Received April 6, 2021, accepted April 21, 2021, date of publication April 23, 2021, date of current version

May 4, 2021.
Digital Object Identifier 10.1109/ACCESS.2021.3075288

Overview and Partial Discharge Analysis of

Power Transformers: A Literature Review
MD RASHID HUSSAIN 1 , (Member, IEEE), SHADY S. REFAAT 2, (Senior Member, IEEE),
AND HAITHAM ABU-RUB 2 , (Fellow, IEEE)
1 Department of Electrical Engineering, Qatar University, Doha, Qatar
2 Department of Computer and Electrical Engineering, Texas A&M University, Doha, Qatar
Corresponding author: Md Rashid Hussain ([email protected])
This publication was made possible by NPRP Grant [10-0101-170085] from the Qatar National Research Fund (a member of Qatar
Foundation). Open Access funding provided by the Qatar National Library.

ABSTRACT The high voltage power transformer is the critical element of the power system, which requires
continuous monitoring to prevent sudden catastrophic failures and to ensure an uninterrupted power supply.
The most common failures in the transformer are due to partial discharge (PD) in electrical insulations
which are the results of the insulation degradation over time. Different approaches have been proposed to
monitor, detect, and locate the partial discharge in power transformers. This paper reviews and evaluates
the current state-of-the-art methods for PD detection and localization techniques, and methodologies in
power transformers. Detailed comparisons of PD detection techniques have been identified and discussed in
this paper. The drawbacks and challenges of different partial discharge measurement techniques have been
elaborated. Finally, brief reviews of PD denoising signals, feature extraction of PD signals, and classification
of partial discharge sources have been addressed.

INDEX TERMS Power transformer, partial discharge, condition monitoring, fault diagnosis, feature
extraction.

I. INTRODUCTION The power transformer during the stage of operation may

The power transformers are the utmost fundamental part of undergo various types of faults with different levels of
the power system utilities [1]. The performance and con- severity [5]. Therefore, proper examination for detecting
stancy of the power system utilities are directly dependent on the transformer’s health level is essential for the continuous
the power transformer [2]. Therefore, its health is essential for operation of the electrical utilities. The condition monitoring
the power system’s stability and reliability. Any failure may systems of a power transformer can be classified into two
result in high capital loss with disruption of power supply [3]. main categories; online and offline. The online condition
The power transformers are exposed to different stress con- monitoring methods are preferred over offline methods even
ditions in the form of electrical, mechanical, environmental, though the reliability of offline testing is higher [2]. Offline
and thermal stress. These stresses are potential sources for condition monitoring methods are usually performed during
different internal and external faults in the power transformer. the manufacturing phase of the power transformer to examine
The majority of these faults occur due to the impending the manufacturing defects, which are present in the form
deterioration of the insulation system [2]. To avoid the electric of voids, cracks, and bubbles in the insulation [6]. Offline
supply disruption, condition monitoring is performed. Mon- testing methods include advanced electrical measurements,
itoring is a form of predictive maintenance that determines which are efficient enough to examine the PD activities.
the operating state and the assessment of the functionality of However, offline monitoring lacks the actual electrical and
the power transformer. This helps in preventing possible fail- thermal conditions of the insulations that are different during
ures by taking early action through scheduling maintenance the operating phase of the transformer [7].
tasks [4]. The power transformer is a complex structure experienc-
ing faults, which are categorized into internal and external
The associate editor coordinating the review of this manuscript and faults. Fig. 1 shows the classification of fault in the power
approving it for publication was Giambattista Gruosso . transformer [8]. Global statistics show that 70% - 80% of

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 9, 2021 64587
 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

PD is contemplated as the root cause of the insula-

tion degradation where a complete breakdown can lead to
transformer failure [25], [26]. Different PD analysis tech-
niques for condition monitoring are performed for PD detec-
tion, identification, and diagnosis [27]. Several techniques
were developed to detect PD including electrical detec-
tion [28]–[32], electromagnetic detection [26], [33]–[37],
optical detection [38]–[40], acoustic detection [41]–[45], gas
presence detection [16], [46]–[48], and combinational meth-
ods [41], [49], [50]. PD sources signals are received through
a detector and are further analyzed to identify the locations
and severity of insulation defects. The complex geometry
FIGURE 1. Classification of fault in power transformer. of power transformers with different noise sources becomes
quite challenging to identify the severity of PD defects and
transformer’s faults are internal faults [9]. The initiation localization. The external noises are in the form of natu-
of internal fault starts with the trivial discharge inside the ral sources such as electrical storms, electrostatic interfer-
transformer insulation, which is a transient state. Further, ence, electromagnetic interference (through current cables),
the trivial value of the discharge in the insulation can grow radiofrequency interference (from radio signals), and cross
rapidly and lead to a complete breakdown. Internal faults talk (cables separated by small distance). Whereas, the inter-
arise due to fault in different areas which include the wind- nal noises are mainly caused by the vibrations of the trans-
ing (axial displacement, buckling deformation, disc space former core and windings. The PD detection system is desired
variation, and short circuit turns) [10], tank, insulating oil to precisely identify the severity and locate the defects.
(oxidation, water penetration, dissolution due to temperature The detection system process includes the denoising pro-
rise, and acidity) [11], core (core insulation failure, shorted cess of PD activity, feature extraction, PD classification, and
lamination) [12], terminal (open leads, loose connections, PD clustering methods.
short circuits) [13], cooling system, and tap changer (mechan- The main contributions of this paper are as follows:
ical, electrical, short circuit, overheating) [14]. External • A brief description of the power transformer, its
faults occur due to several reasons including the external advances, and design for the improved insulation.
short-circuit of the power system, overflux, and overload. • Review of state-of-the-art techniques of power trans-
Fig. 2 shows different areas of fault location for transformer former faults, with a focus on various power transformer
which are located in the substation (> 100 kV) [15]. PD forms types, detection, and measurement in power
transformer during the operating state.
• A comprehensive insight into various PD diagnostics
techniques for different types of defects in power trans-
former during operation.
• Further, the common techniques for denoising, feature
extraction, classification, clustering of PD data for local-
ization, and severity of PD source with current advance-
ment have been discussed.
The paper is organized as follows. Section II introduces
partial discharge in the transformer. Section III provides PD
detection techniques for the transformer. Section IV gives the
FIGURE 2. Transformer failure for transformer at substation
(>100 kV) [15]. most efficient PD diagnostics techniques used in the power
transformer. The information regarding different denoising
The power transformer fault detection and condition mon- techniques is discussed in Section V. PD feature extraction,
itoring are crucial to increase the electric system reliability. PD classification, and PD clustering are comprehensively
Several common online condition monitoring techniques are discussed in Section VI, VII, and VIII, respectively. Finally,
used. These include dissolved gas analysis (DGA) [16], [17], section IX concludes the paper.
partial discharge measurement [18], power factor measure-
ment [19], frequency response analysis (FRA) [11], vibra- II. PARTIAL DISCHARGE IN POWER TRANSFORMERS
tion and acoustic analysis [20], dielectric spectroscopy [21], PD is the electrical breakdown in the insulation that does
differential protection [22], transformation ratio [23], and not completely bridge the electrodes resulting in localized
insulation resistance [24]. Among these techniques, partial electric stress. This process is a slow insulation degradation
discharge monitoring can effectively diagnose the trans- and reducing the insulation life of the power transformer [51].
former’s condition with the possibility of advancement in the PD arises when there is a higher electric field than the thresh-
future. old value causing a partial breakdown of the surrounding

64588 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

medium [52]. The transient behavior of the PD has the prop- the insulation system is an impossible task since an ideal
erty of pulsating currents of duration between nanoseconds to insulator cannot be manufactured.
microseconds [53]. PD discharge level cannot always justify Different types of defects causing PD can be present within
the severity due to the possibility of rapid treeing phenomena. the power transformer oil-paper insulation. The defects are
In general, the complete breakdown destroys the insulator, (a) voids formation due to the separation of layers of paper
which does not retain any information regarding PD type [53]. wrapping the windings, where oil impregnation is absent.
Therefore, it requires continuous monitoring to resolve the (b) Moving metallic particles in the insulating oil due to the
issues in the early stage [54], [55]. Each type of defect has its aging process of metallic tanks and manufacturing flaws.
distinguished degradation properties whose PD pattern can be (c) Voids formation in the bushings due to environmental
used to figure out the insulation condition [56]. effects, humidity, and surge voltage (d) Gas bubbles in the
PD phenomenon has a stochastic behavior due to the insulating oil due to aging, impurities in oil, and trapped
unpredictable nature of pulsating PD occurrences. Some steps moisture. (e) Trapped moisture in solid insulation during
of the PD phenomenon can be predicted statistically, but not the manufacturing process. (f) The localized static electric
accurately due to its complexity. PD pulse properties such as charge due to the flow of oil, and the increase of the electric
amplitude, shape, and time of occurrence can examine PD field resulting in the initiation of PD. (g) Tracking in solid
phenomena keeping into consideration their random behav- insulation [59].
ior. Stochastic behavior of PD can be due to the probability Different physical and chemical processes in defects of the
of introduction of the initial electrons, region of electric insulation system provide the foundation for PD monitoring
field strength, development of discharge in defects, ion space techniques. PD investigation started in the 1960s, and for
charge generation rate, variation in the gas constitution and a decade, the research was performed to study void dis-
density, the existence of ionizing radiation, and formation of charge phenomena [60]. Further, in the late 1970s satisfactory
PD pulse from the remainder of previous PD pulse [57]. advancement took place towards distinct PD mechanisms
Partial discharge in a power transformer can be categorized such as treeing, flashover, sparks, avalanche, and stream-
into five types: internal discharge, surface discharge, corona ers. [61]–[63]. The physical events occurring due to PD in
discharge, electrical treeing, and barrier discharge [58]. Inter- the insulation systems of the power transformer are [64]:
nal discharge occurs due to the formation of cavities inside the (a) Generation of mechanical vibrations results in the creation
insulator during the manufacturing process or due to aging. of acoustic waves in the ultrasonic region. (b) Emission of
Also, due to higher voltage/electric stress in the cavities electromagnetic waves in the ultra-high frequency (UHF)
as compared to the surrounding medium. PD occurs if the region. (c) Ozone and nitrogen-based oxide formation due
voltage stress inside the cavity is higher than the inception to a chemical reaction. (d) Emission of thermal and light
voltage. Surface discharge starts at the high electric field energy [64].
region and then propagates to the lower stress region. The Recently, data analytics and sensing technology are cre-
main cause of the surface discharge is due to the fact that ating possibilities for the advancement of the auto-detection
dielectric strength at the interface of the insulations is less of PDs through PD monitoring systems [65]–[67]. A typical
than the insulation. The surface discharge may occur is dif- PD monitoring system comprises a PD signal collection unit,
ferent areas which includes the cable terminal, bushings, a feature extraction unit, and a data analysis unit. PD signal
line-insulator surface. Corona discharge can occur at the collection unit encompasses sensors designed to sense the
sharp conducting points, high voltage bare conductors, and physical phenomena of PD, which emits energy of different
even sharp points at ground potential in air and transformer kinds. PD signals from the detectors can be represented into
oil. Electrical treeing may occur at a high electric field region two different patterns, which are termed as time-resolved
inside the dielectric material due to defects in the form of partial discharge pattern (TRPD) and phase-resolved partial
a gaseous void, sharp electrode-edge, or metallic particle. discharge pattern (PRPD [68]. PRPD represents the ϕ- q-n
The voids under high electric stress generate by-products waveform pattern, where ϕ, q, and n are the PD pulse phase
(ultra-violet light and ozone gas) that decompose the insulator angle, the amplitude of the apparent charge or discharge volt-
and produce new voids. The process of voids generation is age, and the number of pulses, respectively. TRPD signifies
repeated, creating weak points and form the electrical tree the q-t waveform, where ‘t’ is the time of the waveform and
that finally leads to the breakdown. Electrical treeing can also ‘q’ is the same parameter as of PRPD. PRPD’s most common
be formed on the surface of the dielectric under high electric illustration is the phase window method, which splits the
field stress due to contamination which causes flashover on power cycle angle of 360◦ into a smaller phase window for
the surface. Dielectric barrier discharge generally occurs due feature generation [69]. The feature extraction for the PD
to the presence of insulators (generally silica, silica glass, signals unit extracts the momentous qualities (features) from
ceramics, etc.) between the electrode pair [58]. the raw data [70]. Further, these features are used in the PD
The occurrence possibilities of PDs in the power trans- data analysis unit, which is generally furnished with pattern
former can be lowered by better design of the insulation recognition methods such as artificial neural network, fuzzy
system, reducing manufacturing defects in the drying, and clustering, and expert system for distinguishing PD from
impregnation process. Complete elimination of cavities in noise or source of PD and its location [71], [72].

VOLUME 9, 2021 64589

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

FIGURE 3. Partial Discharge detection techniques.

III. PD DETECTION TECHNIQUES

PDs consist of different physical phenomena having dis-
tinguished characteristics mainly electromagnetic waves,
current pulses, heat, vibrations, and acoustic waves.
[20], [73]–[75]. PD detections and measurements are feasible
due to the distinguished physical phenomena, which can
be allotted into two groups: electrical and non-electrical
methods. Fig. 3 shows different types of PD detection tech-
niques for power transformers. This section provides a brief
description of the measurement techniques.

A. ELECTRICAL (EE) DETECTION METHOD

EE detection methods apply the current pulse generated sig-
nal detection method. The circuit is directly connected to the
testing region where PD activity is present and the current
pulse is detected [76]. The International Electrical and Elec-
tronics Engineers (IEEE) and the International Electrotech-
nical Commission (IEC) adopt this method [77]. When using
this method, the level of charge due to PD is calculated to
identify the condition level of insulation. The International
Council on Large Electric Systems (CIGRE) proposed sur-
veying the available detection systems for PD in power trans-
formers [78]. The common electrical detection approaches
used for condition monitoring of power transformers are
indirect measurement circuit with external coupling capac- FIGURE 4. (a) IEC 60270 based indirect measurement circuit using
itor and coupling capacitor through bushing taps, as shown external coupling capacitor. (b) Coupling capacitor through bushing
taps [79].
in Figs. 4(a) and 4(b) [79], [80]. In Fig. 4(a), the indirect mea-
surement test is performed, where the coupling capacitor (Ck ) as online testing is prone to electromagnetic interference but
is kept in a parallel configuration with the tested insulation is sufficient for offline testing (eg. routine tests of manufac-
system capacitance (CA ). The apparent charge is calculated tured products or pre-commissioning routine tests) [81], [82].
by the PD measurement device attached to the measuring Nevertheless, this method provides indications of the proper
impedance (ZM ). understanding of the insulation condition.
For online testing, this method is not used due to a bulky
high voltage coupling capacitor. But, if the bushings tap is B. ELECTROMAGNETIC DETECTION METHOD
available on the power transformer, as shown in Fig. 4(b), The electromagnetic (EM) method during the initial
online testing can be conducted. Currently, the development investigation showed that at a fixed PD location and
methods are not efficient enough to locate the PD activities constant EM signal propagation, the linear correlation

64590 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

between the PD charge and the possible signal sources can

be obtained [83], [84]. The EM method was first introduced
by W. R. Rutgers for the power transformer in 1997 [85].
Antennas such as conical, spiral, and Vivaldi are used as
sensors for ultra-high frequency (UHF) detection [26], [86].
UHF sensors are actively researched due to their advantages
such as resistance to low-frequency signals, the negligi-
ble effect of noise due to internal transformer construc-
tion by implementing denoising techniques and removing
white noise, and corona-free pulse disturbance [33], [87].
UHF detection is affected by radio interference and switching
events. The hindrance can be eliminated by careful denoising
techniques, which are illustrated in section V. The major
challenge for the implementation of UHF sensors is the cal-
ibration process since the measured amplitude is dependent FIGURE 5. Circuit diagram to examine the impact of different types of PD
on a variety of elements. on UHF calibration in power transformer [88].
In [88], the authors examined the effect of different
PD types on UHF calibration in a power transformer whose
circuit diagram is shown in Fig. 5. PD sources examined are
in the form of corona discharge, internal discharge, surface
discharge on polyethylene, surface discharge on pressboard,
and void discharge. Six drain valves help to install differ-
ent UHF probes at different locations. The authors in [88]
introduced the best possible detection frequency range to
measure the UHF signal concerning PD apparent charge.
Also, they illustrated the big challenge to reduce calibration FIGURE 6. Power transformer recording system for the detection of
error due to active transformer parts thereby, it was ineffi- AE signals from PD.
cient to use the UHF probe. However, the maximum charge
estimation method was proposed where the least feasible operation of the power transformer [95]. Broadband piezo-
ratio between the UHF quantifier parameter and IEC apparent electric transducers are common transduction elements for
charge is achieved in the measurement performed in the many ultrasonic systems. These are fixed on the transformer
laboratory [88]. tank through a magnetic holder for detecting AE signals. The
Different types of current transformers like Rogowski coil, AE signals further undergo amplification, filtration process
high-frequency current transformer (HFCT), and radiofre- and are finally fed and recorded by the AE analyzer.
quency current transformer (RFCT) have been extensively The AE method can be used for detecting multiple
explored as sensors for the PD detection in power trans- PD sources [93]. To overcome the incapability of detection
former [89]–[91]. The EM method can be used for localizing of PD level and calibration, the AE method is combined with
multiple PDs and identifying individual PD sources through other methods such as UHF, optical detection, and electrical
feature extraction and de-noising techniques. detection. The drawbacks of this method are the complex
behavior of acoustic emission, low intensity detected sig-
C. ACOUSTIC EMISSION (AE) METHOD
nals, and high price. AE detection devices include micro-
AE in the power transformer can be produced by a mechanical phones [96], piezoelectric transducers [93], accelerometer,
explosion due to the vaporization of oil near the vicinity of the fiber optic (FO) sensor [93], [97]. Among these AE detection
streamer, electrical arc, and mechanical vibration [92]. The devices, Fiber Optic sensors provide superior results due to
signals are in the form of pressure waves with unique charac- wide acoustic field detection and improved signal-to-noise
teristics for the distinct AE source and can be used to localize ratio (SNR). By denoising and optimization techniques, noise
the AE source by the detection of ultrasonic signals with a due to internal transformer design can be eliminated, and
frequency ranging from 40 kHz to hundreds of kHz [93], [94]. multi-PD sources can also be detected. However, the main
However, the acoustic PD signals can be affected by the problem of the AE technique is the helplessness of PD source
high-frequency signals which can be removed by denoising localization in the transformer winding due to rapid signal
techniques. The advantages of having no EM interference and attenuation by passing through different mediums [98].
at the same time having an economically friendly technique,
have driven the AE method as the most applicable technique D. GAS PRESENCE METHOD
for power transformers. Fig. 6 represents the block diagram of Dissolve gas analysis utilizes the concept of detection of
the power transformer recording system for the detection of gas es released due to the degradation of the transformer
AE signals from PD, which is implemented during the normal oil and cellulose insulation [99]. Degradation of transformer

VOLUME 9, 2021 64591

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

oil results in the liberation of gases in the form of hydrogen

(H2 ), methane (CH4 ), ethane (C2 H6 ), acetylene (C2 H2 ), and
ethylene (C2 H4 ). While cellulose insulation mainly releases
gases such as carbon monoxide (CO) and carbon dioxide
(CO2 ). The DGA is limited to offline testing in the form of gas
chromatography [100] and air circulation pump method [101]
since this method is time-consuming. The air circulation
pump method is a type of hydrogen detection system where
hydrogen is detected by the circulation effect of the air pump.
The online testing is developed are mainly in the form
of hydrogen detection [102], photoacoustic spectroscopy
FIGURE 7. A typical gas chromatograph [107].
(PAS) [103], fiber Bragg grating (FBG) sensor [102],
oil-immersed sensors [104], and membrane-based tech-
nique [105]. A typical gas chromatography system is shown
in Fig. 7 [106], [107]. The injection port vaporizes the oil the oil with the additional benefit of having strong dielectric
sample. The formed gases, fed into the column, contain property and immunity from electromagnetic interference.
lighter gases in the form of argon, helium, nitrogen, and The working principle of FBG is depicted in Fig. 8 [111].
hydrogen. Each gaseous component is filtered out by the FBG works as the narrowband reflective optical sensors,
column individually based on the retention time and comes where a single wavelength of light is reflected by grating, and
into contact with the heat detectors [107]. The detected PD the other wavelengths are transmitted.
signals are then recorded and plotted with the data acquisition
system creating chromatograms. The identification of the
gases is performed based on the concentration of the gas and
retention time. Generally, hydrogen gas detection is preferred
over other detections due to better accuracy. During over-
heating and discharges, if the hydrogen gas level rises above
the warning value, the internal insulations need to be diag-
nosed [108]. Several types of research have employed FBG
sensors inside the tank with the normal operating temperature
(60◦ C to 90◦ C) for detecting the hydrogen gas concentration.
This shows a promising future due to non-interferences from
other gases in addition to the high sensitivity of hydrogen FIGURE 8. Working of Fibre Bragg Grating Sensors [111].
gas detection at 80◦ C [109]. Also, enhanced FBG sensors
(Pd-capped Mg–Ti thin-film-based hydrogen sensor) have Initially, in the case of fluorescent optical sensors,
been developed which show sensitivity at a broader range PD detection from light emission was found possible for
of temperature (10◦ C to 80◦ C), having considerably higher air and not for transformer oil. In 2013, the optical method
sensitivity than the conventional FBG [110]. was found to be reliable for PD measurement in power
transformer through the fluorescent sensor with an uncom-
mon technique [112]. However, the research for PD detec-
E. OPTICAL METHOD tion in transformer oil through fluorescent sensors led to
The optical method can become a support tool for detect- questionable results with many flaws. The experiments have
ing the PD activities in power transformer oil. Researchers been continuously carried out to connect photon activity,
have carried out different methods for the PD analysis PD through optical signal, and PD charge constraints in the
through the light detection method for transformer oil/paper oil. In 2014, the measurement was possible for power trans-
insulation. Common PD optical detection sensors include former oil [112]. However, this was challenging especially
Mach–Zehnder interferometry (MZI), Fabry–Perot interfer- for old transformer oil [113]. The advantages of this method
ometer (EFPI), and fiber Bragg grating (FBG) [39]. MZI is are high-frequency response, immunity to electromagnetic
the early optical fiber-based sensor that uses a single-mode interference, measuring a broad range of chemical elements,
fiber and laser. The beam of light from the source is initially and physical parameters [31]. However, the major drawbacks
split into two fibers by a fiber coupler. The first sensing are: the detection of PD cannot be calibrated, no information
fiber optic coil is positioned to the zone of the PD signals on PD magnitude, and limitation to identify discharges within
in the oil tank and the second fiber is used as the reference transparent media. The PD detection and localization in trans-
for the optical route of the light. EFPI sensor is based on a former oil are still being researched currently. X-ray emission
single optical fiber using a silica diaphragm embedded in the from PD source is also being researched for detection testing
capsuled-shaped silica glass tube. Currently, FBG sensors are due to the advantage of bypassing the complex geometry of
used in power transformers as they can be directly kept inside the power transformer.

64592 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

FIGURE 9. Photo-acoustic spectroscopy-based DGA system [115].

TABLE 1. Advantages and Drawback of different Partial Discharge Detection Techniques in Power Transformer.

F. COMBINATIONAL METHOD AE sensors [117]. Innovative forms of AE detection methods

For the detection and localization of PD faults in power include AE sensors for locating the PD with a signal from EE
transformers, the combination of DGA and AE methods for the reference time which allows locating and verifying
has been used. For the existence of PD in transformers, that the signal detected is not the noise [118].
offline DGA has been implemented first, then to locate PD The combinations of AE, EE, and DGA have been used
sources, AE detection for 24 hours has been implemented for the PD identification method to find the overall insulation
for copying the daily load cycle [114]. Photo-Acoustic Spec- condition of the transformer [119]. By using EE and EM
troscopy (PAS) is an example of a combination of the DGA techniques, a noise rejection system could be created for the
and AE method. Fig. 9 illustrates the working principle of PD detection, where data from EE detection can be used to
PAS [115]. The infrared source provides kinetic energy to the detect the PD source. AE method can be enhanced to have
fault gases. Whereas the microphone detects the pressure sig- better sensitivity by combining with the EE method where
nal and transfers it into the electrical signals. Different gases EE detection is used for triggering.
detected are based on the intensity of sound waves produced, In AE and optical combination method, Fabry-Perot fiber
which are filtered through optical filters. The combination of (AE sensor) has been used for PD localization, and flu-
EM and acoustic methods have been used in which ultrasonic orescent optical fiber (optical sensor) is used as the con-
and UHF sensors were arranged in different shapes to provide firmation that the reference signal has been initiated from
worthy performance to detect PD sources for a particular the PD source [120]. Table 1 compares the advantages and
distance [116]. An example was proposed with the hybrid PD drawbacks of different PD detection techniques used in power
detecting system by transient earth wire voltage (TEV) and transformers.

VOLUME 9, 2021 64593

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

IV. PD DIAGNOSTICS IN POWER TRANSFORMER initial step of calibration and reflects only the effect of the
PD diagnosis of power transformers is a powerful tool to sensor. For the competency of the measurement, the UHF
classify various faults. The main purpose of PD diagnosis is antenna is connected to the transformer for measuring the
to identify the cause of PD in the insulation and distinguish calibration sensitivity. In [122], firstly, an identified UHF
the type of defects. Since power transformers have a highly calibration impulse is introduced without an antenna for cal-
complex insulation system with almost inaccessible inner ibrating cable and measurement devices. Then, the sensor
components, PD diagnosis is demanding and challenging. feature is added to the calibrated path by applying AF. Intro-
Online testing is performed and limited to the transformer ducing the identified transfer function through frequency-
tank and transformer terminals due to the compact structure. dependent, AF can provide variation of the calibration point
For achieving a proper diagnosis, advanced testing devices from calibrator to antenna in the transformer. The calibra-
along with experienced personnel is mandatory. tion process can be shortened by using scalar correction
As per IEC 60270 standards, the PD measurement by the factor AFs, showing PD frequencies with enough accuracy.
electrical method has excessive noise content due to sensi- Power transformer online monitoring with diagnostics has
tivity restrictions [121]. The UHF method has a high EM become a necessity now due to the majority of transform-
frequency range (300 MHz–3000 MHz) and the PD in trans- ers being installed more than four decades back [126]. Per-
former oil releases EM waves in the same range. UHF sensors forming continuous monitoring help to detect and resolve
installation can be performed inside the transformer by an sudden faults which minimize any future hazards. The data
oil filling valve during the state of operation, which provides generated through monitoring are very high and therefore
decent PD signal detection due to EM resistance from the need further analysis. One of the common analyses is phase-
surroundings by the transformer tank [122]. To record PD resolved PD pattern analysis based on pattern recognition.
activity by EE or EM method, localization of PD can be The characterization of PD patterns received from online
performed by the time of arrival of acoustic signals through monitoring is performed and compared with the recognized
piezo-electric sensors that are fastened on the wall of the pattern [33].
transformer tank [123]. The problem is that the acoustic
signals contain distortion due to the complex structure of V. PD DENOISING TECHNIQUES IN POWER
the transformer. It can be resolved by denoising and crafting TRANSFORMER
averaged signals where acoustic PD signals are overlapped, Partial discharge pulses are irregular, short-lived, and non-
and the noise is nullified by averaging. periodic. The obtained partial discharge signals extracted
The physical parameters are the main distinction between from the PD sensors contain excess discharge impulse, which
the EM and EE PD measurement method, where the former is challenging for the processing task. The obtained sig-
measure voltage (in mV) by sensing EM radiation through nals need to be disintegrated further by signal processing
UHF sensors and the latter measure apparent charge level techniques. Taking multiple PD sources generated at dif-
(in pC) by integrating the recharging current [124]. In factory ferent insulation into consideration, the signal processing
acceptance tests (FAT), the apparent charge (pC) is accept- techniques becomes handy. The process is conducted by
able considering the fact that the real PD value (pC/mV) is applying the time and frequency characteristics of obtained
undetermined s the measurement is not directly taken [125]. PD signals to create unique collections of time-frequency
UHF sensors can overcome the challenge of online monitor- maps. Each collection is allocated to a PD source. Various sig-
ing due to surrounding noise and the occurrence of corona nal denoising techniques are fast Fourier transform, low pass
discharges since the UHF antenna can measure PD incidence filtering, Wigner-Ville Distribution, short-time Fourier trans-
in transformer oil as the transformer tank behaves as the form, least mean squares (LMS) approach, frequency-domain
Faraday cage [84]. By this advantage, UHF sensors can be adaptive filtering (FDAF), recursive least squares (RLS),
implemented for offline as well as online routine tests. EE and and exponentially weighted recursive least squares (EWRLS)
EM measurements are predominantly affected by the type methods, matched filtering, notch filtering, wavelet denois-
and the level of PD source, signal diminution in the connect- ing, artificial neural network, empirical mode decomposition,
ing path, sensitivity of the sensors, and measurement device and blind equalization [127]–[130]. Some common denoising
sensitivity. techniques are explained as follows:
The effect of quadrupole or coupling capacitor in the sen-
sitivity of electrical measurement can be amended by the cal- A. FAST FOURIER TRANSFORM
ibration process. For this, the parameter antenna factor (AF) FFT is the algorithm developed for computing discrete
should be identified. AF depends on the design in terms of Fourier Transform (DFT) which is applied to the PD sig-
EM waves and can be estimated by an oil-filled gigahertz- nal to transform from the time domain to the frequency
transversal-electro-magnetic setup (GTEM) cell [122]. The domain [27]. FFT is effective for the slow varying signals
construction of the GTEM cell comprises of an elongated having stationary components. Since the property of PD sig-
coaxial cable in a cell, and with isolation from surrounding nals is rapidly changing, non-periodic, irregular, and transient
EM disturbance, a known EM field is introduced to the in nature, therefore, Wavelet Transform is preferred over this
equipment under test (EUT). GTEM cell is considered as the method [131].

64594 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

B. WAVELET TRANSFORM is centered on a mathematical operator that is implemented

Fourier transform decomposes any signal into sinusoidal between the measured signal, and the structured element
waveforms along with the frequency domain, but the time (flat, sinusoidal, or triangular element) [138]. The filtration
data is absent. Wavelet transform is used to decompose of the PD signal is done by the overlapping of the struc-
the signals and is a small waveform that has a very short tured element and the measured signal, where the filtered
period and with zero average magnitudes [132]. These are signal is denoted as a morphological feature [40]. In [136],
time-domain signals in the form of two-dimensional sets of the mathematical morphology was found to be noteworthy
coefficients and, therefore, confining them into the time and for denoising the PD signals.
frequency domain. They are employed for finding parameters
E. BLIND EQUALIZATION (BE)
such as the breakdown points and noise elimination [133].
The signal is fragmented into different wavelet coefficients Blind equalization is the type of de-noising technique where
into different frequency ranges. The wavelet transform can higher frequencies are eliminated, which are generally pro-
rebuild the PD signals when the wavelet function counterparts duced by communication systems and radio transmission at
the PD signals. By the thresholding, the wavelet coefficient substations [130]. Blind source separation (BSS) and BE
of PD signals is retained and the rest are eliminated [130]. are targeted to retrieve source signal without evaluating the
The two different ways to prepare the wavelet trans- source [139]. Also, there is a drawback that it requires more
form are continuous wavelet transform and discrete wavelet numbers of PD sensors than that of the number of PD sources.
transform [134]. The former acquires the surface of wavelet In [130], the author proposed an automatic BE technique
coefficients, while the latter has the discretized scale and for denoising PD signals in power transformers where tech-
translation. niques BSS and BE were used for recovering the source signal
without evaluating the source. The denoised PD signal vali-
C. ENSEMBLE EMPIRICAL MODE DECOMPOSITION dated the effectiveness of the PD extraction by suppressing
Ensemble empirical mode decomposition (EEMD) is a the high noise level.
method which is used to extract Intrinsic Mode Func- F. ARTIFICIAL NEURAL NETWORK
tion (IMF) from the signals and can be handy for non- The multilayer feed-forward neural network (MLPFNN) has
stationary and non-linear signal [129]. Hilbert Huang trans- become the center of attention to denoise the PD discharge
form (HHT) comprises two sections: Ensemble mode decom- signals [128]. Function approximation is applied due to the
position (EMD) and Hilbert spectral analysis. HHT is widely capability to discover the bond between the input and output
used in fault analysis but contains flaws in the EMD tech- data with the weights of these connected input and output
nique, where the problem arises due to the mode mixing data that are updated by the backpropagation algorithm. The
problem during the sifting process. EEMD is a noise-assisted Lavenberg-Marquardt algorithm that is used to manufacture
analysis technique that is more robust with better accu- perfectly denoised PD signals by changing the number of
racy [135], [136]. PD signal is disintegrated by eliminating nodes in the hidden layer is implemented in the hidden
the riding waves and smooth irregular amplitude that could layers for enhancing the performance of MLPFNN [140].
be rooted in the IMF. This is termed as a sifting process To improve the quality of the denoised signal, the number of
that limits the upper and lower measured signal, evaluates nodes in the ANN architecture is increased keeping in mind
the average values of the boundaries, and deducts them from the processing time.
the measured signal. In general, IMF is a single component Some research works have been performed to inves-
signal, but the IMF repeatedly clinches disparate frequency tigate the capability of ANN-based PD signal denoising
signals. This procedure is further reiterated on the residual techniques [128], [141]. In [128], the authors implemented
part of the measured signal until all the IMF is acquired. The the comparison of the performance of ANN with fast
signal can be rebuilt by adding all IMF. For improving the Fourier transform (FFT) and wavelet transform (WT) to con-
discrimination of noise and PD signal in EEMD, synthesized clude the ANN-based denoising technique to be superior
white noise is added to the measured signal as sometimes they to other techniques. In [141], the authors performed ANN
fail to distinguish them [137]. (curve fitting and function approximation) and WT (energy
In [136], the self-adaptive denoising techniques by using conservation-based method (ECBT)) denoising technique to
EEMD were proposed to eliminate the drawback of WT and remove white noise of radio frequency (RF) signals for
EMD. By utilizing the kurtosis-based selection criterion, the ceramic disc insulators having defects in the form of internal
EEMD method was found to be effective in reducing the noise voids, cracks, and sharp points. The result of ANN-based
to a great extent and retrieve PD impulses without conceding denoising was found to be better than ECBT.
the quality of PD impulses.
VI. FEATURE EXTRACTION OF POWER TRANSFORMER
D. MATHEMATICAL MORPHOLOGY Feature extraction is an important process for the
This technique necessitates earlier knowledge of the repet- analysis of the PD signals of the power transformer.
itive frequency of the signal, which cannot be availed in Common feature extraction techniques include Fourier
some applications. The theory of mathematical morphology transform, wavelet packet decomposition, Hilbert–Huang

VOLUME 9, 2021 64595

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

transforms (HHT), stochastic neighbor embedding (SNE), actual data can be achieved through a scree plot, which is the
principle component analysis (PCA), kernel principal compo- graph of eigenvalue magnitude vs. its number [145]. In [91],
nent analysis (KPCA), support vector machine (FSVM), and PCA was implemented for the autonomous localization of the
artificial neural networks (ANNs) [51], [142]. This section PD source within transformer winding showing the capability
is focused on the statistical overview of feature extraction to locate the PD source.
in power transformers. Fig. 10 represents the flowchart of Machine learning techniques (Artificial neural networks)
the partial discharge monitoring system. The monitoring have currently shown decent efficiency for PD detection and
system comprises of three sections, namely PD signal collec- recognition [151]–[153]. In [154], the authors used four types
tion, feature extraction for PD signal, and PD data analysis. of artificial PD defects (floating, surface, rod-plane, and air
After applying the denoising techniques and localization of gap discharge) for PD identification, which is similar to the
PD signals, the filtered data can be represented by two intelligent framework for power transformer assessment as
different patterns, namely PRPD and TRPD. While analyzing proposed in [155]. A deep learning approach called sparse
PD, very high dimensional data are quite common, which auto-encoder (SAE) was used for feature extraction. The deep
require dimension reduction techniques. PRPD characteriza- learning method of SAE and Softmax exhibited promising
tion can be classified into two different groups; the number of results of more than 96% accuracy.
PD pulse vs. phase angle and amplitude of charge vs. phase
angle which can be categorized further in positive and nega- VII. PD CLASSIFICATION IN POWER TRANSFORMER
tive half-cycles [143]. Statistical features that can be obtained PD classification in the transformer is a valuable supervised
from these distributions are namely skewness, mean, vari- learning technique since it can organize various kinds of
ance, kurtosis, and Weibull [144]. The advantage of statistical defects into their respective category for condition assess-
features is decreased computation time. In [145], the author ment. These appropriate classifiers are necessary as any inde-
included a statistical feature analysis for the extraction of cision may result in the wrong classification of the PD model.
PD signals in transformer insulation defects. Kurtosis and Also, the precision of the PD classification depends on the
skewness show the sharpness and symmetry of the distri- features extracted through the PD pattern. Recent artificial
bution respectively. Weibull distribution portrays the pulse intelligence techniques have shown decent PD classification
height analysis pattern where the PD pulse rate can be shown of PD defects in power transformers.
in the probability distribution curve. The features of Weibull ANN can be the proper classification for PD pattern anal-
distribution in addition to other parameters after analysis are ysis due to the insensitivity to the minor input deviation
served to the intelligent classifiers. and making appropriate decisions during the training pro-
Fractal features can be applied to model PD and pattern cess while feeding the data that is quite similar to the input
recognition due to the ability to model complicated struc- data [156]. ANN is the supervised learning process where
tures and natural phenomena where presently, the mathe- training employs forward and backward processes, and ini-
matical means are inadequate [146]. In [147], the author tialization is performed through the weights and biases hav-
performed fractal-based feature extraction for the identifi- ing small value [157]. The activation function provides the
cation of PD patterns for high voltage power transformers. computation of the feature vector in their respective output
PRPD patterns can be processed through box-counting tech- layers. The layers in between the input and output layers
niques that use two fractal features; fractal dimension (com- that are connected and cascaded together to form a network
putation for image surface) and lacunarity (fractal surface are termed hidden layers. Hidden layers attribute to acquire
compactness) [147]. Even though fractal dimension does not the PD features from various sources and send the extracted
affect due to variation in scale and promising measurement sources to the output. The classification is done based on
of the surface coarseness, the ineffectiveness to distinguish the types of PD defects to be classified. In [158], the author
feature of the same value of fractal surface resulted in the proposed a convolutional Neural network (CNN) architecture
development of another variable termed lacunarity [148]. The for UHF signal PD pattern source recognition which is shown
research was performed based on the removal data contam- in Fig. 11. The input to CNN is 1 × 128 × 256 generated
inated by noise in which PRPD patterns were transformed by Short Time Fourier Transform (STFT). The first three
into a binary image, and fractal features calculations were hidden layers comprise filters, pooling layers, and dropout
performed through ImageJ [143]. Principle component anal- layers. The final two hidden layers are fully connected with
ysis (PCA) is used to filter essential data from big data 500 hidden units each, and the output layer is a fully con-
groups thereby serves as the data reduction techniques [149]. nected linear layer. By the proposed architecture, the data
PCA, also termed as Karhunen-Loève (K-L) method does accumulated from gas-insulated switchgear showed decent
not negotiate with the data information, and with very low accuracy and further help in developing more ideas for PD
depletion of information, data can be shrunken to compact UHF signal recognition for power transformers.
space. Space reduction is attained through data projection Adaptive Neuro-Fuzzy Inference System (ANFIS) is the
in the broadest variance at the lesser dimension that will hybrid system of neural network and fuzzy system that uses a
boost the scatter of the desired samples [150]. The number neural network to eliminate the necessity of finding a suitable
of principal components for attaining the precise value of fuzzy network for operation [159]. Takagi and coworkers

64596 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

FIGURE 10. Flowchart for partial discharge monitoring system.

FIGURE 11. Convolutional Neural Network- based PD classification [158].

created ANFIS that is based on the fuzzy Sugeno model, The decision tree method comprises a type of flowchart
which is a proficient tool for classifying PD patterns by assembly in which the internal node is used for the fea-
implementing If-Then rules through a decision tree andspec- ture testing, the leaf node shows the class label, and the
ified input-output data [160]. For improving the efficiency route between the root and the leaf shows the classification
of the training, input variables are regularized between zero rule [182]. This method has been used extensively in the
and one. In [161], the authors implemented a fuzzy and PD classifications under different PD conditions due to the
ANFIS model for PD fault detection using DGA for power advantage of the visible rule for PD classification, unlike
transformer where the ANFIS model was found to be superior SVM or ANN. A decision tree had been employed to identify
with an accuracy of 98%. the void size and differentiating the multiple PD sources
Support vector machine (SVM) is derived from statistical in power transformers [183]. K-nearest neighbor (KNN) is
learning, where the regulation is managed for several tasks by a simple and non-parametric algorithm that classifies the
applying base algorithms and kernel functions [162]. In this training sets by recognizing the collection of k objects nearest
method, PD pattern data can be portrayed by vector dimen- to test objects and allotting the type through correlation of
sion depending on the number of input features and perform the respective class in the neighborhood [184]. The main
decently in the complications related to non-linearity, lesser elements of KNN are labeled objects, constant ‘k’, and the
sample magnitude, and large dimensions [163]. Kernel’s quantity of nearest neighbors. The classification of KNN is
method is the additional tool to overcome the inefficiency to centered on fresh data points according to greater votes for
analyze non-linear problems. the neighboring data points.
The authors in [164], performed the classification of PD
pattern based on SVM for the floating metallic particles in
the transformer oil showing promising results irrespective VIII. PD CLUSTERING IN POWER TRANSFORMER
of big data and complexity. In [165], the authors per- PD signal clustering is the unsupervised learning technique
formed different PD classifications for the AE-based sig- where the data are organized into clusters such that each
nals for transformer insulation where SVM performance was cluster element is closely associated with the other. The clus-
found to be superior to the decision tree and K-nearest tering technique is extensively used in PRPD and TRPD for
neighbor. distinguishing the characteristics of PD pulse in the multiple

VOLUME 9, 2021 64597

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

TABLE 2. Recent timeline of partial discharge analysis in power transformer.

64598 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

TABLE 2. (Continued.) Recent timeline of partial discharge analysis in power transformer.

VOLUME 9, 2021 64599

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

TABLE 2. (Continued.) Recent timeline of partial discharge analysis in power transformer.

PD sources and arranging in groups. Table 2 shows a recent clustering, objects are initially taken as separate clusters.
timeline of partial discharge analysis in power transformers. Further, according to the distance between the two objects,
K-means (KM) is an efficient and uncomplicated centroid- the individual clusters are merged, and the procedure is fol-
based clustering algorithm where parameter ‘K’ represents lowed until the conditions are fulfilled. Divisive clustering
the pre-defined number of clusters chosen for iteration. has initial single clusters where all objects are allocated and
K-means clustering is applied until the convergence between further separated into different clusters according to the con-
the assignment step, and the update step is achieved [185]. dition [3]. This method can be productive for examining big
However, this method has limitations in the form of local structures but slow in processing. Also, it lags modification
minimum convergence and pre-assigned value of K, which after developing the splitting/merging decision.
may be challenging due to the lack of information regarding
the numbers of different PD sources [186]. Another version
IX. CONCLUSION
of K-means is Fuzzy C-means (FCM) in which in each cluster
This paper enumerates the comprehensive survey of mod-
every object is assigned a fuzzy degree [187], [188]. Soft
ern techniques for PD signal analysis of power transform-
clustering is done for each object where each object can
ers. PD detection, localization, and severity of fault can be
be allocated in different clusters while optimizing. In [189],
analyzed through feature representation, classification, and
the author proposed K-means with SVM clustering-based
clustering techniques, which are extensively reviewed. Dif-
techniques for DGA in the power transformer for improving
ferent methodologies for denoising the PD signals have been
accuracy. The result found was better for KMSVM relative to
introduced. PD detection in the power transformer is essential
SVM and k-mean clustering with a reduction in the training
since the power system network depends completely on con-
set and training time.
tinuous operation. The paper explains the overview of partial
The density-based spatial clustering of applications with
discharge with different types of defects in power transform-
noise (DBSCAN) is the clustering algorithm prepared by
ers. Different types of PD detection techniques (electrical
Martin Ester and coworkers in 1996 [190]. Unlike K-means,
and non-electrical) have been explained, and the advantages
DBSCAN does not require allocation of the number of clus-
and disadvantages of each technique have been elaborated.
ters but assigns the data which are dense and closely related in
The importance of PD diagnostics for identifying the type of
clusters. However, DBSCAN fails in the proper clustering of
power transformer PD defects has been elucidated. The PD
similar density data and high dimensionality data. DBSCAN
monitoring system consists of different steps for analyzing
works on the two parameters: the number of minimum points
PD defects. This includes PD detection, denoising, feature
in the neighborhood of point p (Minpts), and the radius
extraction, classification, and clustering. Each step has been
of neighborhood p (Eps). In [191], the authors performed
elaborated, including the insight of the modern methods. The
automatic pulse grouping through DBSCAN for PD source
online PD measurements in a power transformer are profi-
separation in power transformer where the effectiveness of
cient means for PD analysis with the challenge of the onsite
the DBSCAN algorithm over conventional means is evident.
noise and due to the complex structure of the power trans-
The PD sources are notable by the DBSCAN algorithm,
former. PD sensing techniques can be advanced to reduce
whereas conventional cannot completely separate the PD
the impact of white noise in online sensing and to locate
sources. Furthermore, by the PRPD diagram, the different
PD activities in the power transformer.
sources are noticeably recognized.
Hierarchical cluster analysis is the clustering algorithm
where the clusters are generated in the order of domi- ACKNOWLEDGMENT
nancy from top to bottom (divisive) or bottom to the top The statements made herein are solely the responsibility of
way (agglomerative) [149]. In the agglomerative hierarchical the authors.
64600 VOLUME 9, 2021
M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

REFERENCES [23] C. Yao, Z. Zhao, Y. Mi, C. Li, Y. Liao, and G. Qian, ‘‘Improved online
[1] A. Akbari, P. Werle, H. Borsi, and E. Gockenbach, ‘‘Transfer function- monitoring method for transformer winding deformations based on the
based partial discharge localization in power transformers: A feasibility lissajous graphical analysis of voltage and current,’’ IEEE Trans. Power
study,’’ IEEE Elect. Insul. Mag., vol. 18, no. 5, pp. 22–32, Sep. 2002. Del., vol. 30, no. 4, pp. 1965–1973, Aug. 2015.
[2] M. Meira, C. R. Ruschetti, R. E. Álvarez, and C. J. Verucchi, ‘‘Power [24] T. K. Saha, ‘‘Review of modern diagnostic techniques for assessing insu-
transformers monitoring based on electrical measurements: State of the lation condition in aged transformers,’’ IEEE Trans. Dielectrics Electr.
art,’’ IET Gener., Transmiss. Distrib., vol. 12, no. 12, pp. 2805–2815, Insul., vol. 10, no. 5, pp. 903–917, Oct. 2003.
Jul. 2018. [25] R.-J. Liao, L.-J. Yang, J. Li, and S. Grzybowski, ‘‘Aging condition
[3] A. R. Abbasi, M. R. Mahmoudi, and Z. Avazzadeh, ‘‘Diagnosis and assessment of transformer oil-paper insulation model based on partial
clustering of power transformer winding fault types by cross-correlation discharge analysis,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 18, no. 1,
and clustering analysis of FRA results,’’ IET Gener., Transmiss. Distrib., pp. 303–311, Feb. 2011.
vol. 12, no. 9, pp. 4301–4309, 2018. [26] A. Darwish, S. S. Refaat, H. A. Toliyat, and H. Abu-Rub, ‘‘On the
[4] S. Saponara, L. Fanucci, F. Bernardo, and A. Falciani, ‘‘Predictive diagno- electromagnetic wave behavior due to partial discharge in gas insulated
sis of high-power transformer faults by networking vibration measuring switchgears: State-of-art review,’’ IEEE Access, vol. 7, pp. 75822–75836,
nodes with integrated signal processing,’’ IEEE Trans. Instrum. Meas., 2019.
vol. 65, no. 8, pp. 1749–1760, Aug. 2016. [27] J. Fuhr and T. Aschwanden, ‘‘Identification and localization of PD-
[5] R. P. Medeiros, F. B. Costa, and K. M. Silva, ‘‘Power transformer differ- sources in power-transformers and power-generators,’’ IEEE Trans.
ential protection using the boundary discrete wavelet transform,’’ IEEE Dielectrics Electr. Insul., vol. 24, no. 1, pp. 17–30, Feb. 2017.
Trans. Power Del., vol. 31, no. 5, pp. 2083–2095, Oct. 2016. [28] A. Rodrigo Mor, L. C. Castro Heredia, D. A. Harmsen, and
[6] S. M. Bhatt, D. Kumar, and K. Patel, ‘‘Partial discharge analysis in time F. A. Muñoz, ‘‘A new design of a test platform for testing multiple
and time-frequency domain of solid dielectric in power transformer,’’ in partial discharge sources,’’ Int. J. Electr. Power Energy Syst., vol. 94,
Proc. 5th IEEE Uttar Pradesh Sect. Int. Conf. Electr., Electron. Comput. pp. 374–384, Jan. 2018.
Eng. (UPCON), Nov. 2018, pp. 1–5. [29] J. Rubio-serrano, M. V. Rojas-moreno, J. Posada, J. M. Martínez-tarifa,
[7] A. Santosh Kumar, R. P. Gupta, K. Udayakumar, and A. Venkatasami, G. Robles, and J. A. Garcia-souto, ‘‘Electro-acoustic detection, iden-
‘‘Online partial discharge detection and location techniques for condition tification and location of partial discharge sources in oil-paper insula-
monitoring of power transformers: A review,’’ in Proc. Int. Conf. Condi- tion systems,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 19, no. 5,
tion Monitor. Diagnosis, 2008, pp. 927–931. pp. 1569–1578, Oct. 2012.
[8] A. J. M. Cardoso and L. M. R. Oliveira, ‘‘Condition monitoring and [30] R. Braunlich, M. Hassig, J. Fuhr, and T. Aschwanden, ‘‘Assessment of
diagnostics of power transformers,’’ Int. J. COMADEM, vol. 2, pp. 5–11, insulation condition of large power transformers by on-site electrical
Jul. 1999. diagnostic methods,’’ in Proc. Conf. Rec. IEEE Int. Symp. Electr. Insul.,
[9] S. Li and J. Li, ‘‘Condition monitoring and diagnosis of power equip- Apr. 2000, pp. 368–372.
ment: Review and prospective,’’ High Voltage, vol. 2, no. 2, pp. 82–91, [31] M. M. Yaacob, M. A. Alsaedi, J. R. Rashed, A. M. Dakhil, and S. F. Atyah,
Jun. 2017. ‘‘Review on partial discharge detection techniques related to high voltage
[10] M. Bigdeli, M. Vakilian, and E. Rahimpour, ‘‘Transformer winding power equipment using different sensors,’’ Photonic Sensors, vol. 4, no. 4,
faults classification based on transfer function analysis by support vector pp. 325–337, Dec. 2014.
machine,’’ IET Electr. Power Appl., vol. 6, no. 5, pp. 268–276, May 2012. [32] Q. Khan, S. S. Refaat, H. Abu-Rub, and H. A. Toliyat, ‘‘Partial discharge
[11] N. Hashemnia, A. Abu-Siada, and S. Islam, ‘‘Detection of power trans- detection and diagnosis in gas insulated switchgear: State of the art,’’
former bushing faults and oil degradation using frequency response anal- IEEE Elect. Insul. Mag., vol. 35, no. 4, pp. 16–33, Jul. 2019.
ysis,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 23, no. 1, pp. 222–229, [33] H. H. Sinaga, B. T. Phung, and T. R. Blackburn, ‘‘Partial dis-
Feb. 2016. charge localization in transformers using UHF detection method,’’
[12] Q. Fu, J. Zhu, Z.-H. Mao, G. Zhang, and T. Chen, ‘‘Online condition IEEE Trans. Dielectrics Electr. Insul., vol. 19, no. 6, pp. 1891–1900,
monitoring of onboard traction transformer core based on core-loss calcu- Dec. 2012.
lation model,’’ IEEE Trans. Ind. Electron., vol. 65, no. 4, pp. 3499–3508, [34] M.-K. Chen, J.-M. Chen, and C.-Y. Cheng, ‘‘Partial discharge detection
Apr. 2018. by RF coil in 161 kV power transformer,’’ IEEE Trans. Dielectrics Electr.
[13] C. Aj, M. A. Salam, Q. M. Rahman, F. Wen, S. P. Ang, and W. Voon, Insul., vol. 21, no. 3, pp. 1405–1414, Jun. 2014.
‘‘Causes of transformer failures and diagnostic methods—A review,’’ [35] H. Jahangir, A. Akbari, P. Werle, and J. Szczechowski, ‘‘UHF PD
Renew. Sustain. Energy Rev., vol. 82, pp. 1442–1456, Feb. 2018. measurements on power transformers-advantages and limitations,’’
[14] Y. Han and Y. H. Song, ‘‘Condition monitoring techniques for electrical IEEE Trans. Dielectrics Electr. Insul., vol. 24, no. 6, pp. 3933–3940,
equipment-a literature survey,’’ IEEE Trans. Power Del., vol. 18, no. 1, Dec. 2017.
pp. 4–13, Jan. 2003. [36] M. Ariannik, M. A. Azirani, P. Werle, and A. A. Azirani, ‘‘UHF measure-
[15] S. Tenbohlen et al., ‘‘Development and results of a worldwide transformer ment in power transformers: An algorithm to optimize accuracy of arrival
reliability survey,’’ in Proc. CIGRE SC A2 Colloq., Shanghai, China, time detection and PD localization,’’ IEEE Trans. Power Del., vol. 34,
Sep. 2015. no. 4, pp. 1530–1539, Aug. 2019.
[16] J. Faiz and M. Soleimani, ‘‘Dissolved gas analysis evaluation in electric [37] C. P. Beura, M. Beltle, and S. Tenbohlen, ‘‘Positioning of UHF
power transformers using conventional methods a review,’’ IEEE Trans. PD sensors on power transformers based on the attenuation of UHF
Dielectrics Electr. Insul., vol. 24, no. 2, pp. 1239–1248, Apr. 2017. signals,’’ IEEE Trans. Power Del., vol. 34, no. 4, pp. 1520–1529,
[17] M. Duval, ‘‘A review of faults detectable by gas-in-oil analysis in trans- Aug. 2019.
formers,’’ IEEE Elect. Insul. Mag., vol. 18, no. 3, pp. 8–17, May 2002. [38] S. Biswas, C. Koley, B. Chatterjee, and S. Chakravorti, ‘‘A methodol-
[18] Y. Kawaguchi and S. Yanabu, ‘‘Partial-discharge measurement on high- ogy for identification and localization of partial discharge sources using
voltage power transformers,’’ IEEE Trans. Power App. Syst., vol. PAS-88, optical sensors,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 19, no. 1,
no. 8, pp. 1187–1194, Aug. 1969. pp. 18–28, Feb. 2012.
[19] S. Zhang, ‘‘Analysis of some measurement issues in bushing power factor [39] A. Razzaq, H. Zainuddin, F. Hanaffi, and R. M. Chyad, ‘‘Transformer oil
tests in the field,’’ IEEE Trans. Power Del., vol. 21, no. 3, pp. 1350–1356, diagnostic by using an optical fibre system: A review,’’ IET Sci., Meas.
Jul. 2006. Technol., vol. 13, no. 5, pp. 615–621, Jul. 2019.
[20] A. Secic, M. Krpan, and I. Kuzle, ‘‘Vibro-acoustic methods in the condi- [40] A. Baug, N. R. Choudhury, R. Ghosh, S. Dalai, and B. Chatterjee, ‘‘Identi-
tion assessment of power transformers: A survey,’’ IEEE Access, vol. 7, fication of single and multiple partial discharge sources by optical method
pp. 83915–83931, 2019. using mathematical morphology aided sparse representation classifier,’’
[21] M. Wang, A. J. Vandermaar, and K. D. Srivastava, ‘‘Review of condition IEEE Trans. Dielectrics Electr. Insul., vol. 24, no. 6, pp. 3703–3712,
assessment of power transformers in service,’’ IEEE Elect. Insul. Mag., Dec. 2017.
vol. 18, no. 6, pp. 12–25, Nov. 2002. [41] S. S. Refaat, M. Sayed, M. A. Shams, and A. Mohamed, ‘‘A review
[22] M. O. Oliveira, S. Member, and A. S. Bretas, ‘‘Application of discrete of partial discharge detection techniques in power transformers,’’ in
wavelet transform for differential protection of power transformers,’’ in Proc. 20th Int. Middle East Power Syst. Conf. (MEPCON), Dec. 2018,
Proc. IEEE Bucharest PowerTech, Jul. 2009, pp. 1–8. pp. 1020–1025.

VOLUME 9, 2021 64601

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

[42] R. A. Hooshmand, M. Parastegari, and M. Yazdanpanah, ‘‘Simultaneous [63] K. D. Bergeron, ‘‘Theory of the secondary electron avalanche at electri-
location of two partial discharge sources in power transformers based cally stressed insulator-vacuum interfaces,’’ J. Appl. Phys., vol. 48, no. 7,
on acoustic emission using the modified binary partial swarm optimi- pp. 3073–3080, 1977.
sation algorithm,’’ IET Sci., Meas. Technol., vol. 7, no. 2, pp. 119–127, [64] M. Mondal and G. B. Kumbhar, ‘‘Detection, measurement, and classifi-
Mar. 2013. cation of partial discharge in a power transformer: Methods, trends, and
[43] Y. Wang, Y. Fan, S. Qin, D. Chang, X. Shao, H. Mu, and G. Zhang, ‘‘Par- future research,’’ IETE Tech. Rev., vol. 35, no. 5, pp. 483–493, Sep. 2018.
tial discharge localisation methodology for power transformers based on [65] M. D. Judd, L. Yang, and I. B. B. Hunter, ‘‘Partial discharge monitoring
improved acoustic propagation route search algorithm,’’ IET Sci., Meas. for power transformers using UHF sensors Part 2: Field experience,’’
Technol., vol. 12, no. 8, pp. 1023–1030, Nov. 2018. IEEE Electr. Insul. Mag., vol. 21, no. 3, pp. 5–13, May/Jun. 2005.
[44] T. B. Shanker, H. N. Nagamani, D. Antony, and G. S. Punekar, ‘‘Effects of [66] S. Kanakambaran, R. Sarathi, and B. Srinivasan, ‘‘Identification and
transformer-oil temperature on amplitude and peak frequency of partial localization of partial discharge in transformer insulation adopting
discharge acoustic signals,’’ IEEE Trans. Power Del., vol. 33, no. 6, cross recurrence plot analysis of acoustic signals detected using fiber
pp. 3227–3229, Dec. 2018. Bragg gratings,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 24, no. 3,
[45] Y.-B. Wang, D.-G. Chang, Y.-H. Fan, G.-J. Zhang, J.-Y. Zhan, pp. 1773–1780, Jun. 2017.
X.-J. Shao, and W.-L. He, ‘‘Acoustic localization of partial discharge [67] B. A. de Castro, D. de Melo Brunini, F. G. Baptista, A. L. Andreoli, and J.
sources in power transformers using a particle-swarm-optimization- A. C. Ulson, ‘‘Assessment of macro fiber composite sensors for measure-
route-searching algorithm,’’ IEEE Trans. Dielectrics Electr. Insul., ment of acoustic partial discharge signals in power transformers,’’ IEEE
vol. 24, no. 6, pp. 3647–3656, Dec. 2017. Sensors J., vol. 17, no. 18, pp. 6090–6099, Sep. 2017.
[46] J. Dai, H. Song, G. Sheng, and X. Jiang, ‘‘Dissolved gas analysis of insu- [68] X. Peng, F. Yang, G. Wang, Y. Wu, L. Li, Z. Li, A. A. Bhatti, C. Zhou,
lating oil for power transformer fault diagnosis with deep belief network,’’ D. M. Hepburn, A. J. Reid, M. D. Judd, and W. H. Siew, ‘‘A convolu-
IEEE Trans. Dielectrics Electr. Insul., vol. 24, no. 5, pp. 2828–2835, tional neural network-based deep learning methodology for recognition of
Oct. 2017. partial discharge patterns from high-voltage cables,’’ IEEE Trans. Power
[47] E. A. Mackenzie, J. Crossey, A. Depablo, and W. Ferguson, ‘‘On-line Del., vol. 34, no. 4, pp. 1460–1469, Aug. 2019.
monitoring and diagnostics for power transformers,’’ in Proc. IEEE Int. [69] L. Hao, P. L. Lewin, and S. J. Dodd, ‘‘Comparison of support vector
Symp. Elect. Insul., Jun. 2010, pp. 1–5. machine based partial discharge identification parameters,’’ in Proc. Conf.
[48] J. Faiz and M. Soleimani, ‘‘Assessment of computational intelligence and Rec. IEEE Int. Symp. Electr. Insul., Jun. 2007, pp. 110–113.
conventional dissolved gas analysis methods for transformer fault diagno- [70] T. Wang, Y. He, T. Shi, and B. Li, ‘‘Transformer incipient hybrid fault
sis,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 25, no. 5, pp. 1798–1806, diagnosis based on solar-powered RFID sensor and optimized DBN
Oct. 2018. approach,’’ IEEE Access, vol. 7, pp. 74103–74110, 2019.
[49] X. Wang, B. Li, H. T. Roman, O. L. Russo, K. Chin, and K. R. Farmer,
[71] K. Firuzi, M. Vakilian, V. P. Darabad, B. T. Phung, and T. R. Blackburn,
‘‘Acousto-optical PD detection for transformers,’’ IEEE Trans. Power
‘‘A novel method for differentiating and clustering multiple partial dis-
Del., vol. 21, no. 3, pp. 1068–1073, Jul. 2006.
charge sources using s transform and bag of words feature,’’ IEEE Trans.
[50] W. Si, C. Fu, and P. Yuan, ‘‘An integrated sensor with AE and UHF Dielectrics Electr. Insul., vol. 24, no. 6, pp. 3694–3702, Dec. 2017.
methods for partial discharges detection in transformers based on oil
[72] P.-H. Chen, H.-C. Chen, A. Liu, and L.-M. Chen, ‘‘Pattern recognition
valve,’’ IEEE Sensors Lett., vol. 3, no. 10, pp. 2019–2021, Oct. 2019.
for partial discharge diagnosis of power transformer,’’ in Proc. Int. Conf.
[51] W. J. K. Raymond, H. A. Illias, A. H. A. Bakar, and H. Mokhlis, ‘‘Par-
Mach. Learn. Cybern., vol. 6, Jul. 2010, pp. 2996–3001.
tial discharge classifications: Review of recent progress,’’ Measurement,
[73] S. Bamji, A. Bulinski, and M. Abou-Dakka, ‘‘Luminescence and space
vol. 68, pp. 164–181, May 2015.
charge in polymeric dielectrics–[whitehead memorial lecture (2008),’’
[52] L. Satish and B. I. Gururaj, ‘‘Partial discharge pattern classification using
IEEE Trans. Dielectrics Electr. Insul., vol. 16, no. 5, pp. 1376–1392,
multilayer neural networks,’’ IEE Proc. A Sci., Meas. Technol., vol. 140,
Oct. 2009.
no. 4, p. 323, 1993.
[74] G.-M. Ma, H.-Y. Zhou, C. Shi, Y.-B. Li, Q. Zhang, C.-R. Li, and Q. Zheng,
[53] B. Karthikeyan, S. Gopal, and S. Venkatesh, ‘‘ART 2—An unsupervised
‘‘Distributed partial discharge detection in a power transformer based
neural network for PD pattern recognition and classification,’’ Expert
on phase-shifted FBG,’’ IEEE Sensors J., vol. 18, no. 7, pp. 2788–2795,
Syst. Appl., vol. 31, no. 2, pp. 345–350, Aug. 2006.
Apr. 2018.
[54] X. Ma, C. Zhou, and I. J. Kemp, ‘‘Interpretation of wavelet analysis and
its application in partial discharge detection,’’ IEEE Trans. Dielectrics [75] J. Li, T. Jiang, C. Cheng, and C. Wang, ‘‘Hilbert fractal antenna for UHF
Electr. Insul., vol. 9, no. 3, pp. 446–457, Jun. 2002. detection of partial discharges in transformers,’’ IEEE Trans. Dielectrics
Electr. Insul., vol. 20, no. 6, pp. 2017–2025, Dec. 2013.
[55] H. Zhang, T. R. Blackburn, B. T. Phung, and D. Sen, ‘‘A novel wavelet
transform technique for on-line partial discharge measurements. 1. WT [76] R. Giussani, I. Cotton, and R. Sloan, ‘‘Comparison of IEC 60270 and RF
de-noising algorithm,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 14, partial discharge detection in an electromagnetic noise-free environment
no. 1, pp. 3–14, Feb. 2007. at differing pressures,’’ in Proc. IEEE Int. Symp. Elect. Insul., Jun. 2012,
[56] S. Venkatesh and S. Gopal, ‘‘Robust heteroscedastic probabilistic neu- pp. 127–131.
ral network for multiple source partial discharge pattern recognition— [77] IEEE Recommended Practice for Partial Discharge Measurement in
Significance of outliers on classification capability,’’ Expert Syst. Appl., Liquid-Filled Power Transformers and Shunt Reactors, IEEE Stan-
vol. 38, no. 9, pp. 11501–11514, Sep. 2011. dard C57.113-2010 (Revision IEEE Standard C57.113-1991), 2010,
[57] R. J. Van Brunt, ‘‘Stochastic properties of partial-discharge phenomena,’’ pp. 1–47.
IEEE Trans. Electr. Insul., vol. 26, no. 5, pp. 902–948, Oct. 1991. [78] M. D. Judd, ‘‘Experience with UHF partial discharge detection and
[58] M. G. Niasar, ‘‘Partial discharge signatures of defects in insulation sys- location in power transformers,’’ in Proc. Electr. Insul. Conf. (EIC).,
tems consisting of oil and oil-impregnated paper,’’ Licentiate thesis, KTH, Jun. 2011, pp. 201–205.
Stockholm, Sweden, 2012. [79] High Voltage Test Techniques-Partial Discharge Measurements, IEC Int.
[59] K. I. Lundgaard, J. Poittevin, J. Schmidt, D. Allen, T. R. Blackburn, Standard 60270, IEC, Geneva, Switzerland, 3rd ed., 2000.
H. Borsi, N. Foulon, J. Fuhr, N. Hosokawa, and R. E. James, ‘‘Partial [80] J. Fuhr, ‘‘Procedure for identification and localization of dangerous PD
discharges in transformer insulation,’’ in Proc. CIGRE Task Force, vol. 15, sources in power transformers,’’ IEEE Trans. Dielectrics Electr. Insul.,
2000, pp. 15–302. vol. 12, no. 5, pp. 1005–1014, Oct. 2005.
[60] G. C. Stone and H. G. Sedding, ‘‘In-service evaluation of motor and [81] S. Zheng, C. Li, Z. Tang, W. Chang, and M. He, ‘‘Location of PDs inside
generator stator windings using partial discharge tests,’’ IEEE Trans. Ind. transformer windings using UHF methods,’’ IEEE Trans. Dielectrics
Appl., vol. 31, no. 2, pp. 299–303, Mar./Apr. 1995. Electr. Insul., vol. 21, no. 1, pp. 386–393, Feb. 2014.
[61] A. C. Ashcraft, R. M. Eichhorn, and R. G. Shaw, ‘‘Laboratory studies of [82] Q. Xie, S. Cheng, F. Lü, and Y. Li, ‘‘Location of partial discharge in
treeing in solid dielectrics and voltage stabilization of polyethylene,’’ in transformer oil using circular array of ultrasonic sensors,’’ IEEE Trans.
Proc. IEEE Int. Conf. Electr. Insul., Jun. 1976, pp. 213–218. Dielectrics Electr. Insul., vol. 20, no. 5, pp. 1683–1690, Oct. 2013.
[62] M. Kawai and D. Milone, ‘‘Flashover test at project UHV on salt- [83] A. Troeger, U. Riechert, S. Burow, and S. Tenbohlen, ‘‘Sensitivity eval-
contaminated insulators,’’ IEEE Trans. Power App. Syst., vol. PAS-89, uation of different types of PD sensors for UHF-PD-measurements,’’ in
no. 5, pp. 755–761, May 1970. Proc. Int. Conf. Condition Monit. Diagnostics, Sep. 2010, pp. 2–5.

64602 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

[84] S. Coenen, S. Tenbohlen, S. Markalous, and T. Strehl, ‘‘Sensitivity of [106] N. Bakar, A. Abu-Siada, and S. Islam, ‘‘A review of dissolved gas analysis
UHF PD measurements in power transformers,’’ IEEE Trans. Dielectrics measurement and interpretation techniques,’’ IEEE Elect. Insul. Mag.,
Electr. Insul., vol. 15, no. 6, pp. 1553–1558, Dec. 2008. vol. 30, no. 3, pp. 39–49, May 2014.
[85] W. R. Rutgers and Y. H. Fu, ‘‘UHF PD-detection in a power transformer,’’ [107] O. D. Sparkman, Z. E. Penton, and F. G. Kitson, ‘‘Gas chromatography,’’
in Proc. 10th Int. Symp. High Voltage Eng., vol. 4, 1997, pp. 219–222. in Gas Chromatography and Mass Spectroscopy: A Practical Guide.
[86] W. Wiesbeck, C. Sturm, W. Soergel, M. Porebska, and G. Adamiuk, Waltham, MA, USA: Academic, 2011.
‘‘Influence of antenna performance and propagation channel on pulsed [108] J. Jiang, H.-T. Song, G.-M. Ma, C. R. Li, Y.-T. Luo, and H.-B. Wang,
UWB signals,’’ in Proc. Int. Conf. Electromagn. Adv. Appl., Sep. 2007, ‘‘Dissolved hydrogen detection in power transformer based on etched
pp. 915–922. fiber Bragg grating,’’ in Proc. IEEE Int. Instrum. Meas. Technol. Conf.
[87] T. Jiang, J. Li, Y. Zheng, and C. Sun, ‘‘Improved bagging algorithm for (IMTC) Proc., May 2015, pp. 422–427.
pattern recognition in UHF signals of partial discharges,’’ Energies, vol. 4, [109] A. Mukhtaruddin, M. Isa, M. R. Adzman, S. I. S. Hasan,
no. 7, pp. 1087–1101, Jul. 2011. M. N. K. H. Rohani, and C. C. Yii, ‘‘Techniques on partial discharge
[88] H. Jahangir, A. Akbari, P. Werle, and J. Szczechowski, ‘‘Possibility of detection and location determination in power transformer,’’ in Proc. 3rd
PD calibration on power transformers using UHF probes,’’ IEEE Trans. Int. Conf. Electron. Design (ICED), Aug. 2016, pp. 537–542.
Dielectrics Electr. Insul., vol. 24, no. 5, pp. 2968–2976, Oct. 2017. [110] T. Mak, R. J. Westerwaal, M. Slaman, H. Schreuders, A. W. van Vugt,
[89] A. M. Gaouda, A. El-Hag, T. K. Abdel-Galil, M. M. A. Salama, and M. Victoria, C. Boelsma, and B. Dam, ‘‘Optical fiber sensor for the
R. Bartnikas, ‘‘On-line detection and measurement of partial discharge continuous monitoring of hydrogen in oil,’’ Sens. Actuators B, Chem.,
signals in a noisy environment,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 190, pp. 982–989, Jan. 2014.
vol. 15, no. 4, pp. 1162–1173, Aug. 2008. [111] B. Sarkar, C. Koley, N. K. Roy, and P. Kumbhakar, ‘‘Condition monitoring
[90] J. Seo, H. Ma, and T. K. Saha, ‘‘A joint vibration and arcing measurement of high voltage transformers using fiber Bragg grating sensor,’’ Measure-
system for online condition monitoring of onload tap changer of the power ment, vol. 74, pp. 255–267, Oct. 2015.
transformer,’’ IEEE Trans. Power Del., vol. 32, no. 2, pp. 1031–1038, [112] J. Zhou, M. Zhu, Y. Yun, J. Tao, and F. Liu, ‘‘Partial discharge optical
Apr. 2017. pulse signal characteristics for corona defect in oil immersed trans-
[91] M. S. Abd Rahman, P. L. Lewin, and P. Rapisarda, ‘‘Autonomous local- former,’’ Res. J. Appl. Sci., Eng. Technol., vol. 5, no. 10, pp. 3052–3056,
ization of partial discharge sources within large transformer windings,’’ Mar. 2013.
IEEE Trans. Dielectrics Electr. Insul., vol. 23, no. 2, pp. 1088–1098, [113] J. Fabian, M. Neuwersch, C. Sumereder, M. Muhr, and R. Schwarz,
Apr. 2016. ‘‘State of the art and future trends of unconventional PD-measurement at
[92] H. Kalathiripi and S. Karmakar, ‘‘Partial discharge study in transformer power transformers,’’ J. Energy Power Eng., vol. 8, no. 6, pp. 1093–1098,
oil using acoustic emission technique and UV-visible spectroscopy,’’ in Jun. 2014.
Proc. 3rd Int. Conf. Condition Assessment Techn. Electr. Syst. (CATCON),
[114] T. B. Shanker, V. Vaidhyanathan, and A. S. Mohamed, ‘‘Comparison of
Nov. 2017, pp. 361–366.
partial discharge performance of identical generator transformers in a
[93] S. Qian, H. Chen, Y. Xu, and L. Su, ‘‘High sensitivity detection of thermal power station by acoustic emission technique—Case studies,’’
partial discharge acoustic emission within power transformer by sagnac in Proc. Int. Conf. High Voltage Eng. Technol. (ICHVET), Feb. 2019,
fiber optic sensor,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 25, no. 6, pp. 1–5.
pp. 2313–2320, Dec. 2018.
[115] Z. Qi, Y. Yi, W. Qiaohua, W. Zhihao, and L. Zhe, ‘‘Study on the online
[94] D.-J. Kweon, S.-B. Chin, H.-R. Kwak, J.-C. Kim, and K.-B. Song, ‘‘The
dissolved gas analysis monitor based on the photoacoustic spectroscopy,’’
analysis of ultrasonic signals by partial discharge and noise from the
in Proc. IEEE Int. Conf. Condition Monitor. Diagnosis, Sep. 2012,
transformer,’’ IEEE Trans. Power Del., vol. 20, no. 3, pp. 1976–1983,
pp. 433–436.
Jul. 2005.
[116] Y. Luo, X. Xin, F. Du, X. Tang, and Y. Li, ‘‘Comparison of DOA
[95] T. Boczar, A. Cichon, and S. Borucki, ‘‘Diagnostic expert system of trans-
algorithms applied to ultrasonic arrays for PD location in oil,’’ IEEE
former insulation systems using the acoustic emission method,’’ IEEE
Sensors J., vol. 15, no. 4, pp. 2316–2323, Apr. 2015.
Trans. Dielectrics Electr. Insul., vol. 21, no. 2, pp. 854–865, Apr. 2014.
[96] M. Kucera, R. Jarina, P. Brncal, and M. Gutten, ‘‘Visualisation and [117] Y.-J. Kim and M. Hikita, ‘‘A proposal for hybrid PD sensor identifying
measurement of acoustic emission from power transformers,’’ in Proc. absolute distance from PD source,’’ in Proc. Annu. Rep. Conf. Electr.
12th Int. Conf. Meas., May 2019, pp. 303–306. Insul. Dielectric Phenomena, Oct. 2013, pp. 1177–1180.
[97] D.-G. Kim, U. Sampath, H. Kim, and M. Song, ‘‘A fiber optic multi-stress [118] W. Hauschild and E. Lemke, High-Voltage Test and Measuring Tech-
monitoring system for power transformer,’’ in Proc. 25th Int. Conf. Opt. niques. Berlin, Germany: Springer-Verlag, 2014.
Fiber Sensors, Apr. 2017, pp. 1–4. [119] A. Kraetge, S. Hoek, M. Koch, and W. Koltunowicz, ‘‘Robust measure-
[98] W. Sikorski, K. Siodla, H. Moranda, and W. Ziomek, ‘‘Location of partial ment, monitoring and analysis of partial discharges in transformers and
discharge sources in power transformers based on advanced ausculta- other HV apparatus,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 20, no. 6,
tory technique,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 19, no. 6, pp. 2043–2051, Dec. 2013.
pp. 1948–1956, Dec. 2012. [120] Z. Yin, R. Zhang, J. Tong, and X. Chen, ‘‘An all-fiber partial discharge
[99] N. Mahmoudi, M. H. Samimi, and H. Mohseni, ‘‘Experiences with trans- monitoring system based on both intrinsic fiber optic interferometry
former diagnosis by DGA: Case studies,’’ IET Gener., Transmiss. Distrib., sensor and fluorescent fiber,’’ in Proc. Int. Conf. Opt. Instrum. Technol.,
vol. 13, no. 23, pp. 5431–5439, Dec. 2019. Opt. Sensors Appl., Dec. 2013, Art. no. 904414.
[100] M. Duval, ‘‘Dissolved gas analysis: It can save your transformer,’’ IEEE [121] A. Elfaraskoury, M. Mokhtar, M. Mehanna, and O. Gouda, ‘‘Conven-
Elect. Insul. Mag., vol. 5, no. 6, pp. 22–27, Nov. 1989. tional and un-conventional partial discharge detection methods in high
[101] H.-J. Seo, K.-H. Hwang, and D. Rhie, ‘‘Development of gas detector for voltage XLPE cable accessories,’’ Adv. Elect. Eng. Syst., vol. 1, no. 4,
hydrogen dissolved in transformer oil,’’ in Proc. IEEE Instrum. Meas. pp. 170–176, 2012.
Technol. Conf., May 2005, pp. 608–611. [122] M. Siegel, M. Beltle, S. Tenbohlen, and S. Coenen, ‘‘Application of
[102] G.-M. Ma, C.-R. Li, R.-D. Mu, J. Jiang, and Y.-T. Luo, ‘‘Fiber Bragg UHF sensors for PD measurement at power transformers,’’ IEEE Trans.
grating sensor for hydrogen detection in power transformers,’’ IEEE Dielectrics Electr. Insul., vol. 24, no. 1, pp. 331–339, Feb. 2017.
Trans. Dielectrics Electr. Insul., vol. 21, no. 1, pp. 380–385, Feb. 2014. [123] S. Markalous, S. Tenbohlen, and K. Feser, ‘‘Detection and location of
[103] D. Skelly, ‘‘Photo-acoustic spectroscopy for dissolved gas analysis: Ben- partial discharges in power transformers using acoustic and electromag-
efits and experience,’’ in Proc. IEEE Int. Conf. Condition Monitor. Diag- netic signals,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 15, no. 6,
nosis, Sep. 2012, pp. 39–43. pp. 1576–1583, Dec. 2008.
[104] S. E. U. Lima, O. Frazao, R. G. Farias, F. M. Araujo, L. A. Ferreira, [124] T. Hoshino, H. Koyama, S. Maruyama, and M. Hanai, ‘‘Comparison of
J. L. Santos, and V. Miranda, ‘‘Mandrel-based fiber-optic sensors for sensitivity between UHF method and IEC 60270 for onsite calibration in
acoustic detection of partial discharges—A proof of concept,’’ IEEE various GIS,’’ IEEE Trans. Power Del., vol. 21, no. 4, pp. 1948–1953,
Trans. Power Del., vol. 25, no. 4, pp. 2526–2534, Oct. 2010. Oct. 2006.
[105] J. Sheng, G. Wu, LaishengTong, LijunZhou, and JunZhang, ‘‘Study on [125] G. C. Stone, M. K. W. Stranges, and D. G. Dunn, ‘‘Recent developments
gas and oil separating plant used for online monitoring system of trans- in IEEE and IEC standards for off-line and on-line partial discharge
former,’’ in Proc. Conf. Rec. IEEE Int. Symp. Electr. Insul., Sep. 2004, testing of motor and generator stator windings,’’ in Proc. IEEE Petroleum
pp. 93–96. Chem. Ind. Tech. Conf. (PCIC), Sep. 2014, pp. 79–84.

VOLUME 9, 2021 64603

 M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

[126] C. Homagk, K. Mossner, and T. Leibfried, ‘‘Investigation on degradation [146] L. Satish and W. S. Zaengl, ‘‘Can fractal features be used for recognizing
of power transformer solid insulation material,’’ in Proc. Annu. Rep. Conf. 3-d partial discharge patterns,’’ IEEE Trans. Dielectrics Electr. Insul.,
Electr. Insul. Dielectric Phenomena, Oct. 2008, pp. 75–78. vol. 2, no. 3, pp. 352–359, Jun. 1995.
[127] S. Sriram, S. Nitin, K. M. M. Prabhu, and M. J. Bastiaans, ‘‘Signal [147] H. Chen, ‘‘Partial discharge identification system for high-voltage power
denoising techniques for partial discharge measurements,’’ IEEE Trans. transformers using fractal feature-based extension method,’’ IET Sci.,
Dielectrics Electr. Insul., vol. 12, no. 6, pp. 1182–1191, Dec. 2005. Meas. Technol., vol. 7, no. 2, pp. 77–84, Mar. 2013.
[128] N. A. Yusoff, M. Isa, H. Hamid, M. R. Adzman, M. N. K. H. Rohani, [148] D. P. Ferro, M. A. Falconi, R. L. Adam, M. M. Ortega, C. P. Lima,
C. Y. Chai, and N. N. Ayop, ‘‘Denoising technique for partial discharge C. A. de Souza, I. Lorand-Metze, and K. Metze, ‘‘Fractal characteristics
signal: A comparison performance between artificial neural network, fast of May-Grünwald-Giemsa stained chromatin are independent prognostic
Fourier transform and discrete wavelet transform,’’ in Proc. IEEE Int. factors for survival in multiple myeloma,’’ PLoS ONE, vol. 6, no. 6,
Conf. Power Energy (PECon), Nov. 2016, pp. 311–316. Jun. 2011, Art. no. e20706.
[129] M.-Y. Lin, C.-C. Tai, Y.-W. Tang, and C.-C. Su, ‘‘Partial discharge [149] T. Babnik, R. K. Aggarwal, and P. J. Moore, ‘‘Principal component and
signal extracting using the empirical mode decomposition with wavelet hierarchical cluster analyses as applied to transformer partial discharge
transform,’’ in Proc. 7th Asia–Pacific Int. Conf. Lightning, Nov. 2011, data with particular reference to transformer condition monitoring,’’ IEEE
pp. 420–424. Trans. Power Del., vol. 23, no. 4, pp. 2008–2016, Oct. 2008.
[130] J. C. Chan, H. Ma, and T. K. Saha, ‘‘Automatic blind equalization and [150] M. S. Abd Rahman, P. Rapisarda, and P. L. Lewin, ‘‘The use of three
thresholding for partial discharge measurement in power transformer,’’ dimensional filters for on-line partial discharge localisation in large trans-
IEEE Trans. Power Del., vol. 29, no. 4, pp. 1927–1938, Aug. 2014. formers,’’ in Proc. IEEE Electr. Insul. Conf. (EIC), Jun. 2014, pp. 10–14.
[131] S. K. Rajendra, M. Shrimali, S. Doshi, and M. Sharma, ‘‘Detection of [151] L. Satish and W. S. Zaengl, ‘‘Artificial neural networks for recognition
power transformer winding faults using orthogonal wavelet filter bank,’’ of 3-d partial discharge patterns,’’ IEEE Trans. Dielectrics Electr. Insul.,
in Proc. 5th Int. Conf. Signal Process. Integr. Netw. (SPIN), Feb. 2018, vol. 1, no. 2, pp. 265–275, Apr. 1994.
pp. 431–436. [152] S. Strachan, S. Rudd, S. McArthur, M. Judd, S. Meijer, and E. Gulski,
[132] J. Seo, H. Ma, and T. Saha, ‘‘Probabilistic wavelet transform for partial ‘‘Knowledge-based diagnosis of partial discharges in power transform-
discharge measurement of transformer,’’ IEEE Trans. Dielectrics Electr. ers,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 15, no. 1, pp. 259–268,
Insul., vol. 22, no. 2, pp. 1105–1117, Apr. 2015. Feb. 2008.
[133] J. Li, T. Jiang, S. Grzybowski, and C. Cheng, ‘‘Scale dependent wavelet [153] J. Li, R. Liao, S. Grzybowski, and L. Yang, ‘‘Oil-paper aging evalua-
selection for de-noising of partial discharge detection,’’ IEEE Trans. tion by fuzzy clustering and factor analysis to statistical parameters of
Dielectrics Electr. Insul., vol. 17, no. 6, pp. 1705–1714, Dec. 2010. partial discharges,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 17, no. 3,
pp. 756–763, Jun. 2010.
[134] S. Poornima and K. Ravindra, ‘‘Comparison of CWT & DWT based algo-
[154] L. Duan, J. Hu, G. Zhao, K. Chen, J. He, and S. X. Wang, ‘‘Identification
rithms in combination with ANN for protection of power transformer,’’
of partial discharge defects based on deep learning method,’’ IEEE Trans.
in Proc. Int. Conf. Signal Process., Commun., Power Embedded Syst.
Power Del., vol. 34, no. 4, pp. 1557–1568, Aug. 2019.
(SCOPES), Oct. 2016, pp. 1781–1785.
[155] H. Ma, T. K. Saha, C. Ekanayake, and D. Martin, ‘‘Smart transformer
[135] K. Fang, H. Zhang, H. Qi, and Y. Dai, ‘‘Comparison of EMD and EEMD
for smart grid—Intelligent framework and techniques for power trans-
in rolling bearing fault signal analysis,’’ in Proc. IEEE Int. Instrum. Meas.
former asset management,’’ IEEE Trans. Smart Grid, vol. 6, no. 2,
Technol. Conf. (I2MTC), May 2018, pp. 1–5.
pp. 1026–1034, Mar. 2015.
[136] J. Chan, H. Ma, T. Saha, and C. Ekanayake, ‘‘Self-adaptive partial dis-
[156] A. A. Mazroua, M. M. A. Salama, and R. Bartnikas, ‘‘PD pattern recog-
charge signal de-noising based on ensemble empirical mode decomposi-
nition with neural networks using the multilayer perceptron technique,’’
tion and automatic morphological thresholding,’’ IEEE Trans. Dielectrics
IEEE Trans. Electr. Insul., vol. 28, no. 6, pp. 1082–1089, Dec. 1993.
Electr. Insul., vol. 21, no. 1, pp. 294–303, Feb. 2014.
[157] M. T. Gençoğlu and M. Cebeci, ‘‘Investigation of pollution flashover
[137] R. Jia, Y. Xie, H. Wu, J. Dang, and K. Dong, ‘‘Power transformer partial on high voltage insulators using artificial neural network,’’ Expert Syst.
discharge fault diagnosis based on multidimensional feature region,’’ Appl., vol. 36, no. 4, pp. 7338–7345, May 2009.
Math. Problems Eng., vol. 2016, Aug. 2016, Art. no. 4835694.
[158] G. Li, M. Rong, X. Wang, X. Li, and Y. Li, ‘‘Partial discharge patterns
[138] M.-X. Zhu, J.-N. Zhang, Y. Li, Y.-H. Wei, J.-Y. Xue, J.-B. Deng, recognition with deep convolutional neural networks,’’ in Proc. Int. Conf.
H.-B. Mu, G.-J. Zhang, and X.-J. Shao, ‘‘Partial discharge signals separa- Condition Monitor. Diagnosis (CMD), Sep. 2016, pp. 324–327.
tion using cumulative energy function and mathematical morphology gra- [159] T. Kari, W. Gao, D. Zhao, Z. Zhang, W. Mo, Y. Wang, and L. Luan,
dient,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 23, no. 1, pp. 482–493, ‘‘An integrated method of ANFIS and dempster-shafer theory for fault
Feb. 2016. diagnosis of power transformer,’’ IEEE Trans. Dielectrics Electr. Insul.,
[139] C. Boya, M. Ruiz-Llata, J. Posada, and J. A. Garcia-Souto, ‘‘Identification vol. 25, no. 1, pp. 360–371, Feb. 2018.
of multiple partial discharge sources using acoustic emission technique [160] S. Hongsheng and Z. Feng, ‘‘A novel learning method for ANFIS using
and blind source separation,’’ IEEE Trans. Dielectrics Electr. Insul., EM algorithm and emotional learning,’’ in Proc. Int. Conf. Comput. Intell.
vol. 22, no. 3, pp. 1663–1673, Jun. 2015. Secur. (CIS ), Dec. 2007, pp. 23–27.
[140] S. Nkosi and P. Bokoro, ‘‘Improving the diagnosis of incipient faults in [161] S. A. Khan, M. D. Equbal, and T. Islam, ‘‘A comprehensive comparative
power transformers using dissolved gas analysis and multi layer percep- study of DGA based transformer fault diagnosis using fuzzy logic and
tron,’’ in Proc. IEEE 28th Int. Symp. Ind. Electron. (ISIE), Jun. 2019, ANFIS models,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 22, no. 1,
pp. 112–117. pp. 590–596, Feb. 2015.
[141] A. A. Soltani and A. El-Hag, ‘‘Denoising of radio frequency partial [162] L. V. Ganyun, C. Haozhong, Z. Haibao, and D. Lixin, ‘‘Fault diagnosis of
discharge signals using artificial neural network,’’ Energies, vol. 12, power transformer based on multi-layer SVM classifier,’’ Electric Power
no. 18, p. 3485, Sep. 2019. Syst. Res., vol. 74, no. 1, pp. 1–7, Apr. 2005.
[142] M. Zhao and G. Xu, ‘‘Feature extraction of power transformer vibration [163] L. Hao and P. Lewin, ‘‘Partial discharge source discrimination using a
signals based on empirical wavelet transform and multiscale entropy,’’ support vector machine,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 17,
IET Sci., Meas. Technol., vol. 12, no. 1, pp. 63–71, Jan. 2018. no. 1, pp. 189–197, Feb. 2010.
[143] W. J. K. Raymond, H. A. Illias, and A. H. A. Bakar, ‘‘Classification of [164] K. Ibrahim, R. M. Sharkawy, M. M. A. Salama, and R. Bartnikas,
partial discharge measured under different levels of noise contamination,’’ ‘‘Realization of partial discharge signals in transformer oils utilizing
PLoS One, vol. 12, no. 1, 2017, Art. no. e0170111. advanced computational techniques,’’ IEEE Trans. Dielectrics Electr.
[144] M. Bakrutheen, M. W. Iruthayarajan, and A. Narayani, ‘‘Statistical fail- Insul., vol. 19, no. 6, pp. 1971–1981, Dec. 2012.
ure reliability analysis on edible and non edible natural esters based [165] S. Kanakambaran, R. Sarathi, and B. Srinivasan, ‘‘Robust classification of
liquid insulation for the applications in high voltage transformers,’’ partial discharges in transformer insulation based on acoustic emissions
IEEE Trans. Dielectrics Electr. Insul., vol. 25, no. 5, pp. 1579–1586, detected using fiber Bragg gratings,’’ IEEE Sensors J., vol. 18, no. 24,
Oct. 2018. pp. 10018–10027, Dec. 2018.
[145] V. Darabad, M. Vakilian, B. T. Phung, and T. Blackburn, ‘‘An efficient [166] R. Sharkawy, K. Ibrahim, M. M. A. Salama, and R. Bartnikas, ‘‘Particle
diagnosis method for data mining on single PD pulses of transformer swarm optimization feature selection for the classification of conduct-
insulation defect models,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 20, ing particles in transformer oil,’’ IEEE Trans. Dielectrics Electr. Insul.,
no. 6, pp. 2061–2072, Dec. 2013. vol. 18, no. 6, pp. 1897–1907, Dec. 2011.

64604 VOLUME 9, 2021

M. R. Hussain et al.: Overview and PD Analysis of Power Transformers: A Literature Review

[167] S. Coenen and S. Tenbohlen, ‘‘Location of PD sources in power trans- [187] H. Ma, C. Ekanayake, and T. K. Saha, ‘‘Power transformer fault diagnosis
formers by UHF and acoustic measurements,’’ IEEE Trans. Dielectrics under measurement originated uncertainties,’’ IEEE Trans. Dielectrics
Electr. Insul., vol. 19, no. 6, pp. 1934–1940, Dec. 2012. Electr. Insul., vol. 19, no. 6, pp. 1982–1990, Dec. 2012.
[168] J. Li, T. Jiang, C. Wang, and C. Cheng, ‘‘Optimization of UHF Hilbert [188] H. Kai-Feng and C. Yu-Hua, ‘‘An improved fuzzy C-means clustering
antenna for partial discharge detection of transformers,’’ IEEE Trans. algorithm,’’ in Proc. Int. Conf. Autom. Control Artif. Intell. (ACAI), 2012,
Antennas Propag., vol. 60, no. 5, pp. 2536–2540, May 2012. pp. 1199–1204.
[169] H. H. Sinaga, B. T. Phung, and T. R. Blackburn, ‘‘Recognition of single [189] A. Singh and G. Upadhyay, ‘‘Dissolved gas analysis of power transformer
and multiple partial discharge sources in transformers based on ultra- using K-means and support vector machine,’’ in Proc. IEEE 1st Int. Conf.
high frequency signals,’’ IET Gener., Transmiss. Distrib., vol. 8, no. 1, Power Electron., Intell. Control Energy Syst. (ICPEICES), Jul. 2016,
pp. 160–169, Jan. 2014. pp. 1–5.
[170] L. Cui, W. Chen, B. Xie, J. Du, Z. Long, and Y. Li, ‘‘Characteristic [190] K. Firuzi and M. Vakilian, ‘‘Multi-source partial discharge signals dis-
information extraction and developing process recognizing method of crimination by six bandpass filters and DBSCAN clustering,’’ in Proc.
surface discharge in oil immersed paper insulation,’’ in Proc. ICHVE Int. 12th Int. Conf. Properties Appl. Dielectric Mater. (ICPADM), May 2018,
Conf. High Voltage Eng. Appl., Sep. 2014, pp. 3–6. pp. 68–71.
[171] I. Bua-Nunez, J. E. Posada-Roman, J. Rubio-Serrano, and [191] J. C. Chan, H. Ma, and T. K. Saha, ‘‘Time-frequency sparsity map on
J. A. Garcia-Souto, ‘‘Instrumentation system for location of partial automatic partial discharge sources separation for power transformer
discharges using acoustic detection with piezoelectric transducers and condition assessment,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 22,
optical fiber sensors,’’ IEEE Trans. Instrum. Meas., vol. 63, no. 5, no. 4, pp. 2271–2283, Aug. 2015.
pp. 1002–1013, May 2014.
[172] M. Harbaji, K. Shaban, and A. El-Hag, ‘‘Classification of common partial MD RASHID HUSSAIN (Member, IEEE)
discharge types in oil-paper insulation system using acoustic signals,’’ received the B.Tech. and M.Tech. degrees in elec-
IEEE Trans. Dielectrics Electr. Insul., vol. 22, no. 3, pp. 1674–1683, trical engineering from Aligarh Muslim Univer-
Jun. 2015. sity, with the specialization in high voltage and
[173] H. Mirzaei, A. Akbari, E. Gockenbach, and K. Miralikhani, ‘‘Advancing insulation. He has worked as a Research Associate
new techniques for UHF PDdetection and localization in the power
with Texas A&M University, Qatar. He is cur-
transformers in the factory tests,’’ IEEE Trans. Dielectrics Electr. Insul.,
rently working as a Research Assistant with Qatar
vol. 22, no. 1, pp. 448–455, Feb. 2015.
University. His main research interests include
[174] K. Wang, J. Li, S. Zhang, F. Gao, H. Cheng, R. Liu, R. Liao, and
S. Grzybowski, ‘‘A new image-oriented feature extraction method for partial discharge analysis, nano-fluids, condition
partial discharges,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 22, no. 2, monitoring of electric and electronic equipment,
pp. 1015–1024, Apr. 2015. high-voltage measurement and insulation development, and nano-dielectrics,
[175] S. D. Mitchell, M. Siegel, M. Beltle, and S. Tenbohlen, ‘‘Discrimination finite element analysis, and biomedical engineering. He has published some
of partial discharge sources in the UHF domain,’’ IEEE Trans. Dielectrics research articles in his relevant fields
Electr. Insul., vol. 23, no. 2, pp. 1068–1075, Apr. 2016.
[176] R. Rostaminia, M. Saniei, V. Parvin, M. Vakilian, and S. S. Mortazavi, SHADY S. REFAAT (Senior Member, IEEE)
‘‘Accurate power transformer PD pattern recognition via its model,’’ IET received the B.A.Sc., M.A.Sc., and Ph.D. degrees
Sci., Meas. Technol., vol. 10, no. 7, pp. 745–753, Oct. 2016. in electrical engineering from Cairo University,
[177] R. Ghosh, B. Chatterjee, and S. Dalai, ‘‘A method for the localization of Giza, Egypt, in 2002, 2007, and 2013, respec-
partial discharge sources using partial discharge pulse information from tively. He has worked at the industry for more
acoustic emissions,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 24, no. 1, than 12 years as an Engineering Team Leader,
pp. 237–245, Feb. 2017. a Senior Electrical Engineer, and an Electrical
[178] J. Du, W. Chen, L. Cui, Z. Zhang, and S. Tenbohlen, ‘‘Investigation on Design Engineer on various electrical engineering
the propagation characteristics of PD-induced electromagnetic waves in projects. He is currently an Associate Research
an actual 110 kV power transformer and its simulation results,’’ IEEE Scientist with the Department of Electrical and
Trans. Dielectrics Electr. Insul., vol. 25, no. 5, pp. 1941–1948, Oct. 2018.
Computer Engineering, Texas A&M University, Qatar. He has participated
[179] C. Gao, W. Wang, S. Song, S. Wang, L. Yu, and Y. Wang, ‘‘Localization
and leads several scientific projects over the last eight years. He has success-
of partial discharge in transformer oil using Fabry-Pérot optical fiber
sensor array,’’ IEEE Trans. Dielectrics Electr. Insul., vol. 25, no. 6, fully realized many potential research projects. He has published more than
pp. 2279–2286, Dec. 2018. 105 journal articles and conference papers. His research interests include
[180] M. A. Ansari, D. Martin, and T. K. Saha, ‘‘Investigation of distributed electrical machines, power systems, smart grid, big data, energy management
moisture and temperature measurements in transformers using fiber systems, reliability of power grids and electric machinery, fault detection,
optics sensors,’’ IEEE Trans. Power Del., vol. 34, no. 4, pp. 1776–1784, and condition monitoring and development of fault-tolerant systems. He is
Aug. 2019. a member of the Institution of Engineering and Technology (IET) and the
[181] M. Zhu, Y. Wang, D. Chang, G. Zhang, X. Shao, J. Zhan, and J. Chen, Smart Grid Center-Extension in Qatar (SGC-Q).
‘‘Discrimination of three or more partial discharge sources by multi-
step clustering of cumulative energy features,’’ IET Sci., Meas. Technol., HAITHAM ABU-RUB (Fellow, IEEE) received
vol. 13, no. 2, pp. 149–159, Mar. 2019. the Ph.D. degree. He is currently a Full Professor.
[182] O. Ozgonenel and S. Karagol, ‘‘Power transformer protection based He has served five years as the Chair of the Elec-
on decision tree approach,’’ IET Electric Power Appl., vol. 8, no. 7, trical and Computer Engineering Program, Texas
pp. 251–256, Aug. 2014. A&M University, Qatar, where he is currently
[183] M. Wu, H. Cao, J. Cao, H.-L. Nguyen, J. B. Gomes, and serving as the Managing Director of the Smart
S. P. Krishnaswamy, ‘‘An overview of state-of-the-art partial discharge Grid Center. Since 2006, he has been with Texas
analysis techniques for condition monitoring,’’ IEEE Elect. Insul. Mag., A&M University. He has research and teaching
vol. 31, no. 6, pp. 22–35, Nov. 2015.
experience at many universities in many countries,
[184] M. E. A. Senoussaoui, M. Brahami, and I. Fofana, ‘‘Combining and
including Poland, Palestine, USA, Germany, and
comparing various machine-learning algorithms to improve dissolved gas
analysis interpretation,’’ IET Gener., Transmiss. Distrib., vol. 12, no. 15,
Qatar. He has published more than 400 journal articles and conference
pp. 3673–3679, Aug. 2018. papers, five books, and six book chapters. He has supervised many research
[185] S. D. J. McArthur, S. M. Strachan, and G. Jahn, ‘‘The design of a multi- projects on smart grid, power electronics converters, and renewable energy
agent transformer condition monitoring system,’’ IEEE Trans. Power systems. His main research interests include power electronic converters,
Syst., vol. 19, no. 4, pp. 1845–1852, Nov. 2004. renewable energy, electric drives, and smart grid. He was a recipient of
[186] C. Zhang and S. Xia, ‘‘K-means clustering algorithm with improved the American Fulbright Scholarship, the German Alexander von Humboldt
initial center,’’ Proc. 2nd Int. Workshop Knowl. Discovery Data Mining, Fellowship, and many national and international awards and recognitions.
Jan. 2009, pp. 790–792.

VOLUME 9, 2021 64605