Professional Documents
Culture Documents
Uno Gäfvert
ABB AB, Corporate Research*
SE-721 78 Västerås, Sweden
E-mail: uno.gafvert@se.abb.com
Abstract: Real insulation systems are often transformers and the glass-mica-epoxy system of
heterogeneous and sometimes non-linear. In this paper machine insulation.
we discuss the application of dielectric frequency Most real systems are, in fact, to some extent non-linear.
response methods to a number of such practically The linear behaviour can only be considered as a low
important systems. field limit. Examples of such non-linear dielectric systems
Important examples of composite insulation systems are are dielectric liquids in narrow ducts and insulation in
the oil/cellulose insulation of power transformers and cross-linked polyethylene (XLPE) cables with water
the glass-mica-epoxy system of machine insulation. trees.
A few examples of non-linear effects and their Partial discharge and dielectric response measurements
implication for dielectric response analysis are discussed. are two complementary methods for assessment of high
These include narrow liquid ducts in transformer voltage insulation. Important relations between the two
insulation, exhibiting a voltage dependent conductivity, methods are discussed in the final section of this paper.
polymer cables with water trees exhibiting non-linear
COMPOSITE SYSTEMS
behaviour that is the basis for diagnosis . The relation
between partial discharges and dielectric response is Many practical insulation materials are composites or
elucidated and finally the special problems of combination of two or more materials. The structure of
measuring non-linear field grading material is discussed. the insulation has fundamental influence on its dielectric
response. For (more or less) random mixtures there are
INTRODUCTION established formulas [4] and modelling schemes [5, 6]
Dielectric response stands for a family of methods used for calculation of the response. The general problem is
for characterisation of dielectric materials as well as to calculate the response of the composite from
practical insulation systems. Classical methods of knowledge of the properties of its constituents.
measuring dielectric response include (power frequency) Two examples from “the real world”, transformer and
capacitance-Tan(?) and insulation resistance. More machine insulation, are used to illustrate the study of
recently extensions such as measurements of dielectric insulating systems consisting of several components. In
frequency response, polarisation and de-polarisation spite of their complexity the overall behaviour can be
currents and return voltage have found increasing reasonably well understood based on geometric
popularity. Several authors have pointed out that all of considerations and knowledge of the dielectric
these methods are “in principle” equivalent [1, 2, 3] and properties of the constituent materials.
that may be true in the ideal case. In this paper the
discussion is based on dielectric frequency response Moisture in power transformer insulation
where capacitance and loss are measured as a function An important application of dielectric response analysis
of frequency. In case of non-linear response, the is to assess the moisture content in the cellulose in
harmonics in the response are important for the analysis. transformer insulation [7, 8]. High moisture content
The ideal sample for a dielectric response measurement accelerates ageing and lowers the breakdown strength
consists of a single, linear, homogenous material in of the insulation. In a power transformer the combination
parallel plate capacitor geometry. “Real” systems are of materials is, of course, far from random. The oil
generally composite and often non-linear and may have cellulose insulation system of a power transformer is an
a complex geometry. In this paper a number of such example of an inhomogeneous insulation system where
“real” situations, typical of practical insulation systems, the structure has a major influence on the dielectric
are discussed. Important examples of heterogeneous response.
insulation are the oil/cellulose insulation in power The basic structure of this insulation is illustrated in Fig.
*
Also at Kungliga Tekniska Högskolan, Stockholm
1. The barriers and spacers consist of oil impregnated
cellulose material and the whole structure is immersed in
oil.
3
10
Spacers C' Meas.
2
10
C'' Meas
0
10 C'' XY-mod
-1 C' FEM
10
Inner Winding
C'' FEM
-2
10
-3 -2 -1 0 1 2 3
10 10 10 10 10 10 10
Fig. 1 Section of the insulation duct between Frequency [Hz]
transformer windings with cylindrical barriers and
Fig. 3 Frequency dependent capacitance and loss
axial spacers.
modelled with the XY-model and frequency dependent
For modelling purposes we can use the simplified FEM calculation compared with measured data.
“insulation module” (Fig. 2). The parameter X is the
The four capacitors in the XY-model do not provide a
relative amount of solid insulation (barriers) in the radial
true representation of the electric field distribution. A
direction and Y is the relative amount of spacers.
finite element (FEM) calculation [9] can be used to get a
Y 1-Y more realistic field pattern. Instead of four elements the
mesh now contains a few thousand elements. In Fig. 4
the field patterns calculated at 50 Hz and at 0,005 Hz are
Spacer Oil 1-X
shown. The same dispersive materials properties (?oil,
?spacer and ?barrier) are used as in the XY-model. At 50 Hz
X
Barrier the permittivities (the real parts) control the field
distribution. The field in the barrier is therefore lower
Fig. 2 XY-model: Insulation module used for than in the oil. At 0,005 Hz the field is controlled by the
calculating the response of transformer insulation. conductivities and it is concentrated in the solid
insulation.
With knowledge of the dielectric properties of the solid
material (barriers and spacers) and that of the oil one can a)
easily calculate the response of the transformer
insulation. We refer to this simple model as the XY
model(1).
Y 1? Y
? (? , T ) duct ? ? (1)
1? X X 1? X X
? ? b)
? spacer ? barrier ? oil ? barrier
It is important to note that ?oil, ?spacer and ?barrier, are, in
general, complex frequency dependent quantities. The
influence of temperature and moisture must be known.
The oil can, for low voltages, simply be characterised by
its relative permittivity, ?´oil (typically 2,2) and its DC Fig. 4 Distribution of equipotential lines in XY model
conductivity, ? (2). at a) 50 Hz and at b) 0,005 Hz.
? In most parts of the insulation the field is homogenous
? (? ) oil ? ? ?oil ? i ? oil (2) but in particular at 0,005 Hz, the field is significantly
? ?? 0 inhomogeneous at the corner of the spacer.
The crude XY-model can be used for diagnostic Fig. 3 also shows a comparison between the full
purposes [7, 8]. To determine the moisture in the solid frequency dependent FEM calculation, the calculation
insulation the procedure is to calculate the response of based on simple XY-model and the measured data.
XY-model (1), with materials data inserted from earlier The agreement is very good so, in this particular case,
measurements on pressboard for different moisture the simple XY-model is sufficient, but clearly the
levels. In Fig. 3 the output of this fitting procedure is frequency dependent FEM calculations can be used as a
presented. By inserting earlier measured material data tool for analysing the dielectric response of systems
into the model the moisture content was determined to with more complex geometries.
3 % moisture in the solid insulation and the oil
conductivity to 6 pS/m.
Fig. 8 shows the measured Tan(?) as a function of time
and frequency during a process cycle. In the beginning
Curing of machine insulation
(Phase I) the temperature is rapidly increased to 80 C°,
Another practically important composite system is the resin is melting and the losses increase rapidly. In
insulation of electrical machines. To achieve optimum Phase II the temperature is further increased but the
quality, the curing process must be finely tuned. Too onset of cross-linking eventually limits the motion of the
much or uneven curing may lead to a brittle insulation molecules and also the losses.
and poor adhesion to the conductors. Too little curing
will result in high dielectric losses. Dielectric response
can be used to study the curing process [10, 11].
Fig. 5 shows the layer structure of tape insulation in the
machine stator.
101
T=20 °C ? =0
p=1 MPa ? =0.2
? =0.4
0 ? =0.6
Fig. 6 Cross section of one layer of the insulation tape 10 ? =0.8
? =1
with glass carrier at the top and the epoxy-mica
Tan-Delta
T [?C] TMax
10-3 -2
(II) (III) 10 10-1 100 101 102 103 104 105 106
Frequency [Hz]
80 80
20 (I) (IV) Fig. 9 Frequency dependence of loss factor tan(?) at
10
T=20 ?C when the epoxy-mica tape has been cured to
different degrees, ? [13]. ? 2004 IEEE.
1200 t=f(TMax) Time [s]
The ? =0 curve displays a broad dipolar peak around 100
Fig. 7 Temperature profile, T=T? (t), to cure samples of Hz and in the low frequency end, a behaviour that can
machine insulation to different degrees. [13] ? 2004 be interpreted as DC conduction. As the curing
IEEE.. progresses, the peak rapidly shifts towards lower
frequencies. A “master curves” fit [15] formed by non-linear effects can nevertheless be important.
frequency shifting spectra measured at different (low) Dielectric liquids, such as transformer oil, can be
temperatures yields an activation energy of 0,8 - 1 eV. described as weakly dissociated electrolytes [16, 17].
Modelling the dielectric behaviour of machine insulation When a (DC) electric field is applied to a liquid the ions
during processing is quite complex, but simple geometric will travel with a velocity determined by their mobility, ?,
consideration has still proved to be fruitful [13, 11]. In and the electric field, E.
Fig. 10 two network models are illustrated. v ? E ?? (3)
H-model S-model If the gap length is d and the applied voltage is U then
the time, ?transit, it takes for an ion to cross the gap is
ZFiberglass d d2
ZFiberglass ZEpoxy ? transit ? ? (4)
ZEpoxy E ?? U ??
ZEpoxy
In reality there is a good chance that the ion recombines
ZEpoxy ZMica with a counter ion before it reaches the electrode of
ZMica
opposite polarity. The dielectric relaxation time ?relax is a
measure of the mean life of a free ion before it
Fig. 10 Two simple network models for modelling the recombines
epoxy-mica tape system [13]. ? 2004 IEEE.
?r?0
Fig. 11 illustrates the results obtained by means of the ? relax ? (5)
?
two models. The idea behind the simulation principle is
much the same as that of the XY-modelling for ?r is the permittivity of the liquid and ? the conductivity.
transformer insulation, i.e., the dielectric spectra ?0 =8,854 10-12 As/Vm. For the liquid to behave as a linear
measured for the constituent materials are used in resistor, ?transit must be much larger than ?relax . This can
combination with basic geometric parameters. be translated to a criterion on the applied voltage U for
The H-model with its transversal element is a reasonable Ohmic behaviour.
successful model for the resin-rich system. ? ?d 2
U ?? U crit ? (6)
? ?? r ?? 0
3
10
Measured If the applied voltage is comparable or exceeds the
2
H-Model critical voltage Ucrit in (6) the result may be that the oil
10
S-Model gap is depleted of charge carriers and the current
1 T=20 °C voltage characteristics will be sub-Ohmic. (Fig. 12). The
10
Tan-Delta
p=1 MPa critical voltage is very low for low conductivity liquids
0
? =0 such as for example newly produced transformer oil.
10
Typical values for ions in a good transformer oil at room
temperature are: mobility, ?=10-9 m2/Vs and conductivity,
10-1 ??=10-13 S/m. For a one-millimetre oil gap (d = 1 mm) this
leads to a critical voltage Ucrit of 5 V
-2
10 -4 -2 0 2 4 6
10 10 10 10 10 10
2
Frequency [Hz] 10
Fig. 11 Comparison between measured and modelled Ohmic Current
Total Current
data for uncured, ? =0, epoxy-mica tape system. [13]. 10
1
? 2004 IEEE
j · d/(?· Ucrit )
0
10
NON-LINEAR SYSTEMS
Nearly all insulation systems are, in fact, more or less -1
10
non-linear. It is only in the low voltage limit that the
systems behave linearly. In this section we briefly -2
10
discuss a few important cases: Narrow liquid ducts, 10 -1 10
0
101 102
water-treed cables and insulation with partial discharges.
E?d/Ucrit
Narrow ducts of dielectric liquids
Fig. 12 Normalized current density vs. normalized
Liquid ducts are of course part of transformer insulation. field for a pure oil gap. The “Ohmic Current” is a
Above we used a perfectly linear model (2) to describe hypothetical current expected in the case of linear
the response of transformer oil. At low voltages conductance.
dielectric liquids are almost perfectly Ohmic, but
So, if DC voltage is used for measuring e.g. the oil In the graph the real part value of the permittivity at low
conductivity very low voltages should always be used. voltage, ? ? (2,3) has been subtracted and the
If an AC voltage is applied to a liquid the ions will difference, ? ????is shown.
oscillate back and forth. The amplitude of the oscillation,
? ?' ?"
? x, will be (7) 10
-1 -1
10
U ??
?x? (7)
? ? f ?d
-2 -2
where f is the frequency of the AC voltage. The criterion 10 10
for linearity is now that the oscillation amplitude ? x will
be much smaller than the gap length d. This gives a new
critical field, Ucrit,AC , that
-3 -3
10 10
? ? f ?d 2
U ?? U crit , AC ? . (8)
?
1: 3 kV 3: 6 kV
At a very low frequencies, e.g. at f=0,001 Hz, Ucrit,AC is 2: 4.5 kV 4: 3 kV
-4 -4
only 3V. Clearly very low voltages are to be 10 10
-1 0 -1 0
recommended if one measures on insulation with small 10 10 10 10
Frequency (Hz) Frequency (Hz)
oil gaps, in particular with highly resistive oils.
Fig. 14 Voltage dependent responses of a water tree
At 50 Hz, Ucrit,AC is very high for a gap as big as 1 mm, aged XLPE cable [22].? 2004 IEEE
however for very small gaps, such as those in liquid
impregnated systems [18], the non-linear effect can be Both the real part ? ???and the loss ??? show clear voltage
significant also at power frequencies. Ions that are dependence. One may notice that for the highest
swept out of the liquid phase are immobilized. The result voltages the slope of the loss curve approaches –1 in
is that the losses will decrease as the voltage is the log-log coordinates. This indicates that there are
increased (“The Garton effect”). water trees that penetrate the whole insulation and a DC
like leakage current develops at high voltages. Such
At very high fields the losses may increase again due to, behaviour is a sign of severe degradation and significant
for example, field enhanced dissociation and charge risk of breakdown exists if the cable is left in service.
injection [19, 20].
The relation between breakdown strength and dielectric
XLPE with water trees loss ??? at 0,1 Hz is shown in Fig. 15 for lightning impulse
Water treeing is an ageing mechanism in polymer (LI) strength and in Fig. 16 for AC step test. The
isolated cables for medium voltage. A water tree [21] is a breakdown voltage Ubd is normalised in both graphs with
tree or bush like structure that can develop in polymeric the nominal service voltage, U0.
insulation under the influence of moisture and electric
? " at U 0 and 0.1 Hz
field. Ageing due to water treeing significantly reduces 10
0
-2
10
10 -3
-4
10
-5
10
0 5 10 15 20 25 30 35 40
U /U
bd 0
Fig. 15 Loss ? ? as a function of Ubd/U0 for aged XLPE
cables. (LI test) From [22].? 2004 IEEE
Fig. 13 Water trees found in a field-aged XLPE cable.
Dielectric response methods can be used to diagnose
water tree infested cables [22, 23, 24, 25]. The voltage
dependence of both capacitance and loss is a clear
indication of water tree aging. Fig. 14 shows the results
of measurements on a 12 kV cross-linked polyethylene
(XLPE) cable from the 70-ies.
0,03
0
? " at U and 0.1 Hz
0
10 0,025
Tan-Delta
10 -1 0,02
-2
10 0,015
-3
10 0,01
-4
10 0,005
-5
0 5 10 15 20
10 Applied Voltage [kV]
0 2 4 6 8 10 12
U /U Fig. 17 Tan(?) “tip-up” measurement on an aged
bd 0
stator bar.
Fig. 16 Loss ? ? as a function of normalised breakdown
strength, Ubd/U0, (AC step). [22] ? 2004 IEEE There is naturally a contribution from partial discharges
(PD) to the non-linear dielectric response ?29, 30, 31,
The non-linear response will manifest itself as
32]. To a first approximation a single partial discharge, of
containing harmonics in the response current. Also the
induced charge, qi , can be represented by a delta
harmonic content can be used as diagnostic indicators.
For moderately aged cables the odd harmonics (3f, 5f, 7f function like current pulse ? iPD (9) in the outer circuit.
…) dominate but for severely water-treed cables, with ? iPD ( t) ? qi ?? (t ? ti ) (9)
leakage current, the contribution from the even
harmonics (2f, 4f, 6f, …) becomes significant [22]. A PD pulse contributes both to the capacitance and loss.
If a sinusoidal voltage of frequency f (U?sin(2?ft) ) is
Several models have been applied to explain the
applied the contributions are (10):
dielectric response of water trees. One approach is to
qi
use an equivalent circuit [26] in order to describe the ? Ci?? ?cos(? i )
water-treed insulation. The un-treed XLPE is represented U
by a loss-free capacitor. And the deteriorated region is q
? Ci?? i ?sin(? i ) (10)
modelled by a capacitor with a parallel resistor U
representing the increased conductivity within the water ? i ? 2? f ?ti
tree region. With this model it is possible to explain the
increase in both capacitance and loss, but it is difficult In general numerous pulses exist and we therefore have
to explain both the frequency dependence and the to sum over index i in (9) and (10). The phase position of
non-linearity. the discharge is important as well. Discharges near the
peak of the voltage will have a maximum impact on the
A more physical approach is to [27, 28] assume that the
loss whereas discharges near the zero crossings will
water-treed region consists of water droplets forming
have the maximal impact on the capacitance.
“strings of pearls” interconnected by narrow channels
of crazed insulation. This model explains, to some extent, Not only the fundamental components (10) but also
the leakage current response of deteriorated XLPE by higher harmonics in the response current will be affected
establishment of field dependent electric contacts by the discharge distribution. Fig. 18 shows the current
between water droplets bridging the whole insulation from partial discharges in a small cavity bounded by
thickness. The predicted frequency dependence is dielectric walls (poly ethylene). The real and linear
however in such a case not in complete agreement with displacement current is outbalanced and only the
experiments. non-linear component is recorded.
PD-current [10-8 A]
90
45
PD-current [10-8 A]
45
0
0
-45
-45
-900 90 180 270 360
Phase [Degrees]
-90
0 90 180 270 360
Fig. 20 Corona current from a point-plane gap.
Phase [Degrees]
Fig. 18 PD-current from an artificial cavity.
It is clear that the current contains a lot of harmonics
and Fig. 19 shows the first eight absolute harmonic -7
|a |
1 10 n
values on the cosine components (a n) and sinus
Amplitude [A]
components (b n) in the Fourier transform of the |b |
n
PD-current (11). -8
8 10
a0 ?
iPD (t ) ? ? ? an ?cos(2? f ?t ) ?bn ?sin(2? f ?t) (11)
2 n? 1
4 10-8
3 10-7
|an| 0 10
0
1 2 3 4 5 6 7 8 9
Amplitude [A]
2 10 -7
|bn| Harmonic number, n=f/fo
Fig. 21 First eight Fourier coefficients in the spectrum
of the point-plane corona current from Fig. 20.
1 10-7 Again, the symmetry of the PD pattern is reflected in the
Fourier components. The loss contribution b1 is much
larger than the capacitance contribution a2 and both odd
0 100 and even harmonics are significant.
1 2 3 4 5 6 7 8 In fact, the harmonics in the dielectric response can be
Harmonic number, n=f/fo
used for classification of PD spectra. In Table 1 a few
Fig. 19 First eight Fourier coefficients in the spectrum idealised symmetry cases are listed [31].
of the PD current in an insulated cavity from Fig. 18. The relations (9) and (10) do not indicate any
The fundamental components (a 1 and b 1) are the PD dependence of the frequency of the applied voltage on
contributions to the capacitance and loss, as defined in the PD current. In reality there exists a significant
(10). In this case a 1 is, in fact, larger than b 1. The frequency dependence [33, 34]. Fig. 22 shows the
symmetry, in the PD pattern, between the positive and number of discharges per cycle as a function of the
negative half-periods is reflected in the harmonic frequency of the applied voltage. The study object is an
distribution [31]. In this case harmonics with even n are old bitumen-mica insulated stator bar. There are
in general much smaller than the odd ones. mechanisms that can, at least qualitatively, explain the
Another canonical geometry is the point-plane gap. Fig. frequency dependence. At the high frequency end (Fig.
20 shows the corona current from such a configuration 22) the number of discharges drops sharply. This is most
In this case (Fig. 21) the loss contribution, b 1, is much probably the effect of a statistical time lag. If time
larger than the capacitance contribution a 1. interval, when the voltage is above the inception level,
becomes short enough, it happens that seed electrons
are not available to start an avalanche [35].
nature of material degradation.
Table 1 Relation between the shape of the PD pattern
q (? ) and coefficients of the Fourier harmonic 1000
components.
Phase distribution Influence on the 800
q (? ) Fourier coefficients
?
200
0
bn ? 0, ? n 10-3 10 -2 10-1 100 10 1 102
Even (symmetric) around Frequency [Hz]
origin
Fig. 22 Measured number of discharges per cycle for a
q (? ) ? q (? ? ) bitumen-mica insulated machine bar. Applied voltage
q (? ) is 9 kVrms [33].
q (? ) ? ? q (? ? ? ) an ? 0, bn ? 0, ? even n
q (? )