In addition, PD
inception and extinction voltages cannot be evaluated.
Both kinds of investigations are nevertheless used and can
claim several successful applications. But while off line tests
are considered a necessary and ideal practice for the
commissioning of HV cable systems, the market is generally
looking at on line applications with growing interest,
particularly for periodic diagnostic investigations during
equipment life. In fact, off line testing are more expensive
than on line ones and force an outage of the asset. Actually,
diagnosis by on line measurements has already become a
common practice for rotating machine [3].
Another issue to be considered is the fact that insulation
systems are subjected to changes during time under stress. For
instance, thermal cycling and vibration during operation often
induce interface delaminations. Chemical contaminants,

nuclear radiation or lightning overvoltages may also initiate


defects. Particularly for organic insulation systems, stressinduced defects may disrupt insulation system integrity in a
relatively short time. Hence, it is important to evaluate how
frequently PD assessments should be carried out in order to
minimize failure risk. In general, the more frequent the
assessment the more the chances to detect a failure mechanism
at an early stage. Thus, when dealing with a fundamental asset
of the grid, as well as when a failure would have unacceptable
consequences in terms of unavailability and costs, the option
of a continuous assessment of PD might be considered. In
other words, a correct and serious approach towards asset
management may call for the use of permanently installed
monitoring systems on those assets for which failure risk must
be minimized.
However, monitoring of electrical assets poses other
concerns. Costs are sometimes deemed to be relatively high,
particularly for cable systems, while efficient solutions are
needed to automatically reject noise and handle the large
amount of data generated by monitoring systems. In
particular, proper automatic identification devices must be
used, able to provide suitable alerts in case internal PD
activities are detected. Finally, an ideal solution would also
include a combined evaluation of PD and other quantities
directly or indirectly related to aging and degradation of
electrical systems as, e.g., vibration patterns for rotating
machines, Dissolved Gas Analysis (DGA) for transformers
and core temperature for cables.
The above issues are addressed by the monitoring system
presented in this paper, which takes advantage of an
innovative detection and analysis approach which enables to
successfully reject noise and distinguish among different kind
of PD. These features enable to follow the evolution of those
activities which may pose a threat on the electrical system.
III. PD DETECTION AND ANALYSIS
For several reasons, interpreting PD phenomena in power
apparatus can become an hard task. On one side, pulses due to
noise and/or multiple PD phenomena can be recorded,
providing a pattern that is the superposition of several
contributions. On the other side, PD phenomena are
sometimes complex to identify since several parameters that
affect PD (e.g., the ratio between applied and inception
voltage, cavity size, humidity, etc.) are unknown. To
overcome these difficulties, the diagnostic system proposed in
this paper is realized through (a) an innovative signal
processing hardware aimed at pulse separation, capable of
separating PD from noise, disturbance and PD due to different
sources, (b) fuzzy logic artificial intelligence (FLAI) routines
that provide an information about the nature of PD sources. In
the following, a brief description of the system will be
provided, while a thorough description can be found in [4-7]
Signal processing approach: separation
The pulses arriving at a PD detector input are normally
characterized by a large variance in waveforms. Indeed, noise

pulses are, generally, completely different from PD pulses.


Moreover, due to attenuation and dispersion phenomena, PD
pulses coming from different sources can be remarkably
different. To exploit this information, normally lost in
conventional detectors, an innovative ultra bandwidth
hardware was developed. The hardware is capable to acquire
PD pulses with a bandwidth of 40 MHz, and sampling them at
a rate of 100 MS/s. Pulse acquisition is based on trigger
conditions: as soon as the signal at the detector input exceeds
a predefined level, acquisition is started. Acquisition is
stopped after a predefined time.

generally available on several parameters that directly affect


PD activity. As an example, PD activity is very sensitive to
the overvoltage ratio (i.e., the ratio between applied and
inception voltage), which is generally unknown during online
tests. In order to cope with these limitations, the identification
system was designed using fuzzy logic techniques. Fuzzy
logic can deal with inherently ambiguous information by
assigning, with different likelihoods (or membership values),
a PD source to several categories simultaneously [7].
Pulse type B

F
b1)

b2)

a) Recorded PD pattern

T
Fig. 1. Example of projection of detected pulses into the T-F map (one fast
pulse with high frequency content and one slow pulse with small frequency
content).

After completing an acquisition, while the input stage


resets for a new acquisition, a digital signal processor (DSP)
unit evaluates the features of the most recently-recorded pulse
and stores it for further processing. In particular, the DSP unit
evaluates, for each acquired pulse, polarity, peak value, and
equivalent timelength, T, and bandwidth, W. These two latter
quantities provide a synthetic description of the pulse
waveform in the time and frequency domain, and can be used
to separate pulses having different shapes [4]. As an example,
Fig. 1 shows how fast and slow pulses are characterized
by different T and W values.
Using this signal processing scheme, a large amount of PD
pulses can be recorded and processed, thus allowing for
denoising and for separation of PD coming from different
sources. In particular, TW values for each pulse are plotted on
a Cartesian plane (TW map). Groups of pulses having
different shapes can be separated by recognizing different
clusters in the TW map. Clustering routines can provide an
unsupervised classification of pulses, thus performing
separation automatically. An example of pulse separation is
shown in Fig. 2: after acquisition, it is possible to recognize
that there exist two cluster of pulses characterized by different
TW values. If separation according to TW values is performed,
noise can be removed from the pattern, focusing on PD pulses
only. In a similar way, pulses due to different PD sources can
be separated, allowing to obtain several sub-patterns [6].
Fuzzy logic artificial intelligence routines: identification
Separation is of paramount importance for PD source
recognition since it enables to focus on the pattern created by
a single PD source at a time. However, PD identification still
remains a complex task since incomplete information is

Cluster B

Pulse type A

Cluster A

c) T-W map

d1)

Sub-Pattern B

d2)

Sub-Pattern A

Fig. 2. Example of a typical PD pattern (containing peak, phase and number of


PD), detected on a cable system, where noise and PD pulses are overlapped
(a), relevant pulses (b1 and b2, for PD pulse and noise, respectively), clusters
in the T-W map (c: B and A group PD and noise, respectively), and subpatterns obtained through separation of pulses in the T-W map (d1 and d2)
achieved by means of a fuzzy classification algorithm.

The FLAI routines used in the identification system are


designed according to a three-stage tree-like structure. At the
1st level, identification routines assign the defect to three PD
source macrocategories (internal, surface, and corona), to
noise or invalid data (e.g., PD not acquired correctly). This
identification is very reliable and provides, in most cases, the
fundamental information to carry out (if necessary)
maintenance operations. By proceeding to further levels, more
specified (but, sometimes, less reliable) information can be
found. Thus, the 2nd identification level provides additional
information about internal defects, i.e., defect position with
respect to electrodes, presence of electrical treeing [8]. The
information about treeing activity is particularly important to
carry out maintenance/replacement operations in organic
insulations as, e.g., polymeric cable systems, since after
treeing initiation the insulation system residual life will be
very short. Eventually, the 3rd level identification routines
assign the PD activity to macrocategories that are specific to a

In the following, an example of installation and operation


of a monitoring system on a HV cable link is presented. A
permanent monitoring system was installed in order to assess
PD activity and its possible time evolution in a 400 kV cable
system including two terminations (one outdoor and one
indoor) and two joints. PD signals were detected from the
terminations by means of High Frequency Current
Transformers (HFCT) installed around the grounding leads of
the two terminations and from the joints through internal
capacitive taps. The monitoring system consisted of:
1.
a PD detector working according to the approach
described in Section III,
2.
an industrial PC aimed at supervising the acquisition
and analysis phase, while providing connectivity tools.

550

6.0

PD detector full scale = 500 mV


500

5.5

450

5.0

400

4.5

350

4.0
Amplitude
Frequency

300

Frequency (MHz)

IV. MONITORING OF A POWER CABLE SYSTEM

characterized by pulses having an average equivalent


frequency of roughly 4.5 MHz and reached very high values
of amplitude (more than 1 V). Moreover, its phase position
was typical of discharges occurring at insulator/air interface
(i.e., centred around 90 and 270 degrees) [10]. Thus, the
system was trained to recognize this phenomenon on the basis
of criteria based mainly on PD phase and amplitude values
and provide the system manager with a Yellow (not severe)
alert, since this phenomenon was obviously not harmful for
the insulation system. It is noteworthy that, given the marked
dependence of this phenomenon on weather conditions,
appropriate interaction between the PD detector and a weather
station could further help in strengthening identification,
avoiding false positive detections.

Amplitude (mV)

given type of apparatus. As an example, for rotating


machines, the following defects are recognized by the
identification system: PD in microvoids or delaminations, slot
PD, stress grading PD, bar-to-bar or bar-to-ground PD [9]

3.5

250

3.0
0

Monitoring time (days)

Fig. 4. Behavior of PD pulse magnitude and frequency content for a PD


phenomenon (surface/corona) external to the cable.

Fig. 3. Example of a typical PD pattern relevant to surface/corona PD


occurring on the external surface of the outdoor termination.

Connectivity is an important issue in PD monitoring systems.


Indeed, improperly designed data streams are prone to be too
large to be transmitted and checked accurately. For the system
described here, the PC could be controlled through a
telephone line modem and send warning SMS through a GSM
modem. After an initial training phase, the system was
normally operating unsupervised, being warning SMS the
preferred way to get information in real time. If desired, e.g.,
upon receiving a warning SMS, the operator could also
connect to the PC in order to download historical data, check
the obtained results and, if needed, change acquisition settings
and warning strategies to improve detection and processing
effectiveness.
As regards test results, during the monitoring period
surface/corona phenomena occurring on the outdoor
termination were detected. An example of the pattern relevant
to this phenomenon is reported in Fig. 3. By analyzing pattern
quantities (phase, amplitude variance, etc.) the time behaviour
of this phenomenon it can be observed that it was related to
atmospheric conditions. In fact, it incepted almost regularly
during nights and raining days.
As can be observed from Fig. 4 this activity was

Fig. 5. Example of a typical PD pattern relevant to the internal PD


phenomenon occurring in the defective joint.

Besides the surface activity described above, an additional


PD intermittent activity incepted during the monitoring period
in one of the two joints. An example of the typical pattern of
this activity at a very early stage detected from the capacitive
tap of the defective joint is reported in Fig. 5.
This activity was characterized by pulses having an
average equivalent frequency of roughly 8-9 MHz and low
values of amplitude. This pattern was recognized as a defect
internal to the insulation system by the expert system of the
monitoring tool (FLAI). Therefore, it activated the warning
system that broadcasted a Red (severe) alert signal, indicating
the presence of PD in the cable system. The differences in the
frequency content of the discharge pulses relevant to the two
phenomena (internal PD in the joint and external PD in the

outdoor termination) allowed to identify two separated areas


of the classification map, so that the system could
automatically separate the two activities and process them
accordingly. An example of the simultaneous presence of the
two activities and the corresponding separation is reported in
Fig. 6.
As mentioned before, the internal PD activity was
characterized by an intermittent occurrence in the first days.
Then, it stabilized and started to increase in amplitude, until
breakdown of the joint occurred. The time behaviour of
amplitude and frequency content of the pulses of the internal
phenomenon is reported in Fig. 7
It might be noted that the amplitude of the internal PD
activity in the period close to breakdown reached a significant
level, which might have warned the asset manager to take a
corrective action. However, PD magnitude levels prior
breakdown were also comparably lower than those associated
to external surface/corona phenomena (i.e., disturbance),
providing a further evidence that adequate processing tools
are needed to remove phenomena that are not harmful while
focusing on those that are a real threat to insulation system
reliability.

provide useful information for condition-based maintenance


practices only if the monitoring system is (a) able to reject
noise and disturbance from PD measurements, (b) separate
contributions to the overall pattern due to different PD sources
and, eventually, (c) provide a reliable indication about the
nature of the PD phenomena recorded, thus enabling
intelligent warning systems to be devised. All these features
are necessary to develop a system that is able to provide the
final user with a summary of the apparatus conditions, without
the need to process the large amount of data usually provided
by monitoring system.
The system described in this paper provides effective
solutions to implement a monitoring scheme that is effective,
i.e., able to provide timely warning messages, and viable, thus
allowing
for
maintenance
cost
reduction
and
reliability/availability maximization. At the present stage, the
system is also able to collect quantities from other sensors as,
e.g., temperature and/or vibration monitors. Further
development will focus on the apparatus-specific strategy to
exploit these additional information in order to reinforce PDbased diagnosis of insulation conditions.
VI. REFERENCES

B
Complete PD pattern

Classification map

Cluster B
Cluster A
Fig. 6. Separation of external discharges (cluster A) from internal discharges
in cable joint (cluster B) based on the characteristics of PD pulses.
500

Amplitude (mV)

12
300
10
200
8

Frequency (MHz)

14

Amplitude
Frequency

400

100
6
0
1

Monitoring time (days)

Fig. 7. Behavior of PD pulse magnitude and frequency content for a PD


phenomenon internal to the cable system.

V. CONCLUSIONS
Partial discharge monitoring of power apparatus can

[1] IEEE Guide for Partial Discharge Testing of Shielded


Power Cable Systems in a Field Environment, draft 11,
2005.
[2] A. Cavallini and G. C. Montanari, Effect of supply
voltage frequency on testing of insulation system, IEEE
Trans. on Dielectrics and Electrical Insulation, vol. 13, n.
1, pp. 111-121, February 2006.
[3] G. C. Stone, Tutorial on Rotating Machine Off-line and
On-line Partial Discharge Testing, EPRI/CIGRE
Colloquium on Maintenance of Motors and Generators,
Vol.3, Florence (Italy), April 1997
[4] A. Contin, A. Cavallini, G.C. Montanari, G. Pasini, F.
Puletti, Digital detection and fuzzy classification of
partial discharge signals, IEEE Trans. on Dielectrics and
Electrical Insulation, vol. 9, n. 3, pp. 335-348, June 2002
[5] A. Cavallini, G. C. Montanari, A. Contin, F. Puletti, A
new approach to the diagnosis of solid insulation systems
based on PD signal inference, IEEE Electr. Insul. Mag.,
vol. 19, no. 2, pp. 23-30, April 2003.
[6] A. Cavallini, A.Contin, G.C. Montanari, F. Puletti,
Advanced PD inference in on-field measurements. Part
I. Noise rejection, IEEE Trans. on Dielectrics and
Electrical Insulation, vol. 10, no. 2, pp. 216-224, April
2003..
[7] A. Cavallini, M. Conti, A. Contin, G.C. Montanari,
Advanced PD Inference in On-Field Measurements.
Part.2: Identification of Defects in Solid insulation
Systems, IEEE Trans. on Dielectrics and Electrical
Insulation, vol.10, no. 3, pp. 528-538, June 2003.
[8] A. Cavallini, M. Conti, G.C. Montanari, C. Arlotti, A.
Contin, PD inference for the early detection of electrical
tree in insulation systems, IEEE Trans. on Dielectrics
and Electrical Insulation, Vol. 11, n. 4, pp. 724-735,
August 2004

[9] J. Borghetto, A. Cavallini, A. Contin, G.C. Montanari, M.


de Nigris, R. Passaglia, G. Pasini, "Partial discharge
inference by an advanced system: analysis of online
measurements performed on hydrogenerators", IEEE
Trans. on Energy Conversion, Vol. 19, n. 2, pp. 333-339,
June 2004
[10] A. Cavallini, G. C. Montanari, S. Chandrasekar and F.
Puletti, A Novel Approach for the Inference of Insulator
Pollution Severity, IEEE ISEI, Toronto, Canada, June
2006
VII. BIOGRAPHIES
Andrea Cavallini (M1995) received from the University
of Bologna the master in Electrical Engineering in 1990
and the PhD degree in Electrical Engineering in 1995. He
was researcher at Ferrara University from 1995 to 1998.
Since 1998, he is associate professor at Bologna
University. His research interests are: diagnosis of
insulation systems by partial discharge analysis, reliability
of electrical systems and artificial intelligence. Since 2004,
he is the Italian representative of Cigr SC D1.

Gian Carlo Montanari (M87-SM90-F00) took the


Master degree in Electrical Engineering at the University of
Bologna. He is currently Full Professor at the University of
Bologna. He has worked since 1979 in the field of aging and
endurance of solid insulating materials and systems, of
diagnostics of electrical systems and innovative electrical
materials. He has been also engaged in the fields of power
quality and energy market, power electronics, reliability and
statistics of electrical systems. He is IEEE Fellow and
member of Institute of Physics. Since 1996 he is President
of the Italian Chapter of the IEEE DEIS. He is convener of the Statistics
Committee and member of the Space Charge, Multifactor Stress and Meetings
Committees of IEEE DEIS. He is Associate Editor of IEEE Trans. on DEI. He
is founder and President of the spin-off TechImp, established on 1999. He is
author or coauthor of about 450 scientific papers.
Francesco Puletti was born in Citt di Castello (PG), on
08/12/1974. He graduated in Electrical Engineering in
19/03/99. He has carried out research activity in the topic
of insulating system diagnosis by means of innovative
partial discharge measurement techniques in the
Laboratory of Material Engineering and High Voltage of
University of University of Bologna. He is now CEO of
TechImp, spin-off of Bologna University operating in the
field of the diagnostics of electrical systems

You might also like