Partial discharges (PD) TEST

 

Partial discharges (PD)

Partial discharges (PD) may occur in electrical insulation systems that operate at 3.3 kV and above. PD only occurs when a gas filled voids are present within the insulation or a gas (usually air) is present on the insulation surface when there is a high electric stress [23]. If the stress is high enough, the gas will experience electrical breakdown, creating a spark consisting of energetic electrons which will break molecular bonds in any organic polymer. Thus PD will age the insulation and may eventually cause failure. PD occurs in a wide variety of high voltage electrical apparatus such as transformers, gas insulated switchgear, power cables and rotating machines. Since each discharge causes of a flow of charge, the PD can be detected by measuring the current pulses on the terminals of high voltage equipment. Off-line PD testing has been as a factory test for almost 100 years on equipment such as power cables. The purpose is to detect flaws created during manufacturing that lead to PD, and thus lead to insulation failure. In the past 30 years or so, owners of high voltage equipment are also measuring PD over time on installed equipment. Many aging processes can create voids that can lead to PD, and thus PD is often a symptom of thermal and thermo-mechanical aging processes. By monitoring the evolution of PD over time either in off-line tests or by on-line monitoring while the equipment is operating normally, equipment owners have a powerful tool for determining when maintenance or equipment replacement is needed. More commonly, machine owners have been using off-line and on-line PD testing to assess the condition of the stator winding insulation in order to determine if maintenance is needed. Problems such as loose coils in the stator slots, contamination leading to electrical tracking and thermal aging of the insulation are easily detected

There are many different types of PD testing equipment that have been used for coils and stator windings. Most use a capacitor to detect the PD pulse currents in the presence of the 50/60 Hz high voltage. The instrumentation to measure the PD current pulses most commonly includes an analog to digital converter that determines the number, magnitude and phase position (with respect to the 50/60 Hz ac cycle) of the PD. However almost every brand of PD detector works in a different part of the frequency spectrum. Since each partial discharge pulse is the result of a brief flow of electrons lasting only a few nanoseconds, by the Fourier transform, frequencies from 0 Hz up to several hundred MHz are created by each discharge. Thus PD can be detected in a very wide range of frequencies, and this will impact what is actually measured. This paper discusses how the frequency range affects the detection of PD in stator windings.

2 PD – A COMPARISON TEST
Partial discharges (PD) are small electrical sparks that occur when voids exist within or on the surface of high voltage insulation of stator windings in motors and generators. These PD pulses can occur because of the manufacturing/installation processes, thermal deterioration, winding contamination or stator bar movement during operation. As the insulation degrades, the number and magnitude of PD pulses will increase. Although the magnitude of the PD pulses cannot be directly related to the remaining life of the winding, the doubling of PD pulse magnitudes approximately every 12 months has been used as a “rule of thumb” to indicate rapid deterioration is occurring [If the rate of PD pulse activity increases rapidly, or the PD levels are high compared to other similar machines, this is an indicator that visual inspections and/or other testing methods are needed to confirm the insulation condition Furthermore, if the PD magnitudes by the same test method from several identical windings are compared, the windings exhibiting higher PD activity are generally closer to failure .

.

VOLTAGE
It is well known that voltage has a dramatic effect on the PD magnitude, at least if the insulation is not excessively overstressed. Figure 1 shows the effect of voltage on the PD magnitude of a 13.8 kV epoxy-mica insulated stator coil, a 5% reduction in voltage around 8 kV causes 30% reduction in the PD magnitude, a 10% reduction, 60%. Thus, a small decrease in applied voltage causes an even larger decrease in the detected PD pulse magnitude. In addition, at a given test voltage, the number of pulses per second drops exponentially with pulse magnitude. Hence, there is a significantly lower PD pulse count rate at the lower voltages.

Figure 1. Pulse magnitude and pulse phase analysis of the PD on the stator coil energized to 8 kV

This has a significant implication for on-line PD testing, where the voltage across each coil decreases linearly between the phase terminal and neutral. In a typical machine with 100 coils, the coil connected to the phase end will have the highest PD activity (magnitude and number), A coil connected one coil down, will have a PD magnitude that is about half of that in the phase end coil, with the same amount of deterioration. A coil two coils down from the phase end, will have a PD magnitude one quarter of the magnitude in the phase end coil, and even fewer pulses. This analysis plus many years of visual observation of PD effects in actual machines show that only the first few coils in a winding suffer the most from PD. There is no PD at the neutral end.

From the data described above, it is clear that the coil voltage has a much higher impact on the PD magnitude than the winding attenuation effect. The result is that one can choose whatever bandwidth one desires for PD measurements in operating stators; if the PD sensor is located at the phase terminals, it is near the only coils likely to be subject to high PD activity Therefore, one can detect the PD at either low frequency or high frequency, since pulse attenuation at higher frequencies is relatively small, because the most active PD sites are close to the PD sensor.

Therefore, PD sensors should be located near the first few coils in a stator winding.

6 BANDWIDTH
The frequency domain characteristics of a pulse stream can be calculated using standard Fourier transforms [39]. Alternatively, several papers [40][41][42] show how to calculate the spectrum, including the all-important upper frequency at which the signal level starts decaying to 0. From these calculative methods, it is clear that pulses, such as those shown in Figure 2 produce frequencies into ~300 MHz. To a first approximation, this can be seen by making, the unipolar pulse shown in Figure 2 into the first half cycle of a sinusoid. The unipolar pulse in Figure 2 has a duration of ~3ns. In constructing the sinusoid, there is a period of ~6 ns. Since frequency is the inverse of the period, the pulse in Figure 2 has harmonics at 160 MHz. For shorter duration pulses, frequencies to 350 MHz are present.

Figure 2. Spectrum analyzer output for a stream of PD pulses occurring in an operating generator, which shows the frequency domain response. Note that at high frequency the signal levels are relatively low. The spectrum on the left covers the range from 100kHz to 35 MHz.

LF VS VHF OFF-LINE PD ON MOTOR STATOR
Off-line PD tests were performed using LF and VHF PD detectors on a 13.2 kV, 6000 HP motor stator winding [37]. The LF test was performed with a PDTech DeltaMaxx analyzer using a 1000 pF PD detection capacitor. It operated in the wideband mode in frequency range 40 – 800 kHz. C phase had the highest activity and is shown in Figure 3. Figure 3 (left) shows the phase resolved PD (PRPD) pattern obtained after stabilization at 8 kV (just above rated line to ground voltage). Figure 3 (right) shows the VHF PRPD plot measured on the same phase of the stator winding at the same voltage using an 80 pF PD sensor and an Iris Power TGA-B instrument. The PD detection frequency range is 40 -350 MHz.

Figure 3. Off-line PD test on a stator winding measured in the LF (top) and VHF (bottom) frequency ranges. The vertical scale is the apparent pulse magnitude in either nC or mV. The horizontal scale is the phase angle of the 60 Hz AC cycle. The colour of the dots represents the number of PD pulses per second. Note that the VHF PRPD plot shows the polarity of the PD pulses. In the LF PRPD plot the pulse polarity is suppressed, but can be inferred from its position on the AC cycle.

The peak PD magnitude (Qm) for the VHF measurement (calculated using the definition for digital instruments in IEC 60034-27 [24] is +816 mV and -912 mV. The Qm is the magnitude at a PD pulse repetition rate of 10 pulses per second (pps). The LF PD instrument calculated Qm to be 2.5 nC, using the same 10 pps definition. Note however some LF instruments calculate the peak PD magnitude using a method based on an analog definition of the largest repetitive magnitude in a train of pulses, as defined in IEC 60270 [23].

C phase of this stator winding has a ratio of 2.7 pC/mV between the LF and the VHF detectors. However note the cautions in IEC 60034-27 [24] that different brands of PD instruments are likely to give much different pC levels even under the same test conditions. The PRPD patterns and the relationship between positive and negative PD is essentially the same between LF and VHF.

PD occurs into frequencies up to ~300 MHz

LF VS. VHF OFF-LINE PD IN A TURBINE GENERATOR STATOR
The stator winding of an 18 kV, 200 MVA hydrogen-cooled generator was also given an off-line PD test using both LF and HF instruments described above. The tests were done at about 9.7 kV, a little below the rated line to ground voltage, in atmospheric pressure air. Fig 2a shows the PRPD plot from Phase C, which had the highest activity. The plot shows classic internal groundwall activity (see IEC 60034-27 for PRPD plots associated with each type of PD source), with approximately equal positive and negative PD activity. A Qm of 1.1 nC was measured.

Figure 4. LF (top) and VHF (bottom) off-line PD test on a 200 MVA generator stator

Figure 4 shows the PDPD plot measured in the VHF range on C phase under the same test conditions as for the LF test.

It shows the same PRPD pattern as for the LF test. The Qm (calculated at 10 pps) is +106 mV and -121 mV. Thus the ratio between the LF test in pC and the VHF test in mV is 9.1 pC/mV, which is substantially different than the ratio measured on the motor. This variability exemplifies the difficulty in calibration as described below.  

SIGNAL TO NOISE RATIO
During on-line testing, the noise (i.e. electromagnetic pulses from sources different to PD from stator winding) becomes relevant, because the stator winding subjected to testing is connected to the rest of the system with many sources of noise (sparks from busbars, corona from bolts, tracking from insulators, PD from current transformers, sparks from tools, radio broadcasting signals, etc). This electrical interference is a factor which can have an influence on the measurement bandwidth. The electrical noise can be as much as l000x (60 dB) larger than the PD signals from an operating motor or generator, especially if the stator winding is cooled by high pressure hydrogen gas. [Figure 5]

The question to be asked when trying to detect PD in an operating motor-or generator is not-how large is the PD signal, but how large is the PD signal in comparison to the ‘noise’, i.e. the signal-to-noise ratio (SNR). Boggs [34] shows that if the noise has a broadband (white noise) characteristic, contamination theory indicates that the optimum (i.e. highest SNR) frequency band for PD detection, assuming there is little attenuation, occurs at ~250 MHz. This is because the power in broadband (white or electronic) noise increases with the square root of the bandwidth of the measuring system; whereas, the power in the PD signal increases proportionately, with bandwidth, up to the upper limit set by the PD pulse risetime. The result is that the SNR increases with the square root of the bandwidth. A higher SNR reduces the risk of false indications of stator winding problems caused by noise. This is why VHF or UHF with UWB (i.e. ~100 MHz) PD detectors have become dominant for on-line PD defection in rotating machines [33], [35], [36].

Figure 5. Oscilloscope image of the combined PD and noise signals from each phase. The vertical scale is pulse magnitude and the horizontal scale is one 60 Hz AC cycle.

On-line tests, high SNR (VHF or UHF with UWB) reduces risk of false indications

ADVANTAGES OF LF AND VHF PD DETECTION
Off-tine PD testing is best performed with PD detectors operating in the 1 MHz range, to ensure that PD can be detected in all the coils in a stator winding with the minimal amount of attenuation. On-line PD tests are best performed at higher frequencies since this optimizes the signal to noise ratio, as well as enabling the separation of disturbances from PD based on time-of pulse arrivals. The result is a PD signal free of interference, and thus a reduced risk of false alarms. Although more pulse attenuation will occur at the higher measurement bandwidth needed to separate out the noise, this attenuation is relatively minor, since in an operating stator the sensor can be placed very close to the few coils experiencing the HV.

Off-line tests on both coils/bars and complete stator windings should be performed in the LF range, even though Figure 3 and Figure 4 show the PRPD patterns are essentially the same. For coil/bar tests this will allow the PD to be scientifically quantified in terms of apparent charge (pC). For windings, the LF range maximizes sensitivity to PD in more of the coils/bars in the winding.

The basic advantages of the LF test for off-line testing, as described in IEC standards [24][28], include

The LF advantage of greater sensitivity is less important than for off-line tests, since the coil/bar voltage decreases linearly through the circuit from the phase terminal to the neutral end of the winding. As the voltage deceases so does the PD magnitude and the number of defects that produce PD.

There is greater immunity to noise and disturbances from the power system with VHF and UHF methods, which lowers the risk of false indications of stator winding problems.  Also this implies that less expertise is needed to perform and interpret PD results, since there is a lower risk the stator PD is obscured by the noise. This implies a lower marginal test cost.  As a consequence, continuous PD monitoring is less likely to give false indications.

With some of the VHF methods and all of the UHF methods, it is possible to locate with more certainty where the PD is occurring within the winding.

Most capacitive PD sensors in VHF methods can meet the sensor reliability requirements stipulated in IEC 60034-27-2, thus reducing the risk that a PD sensor may fail the machine

Peer-reviewed databases containing hundreds of thousands of test results have been summarized in tables of “high” and “low” PD.  These severity levels have been confirmed by visual inspections on many hundreds of machines [22]

It is clear from the IEC standards that both the LF and VHF ranges can detect severe PD in a winding. The LF method tends to be preferred by OEMs and test service providers who have the expertise to separate the PD from the noise, and judge the severity based on experience with similar machines. Owners of machines tend to use the VHF and UHF methods since utility staff can perform the test and do a basic interpretation with relatively little training and experience.

On-line tests, high SNR (VHF or UHF with UWB) reduces risk of false indications

6.5 MATCHING BANDWIDTH FREQUENCIES
An instrument provider researched doing on-line PD data collection service using a LF instrument connected to existing 80 pF capacitive PD sensors [38]. 80pF sensors terminated in 50ohms provide a high-pass filter of ~ 40MHz. According to the data available for the LF PD monitoring instrument, the input frequency range is up to 20 MHz and bandwidth up to 3 MHz, and it has only one PD sensor input per phase (single ended installation). Note that VHF instruments have a bandwidth from 50 kHz to 350 MHz and when combined with the 80 pF capacitive PD sensor, the system bandwidth is 40 MHz to 350 MHz. The LF instrument configuration used a matching unit (basically an amplifier plus a low pass filter) intended to get from 80 pF PD sensors (EMCs), to a measuring range between 100 kHz to 500 kHz [38].

The measured background noise, the signal to noise ratio (SNR) of the LF instrument connected to the 80pF sensors is only 12% of the SNR from the typical 2nF PD sensors. The research shows that when the matching unit described above is used, the measurement is less reliable since the SNR from the connected 80 pF sensors gets worse and decreases to 11% of the SNR from the 2nF PD sensors. Thus it is clear from these tests that if a low frequency measurement system is used, it is best to connect it to a large capacitance, rather than try to force fit the instrument to a PD sensor made for high frequency applications.

Bandwidth frequency of the sensors should match that of the instrument

7 DISTURBANCES
Much of the noise or disturbances encountered in operating machines is not of the white sort, but is pulse-like, for example: corona on HV buses, switching noise, etc. Traditional filtering cannot completely remove such pulse-like noise, since, as described above, such signals will contain frequency components at all frequencies, especially if the noise pulse risetime is fast. Hence, any filtering of the noise pulses will also filter the PD, resulting in no net gain in SNR. It is apparent that alternative methods to filtering are needed to eliminate pulse-like noise [25]. Several methods have been developed based on time-of-pulse arrival from a pair of sensors, or on more complex differentiation of pulse shapes in the time domain [33], [35]. Such pulse discriminations are now widely used, and the required specialized sensors are installed in ~3500 machines. The discrimination techniques require the measurement of PD at high frequencies (≥40 MHz), since the fast risetime pulse characteristic of PD needs to be preserved so that it can be used as a precise riming signal, Of course, human experts observing data displayed on oscilloscope screens and spectrum analyzers, are also capable of separating the stator PD from the noise.

 DIFFERENTIAL CONFIGURATION (HYDROS)
In a differential configuration, the installation is calibrated in such a way that a noise pulse will be detected at a pair of Iris Power 80 pF PD sensors (EMCs) at the same time. The Iris Power PDA-IV can resolve the pulse arrival time from a pair of sensors to <6 ns. PD originating near one sensor (C1 or C2) will arrive a significant time later at the other sensor. The differential configuration is employed in large hydro generators that have enough room to install PD sensors along the circuit ring bus. An instrument with only one input per phase, will be not able to discriminate the pulses from a pair of PD sensors as either coming from the parallel C1, coming from parallel C2, or noise. Thus such an arrangement will mean the results will be corrupted by electrical noise, greatly increasing the probability of a false indication (and unnecessary testing and stator repairs being planned). Alternatively a human expert (presumably from the vendor) will be needed to interpret all results, rather than the utility performing and interpreting their own tests – greatly increasing test costs.

DIRECTIONAL CONFIGURATION (MOTORS AND TURBOGENERATORS)
For machines with a small bore diameter (< 2 m) such as motors and turbogenerators, a pair of Iris Power 80 pF PD sensors (EMCs) is located outside of the machine along the output bus, in a directional configuration. It is called a directional configuration because pulses coming from beyond the bus segment between the PD sensor pair, arrive first to one sensor and a delayed time after, arrive to the second sensor. The PD sensor closest to the machine is named M and the PD sensor further down the system is named S. PD pulses coming from the stator winding arrive at the M sensor first and are classified by an instrument called the Iris Power TGA-B as stator PD. Otherwise, pulses are classified as noise (pulses from beyond S, and pulses between the pair of PD sensors). This differentiation between stator PD and power system noise occurs because it takes time for electrical pulses to travel, and the TGA-B can resolve the arrival time differences as short as 6 ns. An instrument with only one input per phase, will be not able to discriminate the pulses from a pair of PD sensors as either PD or power system noise. Thus it will be prone to false indications, and unnecessary testing and repairs on the stator.

STATOR SLOT COUPLER (LARGE TURBOGENERATORS)
The SSC is a two port directional electromagnetic coupler that is installed in slots of large turbo generators, either under the wedges, or between the two bars in the slot. An instrument with only one input per phase, will be not able to discriminate the pulses from the two port coupler as either PD from the slot, PD from the endwinding, or noise.

Sensor configuration is necessary to reduce influence of disturbances

8 COLLECTION OF DATA
 PD TEST METHOD
During normal machine operation, the VHF instrument called the PDA-IV or TGA is temporarily connected to the previously installed sensors in each phase. The sensor blocks the power frequency voltage, and passes the high frequency voltage pulse accompanying partial discharge. To avoid any confusion with electrical noise from power tool operation, corona from the switchgear, RF sources, etc., the PDA-IV or TGA separates PD from system noise and disturbances on the basis of time-of-arrival and pulse characteristics, and measures the number, magnitude and ac phase position of the PD pulses.

 DATA PRESENTATION
Two types of plots are generated for each partial discharge test. The first type of plot is two-dimensional (2-D), where the number of partial discharges per second versus PD magnitude is displayed. The greater the number of pulses per second, the more widespread is the deterioration in the winding. The higher the PD magnitude, the more severe is the deterioration. The second type of plot is three-dimensional (3-D), where the quantity (vertical scale) and magnitude (scale coming out of the page) of the PD versus the ac phase angle (horizontal scale) are displayed. Experience has indicated that such pulse phase analysis can be used to identify if multiple deterioration mechanisms are occurring, and what the mechanisms are.

The 2-D and 3-D plots are unwieldy for making comparisons amongst the machines. The PDA-IV or TGA summarizes each plot with two quantities: the peak PD magnitude (Qm) and the total PD activity (NQN). The Qm is defined to be the magnitude corresponding to a PD repetition rate of 10 pulses per second. Qm relates to how severe the deterioration is in the worst spot of the winding, while the NQN is proportional to the total amount of deterioration and is similar to the power factor tip-up. Since the Qm scalar quantity is more indicative of how close the winding is to failure, the peak magnitude (Qm) will be used throughout this paper for comparisons.