Haefely Test AG, Tettex Instruments Division, Bernstrasse 90, P. O. Box, CH–8953 Dietikon - Zurich Switzerland

Phone +41.1.744 74 74, Fax +41.1.744 74 84, www.tettex.com, e-mail: sales@tettex.com

Pattern Recognitionfor Partial Discharge Measurement

Mark G. Turner Dr. Edward Gulski

Tettex Instruments Division,Haefely Test AG, Dietikon,

Switzerland

TU Delft,Netherlands

AbstractBasic PD testing of HV test objects is limited to measuring the inception voltage (inkV) and the largest discharge magnitude (in pC), and comparing these to the testspecifications. If the maximum allowable discharge level is exceeded, it is important toidentify the cause of the discharge.

The TEAS Diagnosis technique of generating fingerprints for individual partialdischarge measurements and comparing them with a databank of know PD patternshas been established for several years now.

This paper discusses the commercial implementation and practical application of thePD pattern recognition technique using the TE 571 Partial Discharge Detector. Variousapplications for digital PD analysis are discussed in the areas of monitoring anddiagnosis. Field measurements on turbo-generators, GIS, accessories and powertransformers underline the flexibility and usefulness of this technology.

2

Pattern Recognition for Partial Discharge Measurement

1. Acceptance of digital PD measurement and diagnosis

1.1 Introduction

Initial reaction to the TE 571 digital partial discharge detector when it was released in 1994could be summarised as follows: "Haefely will probably be successful with this instrument inresearch and development applications. The pattern recognition software is of academicinterest."

In fact, the application of the TE 571 proved to be universal and the appeal of patternrecognition has found a much broader following than anticipated. With more than 100 TE 571units now in the field, mostly installed between 1996 and 1997, it can be said that the partialdischarge community has widely embraced digital instrumentation.

This paper illustrates some areas in which digital discharge detection and diagnosis havefound application.

1.2 A commercial implementation of PD pattern recognition

When the maximum permissible PD level is exceeded during a PD test, it is often important toknow the cause of the discharge. Visual study of the PD patterns with reference to test voltagephase provides the experienced test engineer with valuable information. It has long beenaccepted that the shape of these patterns has been found to have a strong relationship withthe type of causal defect (CIGRÉ Working Group 21.03 "Recognition of Discharges"). PDpattern recognition is, therefore, an established practice.

Digital PD detectors provide the possibility for post-processing of the instantaneous PDsignals. Most commercially available instruments have a computerised option which canprovide 2 dimensional or 3 dimensional graphs of the PD activity (Figure 1). The test engineercan then visually study and attempt to interpret the various graphs. Ideally, the user shouldremain objective, study all graphs from all angles and have a perfect memory of previous testresults.

Figure 1 TE 571 3-D PD display of Hn(φφ,q)

3

Digital PD measurement technology has provided an easier solution. A fast and accurateanalysis of PD test results can be performed with pattern recognition software. The TE 571-DSW TEAS diagnostic software enables a characteristic fingerprint of the measurement to begenerated. Comparison of this fingerprint with a databank of known cases clarifies theinterpretation.

1.2.1 The principal of operation and the benefits

Post processing of PD data is performed in the standard TE 571 PD detector. The PD pulsesare plotted with respect to intensity and phase angle in a number of different combinations. TheTE 571-DSW TEAS software then has to form a fingerprint from this data and this is achievedby comparing the distribution of the PD data to a normal distribution. Standard statisticalparameters such as asymmetry, kurtosis and skewness can then be calculated. In addition,values are also taken for differences between the negative and positive test voltage halfcycles, and the number of peaks in the PD data distribution. The result is a fingerprintconsisting of 29 parameter values and this can then be compared with a databank of previousresults relevant to the test object. This process is called classification of the fingerprint.

Construction of the databank determines the potential outcome of the diagnosis. A number ofpreviously calculated fingerprints are arranged in groups which represent, for example, goodand bad test objects or specific fault conditions.

Classification produces a percentage score for the fingerprint compared to the different groupsof fingerprints contained in the databank. A score of 95% means that the fingerprint is in the top5% of typical fingerprints in the group. Plainly speaking, this means that it is highly likely tobelong to that group. The databank is judged to be well designed if it produces a high similarityfor the correct group and low for all others.

The TEAS-Diagnosis databanks can be designed by the user to perform multi-level diagnosis.Good/bad diagnosis is possible based upon a databank containing fingerprints of "regular" and"irregular" PD patterns in different, but related, test objects [3]. For example, a 370 MVA auto-transformer with a PD fault will show a high percentage match with PD fault patterns from othertransformer tests. A 370 MVA auto-transformer in good condition will show a high percentagematch with PD patterns from other transformers in good condition, and little or no match withfaulty transformer records.

An extension of good/bad diagnosis is practical for some test objects. Faulty phaseidentification in 3-phase transformers can be made with a simple databank based upon themeasurements from the individual phases (Figure 2). If one phase shows large dissimilarity tothe others then it can be identified as the main source of PD.

Figure 2 Diagnosis of transformer phases

4

Identification of source of fault can also be performed using the TEAS-Diagnosis [5]. In thiscase the databank must contain typical examples of the likely faults in the test object e.g.damaged screen, internal corona, tracking between layers, HV set-up noise. The databankthen acts as a library for the user's years of experience in testing and repair.

Life cycle analysis is a growing application for the TEAS-Diagnosis method. Periodic tests canbe fingerprinted and a databank constructed that characterises typical patterns expected fromthe test object during different phases of the ageing process. In power transformers thismethod can provide information about the windings etc. but does not indicate the condition ofthe insulating oil.

The potential benefits from the TE 571-DSW TEAS software are clear. Application details fortransformers, generators and GIS are described in the second part of this paper. But thequestion now discussed is: to what extent has PD pattern recognition by software beenaccepted by the partial discharge community ?

1.3 PD pattern recognition software as a TE 571 option

1.3.1 To what extent has TE 571-DSW TEAS Diagnosis shared in the successof the TE 571 detector?

A basic analysis of over 100 instruments currently in the field (Figure 3) shows that more than50% are equipped with the TE 571-DSW TEAS software. This ratio has been fairly constantsince the introduction of the instrument in 1994.

TE 571 Detectors with Diagnosis Option

53%with DSW

47%without

DSW

Figure 3 Ratio of TE 571 detectors with/without TE 571-DSW TEAS software

5

1.3.2 What are the applications for TE 571-DSW TEAS software?

Although the ratio of users with and without the diagnostic software is about 50/50, there arecertain applications which have a greater demand for PD pattern recognition than others (seeFigure 4). When this data is considered, it must be remembered that the size of population foreach application is affected by factors such as industry trends and market specialisation by theHV system supplier. Nevertheless, certain patterns are apparent.

TE 571 Applications with / without TE 571-DSW TEAS Diagnosis

0%

10%

20%

30%

40%

50%

60%

CTP

T

Ent

wic

klun

g

Gen

erat

oren GIS

Kab

el

Kon

dens

ator

en Traf

os

mit

ohnew/out

with

Figure 4 Percentage of TE 571 detectors by test application

As expected, the PD pattern recognition software is supplied for most R&D applications.

Customers testing generators always require the package and it is becoming well establishedin this application (see part 2.1.3 of this paper re. Turbo-generators).

GIS is a growing test area and early signs are that the diagnostic capabilities are well received(see part 2.3 of this paper re. GIS on-line).

Cable is predominantly a routine test application in which the operators immediately recognisescharacteristic production or set-up faults. Diagnostic software has found some interest,however, in the area of HV power cables and research / type test laboratories favour it.

Transformers is the strongest application area for TE 571-DSW TEAS software. This formspart of the Tettex solution for transformer testing and diagnosis together with C Tan δ andRVM.

1.4 Conclusion

The number of TE 571 PD detectors now in the field is considered significant enough to reflectmarket trends. Similarly, the suitability of TE 571-DSW TEAS diagnostic software for differentapplications can be determined. At this time, pattern recognition by software is a significantrequirement in PD test applications for transformers, HV power cables, generators and R&D.

6

2. Practical experience and development of test methods using theTE 571 digital partial discharge detector

2.1 Introduction

The main goal of PD diagnosis is to recognise the insulation defect causing the discharge e.g.internal or surface discharges, corona, treeing, etc. This information is vital for estimating theharmfulness of the discharge.

Manufacturers of HV equipment, together with producers and distributors of electrical power,have a growing interest in off-line, on-site and on-line analysis of PD in existing HVcomponents. The objective of such analysis is the early detection, location and recognition ofpossible insulation failures in HV equipment. As a result, maintenance actions can be preciselyplanned to prevent unexpected interruptions in equipment utilisation. Furthermore, based uponthe knowledge of type of discharge and behaviour over time, important information can beobtained regarding degradation processes.

The ability of digital PD analysers to processand store specific information concerningdischarge activity can be used for differentpurposes: discharge recognition, conditionmonitoring etc. To exploit these possibilities,a specific commercialised fingerprinttechnique TEAS® has been successfullyintroduced for off-line and on-line PDmeasurements of different HV components[3-15].

Figure 1Schematic diagram of a transformer test circuit usingthe multi-channel TE 571-MPR PD analyser.

2.1.1 PD database for decision support

The development of a PD database to support the discharge evaluation during periodic off-lineinspections of HV components is of great importance.

It is known that important conclusions are made regarding the condition of the test objectinsulation, based on periodic off-line PD measurements. It is also known that in several types ofHV apparatus a certain level of discharge is allowed. For example, turbo-generators areexpected to have a discharge level of < 10nC...50 nC and power transformers < 500 pC.Interpretation of the measuring results depends on the subjective opinion of test engineers.

With the advance of digital processing, the task of data acquisition and evaluation can now beperformed efficiently. This can provide interpretation of PD patterns and classification withrespect to type of discharge.

7

In the past a strong relationship has been foundbetween the shape of PD patterns which occurin the 50 Hz(60 Hz) sine wave and the type ofdefect causing them. From a practical point ofview it was shown that two types of PDpatterns are of interest by interpreting PDmeasurements:

regular PD patterns which are characteristicfor a particular type of HV component withinsulation in good condition, see figures 3 & 8;

irregular PD patterns representing certainunacceptable discharge sources. These maybe related to manufacturing defects or theeffects of ageing during service life, see figures4 and 9.

This study describes different PD databasescreated for two main areas of application:induced voltage tests of 80 different powertransformers and reactors and off-line PDmeasurements made over the last two years ontwenty different 6 MW and 63 MW turbo-generators. In both cases, the digital PDanalysis technique TEAS® was used for signalprocessing, statistical analysis and generationof the PD database.

Figure 2PD diagnosis test on a 50/10kV 14/10MVA power transformer using TE 571-MPR PD analyser(HV Laboratory, TU Delft, Netherlands).

2.1.2 PD database for power transformers and reactors

When a measurement has been made on a test object, it can be compared to the PD databasecomprising of a collection of previous PD tests. For reasons of clarity, the entire PD databasehas been divided into two separate parts. The first part is constructed of measurements madeon reactors, whereas the second part concerns only auto-transformers and three phasetransformers.

The main goal of this PD database was toanswer questions about general trends inregular or irregular PD patterns occurringduring induced voltage testing of powertransformers and reactors. In the following,two examples are given showing anapplication of both PD databases duringclassification of an unknownmeasurement.

Figure 3 Regular 3-D discharge pattern Hn(ΦΦ,q) for a203 MVA transformer in good condition.

8

As observed in [5,6], the classification of an acceptable PD pattern with a database (figure 5)gave a multiple recognition in many cases. In most of these cases a low discharge magnitudeand low discharge intensity had been observed (+). No recognition was found for unacceptablePD problems. When a typical defect was classified, recognition was found for only a few prob-lems (figure 6), most of which had shown unacceptable PD patterns.

Figure 4Irregular 3-D discharge pattern Hn (ΦΦ,q) fora 55 MVA reactor containing PD on adamaged screen inside the test object.

Figure 5Recognition of regular PD patterns usingcomputer aided PD database for reactors.Typical overlap with other reactors showingregular PD patterns. (+), (-), ( ) represents atest object in the database characterised bya (regular), (irregular) (unknown) pattern

Figure 6Recognition of irregular PD patterns usingcomputer aided PD database for reactors.Typical overlap with other reactors showingirregular PD patterns. (+), (-), ( ) representsa test object in the database characterisedby a (regular), (irregular) (unknown) pattern

9

2.1.3 PD data base for turbo-generators

When discharge data are measured during periodic inspections every few years, PD patternsof separate coils can be compared to those observed during previous inspection. Based onthese experiences, several characteristics have been found to describe typical insulationproblems of stator insulation [4,7]. Several groups of PD patterns have been found describingthe case of irregular PD patterns. Based on several inspections and repairs the followingdischarges sources were found during periodic PD-measurements:

a) PD in the HV bushings,

b) PD in a slot section caused bydamaged outer corona protection,

c) PD in the end winding section.

Figure 7Off-line periodic inspection of a 63MWturbo-generator using TE 571 PD analyser(ABB Dolmel Ltd, Poland).

As a result two PD databases, one for 6 MW and one for 63 MW units, have been developed tosupport the recognition of insulation degradation during periodic off-line inspection. In figures10 and 11 examples of classification of a particular defect with the databases is shown. Bothexamples confirm that the significance of PD patterns measured for particular defects can beused for identification of these defects.

Figure 8Example of regular PD pattern observed fora 63MW turbo-generator. The outer andinner sinusoidal shapes of the 3-D phaseresolved distribution Hn(ΦΦ,q) are typical forturbo-generators in good condition.

Figure 9Irregular PD pattern observed for a 6MWturbo-generator. This Hn(ΦΦ,q) phaseresolved distribution was observed forend-windings discharges.

10

Figure 10Recognition of an irregular PD pattern by the63MW turbo-generator database. A highpercentage represents a classification as aslot discharge.

Figure 11Recognition of an irregular PD pattern bythe 6MW turbo-generator database. A highpercentage represents a classification as anend-windings discharge.

2.1.4 Conclusion

The digital classification of PD patterns observed during periodic PD measurements on HVcomponents has made it possible to develop a decision support database for discharge faults.Two different ways of constructing a PD database have been shown. One provides adistinction between objects in good condition and objects showing unacceptable discharges.The other example confirms the possibility to evaluate the source of the discharge in theinsulation.

It has been demonstrated that, using this technique, clear distinction is possible betweencomponents in good condition and those which show internal or external discharges originatingfrom insulation degradation.

11

2.2 Digital PD location in HV cables using the travelling wave method

Partial discharges occur in gas-filled cavities in a dielectric and cause a gradual erosion of theinsulation material. For this reason location of PD in HV cables is important for quality control. Awidely used method for PD location is to use travelling waves, introduced in 1960 by F.H.Kreuger (see figure 12). A few years ago this method was automated and commercialisedunder the name PDLOC® and is now in world-wide use in shielded laboratories (see figures 13-15). If measurements are performed in an insufficiently shielded area, or the level of PD(normally just a few pC) is of a similar order to the background noise, the combination of noiseand disturbances may easily influence the PD sequence required for the location.

Figure 12 Figure 13Principle of travelling wave method. The two PD detector type TE 571-4 for location oftravelling waves caused by a PD pulse at site X defects in HV cablescan be detected at a cable end with the PD detector. (NKF Kabel B.V. Delft, The Netherlands)

In such cases a solution for noise and disturbance suppression is necessary and this can beachieved using digital filters. Figure 14 shows an example of the use of digital filters to improvePD location in the presence of noise and disturbances. In this particular case a matched filterwas used to suppress HF noise. An LF filter and Fourier filter are also provided.

Figure 14 Figure 15HF filter application. Upper curve represents PD PDLOC indication of discharge location atpulse sequence before filtering and lower curve 1194m in a 1785m long plastic insulated cable.after filter application.

12

2.3 Recognition of defects in GIS

Acceptance tests and periodic off-line measurements of SF6 gas insulated test objects arerestricted to measurement of PD inception voltage (in kV) and maximum discharge magnitudein pC and comparing these to the test specifications. The test objects may be GIS substationsor GIS components such as switchgear, disconnectors and bus bars.

If the permitted PD level is exceeded thenthe main goal of evaluation in GIS is tolocalise the discharge source. For periodicinspection it is also possible to useVHF/UHV sensors to measure PD signalson-line. The VHF/UHF detection circuitusually consists of a sensor and a spectrumanalyser (see figure 16).

Figure 16Four components are of importance for VHF/UHFPD measurements in GIS: (1) discharging defect,(2) excitation of travelling waves, (3) transferfunction sensor, (4) data processing.

The main objective of a PD measurement,whether it is based upon IEC 270 orVHF/UHF, is to assist with recognition andlocation of the discharging defect. Tosupport the evaluation process during ameasurement it is possible to use referencePD patterns of typical defects. Someexamples of typical GIS defects aredescribed below.

Figure 17420kV GIS test set-up (ABB/CESI, Milan, Italy).

Protrusion on the HV conductor represents sharp conducting particles which may occur onthe HV electrode inside the GIS installation. In figure 18 a phase-resolved plot is shown.

Figure 18 Figure 19Protrusion on the conductor at 220 kV Protrusion on the enclosure at 90 kV

13

Protrusion on the enclosure represents sharp conducting particles on the surface of theenclosure. In figure 19 the phase-resolved plot is shown. It follows from this comparison thatthe asymmetry between discharges in the positive and negative half of the applied AC voltagein case of a protrusion on the enclosure and a protrusion on the conductor is very typical forboth defects.

Figure 20 Figure 21Particle on insulator at 294 kV Free moving particle at 261 kV

Particle on an insulator means a small conducting particle contacting the surface of aninsulator (spacer) and distorting the field by producing a local field concentration. As a resultthe breakdown voltage along the surface is diminished and in some cases discharges mayoccur before the breakdown occurs. In figure 20 the phase-resolved plot is shown.

Figure 22 Figure 23Floating electrode in 4 bar SF6 Internal fault in switchgear at 138 kVFree moving particle inside the enclosure means a conducting particle which is not fixed toany of the electrodes or insulators may move (jump) inside the enclosure with a certainfrequency. As a result PD occur producing patterns as shown in figure 21. In contrast to thethree defects mentioned above, a typical sinusoidal shape can be observed in the phase-resolved plot for this defect.

Internal defect in the moving parts. Circuit breakers and disconnectors are mechanicallyand electrically stressed during their service life. As a result, ageing processes occur inside theelements. An example of internal discharges in the grading capacitances of switchgear isshown in figure 23.

Foreign particles and ageing processes of solid materials are not the only contributors to GISfailures. Floating parts in the installation i.e. electrodes imperfectly connected to HV potentialmay cause regularly repeating discharge groups of the same amplitude, see figure 22. Thispattern confirms the observation made before that each of the GIS defects is characterised byits own PD pattern. It can be seen that PD quantities processed by the TE571 can create yetfurther information for evaluation and diagnosis of PD measurements in GIS.

14

Figure 24Statistical analysis made by TEAS applied to four different PD patterns: (a) protrusion on the conductor,(b) protrusion on the enclosure, (c) free moving particle, (d) particle fixed to an insulator.

2.3.1 Conclusion

The systematic approach of examining digitally acquired PD quantities lays the foundation fora more systematic analysis of the different digital techniques and statistical tools which are inuse in the field of recognition and diagnosis of discharges in GIS components. Figure 24shows an example of statistical analysis using digital tools applied to four different PD faultpatterns: protrusion on the conductor, protrusion on the enclosure, free moving particles and aparticle fixed to an insulator. Discrimination, recognition and classification of these faults isshown to be possible using digital tools.

15

2.4 PD Pattern analysis of on-line measurements on rotating machines

In addition to periodic off-line PD testing, on-line PD measurement is an accepted method forrotating machines [13]. Using experience gained from off-line PD tests, this method can beutilised for condition based monitoring of the stator insulation [7,14].

The PD signals are measured by a specially adapted TE 571 PD detector (figures 25 and 26),using capacitive or inductive couplers, while the generator is in regular operation. Thecouplers are permanently installed on the generator (at least one on each phase) and an on-line test can then be performed. This type of measurement is easy performed withoutinterrupting the operation of the generator. As a result, the PD measurement is performed on asample under operational thermal and mechanical stresses.

Figure 25 Figure 26Measuring set-up for HF PD detection on machines VHF PD coupler of TE 571. Type: split ring

Rogowski coil, ∅∅ 160mm, impedance 50ΩΩ ,bandwidth 5 - 100MHz, sensitivity 96 mV/A.

Two difficulties arise when such PD measurements have to be performed:

- system interference may occur in the measuring circuit due to the power plant and from rotor excitation;

- complex propagation processes of PD signals through the stator winding occur, resulting in cross-talk. This is due to the fact that all three phases are energised at the same time.

A spectrum analyser (SA) can be used as a tuned filter to suppress external noise. The SA istuned to a frequency in the range of 10 MHz - 100 MHz where PD from the stator insulationdominate the noise signals. This measurement method is known as the VHF PD detectiontechnique due to the frequency range involved. The level of PD signals at this selectedresonant frequency f0 is demodulated to some hundreds of kHz and displayed on a 50 Hz time-base. As a result, the measured signals can be further processed by a conventional PDanalyser with the goal of using the broader experiences of phase resolved PD patternrecognition [4,15].

2.4.1 Partial discharge patterns

As mentioned above, the measured PD patterns will reveal the single phase PD responsetogether with PD responses from the other phases in an effect known as cross-talk. Theposition of the single phase patterns with respect to the power cycle of phase U is illustrated infigure 27, showing the 120° shift between the phases.

16

Figure 27Positions of the single phase patterns with respect to the power cycle of phase U.

Selection of a suitable resonant frequency for measurement is an exceptionally delicateprocedure. This is illustrated by figure 28 which shows five PD patterns measured at the samephase of a generator at different resonant frequencies. At f0 = 18 MHz the PD pattern is that ofthe actual phase. At f0 = 30 MHz and f0 = 62 MHz the measured PD pattern is that of the actualphase together with cross-talk. No response at all is measured at f0 = 48 MHz and at f0 = 64MHz only cross-talk is measured.

Figure 28Example to illustrate the influence offo. Measurement at different fo onphase W of a 650 MW generatorproduces different patterns.

17

Results of on-line measurements performed on a 155 MW and a 650 MW turbo-generatorclearly illustrate this influence of the resonance frequency upon the following responses (seefigure 29):

- the PD response of the measured phase, i.e. the PD activity originating from that phase

- the cross-talk PD response, i.e. the PD activity originating from the other phases;

- the disturbance response, i.e. disturbances originating from the power plant and from the generator itself (e.g. rotor excitation).

2.4.2 Evaluation for condition monitoring

When a suitable frequency is found and selected for the measurement of the phase’s own PDpattern, the pattern’s characteristics can be used for identification of the insulation state.Experience resulting from analysis of off-line PD tests can be used to assist the interpretationof PD patterns. As an example, figure 29 shows the three-dimensional Hn(Φ,q) distribution of ameasurement at phase U of a 155 MW turbo-generator. The measurement was performed at f0= 53 MHz. The pattern shows the phase’s own PD, cross talk PD and disturbances. Thephase’s own PD pattern shows the characteristics of a regular PD pattern of insulation with nosignificant degradation [7].

Figure 29Hn(ΦΦ,q) distribution as measured on phaseU of a 155 MW turbo-generator at fo = 53MHz. The pattern shows a regular shape forthe phase PD pattern (no degradation)together with cross-talk and disturbances.

2.4.3 Conclusion

Several conclusions can be drawn, based upon on experience gained from the on-linetechnique for PD tests presented above.

On-line VHF detection of PD processes in the insulation of a generator phase can beperformed with a number of suitable SA resonant frequencies.

Careful selection of resonant frequencies can provide information about the insulationcondition of the phases of a generator by analysis of the PD patterns. It can be expected that,in the course of time, local insulation degradation and disturbances will be identifiable by PDpattern deviation.

18

3. LITERATURE

[1] E. Gulski, P.N. Seitz, Computer-aided Registration and Analysis of PD in HVEquipment, Proc. 8th ISH, Yokohama, Japan, 1993

[2] E. Gulski, R. Oehler, New generation of computer-aided PD measurement systems,9th ISH, Graz, Austria, 1995

[3] E. Gulski, Digital Analysis of PD, IEEE Trans. on D and EI, Vol. 2, pp 822-837, 1995

[4] A. Zielonka, E. Gulski, K. Andrzejewski, Application of Digital PD MeasuringTechniques for the Diagnosis of HV Generator Insulation, Proc. CIGRE Session 1996,paper 15/33-06

[5] E. Gulski, H.P. Burger, G.H. Vaillancourt, R.Brooks, Digital Tools for PD AnalysisDuring Induced Test of Large Power Transformers, CEIDP, October 20-23 1996, SanFrancisco, p 36-39.

[6] E. Gulski, H.P.Burger, G.H. Vaillancourt, R. Brooks, PD database for powertransformers and reactors, ISH 1997, Montreal

[7] E. Gulski, J.P. Zondervan, A. Zielonka, R.Brooks, PD database for stator insulation ofturbogenerators, CEIDP, October 19-22, 1997, Minneapolis, p 546-549.

[8] E. Gulski, A.R. Samuel, L. Kehl, H.T.F. Geene, Digital Discharge Location in HVCables with Travelling Wave System, 1996 IEEE International Symposium on EI, June16-19, 1996, Montreal, Canada

[9] S. Meijer, E. Gulski, J.J. Smit, R. Brooks, Comparison of Conventional and VHF/UHFPartial Discharge Detection Methods for SF6 Gas Insulated Systems, 10th Int. Symp.on HV Engineering, Montreal, Vol. 4, pp. 187-190, 1997.

[10] S. Meijer, E. Gulski, W.R. Rutgers, Evaluation of Partial Discharge Measurements inSF6 Gas Insulated Systems, 10th Int. Symp. on HV Engineering, Montreal, Vol. 4, pp.469-473, 1997.

[11] S. Meijer, W.R. Rutgers and J.J. Smit, Acquisition of partial discharges in SF6

insulation, Conference on Electrical Insulation and Dielectric Phenomena, pp. 581-584,1996.

[12] E. Gulski, S. Meijer, W.R. Rutgers, R. Brooks, Recognition of PD in SF6 insulationusing digital data processing, Conference on Electrical Insulation and DielectricPhenomena, pp. 577-580, 1996.

[13] G.C. Stone, Tutorial on Rotating Machine Off-line and On-line PD Testing, Coll. onMaintenance and Refurbishment of Utility Turbogenerators, Hydrogenerators andLarge Motors, Firenze, 1997

[14] E. Binder, H. Egger, A. Hummer, M. Muhr, J. Schernthanner, Predictive Maintenanceof Generators, CIGRÉ Sess. 1992, pap. 11-305

[15] J.P. Zondervan, E. Gulski, J.J. Smit, R. Brooks, PD Pattern Analysis of On-lineMeasurements on Rotating Machines, Proc. CEIDP, Minneapolis,1997

[16] A. Krivda, E. Gulski, Influence of Aging on Classification of Partial Discharges inCavities, Jap Jnl App Phy Vol 33 (1994).

25.11.98