THIS IS A PRE-PRINT COPY
(c) 2009 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including
reprinting/ republishing this material for advertising or promotional purposes, creating new collective works for resale or
redistribution to servers or lists, or reuse of any copyrighted components of this work in other works.
Abstract—Partial discharges are not only a consequence but a symptom of insulation degradation in electrical equipment. Partial
discharges are revealed outside the insulation as very narrow, high frequency pulses, superimposed to the grid frequency high level
voltages or currents, and thus fairly hard to detect. This paper presents an application of a new inductively coupled sensor that allows
these signals to be measured in an accurate and inexpensive manner. The sensor model and its frequency and geometry dependent
behavior are studied to increase the probe sensitivity as much as possible. Its measuring capability in real insulation systems subjected
to high partial discharge activity will be tested.
Index Terms—Partial discharges, high frequency pulses, inductive sensor.
Ideal insulation systems are defined as having the capability of blocking any current when a voltage is applied to them. This
property has limitations since there is a voltage threshold above which the dielectric loses its insulating properties and conducts a
certain current responsible for its breakdown. Despite the fact that it is possible to measure this breakdown voltage accurately,
unexpected failures have been reported in electrical equipment working at voltages well below its breakdown value due to
premature insulation ageing. One of the most wide-spread causes of damage are local discharges arising at defect sites within the
dielectric or small volumes where the electric field is intensified. These are the so-called partial discharges (PD) that do not
cause an immediate insulation failure in the bulk of the insulating material, but accelerate the material ageing , .
A great deal of effort has been made to measure these partial discharges at electrical machines and power cables, since they
provide useful information regarding their insulation status. This is very important for predictive maintenance of high voltage
equipment, but in certain applications even at low voltage machines, partial discharges can arise .
Partial discharges are low-magnitude current pulses flowing for some few nanoseconds through insulators usually
withstanding high voltages at power frequency. The superposition of low-magnitude high-frequency pulses on high-magnitude
low-frequency voltages is a real instrumentation problem that has been dealt with by different types of circuits and sensors ,
The traditional approach for measuring a PD consists of a capacitive divider, where a high voltage capacitor is connected in
Inductive Sensor for Measuring High Frequency
Partial Discharges within Electrical Insulation
G. Robles, Member, IEEE, J.M. Martínez, Member, IEEE, M.V. Rojas, and J. Sanz, Member, IEEE
series with a measuring impedance, represented in Figure 1 as another capacitor. When a PD takes place, its transmission path
will be the two existing capacitive branches: the measuring branch and the equivalent capacitor that models any insulating
system. Current pulses outside the insulation are measured as voltage pulses in the measuring impedance . Despite the fact
that this circuitry is used in current standards, its bandwidth limitations (up to 500 kHz) do not allow registering the pulse itself,
but it does allow registering of the pulse amplitude. Moreover, it can register the superposition of two successive PDs leading to
integration and superposition errors .
Fig. 1. Basic partial discharges measuring circuit.
The amplitude, together with the occurrence of the PD referring to the phase of the applied 50 Hz sinusoid provide sufficient
information so as to know the type and origin of the partial discharge .
However, recent works point out that measuring apparent charge amplitude and applied phase voltage must be completed with
information regarding the PD pulse waveform, and this requires wider frequency band detection. An inductive coupling is then
suggested by means of high frequency current transformers (HFCT) clamping the capacitor measurement branch . These
sensors provide excellent results when measuring PD pulse waveforms, but are made of expensive ferromagnetic materials able
to conduct up to 50 MHz flux lines . In addition, they may become saturated when connected to high voltage leads. In order to
overcome these drawbacks, air cored transformers have been used as well, but due to the high frequencies involved, quite
complex distributed parameters models are needed in order to study their behavior . This was the reason for designing a new
air-cored inductive coupling sensor capable of measuring PD pulse waveforms in a simple and inexpensive setup .
In this paper, this new sensor’s applicability to the measurement of partial discharges is analyzed. The model of the sensor is
introduced in the next section. Then, an explanation of its behavior in the frequency band of interest and a study of the sensitivity
using a theoretical and experimental view are presented. Afterwards, the sensor will be used to measure a known partial
discharge activity appearing within an electric motor insulation system. The pulse waveforms and the PD pulse magnitude-phase
pattern are compared with signals given by conventional devices. For this purpose, measurements of the well-known PD pattern
due to air ionization will be done as well.
II. SENSOR MODEL
This new probe has a very simple geometry, as it is made of a strip forming a rectangular loop on a printed board placed close
and parallel to the primary conductor; this is also printed to keep the parameter a constant, Figure 2 and photograph in Figure 3.
Fig. 2. Inductive Sensor Schematic
Fig. 3. Inductive Sensor detail. Two connectors to the primary conductor are clearly seen at the left and right of the picture. The output is taken from the
coaxial connector at the bottom.
The operating principle is also very straightforward: the current derived from the partial discharge flows through the primary
conductor and creates a circular magnetic field which varies with time. According to Faraday’s Law, the magnetic flux through
the loop surface induces a voltage in the turn ends proportional to the derivative of the primary current as shown in Equation (1):
Where e is the voltage across the ends of the turn, M is the mutual inductance and i is the current to measure. As can be seen in
equation (2) M depends on the geometry of the turn and on its separation, a, from the primary conductor, and the turn. To
achieve a good sensitivity, the turn should be close and parallel to this conductor.
The sensitivity also depends on the frequency of the measured signal. PDs can easily reach 100 MHz, so the output voltage of
the probe is expected to be high . The main limitation in raising the frequency of the signal is imposed by the frequency
response of the electric model of the sensor. Considering a lumped parameters model under the constraints stated in , the
equivalent circuit of the probe shown in figure 4 is obtained.
Fig. 4. Electric equivalent circuit
The Equations for the inductance L, capacitances C1 and C2 and resistance R are respectively,
4cos( /2 )sin( /2 ) 1(6)
2 cos( / ) 1
Where l1 and l2 are the lengths of the probe as shown in figure 2, h is the width and d the thickness of the loop wire. Finally, σ
and δ are the conductivity and the skin depth of the copper strip respectively .
In figure 4, C1 is the capacitance between the long sides of the rectangle whereas C2 is between the short sides. Since the loop
is designed making l1>>l2, the latter capacitance is negligible compared to the former and it is no longer considered in the
The impedance Z0 = 50 Ω represents the input impedance of the measuring instrument since the coaxial cable and the
oscilloscope have matched impedances avoiding reflected waves.
Two probes, #1 and #2 with lengths, l1, of 6 cm and 12 cm respectively, were used to test their behavior measuring PDs. As
shown in Table 1, the electrical parameters are different for each probe and hence, the frequency response changes. Notice, that
the resistance is not included in the table because it changes with frequency. Figure 5 shows the Bode plot of the transfer
function Vout/I for both probes when the impedance Z0 is considered. This probe shows a high pass behavior especially in probe
#1 where it is more pronounced. Hence, disturbances with low frequency components will be effectively rejected. The plot
corresponding to loop #1 has a resonance at 680 MHz whereas the plot corresponding to loop #2 has a resonance at 350 MHz.
The impedance Z0 includes a low frequency pole that slightly reduces the bandwidth of the probe. This load effect, introduced
when the oscilloscope is connected, does not effectively damage the signal shape as will be shown below. Nevertheless,
connecting an amplifier in the same printed board would prevent this load effect from happening. Moreover, the amplifier could
include an integrating stage to give the primary current directly instead of its derivative. These improvements are left for future
works and this paper will highlight the performance of the plain probe measuring PDs.
ELECTRICAL AND DIMENSIONAL PARAMETERS FOR BOTH PROBES
Parameter Probe #1 Probe # 2
d [µm] 35 35
a [mm] 1.02 1.02
Fig. 5. Frequency response for both probes.
III. SENSOR PERFORMANCE
In order to study how the geometry affects the sensor behavior, PD measurements were taken using the former two probes
with different lengths l1. Additionally, the reliability of the probes was tested measuring the signals simultaneously with one of
these two commercial high frequency current transformers with the following bandwidths: TechImp Clamp HFCT 39, 80 MHz,
and Bergoz FCT-028-05:1-LC-B, 450 MHz. The device under test was the phase to phase insulation from a 370 W induction
motor AEG Type AM 71 NY4 Q4.
Fig 6. Experimental setup for partial discharges calibration and testing.
The measuring procedure in these experiments consists of three steps. Firstly, the amplitude of a known signal injected by a
calibrator is compared with the amplitude of the signals given by probes #1 and #2. Secondly, a sensitivity test is done for
different charge injections. In these experiments, two calibrators from different manufacturers were used: Lemke LDC-5 with a
clear output signal and TechImp’s with a better charge selection range. Finally, the calibrator is disconnected and a high voltage
transformer HIGHVOLT WGBS 6.6 kVA/100 kV, with tunable voltage is connected to the test specimen. Figure 6 shows the
experimental setup mounted for the measurements. The upper capacitor in the divider is a HIGHVOLT 1 nF, 100 kV, the lower
is a Lemke LDM-5/U measuring impedance. The sensor and the HFCT are connected to the wire to ground so there is no need to
have special care with cable insulation.
A. Signal Test
As a previous approach to the probe performance, these measurements are based on a known pulse with the same
characteristics as the phenomenon to measure. This is the reason for using a standard calibrator, in this case Lemke LDC-5,
because it injects a known and constant charge magnitude into the detection circuit. Both probes with different lengths were
connected to measure the same charge flow, 500 pC in this particular case, and the results are shown in Figures 7 and 8.
The purpose of these Figures is twofold: to show the real output of the sensor and to demonstrate that the signal given by the
inductive sensor doubles when the length l1 doubles. Channel 1 represents the signal obtained with the TechImp HFCT and
channel 2 is the raw signal given by the new probe which has to be integrated to obtain the same waveform as channel 1. This is
in agreement with the sensor principle, since the higher the effective area is, the more the magnetic flux is linked. Hence,
increasing the length of the probe increases the sensitivity of the sensor in the same magnitude. However, as derived from the
Bode plot, the effective bandwidth (avoiding resonance) of the sensor is reduced as the length increases, and simultaneously, its
high-pass behavior is limited. Therefore, there is a compromise between bandwidth and sensitivity. Moreover, increasing the
length would increase the total length of the probe and the restriction of being far smaller than the wavelength of the signal
would not hold. In such case, the lumped parameters model would no longer be applicable and reflections of the wave travelling
through the length of the probe would occur.
Fig. 7. Calibration PD pulse detected using probe #1 (l1=6cm). Channel 1 is the signal obtained with the HCFT. Channel 2 is the signal obtained with the
inductive probe. The vertical scales are 50 mV/div and 5 mV/div for channel 1 and channel 2, respectively. The timebase is 40 ns/div and the sample frequency
is 2.5 GS/s.
Fig. 8. Calibration PD pulse detected using probe #2 (l1=12cm). Channel 1 is the signal obtained with the HCFT. Channel 2 is the signal obtained with the
inductive probe. The vertical scales are 50 mV/div and 10 mV/div for channel 1 and channel 2, respectively. The timebase is again 40 ns/div and the sample
frequency is 2.5 GS/s.
B. Sensitivity Test
The setup is the same as in Fig. 6 using the TechImp calibrator because it has a larger number of steps in the charge range. In
the experiment 10 pC, 30 pC, 40 pC, 70 pC and 100 pC were injected as reference charge magnitudes. The charge injection is
plotted versus the integrated amplitude of the output of the 12 cm inductive sensor and divided into the mutual inductance,
giving the results shown in Fig. 9. The relationship is a straight line, as expected, so the sensitivity yields 6 µA/pC for this
calibrator and this setup. The probe is able to measure PDs as low as 10 pC without amplifying the signal.
20 4060 80100
Injected Charge [pC]
Fig. 9. Amplitude of the integrated output of the 12 cm probe versus charge injected with the Techimp calibrator for the current setup.
C. PD created from power voltage
Once controllable calibration signals are detected properly, the sensor is connected to measure real PD created by power
frequency high voltages. In order to guarantee partial discharge activity within the dielectric layers, a high voltage (1400 V, 50
Hz) is applied to the phase to phase insulation system of the 370 W induction motor.
Previously, amplitude measurements from the calibration signals were used to set an appropriate trigger level in the
oscilloscope so that many PD pulse events coming from both sensors were visible.
Figure 10 represents one partial discharge pulse measured simultaneously with both sensors connected individually one after
the other to the same primary conductor to measure the same charge flow. The outputs are connected separately to different
channels of the oscilloscope. Again, notice that the channels are in different scales and one signal is double the other maintaining
the same wave shape.
Fig. 10. PD pulse detected at 1400 V using both probes simultaneously. Channel 1 is the 6 cm probe with 100 mV/div and channel 2 is the 12 cm probe with
Likewise, two examples of these measurements with probe # 1 are presented in figures 11 and 12. The upper plot represents
the signal obtained with the high frequency current transformer and the lower plot represents the signal acquired with the
inductively coupled sensor after being integrated off-line with Matlab. This numerical integration leads to wave shapes very
similar to the one simultaneously acquired by the HFCT. The upper plot measurements in figure 11 have been taken with the
commercial transformer by TechImp, with a bandwidth up to 80 MHz, whereas the Bergoz transformer in figure 12 had a
bandwidth up to 450 MHz. The differences between both figures are negligible, so a bandwidth of 80 MHz is sufficiently high
for this kind of electrical PD detection. The load effect shown in the bode plot in Figure 5 does not damage the response of the
probes, and the outputs of the HFCT and the designed probe are the same.
0 0.51 1.522.533.54
Fig. 11. Current pulse corresponding to a PD at 1400 V measured with a the Techimp HFCT (80 MHz; upper plot), and with the inductive probe (lower plot).
00.51 1.52 2.533.54
0 0.51 1.52 2.53 3.54
Fig. 12. Current pulse corresponding to a PD at 1400 V measured with the Bergoz HFCT (450 MHz; upper plot), and with the inductive probe (lower plot).
IV. INDUCTIVE SENSOR MEASURING PARTIAL DISCHARGE PATTERNS
Despite the fact that PD pulse waveform is a promising technique used in diagnosis, for any detection system it is necessary to
measure proper partial discharge patterns referred to the phase of the applied voltage (PRPDA-Phase-Resolved Partial Discharge
Analysis). In fact, partial discharges arising from different types of insulation defects lead to different partial discharge patterns
. Thus the presented inductive sensor must be able to measure these high frequency pulses superimposed to 50 Hz signals that
have much higher magnitudes. Moreover, the bandwidth of the sensor must allow both the measurement of signals in the range of
nanoseconds and the occurrence of those PD events within one cycle, in order to help in PD source identification.
Overstressing the phase to phase insulation system in one motor results in a typical partial discharge pattern arising from
internal and surface discharges. These kinds of discharges come from microscopic gaseous voids inside solid insulation bulk and
they appear when both the applied voltage and its slope are high enough . As can be seen in figure 13, the inductive coupled
sensor detects these signals as expected, since its phase occurrence is in the [0º, 90º] and [180º, 270º] ranges. The 50 Hz signal
represents the high voltage applied to the device under test and it is acquired with a capacitive divider.
Fig. 13. PD pulses within electrical motor insulation system referring to the applied high voltage (1400 V). Plot obtained with a Lecroy WavePro 950 at 25
On other hand, figure 14 shows the result of measuring another potential cause of PD occurrence: point-plane conductors
within air . This configuration is responsible for a specific type of discharge called corona discharges that typically appear
around 270º phase of applied voltages . In addition, one high frequency corona pulse is presented in figure 15.
Fig. 14. Detected corona PD pattern referring to the applied high voltage (2400 V). Plot obtained with a Lecroy WavePro 950 at 25 MS/s.
Fig. 15. Detected corona PD pulse from one commercial HFCT (upper) and the integrated one from the new sensor probe (lower).
As can be seen from figures 13 and 14, this sensor successfully decouples sinusoidal power frequency signals from high
frequency PD pulses and measures nanosecond PD pulse waveforms as well.
This paper presents a new inductively coupled probe that can be used for partial discharge detection. The sensor behavior has
been tested for two different lengths and its capability for broadband partial discharges measurements has been compared with
commercial modern probes. This sensor is notably simpler and less expensive than others. It can be embedded within real
insulation systems to measure without interrupting the electrical circuit which opens the possibility of taking on-line
measurements to help in power cable and machine diagnosis. In addition to measuring classical partial discharge patterns, this
probe has the appropriate bandwidth to read partial discharge pulses in the range of nanoseconds. In this way, this system is
completely compatible with modern partial discharge recognition systems. It would be of interest to study how to increase
sensitivity using hardware amplifiers and to design proper integration circuits as a future work.
 Dissado L., Fotherhill J.; “Electrical Degradation and Breakdown in Polymers”; IEE Materials and devices series, 1992.
 Stone G., Boutler E.A., Culbert I., and Dhirani H.; “Electrical Insulation for Rotating Machines: Design, Evaluation, Aging, Testing and Repair”; IEEE
Press Series on Power Engineering, 2004.
 Boggs, S.A.; “Partial discharge- Part II: Detection and Sensitivity”; IEEE Electrical Insulation Magazine, vol. 6, pp. 35-42, Sep/Oct. 1990.
 Bilodeau, T.M., Fitzpatrick, G.J., Shea, J.J., and Dollinger, R.E, “The Design and Application of a Novel High Speed Partial Discharge Diagnostic
System”; IEEE Transactions on Instrumentation and Measurements; vol. 37, pp. 25-29, Mar. 1988.
 Bartnikas, McMahon, “Engineering Dielectrics: Volume I, Corona Measurement and Interpretation”; ASTM; 1979.
 IEC Technical Specification “Off-Line partial discharge measurements on the stator winding insulation of rotating electrical machines”; IEC TS 60034-
27; December 2006.
 Cavallini A., Montanari G.C., Contin A. and Puletti F.; “A New Approach to the Diagnosis of Solid Insulation Systems Based on PD Signal Inference”;
IEEE Electrical Insulation Magazine, vol. 19, pp 23-30; March/April 2003.
 V. Dubickas and H. Edin; “High frequency Model of the Rogowski Coil with a Small Number of Turns”; IEEE Transactions on Instrumentation and
Measurements; vol. 56, pp. 2284-2288, Dec. 2007.
 Robles G., Martínez J.M., Rojas M., Sanz J.; “Inductively coupled probe for the measurement of partial discharges”; Review of Scientific Instruments;
The American Institute of Physics; vol. 79, pp 055104, may 2008
12 Download full-text
 Mandani M.R. and Miller T.A., “Current Density Distribution Measurement of Negative Point-to-Plane Corona Discharge” IEEE Transactions on
Instrumentation and Measurements; vol. 47, pp. 907-913, Aug. 1998.
Guillermo Robles (M’03) was born in Madrid, Spain, in 1969. He received both his degree in Electronic Engineering in 1993 and his PhD
in Electronic Engineering in 2002, from the Universidad Pontificia de Comillas de Madrid. Since 2002, he is a Professor in the Department
of Electrical Engineering in the Universidad Carlos III de Madrid. His main research fields are related to the developing of sensors,
instrumentation and measurement techniques for high frequency currents due to partial discharges in noisy environments. He is also
experienced in the study and characterization of magneto-optic sensors based on the Faraday Effect to measure currents and in the
characterization of the behaviour of magnetic materials at high frequencies.
Juan M. Martínez-Tarifa was born in Lorca, Spain in 1975. He received his M.Sc. degree in Electronic Engineering from the Universidad
de Granada, Spain in 1999, and his M.Sc. degree in Physics from the Universidad de Granada, Spain in 2000. He received his Ph.D. degree
in Electrical Engineering from the Universidad Carlos III de Madrid, Spain in 2005. From 2000 to the present time he has been an Assistant
Professor in the Department of Electrical Engineering in the Universidad Carlos III of Madrid. He is currently Technical Supervisor at the
High Voltage Research and Tests Laboratory (LINEALT) at UC3M where he is working on insulation systems diagnosis within power cables
and electrical machines.
Mónica V. Rojas-Moreno was born in Duitama, Colombia in 1979. She graduated with a degree in Electrical Engineering from the
Universidad Industrial de Santander (UIS), Colombia in 2003. From 2003 to 2007, she was Professor in the Escuela de Ingenierías Eléctrica,
Electrónica y Telecomunicaciones in the Universidad Industrial de Santander. She is currently pursuing a PhD degree in Electrical
Engineering at the Universidad Carlos III de Madrid (UC3M), Spain. Her areas of research interest include Electromagnetic Fields, Partial
Discharges, Instrumentation, High Frequency Signals and Modelling of Circuits.
Javier Sanz-Feito was born in Madrid, Spain, in 1954. He received his M.Sc. and Ph.D. degrees in electrical engineering from the
Universidad Politécnica de Madrid (UPM) in 1976 and 1980, respectively. From 1976 to 1977 he was employed by INITEC S.A, a Spanish
engineering consultant. From 1977 to 1981, he was an Assistant Professor with UPM. From 1981 to 1982, he was a Professor of Electrical
Engineering at Universidad de Oviedo, Spain. In 1992, he joined the Universidad Carlos III de Madrid (UC3M), where he has been Dean
of the Higher School of Engineering (EPS) and Head of the Electrical Engineering Department. He is currently Director of the High Voltage
Laboratory at UC3M. He has conducted research in modeling and control of electrical machines and drives, monitoring of power transformers and
multifactorial aging processes in electrical insulating materials.