SENSITIVITY ASPECTS OF ON-LINE PD DIAGNOSIS OF MV POWER CABLES

Transcription

1 SENSITIVITY ASPECTS OF ON-LINE PD DIAGNOSIS OF MV POWER CABLES Frank WESTER, Edward GULSKI, Johan SMIT, Edwin GROOT*, Mark VAN VLIET* Delft University of Technology The Netherlands * NUON The Netherlands f.j.wester@its.tudelft.nl INTRODUCTION For the determination of the insulation condition of the distribution power cable grid, several on-site diagnostic measuring methods have been developed [1-3]. Most of the developed techniques work off-line, which means that during the test the cable section has to be disconnected from the network and energised by an external voltage source, e.g. 50 Hz AC, VLF or DAC energising. In that way, applying to the cable sample AC test voltages between PD inception voltage (PDIV) and 2U 0 the discharge behaviour of cable insulation can be analysed to obtain most complete information about the insulation condition. Furthermore, on-line diagnostics for power cables have been introduced during the last years [4-8]. It is known that in contrast to an off-line diagnosis, on-line PD detection can only be performed at the service voltage U 0. In that way, the operational PD behaviour is captured and evaluated for diagnostic purposes. From economical point of view, the network switching costs which accompanies applying off-line diagnosis is important factor in using diagnostics. Therefore, to obtain time and cost reduction in applying advanced on-site PD diagnostics to a large number of different cable sections, a certain preselection is necessary. This pre-selection could be performed in different ways, e.g. using failure history, position in the network or based on an on-line diagnosis. On-line PD measurement systems are well known systems for other high voltage equipment then power cables (e.g. GIS, generators and transformers) [9-11]. Over time, they have proven their effectiveness and usefulness. However, for power cables several external factors may be of influence on the test results, as is described in this paper: distribution power cable-systems are complex systems consisting of different cable accessories and cable parts; on-line energising may be accompanied by external disturbances; the electrical (only U 0 ) and thermal (actual) load condition may influence the discharging defects. In particular, to support the off-line diagnosis by prioritising particular cable sections on the one hand, and after an onsite inspection to trend the discharge process in a suspicious cable section, on-line PD diagnostics is becoming increasingly interesting. Unfortunately, in contrast to off-line PD diagnosis, the sensitivity of on-line PD detection as well as the interpretation of measuring results is strongly dependent on local external disturbances and operation conditions of the cable section. ON-LINE VHF PD DETECTION Some of the available on-line PD diagnostics detect defects directly on the outside of the cable system [5,6], with inductive/capacitive sensors. In this way, PD signals can be detected in the frequency range up to several hundreds of MHz. However, if the cables are not situated in cable ducts but are directly buried underground, the cable and its joints are not accessible. Another way of detecting PD on-line in power cables is to connect the sensor at the earth connection [4,7,8] at the end of a cable section in one of the substations. Due to the applied materials (paper/oil or XLPE) in power cables, the UHF signals will be attenuated very fast during their propagation to the sensor [9], and only the signals in the HF range can be detected at the end of a cable section. The on-line PD detection system as described in this paper uses a HF inductive sensor at the cable end. In contrast to off-line PD detection systems, on-line detection systems don t use an external power supply to energise the cable sample, but use the line power of the network. Active defects will discharge, and the PD pulses will propagate (as TEM waves [9]) trough the cable. At the end of a cable section, in a substation, the earth screen is stopped just before the cable is connected to the switchgear. At this point the earth screen is connected to the substation earth, resulting PD pulses to couple out, where they can be detected. In order to detect PD sources in power cable sections on-line, a VHF detection circuit can be used [10,11]. As compared to IEC related detection methods, using the VHF detection method has a more complex character. Standard calibration is pc is not possible, whereas PD is detected in µv. The PD sensor, a high frequency split-core current transformer is attached around the earth connection of the cable section, figure 1. This 5 MHz HFCT is connected to a spectrum analyser (SA), which is connected to a computer for controlling and data processing. If PD is active in the Figure 1: On-line diagnostic set-up as applied for field measurements. A HF sensor is attached to the earth-connection of the cable sample The detected PD signals are measured by the SA and processed by the computer. TUD_Fwester_B1 Session 1 Paper No

2 a c Figure 2: On-line and off-line measurement results obtained from a PILC sample with an internal defect in the insulation, using respectively VHF PD detection and conventional PD detection; a detected signal spectrum and noise spectrum, by tuning to a centre frequency with high signal to noise ratio a phase resolved pattern can be obtained; b on-line phase resolved PD pattern measured on the earth screen (phase angle is related to mains voltage of the SA); c off-line phase resolved PD pattern measure at conductor (phase angle is related to the actual test voltage). diagnosed cable, PD pulses will pass the HF sensor and will be detected by the SA. Using spectrum analysis [9,10], the distinction between external disturbances and internal PD from cables are obtained. In particular, by tuning the SA to frequency ranges with a high signal to noise ratio, typical phase-resolved PD pattern are measured and analysed for the presence of internal discharges. For this purpose, the SA has to be set to the zerospan mode on a certain centre frequency (CF), and trigger to the 50 Hz power supply. As a result, phase-resolved PD patterns, known from off-line PD detection and recognition, are obtained, see figure 2. b Figure 3: Different influencing aspects for the sensitivity of VHF on-line PD detection of MV power cables. In the laboratory, investigations have been performed on different cable samples with known defects. In figures 2, laboratory results obtained from a short length PILC sample is shown. The phase resolved PD patterns of the on-line measurements are good comparable to those obtained from conventional off-line techniques. The only real difference between the two patterns in figure 2b and c is the phase shift of the PD pattern. This difference is caused by the fact that the triggering during the on-line measurements is related to the mains voltage instead of the actual test voltage, which were not synchronous during the measurement. From the laboratory investigations, with regard to VHF PD detection the following conclusions can be drawn: 1. The VHF PD method can detect PD activity in power cables (in both PILC and XLPE cables), through measuring the spectrum of the signals in the earth shield. 2. PD activity detected using HFCT is accompanied by an increase in the signal spectrum in the frequency range from about 10 khz to 10 MHz (especially between 1 MHz to 5 MHz); 3. The phase resolved PD patterns could be used to check whether a suspicious increase of the measured spectrum relatively to the noise spectrum, is caused by PD activity or by external noise. SENSITIVITY ASPECTS FOR ON-LINE DIAGNOSTIC Although on-line PD diagnostics are ideal for low cost preselection of power cables containing PD related defects, it should be kept in mind that some relevant sensitivity aspects influence the applicability of on-line diagnostics, see figure 3. As a result, inaccurate results may be detected in the field, and in some cases measurements are not possible at all. Furthermore, on-line PD systems do not fulfil the IEC recommendations about PD detection, amplification and calibration, which is necessary to detect discharges in pc. As a result, the amplitude of measured PD activity is represented by the on-line system, in µv values. This implies that the usual interpretation of PD amplitudes in [pc/nc] cannot be performed. Sensitivity aspect: Cable load TUD_Fwester_B1 Session 1 Paper No

3 Figure 4: Example of the influence of the cable load on internal PD activity, as detected on a feeding cable for a factory plant. The weekends and free days are visible in the course of the graph. This measurement implies that the time of on-line measurements during the day is important. Measurement results obtained with on-line PD detection system as described in [8]. Figure 6: External disturbances occurring in the same frequency range as PD, can have a tremendous effect on an on-line PD measurement results. This example shows external disturbances, obtained during a measurement near a rectification installation for the power supply of the tram in Amsterdam. The HF pulses of this switched power supply are clearly visible. Figure 5: Dependent on the switchgear installation type, the earth connection of the power cable can be easily and safely accessible for the PD sensor (left). In other installations, attachment of the sensor is of no use because the PD will pass trough the sensor at all (right & middle). The moment of on-line measurements can be of great influence. PD activity may changes with the power load of the cable sample. Figure 4 shows the measurement result from a three weeklong on-line PD measurement on a feeding power cable for a factory plant. Clearly visible is that the detected PD activity is changing with the daily load curve (weekends and free days are visible) of the cable sample. The course of the reflected graph implies that different measurement results may be obtained, dependent on the specific load of the power cable at the moment of measuring. Therefore, especially for cable sections with strongly varying power loads, the time of measuring is important. Sensitivity aspect: Installation type For optimal on-line PD measurements, the HF sensor should be attached as close as possible to the earth connection of the power cable, which connects the cable earth screen to the central station earth. If the HF sensor is connected too far away from the cable end, PD pulses will be attenuated before they can be detected. Obviously, for on-line PD measurements in the field, the earth connection of the cable should be easily and safely accessible around the switchgear it is connected to, see figure 5 (left side). However, it may occur for some installation types, that the cable earth screen is directly attached to the metallic box of the switchgear. As a result, the PD pulses propagate into the switchgear. Figure 7: On-line PD measurement on three-core distribution cables may result in multiple PD sources from different phases. As a result, the obtained phase resolved PD patterns might be difficult to discriminate from noise. Sensitivity aspect: External disturbances External disturbances present during on-line PD detection can be 50Hz related or non-50hz related noise, as they may occur in the same frequency range of the PD discharges for the SA. 50Hz related background noise (e.g. due to control pulses of a switched power supply) are in the same frequency range as the signal spectra of PD pulses, see figure 6. Due to the external signals, it is possible that PD activity may not be recognised (detected) in the phase-resolved pattern. Not-50Hz related background noise (e.g. due to coupled in radio signals or other audio signals) might effect that a possible phase resolved pattern from internal PD activity is overshadowed and therefore not recognisable. Sensitivity aspect: Multiple PD sources Most of the distribution power cable network consists of three core power cables, often paper insulated. As a result of the layered insulation of PILC, there is a probability that more than one small defect in a cable section are active. If these defects are than also related to more than one phase, the recognition of PD patterns will become more complex. The measurement result may show multiple PD patterns with phase shifts of less than 180 degrees, as is shown in figure 7. In this example, the PD patterns related to the three phases, indicated by A, B and C are distinguishable. Furthermore, the obtained PD patterns from multiple defects will be more difficult to discriminate from typical noise patterns. TUD_Fwester_B1 Session 1 Paper No

4 Figure 8: Phase-resolved on-line PD patterns (top) obtained from both sides of a 1974m long 10 kv PILC with a PD source in the cable joint at 1107 m, as is determined with off-line OWTS diagnosis (below). A PD pattern is just recognisable (as a result of the low noise level) at the cable end, which is the closest to the PD source. The attenuation effect of the cable length on the detected PD pattern levels on both cable sides. Sensitivity aspect: Signal attenuation It is known from PD detection in transformers and generators that in comparison to the overall sensitivity of IEC60270 detection (range of hundreds of khz) increasing the frequency range of detection systems enhances the local (close to the sensor) sensitivity and decreases the overall sensitivity. Signal attenuation effects in case of VHF detection on power cables are strongly visible in the field. Figure 8 shows the measurement result from a 1974m long PILC section, with a PD source at 1107m, as known from off-line PD detection and localisation with OWTS [13]. A PD pattern is just recognisable due to the low noise level in the measurement at the cable end closest (900m) from the PD source. In the measurement results at the other cable end no PD pattern is detected. The difference in propagation path from the PD source to both cable ends is only 200m of cable and one joint, but still it has a big effect on the measuring results. In addition to the results as described above, the influence of the propagation path was investigated. Different lengths of power cable of various conductor thicknesses and multiple types of joints are used to find the effect of the different components on the propagation. These signal attenuation measurements are performed by injection of standard calibration PD pulses of various amplitudes in one cable end of a cable section. The PD pulses are injected between one HV conductor and the earth screen, while the other two phases were floating. At the opposite cable end, the PD pulses are detected with the VHF detection circuit after a background noise measurement was performed. For noise independent comparison, the signal to noise spectra were studied for a fixed frequency span of 100kHz-10MHz. Next, the energy content of the different S/N spectra was calculated according to [12], in order to capture Figure 9: Signal attenuation measurement results for power cables of different length and cross section, containing various types of cable joints. The energy content of the different S/N spectra, after injection of 20nC, is calculated for independent comparison between the obtained results from the different cable sections. the spectra into a single value for mutual comparison. The obtained results for the 12 cable sections are reflected in figure 9. The injected artificial PD pulse amplitude in this example was 20nC. The lower threshold in the graph reflects the energy content of the average noise level. Below this value, no PD activity can be distinguished from the frequency spectrum. It can be concluded that the attenuation of PD signals for online measurements is influenced, in order of importance, by: the cable cross-section (conductor thickness); the length of the cable section; number of joints in the propagation path. From the investigations on the signal attenuation, it is shown that the cable cross-sections cause the major influence on the obtained measurement results. As shown in figure 9, cable sections of almost equal length, but different conductor thickness, give various measurement results. The trend lines show that cable sections with a large conductor thickness (240mm 2 or 150m 2 ) generally affect the PD signals considerably more as compared to smaller cross-sections (95mm 2 or 50mm 2 ). The influence of the cable length on the detection sensitivity is clearly visible in figure 9. For longer cable sections, the signal attenuation is generally higher, as the trend-lines show. For cable lengths longer than one kilometre, detection of PD of 20nC is hardly possible. Furthermore, the number of joints in the propagation path has influence to the signal attenuation, however the influence is unambiguous. The groups of joints in the graph, where the diversity in the length of the cable samples is small, show the effect of the joints. The number of joints in those cable sections has no clear contribution to the PD signal attenuation. TUD_Fwester_B1 Session 1 Paper No

5 a b c d Figure 10: Categorisation of on-line PD diagnostics for field application. For disturbances category I and or II, by off-line the PD presence diagnostics of is PD recommended related to for multiple in dept analysis of the insulation condition of the cable section; phases. If the noise level is low, and there is no detectable PD a Cat I: clear PD pattern, active PD source at service voltage; b Cat I: clear PD pattern, although the presence of disturbances; pattern, this indicates that no PD source is active during c Cat II: doubtful PD pattern (affected by external noise, or due to propagation service of distance), the power possible cable. active PD source during operation; d Cat III: no PD pattern, if the cable is within the length limitations of figure 10, no active PD source present in the cable section. PRACTICAL APPLICATION Furthermore, during the on-line measurements, it has to be kept in mind what the cable s cross-section is. As investigated in figure 9, the maximum length of a cable section to be sensitively detected may vary for different cable sections. Artificial PD pulses of 20nC can be detected over a distance of ca one kilometre. However, the same investigation showed that PD pulses of 5nC are attenuated to the lower threshold of figure9 after averagely 750m of cable section with large cross-sections. The approximate limitations of cable section length to be detected sensitive enough are reflected in table 1. After a feasibility study in the laboratory, where the VHF method was compared to conventional 50Hz PD detection and OWTS PD detection, the VHF PD detection method has proved to be effective to identify the presence of PD sources in power cables. Furthermore, several influencing factors are indicated, which affect the sensitivity and possibility of the on/line measurements. More than 70 on-line measurements are performed, to prove the effectiveness of this diagnostic as a pre-selective tool for off-line diagnosis. However, using offline diagnostics, the diagnostic information is much more extensive and accurate (PD inception voltage, PD extinction voltage, PD levels at different test voltages and localisation of PD [13]). From the field experiences, the VHF on-line PD detection results can be categorized in three groups, as reflected in figure 10: I: clear PD pattern; presence of PD source in cable sample; II: doubtful PD pattern; presence of PD source is not clear as a result of high disturbances or multiple PD sources; III: no PD pattern; no PD activity in the cable sample, the noise level is low. In case of clear PD patterns, PD sources are active under operation voltage of the cable, which means a degradation of the insulation materials. Therefore, more accurate diagnosis is necessary by off-line PD measurements. Doubtful PD patterns indicate that PD related defect might be active in the cable insulation. The patterns can be influenced by external Table 1: Approximated limitation of cable section lengths for sensitive on-line PD detection. Cable cross-section Large (>150mm2) Small (<150mm2) Approximate length limitations Single end Both end detection detection 750m 1500m 1000m 2000m Power cables with larger lengths than indicated in table 1, can of course be on-line diagnosed for PD activity. However, the probability of missing a defect in the cable section becomes bigger, except for those cases where very high PD activity is active. As can be concluded from the comparison of the field, VHF on-line PD diagnostics can be used as a pre-selective tool for off-line PD diagnosis of cable sections. In those cases, where doubtful PD patterns (cat II) are detected on-line, off-line TUD_Fwester_B1 Session 1 Paper No

6 diagnosis is recommended, even if there is a chance that no PD source will be found. CONCLUSIONS This paper handles different sensitivity and possibility aspects for on-line PD detection as a pre-selection tool for off-line diagnostics on MV power cables. The following can be concluded: 1. On-line PD detection can be performed using an inductive PD sensor on the earth connection at the end of a cable section, connected to a SA. Tuning to a frequency with good SNR, phase resolved PD patterns could be obtained. 2. Applying VHF PD detection in practice, several influencing factors regarding the sensitivity and detection possibility are relevant: a. PD activity in a cable section can change with the cable load; b. depending on the switchgear type, the HF PD sensor can sometimes be difficult to connect; c. external disturbances can result in unreadable PD patterns; d. phase-resolved PD patterns in case of multiple PD sources in more than one phase are difficult recognisable; 3. Signal attenuation during propagation of PD pulses through the cable is influenced by: a. the cross-section of the cable, the thicker the cable conductor the higher the attenuation; b. the length of the cable section; c. the number of joints in the propagation path; 4. On-line PD patterns can be categorised in three groups: a. Clear PD pattern; b. Doubtful PD pattern (due to e.g. disturbances); c. No PD pattern, low noise level; 5. For sensitive PD detection, the maximum length of a cable section is limited, for single and both end detection. As a result, combining the less sensitive on-line PD detection and high sensitive off-line diagnosis (e.g. OWTS) can be of benefit by insulation condition assessment of distribution power cable networks. In particular, in combination with failure history and the position in the network, the on-line PD detection could be useful for prioritising particular cable sections for an off-line in-depth analysis. REFERENCES [1] E. Gulski, F.J. Wester, J.J. Smit, P.N. Seitz, M. Turner, 2000, On-site PD Diagnostics of Power Cables using Oscillating Wave Test System, Proceedings 2000 International Symposium on Electrical Insulation. [2] E. Pultrum, E. Hetzel, 1997, VLF PD Detection as a Diagnostic Tool for MV Cables, Proceedings IEEE PES Summer Meeting. [3] E. Lemke, P. Schmiegel, H. Elze, D. Russwurm, 1996, Procedure for Evaluation of Dielectric Properties Based on Complex Discharge Analysing, Proceedings 1996 International Symposium on Electrical Insulation. [4] C.G. Henningson, K. Poster, B.A. Fruth, D.W. Gross, 1996, Experiences with an On-line Monitoring System for 400kV XLPE Cables, Proceedings IEEE 1996 Transmission and Distribution Conference. [5] V.R. Garcia-Colón, 2001, Power Cable On-line Diagnostics using Partial Discharge Wide Band Techniques, Proceedings LESCOPE [6] N. Ahmed, N. Srinivas, 1998 On-line Partial Discharge Detection in Cables, IEEE Transactions on Dielectrics and Electrical Insulation, Vol5, Issue2. [7] B.T. Phung, Z. Liu, T.R. Blackburn, 1999, On-line Partial Discharge Measurements on High Voltage Power Cables, Proceedings IEE HV Engineering Symposium. [8] C.M. Walton, 2001, Detecting and locating MV failure before it occurs. Experience with live line partial discharge detection on underground paper insulated 11 kv cables in London, Proceedings CIRED [9] S. Meijer, 2001, Partial discharge diagnosis of highvoltage Gas-Insulated Systems, PhD Thesis, Optima Grafische Communicatie, Rotterdam, The Netherlands. [10] J.P. van Bolhuis, 2002, Applicability of recovery voltage and on-line partial discharge measurements for condition assessment of high voltage power transformers, PhD Thesis, Optima Grafische Communicatie, Rotterdam, The Netherlands. [11] J.P. Zondervan, E. Gulski, J.J. Smit, T. Grun, M. Turner, 1999, A new multipurpose partial discharge analyzer for on-site and on-line diagnosis of HV components, Proceedings 11 th International Symposium of High Voltage Engineering. [12] S. Meijer, J.J. Smit, A. Girodet, T.S. Ramu, 2001, Spectral analysis of partial discharges in GIS, Proceedings 12 th International Symposium of High Voltage Engineering [13] F.J. Wester, E. Gulski, J.J. Smit, F. de Vries, 2001, Advanced PD diagnostics for CBM of MV Power Cables, Proceedings 12 th International Symposium of High Voltage Engineering. TUD_Fwester_B1 Session 1 Paper No