You are on page 1of 12

21, rue d'Artois, F-75008 Paris Session 2000

http://www.cigre.org 15-302 © CIGRÉ

Partial Discharges in Transformer Insulation


by

Task Force 15.01.04•

SUMMARY
This paper discusses the opportunities for realising the control during factory testing, and a growing concern
full potential of modern digital partial discharge detec- for the technical condition of an ageing transformer
tors in the condition monitoring of large oil-filled trans- population has led to a focus on diagnostic methods.
formers, through the correlation of discharge patterns According to IEC 60076-3 partial discharge measure-
with typical discharge-inducing defects. The types of ment may be used for the commissioning test of a trans-
discharge which can be present in transformers are former. For transformers in service it would be useful to
analysed and this leads to a discussion on appropriate obtain a more detailed analysis based on partial dis-
test instrumentation to accurately and objectively ac- charge measurement in cases where gas-analyses has
quire, store and process discharge data. The range of given warnings of internal discharges.
discharge-inducing defects to be found in transformers
is discussed and this leads into the results of three in- In 1996 Cigré WG 15.01 established TF 15.01.04 to
vestigations by task force members of model geometries consider the possibilities for “Characterisation of differ-
representing such defects variously. In these studies ent discharges and their harmfulness to transformer in-
correlations were sought among the patterns of dis- sulation”.
charge activity and the nature/form of defect model. On
the basis of these investigations, key parameters are Partial discharge measurement has drawn much atten-
elucidated to include the phase angle of discharge ac- tion during recent years. Many studies relevant to GIS
tivity (start, mean and end phase), discharge repetition and machine insulation have been published. However,
rate and discharge magnitude. No evidence has been surprisingly little work has been published on dischar-
found to support the use of mathematical distributions ges in liquid impregnated insulation systems relevant to
of these primary parameters. In addition, from these in- power transformers. (Although problems associated
vestigations, the parameters which may influ- with the complex nature of the transformer circuit in
ence/control the results obtained (and hence which must measuring partial discharges have been recognised [1].)
be controlled) are discussed.
In this paper the task force wishes to draw attention to
Keywords some fundamental aspects of PD initiation and occur-
rence in impregnated insulation systems.
Transformer, Impregnated Insulation, Paper, Oil, Dis-
charges, Patterns, Defects, Moisture, Gassing.
*Members:
L.E.Lundgaard (NO) secretary, J.Poittevin (FR)(SC12),
1. INTRODUCTION J. Schmidt (SC12)(HU), D.Allen (AU), T.R.Blackburn
(AU), H.Borsi (DE), N.Foulon (FR), J.Fuhr (CH),
Generally, transformers are considered to be reliable N.Hosokawa (JA), R.E.James (AU), I.J.Kemp(GB),
apparatus. However, recent trends to improve quality O.Lesaint(FR), B.T.Phung(AU)

*
L.E.LUNDGAARD – SINTEF Energy Research, Sem Saelandsvei 11, N-7465 Trondheim Norway
Also some aspects of measuring techniques are dis- The acceptance criteria for the discharge level set in a
cussed. Finally some examples of representative defects commissioning test, is 300/500 pC measured at 1.5 UM
and their characteristic PD-signatures will be discussed during a 30 minute period after a short time over-stress
in the framework of some simple – but robust – physi- to 1.7 UM. However, the measured magnitude of a dis-
cal models. Our main goal is to create a datum for fur- charge will, as with rotating machines, depend on the
ther research in this field. coupling of the discharge pulse to the terminals. Fur-
thermore, the level referred to in the standards (e.g. IEC
2. INSULATION SYSTEM AND DISCHARGES IN 60270), is the largest repeatedly occurring apparent dis-
TRANSFORMERS. charge (measured with a quasi-peak detecting instru-
ment with a time constant less than 0,44 second). This
2.1. The insulation system of a transformer has to be distinguished from the maximum PD level re-
The insulation system in a transformer consist of oil and corded with modern instrumentation, which is able to
solid insulation (i.e. pressboard, paper and wood) used detect every single discharge occurring within the test-
for creating supports and cooling ducts for the wind- ing period.
ings. Deterioration of the cellulose is an irreversible
process, while the oil may be considered, at least par- The apparent charge is not always the best unit to quan-
tially, as self-healing. tify the magnitude of a discharge in an insulation sys-
tem, because a different apparent charge can be attrib-
An impregnated insulation system is very complex. uted to the same discharge source depending on its lo-
Moisture varies from one transformer to the other, and cation in the local field and its coupling to the winding.
will also vary with time and temperature. Therefore dif- The distances in the insulation system (i.e. voltage
ferences in conductivity of solid and liquid insulation is class) influence this coupling [5]. As an example a de-
to be expected. Liquid motion due to forced cooling and fect showing some 10,000 pC in a 20 kV system will
field enhanced charge injection will move space charges show some 1000 pC if located similarly in a 200 kV
around in the open oil volumes, thus influencing the lo- system. Theoretically, the acceptance level should
cal electric field stresses. The liquid movement may therefore be reduced for the higher voltage classes.
even give rise to electrostatic charge generation.
Referring to the IEC standard 60270, integrating the en-
The water content in the insulation, and the dynamic ergy dissipation that takes place over one period is an
fluctuation of humidity in pressboard and oil with trans- alternative of focusing on the discharge magnitude. The
former temperature is an important feature of the insu- dissipated energy may be estimated by summing the
lation system [2, 3, 4]. In particular humidity at the sur- charge multiplied by the instantaneous voltage (Σqi⋅Vi),
face of the paper is important for the occurrence of dis- provided a proper calibration has been made and the
charges. polarities of both charge and voltage are considered.
However, one should bear in mind that for many small
Typically a new transformer will have a moisture con- distributed discharges and for one large localised dis-
tent less than 1% (by weight). Before the vapour phase charge the dissipated energy may sum up to same.
technique was introduced in the 1960's the drying could
be less thorough. Over time the moisture content will The “scaling” of the discharge severity with the voltage
increase, both due to breathing and to ageing of the class may therefore be done in two ways:
cellulose in the insulation. The humidity of an old trans-
former can typically reach the 3% range. § By adapting the acceptance limit of the apparent
charge to the voltage class.
The field distribution will be homogeneous (e.g. be- § By use of the energy dissipated in the discharge.
tween windings) or semi-homogeneous (e.g. wedges
between turns and in-take leads). Typically the maxi- It has been assumed that surges may initiate discharges
mum fields occurring during service, (occurring e.g. at that afterwards will continue because the ac stress was
intake leads and at the windings’ ends), can be some 6 above the PD-extinction level. However, it has been
kVRMS/mm. In a core-form transformer the voltage will found that the ac level had to be very close to the PD-
– also in service - be in the order of 0.1 - 5 kVRMS be- inception to make it possible for an impulse to trigger
tween turns and some 4 – 5 kVRMS between discs. In sustained discharges [6, 7, 8].
shell-form transformer, the voltage between “discs” in
service is greater than this. The highest voltage would 2.2. Discharge types.
be in the order of 30 kVRMS. This is due to the differ- One may divide the discharge types in transformer in-
ence in winding arrangement. The number of “discs” in sulation into two types:
shell-form transformer is smaller than that of discs in • Discharges in an open oil volume (e.g. sliding dis-
core-form transformers. Therefore, the voltage distrib- charges on surfaces).
uted between “discs” (and therefore also clearances) • Discharges occurring within a confined space (e.g.
becomes greater than that between discs in core-form bubble, delamination or wedge).
transformer.
Often, discharges are thought of as very short duration charge at the opposite side that reduces the voltage
events. However, studies of discharges in open oil vol- across the void. This discharge type may, depending on
umes at impulse stresses, has shown that the final local conditions, be either a Townsend- or a streamer-
propagation length of such a discharge is related to the type discharge if it occurs in a vapour filled void. If it
total charge injected [9]. During the propagation many occurs in oil in a wedge, the event will - because the
current pulses (of some 10 nanoseconds duration) occur distance is short - develop over a short time. For these
and reilluminations are seen along the stems of the discharge types a measuring impedance with short time
streamer discharge structure [10, 11]. constant is adequate for a correct integration.

Figure 1 and Figure 2 show the structures of positive 3. INFLUENCE OF MEASUREMENT AND TEST
and negative discharges in open oil volumes, together CONDITION
with oscillograms of applied voltage, current and inte-
grated charge. The oscillograms show that, due to the 3.1. General considerations
time to reach the final size of the discharge, long time
constants are needed in the measuring impedance to 3.1.1. The frequency response of the measuring im-
obtain the properly integrated charge. The initiation pedance and discharge signatures.
voltage and the propagation velocity are different for When the instrumentation used is in accordance with
positive and negative discharges [11]. In addition, for ac IEC 60270 one can, because the instrumentation can be
stressing, two major types of discharges, namely the calibrated, talk about a real measurement. However, as
bubble- and the filamentary- type, both occurring at shown in Figure 3, even in the 10 – 1000 kHz regime,
positive and negative polarity, can be distinguished: the PD-magnitude measured in a liquid insulation –
• The slow bubble-type discharge, where the even in a calibrated system – is a highly questionable
“discharge” channel/void contain mainly vaporised parameter [16]. As shown, if the integration time con-
liquid [12]. stant is too small the discharge magnitude will be meas-
• The faster filamentary discharge [12], where there ured as smaller than it in reality is.
is a high temperature plasma - resulting in full dis-
sociation - within the channels [13]. This may, as shown, have a significant effect especially
Studies have shown that the discharges occurring in a when, in an attempt to eliminate interference due to ex-
point to plane gap under impulse and ac condition are ternal “noise”, very high frequencies are chosen for the
similar [14]. For this discharge type it is normal to dif- detector.
ferentiate between the initiation and the propagation
100
phase of the discharges (or streamers).
80
Air
60
[%]

40

20
Oil
Figure 1: a): Discharge from an earthside positive point, 0
b): oscillogram from positive discharge. 1E+004 1E+005 1E+006
Frequency
Resonance [ Hzf0] [Hz]
frequency

Figure 3: Percentage of the “real” PD charge measured


with RLC impedances of various resonance frequency
(corona discharges in air (upper trace) and oil (lower
trace) in a point-plane gap)

As a further exemplar of this phenomenon, consider


Figure 4 and Figure 5. Figure 4 illustrates the typical
Figure 2: a): Discharge from an earthside negative point phase resolved amplitude distribution of discharges in a
b): oscillogram from negative discharge. point-plane gap and shows major variations in the am-
plitude and repetition rate between positive and nega-
In the case of voids occurring in the insulation system tive half cycles. The ratio between the level of positive
(e.g. bushing, delaminations), the discharge type will be and negative discharges is a characteristic feature in the
more similar to gas discharges occurring in solid insu- discharge pattern from a sharp point. By changing the
lation. There will normally be one single current pulse frequency response of the measuring chain this ratio
associated with the gap breakdown [15], depositing a will be influenced as shown in Figure 5. This fact shows
that a documentation of frequency response for a meas- given measurement, the resolution has to be better than
uring system is essential. In particular this is of rele- that which can be achieved with an 8 bit converter.
vance when taking modern VHF and UHF methods into
use. 3.2. Test Condition

3.2.1. Voltage shape


To control the voltage shape is important. The shape of
the applied voltage is frequently seen to be distorted
both during an induced test for a transformer and during
model tests in the laboratory using test transformers.
Distortion of the test voltage may heavily influence the
PD-pattern [17]. Since a distorted voltage-shape does
not necessarily show in the crest-value of the voltage,
the shape must be inspected on an oscilloscope. This
effect, in real transformer testing, is illustrated in Figure
6. Here it is seen that the highest pulse rates occur
where the voltage gradient is steepest. The pattern will
be different without overharmonics.
100
100

80

Number of discharges
50
Voltage p.u.[%]

60
0
Figure 4: Phase-resolved signature from a sharp metal- 40
lic point recorded over a 300 second period. -50
20
4,5
-100
Qavg(+)/Qavg(-)
4,0 0
Qmax(+)/Qmax(-)
3,5
0 90 180 270 360
3,0 Phase angle [deg]
2,5
Figure 6: PD-pulse number (full drawn curve) vs. phase
2,0
for distorted voltage shape (dotted curve) during in-
1,5
duced test.
1,0
0,5
4. PD PATTERN ANALYSES TECHNIQUES.
0,0
8 MHz

10 MHz

20 MHz

45 MHz

75 MHz

90 MHz
IEC 60270

Pattern recognition of signatures from partial discharges


involves no single technique. There are a multitude of
Center Center freq.
freq. of of narrow band
narrowband (3 Mhz BW)
(3 MHz BW) filter
filter signatures and ways of analysing these.

Figure 5: The effect of the frequency response on the The first significant “expert system”, developed for
ratio of discharge amplitudes observed in positive and analyses PD signatures observed on an CRT, was pub-
negative half cycles lished by CIGRÉ in 1969 [18]. In the late 80’s comput-
erised data acquisition became commonly available, fa-
3.1.2. Measuring time cilitating effective acquisition, storage and processing
This must be as long as is feasible since discharge repe- of discharge patterns, e.g.:
tition rates in oil can be very low (see Figure 4). Cer- • Phase-Resolved Partial Discharge recording [19]
tainly >10 minutes and, as with factory tests, preferably which has been combined with various computer-
from one half to one full hour. This allows the pattern to ised pattern recognition methods, e.g. statistical
develop. However, it must be remembered that this is a analyses [20], neural network analyses [21], and
presentation of “statistical” phase position and masks fractal analyses [22]. The basic concept is that the
any temporal effects, e.g. bursts of discharge activity. discharge occurrence is statistically variable, but
over time locked to the power cycle. Patterns from
3.1.3. Dynamic range sources with low discharge rates are easily re-
Discharges in oil do appear with a large dynamic range. vealed.
To get both a good depiction of a pattern of large dis- • Frequency resolved discharge patterns have been
charges and small ones typically prevalent within a used to find “slots” in the frequency domain where
• noise is low. It has also been proposed for distin- formers in the field. To initiate discharges there has to
guishing between different defects in electrical ma- be a defect in the insulation system. Such defects in-
chines [23, 24, 25]. clude:
• Pulse sequence analyses with “voltage difference” Delamination may occur when thinner pressboard
plots have proved useful for revealing space charge sheets are glued together to form thicker barriers. Voids
effects due to discharges in solid insulation [26]. within delaminations may remain for a very long time.
For transformers it has been shown that discharge Voids may occur variously. They may occur in glue and
data of this type gives patterns with clear structures in connections with enamelled thread, furthermore in-
such as that shown in Figure 7. The method is less sufficient impregnation may give rise to voids. Some-
suited for situations with low discharge rates (i.e. < times (i.e. during commissioning) it is found that voids
1 per cycle) and for cases where conduction cur- disappear from one day to the next, as gas is absorbed
rents are significant compared to the charges re- by the oil and the cavity is filled with oil. Voids may
leased in the discharges. It has also been reported also occur in bushings. High moisture content combined
that these signatures are sensitive to the threshold with heat and high fields may create local high water
of the data acquisition system. vapour pressure and “puffing” effect of boards and
• Time resolved partial discharge recording [27] will winding insulation. Discharges in this spongy material
in principle offer possibilities to establish all of the may start thereafter.
patterns mentioned above. Bubbles may occur due to gas evolution from dis-
charges, and evaporation of water droplets. Experience
Phase resolved partial discharge plots are the most shows that a gas bubble in an open oil volume will be
common in use. Their advantage is that they easily may ripped into smaller bubbles which will quickly vanish
be combined with simple physical models and reason- when the first discharge occurs inside this bubble. Bub-
ing (in addition to or instead of synthetic intelligence), bles will therefore only exist at locations where they are
and they resemble what has previously been observed supported mechanically by solid insulation (e.g. in
on a CRT. wedges).
Free metallic particles may be left from the production
In liquid dielectrics, discharges have been seen to ap- process. Smaller particles will be carried by oil flow,
pear stochastically in bursts, and also to change with while larger ones will sink to the bottom. Dielectropho-
time in a systematic way. In addition, the electric field retic forces will attract high permittivity and conductive
distribution is influenced by space charges injected into particles towards high stress regions. However, free
and carried by the liquid. For these reasons, it is ex- particles can easily adhere to surfaces e.g. the surface of
pected that time resolved discharge recording will sup- a winding.
ply a new and necessary axis for an improved charac- Fixed metallic particles (e.g. in wood details, or fixed to
terisation of discharge behaviour in impregnated insu- paper in windings) may occur.
lation in the future. Moisture may occur due to ageing and may also be in-
troduced into the insulation during site erection. Mois-
ture will contribute in several ways to discharge incep-
tion and extinction. During a heating cycle in service
(e.g. when energising a transformer that has been off-
line), moisture may be pressed out of the solid insula-
Ui+1 - Ui tion. Due to poor solubility in the oil there will be a su-
per-saturation of the oil next to the cellulose surface re-
sulting in water droplets and bubbles [3]. Evaporation
of the water will give micro-bubbles giving rise to dis-
charges. Increased moisture content in the cellulose will
Ui – Ui-1 make the cellulose more conductive. In these condi-
tions, cellulose fibres will start acting like metallic par-
Figure 7: Pattern experienced during transformer test- ticles. Fibres sticking out from paper surfaces and fibres
ing, believed to origin from a 100 – 150 pC surface dis- moving in the oil may initiate discharges. Increased
charge. moisture content will also increase the dielectric losses.
During a cooling cycle, water may condense and be lo-
5. EXAMPLES OF PD PRODUCING DEFECTS IN cally absorbed by pressboard. Pressboard has been seen
TRANSFORMERS AND THEIR PATTERNS to “puffed up” due to heat developed by dielectric
losses and evaporation of water within the pressboard.
5.1. PD sources in a transformer. As a consequence internal discharges may become
In a well-designed transformer insulation, which is prevalent in the cavities within the pressboard.
properly dried and impregnated, it is very difficult to Bad connection of electrostatic shields will give large
initiate discharges. However, discharges do occur in discharges because the capacitance of the defect is
transformers as found both from PD-measurement dur- large. The “bad connection” will usually have a defined
ing commissioning and from gas analyses from trans- breakdown voltage (UBD). The result is discharges ap-
pearing on the rising flank with voltage independent charge effects from the preceding half cycle are mini-
amplitudes (Q= C⋅UBD). mal. The discharge amplitudes do increase with voltage
along the rising flank. Sometimes Trichel-like behav-
Static electrification will give rise to local charge depo- iour can be seen if there is more than one discharge per
sition. The resulting field enhancement may initiate dis- half cycle.
charges. Discharge tracks due to such discharges have
been seen along pressboard surfaces. Once the discharges occur in a void or in a wedge,
Surface tracking resulting from discharge propagation space charge effects start to move the centre of the pat-
has been found along barrier surfaces and along sup- terns and the phase start point for discharges down the
ports. Carbonised tracks can act as conductive protru- rising flank of the applied voltage, and even into the
sions, and may over time increase in length. preceding half cycle.
110.000mm 110.000mm

5.2. Models of defects 10.000mm 10mm

To investigate the likely behaviour of a plant insulation

23.000mm

1.000mm

1.000mm

1.000mm
system with defects, model studies are usually per-

23mm
a) b)
formed. In any model experiment sample preparation
should be well controlled [28] and the scaling problem R5mm R5mm

has to be considered. Even if relevant field stresses are


110.000mm 110.000mm
38mm
10.000mm

applied, the capacitive coupling between electrodes and

23.000mm

1.000mm
a defect will be different in a model. In a real HV trans-

1.000mm

1.000mm
c) d)

20mm
former the insulation distances are longer, thus the ca- R21mm

pacitances between the defect and the electrodes will be R5mm


R5mm

much smaller than when the same defect is introduced

2.662mm
110mm 40.000mm

in a small-scale experiment thereby limiting the dis- f)

26.000mm
charge currents. The charges created from charge sepa-

29.300mm
ration in the head of a propagating streamer must either e)

3mm

5mm
be compensated by conduction or displacement current
in the “other end”. R5mm

5.3. Behaviour and patterns from defects


Three investigations have been undertaken by members Figure 8: Some examples of pd-producing models.
of the TF to determine whether correlations exist among a) point to plane gap, b) rod to plane with pressboard,
the different types of the discharging defect which can c) elevated rod, d) wedge with VDE electrode
occur in power transformers and the associated dis- e) oil-filled void, f) interturn
charge patterns. It should be stressed that these were
independent studies and not undertaken as a TF group The pulse repetition rate is low for single large defects
activity. Differences in the outcome of the results may like sharp points (possibly due to lack of starting elec-
among other things depend on the local field in the trons), while it will be high for distributed defects (e.g.,
tested models. over-stressed wedge/multiple voids) and for
“capacitances” (e.g. large voids, floating objects) at
In the first, L.Lundgaard and colleagues have investi- high overstress above the inception voltage.
gated a range of geometries representing variously de- The discharge level has been seen to be high (i.e. up to
fect conditions in a power transformer. Six of these thousands of pC) for larger injected bubbles and cavi-
geometries are shown in the schematic diagram, ties where gaseous by-products are unable to escape.
Figure 8. The remainder involved various plane-plane Creeping (surface) discharges also produce large dis-
gaps with either bare or insulated electrodes with con- charges. Small voids and wedges show discharges of
ducting particles present. some hundred pC.
During the experiments, PRPDA (φ, q, n) recordings Polarity dependent differences in discharge level are
were taken at different voltage levels, and the maximum seen from sharp electrode bound defects. At inception
and average amplitudes and discharge frequency (repe- the negative discharges may dominate, while at higher
tition rate) were calculated for each polarity and plotted voltages when patterns are seen in both half cycles, the
vs. voltage. discharges will usually be bigger when the metal sur-
face has a positive polarity. This is attributed to easier
From these studies, a variety of observations can be propagation for positive discharges.
made, based on the specific measured parameters of Temporal instability/ burst behaviour is seen for many
discharge activity. These are as follows: types of defects. No systematic investigations of this
Phase angle of discharge pattern; When discharges oc- property have been done for these kinds of geometries.
cur at a sharp point into the oil or along a pressboard However, at elevated moisture levels surface discharges
surface the largest amplitudes in the patterns are centred are seen to vanish with time, and reappear after a resting
around the voltage peaks like shown in Figure 4. Space period. This behaviour is attributed to drying of the pa-
per surfaces due either to dielectric losses or heat from When first seen, it occurs at the zero-crossing. At higher
the discharges. voltages this structure (1) is shifted to the left due to
charges remaining from discharges in the previous half-
In addition to these general observations, the study gen- cycle. Additionally, “structure 2” also appears when
erated some patterns which is considered worth repro- voltage is raised. The valley between 1 and 2 may be
ducing more completely here. explained by “corona stabilisation” or it may simply be
Wedge discharges another location starting to discharge. Both these struc-
Typically, wedge discharges (Model 8f) in a dry wind- tures show that the number of discharges decays
ing will give patterns like those shown in Figure 9a monotonically with increasing amplitudes. Structure 1
[29]. When raising the voltage the sharp “structure 1”, and 2 resemble what is seen from a wedge model made
(possibly indicating a defined threshold and abundant from a covered VDE electrode on a pressboard covered
starting electrons [30]), is the first structure observed. plane electrode (model 8d) (Figure 9c).

At increased moisture levels (4 % in pressboard) the


structure (model 8f) is changed (Figure 9b). The pat-
terns do not start before the zero crossing.

Surface discharges
Sliding discharges produced from a rod placed onto a
Structure 2 pressboard covered electrode has been seen to produce
patterns like those shown in Figure 10. The fan-shaped
Structure 1
larger discharges are associated with creeping dis-
charges. The dense lower level pattern has a high dis-
a)
charge rate, while the larger discharges occur at a low
rate reminding of what is seen in a point to plane gap.

b) a)

c)
b)

Figure 9: Patterns from paper-oil-paper wedges, a) In-


terturn discharge <1% moisture, b) Interturn discharges Figure 10: Signatures from rod to plane models a) Low
≈4% moisture, c) Paper covered sphere towards press- Moisture 10 kV, b) High moisture, 16 kV.
board covered plane 7% moisture.
Effect of time.
In addition to the preceding statistical studies, temporal
stability was investigated for several models. The gen-
eral behaviour is one of high discharge level after a
voltage step or increase with the PD activity thereafter
ceasing. If model objects thereafter are left unenergized
overnight then they return to their virgin state (provided
no irreversible ageing has taken place). This process is
believed to be caused by either dielectric losses or dis-
charges heating the pressboard surface and thereby
drying it. Leaving the object at rest overnight re- Key
established its “virgin” state.

The other investigation of model geometries has been


undertaken by I. Kemp and colleagues with funding
from The National Grid Company plc. [31]. In their in-
vestigation, they utilised 10 geometries to simulate de-
fect conditions in an oil-filled power transformer. These
are listed and shown schematically in Figure 11.
Model 1 Model 2
The data was initially correlated as phase resolved plots.
In addition, different mathematical distributions were
subsequently investigated, chosen on the basis that in
previous investigations of other plant systems results
have indicated that these may be relevant to classifying
a discharge fault condition. Parameters investigated
were PD frequency (repetition rate); Charge Mean,
Standard Deviation, Skewness and Kurtosis; Start, End
and Mean Phase relationship and Phase Width; and
Model 3 Model 4
Phase Standard Deviation, Skewness and Kurtosis.

Lack of space precludes the reproduction of all data on


a test by test basis for this study. However, from a rig-
orous analysis of all the data produced, the following
observations can be made:
(1) corona-type discharges in air or oil can always be
distinguished from each other and from other forms of
fault condition resulting in discharge activity.
(2) of the other fault conditions investigated, although Model 5 Model 6
each in some tests produced unique features, these were
not sufficiently consistent to have confidence in distin-
guishing all conditions from the discharge data. This
inconsistency stemmed primarily from changes in the
discharge pattern associated with different stress levels
in the test programme or with “ageing” as the test pro-
ceeded. Both of these were also noted as important in
the other investigations. Since neither of these could be
inferred in a “real” situation, although these could be Model 7 Model 8
viewed as consistent within a laboratory test pro-
gramme, they must be viewed as inconsistent for any
practical purposes.
(3) if the ten fault conditions were, however, sub-
grouped into one of four categories, i.e. enclosed type
discharge (Models 1, 2, 3, 4 and 5), gap-type (void) dis-
charge (Models 8 and 9), discharges from a sharp pro-
trusion (6) or surface-type discharge (Model 7) then,
with confidence, each of these sub-groups could be dis-
Model 9 Model 10
tinguished from the discharge data irrespective of the
stress level or “ageing”. The models were:
Figure 11: Discharge-inducing defect geometries
Model 1: Sharp point causing void. pC. (Of course, as an indication of a particular type of
Model 2: Sharp point embedded in insulator. fault condition, discharge magnitude must be treated
Model 3: Punctured paper increasing stress. with care as indicated earlier).
Model 4: Loose paper causing void.
Model 5: Winding movement. In addition to the two preceding investigations, James
Model 6: Conductor to conductor discharge. and colleagues [32] investigated the effects of moisture
Model 7: Surface discharge. and temperature changes in the PD characteristics of an
Model 8: Void adjacent to insulator. oil impregnated insulation system simulating a power
Model 9: Void on conductor surface. transformer duct using 3 mm thick pressboard spacers
Model 10: Corona gas discharge. between paper covered conductors and an upper insu-
lated electrode. The paper thicknesses were 0.6 and 0.5
Interestingly, not all of the discharge data was required mm respectively. Phase patterns were recorded at
to enable the 4 sub-groups of discharging fault condi- overload temperatures of 150ºC (established by current
tions to be distinguished. circulation through the copper conductors) before and
after a sudden increase in PD magnitudes (Figure 12a,
Most of the features yield information that may be use- b). To avoid possible failure the temperature was low-
ful in determining an unknown fault in a high voltage ered to the bulk oil temperature of 80ºC and the test
power transformer. Certain features, however, perform continued for 400 hours at 200 Hz.
better than others and provide consistent information on
the type of pd source. The implementation of an auto-
mated identification system, using a restricted set from
the features listed, should provide the opportunity of
producing an instrument capable of carrying out a more
rapid identification of fault type.

The features which offer the best performance are as


follows:
Frequency of discharge (number per half cycle) – a
very low repetition rate indicates a surface type dis-
charge, a rate between 1 and 20 indicates a gap type
discharge and a rate above 20 indicates an enclosed type
discharge. Over 100 per half cycle indicates discharges
from a sharp protrusion. Air corona was of the order of
50 per half cycle.
Phase mean – an enclosed type discharge is indicated
by a phase angle ≅ 45°, a gap type discharge is indi-
cated by a phase angle between 50°-70°, between 70°-
90° indicates a surface type discharge. A phase angle
>90° indicates air or oil corona.
Start phase – a consistent value of around 0° for both
positive and negative parts of the ac cycle is indicative
of either an enclosed type or a surface type discharge
that is not close to the inception voltage (examination of
the pd frequency will allow separation of the fault type).
An inconsistent value of over 10° on both halves of the
cycle indicates a gap type discharge. Differing values
on each distribution may be characteristic of air or oil
corona.
Phase width – a value of >110° is highly indicative of
an enclosed type discharge. Values of <100° are char-
acteristic of either gap or surface type discharges (these
discharges can be more conclusively identified by com-
paring the start phase.
Charge mean – although this feature in isolation cannot
provide identification of a fault type, used in combina-
tion with other features it provides greater confidence in
Figure 12: PD phase pattern during and after overload.
fault identification. Enclosed type discharge consis-
a): 3 minutes before large discharges at 150oC (2049
tently give higher values, over 100 pC. Values between
sec), b): 5 second during large discharges at 150oC
20 pC-80 pC are indicative of a gap type discharge. Sur-
(2210 sec), c): 80°C, after overloading (55 hours).
face type discharge produces values from 30 pC to 180
After completion of the test it was confirmed that bub- defects likely to be found in oil-filled power transform-
bles were trapped under a spacer at the pa- ers exist. Despite the fact that the empirical evidence
per/pressboard interface and that discharging had oc- has highlighted parameters that require to be controlled
curred in this location. No marking was detected at the and which weren’t systematically controlled among
conductor surfaces. these studies, there remains evidence of correlations
among these studies. Inspection of the results under 5,
It will be noted that the “rabbit’s ear” in Figure 12a is in particular, for similar geometries, show distinct
present throughout the series, the lowering of the tem- similarities. Certainly the evidence is strong enough to
perature only eliminating the large discharges. Such a suggest that a more rigorous approach, with systematic
pattern was also observed in a different test rig in which variation of the parameters now recognised, in hind-
gas was trapped within a cavity [33]. sight, through the results of these studies to influ-
ence/control the processes involved (and hence influ-
The total time during overloading at 150oC was about ence the pattern of discharge activity), would yield the
2000 s. During this period, the integrated quantities correlations to make full potential of modern digital
(peak PD magnitude qm, average PD magnitude qa, cur- partial discharge measurement instrumentation. Inter-
rent I, power P, quadratic rate D and repetition rate r) estingly, the results suggest that it is the primary, core
increased steadily. These were evaluated separately for measurements of partial discharge activity, i.e. magni-
the two groups of half-cycles. The positive half-cycle tude, phase relationship and repetition rate, which are
values were consistently larger than the negative half- important in correlating specific discharge patterns to
cycle values. An exception was the repetition rate which defects. It is not the more complex mathematical distri-
displays an unusual pattern. In the initial stage of over- butions of these “primary” φ, q, n patterns, e.g., skew-
loading from t ≅ 350 s to t ≅ 1100 s, r+ and r- were very ness and kurtosis, within different axes of the φ, q, n 3D
similar. From then on, there was a distinct change in the distribution.
trend where the repetition rate became larger for the
negative half-cycle, see Figure 13. This may have indi-
cated that the material damage was already occurring. 7. DAMAGE TO THE INSULATION SYSTEM

3000 Even once detected, the risk to the remnant life of the
r insulation from a given defect remains to be assessed.
2500 Little evidence is available for transformer insulation, in
r+
this aspect, in the literature. However, a range of obser-
2000 r-
vations can be made towards this objective.
1500
In a point-to-plane gap the current pulses of some hun-
1000 dred microamperes are seen for small bubble type dis-
charges. When filamentary ones occur, current pulses
500 (in the 10 ns range) can be in the ampere range. If the
0 discharge energy is in the microjoule range only x-wax
is created and methane, H2 and C2H2 is formed. Dis-
0 1000 2000 3000 charge energies in the 1 – 10 millijoule range are seen
to destroy paper and unsaturated hydrocarbon gases are
Figure 13: Discharge repetition rate vs. time (seconds) produced.

The overall pattern developed further, becoming more If the gas formed by the discharges in the oil is trapped
intense and skewing to the voltage zeros with an in- in the insulation, voids can be created. If the void then
crease in the negative cycle magnitudes compared with starts to discharge then there will be a steady bombard-
those in the positive, compare ment of the insulator surfaces enclosing the void re-
Figure 12a, b and c. Some increase in maximum PD sulting in bond breaking of the cellulose molecules.
magnitudes to 2,5 nC was noted after 100 hours but this
reduced to less than 2 nC and remained constant. The Rupture of fibres and pitting of the surfaces involved (in
average electric stress in the uniform part of the field a point plane gap with the plane covered by cellulose)
(paper/spacer region) was approximately 2.5 kV/mm has been observed, and the erosion was observed to be
throughout the test period. proportional to the discharge energy [34].

Carbonised tracks both on surfaces and within press-


6. DISCUSSION board sheets has been seen in both models and real
transformers [35]. A stack of paper was found to have
From these various model studies which have been un- excellent corona resistance compared to oil-
dertaken, there is evidence that correlations between impregnated (non layered) pressboard sheets.
discharge patterns of activity and discharge-inducing
Ageing from discharges is less severe to the oil, as the of the discharge in the windings are under study, and
oil can be considered self-healing and the oil easily may the first results are promising for the application to ac-
be reclaimed. However, gas formation is an indication tual transformers.
that something is wrong within the insulation system.
One would expect that the gas formation to be different Last, one should not forget the modern VHF and UHF
in a “bubble” discharge from that of a “filamentary techniques applying antennas for detection of dis-
streamer”. In a bubble the discharges occur within a charges. These techniques are believed to be suited for
confined vapour filled “void”. Here the deterioration is in-service measurement.
mainly due to hot electrons hitting the walls and result-
ing in bond breaking of the oil molecules. In the fila-
mentary streamer the channels are believed to be filled 9. CONCLUSIONS
with a plasma (of several thousand degrees) with a high
degree of dissociation. In the latter case one will get hy- As indicated in the studies by task force members, there
drogen and various recombinations of hydrogen and is tentative evidence that correlations do exist between
carbon. (It is likely that these conditions are of higher elements of discharge patterns from different discharge-
relevance to the gas formation than the mere energy in- producing fault/defect conditions to be found in oil-
put.) The high temperatures of the filamentary stream- filled power transformers. As with all early investiga-
ers may result in damage to the cellulose like e.g. car- tions, in addition to providing valuable insight into the
bonised tracks. relationships between discharge patterns and specific
defects, these studies have provided significant insight
into the parameters which must be controlled in future
8. PRACTICAL APSECTS studies to ensure that data is both accurate and objec-
tive. Temperature and moisture content in both the oil
Even in the laboratory there may be difficulties sepa- and any solid insulating materials are important, as is
rating external noise signals from real discharges in a the voltage level. In addition, in making such investiga-
transformer. This is always the main problem when tions, partial discharge measurement instrumentation
measuring discharges on-site. Several possibilities ex- and, in particular, the bandwidth of such instrumenta-
ists: tion, must be given careful consideration. Finally, care
should be taken to ensure that during extended tests, if
Acoustic methods are used to measure sound waves there is degradation within the discharging defect,
from the discharges. When the discharge is correlated to physical/chemical changes associated with the degrada-
the acoustic signal, the source is always inside the trans- tion are monitored. The degradation may change the
former tank. With acoustic methods it is also possible to nature of the discharging defect, with associated
locate the discharges by triangulation, and a precision changes to the discharge pattern.
better than 10 cm may be obtained. The success of this With further laboratory studies under the controlled
method is based on the sensitivity of the acoustic sen- conditions detailed above, coupled to practical dis-
sors. This method has been used during factory testing charge pattern measurements from discharging trans-
for some tens of years, and is now also emerging for on- formers in the field, from the investigations of this TF,
site measurements. Experience shows that the method is it is likely that the potential of discharge pattern meas-
less sensitive for sources inside the winding structure. urements to determine the extent, nature and form of
discharging defects in power transformers, will be real-
Simultaneous multiterminal electric PD detection has ised.
successfully proven useful for pinpointing a discharge
to a specific winding and phase. Comparison between REFERENCES
HV and ground side signals can in some cases also sup-
ply information for location. [1] CIGRE WG12.01: ”Measurement of partial dis-
charges in transformers”, part 1 & 2, Electra no
During one half-cycle discharge signals of opposite po- 19, pp 13-65, 1971.
larities may occur, indicating more than one PD source. [2] Y.Sheiretoc, M.Zahn; “Dielectrometry measure-
For analyses like this it is important to have a high dy- ments of Moisture Dynamics in Oil-impregnated
namic range of the measuring equipment, as signal lev- Pressboard” IEEE TDEI, Vol. 2, No. 3 June 1995,
els easily may vary with a factor of 100. pp 329-351.
[3] V.G.Davydov, O.M.Roizman W.J.Bonwick;
Frequency analysis of the discharge signal can be used ”Transformer Insulation behaviour during over-
to sort discharges into families. An expert is able to say load”, EPRI Substation Equipment Diagnostic
whether the discharges are coming from the inside or Conference V, TR-111282, 1998, pp. II-53 to II-
the outside of the transformer. 72.
[4] J.Fabre, A.Pichon; “Deteriorating Processes and
Some new methods based on the change of the fre- products of paper in oil. Application to Trans-
quency spectrum of the signal linked with the position formers.”, Cigre, report 137, 1960.
[5] S.A.Boggs: ”Partial Discharge – Part III: Cavity- grated and statistical parameters”, 8th ISH, Vol. 3,
Induced PD in Solid Dielectrics”, IEEE Paper 62.06, Yokohama, 1993.
El.Ins.Mag., Nov,/Dec., Vol.6 No.6. 1990, pp 11- [22] L.Satish, W.S.Zaengl:”Can Fractal Features Be
20. Used for Recognizing 3-d partial Discharge Pat-
[6] P.Hessen, W.Lampe; ”Partial Discharges Trig- terns”, IEEE TDEI, Vol2, No.3, June 1995, pp
gered By Switching Surge In Power Transform- 352-359.
ers”, IEEE Trans. PAS Vol. PAS-97, No.3 May- [23] R.T.Harrold,F.T.Emery, F.J.Murphy, S.A.Drinkut:
June 1972, pp1225-1234. ”Radio Frequency sensing of incipient arching
[7] S.Cesari, S. Yakov: ”Switching Surge Response faults within large turbine generators”, IEEE PAS
Of Transformer Insulation Designed On The Basis Vol. 98 No. 4 July/Aug 1979. pp 1167-1173.
Of Power Frequency Pd Inception Volt-Time [24] I.J.Kemp, J.W.Wood, H.Zhu, H.G.Sedding and
Curves.”, IEEE Trans PAS, Vol. PAS-100, No 7, W.K.Hogg, “Towards a new partial discharge
July 1981, pp 3263-3273. strategy based on the transfer function of machine
[8] Y.Yanari, M.Honda, Y.Tanigychi, Y.Ebisawa: stator windings”, Proc.IEE Sci. Meas. And Tech-
”Problems Of Long Term Reliability For UHV nology, Vol.143, No.1, pp57-62, 1996.
Transformer Insulation”, IEEE PAS, Vol. Pas-102, [25] C.Zhou and I.J.Kemp, “Theoretical investigations
No 6, June 1983, pp1693-1701. into propagation in large power tranformers”,
[9] O.Lesaint, G.Massala: ”Positive Streamer propa- Proc. Int.Conf. on Dielectrics and Insulation, Bu-
gation in Large Oil gaps”, IEEE TDEI, Vol. 5 dapest, Hungary, Sept.1997.
No.3, 1998, pp 360-370. [26] M. Hoof, B. Freisleben, R.Patch: “PD source
[10] P.Rain, C.Boisdon, O.Lesaint, R.Tobazeon: “Be- Identification with Novel Discharge Parameters
haviour of Streamers under Divergent ac Fields in using Counterpropagation Networks”, IEEE TDEI,
Transformer Oils at Large gaps”, IEEE Vol. 4, No. 1, pp 17-31, Feb 1997.
Trans.El.Ins., Vol.26 No.4, Aug 1991 pp 715-725. [27] J.Poittevin, P.Andre; “New Digital Partial Dis-
[11] D.Linhjell, L.Lundgaard, G.Berg: “Streamer charge Measurements on Transformer”, Cigré
propagation under Impulse Voltage in Long Point- 15/21/33-17, 1986.
plane Oil Gaps”, IEEE TDEI, Vol.1 No.3, June [28] ERA Guide 70-25: “Principles and guide to prac-
1994, pp 447-458. tice for vacuum impregnation of cellulose materi-
[12] O.Lesaint, P.Gournay: “Initiation and propagation als prior to testing samples", 1970.
Threshold of positive prebreakdown phenomena in [29] G.Berg, L.Lundgaard: ”Discharges in combined
Hydrocarbon Liquids”, IEEE TDEI, Vol.1 No.4, transformer oil/paper insulation”, 13th IEEE ICDL
1994, pp 702-708. Conf., Nara, 1999, pp 144-147.
[13] P.Bårmann, S.Kröll, A.Suneson: “Spectroscopic [30] F.Gutfleisch, L.Niemeyer: ”Temporal evolution of
Measurements of Streamer Filaments in Electric PD in sperical epoxy voids”; 9 th ISH, paper 5667,
Breakdown in Liquid”, Journal of Physics D (Ap- Graz. 1995.
plied Physics). 1996 , 29 (5) , pp. 1188-96. [31] J.R.Hyde, “Condition Monitoring of High Voltage
[14] L.Lundgaard, O.Lesaint: “Discharges in Point- Power Transformer Insulation Systems by Partial
PlaneGap under ac and Impulse Stress”, IEEE Discharge Measurements”, PhD work in process,
CEIDP, Virginia each, 1995, pp 596-599. Glasgow Caledonian University.
[15] F.Colla, M.Pompili,C.Mazetti, R.Bartnikas: [32] R.E.James, Z.Liu, B.T.Phung and T.R.Blackburn,
“Partial discharge behaviour of oil impregnated "Effect of moisture and temperature on the PD be-
paper bushings”, IEEE Int. Symp. on Electrical haviour of an oil-impregnated insulation system",
Insulation, Arlington, 1998, pp35-38. Australasian Universities Power Engineering Con-
[16] H.Debruyne, O.Lesaint: ”On the significance of ference, Sydney, 29Sept.-1Oct. 1997, pp.607-614,
PD-measurements”, 11th Int.Symp on High Volt. ISBN 0 7334 1594 6.
Eng. ISH’99, Vol. 5, pp 45-48, London,1999. [33] B.T. Phung “ Computer-Based Partial Discharge
[17] M.Florkowski, ”Distortion Of Partial Discharge Detection and Characterisation” Ph.D. Thesis,
Images Caused By High Voltage Harmonics”, 1998, School of Electrical Engineering, University
ISH87, Montreal, pp 95-98. of New South Wales, Sydney 2052, AUSTRALIA
[18] Kreuger/WG21/03; ”Recognition of Discharges”, [34] F.Viale, J.Poittevin, B.Fallou, F.Morel,R.Buccanti,
Electra, no. 11, Dec 1969. pp 61-98. S.Yakov, S.Cesari. E.Serena: “Study Of A Corre-
[19] B.Fruth, J.Fuhr: “Partial discharge pattern recog- lation Between Energy Of Partial Discharges And
nition. A Tool for Diagnoses and Monitoring of Degradation Of Paper-Oil Insulation”. Cigré, pa-
Ageing”, Cigre report 15/33-12, Paris, 1990. per 15-12, Paris, 1982.
[20] E.Gulski: “Computer-Aided Recognition of Partial [35] N.Izeki, A.Kurahashi, K.Matsuura: ”Behaviour of
Discharges using statistical Tools”, PhD, Delft oil corona and damage of Transformer Insulation”,
University Press. 1991. Paper 71 TP 203 PWR, Winter meeting 1971, pp
[21] B.T.Phung, T.R.Blackburn, R.E.James: ” A neural 2330-2338.
network for discharge recognition based on inte-

You might also like