The Case for Frequency Domain PD Testing
THE CASE FOR FREQUENCY DOMAIN PD TESTING IN THE CONTEXT OF DISTRIBUTION CABLE Steven Boggs Electrical Insulation Research Center Departments of Electrical Engineering and Physics University of Connecticut Department of Computer and Electrical Engineering University of Toronto [email protected]
INTRODUCTION When PD occurs in the dielectric between two electrodes, the electric field in the region of the partial breakdown is shorted out by the conductivity of the ionized gas (Figure 1). If the system is adiabatic, i.e., isolated from all power supplies at the time of the PD, the energy in the system after the PD will be lower than that before the PD, and the change in energy is deposited in the discharge to dissociate the gas. At the end of the discharge process, the gas is heavily ionized, and these ions drift in the electric field. In the case of cavity PD in a solid dielectric, the ions drift to the cavity walls and (nearly) cancel the field within the cavity. Thus if we imagine that we raise the voltage slowly on a cavity-containing system, the first PD is likely to occur near the peak voltage. At the end of that PD, the applied voltage will be near peak, but the electric field in the cavity will be near zero. Thus when the voltage swings to the opposite peak polarity, the field in the cavity will be twice that required to initiate PD, which is why the PD inception voltage is substantially greater than the PD extinction voltage. These phenomena are shown schematically in Figures 2. As noted above, the energy in an adiabatic system which suffers PD is reduced, which means that the voltage on the system is lower after the PD than before as a result of the PD shorting out part of the electric field which effectively increases the capacitance between the electrodes. The PD signal in the external circuit is generated by the power supply recharging the sample, as seen in Figure 1. PD pulses are typically very narrow at their source, in the range of 1 ns with a bandwidth of >300 MHz. However in general purpose PD detection apparatus, a relatively narrow band detection circuit is employed because lumped element circuits inevitably have a wide range of resonances which will be excited by such a wide band pulse. The detection circuit integrates these resonances into a signal with an amplitude which is proportional to the apparent charge in the PD pulse and relatively independent of resonances within the test object. In the context of PD in cable, the PD source, such as cavity discharge, is loaded by the characteristic impedance of the transmission line, so that the change in potential between the conductor and ground caused by the increase in
capacitance produced by the PD canceling part of the electric field is recharged through the impedance of the transmission line. Thus the signal magnitude can be predicted if the PD pulse waveform is known. For a 1.5 ns FWHM Gaussian PD pulse of 1 pC magnitude, the peak current will be about 0.6 mA which will be fed equally by the cable on each side of the PD site, so that the 0.3 mA will be loaded across about 35 Ohms (typical cable characteristic impedance) to produce a pulse of about 10 mV which propagates away from the PD site in both directions. As a result of the transmission line nature of the system, this pulse will be very narrow. However, typical distribution cable has substantial high frequency attenuation as a result of loss cause by the semiconducting layers [1] and dielectric (Figure 3). As a result, the pulse width will increase and the pulse amplitude will decrease as high frequency energy is lost to the semiconducting layers and, to a lesser degree, to the dielectric and skin effect in the conductors (Figure 4) [2-4]. As charge is conserved, the time integral of the pulse amplitude will remain approximately constant. This is possible as the energy is proportional to the square of the amplitude (voltage), but the charge is proportional to the time integral of the pulse amplitude. As a result of this high frequency attenuation, the optimum bandwidth in time domain PD measurements of distribution cable is typically about 20 MHz [1,4].
TIME DOMAIN PD TESTING Time domain PD testing of distribution cable is normally carried out by triggering a PD detector on the first pulse, the one which propagates from the PD source to the cable termination at which the PD detector is located. Once the first pulse is detected, data can be digitized for some microseconds, and sophisticated signal processing techniques can be employed to find the second pulse, which was reflected off the “far end” of the cable (Figure 5). Under most conditions, PD detection sensitivity is determined by the ability to trigger on the “first” pulse above the noise. This pulse must be detected with only rudimentary frequency domain filtering of major noise sources such as radio stations, etc. Sophisticated signal recognition techniques, such as wavelet analysis, cannot be used to enhance detection of the first pulse, as such techniques are far too computationally intensive to be applied to a continuous 20 MHz bandwidth data stream in real time. Thus the need to trigger on the first PD pulse is a major impediment to high sensitivity PD detection under noisy field conditions. Once a PD source is detected, location is based on the relative time of arrival of the first and second pulses. If a phase reference is available, phase resolved PD analysis can be undertaken to help identify the cause of the PD. However since time domain testing is usually carried out off-line at elevated voltage, the acceptable test time may limit the number of PD pulses which can be detected to below that which makes phase analysis very useful.
FREQUENCY DOMAIN PD TESTING Frequency domain testing has several major advantage over time domain testing, including: 1. PD can be detected, characterized, and located without having to trigger on the first pulse. 2. Since the frequency domain testing is usually carried out in service, and the PD is measured from various points (joints) along the cable, the cable between the point of detection and the cable termination acts as a high frequency filter which removes much of the noise which interferes with sensitive PD detection. 3. Since the PD detector is closer to the PD source and much of the interfering noise is filtered out by the cable, the bandwidth of the PD signal at the sensor can be used to judge location, and very sensitive PD detection is possible (Figure 6) 4. If the spectrum analyzer is triggered synchronously with the power frequency, the spectrum analyzer display becomes a phase/frequency fingerprint of the PD signal (Figure 7). 5. Because PD detection is undertaken in service, one can assume that any PD source which could be active is likely to be active. As explained above, if the cable is taken out of service to do off-line PD testing, the voltage must be raised to the range of 2 pu in order to assure that all PD sources which could be active at normal operating voltage will be active. Thus PD testing at 2 pu off-line is roughly equivalent to testing at normal operating voltage in service.
No Need to Trigger Off the “First” PD Pulse A conventional RF spectrum analyzer triggers a fixed bandwidth swept filter which then sweeps through a defined range of frequencies at a predefined frequency sweep rate. We assume that we trigger the frequency sweep synchronously with power frequency at the same point on cycle for each frequency sweep. The frequency spectrum can be accumulated over a large number of frequency sweeps. PD sources are typically strongly correlated with the power frequency phase. If such a PD source is measured with a spectrum analyzer in which the sweep takes several cycles, the PD signal will occur when the swept filter of the spectrum analyzer is at one of several frequencies which correspond to the phase at which the PD occurs (Figure 7). In other words, assume that we trigger the swept filter at positive zero crossing of the voltage, and the sweep takes 5 power frequency cycles. Further for illustrative purposes, we assume that the PD always occurs near negative zero crossing. Then the PD pulse could occur near any of the five negative zero crossings during the sweep of the filter. Thus the signal
which will appear on the spectrum analyzer is the PD pulse energy within the bandwidth of the swept filter at the frequency of the swept filter at the time of the PD pulse, as illustrated in Figure 7. Since we are sweeping continuously, we will have random PD pulses at the 5 frequencies which correspond to the phase of the PD pulse relative to the trigger of the swept filter. Thus the spectrum analyzer display represents a phase/frequency display of the pulse, as from the display, we can see the phases at which PD pulses occur, and we can see the frequency content of the PD pulses at several frequencies scattered over a broad range of frequency. The display is therefore rich in detail which can be used to characterize the PD pulse, by location based on frequency content and by type based on phase stability, frequency content, etc. Further, the filter can be swept at a wide range of rates and frequency ranges to tradeoff between phase resolved detail and frequency resolved detail. Most importantly, all of this detail is available without ever having to trigger on a PD pulse above broadband noise. Indeed, the noise spectrum can be measured and subtracted from the PD signal spectrum. This results in a PD detection sensitivity which is several orders of magnitude greater than for time domain detection from the cable terminals. This greatly increased sensitivity facilitates improved characterization of the cable condition through detection of signals which are far below the detection limit of the best time domain PD detection systems. If the PD signal is sufficiently large, time domain detection is also possible by operating the spectrum analyzer in the “zero span” mode, in which the frequency of the filter is fixed in a frequency region with little noise. This facilitates very sensitive time domain PD detection by eliminating most of the background noise.
PD Location Based on High Frequency Attenuation As noted above, cable has substantial high frequency attenuation (Figure 3,4). In frequency domain PD testing, this attenuation is used in two contexts. First, since PD is normally detected using capacitive/inductive couplers placed in manholes along the cable route, the cable between the PD coupler and the termination acts as a filter to remove high frequency noise. Further, since the PD coupler is relatively close to the PD source, the bandwidth of the PD signal is quite large and extends into the region from which noise was eliminated by the cable attenuation between the coupler and the termination [5]. This combination of filtering action by the cable and increased PD signal bandwidth improves PD detection sensitivity substantially at most locations along the cable.
Water Tree-Induced Degradation Water treeing in XLPE and HMWPE is the dominant mechanism of cable degradation and failure of distribution cable. Time domain PD detection can only detect water trees if an electrical tree has initiated, as water trees do not cause PD
[6]. But electrical trees generally grow to failure rapidly once initiated. Early time domain PD detection was carried out at 2.5 to 3.5 pu, voltage levels which may be effective in converting water trees to electrical trees but which also results in increased cable failure rates after the test. Today, time domain PD testing is usually carried out in the range of 2 to 2.5 pu where the test is less damaging but where water trees are much less likely to be detected. Thus time domain tests are generally incapable of providing a good characterization of the condition of water treed cable. Other test techniques, such as dielectric loss, various forms of dielectric spectroscopy, etc. can be used in conjunction with time domain PD testing to provide an indication of the overall condition of the entire length of cable tested but can provide no indication of which parts of the cable are “good” and which are “bad”. In service frequency domain testing is sufficiently sensitive to characterize the condition of the cable as measured at various manholes along the cable. Water trees cause very small reflections of high frequency signals propagating along the cable, and these many small reflections raise the noise background on the cable. This can be used to characterize the condition of the cable with respect to water treeing. As the effect is relatively local, the condition of various sections of the cable can be characterized relatively independently. The effectiveness of this approach has been verified by independent, double blind testing in which inservice PD testing was applied in an area of unknown condition on the part of the test provider who also did not know what would be done with the data that were provided. The cables identified as “bad” and “very bad” (the worst two of five categories) were removed and sent to an independent test lab where the AC strength was determined. The cables which were identified as “very bad” were substantially worse and older than any of several hundred cable sections which had been removed at random, and the cables identified as “bad” were as bad and as old as any which had been removed from the system at random. The AC breakdown strength of the “very bad” cables was in the range of 200 to 300 V/mil, and that of the “bad” cables was 200 to 400 V/mil, both very near the lower limit at which cables fail extremely rapidly in service (200 V/mil). This double blind test demonstrates the effectiveness in-service condition characterization using frequency domain detection.
Sensitivity for In-Service PD Detection Thus a final major advantage of frequency domain PD testing is that it is sensitive enough to be carried out in service. Since the cable is not taken out of service, one can assume that any PD source likely to be active during normal operation is active. Further, in-service PD testing imposes no risk to the integrity of the cable. A recent research study indicates that PD detection above 2 to 2.5 pu reduces substantially the subsequent PD inception voltage, which means that the PD test damaged the cable (Figure 8). On the other hand, off-line PD testing must be carried out at no less than 2 pu to assure that all PD sources which can be active during normal operation are likely to be initiated during the test. In fact, even
testing at 2 pu does not assure PD initiation, as initiation requires a free electron. The primary source of free electrons is cosmic rays, which generate about 3 3 3 electrons/cm -s in air. For a 1 mm cavity, the waiting time for a free electron would be in the range of 5 minutes, which is longer than the desirable test time at 2 pu. Thus even at a test voltage of 2 pu, PD may not be initiated in relatively small cavities during an off-line test, even if such cavities would be in active discharge during normal operation. In service testing has the advantage that the cable has been exposed to normal system disturbances for a long time, and if a PD source can be initiated it is likely to have been initiated.
SUMMARY In summary, frequency domain PD testing offers a rich set of PD characterization tools, the most basic of which have been described above. PD detection sensitivity under field conditions is one to two orders of magnitude greater than for time domain testing as a result of (i) no need to trigger on the first PD pulse above the broadband noise and (ii) the filtering effect of the cable between the PD detection site and the terminations. As a result of this greatly increased sensitivity and the rich set of characterization tools, frequency domain PD testing has been developed into a highly sensitive and reliable tool for characterizing the condition of distribution cable during normal operation, the sensitivity and accuracy of which have been confirmed through independent analysis.
ACKNOWLEDGEMENTS I am please to acknowledge useful discussions with Dr. Nezar Ahmed and Dr. Liming Zhou of DTE Energy. Dr. Zhou produced Figure 3 while a PDF at the University of Connecticut.
REFERENCES 1.
Boggs, S.A. and G.C. Stone. “Fundamental Limitations to the Measurement of Corona and Partial Discharge”. 1981 Annual Report of the Conference on Electrical Insulation and Dielectric Phenomena, reprinted in IEEE Trans. EI-17, April, 1982. p. 143.
2.
Stone, G.C. and S.A. Boggs. “Propagation of Partial Discharge Pulses in Shielded Power Cable”. 1982 Annual Report of the Conference on Electrical Insulation and Dielectric Phenomena, National Academy of Sciences, Washington, DC. p. 275-280.
3.
Zhou, L.M. and S.A. Boggs. “Effect of Shielded Distribution Cable on Very Fast Transients”. IEEE Trans PD-15, No. 3, July 2000. pp. 857-863.
4.
Boggs, S.A., A. Pathak, and P. Walker. “Partial Discharge XXII: High Frequency Attenuation in Shielded Solid Dielectric Power Cable and
Implications Thereof for PD Location”. IEEE Electrical Insulation Magazine, 12, January/February 1996. pp. 9-16. 5.
Ahmed, N., and N. Srinivas. “On-line Partial Discharge Diagnostic System in Power Cable System”. 2001 IEEE Transmission and Distribution Conference and Exposition, Atlanta, GA, 28 Oct. – 2 Nov. 2001. Vol. 2, pp. 853-858.
6.
Boggs, S.A. and R.J. Densley. “Fundamentals of Partial Discharge in the Context of Field Cable Testing”. IEEE Electrical Insulation Magazine, Vol. 16, No. 5, Sept/Oct 2000. pp. 13-18.
Steven Boggs was graduated from Reed College with a BA in Physics and from the University of Toronto with a Ph.D. in Physics and an MBA. His career in the power industry started with Ontario Hydro Research, where he spent 12 years working in the areas of soil thermal properties, SF6 insulated switchgear, and solid dielectrics. Steve was elected a Fellow of the IEEE for his contributions to SF6insulated substation technology. Steve spent 6 years as Director of Engineering and Research for Underground Systems, Inc., where he was also responsible for USi's capacitor subsidiary, Chicago Condenser Corporation. Steve wrote the software used for capacitor winding design, designed PD detection systems for capacitors, etc. Under EPRI sponsorship, he also made the first measurements of ac loss in high temperature superconducting conductor elements. With a USi colleague, he holds a patent on the room temperature dielectric HTSC cable design adopted by Pirelli and American Superconductor. Steve is presently Director of the Electrical Insulation Research Center at the University of Connecticut and Research Professor of Materials Science, Electrical Engineering and Physics. He is also an Adjunct Professor of Electrical Engineering at the University of Toronto. His work ranges from development of an on-line ultrasonic PD monitoring system for a large GIS to fundamental investigations of high field phenomena in dielectrics based on proprietary software for transient nonlinear finite element analysis with coupled electric and thermal fields.
Measure voltage across resistor
Measure voltage across coupling unit
Measure voltage across resistor
Measure voltage across coupling unit
Figure 1. Before the PD, the electric field penetrates the cavity as seen in the upper equipotential plot. The PD causes the gas in the cavity to become conducting, which shorts out the field in the cavity (lower equipotential plot), lowering the overall energy in the test object. The PD signal in the coupling unit is generated by recharging of the test object by the power supply.
After First Discharge
Before First Discharge
+
+ + + + +
+ +
-
-
-
-
-
-
Applied Voltage
Field in Cavity
Figure 2. After the first PD event, ions drift to the walls of a cavity, resulting in very little field within the cavity after the PD. However, the first PD is likely to occur near the peak AC voltage, so that as the voltage polarity reverses, the field in the cavity increases to twice that required to initiate PD. This is why the PDIV (PD initiation voltage) can be twice the PDEV (the PD extinction voltage) and why an off line test much be carried out at no less than 2 pu in the hope that most of the PD sites which could be active at normal operating voltage will be initiated.
Attenuation Constant (dB/m)
100
HF Meas. (Total Loss)
10-1
10-2
Skin effect + Diele. Loss
Skin Effect
10-3
LF Meas. (Diele. Loss)
10-4
10-5 0.01
0.10
1.00
10.00
Frequency (MHz)
Figure 3. Loss budget for PILC cable, which is relatively “ideal” as a result of its concentric lead sheath. The high frequency loss measurements (green dots) were made with a transmission line impedance analyzer which measures all cable losses. The lower frequency losses were made with a lumped element impedance analyzer which measures only dielectric losses. The skin effect losses were computed from the conductor and sheath conductivities. Both taped shield and concentric neutral cables are less “ideal”, in the former case because some of the ground current spirals down the tape and in the latter case because the cable is not fully shielded at high frequency.
100.00
0m
100 m 300 m
1000 m
Figure 4. Typical PD pulse waveform as a function of distance propagated in a distribution cable.
Termination Reflection off far end Splice
PD Source
PD pulse propagating away from coupler
Splice
PD pulse propagating toward coupler
Noise from termination
Figure 5. Typical configuration for time domain testing. The PD source shown near the center of the cable produces two pulses propagating in the cable, one which propagates toward the test termination (blue) and one which propagates in the opposite direction (green). The pulse which propagates toward the test termination (right) must be detected by triggering the digital signal acquisition system in the presence of noise entering the termination from the environment (red), which will include noise from spark plugs in cars, noise from radio stations, etc. The worst of stationary sources, such as radio stations, can be removed with selective notch filters; however, much environmental noise remains. If the trigger level is set too low, the false triggers will overload the detection circuitry. The ability to trigger the system on the “first” PD pulse in the presence of noise determines the overall PD detection sensitivity. Once the first pulse is detected, data can be acquired for some microseconds, and sophisticated signal processing can be applied to detect the second, reflected pulse, in noise. Such signal processing cannot be applied to the first pulse, as it is too computationally intensive to be applied continuously in real time.
Termination Coupler Splice
Coupler Splice
Noise from termination Figure 6. In frequency domain PD testing, the PD coupler is usually relatively near the PD source and far from the terminations. As a result, high frequency noise which enters through the terminations is attenuated by the cable (Figure 3), which reduces interference. Since the PD source is closer to the PD coupler, the pulse magnitude and pulse bandwidth are larger than if the PD pulse were detected at the termination, as for time domain testing. The combination of increased PD signal amplitude and bandwidth, combined with reduced high frequency noise from the filtering effect of the cable and no need to trigger on the PD pulse for PD detection result in increased PD detection sensitivity by several orders of magnitude relative to conventional PD detection at the termination.
Figure 7. The lower plot shows the AC power frequency (50 Hz) voltage over 100 ms (5 cycles) along with the spectrum analyzer filter center frequency as it sweeps from 0 to 500 MHz over the 100 ms. Thus if a PD occurs at the first negative zero crossing, the filter would be tuned to 50 MHz, and the signal amplitude on the spectrum analyzer would represent the PD signal energy within the bandwidth of the filter when the filter is tuned to 100 MHz. If a PD occurred at the same phase of the next cycle, the filter would be tuned to 150 MHz, and the signal amplitude on the spectrum analyzer would be the PD signal energy within the bandwidth of the filter when tuned to 150 MHz. The top graph shows the signal which would accumulate on the spectrum analyzer as a result of a PD source which discharges at around the negative zero crossing and is relative close to the PD coupler. The plot indicates that the signal energy is relatively constant to about 300 MHz, after which it falls off. This would indicate a PD pulse width of 1.5 to 2 ns, which would mean that the PD source was very close to the PD coupler, as little high frequency attenuation of the PD pulse has taken place.
30 Field Test
PDIV (kV)
25 20 15
Subsequent Lab Test
10 5 0 0
1
2
3
4
5
6
7
8
Sample Number
Figure 8. Data for PD inception voltage (PDIV) during a field PD test on distribution cable and the PDIV on a subsequent laboratory test after the cable was removed from the field. Note that as the field test voltage is raised above about 17 kV (2 pu for a 15 kV system), increasing damage is done to the cable, as evidenced by the reduction in PDIV on the subsequent test in the laboratory. Thus field PD testing above 2 pu appears to be impose significant risk to the cable.
INTRODUCTION When PD occurs in the dielectric between two electrodes, the electric field in the region of the partial breakdown is shorted out by the conductivity of the ionized gas (Figure 1). If the system is adiabatic, i.e., isolated from all power supplies at the time of the PD, the energy in the system after the PD will be lower than that before the PD, and the change in energy is deposited in the discharge to dissociate the gas. At the end of the discharge process, the gas is heavily ionized, and these ions drift in the electric field. In the case of cavity PD in a solid dielectric, the ions drift to the cavity walls and (nearly) cancel the field within the cavity. Thus if we imagine that we raise the voltage slowly on a cavity-containing system, the first PD is likely to occur near the peak voltage. At the end of that PD, the applied voltage will be near peak, but the electric field in the cavity will be near zero. Thus when the voltage swings to the opposite peak polarity, the field in the cavity will be twice that required to initiate PD, which is why the PD inception voltage is substantially greater than the PD extinction voltage. These phenomena are shown schematically in Figures 2. As noted above, the energy in an adiabatic system which suffers PD is reduced, which means that the voltage on the system is lower after the PD than before as a result of the PD shorting out part of the electric field which effectively increases the capacitance between the electrodes. The PD signal in the external circuit is generated by the power supply recharging the sample, as seen in Figure 1. PD pulses are typically very narrow at their source, in the range of 1 ns with a bandwidth of >300 MHz. However in general purpose PD detection apparatus, a relatively narrow band detection circuit is employed because lumped element circuits inevitably have a wide range of resonances which will be excited by such a wide band pulse. The detection circuit integrates these resonances into a signal with an amplitude which is proportional to the apparent charge in the PD pulse and relatively independent of resonances within the test object. In the context of PD in cable, the PD source, such as cavity discharge, is loaded by the characteristic impedance of the transmission line, so that the change in potential between the conductor and ground caused by the increase in
capacitance produced by the PD canceling part of the electric field is recharged through the impedance of the transmission line. Thus the signal magnitude can be predicted if the PD pulse waveform is known. For a 1.5 ns FWHM Gaussian PD pulse of 1 pC magnitude, the peak current will be about 0.6 mA which will be fed equally by the cable on each side of the PD site, so that the 0.3 mA will be loaded across about 35 Ohms (typical cable characteristic impedance) to produce a pulse of about 10 mV which propagates away from the PD site in both directions. As a result of the transmission line nature of the system, this pulse will be very narrow. However, typical distribution cable has substantial high frequency attenuation as a result of loss cause by the semiconducting layers [1] and dielectric (Figure 3). As a result, the pulse width will increase and the pulse amplitude will decrease as high frequency energy is lost to the semiconducting layers and, to a lesser degree, to the dielectric and skin effect in the conductors (Figure 4) [2-4]. As charge is conserved, the time integral of the pulse amplitude will remain approximately constant. This is possible as the energy is proportional to the square of the amplitude (voltage), but the charge is proportional to the time integral of the pulse amplitude. As a result of this high frequency attenuation, the optimum bandwidth in time domain PD measurements of distribution cable is typically about 20 MHz [1,4].
TIME DOMAIN PD TESTING Time domain PD testing of distribution cable is normally carried out by triggering a PD detector on the first pulse, the one which propagates from the PD source to the cable termination at which the PD detector is located. Once the first pulse is detected, data can be digitized for some microseconds, and sophisticated signal processing techniques can be employed to find the second pulse, which was reflected off the “far end” of the cable (Figure 5). Under most conditions, PD detection sensitivity is determined by the ability to trigger on the “first” pulse above the noise. This pulse must be detected with only rudimentary frequency domain filtering of major noise sources such as radio stations, etc. Sophisticated signal recognition techniques, such as wavelet analysis, cannot be used to enhance detection of the first pulse, as such techniques are far too computationally intensive to be applied to a continuous 20 MHz bandwidth data stream in real time. Thus the need to trigger on the first PD pulse is a major impediment to high sensitivity PD detection under noisy field conditions. Once a PD source is detected, location is based on the relative time of arrival of the first and second pulses. If a phase reference is available, phase resolved PD analysis can be undertaken to help identify the cause of the PD. However since time domain testing is usually carried out off-line at elevated voltage, the acceptable test time may limit the number of PD pulses which can be detected to below that which makes phase analysis very useful.
FREQUENCY DOMAIN PD TESTING Frequency domain testing has several major advantage over time domain testing, including: 1. PD can be detected, characterized, and located without having to trigger on the first pulse. 2. Since the frequency domain testing is usually carried out in service, and the PD is measured from various points (joints) along the cable, the cable between the point of detection and the cable termination acts as a high frequency filter which removes much of the noise which interferes with sensitive PD detection. 3. Since the PD detector is closer to the PD source and much of the interfering noise is filtered out by the cable, the bandwidth of the PD signal at the sensor can be used to judge location, and very sensitive PD detection is possible (Figure 6) 4. If the spectrum analyzer is triggered synchronously with the power frequency, the spectrum analyzer display becomes a phase/frequency fingerprint of the PD signal (Figure 7). 5. Because PD detection is undertaken in service, one can assume that any PD source which could be active is likely to be active. As explained above, if the cable is taken out of service to do off-line PD testing, the voltage must be raised to the range of 2 pu in order to assure that all PD sources which could be active at normal operating voltage will be active. Thus PD testing at 2 pu off-line is roughly equivalent to testing at normal operating voltage in service.
No Need to Trigger Off the “First” PD Pulse A conventional RF spectrum analyzer triggers a fixed bandwidth swept filter which then sweeps through a defined range of frequencies at a predefined frequency sweep rate. We assume that we trigger the frequency sweep synchronously with power frequency at the same point on cycle for each frequency sweep. The frequency spectrum can be accumulated over a large number of frequency sweeps. PD sources are typically strongly correlated with the power frequency phase. If such a PD source is measured with a spectrum analyzer in which the sweep takes several cycles, the PD signal will occur when the swept filter of the spectrum analyzer is at one of several frequencies which correspond to the phase at which the PD occurs (Figure 7). In other words, assume that we trigger the swept filter at positive zero crossing of the voltage, and the sweep takes 5 power frequency cycles. Further for illustrative purposes, we assume that the PD always occurs near negative zero crossing. Then the PD pulse could occur near any of the five negative zero crossings during the sweep of the filter. Thus the signal
which will appear on the spectrum analyzer is the PD pulse energy within the bandwidth of the swept filter at the frequency of the swept filter at the time of the PD pulse, as illustrated in Figure 7. Since we are sweeping continuously, we will have random PD pulses at the 5 frequencies which correspond to the phase of the PD pulse relative to the trigger of the swept filter. Thus the spectrum analyzer display represents a phase/frequency display of the pulse, as from the display, we can see the phases at which PD pulses occur, and we can see the frequency content of the PD pulses at several frequencies scattered over a broad range of frequency. The display is therefore rich in detail which can be used to characterize the PD pulse, by location based on frequency content and by type based on phase stability, frequency content, etc. Further, the filter can be swept at a wide range of rates and frequency ranges to tradeoff between phase resolved detail and frequency resolved detail. Most importantly, all of this detail is available without ever having to trigger on a PD pulse above broadband noise. Indeed, the noise spectrum can be measured and subtracted from the PD signal spectrum. This results in a PD detection sensitivity which is several orders of magnitude greater than for time domain detection from the cable terminals. This greatly increased sensitivity facilitates improved characterization of the cable condition through detection of signals which are far below the detection limit of the best time domain PD detection systems. If the PD signal is sufficiently large, time domain detection is also possible by operating the spectrum analyzer in the “zero span” mode, in which the frequency of the filter is fixed in a frequency region with little noise. This facilitates very sensitive time domain PD detection by eliminating most of the background noise.
PD Location Based on High Frequency Attenuation As noted above, cable has substantial high frequency attenuation (Figure 3,4). In frequency domain PD testing, this attenuation is used in two contexts. First, since PD is normally detected using capacitive/inductive couplers placed in manholes along the cable route, the cable between the PD coupler and the termination acts as a filter to remove high frequency noise. Further, since the PD coupler is relatively close to the PD source, the bandwidth of the PD signal is quite large and extends into the region from which noise was eliminated by the cable attenuation between the coupler and the termination [5]. This combination of filtering action by the cable and increased PD signal bandwidth improves PD detection sensitivity substantially at most locations along the cable.
Water Tree-Induced Degradation Water treeing in XLPE and HMWPE is the dominant mechanism of cable degradation and failure of distribution cable. Time domain PD detection can only detect water trees if an electrical tree has initiated, as water trees do not cause PD
[6]. But electrical trees generally grow to failure rapidly once initiated. Early time domain PD detection was carried out at 2.5 to 3.5 pu, voltage levels which may be effective in converting water trees to electrical trees but which also results in increased cable failure rates after the test. Today, time domain PD testing is usually carried out in the range of 2 to 2.5 pu where the test is less damaging but where water trees are much less likely to be detected. Thus time domain tests are generally incapable of providing a good characterization of the condition of water treed cable. Other test techniques, such as dielectric loss, various forms of dielectric spectroscopy, etc. can be used in conjunction with time domain PD testing to provide an indication of the overall condition of the entire length of cable tested but can provide no indication of which parts of the cable are “good” and which are “bad”. In service frequency domain testing is sufficiently sensitive to characterize the condition of the cable as measured at various manholes along the cable. Water trees cause very small reflections of high frequency signals propagating along the cable, and these many small reflections raise the noise background on the cable. This can be used to characterize the condition of the cable with respect to water treeing. As the effect is relatively local, the condition of various sections of the cable can be characterized relatively independently. The effectiveness of this approach has been verified by independent, double blind testing in which inservice PD testing was applied in an area of unknown condition on the part of the test provider who also did not know what would be done with the data that were provided. The cables identified as “bad” and “very bad” (the worst two of five categories) were removed and sent to an independent test lab where the AC strength was determined. The cables which were identified as “very bad” were substantially worse and older than any of several hundred cable sections which had been removed at random, and the cables identified as “bad” were as bad and as old as any which had been removed from the system at random. The AC breakdown strength of the “very bad” cables was in the range of 200 to 300 V/mil, and that of the “bad” cables was 200 to 400 V/mil, both very near the lower limit at which cables fail extremely rapidly in service (200 V/mil). This double blind test demonstrates the effectiveness in-service condition characterization using frequency domain detection.
Sensitivity for In-Service PD Detection Thus a final major advantage of frequency domain PD testing is that it is sensitive enough to be carried out in service. Since the cable is not taken out of service, one can assume that any PD source likely to be active during normal operation is active. Further, in-service PD testing imposes no risk to the integrity of the cable. A recent research study indicates that PD detection above 2 to 2.5 pu reduces substantially the subsequent PD inception voltage, which means that the PD test damaged the cable (Figure 8). On the other hand, off-line PD testing must be carried out at no less than 2 pu to assure that all PD sources which can be active during normal operation are likely to be initiated during the test. In fact, even
testing at 2 pu does not assure PD initiation, as initiation requires a free electron. The primary source of free electrons is cosmic rays, which generate about 3 3 3 electrons/cm -s in air. For a 1 mm cavity, the waiting time for a free electron would be in the range of 5 minutes, which is longer than the desirable test time at 2 pu. Thus even at a test voltage of 2 pu, PD may not be initiated in relatively small cavities during an off-line test, even if such cavities would be in active discharge during normal operation. In service testing has the advantage that the cable has been exposed to normal system disturbances for a long time, and if a PD source can be initiated it is likely to have been initiated.
SUMMARY In summary, frequency domain PD testing offers a rich set of PD characterization tools, the most basic of which have been described above. PD detection sensitivity under field conditions is one to two orders of magnitude greater than for time domain testing as a result of (i) no need to trigger on the first PD pulse above the broadband noise and (ii) the filtering effect of the cable between the PD detection site and the terminations. As a result of this greatly increased sensitivity and the rich set of characterization tools, frequency domain PD testing has been developed into a highly sensitive and reliable tool for characterizing the condition of distribution cable during normal operation, the sensitivity and accuracy of which have been confirmed through independent analysis.
ACKNOWLEDGEMENTS I am please to acknowledge useful discussions with Dr. Nezar Ahmed and Dr. Liming Zhou of DTE Energy. Dr. Zhou produced Figure 3 while a PDF at the University of Connecticut.
REFERENCES 1.
Boggs, S.A. and G.C. Stone. “Fundamental Limitations to the Measurement of Corona and Partial Discharge”. 1981 Annual Report of the Conference on Electrical Insulation and Dielectric Phenomena, reprinted in IEEE Trans. EI-17, April, 1982. p. 143.
2.
Stone, G.C. and S.A. Boggs. “Propagation of Partial Discharge Pulses in Shielded Power Cable”. 1982 Annual Report of the Conference on Electrical Insulation and Dielectric Phenomena, National Academy of Sciences, Washington, DC. p. 275-280.
3.
Zhou, L.M. and S.A. Boggs. “Effect of Shielded Distribution Cable on Very Fast Transients”. IEEE Trans PD-15, No. 3, July 2000. pp. 857-863.
4.
Boggs, S.A., A. Pathak, and P. Walker. “Partial Discharge XXII: High Frequency Attenuation in Shielded Solid Dielectric Power Cable and
Implications Thereof for PD Location”. IEEE Electrical Insulation Magazine, 12, January/February 1996. pp. 9-16. 5.
Ahmed, N., and N. Srinivas. “On-line Partial Discharge Diagnostic System in Power Cable System”. 2001 IEEE Transmission and Distribution Conference and Exposition, Atlanta, GA, 28 Oct. – 2 Nov. 2001. Vol. 2, pp. 853-858.
6.
Boggs, S.A. and R.J. Densley. “Fundamentals of Partial Discharge in the Context of Field Cable Testing”. IEEE Electrical Insulation Magazine, Vol. 16, No. 5, Sept/Oct 2000. pp. 13-18.
Steven Boggs was graduated from Reed College with a BA in Physics and from the University of Toronto with a Ph.D. in Physics and an MBA. His career in the power industry started with Ontario Hydro Research, where he spent 12 years working in the areas of soil thermal properties, SF6 insulated switchgear, and solid dielectrics. Steve was elected a Fellow of the IEEE for his contributions to SF6insulated substation technology. Steve spent 6 years as Director of Engineering and Research for Underground Systems, Inc., where he was also responsible for USi's capacitor subsidiary, Chicago Condenser Corporation. Steve wrote the software used for capacitor winding design, designed PD detection systems for capacitors, etc. Under EPRI sponsorship, he also made the first measurements of ac loss in high temperature superconducting conductor elements. With a USi colleague, he holds a patent on the room temperature dielectric HTSC cable design adopted by Pirelli and American Superconductor. Steve is presently Director of the Electrical Insulation Research Center at the University of Connecticut and Research Professor of Materials Science, Electrical Engineering and Physics. He is also an Adjunct Professor of Electrical Engineering at the University of Toronto. His work ranges from development of an on-line ultrasonic PD monitoring system for a large GIS to fundamental investigations of high field phenomena in dielectrics based on proprietary software for transient nonlinear finite element analysis with coupled electric and thermal fields.
Measure voltage across resistor
Measure voltage across coupling unit
Measure voltage across resistor
Measure voltage across coupling unit
Figure 1. Before the PD, the electric field penetrates the cavity as seen in the upper equipotential plot. The PD causes the gas in the cavity to become conducting, which shorts out the field in the cavity (lower equipotential plot), lowering the overall energy in the test object. The PD signal in the coupling unit is generated by recharging of the test object by the power supply.
After First Discharge
Before First Discharge
+
+ + + + +
+ +
-
-
-
-
-
-
Applied Voltage
Field in Cavity
Figure 2. After the first PD event, ions drift to the walls of a cavity, resulting in very little field within the cavity after the PD. However, the first PD is likely to occur near the peak AC voltage, so that as the voltage polarity reverses, the field in the cavity increases to twice that required to initiate PD. This is why the PDIV (PD initiation voltage) can be twice the PDEV (the PD extinction voltage) and why an off line test much be carried out at no less than 2 pu in the hope that most of the PD sites which could be active at normal operating voltage will be initiated.
Attenuation Constant (dB/m)
100
HF Meas. (Total Loss)
10-1
10-2
Skin effect + Diele. Loss
Skin Effect
10-3
LF Meas. (Diele. Loss)
10-4
10-5 0.01
0.10
1.00
10.00
Frequency (MHz)
Figure 3. Loss budget for PILC cable, which is relatively “ideal” as a result of its concentric lead sheath. The high frequency loss measurements (green dots) were made with a transmission line impedance analyzer which measures all cable losses. The lower frequency losses were made with a lumped element impedance analyzer which measures only dielectric losses. The skin effect losses were computed from the conductor and sheath conductivities. Both taped shield and concentric neutral cables are less “ideal”, in the former case because some of the ground current spirals down the tape and in the latter case because the cable is not fully shielded at high frequency.
100.00
0m
100 m 300 m
1000 m
Figure 4. Typical PD pulse waveform as a function of distance propagated in a distribution cable.
Termination Reflection off far end Splice
PD Source
PD pulse propagating away from coupler
Splice
PD pulse propagating toward coupler
Noise from termination
Figure 5. Typical configuration for time domain testing. The PD source shown near the center of the cable produces two pulses propagating in the cable, one which propagates toward the test termination (blue) and one which propagates in the opposite direction (green). The pulse which propagates toward the test termination (right) must be detected by triggering the digital signal acquisition system in the presence of noise entering the termination from the environment (red), which will include noise from spark plugs in cars, noise from radio stations, etc. The worst of stationary sources, such as radio stations, can be removed with selective notch filters; however, much environmental noise remains. If the trigger level is set too low, the false triggers will overload the detection circuitry. The ability to trigger the system on the “first” PD pulse in the presence of noise determines the overall PD detection sensitivity. Once the first pulse is detected, data can be acquired for some microseconds, and sophisticated signal processing can be applied to detect the second, reflected pulse, in noise. Such signal processing cannot be applied to the first pulse, as it is too computationally intensive to be applied continuously in real time.
Termination Coupler Splice
Coupler Splice
Noise from termination Figure 6. In frequency domain PD testing, the PD coupler is usually relatively near the PD source and far from the terminations. As a result, high frequency noise which enters through the terminations is attenuated by the cable (Figure 3), which reduces interference. Since the PD source is closer to the PD coupler, the pulse magnitude and pulse bandwidth are larger than if the PD pulse were detected at the termination, as for time domain testing. The combination of increased PD signal amplitude and bandwidth, combined with reduced high frequency noise from the filtering effect of the cable and no need to trigger on the PD pulse for PD detection result in increased PD detection sensitivity by several orders of magnitude relative to conventional PD detection at the termination.
Figure 7. The lower plot shows the AC power frequency (50 Hz) voltage over 100 ms (5 cycles) along with the spectrum analyzer filter center frequency as it sweeps from 0 to 500 MHz over the 100 ms. Thus if a PD occurs at the first negative zero crossing, the filter would be tuned to 50 MHz, and the signal amplitude on the spectrum analyzer would represent the PD signal energy within the bandwidth of the filter when the filter is tuned to 100 MHz. If a PD occurred at the same phase of the next cycle, the filter would be tuned to 150 MHz, and the signal amplitude on the spectrum analyzer would be the PD signal energy within the bandwidth of the filter when tuned to 150 MHz. The top graph shows the signal which would accumulate on the spectrum analyzer as a result of a PD source which discharges at around the negative zero crossing and is relative close to the PD coupler. The plot indicates that the signal energy is relatively constant to about 300 MHz, after which it falls off. This would indicate a PD pulse width of 1.5 to 2 ns, which would mean that the PD source was very close to the PD coupler, as little high frequency attenuation of the PD pulse has taken place.
30 Field Test
PDIV (kV)
25 20 15
Subsequent Lab Test
10 5 0 0
1
2
3
4
5
6
7
8
Sample Number
Figure 8. Data for PD inception voltage (PDIV) during a field PD test on distribution cable and the PDIV on a subsequent laboratory test after the cable was removed from the field. Note that as the field test voltage is raised above about 17 kV (2 pu for a 15 kV system), increasing damage is done to the cable, as evidenced by the reduction in PDIV on the subsequent test in the laboratory. Thus field PD testing above 2 pu appears to be impose significant risk to the cable.
