Page 1 of 13

Localizing partial discharge in power transformers by combining acoustic

and different electrical methods

By

Stefan M. Hoek, Rene Hummel, Alexander Kraetge, Benedikt Kästner and Ulrike

Broniecki

Page 2 of 13

Page 3 of 13

LOCALIZING PARTIAL DISCHARGE IN POWER TRANSFORMERS BY COMBINING ACOUSTIC AND DIFFERENT ELECTRICAL

METHODS

Stefan M. Hoek1, Rene Hummel1, Alexander Kraetge1, Benedikt Kästner2, Ulrike Broniecki2 1OMICRON electronics, Austria

2Technical University Berlin, Germany

Abstract

Power transformers are important nodes in the electrical power grid. The reliability of such electrical

systems depends on the quality and availability of the power apparatus. Common reasons for

breakdowns are problems in the insulation system. Detecting partial discharge (PD) in the insulation

and windings of a power transformer at an early stage reduces the risk of total breakdown of power

apparatus.

One method to detect PD is the acoustic measurement. With this technique a detection and

localization of PD is possible by placing acoustic sensors on the surface of the transformer tank. The

low level of electrical interferences from outside the measurement setup constitutes one of the

strengths of this method. A further advantage is the ability of identifying the position of the PD

source, for it is very important to estimate the risk and to solve insulation problems fast and effective.

1 Partial discharge in high voltage apparatus

Partial discharge measurements on transformers are an accepted tool of quality control, in factory and

on site. Different PD measuring techniques are using different physical peculiarities of the PD

phenomenon, e.g. electric discharge currents (acc. to IEC 60270 [1]), gas formation (DGA - dissolved

gas analysis), electromagnetic (UHF Measurement) or acoustic radiation (some tens of kHz). Partial

discharge measurements according to IEC60270 standard are often the basis for acceptance tests of

the insulation system of high voltage (HV) equipment [2]. The main benefits of acoustic PD

measurement are the possibility of detecting PD without intervention into the device under test (DUT)

and the possibility of localizing the PD source by the accuracy of some centimeters. In case of an

evidence for PD, the location of the potential PD source can be important to verify PD and to estimate

the risk of a complete failure. The knowledge of PD location is also crucial for the assessment of the

asset and the process of maintenance or repair.

2 The propagation behavior of acoustic PD signals in transformers

The acoustic response of PD inside a transformer is typically measured by a piezo-electrical sensor in

the frequency range of some tens of kHz up to some hundreds of kHz [3]. Due to the resonant

character of the sensors, the measured acoustic PD signal is inherent overlaid by oscillations as

illustrated in Figure 1. For that reason the determination of the frequency content and proper signal

form is difficult [4]. Using the difference in arrival time of the acoustic PD signal at multiple sensors,

algorithms compute the location of the PD source.

The complex physical processes involved in sound propagation and the large structural differences

between different DUTs may be challenging during the measurement. The following parameters have

to be considered:

The PD source position and the inner structure of the transformer mainly influence the

propagation path.

Page 4 of 13

More than one propagation path from source to the sensor is possible (direct oil, reflection,

steel path).

The speed of sound depends on the propagation path (crossed medium), the frequency and

temperature.

Depending on the position of the source and the inner structure of the transformer, a direct oil

path propagation may prevent a proper measurement by attenuating the signal too much.

The individual consideration of the measurement setup and the inner structure of the DUT, a

cautious interpretation of the measurement results by an experienced person is essential.

Figure 1

Acoustic PD Signal

Speed and damping of the acoustic waves are dependent on the crossed medium, frequency range and

temperature [5], [6]. Figure 2 and Figure 3 shows the variation of the velocity of sound in transformer

oil for different temperature and frequency. For example, the propagation speed decreases during the

heat-up period of the transformer by approximately 15%, from about 1400 m/s at 20 °C to 1200 m/s at

80 °C.

Figure 2

Dependence of sound velocity on temperature in

transformer oil (f=150 kHz) [5]

Figure 3

Dependence of sound velocity on frequency

in transformer oil (T=60 °C) [5]

The propagation path is often complex. According to Figure 4 multiple propagation paths of the

emitted sound wave are possible. Depending on sensor and PD location, multiple acoustic wave

components of the same PD event are potentially detected by one sensor and overlay the direct oil

signal as illustrated in Figure 5. The acoustic wave can be reflected by the tank wall, core, winding,

flux shields and other components.

Components of the reflected wave will arrive at the sensor position later than the signal travelling a

direct path. Furthermore, the acoustic wave can couple into the transformer wall and propagate

through the steel of the tank to reach the sensor. Due to the higher propagation speed in steel of about

3.000 - 5.000 m/s [3], the so-called steel wave signal can reach the sensor earlier than the waves

1000

1100

1200

1300

1400

1500

1600

-40 -20 0 20 40 60 80 100 120 140

Ve

loci

ty (

m/s

)

Temperature ( °C)

1250

1260

1270

1280

1290

1300

1310

0 200 400 600 800 1000 1200

Ve

loci

ty (

m/s

)

Freqency (kHz)

Page 5 of 13

following the direct oil path. This effect complicates the automated determination of the starting point

of the direct oil signal.

Figure 4

Possible propagation paths in the DUT

Figure 5

Acoustic PD signal components according

to propagation paths

The measurable direct oil signal at the sensor position depends on the intensity of the causative PD

event [4] and on the damping in the propagation path. Therefore, the attenuation by core, winding,

transformer board, flux shielding etc. should be as low as possible. For that reason, the search for

sensor positions that ensure good signal quality is essential during measurement procedure. The

knowledge about the inner structure of the transformer is helpful for good positioning and

repositioning of the sensors.

3 Localization of PD

Different algorithms can be used to perform a time-based localization of PD. The input information

used by the algorithms is the time of arrival of the signals propagating on direct oil path wave at

multiple sensors. The exact time of arrival has to be determined by evaluating the measured signal.

A criterion for the starting point can be found e.g. by investigation of energy steps [7] or by threshold

criteria [8].

Figure 6

Absolute and relative times in a two-sensor-setup

Sensor S 1

Sensor S 2

Occurence of PD time

t 1

t 2

Δ t 1 , 2 Amplitude

Page 6 of 13

The relative arrival times at different sensor positions lead to time differences (Δt1,2). These time lags

are the only available data in all-acoustic measurements, when the data acquisition is triggered by the

acoustic signal at one of the sensors. If the time delay between occurrence of a PD and the arrival of

the associated acoustic wave is available, the absolute propagation times (t1, t2) from source to sensor

can be used for localization, Figure 6.

The exact timing of the emission of the PD signal can be estimated e.g. by an electrical PD

measurement according the IEC 60270 standard or a measurement in the ultra-high frequency (UHF)

range. In the latter case, sensors within the transformer walls can be used to collect the high frequency

electromagnetic wave that is emitted during PD [9]. The principle and a measurement setup are shown

in Figure 7 and Figure 8.

Figure 7

Schema of the measurement setup for the UHF trigger

signal

Figure 8

UHF probe installed

in a DUT

The distance between sensor and source is calculated using the available absolute or relative

propagation times and an assumptive average propagation speed. With the determined distances and

the sensor positions a geometrical localization of the PD source can be performed in several steps.

Figure 9

Principle of the acoustic localization

The arrival time at a single sensor and the timing of PD occurrence leads to the surface in the shape of

a sphere around the sensor position on which the PD source is supposedly located, Figure 10. The

radius r is proportional to the absolute propagation time (t1) and the propagation speed.

In all-acoustic measurements the data of a single sensor does not contain meaningful information. In

this case the data of two acoustical sensors - the relative time Δt1,2 – delivers a distance difference

(Δd1,2) and therefore a hyperbolic sphere (Figure 11).

Page 7 of 13

Figure 10

Spatial information from one absolute time (t1)

Figure 11

Spatial information from one relative time (Δt1,2)

The position of the source can be specified with the information of more sensors. For this purpose

multiple of the described geometrical shapes are intersected. The absolute propagation time of the

signal at a second sensor leads to a second sphere, the resulting intersection shape is a circular. In a

further step the absolute coordinates of the source can be estimated by intersecting the circulars of

three sensors. This procedure is shown in Figure 12. Figure a)-c) shows the spheres around three

acoustic sensors, in section d) the resulting intersection circulars and the estimated point of the

acoustic source are displayed. In an all-acoustic measurement environment the approach is in

principle identical. In this case a fourth sensor delivers the necessary information to estimate a point.

a) Sensor A.1 (blue)

b) Sensor A.2 (green)

c) Sensor A.4 (black)

d) Resulting circulars by use of

Sensors A.1, A.2 and A.4

Figure 12

Source localization with three sensors using absolute times

Page 8 of 13

The depicted method is based on a direct propagation path for the acoustic wave from source to

sensor. As described above, the transformer must not be considered as an empty box and the

propagation speed is highly dependent on the travel path. For that reason, the model is always a

simplification of the real setup. Thus, an inaccurate or wrong localization of the source position is

possible.

To ensure reliable measurement results, a workflow is proposed that is based on an iterative relocation

of the sensors with the intention to find positions with a minimal and undisturbed path between sensor

and source. At the same time, the assembly of the sensors among themselves is restricted by a

minimal distance and a proper placement in multiple axes.

4 Improving sensitivity and accuracy

Due to the inherent inaccuracy of the assumed model, all results should be verified. This can be

accomplished by multiple measurements with identical and varied sensor setups and software

parameters. The outcome for different sensor setups must not be entirely congruent but the resulting

cluster of points leads to the identification of incorrect or ambiguous results and to the proper

localization of the affected part of the DUT, Figure 13.

Figure 13

Plot of multiple localization results to identify mislocation

In addition a verification of the recorded pattern and the identified failure location often can be done

with potentially available data from the DUT manufacturer (inner structure) and other investigations

like electric PD measurement.

a) Unfiltered Signal

b) Filtered signal

c) Filtered signal (zoom)

Figure 14

Effect of high-pass filtering of an acoustic PD signal

The signal quality and hence the sensitivity of the measurement can be enhanced by digital signal

processing. High-pass, low-pass and band-pass filters can be used to reduce the impact of acoustic

disturbances like core noise (Barkhausen effect), oil pumps or fans, Figure 14. A preliminary selection

Page 9 of 13

of the considered frequency range is made by the choice of the sensor type [3] as piezo-electric

sensors have limited bandwidth. Figure 14 shows the effect of a 45 kHz high-pass filter on a

measurement which was recorded with a sensor with 75 kHz resonance frequency.

Another opportunity to raise the signal-to-noise ratio of recurring PD events is provided by the

averaging technique, Figure 15 and Figure 16 [6]. The sample values of multiple PD events following

each other are added and in further consequence divided by the number of iterations. The PD signal is

constructively superposed with each addition. The superimposed noise is statistically distributed and

therefore the corresponding sample values are different for each occurrence of PD. For that reason, an

appreciable reduction of noise can be achieved with a sufficient number of summands (typically 10 to

some hundreds) because of the partially cancellation of sample values. Due to the required correlation

of the time windows this technique is limited to applications with a sufficiently stable electric or

acoustic trigger source.

Figure 15

Principle of signal processing through averaging

by using an electrical trigger (acc. IEC or UHF)

Figure 16

Outcome of the averaging algorithm at different

numbers of runs

5 Study case of localization

The described investigation was performed on a 150 kV/20 kV three-phase power transformer with a

nominal capacity of 16 MVA. Figure 17 shows the spatial computer model and the electric-acoustic

test setup. The electric multi-phase PD measurement pattern (Figure 18) clearly indicates a fault

nearby phase W.

Figure 17

Test setup (el. and acoustic PD test setup) Figure 18

Electric PD pattern

Page 10 of 13

A first measurement was performed with the sensor setup shown in Figure 19 to acquire a large-scale

scan of the windings and the high voltage bushing of phase W. Positioning of sensors was

substantially limited by cooling facilities and stiffeners on the housing. Signal quality on all sensors

was insufficient for reliable determination of arrival times and therefore a localization of the

discontinuity was not possible for this setup. Still, the high amplitude and the short absolute delay

lead to the assumption that the PD source may be located near to the sensor A.3 (red). Therefore, all

sensors were relocated to the relevant area.

Figure 19:

Initial sensor setup Figure 20:

Result of measurement for the

initial sensor setup, sensors A.1-A.4

After sensor relocation the signal form on all sensors indicated propagation paths with little

disturbance, although the signals on sensor A.1 (blue) and A.3 (red) contained clear steel wave

components (Figure 21 and Figure 22). This characteristic was probably caused by the sensor

positions on the top of the DUT that were unavoidable due to the construction of the transformer.

Figure 21

Final sensor setup Figure 22

Result of measurement with automatically computed arrival

times for the final sensor setup, sensors A.1-A.4

Page 11 of 13

The computed spheres at each sensor were checked separately to be located inside the DUT with

plausible radii. Subsequently the circular intersections of all possible sensor triples were reviewed, see

Figure 23. The sensor combinations in a) and b) immediately provide nearby intersection points, c)

and d) suggest a comparable area.

a) Sensors A.2,A.3,A.4

b) Sensors A.1,A.3,A.4

c) Sensors A.1,A.2,A.4

d) Sensors A.1,A.2,A.3

Figure 23

Circular intersections calculated with the automatically computed arrival times for all sensor triples

The starting times were calculated by using the energy criterion as well as the overall propagation

velocity. These starting times were manually adapted to consider the steel wave characteristics on

sensors A.1 and A.3, see Figure 24. Using the adapted parameters, identical intersection points for all

sensor triples were found, see Figure 25.

a) Sensors A.2,A.3,A.4

b) Sensors A.1,A.3,A.4

c) Sensors A.1,A.2,A.4

d) Sensors A.1,A.2,A.3

Figure 24

Final results of measurement with manually

adapted arrival times, sensors A.1-A.4

Figure 25

Final results for all sensor triples

5 Conclusions

This paper describes the basic idea of the method of time based acoustic localization. The PD signals

are captured using three or more piezoelectric acoustic sensors, magnetically mounted on the tank at

Page 12 of 13

different locations. For localizing the source, the time delays between the recorded acoustic signals or

between an electrical signal and the acoustic signals are used to get information about the propagation

of the acoustic signal inside the transformer tank and the distances between signal source and sensors.

The received sensor signals are filtered and processed to obtain the difference between the signal

arrival times at each sensor.

In addition to the acoustic measurement, electrical measurement can be used in parallel to obtain a

trigger signal. This can be essential for the success of localization. Alternatively to the electrical

measurement according the IEC 60270, unconventional measurement techniques e.g. in the UHF

range, can be used to gain a trigger source for an acoustic measurement.

A case study of PD localization is shown. The procedure and successes of an acoustic measurement

with electrical trigger has been proved. The results are analyzed and visualized by using a 3D model.

References 1. IEC 60270 (2000) „High-voltage test techniques - Partial discharge measurements” International

Electrotechnical Commission, Publication 60270, 2000

2. Tenbohlen, S.; Denissov, D.; Hoek, S.M.; Markalous, S.M. “Partial Discharge Measurement in

Ultra High Frequency (UHF) Range” IEEE Transactions on Dielectrics and Electrical Insulation,

Vol 15, No 6, Dec 2008

3. C57.127 (2007) “IEEE Guide for the Detection and Location of Acoustic Emissions from Partial

Discharges in Oil-Immersed Power Transformers and Reactors”, The Institute of Electrical and

Electronics Engineers, Inc. New York, USA, 2007

4. Bengtsson, T.; Leijon, M.; Ming, L. “Acoustic Frequencies Emitted by Partial Discharges in Oil”, 8

th International Symposium on High Voltage Engineering, 1993, p.113-116

5. Howells, E. ; Norton, E.T.: “Parameters affecting the velocity of sound in transformer oil” IEEE

Transactions on Power Apparatus and Systems (1984)

6. Lundgaard, L.E. „Partial Discharge – Part XIV: Acoustic Partial Discharge Detection – Practical

Application“, IEEE El. Insulation Magazine Sep 1992 Vol.8,No.5

7. Große, C.U.; Reinhardt, H.-W.; (1999) „Schallemissionsquellen automatisch lokalisieren“ MP

Materialprüfung, Jahrg. 41, pp.342, Carl Hanser Verlag, München, Germany, 1999

8. Wagenaars, P.; Wouters, P.A.A.F.; van der Wielen, P.C.J.M.; Steennis, E. F. “Algorithms for

Arrival Time Estimation of Partial Discharge Pulses in Cable Systems” IEEE Vancouver, Canada,

2008

9. Coenen, S.; Tenbohlen, S.; Markalous, S.M.; Strehl, T. (2008) “Sensitivity of UHF PD

Measurements in Power Transformers” IEEE Trans. on Dielectrics and Electrical Insulation, Vol.

15, No. 6, pp. 1553-1558

Biography Stefan Hoek is with OMICRON electronics in Austria since May 2008. He works as product

management with main focus on partial discharge analysis. He studied electrical engineering at the

University Stuttgart (Germany) and worked as research assistant at the Institute of High Voltage

Page 13 of 13

Technologies (IEH). His research focus was on partial discharge localization in GIS with help of

measurements in the UHF range.

Rene Hummel is with OMICRON electronics in Austria since 2009. He works as an application

engineer for partial discharge products, conducting trainings and giving on site support. He studied

electrical engineering at the University of Technology in Berlin (Germany) and did research

concerning the method of pulse sequence analysis (PSA) of partial discharges.

Alexander Kraetge is with OMICRON electronics in Austria since 2006. He works as product

manager for partial discharge products. He studied electrical engineering and received his PhD degree

at the University of Technology in Berlin (Germany).

Ulrike Broniecki studied electrical engineering at the TU Berlin (Germany) and works as research

assistant at the department of High Voltage Engineering at TU Berlin since July 2008. Her research is

focused on acoustic partial discharge measurement and localization.

Benedikt Kästner is studying electrical engineering with focus on high voltage engineering at TU

Berlin. Besides, he is with OMICON electronics as a student worker.

AmericasOMICRON electronics Corp. USA3550 Willowbend BlvdHouston, TX 77054, USAPhone: +1 713 830-4660 +1 800-OMICRONFax: +1 713 830-4661info@omicronusa.com

© OMICRON

Asia-PacificOMICRON electronics Asia LimitedSuite 2006, 20/F, Tower 2The Gateway, Harbour CityKowloon, Hong Kong S.A.R.Phone: +852 3767 5500Fax: +852 3767 5400info@asia.omicron.at

Europe, Middle East, AfricaOMICRON electronics GmbHOberes Ried 16833 Klaus, AustriaPhone: +43 5523 507-0Fax: +43 5523 507-999info@omicron.at

www.omicron.at • www.omicronusa.com

OMICRON is an international company serving the electrical power industry with innovative testing and diagnostic solutions. The application of OMICRON products allows users to assess the condition of the primary and secondary equipment on their systems with complete confidence. Services offered in the area of consulting, commissioning, testing, diagnosis and training make the product range complete.

Customers in more than 140 countries rely on the company’s ability to supply leading edge technology of excellent quality. Service centers on all continents provide a broad base of knowledge and extraordinary customer support. All of this together with our strong network of sales partners is what has made our company a market leader in the electrical power industry.