Professional Documents
Culture Documents
Working Group
D1.03
February 2013
RISK ASSESSMENT ON DEFECTS IN GIS
BASED ON PD DIAGNOSTICS
WG D1.03 (TF 09)
Members
U. Schichler, Convenor (DE), W. Koltunowicz, Secretary (AT)
F. Dorier †(FR), F. Endo (JP), K. Feser (DE), A. Giboulet (FR), A. Girodet (FR),
H. Hama (JP), B. Hampton (GB), H.‐G. Kranz (DE), J. Lopez‐Roldan (AU),
L. Lundgaard (NO), S. Meijer (NL), C. Neumann (DE), S. Okabe (JP),
J. Pearson (GB), R. Pietsch (DE), U. Riechert (CH), S. Tenbohlen (DE)
Copyright © 2012
Disclaimer notice
“CIGRE gives no warranty or assurance about the contents of this publication, nor does
it accept any responsibility, as to the accuracy or exhaustiveness of the information. All
implied warranties and conditions are excluded to the maximum extent permitted by
law”.
ISBN: 978-2-85873-218-0
Risk Assessment on Defects in GIS based on PD Diagnostics
Table of Contents
EXECUTIVE SUMMARY .................................................................................................... 4
Glossary.................................................................................................................................................. 5
Introduction ............................................................................................................................................. 6
1 Defect properties ......................................................................................................................... 9
1.1 Breakdown behaviour of defects in gas-insulated systems ....................................... 9
1.2 PD inception electric field strength ............................................................................... 11
2 Sensitivity of PD diagnostic methods ...................................................................................... 13
2.1 Foreign noise and background signals ........................................................................ 13
2.2 Sensitivity of PD detection during testing and in service ......................................... 14
2.2.1 PD measurement during on-site test ................................................................. 14
2.2.2 PD measurement in service ................................................................................ 14
2.3 Sensitivity of PD measurements..................................................................................... 15
3 Location and identification of PD defects .............................................................................. 16
3.1 Location of PD defects .................................................................................................... 16
3.2 Identification of PD defects ........................................................................................... 17
3.2.1 Identification based on electric measurements .............................................. 17
3.2.2 Identification based on acoustic measurements ............................................. 21
4 Assessment of critical defects ................................................................................................... 23
4.1 General aspects .............................................................................................................. 23
4.2 Mobile particles ............................................................................................................... 23
4.3 Loose and electrically floating elements ..................................................................... 24
4.4 Particles on insulation material ..................................................................................... 24
4.5 Protrusions ......................................................................................................................... 24
4.6 Voids in insulation material and cavities in spacer ................................................... 25
4.7 Contact erosion in main circuit ....................................................................................... 25
4.8 Defects in GIS components ............................................................................................ 26
4.9 Defects on interfaces ...................................................................................................... 27
5 Risk assessment............................................................................................................................ 28
5.1 General guideline ........................................................................................................... 28
5.2 Technical impact parameters of PD defects ............................................................... 28
5.3 Calculation procedure of dielectric failure probability ........................................... 29
5.4 Estimation of the consequences ..................................................................................... 31
5.5 Risk diagram .................................................................................................................... 31
5.6 Risk assessment for multiple PD defects ...................................................................... 32
Page 2
Risk Assessment on Defects in GIS based on PD Diagnostics
Conclusion ............................................................................................................................................ 34
References ............................................................................................................................................ 35
Annex ............................................................................................................................................ 39
Annex A: Calculation sheets of dielectric failure probability ............................................ 40
Annex B: Cluster analysis to form databases ....................................................................... 45
Annex C: Examples of risk assessment based on PD diagnostics ...................................... 47
Page 3
Risk Assessment on Defects in GIS based on PD Diagnostics
EXECUTIVE SUMMARY
Page 4
Risk Assessment on Defects in GIS based on PD Diagnostics
Glossary
Availability
Probability that an item is in a state to perform a required function under given conditions at a given
instant of time, assuming that the required external resources are provided [1].
Consequence
Outcome of a failure,e.g. its costs, social-economic and safety implications.
Critical defect
A defect which can reduce the insulation withstand level of the GIS below the coordination withstand
level (Ucw) and which might lead to dielectric failure.
Degradation
Partial discharges are the primary cause of degradation in (solid) parts of GIS.
Defect
Abnormality in a component that can be detected by partial discharge (PD) diagnostics.
Dielectric Failure
Failure is the flashover or breakdown of the insulation in the GIS.
Failure probability
Failure probability is the extent to which a dielectric failure is likely to occur [2]. In this document, an
estimation of failure probability is applied, based on limited information and assumptions only.
Fault
Fault is the state when a dielectric failure is followed by a power arc.
PD Monitoring
PD monitoring is the periodic or continuous activity to detect critical defects.
Risk
Risk is the combination of failure probability and the consequences of the failure [2].
Risk assessment
Risk assessment is the evaluation of identified risks to determine whether action is required
Page 5
Risk Assessment on Defects in GIS based on PD Diagnostics
Introduction
The number of installed GIS substations increases rapidly all over the world. The reliability of GIS is
improving as technology develops and improved designs are introduced. With increasingly reliable
equipment, the role of PD diagnostics becomes more important and the following expectations of
asset manager can be indicated [3]:
move from time based to condition based maintenance,
extend the operation intervals, reduce maintenance costs,
prevent damages, decrease unplanned outages,
enhance the availability and reliability of equipment,
detect defects before breakdown by warning and alarm functions,
reduce risk of expensive collateral damages (as fire in sustations),
extend the economic life time of equipment and avoid premature retirements,
improve safety of service personnel,
reduce environmental risks.
The order of priorities is changing from user to user and a compromise between cost savings, risk
reduction and technical performance improvement has to be found [3].
It is known that the maximum electric field stress is a decisive factor for the withstand voltages of SF6
gaps in quasi-homogeneous fields typical for GIS. Because of that, the dielectric properties of GIS are
mainly governed by the possible defects of its insulation media that may occur inside the GIS
enclosure during production, shipment, assembling on-site and during operation.
The in-service dielectric performance of GIS is satisfactory up to a nominal voltage of 170 kV but the
dielectric failure rate is considered to be too elevated for higher nominal voltages. In this case,
improvements are necessary to reach the target of 0.1 failures per 100 bay years, which is the
acceptable in-service failure rate indicated by utilities to get the same availability for GIS as for air-
insulated switchgear [4].
The worldwide in-service return of experience for GIS is presented in [4]. Referring to that, about 30%
of the dielectric failures are related to design deficiencies. Other failures are related to quality
assurance problems. Failures caused by mobile particles represent 20% of the total. Loose shields
lead to floating element type defects while current carrying contact create galling type defects; both
are also common malfunctions. Problems related to insulators surface contamination by particles and
voids in the bulk insulation cannot be excluded (Figure 1).
The majority of the problems can be eliminated by improved quality assurance measures during
manufacturing and on-site assembly. The improved dielectric testing procedure and more efficient
diagnostic checks, during on-site tests and later in service, have to be adopted. The aim of the on-site
dielectric tests is to detect the presence of foreign bodies (mobile particles), damages at the conductor
and enclosure (protrusions), incorrect assembly (for e.g. loose contacts) and all potential defects
which may evolve and later endanger the dielectric integrity of the GIS in service.
From the insulation co-ordination point of view, all the defects that may decrease the withstand
voltages of the GIS below the required withstand levels according to IEC 60071-1 are critical defects
and should be detected and eliminated. The improved and more rigid on-site testing procedures
proposed by CIGRE JWG 23/33-12 [4] are indicated in IEC 62271-203. For AC test, the value of
.
Ut = 0.36 ULIWL is proposed based on the typical ranges of withstand voltages under standard test
Page 6
Risk Assessment on Defects in GIS based on PD Diagnostics
voltages for the sound GIS [5]. The procedure has been widely adopted and specified by Utilities
worldwide.
The first step of the risk assessment procedure is to perform a sensitive PD measurement to detect
the critical defects. After the detection of a PD signal, it is necessary to obtain information about the
Page 7
Risk Assessment on Defects in GIS based on PD Diagnostics
type of defect (chapter 2) and about its location inside the GIS (chapter 3). Taken together with other
essential information from laboratory measurements, manufacturer's experience, design aspects and
trend analysis of the PD activity, the criticality of the defects can be estimated (chapter 4). This allows
the estimation of the dielectric failure probability (chapter 5.3).
In the last step, the risk assessment is performed based on the estimated dielectric failure probability
and failure consequences (chapter 5.5). The final decision can be different in case of on-site testing or
in service activity. The annex gives examples of risk assessment based on PD diagnostics.
It is not the intention of the CIGRE WG D1.03 (TF 09) to establish a standard procedure. The
presented risk assessment procedure is based on the information in chapters 2, 3 and 4 of this
brochure.
The proposed risk assessment procedure based on PD diagnostics combines technical and other
parameters (e.g. social and economical) and will support test engineers, engineers at condition
monitoring departments and asset managers.
Page 8
Risk Assessment on Defects in GIS based on PD Diagnostics
1 Defect properties
1.1 Breakdown behaviour of defects in gas-insulated systems
The volt-time curves for different types of the defects are presented in Figure 3 [9]. A sharp protrusion
on the inner conductor is critical at lightning impulse voltages. A protrusion of 1 mm length will lead to
breakdown during lightning impulse tests whereas at AC test voltages and switching impulse voltages
even a 5 mm long particle on the inner conductor is still not a critical [10, 11]. This is well explained by
the corona stabilisation effect at AC voltage and SF6 pressure of 0.3 MPa [12].
The influence of a particle on an insulator vs. electrical withstand of GIS configuration for different
lengths of the particle and at varying SF6 pressure is shown in Figure 4a at AC voltage and in
Figure 4b at lightning impulse voltages [9]. The flashover field strength EFo is given as the mean value
of 10 tests.
The criticality of a particle on an insulation material is further demonstrated in Figure 5, where the
flashover field strength at SF6 pressure of 0.4 MPa is compared to the typical design field strength of
GIS of rated voltages from 123 kV to 420 kV for lightning impulse voltages, switching impulse voltages
and AC voltages respectively. It is shown that to detect a particle by flashover on an insulation
material, the lightning impulse is the most critical voltage type. A particle shorter than 2 mm, located in
the highest electric field regions, is critical and reduces the flashover field strength far below the
designed level. Switching impulse voltages and AC voltages are less sensitive. A 4 mm long particle
leads to breakdown during AC test and a 10 mm one at nominal AC voltage.
Mobile particles are critical under AC voltages. The critical particle length for wires is in the range from
3 mm to 5 mm [6, 10, 11]. The experimental data is shown in Figure 3 (curve 1).
Page 9
Risk Assessment on Defects in GIS based on PD Diagnostics
a) b)
Page 10
Risk Assessment on Defects in GIS based on PD Diagnostics
critical defect
detectable length of
type of defect
apparent charge defect at Un
length
according to IEC 60270
moving particle 3 - 5 mm 2 - 10 pC 3 - 5 mm
protrusion on HV
around 1 mm 1 - 2 pC 3 - 4 mm
conductor
protrusion on enclosure 4 - 6 mm 2 pC 10 - 15 mm
3 - 10 mm
particle on insulation 1 - 2 mm about 0.5 pC
depending on location
3 - 4 mm 2 - 3 mm
Void 1 - 2 pC
(diameter) depending on location
The PD magnitude for the same defect placed in GIS of different voltage level will decrease with the
increase of nominal voltage (Figure 8). The apparent charge value is inversely proportional to the total
width of the insulation, so the higher the voltage level of the equipment, the lower the sensitivity of the
PD measurements.
Mobile particles are the most common defects detected on-site. The difference between the PD
inception level (PDIV) and breakdown value is much higher than for other defect types indicated in
Table 1. This means that free particles can be detected with greater simplicity
Page 11
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 12
Risk Assessment on Defects in GIS based on PD Diagnostics
Acoustic method
Acoustic signals from different sources have different bandwidth, both due to their initial characteristic
and due to absorption in the transmission path. Particle generated signals (impacts) have a bandwidth
from the kilohertz range and well into the megahertz range, while signals from protrusions start in the
megahertz range and fall off in the 100 kilohertz range due to sound absorption in the transmission
path (gaseous insulation).
The background noise for the acoustic method is given by:
A) Mechanical environmental noise that can originate from traffic, personnel moving in the plant, rain
etc. This noise is strongest in the low frequency range and falls off with increasing frequency. Sensors
and amplifiers with a low frequency cut off in the 10-kilohertz range suppress the external noise and
yields adequate sensitivity.
B) Magnetostrictive noise from inductive voltage transformers is a stable 100 Hz periodicity signal
being strongest close to the transformer. The occurrence of such noise does depend on the design of
the voltage transformer. Often it will occur during testing at elevated voltage when the core is
saturated.
Page 13
Risk Assessment on Defects in GIS based on PD Diagnostics
C) Magnetostrictive noise from enclosures made from steel, observed in bus runs with high currents.
These signals have a clear 100 Hz component in the envelope of the phase plot. These signals are
very stable and depend on the current value.
D) Heavy electric corona from bushings can impinge on enclosures. A 100 kHz high-pass filter will
remove such noise. However, in this case, the sensitivity for discharges will be strongly reduced.
A PD detection sensitivity equivalent to about 0.3 pC can be achieved for the detection of mobile
particles whereas for particles on insulation on the convex surface of a spacer cone a sensitivity
equivalent to approx. 5 pC was measured during experiments due to the fact that the spacer acts as a
barrier for the sound wave propagation to the enclosure.
UHF method
Signals from radar or mobile telephones in the frequency range between 300 MHz and 2 GHz are the
main interference for the UHF measurements. These noise signals are present at discrete frequencies
in the frequency domain. PD single resonance frequencies in the spectra (without interferences) must
be used for narrow-band UHF detection. In case of broad-band measurements the noise signals can
be removed by gating or filtering techniques. The background noise level is also influenced by the
distance between coupler and PD source: a PD magnitude equivalent to 0.3 pC can be detected for a
defect close to the coupler (test circuits in laboratory with a well-known defect position) [13]. For
complete GIS substations the detection of a defect showing a PD magnitude equivalent to 5 pC should
be fulfilled for any defect position, but often better sensitivities are obtained.
Page 14
Risk Assessment on Defects in GIS based on PD Diagnostics
On-site test: PD measurement at 80% of short time AC voltage withstand voltage test level on-
site (after the test according to CIGRE proposal, [4])
Page 15
Risk Assessment on Defects in GIS based on PD Diagnostics
Conductor
Coupler 1 X1 X2 Coupler 2
X
Δt
X ( X 2 X 1 ) X c t
X1 (1)
2 2
where ∆t is the difference in the arrival time of the signal at couplers 1 and 2, and vc is the
propagation velocity of the signal (0.3 m/ns).
If the electromagnetic wave passes through insulating material with permittivity εr, the propagation
Page 16
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 17
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 18
Risk Assessment on Defects in GIS based on PD Diagnostics
Figure 11: PRPS and calculation of voltage gradients and voltage differences of
consecutive pulses
As stated before, after a PD measurement is finished, three steps have to be taken into account for
PD identification: feature extraction, classification and determination of the confidence interval.
Examples of different methods, which are commonly used, are given in Table 2.
Page 19
Risk Assessment on Defects in GIS based on PD Diagnostics
The aim of feature extraction is to reduce PD-data. More often the pattern itself is the first step to
reduce data, from this in computer aided systems an additional step is performed to extract more
significant features by advanced pattern recognition procedures (e.g. Fourier, Haar transformation,
Markov Models, Fractals, etc.). The dynamic confidence interval used in [36, 37] is calculated by the
conformity of the new measurement to the collection of all references. If the conformity of a
measurement to the most likely defect class is below the conformity to the collection of all references,
then the classification is doubtful. Using this algorithm the confidence in diagnosis depends on a
dynamic boundary of confidence, which is additionally influenced by the degree of the conformity to
the favoured, second and third placed defect class. The confidence interval can also be determined by
statistical procedures. However, because of the limited amount of reference measurements and their
unknown distributions, the boundary is mostly set by expert estimation.
acquisition confidence
feature extraction classification reference
procedure interval
qn Statistical analysis centour score Expert [26, 31]
PRPS Fourier-transform L2-distance Dynamic [36, 38]
PRPS Haar-transform ANN > 50% [37, 38]
PRPS Walsh Hadamard Fuzzy Dynamic [37, 38]
PRPS ∆u ANN (CPN) > 50% [39]
PRPS ∆u + Fourier L2-distance Dynamic [37]
PRPS ∆u/∆φ + Fourier L2-distance Dynamic [37]
qn Fractals human expert human expert [40]
frequency, q no ANN not described [41]
PRPS qn ANN > 50% [42]
elliptic time b q Markov Models Model distance [43]
qn wavelet transform ANN (RBF) > 60% [44]
qn 16 x 20 matrix ANN not described [45]
qn no ANN > 70% [27]
Pulse Shape Q, tr, tf, width, area ANN > 70% [46]
qn Fractals + Statistics ANN > 90% [47]
Fractal, Statistics,
qn ANN Bootstrap-t [48]
Fourier
As shown above, classification can be performed as well by an expert as by a computer system. One
of the methods to classify PD measurements by a computer is the contour score method [26, 31]. This
method uses fingerprint populations of known discharge defects in a reference database. It compares
the fingerprint of the new PD measurement with the centre of each population. The contour score is
defined as the percentile rank of the fingerprints that are further away from the centre of the known
discharge population than the fingerprint of the new PD measurement. This has been illustrated in
Figure 14 for a simplified example with the feature vector dimension of only 2. Classification problems
occur if the fingerprint populations are overlapping. This is always the case if insufficient or not enough
features are used [38].
Page 20
Risk Assessment on Defects in GIS based on PD Diagnostics
The distance in feature space can be calculated by any algorithm used in common pattern recognition
(e.g. manhattan distance, L2 or others) [30].
It has been shown that different identification methods can generate different outcomes especially in
the cases of unknown defects. Therefore it is a common use to apply redundant systems, i.e. to
combine the results of several independent feature extraction and classification methods [30]. But also
the optimisation of the structure of a redundant identification system gives the chance to improve
diagnosis efficiency and diagnosis reliability [37].
50% contour
90% contour
C1
C2
1% contour
* New fingerprint
Feature 1
Figure 14: Contour formation with the contour score method [31, 49]:
□ fingerprints of PD defect A, ○ fingerprints of PD defect B
One major GIS reference database in use [33, 36, 38] in more than 24 populations uses a cluster
analysis as described in Annex B. Therefore any defect identification in GIS has to face the problem,
that one physical defect (e.g. protrusion, floating potential, mobile particle or void) has more than one
representation in feature space and consequently more than one significant PD pattern. This might be
the reason that an expert system proved to be more efficient for defect identification on site than a
human expert [50]. Only if the computer decides the defect type as "unknown" the expert's knowledge
is indispensable.
Conventional PD signals from round end particles will have a low rate (i.e. < 50 Hz). These
particles are discharging only once every period, when they hit the enclosure. When the ends are
sharp or when the particle is long, one starts to get discharges also during flight. These discharges
Page 21
Risk Assessment on Defects in GIS based on PD Diagnostics
can be quite big, and they can have a higher signal rate. Furthermore, the amplitude distribution
over the power cycle becomes asymmetric with respect to the positive and negative half period,
because discharging tendency is different for positive and negative polarity;
Acoustic signals from non-discharging particles have a high degree of symmetry in a phase plot
for the two half cycles of the applied voltage. Once one starts to get discharges from the particle
ends an asymmetry will occur. Also, amplitude vs. flight time plots will show increasing (and
oscillating) amplitude with flight time. Once one starts to get discharges from the ends it will loose
charge and the amplitude level will decay with flight time [53].
The particle plots are especially sensitive to the applied threshold level for the measurements. In case
where pulses are too small and below the threshold level, the elevation time (which is the time
between successive pulses) of the detected particle is not correct. This will result in the conclusion that
the particle is jumping and can even be indicated as harmful.
Floating
Particle Protrusion
shield
Figure 15: Scheme for PD defect identification using acoustic measurements [51]
Page 22
Risk Assessment on Defects in GIS based on PD Diagnostics
AQ ln 2l /r - 1
l =
2 k E 2
where
AQ Amplitude of acoustic signal variation s
Pulsation of oscilating electric field
l Particle length
r Particle radius
k sensor sensitivit y
E Modulus of electrical field
2
1 tmax h
Hh g
2 2
where
H h particle jump height
g 9.81m / s ²
tmax h maximum elevation time
The criticality of a jumping particle depends on its length and mass. Acoustic measurements can
provide, with some assumption, indications about the particle mass and length [53]. From the
Page 23
Risk Assessment on Defects in GIS based on PD Diagnostics
amplitude of the acoustic signal versus the time of flight (Figure 16), the particle jump height as well as
the length of the particle can be estimated by the equations given along with Figure 16.
As mentioned in chapter 1, the critical size of particle leading to breakdown during on-site tests is from
3 mm to 5 mm. A moving particle has the possibility to develop into another type of defect – particle on
an insulator – if during the trajectory movement the particle can reach and stay on the surface of an
insulation material. This defect, described in chapter 4.4, will be much more sensitive to LI rather than
AC overvoltages.
Mobile particle dimensions and its location, evaluation of temporary overvoltages and estimation of the
trajectory in relation to the insulation materials in the compartment must be analysed by the GIS
manufacturer and/or the utility in order to assess the dielectric failure probability of this defect.
4.5 Protrusions
Considering quality assurance procedures in the factory (e.g. dimensional controls carried out on
metallic parts), protrusions are very seldom found in GIS. Protrusions on HV conductors are mostly
due to scratches of metallic surfaces by insufficient care during assembly. Such defects reduce the
lightning impulse dielectric strength of the equipment (Figure 3) and the PD level is very low, close to
the background noise level. The shape of the protrusions tip can change in time with voltage
Page 24
Risk Assessment on Defects in GIS based on PD Diagnostics
application from a sharp to a rounded one. This can influence both, the PD inception voltage and the
breakdown level (Figure 17, [53]).
Page 25
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 26
Risk Assessment on Defects in GIS based on PD Diagnostics
The return of experience is based on interfaces like SF6/oil-bushings, surge arrestors and cable
terminations which show PD activity at nominal voltage [61, 62]. The IEC standards and field
experience of the equipment under question should be considered for the assessment of the PD
measurements.
Page 27
Risk Assessment on Defects in GIS based on PD Diagnostics
5 RISK ASSESSMENT
5.1 General guideline
Risk assessment on defects in GIS based on PD diagnostics is a complex task. One approach to
perform risk assessment is to combine technical and non-technical (social and economical) impact
parameters to estimate the dielectric failure probability and consequences (Figure 18, [63, 64]). The
technical impact parameters in this document are related to insulation defects generating PD.
Sensitive PD measurements are necessary to detect and identify the defect type. To assess the
technical risk the diagnostic confidence of each PD identification process must be considered. Based
on the risk assessment, further actions have to be taken.
Page 28
Risk Assessment on Defects in GIS based on PD Diagnostics
The behaviour of mobile particles is well known by experiments and return of experience (chapter 1
and 4). For this defect a lot of information for risk assessment can be generated by the application of
PD diagnostics and this information should be structured according the technical impact parameters
and aspects listed in Table 3.
- vicinity to spacer
- particle - vibration initiating - trend of - AC voltage level
Mobile dimension movement magnitude - DC stress
Particle and mass - local field strength - activity - superimposed
- jump height - particle trap - time of flight stresses
- dielectric coating
- trend of
- movement
Floating magnitude - AC voltage level
(fixing design) - vicinity to spacer
Element - activity
- number/cycle
- phase angle
Particle on - tip shape - LI
- local field strength * - activity
Insulation - length - VFT
- tip shape - LI
Protrusion - on HV electrode **
- length - VFT
- trend of
magnitude
- intermittent
- size * - AC voltage level
activity
Void - number * - local field strength * - inception by
- phase angle
- shape * transient voltage
stabilisation
- inception /
extinction voltage
Page 29
Risk Assessment on Defects in GIS based on PD Diagnostics
in the recognition of the defect type in question, the dielectric failure probability has to be set to a high
level, which means that additional expertise is necessary. Taking this into account, the calculation of
the breakdown probability has to consider the value of the diagnosis confidence factor C (Table 4).
uncertain untrained 0 - 25 %
A j
IP i
j1 (2)
NANI
where NA is the number of aspects and NI is the number of technical impact parameters. Secondly,
the failure probabilities of all technical impact parameters (IPi) are added to reach the failure
probability FP:
NA
NI NI A j
FP IP
i 1
i
i 1
j1
NANI
(3)
Page 30
Risk Assessment on Defects in GIS based on PD Diagnostics
NA
NI
Aj
TFP 1 1 R min R max R min (4)
j 1
C
i 1 N A N I
Examples of calculation sheets to estimate the total failure breakdown TFP for the defects shown in
Table 4 can be found in Annex A. The defined settings in these calculation sheets are based on the
experience of the WG members. They are not fixed and can be adjusted according to the experience
of the user.
Page 31
Risk Assessment on Defects in GIS based on PD Diagnostics
For the same dielectric failure probability (points A and B), different estimation of the consequences
can change the risk assessment area and finally actions to be performed by utility (Figure 20). In time
the failure probability can also change, e.g. a mobile particle moved to a harmless location, and for the
same estimated consequences the risk assessment area can be shifted from a yellow one (point C) to
the green one (point A).
Page 32
Risk Assessment on Defects in GIS based on PD Diagnostics
Figure 20: Risk diagram with different Figure 21: Risk factors for ranking results
results from risk assessment analyses of multiple PD defects
Page 33
Risk Assessment on Defects in GIS based on PD Diagnostics
Conclusions
PD diagnostics of GIS insulation generally aims to reduce as much as possible the failure rate.
Furthermore, it provides a significant decrease of the outage time, since a failure in GIS mostly causes
a longer outage time than the failure in air-insulated substation. Taking this into account, an
improvement of the reliability and availability of GIS can be obtained and repair works or exchange
measures, if needed, may be initiated before a breakdown or an outage occurs.
As an achievement, a better use of the equipment and an adoption of more efficient maintenance
strategies may be obtained by an advanced knowledge about the actual risk. By means of PD
diagnostics today a qualified assessment of the dielectric properties of the GIS can be achieved. In
this way a life extension of the HV assets is possible and their utilisation is accomplished leading to
considerable savings of life cycle costs. But it should be underlined, that without sensitive PD
measurements giving meaningful raw data, it is impossible to perform a proper risk assessment
analysis.
The report gives recommendations and guidelines for the application of PD measurement techniques
on site and in service. The recommendations are mainly given to utility but also to third party
consulting companies and to GIS manufacturers. It is important to underline the necessity of high PD
measurement sensitivity to detect and locate insulation defects, but the present limits of interpretation
of the results of the PD measurements, as related to the criticality of the defects, have also be noted.
Not all PD defects detected in service are critical and will lead to failure in short time. The PD
measurements have to be repeated to observe the PD signal behaviour in time. The remedial actions
can be scheduled and perform together with the regular maintenance activity.
In many cases the actions cannot be limited only to PD measurements but will request to adopt the
detailed risk assessment analysis.
The proposed risk assessment procedure based on PD diagnostics combines technical and other
parameters (e.g. social and economical) and will support test engineers, engineers at condition
monitoring departments and asset managers.
Page 34
Risk Assessment on Defects in GIS based on PD Diagnostics
References
[1] IEC 60050-191: International Electrotechnical Vocabulary, Dependability and Quality of Service,
1990, Amendment 1: 1999, Amendment 2: 2002
[2] ISO/IEC Guide 73: Risk Management - Vocabulary - Guidelines for Use in Standards, 2002
[3] CIGRE TF D1.02.08: Instrumentation and Measurements for In-service Monitoring of High
Voltage Insulation, CIGRE, Technical Brochure No. 286, 2005
[4] CIGRE JWG 33/23.12: Insulation Co-ordination of GIS: Return of Experience, On Site Tests
and Diagnostic Techniques, Electra No. 176, 1998
[5] A. Diessner, G.F. Luxa, W. Mosca, A. Pigini: High Voltage Testing of SF6 Insulated Substations
on Site, CIGRE, 1986, Report 33-06
[6] E. Colombo, W. Koltunowicz, A. Pigini: Sensitivity of Electrical and Acoustical Methods for GIS
Diagnostics with Particular Reference to On Site Testing, CIGRE Symposium "Diagnostics and
Maintenance Techniques”, Berlin, Germany, 1993
[7] CIGRE WG 15/33.03.05: Partial Discharge Detection System for GIS: Sensitivity Verification for
the UHF Method and the Acoustic Method, ELECTRA No. 183, 1999
[8] CIGRE WG 15.03: Diagnostic Methods for GIS Insulating Systems, CIGRE, 1992, Report
15/23-01
[9] R. Schurer: Der Einfluss von Stoerstellen auf Stuetzeroberflaechen auf die elektrische
Festigkeit von Isolieranordnungen in SF6-isolierten Anlagen. Dissertation Universitaet Stuttgart,
1999
[10] CIGRE WG 15-03: Effects of Particles on GIS Insulation and Evaluation of Relevant Diagnostic
Tools, CIGRE, 1994, Report 15-103
[11] A. Bargigia, W. Koltunowicz, A. Pigini: Detection of Partial Discharges in Gas Insulated
Substations, IEEE Transactions on Power Delivery, 1992, pp. 1239 - 1249
[12] A. Haddad, D.F. Warne: Advances in High Voltage Engineering, IEE, London, United Kingdom,
2004
[13] T. Kato, F. Endo, S. Hironaka: Sensitive Partial Discharge Monitoring System by UHF Method
and Calibration Technique, CIGRE Symposium “Gas Insulated Systems”, Dubai, 2001, pp. 73 -
76
[14] IEC 62271-203: High-voltage Switchgear and Controlgear - Part 203, Gas-insulated metal-
enclosed switchgear for rated voltages above 52 kV, 2011
[15] W. Koltunowicz, M.L. Philip, Y.S. Goh, W.H. Leong, M. Sethuraman, A. Darus, S.A. Fuad:
Partial Discharges Tests on GIS Installation in Malaysia - Techniques and Experiences, CIGRE,
2000, Report 15-304
[16] M.C. Zhang, H. Li: TEM- and TE-Mode Waves Excited by Partial Discharges in GIS, 11th ISH,
London, UK, 1999
[17] G. Schöffner: A Directional Coupler System for the Direction Sensitive Measurement of UHF PD
Signals in GIS and GIL, CEIDP, 2000, pp. 634 - 638
[18] S.M. Hoek, U. Riechert, T. Strehl, S. Tenbohlen, K. Feser: A New Procedure for Partial
Discharge Localization in Gas-Insulated Switchgears in Frequency Domain, 14th ISH, Beijing,
China, 2005, Paper G-005
Page 35
Risk Assessment on Defects in GIS based on PD Diagnostics
[19] B. Hampton, R.J. Meats: Diagnostic Measurements at UHF in gas-insulated Substations, IEE
Proceedings, Vol. 135, No. 2, 1988, pp. 137 - 144
[20] T. Utsumi et al.: Preventive Maintenance System with a Different Gas Injecting Facility For GIS,
IEEE Transactions on Power Delivery, Vol. 8, No. 3, 1993, pp. 1107 - 1112
[21] F. Endo et al.: Insulation Diagnostic System of GIS, 8th ISH, Yokohama, Japan, 1993, Paper 66-
01
[22] M. Hanai, F. Endo, S. Okabe, T. Kato, H. Hama, M. Nagao: New Development for Detecting
Partial Discharge Using an UHF Method and Its Application to Power Apparatus in Japan,
CIGRE, 2006, Report D1-106
[23] L.E. Lundgaard, M. Runde, B. Skyberg: Acoustic Diagnosis of Gas Insulated Substations: A
Theoretical and Experimental Basis, IEEE Transactions on Power Delivery, Vol 5, No 4, 1990,
pp. 1751 - 1759
[24] L.E. Lundgaard: Partial Discharge - Part XIV: Acoustic Partial Discharge Detection - Practical
Application, IEEE Electrical Insulation Magazine, Vol. 8, No. 5, 1992, pp. 34 - 43
[25] S. Markalous, S. Tenbohlen, K. Feser: Detection and Location of Partial Discharges in Power
Transformers Using Acoustic and Electromagnetic Signals, IEEE Transactions on Dielectrics
and Electrical Insulation, Vol. 15, No. 6, 2008, pp. 1576 - 1583
[26] E. Gulski: Computer aided Recognition of Partial Discharges using Statistical Tools, PhD
Thesis, Technical University Delft, The Netherlands, 1991
[27] B.T. Phung: Computer-Based Partial Discharge Detection and Characterisation, PhD Thesis,
University of New South Wales, Australia, 1997
[28] H.-G. Kranz, A. Lapp: Neuro-Fuzzy Diagnosis System with a Rated Diagnosis Reliability and
Visual Data Analysis, IDA-97, London, UK, Springer Verlag, 1997
[29] T. Hücker: Computergestützte Teilentladungsdiagnostik unter Berücksichtigung praxisrelevanter
Randbedingungen, PhD Thesis (in German), Bergische Universität Wuppertal, 1995
[30] H.-G. Kranz: Fundamentals in Computer Aided PD Processing, PD Pattern Recognition and
Automated Diagnosis in GIS, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 7,
2000, pp. 12 - 20
[31] A. Krivda: Automated Recognition of Partial Discharges, IEEE Transactions on Dielectrics and
Electrical Insulation, Vol. 2, No. 5, 1995, pp. 796 - 821
[32] A. Lapp, H.-G. Kranz: The Use of the CIGRE Data Format for Partial Discharge Diagnosis
Applications, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 7, 2000, pp. 102 -
112
[33] H.-G. Kranz, A. Lapp: The Influence of the Quality of the Reference Data Base on PD Pattern
Recognition Results, 10th ISH, Montreal, Canada, 1997
[34] R. Patsch, F. Berton: Pulse Sequence Analysis - Development and Future Prospects, 12th ISH,
Bangalore, India, 2001, Paper 6.69
[35] D. Aschenbrenner, H.-G. Kranz: Diagnosis Potential of Different Partial Discharge Features of
diverse PD Defects in N2/SF6 Mixtures, 7th ICPADM, Nagoya, Japan, 2003, Report S3-2
[36] D. Aschenbrenner, H.-G. Kranz, U. Schichler: Hierarchical Diagnosis Systems for Partial
Discharge Identification and Risk Assessment in GIS, 12th ISH, Bangalore, India, 2001, Paper
6.70
[37] D. Aschenbrenner, H.-G. Kranz, A. Lapp: The Importance of Voltage Independent Partial
Discharge Features for Monitoring Applications and Risk Assessment, 12th ISH, Bangalore,
India, 2001, Paper 6.58
Page 36
Risk Assessment on Defects in GIS based on PD Diagnostics
[38] T. Huecker, J. Gorablenkow: UHF Partial Discharge Monitoring and Expert System Diagnosis,
IEEE Transactions on Power Delivery, Vol. 13, No. 4, 1998, pp. 1162 - 1167
[39] M. Hoof, B. Freisleben, R. Patsch: PD Source Identification with Novel Discharge Parameters
Using Counterpropagation Neural Networks, IEEE Transactions on Dielectrics and Electrical
Insulation, Vol. 4, No. 1, 1997, pp. 17 - 32
[40] L. Satish, W.S. Zaengl: Can Fractal Features be used for Recognizing 3D Partial Discharge
Patterns?, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 2, No. 3, 1995, pp.
352 - 359
[41] T. Kato, F. Endo, S. Hironaka: Sensitivity Calibration of UHF Partial Discharge Monitoring
System in GIS, IEE of Japan, Vol. 122-B, No. 11, 2002
[42] H. Suzuki, T. Endoh: Pattern Recognition of Partial Discharge Patterns Using a Neural Network,
IEEE Transactions on Electrical Insulation, Vol. 27, 1992, pp. 543 - 549
[43] L. Satish, B. I. Gururaj: Use of Hidden Markov Models for Partial Discharge Pattern
Classification, IEEE Transactions on Electrical Insulation, Vol. 28, 1993, pp. 172 - 182
[44] E.M. Lalitha, L. Satish: Wavelet Analysis for Classification of Multi-source PD Patterns, IEEE
Transactions on Dielectrics and Electrical Insulation, Vol. 7, No. 1, 2000, pp. 40 - 47
[45] N. Hozumi, T. Okamoto, T. Imajo: Discrimination of Partial Discharge Pattern using a Neural
Network, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 27, No. 3, 1992, pp.
550 - 556
[46] A.A. Mazroua, R. Bartnikas, M.M.A. Salma: Neural Network System using the Multi-layer-
Perceptron Technique for the Recognition of PD Pulse Shapes due to Cavities and Electrical
Trees, IEEE Transactions on Power Delivery, Vol. 10, 1995, pp. 92 - 96
[47] R. Candela, G. Mirelli, R. Schifani: PD Recognition by Means of Statistical and Fractal
Parameters and a Neural Network, IEEE Transactions on Dielectrics and Electrical Insulation,
Vol. 7, No. 1, 2000, pp. 87 - 94
[48] M. Amini, P. Gallinari, F. d’Alche-Buc: Automated Statistical Recognition of Partial Discharges in
Insulation Systems, 8th ICANN, 1998
[49] S. Meijer: Partial Discharge Diagnosis of High-Voltage Gas-Insulated Systems, PhD Thesis,
Technical University Delft, The Netherlands, 2001
[50] J. Gorablenkow, T. Hücker, U. Schichler: Application of UHF PD Monitoring and Expert System
Diagnosis, IEEE ISEI, Arlington, USA, 1998
[51] L.E. Lundgaard, B. Skyberg, A. Schei, A. Diessner: Method and Instrumentation for Acoustic
Diagnoses of GIS, CIGRE, 2000, Report 15-309
[52] H.-D. Schlemper, K. Feser: Characterization of Moving Particles in GIS by Acoustic and
Electrical Partial Discharge Detection, 10th ISH, Montreal, Canada, 1997, Paper 3109
[53] L.E. Lundgaard: Particles in GIS: Behaviour and Possibilities for Characterisation from Acoustic
Signatures, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 8, No. 6, 2001, pp.
1064 - 1074
[54] A. Girodet, S. Meijer, J. Smit: Development of a Partial Discharge Analysis Method to Assess
the Dielectric Quality of GIS, CIGRE, 2002, Report 15-106
[55] R. Zoetmulder, S. Meijer, J. Smit, A. Girodet: Risk Assessment of GIS containing Free Moving
Particles using Spectral and Partial Discharge Analysis, INSUCON, Berlin, Germany, 2002
[56] R. Schurer, K. Feser: The Effect of Conducting Particles adhering to Spacers in gas insulated
Switchgear, 10th ISH, Montreal, Canada, 1997, pp. 165 - 168.
Page 37
Risk Assessment on Defects in GIS based on PD Diagnostics
[57] H.-D. Schlemper, R. Kurrer, K. Feser: Sensitivity of On-site Partial Discharge Detection in GIS,
8th ISH, Yokohama, Japan, 1993, pp. 157 - 160
[58] A.R. Salinas, J.S. Pearson: Detection of Partial discharges in Gas Insulated Switchgear Using
Diagnostic Technique, Doble 2000 Conference in Boston
[59] M. Runde et al.: Service Experience with Voltage Grading Capacitors, CIGRE, 2006, Report A3-
207
[60] A.A. Ebrahim: Review of UHF Detection of Leader Discharges in SF6 and the Bahrain
Experience, CIGRE Symposium “Gas Insulated Systems”, Dubai, 2001, pp. 77 - 82
[61] U. Riechert, J. Linn, E. Winkler, R. Pietsch: Reasonable Application of UHF Partial Discharge
Measurements in Development, Production and Service of Gas Insulated Switchgear (GIS),
CIGRE Symposium “Gas Insulated Systems”, Dubai, 2001, pp. 66 - 72
[62] K.T. Yoon, G.J. Behrmann, R. Pietsch, J. Pearson: Further Results of the Partial Discharge
Monitoring System at Labrador and Ayer Rajah 400 kV GIS Substation, 13th CEPSI Conference,
Manila, 2000
[63] M. Mathis, V. Koch, P.T. Lee, G. Balzer: Reliability Centered Maintenance with CALPOS Main,
13th ISH, Delft, The Netherlands, 2003
[64] A. Meyna: Einführung in die Sicherheitstheorie, Hanser Verlag, 1982, ISBN 3-446-13581-2
[65] K. Dreisbusch, H.-G. Kranz, A. Schnettler: Determination of a Failure Probability Prognosis
based on PD-Diagnostics in GIS, IEEE Transactions on Dielectrics and Electrical Insulation,
Vol. 15, No. 6, 2008, pp. 1707 - 1714
[66] G. Balzer, K. Bakic, H.-J. Haubrich, C. Neumann, C. Schorn: Selection of an optimal
Maintenance and Replacement Strategy of HV Equipment by a Risk Assessment Process,
CIGRE, 2006, Report B3-103
[67] F.R. Farmer: Reactor Safety and Siting: A Proposed Risk Criterion, Nuclear Safety, Vol. 8, No.
6, 1967
Page 38
Risk Assessment on Defects in GIS based on PD Diagnostics
Annex
Page 39
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 40
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 41
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 42
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 43
Risk Assessment on Defects in GIS based on PD Diagnostics
Page 44
Risk Assessment on Defects in GIS based on PD Diagnostics
Once a database has been built, cluster analysis can be used to check whether the database
is consistent. Cluster analysis tries to recognise groups in a number of feature vectors
without a priori knowledge. A feature vector consists of one or more features, extracted from
a PD measurement. A detailed description of clustering techniques can be found in [31, 67].
Figure A.1 shows a simplified example how cluster analysis works. Here five feature vectors
consisting of only one feature are analysed. The feature vectors can be analysed and
compared in different ways. In this example, simply the distance between both feature
vectors is calculated. The two vectors which are closest together (having the smallest
distance and are the most similar) are linked together and represented by one new feature
vector. In this way, the number of feature vectors is reduced by one.
Figure A.1: Example of cluster analysis of fingerprints using the tree structure
as obtained from different measurements [49]
An example of such cluster analysis for frequency spectra visualized by tree structure is
shown in Figure A.2 [49]. In this cluster, four separate clusters can be recognised, each
representing one type of defect.
Page 45
Risk Assessment on Defects in GIS based on PD Diagnostics
Figure A.2: Cluster analysis applied to the characterizations calculated from the
frequency spectra for four different defects
Page 46
Risk Assessment on Defects in GIS based on PD Diagnostics
During on-site testing of a 145 kV GIS, some PD signals were detected by the UHF method. The PD
coupler close to a SF6/air-bushing registered PD pattern after the high-voltage testing with
U = 220 kV, while the adjacent PD coupler located in the busbar registered no PD activity. The PRPD
patterns changes by time, but the characteristic features were always the same (Figure C.1). The PD
defect was traced to the epoxy-impregnated SF6/air-bushing by switching operations and comparison
with known PD patterns from laboratory (Figure C.2). It was suspected that transport damage was the
reason of the defect. The dismantled bushing was subjected to a further routine test by the
manufacturer. During the routine test with PD measurement according to IEC 60270 the PD
background level of 1.5 pC was not exceeded up to the full test voltage, so there had been no damage
caused during transport. The SF6/air-bushing was then refitted on-site and tested again. The PD
patterns were reproducible and thus correlated with an apparent charge of less than 1.5 pC. Due to
the low PD level and the positive service experience with the related type of SF6/air-bushing by the
manufacturer and the utility, there was no objection to put the SF6/air-bushing into operation. There
was no further objection for the last 14 years.
Figure C.1: PRPD pattern Figure C.2: PRPD chart from laboratory test
(U = 176 kV, f=97 Hz) (f = 50 Hz, type of defect: void)
Page 47
Risk Assessment on Defects in GIS based on PD Diagnostics
A utility heard audible noise from the GIS that not had been observed earlier. They believed the noise
came from a current transformer and wanted it to be tested to investigate the problem. Acoustic
measurements were carried out within one week. At the day of the measurements, the audible noise
was intermittent (coming and going). The audible noise resulted in acoustic signatures as shown in
Figure C.4. The acoustic amplitudes were very large, and the phase plot shows a distinct 100 Hz
phase correlation, i.e. strong indications of an electrically floating element causing large partial
discharges. When the audible noise was gone,
however, small partial discharges were still detectable
with the acoustic method. By searching around for the
location that showed the highest acoustic signal level,
it was soon realized that the signal did not originate
from the current transformer, but from the vertical
busbar (approx. 1.5 m) under the CT. By using two
acoustic emission sensors, the partial discharge was
accurately located to the area under an insulating
support where a tulip contact and a shield were
located. When the GIS the following week was opened
for inspection, the underside of the insulating support
Figure C.4: Acoustic signature
was covered with white dust (by-products from partial
due to a floating element
discharges), and burning marks due to partial
discharges were found in the interface between the
shield and the support. The calculation sheet to
estimate the dielectric failure probability is presented in
Figure C.5 and shows a result of 98,1%.
Page 48
Risk Assessment on Defects in GIS based on PD Diagnostics
Figure C.6: Acoustic signature for a small, harmless particle in a 275 kV busbar
Page 49
Risk Assessment on Defects in GIS based on PD Diagnostics
A mobile particle in a 150kV substation was detected by UHF PD measurement. Measuring data and
identification are shown in Figures C.8a-d. The calculation sheet of dielectric failure probability is filled
in resulting in a value of 61%, which is rather high. At the time of the occurrence, the utility decided to
increase the measuring rate. As a result, the particle seems to be rather inactive and no maintenance
actions were taken. No failure occurred as well and after dismantling the GIS several years later, a
mobile particle of 4 mm long was found, as shown in the Figure C.8.
Page 50
Risk Assessment on Defects in GIS based on PD Diagnostics
In this example, a void in a 380 kV substation is shown. The PD pattern looks like the one shown in
the Figure C.9 below. This PD pattern is already being measured for about 10 years and has not
significantly changed over the years. Filling in the calculation sheet the dielectric failöure probability is
52.5%, which is however almost the maximum of the indicated ranges. During the exercise, a
discussion started about the voltage level and wave shape: because the defect is always active during
normal operating voltage, the transients have no influence (no effect) on the inception voltage. So it
was set to “1”. In case the correct answer is “no” it has to be set to “0”, which reduces the breakdown
probability to 37.5% which lower the risk.
Page 51