198 A. Cavallini et al.

: PD Apparent Charge Estimation and Calibration: a Critical Review

PD Apparent Charge Estimation and Calibration:

a Critical Review
Andrea Cavallini, Gian Carlo Montanari and Marco Tozzi
LIMAT/DIE, University of Bologna
viale risorgimento 2
40136 Bologna, Italy

ABSTRACT
This paper focuses on the limits of partial discharge apparent charge estimation and
calibration using an unified approach. The fundamental problems that prevent
accurate apparent charge estimates (resonances, attenuation and partial discharge
pulse splitting) are tackled through this approach, highlighting the physical limitations
that prevent partial discharge charge to be measured properly. The feasibility of
calibration of Ultra-Wide Bandwidth measurement systems, something that is not
treated in the IEC 60270 standard, is also discussed.
Index Terms — Partial discharges, apparent charge, calibration, UWB detectors

1 INTRODUCTION that are related to apparatus geometry, materials and the

location of the discharge site [8, 9].
DURING Partial Discharges (PD), avalanches of
electrons and ions bombard dielectric surfaces causing Yet, apparent charge can serve different purposes as,
irreversible degradation phenomena (e.g., CH and CC bond e.g., harmonizing readings by different detectors or
breaking if we refer to polymeric insulation systems) [1-4]. different coupler types, verification of proper installation
Each particle in these avalanches acquires an energy (and and compensate in part for attenuation phenomena in
therefore, a capability of interacting with dielectric distributed-parameter test objects. For these reasons, IEC
surfaces) that depends on collisions with gas molecules, the 60270 [10] standardizes apparent charge measurements.
spatial position of other charged particles, etc.. A PD is However, the standard does not provide recommendations
therefore a complex random phenomenon which is for Ultra-wide-band (UWB) detectors as they do not
generally lumped into a single number, i.e., the charge directly quantify the apparent charge of PD current
moved by the avalanche itself. This quantity can provide an pulses. Wide-band detectors, as defined in the IEC 60270
indirect estimate about the number of hot electrons (i.e., standard, have, at maximum, a bandwidth of 400 kHz,
electrons having an energy sufficiently high to cause while the class of UWB detectors, not dealt with in the
irreversible degradation) impinging on the dielectric standard, includes a variety of detectors which are
surfaces. becoming more and more popular these days, that is,
commercial detectors with bandwidth of tens of MHZ,
In the lab, when the geometry of the defects generating Digital Sampling Oscilloscopes (DSO) working up to
PD is known, of small dimension and PD currents are several GHz and spectrum analyzers.
recorded with high accuracy, PD charge can be assessed
with good precision, thus providing useful information Reviewing the concept of apparent charge estimation and
about degradation rate (and, therefore, PD harmfulness) [3]. calibration taking into account UWB detectors is the goal of
In the field, the properties of the so-called ABC circuit this paper. For this purpose, the basic concept of apparent
(which, models only electrostatic equilibriums inside the charge estimation is recalled in order to highlight the
Equipment Under Test, EUT) highlight that only a fraction fundamental assumptions behind this procedure. Then, a
of the so-called apparent charge can be measured [5-7]. The taxonomy of UWB detection systems is set up. The
PD and the apparent charge are related through a categorization provided here will show that some systems
proportionality factor that is well below 1 (which implies can be calibrated (providing readings in pC, besides
that the true charge can be severely underestimated) and features as source separation and noise rejection) under
depends in part on the geometry of the defect (thus, the peculiar conditions, some of them not. Eventually, pros and
error cannot be compensated). Furthermore, attenuation cons of the different types of UWB detectors will be
phenomena can reduce apparent charge readings by factors discussed. Eventually, it is noted that only detectors dealing
with conducted signals are dealt with, since those working
Manuscript received on 29 August 2009, in final form 16 October 2009. with radiated signals cannot be calibrated by definition.

1070-9878/10/$25.00 © 2010 IEEE

IEEE Transactions on Dielectrics and Electrical Insulation Vol. 17, No. 1; February 2010 199

2. CALIBRATION AND APPARENT CHARGE some hundreds of nanoseconds with a rise time (associated
ESTIMATION: A REVIEW with electron displacement) much shorter than the fall time
(associated with ion drift). The effects induced in the EUT
According to the analysis of the so-called ABC circuit consist generally of fast phenomena occurring during the
(see Figure 1, where the meaning of ABC is clarified), the PD pulse itself, followed by the natural response of the
system, usually characterized by longer time constants.
150
Integrated current
Theoretical charge value (94.3
(95.9 pC)

Transferred charge (pC)

100

Figure 1. The ABC circuit. The EUT is represented

as three capacitances. A is the capacitance of the
EUT not affected by the defect (cavity). B is the
50
capacitance of the dielectric in series with the
cavity. C is the capacitance of the cavity

20 0
0 100 200 300 400 500
Time (ns)
(mA)

10 Figure 3. Integration of the current reported in Figure 2 (below).

Measurements have been corrected for DSO input circuit bias.

0 120
0 50 100 150 200 250 300

10 100

Transferred charge (pC) 80

(mA)

0
60

-10 40
0 50 100 150 200 250 300
Time (ns)
20
Figure 2. Calibrator pulse (95.9 pC) injected (top) directly at a 16 GSa/s
DSO input, (bottom) at the same DSO input through a 1 nF capacitive
0
coupler.
-100 0 100 200 300 400 500 600 700
Time (ns)
charge, q, that flows in a coupling capacitor in parallel to an Figure 4.. Integration of the current injected by a pulse generator in a 350
Equipment Under Test (EUT) due to a PD of charge qPD is W induction motor. Due to reflections at the (open) star point, after 600 ns
[5,6]: the charge estimate has not fully stabilized.

⎛ Ck ⎞ ⎛ d ⎞ As an example, Figure 2 shows (top) a 95.9 pC pulse

q ≈ ⎜⎜ ⎟⎟ ⋅ ⎜ ε r ⋅ q PD ⎟ << q PD (1) (obtained using a calibrator with 18 pF series capacitance,
⎝ EUT + Ck
C ⎠ ⎝ D − d ⎠ injecting a nominal charge of 100 pC, real charge 95.9)
injected directly into a Digital Sampling Oscilloscope
D and εr, being the thickness and the permittivity of the (DSO) having input impedance of 50 Ω, and the response of
insulation of the EUT, d (d<<D) is the height of the cavity, a 1 nF coupler to the same pulse (bottom). If reflections are
Ck is the coupling capacitance. Although the term “apparent not taken into account, the response of the coupler circuit
charge” is used to indicate the term in the rightmost obeys, approximately, the step response of a second-order
parenthesis of eq. (1), we will use it to indicate the charge underdamped circuit. In order to assess exactly the charge
that can be estimated processing the current flowing in the transferred by the pulse For the circuit described by the
coupling capacitor. This will be done for the sake of response of Figure 2 (bottom), one should wait at least 300
simplicity, in order to avoid to introduce new terms as, e.g., ns. This is further confirmed by analyzing Figure 3, which
measurable charge.. shows the integral of the current in Figure 2 (bottom).
Equation (1) implies clearly that the true apparent charge In power apparatus, characterized by large equivalent
cannot be recovered from measurements if the geometry of capacitance and inductance, the time constants needed to
the defect is unknown, which will be almost always the achieve a correct estimate of charge can be much larger,
case. Therefore, apparent charge measurements could making direct integration very complex also due to
provide only qualitative indications about real PD charge. superposition phenomena. As an example, Figure 4 shows
The ABC circuit analysis deals only with electrostatics. the response of a 350 W motor to a pulse: at about 600 ns,
In practice, a PD can be schematized as a fast pulse lasting the transferred charge has not yet fully stabilized.
200 A. Cavallini et al.: PD Apparent Charge Estimation and Calibration: a Critical Review

Furthermore, interference from ac bus voltage harmonics According to equation (2), the filter output can be treated
would cause even larger errors [7]. in many ways to achieve a reading that can be reported in
For the above reasons and to avoid interference from the pC. Since the peak output voltage can be measured by
ac mains, a procedure for apparent charge estimation that simple analog circuits and is the value of v(t) with largest
does not attempt to estimate directly the dc component, 〈X〉, signal-to-noise
of the pulse (that is clearly related to the charge transferred ratio, this quantity, supported by the IEC 60270 standard, is
probably the most convenient one.

Figure 5. Rationale of PD apparent charge estimation performed according

to standard IEC 60270: by processing spectral components in [0,fc],
linearity ensures that one would obtain a response that is directly
proportional to 〈X〉. Interference from ac mains and radio station
broadcasting is not considered for the sake of simplicity.

Figure 6. Equivalent circuits representing the electrical effects induced by

by the pulse itself) has been standardized in the course of a PD and by a calibrator in the EUT/coupler circuit. (already in the text).
time. This procedure reconstructs 〈X〉 by processing the
pulse spectral components that, due to the properties of
ideal (Dirac) pulses, have a magnitude that is similar to 〈X〉 3 LIMITS OF CALIBRATION
(see Figure 5). The calibration procedure assumes implicitly that a PD
The standardized procedure, known as quasi-integration, and a calibrator pulse having the same charge q provide the
resorts to a band-pass filter that is able to single out the same effects at the coupler. Referring to the case where the
pulse spectral components that have a magnitude similar or coupler is connected at the EUT terminals, this condition is
equal to the dc components (the filter should have upper met only if the PD can be meaningfully represented as a
cutoff frequency below fc, as sketched in Figure 5). In the current generator connected at the same terminals where the
time domain, the output of such filter is the impulse calibrator has to be connected, that is, the EUT terminals.
response of the filter itself, h(t), times the dc component of Since PD sources are often at some distance from the coupler
the pulse 〈X〉: terminals (e.g., PD source in a cable joint, with calibration
made at terminations), the EUT should be treated as a
v(t ) = X ⋅ h(t ) (2) quadrupole, with PD injection at one port, calibration at a
different one. Only for PD sources very close to EUT line
terminals these two ports could be regarded as a unique one.
Since the output of the filter is in mV, a further step is
In general, however, the difference between PD injection port
needed to achieve pC: calibration. Calibration is performed and calibration port will affect measurements.
by injecting at the EUT terminals a fast pulse of known
charge. Generally, the calibrator is made by fast-front In order to investigate how PD source location can
square wave generator (of peak voltage Uk) in series with a influence measurement accuracy, reference is made here to
capacitor having a capacitance (Ck) much lower than that of the circuit shown in Figure 6, where the coupler is
the EUT [10-13]. The product Uk⋅Ck is the charge injected connected between the EUT high voltage terminal and the
by the generator when the calibrator is short circuited. To ground and the EUT itself is first treated as a quadrupole (to
minimize loading effects of the EUT, the calibrator highlight that the PD injection port may not necessarily be
capacitance should be the lowest possible (compatibly with that where the calibrator is connected), then it is replaced by
the peak voltage requirement on the square voltage its equivalent π circuit. Using nodal analysis it is possible to
generator), even if, as highlighted in Figure 4, some prove that when the transversal impedance of the EUT, Zt,
discrepancies are often likely to occur. It is also important is infinity, the effects of the PD and calibrator pulse are
that the calibrator rise time is small enough to ensure that in equal (direct injection of both pulses into the coupler).
the frequency range used for apparent charge estimation, Clearly, the requirement 1/Zt=0 cannot be met in practice,
the spectral energy of the pulse is comparable to the dc so that it can be stated that calibration provides a reasonable
component. approximation of PD events only when the transversal
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 17, No. 1; February 2010 201

impedance is very large compared to other impedances in voltage at the 50 Ω measuring impedance was sent to an
the circuit. Various phenomena, described in more detail in ideal integrator circuit (an approximation of the quasi-
the following, can cause significant deviations from ideal integrator bandpass filter) and the peak response of this
conditions, such as resonances, attenuation and PD pulse circuit was used as a means to quantify the PD charge. The
splitting in systems behaving as long transmission lines. results, reported in Figure 8, highlight that noticeable errors

Figure 8. Estimated charge for a circuit having (see Figure 6) longitudinal

impedance Z long = j ⋅ ω ⋅ L , with L varying from 1 nH to 2000 nH,
transversal impedance Z t = 1 / ( j ⋅ ω ⋅ C ) , with C=1 uF. Pulses are acquired
using a 1 nF coupling capacitor in series with a 50 Ohm resistor. Computer
simulations carried out using Pspice.

(PD charge estimates range from 137 to 0.5%) can affect

the measurement procedures. With very low inductance
values, due to resonances, overestimates can be obtained.
With very high inductance values, the systems acts as a
Figure 7. Impulse response (in the frequency domain) of the voltage
across a 1 nF capacitive coupler for in Figure 6 (already in the text) Top: filter that removes most of the energy of the signal,
pulse injected at the PD site. Bottom: pulse injected at the EUT terminals providing strong underestimates.
3.2 ATTENUATION
3. 1 RESONANCES
Attenuation phenomena are particularly important in long
In electrical machines, the longitudinal impedance Zlong
transmission lines. With reference to Figure 6, Zlong can be
of the EUT is characterized by the self-inductance of the
approximated by the characteristic impedance of the line:
windings. Due to the capacitive nature of the transverse 4

impedance, resonances can be activated in different ways

by the PD sources and by the calibrator (connected between r + j ⋅ 2 ⋅π ⋅ f ⋅ l
Z0 = (3)
line terminal and ground), affecting noticeably the g + j ⋅ 2 ⋅π ⋅ f ⋅ c
calibration procedure [14-16]. 4

As an example, Figure 7 shows the response of the where r and l are the longitudinal resistance and inductance
circuit reported in Figure 6 when Z long = j ⋅ ω ⋅ L , with (per unit length), g and c the transversal conductance and
capacitance (per unit length). As an example, in power cables,
L=500 nH, Z t = 1 / ( j ⋅ ω ⋅ C ) , with C=1 μF and the coupler by rising the frequency, f, skin effect increases r, whereas the
is a 1 nF capacitor in series with a 50 Ω resistor. Figure 7 semiconductive coating loses dielectric permittivity,
(above) reports the response the voltage across the coupler decreasing c [8, 9]. Overall, the effect is an increase of the
(in the frequency domain) when a pulse having dc characteristic impedance with frequency. Referring to the 100-
component equal to 1 A is injected at the PD site. Figure 7 500 kHz range, investigations have shown that a PD pulse can
(below) is relevant to the injection of the pulse at the EUT lose more than 10-15% of its frequency content when traveling
terminals. As can be seen, the response of the circuit is along a 5 km long cable [9].
completely different in the two cases. In particular, a Attenuation-induced inaccuracies can be emphasized by
marked resonance peak can be observed in the first plot detectors designed in an inappropriate way. As an example,
only. These differences would affect noticeably the if the upper cutoff frequency of the bandpass filter is too
measurement of apparent charge, as the effects due to a PD large, attenuation phenomena may affect measurement
or a calibrator pulse are completely different. accuracy. Figure 9 provides a schematic representation of
To further stress these findings, Pspice simulations aimed this type of error. In particular, assuming that calibration
at reproducing in the time domain the effects associated has been performed using an ideal pulse generator, referring
with calibrator and PD pulses were carried out. This to filter bandwidth A will provide the correct estimate of
investigation was carried out to quantify directly, how pulse dc component (calibrator pulse and PD pulses have
resonances can influence apparent charge estimation. The the same characteristics in this range of frequencies). Filter
202 A. Cavallini et al.: PD Apparent Charge Estimation and Calibration: a Critical Review

B will provide some approximation of the dc component, According to equation (4), it is not surprising that a 100
since calibrator and PD pulses are coincident below fc pC calibrator injected at the far end and only moderately
whereas for frequencies above fc discrepancies exist. Filter affected by attenuation could provide a reading at the
C will provide strongly incorrect estimates, particularly for termination of about 130 pC. On the contrary, the pulse
pulse #3. injected at the connection between cable splices splits in
two pulses traveling in opposite direction. Theoretically, we
would expect a reading of 64.5 pC at the detector end (1.29
x 50). The additional 5.5 pC are likely originated by a
partial reflection of the pulse at the junction of the two
cable splices, which gives rise to a reflection arriving at the
detector end about 600 ns after the calibrator pulse.
This simplified example highlights that the apparent
charge of PD pulses in EUT behaving as transmission lines
can be measured accurately only when they occur in close
proximity of the termination where calibration has been
performed. Otherwise, depending on reflection coefficients
at the EUT ends and on the position of the defect, large
underestimation or overestimation errors are likely to occur.
Figure 9. Estimation of apparent charge using three different frequency
ranges for the bandpass filter. Filter A will provide correct results. Filter B 4 UWB SYSTEMS, IS CALIBRATION FEASIBLE?
and C will be affected by errors, that will be larger for filter C. The concept that UWB systems cannot provide pC
readings seems to be accepted by most people, despite the
Clearly, by reducing as much as possible the filter lower fact that it can be wrong. The reason is that the “umbrella”
cutoff frequency, this error can be remarkably reduced, acronym UWB covers a variety of situations. In some of
even if quantifying this error in analytical form is very them calibration is unfeasible, while in some other cases
complicated. This suggests that detectors having large calibration is possible if suitable tools are employed. First
bandwidth with low cutoff frequency are best suited for this of all, since UWB systems are made by a UWB detector
type of application. and a coupler, the analysis can be split between these two
components, joining the results, eventually, to grasp a
3.3 PD PULSE SPLITTING
comprehensive picture.
In an object behaving as a transmission line (say, a
To proceed further, some order in the current
cable), the PD pulse will split in two, half propagating in
terminology must be done. Indeed, UWB detectors can be
one direction, the other one in the opposite direction [17].
characterized by very different bandwidths. In particular,
The two pulses will arrive at the coupler terminal at
spectrum analyzer working in the zero-span mode are
different times, with a delay that depends on the distance
characterized by narrow bandwidths (say some MHz), with
between the PD source and the terminal opposite to the
central frequencies often in the range of few GHz [18-21].
coupler. Depending on the delay, the two filter responses
We shall term these detectors as narrowband UHF
might be completely or partially overlapping when the
detectors. On the contrary, other UWB detectors have
source is close to the terminal opposite to the coupler, not
analog inputs designed to work properly with signals in the
overlapping at all in the other cases. Therefore, we can
range from few tens of kHz up to some tens of MHz. When
expect that errors up to 50% of the original apparent charge
Digital Sampling Oscilloscope (DSO) are used, the
can be associated with this type of mechanism.
bandwidth of the detector can be in the range of few GHz.
To prove this, two splices of cable (80 and 125 m long) Only these detectors could be termed correctly as Ultra
were connected together. Calibration was performed at the Wide Band (UWB).
80 m cable terminal injecting a 100 pC pulse between core
and sheath. The PD detector was also connected between 4.1 NARROWBAND UHF DETECTORS
core and sheath, directly. After that, the same calibration If one disregards the effect of the coupler, a clear cut can
pulse was first injected at the connection between the two be made here: UWB detectors can be calibrated (as shown
splices, obtaining a reading at the termination of 70 pC. later), narrowband UHF detectors cannot be generally
Then, the calibration pulse was injected at the far end calibrated if the central frequency of their band is well
(termination of the 125 meter splice), providing 130 pC. above the range of frequencies where the PD pulse spectral
These results can be explained as follows. At the terminal energy is still comparable with that of their dc component
where measurements are performed, there is an impedance (as already discussed in the previous section, see Figure 9).
mismatch, being the cable characteristic impedance about This range of frequencies is strongly apparatus-dependent
27 Ω and the detector input impedance 50 Ω. The reflection being, e.g., in the few MHz range for rotating machines, up
of the propagating waves is: to several tens or hundreds of MHz for GIS). Under some
circumstances, also the difference between calibrator and
50 − Z 0 PD pulse rise and fall times might affect considerably
1+ Γ ≈ 1+ = 1.29 (4)
50 + Z 0 calibration accuracy. To prove this, simulations were
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 17, No. 1; February 2010 203

carried out assuming that calibration were to be performed from 16 kHz to 50 MHz, able to digitize the voltage pulses
resorting to a 100 pC pulse given by: appearing at its terminals with a sampling rate of 200 MSa/s
[22]. The two calibrators are characterized by different
x(t ) = A ⋅ (− exp(− t / τ 1 ) + exp(− t / τ 2 )) (5) pulse types, something that is not uncommon given the
2 weak constraints imposed by the IEC 60270 standard. In
10
10 ns
particular, calibrator (A) is able to provide pulses with a
20 ns larger frequency content, if compared with calibrator (B).
30 ns Figure 11 shows the differences in the calibrator pulse
40 ns spectra at the nominal charge of 100 pC recorded through a
Apparent charge (pC)

50 ns Digital Sampling Oscilloscope (DSO) having a sampling

100 ns rate of 16 MSa/s for single-shot measurements. The spectral
1
10
components of the pulses are reported in the figure in pC
since the measured currents were first scaled to pA, then
multiplied by the time length of the recording window, so
that the dc component provides directly the charge
associated with the pulse. The figure also reports the real
0
charge, evaluated by direct integration of the pulses.
10 0
10
1
10 10
2 3
10 To test whether apparent charge estimation could be
Filter center frequency (MHz) achieved in a digital way, the digital 100-500 kHz bandpass
Figure 10. Apparent charge estimation performed through a narrowband filter (that is, a filter with cutoff frequencies that comply
UHF detector as a function of filter central frequency and PD pulse rise with the IEC 60270 standard) embedded in the detector was
time. used as follows:
100 • Calibrator A at nominal charge of 100 pC was used to
Calibrator A (95.9 pC)
Calibrator B (96.4 pC)
achieve the calibration constant (the charge used to
80 achieve the calibration constant in the software was the
actual one provided by the calibrator and estimated
using the DSO, i.e., 95.9 pC).
Magnitude (pC)

60
• The charge of calibrator B at 100 pC (nominal charges)
40
was also estimated according to the above procedure.
The results reported in Tab. 1 show that apparent charges
were measured with a very good SNR for both calibrators
20
(standard deviation <1 pC in both cases), with a minimum
discrepancy between what was measured by the DSO for
0 -2 0 2 4 6 calibrator B (96.4 pC) and what was measured by the PD
10 10 10 10 10
Frequency (MHz) detector (98.3 pC, median value). The magnitude of this
Figure 11. Spectral characteristics of pulses obtained by two different
deviation is, however, in line with those typically observed
calibrators at the nominal charge of (top) 100 pC, (bottom) and 100 pC. using detectors with slightly different characteristics.
(already in the text).

Table 1 Results of apparent charge estimation through digital signal

with τ 1 = 1 ns and τ 2 = 200 ns . On the contrary, processing techniques. Measurements performed using a UWB detector
having 50 MHz bandwidth, 200 MSa/s sampling rate.
measurements were to be carried out on 100 pC pulses
Charge Average Standard
(simulated PD) that, due to attenuation, had τ 1 =10, 20, 30, Calibrator estimated from estimated charge deviation
40, 50 and 100 ns. Apparent charge estimation was DSO (pC) (pC) (pC)
performed with digital filters having cutoff frequencies of A 95.9 95.9 0.68
f 0 ± 1 MHz , with f0 varying from 1 MHz to 1 GHz. The B 96.4 98.3 0.37
results reported in Figure 10 emphasize that apparent
charge estimates can be affected by noticeable errors when
carried out using UHF narrowband detectors. The 5 DISCUSSION
magnitude of the error increases with τ 1 (attenuation) and The above presentation outlines how apparent charge can
f0. provide misleading indications when applied to practical
objects as, e.g., cables, synchronous generators,
4.2 ULTRA WIDE BAND DETECTORS  transformers. The concept that apparent charge can give
UWB detectors work with signals that encompass all the information about PD severity and degradation rate at the
information that is necessary for calibration: extracting it is defect site has been, therefore, de-emphasized in the course
simply a matter of setting up adequate tools. As an of time, highlighting that a correct risk assessment requires
example, let us show how different calibration pulses were trending of PD phenomena and PD pattern analysis to
handled by a detector operating in the range of frequencies recognize pre-fault conditions [23-25].
204 A. Cavallini et al.: PD Apparent Charge Estimation and Calibration: a Critical Review

Yet, apparent charge maintains a role when from the windings. Since PD in the overhang are often
measurements from different sensors/detectors need to be characterized by large repetition rates, it is likely that a
compared. As a matter of fact, while quantities as, e.g., PD NUHF detector is set to operate at frequencies typical of
pulse peak value, are remarkably affected by the bandwidth these phenomena, disregarding other phenomena as slot or
of the measurement chain, apparent charge is generally delamination PD. This can be more marked in the case of
comparable if different detectors are used. Therefore, it can monitoring, where intermittent or new phenomena
be exploited to compare measurements from different providing pulses with low energy content within the
service providers, when changing detector/sensors to frequency range used for PD detection (defined at the
preserve old data, etc. For these reasons, recent standards beginning of the monitoring period) can be totally
[26] have changed terminology in the course of time, neglected.
switching from calibration (which provides the false Furthermore, the knowledge of PD pulse shapes (possible
confidence that absolute measurements are being carried only through UWB detectors) can enable separation of the
out) to normalization. This latter term is more appropriate contributions due to different phenomena in the pattern,
to indicate that what it is performed is a procedure aimed to allowing to perform PD source identification in a
homogenize measurements and check the sensitivity of the convenient way through automatic procedures [22, 24].
measurement chain.
Eventually, UWB detectors prove to be appropriate when
Regarding the measurement chain and, specifically, PD location is required. In power cables, location can be
detectors, it must be reminded that the acronym UWB carried out by three different procedures: the Time Domain
encompasses two families of detectors with characteristics Reflectometry (TDR), the Arrival Time Analysis (ATA)
that are peculiarly different. Although the analog input and the Amplitude/Frequency Mapping (AFM) [27-29]. All
stage in both families can be regarded truly as a UWB one, these methods require UWB detectors in order to establish
the operation mode of one class of detectors, here termed as accurately the arrival time of the pulse. In power
narrowband UHF (NUHF), is that of a wideband (as transformers and in rotating machines, the knowledge of PD
defined in IEC 60270) detector since the difference between pulse shapes can help to evaluate the distance between the
lower and upper cutoff frequency can be of few MHz, but coupler and the PD sources [30].
with a central frequency as large as some GHz. On the
contrary, UWB detectors can have a bandwidth of, at least, 6 CONCLUSIONS
several tens of MHz, with a lower cutoff frequency that is The rationale behind apparent charge estimation, together
the minimum one required to suppress voltage harmonics with the limits of this concept, have been recalled and
from the ac bus (some tens of kHz, generally). discussed in this paper. It has been highlighted that,
The difference in operation mode reflects on the depending on the characteristics of the EUT, injecting the
properties of the two types of detector. As a matter of fact, calibrator pulse at the EUT terminals may induce a response
while most of the information contained in a PD pulses is that can be significantly different from that induced by a PD
retained by UWB detectors, NUHF ones disregard it. By pulse, leading to large errors.
tuning appropriately the center frequency of NUHF, these Furthermore, since PD detection at very high frequencies
detectors are able to operate in very noisy environment seems promising to achieve large SNR in measurements
where, sometimes, UWB ones may need external bandpass when the coupler can be placed in close proximity of the
or notch analog filters to be able to acquire signals with defect, as in cables, the possibility of getting apparent
sufficient Signal-to-Noise Ratio (SNR). However, if the charge estimates from UWB detectors has been examined
center frequency is not small enough, apparent charge
With narrowband UHF detectors, the information about
estimation cannot be performed. UWB detectors, on the
PD apparent charge can be lost, depending on the
contrary can be calibrated and provide apparent charge
characteristics of the bandpass filter. On the contrary,
estimates whenever the coupler itself offers sufficient SNR
depending on SNR in the frequency range used for apparent
at the frequencies generally used for this purpose (100-500
charge estimation, useful information about PD charge can
kHz).
be recovered from UWB detectors. Furthermore, these
Since the central frequency of NUHF detectors is detectors can help locating PD sources using, e.g., Time-
generally set by comparing the spectrum at zero voltage Domain Reflectomery, in measurements carried out at cable
with that at voltages above the PDIV (with the PDIV terminals and separate the contributions given by multiple
generally assessed by observing a sudden change in the PD sources.
spectrum envelope), it comes out that the application of
NUHF detectors to online system can be virtually REFERENCES
impossible, or very difficult, if the system cannot be halted [1] L. Dissado and J. Fotherghill, Electrical Degradation and
Breakdown in Polymers, IEE Materials and devices series, 1992.
to record the spectrum at zero voltage. [2] J. Densley, “Ageing mechanisms and diagnostics for power cables -
UWB detectors behave well in the measurement of large an overview”, IEEE Electr. Insul. Mag., Vol. 17, No. 1, pp. 14-22,
objects with multiple PD sources, where PD pulses can 2001.
[3] L. Testa, Physical model of damage inception and growth in
have very different frequency contents. As an example, in polymeric insulating materials subjected to partial discharges under
rotating machines PD in the overhang have generally a dc and ac voltage, Ph.D. Thesis, University of Bologna, Bologna,
larger frequency content if compared with those coming Italy, 2009.
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 17, No. 1; February 2010 205
[4] S. Serra, G.C. Montanari and G. Mazzanti, “Theory of inception [25] A. Cavallini, M. Conti, G.C. Montanari, C. Arlotti and A. Contin,
mechanism and growth of defect-induced damage in polyethylene “PD inference for the early detection of electrical tree in insulation
cable insulation”, J. Appl. Phys., Vol. 98, pp. 034102.1-034102.15, systems”, IEEE Trans. Dielectr. Electr. Insul., Vol. 11, pp. 724-735,
2005. 2004.
[5] R. Bartnikas and E.J. McMahon, Engineering Dielectrics: Corona [26] IEC 60034-27, Rotating machines – Part 27: Off-line partial
Measurement and Interpretation - Stp 669, ASTM Intl. 1979 discharge measurements on the stator winding insulation of rotating
[6] F. H. Kruger: Discharge detection in high voltage equipment, electrical machines, 2/1237/CD, 2005.
London, UK, Temple Press. 1964. [27] M.S. Mashikian, R. Bansal and R.B. Northrop, “Location and
[7] W. S. Zaengl and K. Lehmann, “A critique of present calibration characterization of partial discharge sites in shielded power cables”,
procedures for partial discharge measurements”, IEEE Trans. IEEE Trans. Electr. Insul., Vol. 5, pp. 833-839, 1990.
Dielectr. Electr. Insul., Vol. 28, pp. 1043-1049, 1993. [28] G. Katsuta, A. Toya, K. Muraoka, T. Endoh, Y. Sekii and C. Ikeda,
[8] G. C. Stone and S. A. Boggs, “Propagation of partial discharge pulses ”Development of a method of partial discharge detection in extra-
in shielded power cables”, IEEE Conf. Electrtr. Insul. Dielectr. high voltage cross-linked polyethilene insulated cable lines”, IEEE
Phenomena (CEIDP), pp. 275-280, 1982. Trans. Power Del., Vol. 7, pp. 1068-1074, 1992.
[9] M. Tozzi, A. Cavallini, G.C. Montanari and G.L. G. Burbui, “PD [29] A.Cavallini, G. C. Montanari and F. Puletti, “A novel method to
Detection in Extruded Power Cables: An Approximate Propagation locate PD in polymeric cable systems based on amplitude-frequency
Model”, IEEE Trans. Dielectr. Electr. Insul., Vol. 15, pp. 832-840, (AF) map”, IEEE Trans. Dielec. Electr. Insul., Vol. 14, pp. 726-734,
2008. 2007.
[10] IEC 60270, Partial Discharge Measurement, 3rd Edition, 2001. [30] A. Akbari, P. Werle, H. Borsi and E. Gockenbach, “Transfer
[11] TF 33.03.05, “Calibration procedures for analog and digital partial function-based partial discharge localization in power transformers: a
discharge measuring instruments”, Electra, No. 180, pp. 122-144. feasibility study”, IEEE Electr. Insul. Mag., Vol. 18, No. 5, pp. 22-
Oct. 1998. 32, Sept.-Oct. 2002.
[12] O. Gunnarsson, A. Bergman and K. E. Rydler, “A method for
calibration of partial discharge calibrators”, IEEE Trans. Instr. and Andrea Cavallini (M’95) received from the University
Meas., Vol. 48, pp. 453-456, 1999. of Bologna the Master degree in electrical engineering in
[13] P. Osvath, E. Carminati and A. Gandelli, “A contribution on the 1990 and the Ph.D. degree in electrical engineering in
traceability of partial discharge measurements”, IEEE Trans. Electr. 1995. He was a Researcher at Ferrara University from
Insul., Vol. 27, pp. 130–134, 1992. 1995 to 1998. Since 1998, he is an associate professor at
[14] I. J. Kemp, H. Zhu, H. G. Sedding, J. W. Wood and W. K. Hogg, Bologna University. His research interests are diagnostics
“Towards a new partial discharge calibration strategy based on the of insulation systems by partial discharge analysis,
transfer function of machine stator windings”, IEE Proc. Sci. reliability of electrical systems and artificial intelligence. Since 2004, he is
Measurement Techn., Vol. 143, pp. 57 – 62, 1996. the Italian representative of Cigrè SC D1.
[15] G. C. Stone, “Partial discharge. XXV. Calibration of PD
measurements for motor and generator windings – why it can’t be
done”, IEEE Electr. Insul. Mag., Vol. 14, No. 1, pp. 9–12, 1998.
[16] M. M. Twiel, B. G. Stewart and I. J. Kemp, “An investigation into Gian Carlo Montanari (M’87-SM’90-F’00) was born on 8
the effects of parasitic circuit inductance on partial discharge November 1955. In 1979, he took the Master degree in
detection”, IEEE Electr. Insul. Conf. (EIC), Cincinnati (USA), pp. electrical engineering at the University of Bologna. He is
213-217, 2001. currently Full Professor of electrical technology at the
[17] Partial Discharge Detection in Installed HV. Extruded Cable Department of Electrical Engineering of the University of
Systems, CIGRE Tech. Brochure 182, CIGRE WG 21.16, 2001. Bologna, and teaches courses on Technology and
[18] S. Tenbohlen, D. Denissov, S. Hoek and S. M. Markalous, “Partial Reliability. He has worked since 1979 in the field of aging and endurance
discharge measurement in the ultra high frequency (UHF) range”, of solid insulating materials and systems, diagnostics of electrical systems
IEEE Trans. Dielectr. Electr. Insul., Vol. 15, pp. 1544-1552, 2008. and innovative electrical materials (magnetics, electrets, superconductors).
[19] M. D. Judd, O. Farish and B. F. Hampton, “The excitation of UHF He has been also engaged in the fields of power quality and energy market,
signals by partial discharges in GIS”, IEEE Trans. Dielectr. Electr. power electronics, reliability and statistics of electrical systems. He is a
Insul., Vol. 3, pp. 213–228, 1996. member of AEI and Institute of Physics. Since 1996 he is President of the
[20] D. Denissov, W. Köhler, S. Tenbohlen, R. Grund and T. Klein, Italian Chapter of the IEEE DEIS. He is convener of the Statistics
“Wide and narrow band PD detection in plug-in cable connectors in Committee and member of the Space Charge, Multifactor Stress and
the UHF range”, International Conference on Condition Monitoring Meetings Committees of IEEE DEIS. He is an Associate Editor of IEEE
and Diagnosis (CMD), pp. 1056–1059, Beijing, China, 2008. Transactions on Dielectrics and Electrical Insulation. He is founder and
[21] P. D. Agoris, S. Meijer, E. Gulski, J. J. Smit, T. J. W. Hermans and President of the spin-off TechImp, established on 1999. He is author or co-
L. Lamballais, “Sensitivity check for on-line VHF/UHF PD detection author of about 550 scientific papers.
on transmission cables”, IEEE Intern. Conf. Properties Applications
Dielectr. Materials (ICPADM), Bali (ID), pp. 204-207, 2006.
[22] A. Contin, A.Cavallini, G. C. Montanari, G. Pasini and F. Puletti “
Marco Tozzi was born in Udine, Italy, on 2 May 1979. He
Digital detection and fuzzy classification of partial discharge
received the Master degree in electrical engineering from
signals”, IEEE Trans. Dielectr. Electr. Insul., Vol. 9, pp. 335-348,
University of Trieste in 2005. At present, he is a Ph.D.
2002.
student at the University of Bologna, Italy. His research
[23] H. Zhu, V. Green and M. Sasic, “Identification of stator insulation
interests include high voltage systems characterization,
deterioration using on-line partial discharge testing”, IEEE Elec.
artificial intelligence techniques and diagnostic of
Insul. Mag., Vol. 17, No. 6, pp. 21-26, 2001.
insulating systems by partial discharges analysis.
[24] A. Cavallini, M. Conti, A. Contin and G.C. Montanari, “Advanced
PD inference in on-field measurements. Part 2: Identification of
Defects in Solid insulation Systems”, IEEE Trans. Dielectr. Electr.
Insul., Vol. 10, pp. 528-538, 2003.