A partial discharge (PD) is a transient breakdown of a part of the insulation system in high voltage equipment. By this process, the entire dielectric is not breached; that is, the breakdown does not occur between two conductors. Only a small portion of the insulation suffers a miniature breakdown.

Partial discharges can be thought of as motion of charges due to a local intensification of the electric field which exceeds the breakdown strength in a small region of the insulation.

A sustained PD activity leads to an accelerated aging of the insulation and can become the cause of an eventual failure. PD processes are complex and recorded signals are statistical in nature, and the effect of previous discharges on the subsequent ones cannot generally be ignored.

Partial Discharge in Transformer

The correlation of the physics of a PD activity with measured electrical, acoustic or chemical signals is extremely intricate due to the presence of various possible conditions and parameters that affect it. In fact, determining such correlations has been an important and integral part of diagnostic studies recently.

Presence of microscopic voids is inevitable in the manufacture of solid dielectrics such as pressboard cylindrical barriers, supports or spacers. These voids will be mostly filled with the impregnating medium which may be resin or oil. The potential difference across the insulator sets up an electric field. The void, with or without the impregnant, has a lower dielectric constant than the surrounding solid, and hence has a higher electric field across it.

When the electric field exceeds its dielectric strength, a local breakdown occurs. The actual physical process is fairly intricate. Ionization has to take place as the voltage rises to peak in a half cycle, in order to generate the charges that jump across the void and reduce the overall electric field.

The electrons so produced move mainly under the influence of the external electric field. These electrons get deposited on the dielectric surface, and thus effectively reduce the field inside the void, which leads to PD extinction as the voltage dips after reaching the peak. The presence of deposited charges causes the discharge pulse to occur at a smaller value of the phase angle during the next half-cycle, as the charges effectively increase the field within the void during this opposite half-cycle.

The process is further complicated by the generation of secondary electrons and the memory effect of the previous discharge pulse. In a void, PD takes place when the following conditions are satisfied, viz. the local electric field should be greater than the breakdown field strength, electrons must be available for starting the physical process, the weak region should be enough in size for the avalanche to occur, and the duration of the applied voltage should be sufficiently long to bridge the weak dielectric region.

Local field intensification also takes place in oil in the solid-oil interface due to improper stress grading or due to sharp points on the tank. When the field exceeds the oil dielectric strength, PD will result in oil.

Diagnostic Techniques for Partial Discharge in Transformer

PD results in a localized and nearly instantaneous release of energy, which produces a number of effects such as local heating, generation of electromagnetic and acoustic waves, chemical reactions, among others. Correspondingly, the following different methods of recording these have been developed.

Dissolved gas analysis: Like many other defects, PD occurring in oil causes the generation of hydrogen gas. Hence, its trace detected during DGA may be indicative of PD activities.

Detection of certain gases such as carbon monoxide or carbon dioxide may indicate the presence of partial discharges in paper insulation. Since the paper insulation can be replaced only by rewinding, even a small trace of carbon monoxide/dioxide becomes a matter of concern.

Electrical method: In this technique, a large valued coupling capacitor is connected across the equipment under test, and the apparent charge provided during the PD discharge is measured by integrating the high-frequency current pulses flowing through this capacitor.

The actual charge transfer across the cavity cannot be measured, and hence the apparent charge transfer has to be measured. The energy dissipated during the discharge is directly proportional to this apparent charge transfer and the square of the inception voltage.

A measuring impedance is connected either in series with the test specimen or with the coupling capacitor. For testing a transformer, a bushing is invariably used as the coupling capacitor. Impedance is connected between the bushing tap and ground for sensing the capacitively coupled PD signal.

Alternatively, a high frequency current transformer (HFCT) is placed in the bushing turret or around the ground lead from neutral to act as a current sensor.

One common method of studying PD phenomena is phi-q-n analysis. Here, phi is the phase angle, with respect to the voltage waveform, at which the PD pulse occurs, q is the magnitude of the discharge pulse in pC, and n indicates the number of such occurrences over a period.

This data acquired over a long duration can exhibit a 3-D pattern which can be used to identify the source of PD inside the transformer. Also, the evolution of the pattern over a much longer period may be used to analyze the introduction/suppression of PD sources. An appropriate statistical analysis of the measured data over many cycles can give useful information.

The method’s offline nature, elaborate setup, high external/ electromagnetic interference, and difficulties in locating PD discharges are the major shortcomings.

The measurements are affected by electrical interference signals from surrounding equipment. Hence, it is not attractive for on-site measurements.

One more limitation is that the maximum observed amplitude of the discharge may not be indicative of either the deterioration taking place or the time to breakdown.

For example, a solid dielectric with an internal void breaks down after the appearance of a tree; the propagation rate for the tree is generally greatest when the detected discharge amplitude and frequency are the lowest. Continuous monitoring at the bushing tap or through a HFCT and the use of phase-resolved digital measurements over a long period with suitable statistical analysis may somewhat overcome these disadvantages.

The PD detection range for the electrical method is larger; it covers a wider area which includes, for example, tap changer and bushings. There is better correlation between instrument readings and actual PD magnitudes as compared to that with the acoustic method which is discussed next.

Acoustic method: This method involves the detection of mechanical signals emitted from the discharge as a result of pressure variations. The corresponding waves propagate through the surrounding oil and other insulating materials, which can be detected at a tank wall using piezoelectric sensors fitted externally.

The acoustic sensors based on the piezoelectric effect are less expensive and the main advantage of the acoustic technique is that there is generally no interference in measurements due to signals from the electric network.

However, the PD detection is possible within a radius of only about 20 to 30 cm from the source since the acoustic signals are attenuated by the media through which they travel. Hence, a number of acoustic sensors may have to be used, located judiciously around the transformer periphery. Acoustic sensors can also be placed internally using waveguides (e.g., fiberglass rods) to enhance the strength of the signals to be measured, but the system is expensive and difficult to install.

A typical acoustic PD detection system consists of a sensor, a filter, a preamplifier and a data acquisition system. Ambient noise of mechanical origin mainly influences the signal-to-noise ratio and sensitivity of the PD detection system.

Sensors used have a typical frequency range of 70 to 150 kHz to suppress the core noise as well as the noise due to mechanical vibrations of the transformer structure. The location of the PD source can be determined based on the measurement of the time of arrival and the amplitude of the signal.

The intensity of the emitted acoustic signal is proportional to the discharge energy. Therefore, a linear relationship between the peak discharge magnitude and the amplitude of the acoustic signal over a wide frequency range is possible.

The shape of the detected signal depends on the source, the detection apparatus and the sensor. Simple hardware requirements and well-established discharge-location algorithms make the method attractive. Also, being immune to electromagnetic noise, it can be effectively used for on-site condition monitoring.

The main limitation of the method is that its sensitivity can be seriously hampered by signal attenuation. Also, the propagation velocity of the ultrasonic waves is strongly dependent on material properties of the media. It is less sensitive to partial discharges occurring in inner windings.

Furthermore, a considerable amount of expertise and experience is required for interpreting acoustic measurements since multiple signals are recorded after attenuation, reflections, and scattering. Thus, when an acoustic wave propagates through the insulating media, its intensity decreases as a function of the distance from the source due to several mechanisms such as geometrical spreading of the wave, acoustic absorption (conversion of acoustic energy to heat) and scattering of the wave-front.

It is therefore important to increase the signal-to-noise ratio through an averaging technique whereby the noise is effectively suppressed. The technique works very well when the signal and the noise are uncorrelated, the signal is of repetitive nature, and the noise is white (meaning that it contains every frequency within the investigated band in equal proportions).

For an effective averaging to happen, a trigger derived from a physical PD effect having higher sensitivity can be used. The electrical PD method or UHF (ultra high frequency) PD method, to be discussed next, can be used for the purpose.

UHF method: This technique involves detection of electromagnetic waves (300–3000 MHz) emitted by PD activities. During a PD pulse, electrons from the outermost orbit are dislodged from atoms or molecules and are accelerated rapidly by the electric field. Electromagnetic radiations occur and spread in all possible directions when electrical charges are accelerated. Bad contacts and electrostatic discharges due to floating metallic parts can also result in radiations.

The signals are attenuated as they travel through complex arrangement of conductors and insulation due to losses and reflections. They are also reflected at tank boundaries. Because of the inverse relationship between the time-domain and frequency-domain representations of a pulse, shorter PD current pulses have more spectral energy at high frequencies.

Signals in the UHF range can be measured by suitable tank-mounted sensors when the discharges occur on a time scale of 1 ns or less. The sensors used in gas-insulated substations (GIS) have been successfully applied for UHF diagnostics of transformers. PD sources can be located by the time-of-flight method.

Signals from sensors are usually filtered and amplified, and analog-to-digital conversion is done to digitize the data. The phase of each recorded PD pulse can be easily determined using a clock and a phase reference signal. The amplitude of the pulses can be directly correlated to the energy content in the UHF signal.

Neural networks can be used for pattern recognition to diagnose the characteristics of the PD source. Dielectric windows need to be provided for enabling the UHF diagnostics of transformers. They provide a robust electrical aperture through which the UHF signals can be detected. Alternatively, the sensor can be installed through a valve, or it can be permanently positioned inside the tank.

The UHF sensors must have a large bandwidth because the frequency content of the signals from a PD source can vary considerably depending on its location and the signal path. Sensitivity over the range of 500 to 1500 MHz is essential.

Although immune from electrical interference, the UHF measurements may be affected by communication noise, thermal noise in the detection system, and noise due to operation of power electronic switches; denoising of UHF signals can be done using the discrete wavelet transform.

It has been experimentally validated that corona signals exist in a frequency range of 300 MHz to 1500 MHz. The signals received for positive and negative DC polarities are distinct. The overall strength of a signal captured from a negative polarity discharge is stronger than that for a positive polarity discharge, on an average by about 10 dB, for the same voltage magnitude.

Also, the negative polarity corona has more uniform frequency content. The corona magnitude for the positive polarity has a tendency to increase in the frequency range of 1000–1500 MHz. Correlations between the physical processes and the received signals have also been identified. The electron avalanche travel duration in the negative polarity corona is much longer than that in the burst pulses or streamers for the positive polarity corona.

At higher voltages, it can be expected that a larger region around the point electrode gets depleted of electrons and, hence, forms a positive space charge.

Consequently, electrons that were further away reach the positive point and thus there is an increase in the current pulse duration and a consequent increase in the lower frequency content of the spectrum. These are the possible mechanisms responsible for a changing UHF pattern with increasing voltage. The analysis of signals received for an applied AC voltage has also been done with respect to physical discharge mechanisms.

Under DC voltage conditions, due to the faster accelerations of electrons and ions in negative streamers and short pulses in positive streamers, the frequencies emitted by the discharges can be expected to be in a higher range. However, these two modes are not observed in the case of AC corona. Hence, higher frequency signals may be much smaller in magnitude for AC as compared to DC.

This discussion is more applicable to point-plane configurations. The analysis and experimental work needs to be extended for practical configurations encountered in transformers. The interest in UHF detection of PD is mainly due to the possibility of online monitoring.

Secondly, with UHF frequencies in the range of 300 to 3000 MHz, the technique is inherently immune to noise. Very small discharges up to 10 pC can be recorded with good sensitivity. In this respect, the UHF method is much superior to the acoustic technique if discharges are confined within insulating barriers.

A main disadvantage of the method is that special arrangements for incorporating UHF sensors have to be made during the manufacturing stage. Also, a complicated algorithm is required for locating discharges. Hence, although the method is well-developed for GIS equipment, it is still evolving for transformers.

FDTD method: Recently, numerical simulations have been attempted to gain further insight into the PD propagation to aid diagnostics. The finite difference time domain (FDTD) technique is widely used to simulate and analyze electromagnetic wave propagation phenomena.

The technique is relatively easy to understand and implement compared to other methods such as FEM and method of moments (MOM) when applied to high-frequency computational electromagnetics.