Investigation of PD Detection on XLPE Cables

Investigation of PD Detection on XLPE Cables Hio Nam O, T.R. Blackburn and B.T. Phung School of Electrical Engineering and Telecommunications The University New South Wales, Australia Abstract- The insulation lifetime of XLPE power cables is determined by several factors. One of the most important of these is the occurrence of partial discharge (PD) in the dielectric. The ability to detect and locate the PD sources is limited by attenuation and distortion of the high frequency PD pulses as they propagate through the cable. This paper presents the results of measurements of PD pulses that are generated by an artificial PD defect with capacitance-coupled sensors used for the detection of PD, rather than a high frequency current transformer (HFCT). The results show that the external capacitance sensors have a number of advantages, such as better sensitivity, for the detection of PD pulses. PD pulse waveforms were evaluated by three different integral methods in order to estimate the best way to characterise the PD pulses. The PD pulse is severely attenuated and distorted with increasing length of the power cable and the frequency of the PD pulse and some means of characterising is necessary. Simulation results are compared with test measurement results and it has been found that the cable model developed and used was able to predict the measurement results accurately. D I. INTRODUCTION ue to technical and economic advantages, cross-linked polyethylene (XLPE) insulated cable systems are now almost universally used in the electrical supply industry and this trend is likely to continue. However, the lifetime of XLPE power cables is determined by several factors. One of the more important of these is degradation and breakdown of the XLPE. There are number of possible mechanisms for electrical breakdown in a solid dielectric [1], of which partial discharges is a major factor. Various defects, such as voids, contaminants and electrical trees can cause partial discharge (PD) activity in the XLPE. Although the magnitude of PDs is usually small XLPE insulation is very sensitive to partial discharges and PD activity can cause rapid progressive deterioration and failure. To ensure the reliability of the whole cable system, PD testing and location are becoming an essential part of the monitoring and testing of high voltage XLPE cables. Partial discharges generate electrical signals with frequencies up to several hundred MHz [2] in cables. The ability to detect and locate a PD source is limited by various factors such as external interference, signal attenuation and the limited bandwidth of PD detection sensors. Therefore it is necessary to understand the high frequency response of power cables and there are a number of publications on HF cable modelling already available [2] [3]. In order to achieve a better understanding of practical PD signal propagation, experiments have been undertaken at UNSW using an artificial PD source. The external sensors are used for PD detection are capacitive instead of the conventional HF-CTs usually used. Capacitance sensors have better sensitivity and the purpose of the tests was to assess the sensitivity and bandwidth of the sensors, investigate the best method to estimate PD pulse character and the propagation of PD pulses, including attenuation and distortion effects imposed by the lossy cable dielectric. II. CHARACTERISTICS AND STRUCTURE OF XLPE CABLE XLPE cables have excellent electrical insulation performance. In addition to good dielectric properties, the thermal resistivity and dielectric dissipation factor are low, giving good thermal rating. They are economical and easy to handle and install [4]. However they have the problem of susceptibily to water trees and the development of PD activity. The single core XLPE cable used for the work described in this paper has a stranded aluminium core and the cable is concentric with semi-conducting layers under and over the XLPE insulation. The metallic shielding comprises helical copper strands and the overall jacket is extruded polyvinyl chloride (PVC). The semi-conducting layers are polyethylene or ethylene copolymer mixed with conductive carbon black [5]. The semi-con layers are often neglected in cable modelling but it has been shown that they are very important in determining PD propagation characteristics [3]. III. HIGH FREQUENCY ATTENUATION In order to understand PD signal propagation characteristics of XLPE cable, an understanding of the high frequency loss parameters of a cable is needed. These loss parameters include skin effect, dielectric loss, and semi-conducting resistivity. XLPE cables have very low dielectric loss factor at power frequency so that one might expect very low high frequency signal attenuation. However the semi-con layers result in significant high frequency signal loss caused by the propagation of a radial displacement current (which increases with frequency) through the resistance of these layers [3]. IV. MEASUREMENT SET UP A 22KV single core XLPE cable was used in the tests. The detail of the cable is shown in Figure 1. The total length of cable was four metres. An artificial PD source defect was inserted one third of the way along the cable. Four external sensors were attached at various locations on the cable. The 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 1

external sensors are capacitive-coupled. Sensors 1 and 2 are attached at each terminated end of the cable and sensor 3 and 4 are attached on either side of the PD source. Sensors 2 and 3 are separated by 1/3 of the cable length. Sensors 4 and 1 are separated by 2/3 of the cable length. As sensors 3 and 4 are at the same position, in the following sections only sensor 3 results are shown. The major benefit of using those external sensors is the ease of location along the length of cable. This is a major advantage for studying PD signal propagation along the cable. The artificial defect is created by a pin shape which is closely attached on the insulation surface. The closer to the insulation surface, the higher is the electric stress in the void, which generates the PDs. The detail of the PD source is shown in Figure 2. A test voltage of 10 kv was applied to the cable. The PD inception voltage was 4 kv. PD pulses generated at the defect travel to both ends of the cable and are then reflected back, depending on the termination.. The sensors capture the PD pulse travel. Figure 3: Comparison of PD signals captured by different types of HF-CTs B. Comparison results In order to compare suitability for the PD tests, the external sensor and HFCT #1 were compared using the set up shown in Figure 1. The comparison is shown in Figure 4. The results indicate clearly that the external sensor has the better sensitivity for the internal PD discharge. Due to the attenuation effect of the semi-con layers on HF PD signals, the HF-CT signal is heavily attenuated and distorted. Figure 5 shows the bandwidths of the three HF-CTs and the external sensor. A network analyzer (9 khz to 3000 MHz) was used to perform two port measurements of the parameters on all the detection sensors. HF-CT #1 and HF-CT #2 were used for the testing due to their good sensitivity. The results illustrate that capacitive sensors have better sensitivity than HF-CTs. Both the HF-CTs and the capacitive sensors have good bandwidth for PD measurements. Figure 1: Laboratory test measurement setup Figure 2: Cross section diagram of the artificial defect Figure 4: the compared result of HFCT and External sensor. V. SELECTION OF DETECTION SENSORS A. Conventional PD detection sensor In order to determine the sensitivity level of the capacitance sensors, HFCTs were used for comparison. Three HFCT with different bandwidths were used. Figure 3 illustrates the results of the captured PD signals using three different type of HFCT. Figure 5: Bandwidth of external sensor 3 and two other HF-CTs. 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 2

From the experience on using the external sensor, the following advantages of capacitive sensors can be listed: 1. Better sensitivity than HF-CT. 2. Provide a better representative PD pattern. 3. Can be sited at any position on the cable on and off site. 4. Suitable for most applications due to its ease of installation. VI. MEASUREMENT RESULTS A. PD signal propagation To establish a reliable monitoring system for the cable, a comprehensive knowledge of PD signal propagation characteristics is required. Three separate PD signals were detected from the output of the sensors 1, 2 and 3 for each PD event. As each sensor is located at a different position this gives information on the PD signal propagation characteristics. A comparison of three separate PD signals is shown in Figure 6. Sensor 3 is located near the PD source so that signal has the greatest amplitude. Sensor 2 is located about 1/3 of the total cable length away and sensor 1 is located about 2/3 of the total cable length away. Therefore, the PD signal of sensor 1 travels 1/3 of a cable length and 2 travels 2/3 while sensor 3 is direct. There is a clear attenuation of the PD signals and higher attenuation of PD signals of sensors 1 and 2 compared with sensor 3. Moreover, Figure 6 shows that there are few significant reflective pulses from the terminations. The oscillation in the PD signal wave tail is not significant and is attenuated quickly. Figure 6: Comparison of PD signals captured at different locations with external sensors. Figure 7: PD pattern of PD defect activity as recorded by an Mtronix PD detector. Figure 8: Frequency spectrum of the PD measurement in figure 7. C. PDs pattern from the external sensors The same PD activity was also recorded using the external sensors at three different positions of cable, shown in Figure 1. Figure 9 is the record of the PD pattern by sensor 1, Figure 10 is from sensor 2 and Figure 11 is from sensor 3. The selections of the bandwidth of those three PD measurements are between 8-8.5MHz which is higher than the PD measurement in section B (Figure 7). Figures 9-11 have a similar PD pattern. Sensor 3 has the highest PD level but the differences of PD level between those three external sensors are relatively low. The high noise levels cover some of the PDs patterns and some of the important information is lost due to the integrating function in the Mtronix monitoring system. The maximum PD magnitude is 50pC which is captured by external sensor 3, sited at the PD source. The HFCTs were also used to detect PD patterns but the HFCT were unable to detect any PDs using the Mtronix monitoring system with the artificial defect test. B. PDs pattern of the measurement A PD scatter plot pattern of the defect, recorded from an Mtronix PD monitoring system MPD540, with direct coupling is shown in Figure 7. The selection of the bandwidth of the PD measurement system is very important for noise suppression. The bandwidth of the PD measurement used was 700 KHz to 1 MHz. Figure 8 shows the frequency spectrum of the PD measurement. The bandwidth of PD measurement (700KHz-1MHz) was taken for the first peak. Figure 7 illustrates that the discharge level is over 100pC. The pattern is shows stationary and wandering impulses and this is the usual pattern of an internal discharge in a void in a solid dielectric. Figure 9: PD pattern from external sensor 1 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 3

q1 q2 q3 q4 q5 q6 Figure 10: PD pattern from external sensor 2 Figure 11: PD pattern from external sensor 3 VII. EVALUATION OF PARTIAL DISCHARGE The shape of the partial discharge pulse that is received by a PD sensor on a cable is influenced significantly by its passage through the cable. It can be distorted and attenuated and thus its apparent magnitude becomes difficult to evaluate. The distortion can also affect the use of pulse shape analysis to identify fault types. The aim of this section is to determine which parts of the distorted PD pulse waveshape are the most suitable for evaluation so that the apparent magnitude of the PD pulse can be best determined. A. Introduction of three different methods According the Australian Standard 60270 [6], the charge q generated by a PD calibrator source is determined from the equation: q = i(t) dt = 1 R u(t) dt Eq (1) Where i(t) is the current pulse generated by the calibrator, u(t) is the voltage pulse measured by an oscilloscope, for example, and R is the load resistance (R =50 Ω in this case). The total charge q is the integration of the current pulse. The typical charge values obtained in PD detection are picocoulombs (pc). In order to apply equation 1 to determine the charge under high voltage testing in this case the percentage of charge was used instead of the normal charge unit (pc). Figure 12: Example of the full PD pulse waveshape obtained with an external sensor. In practical testing the number of PD pulses can be quite numerous and so there is some need to make the evaluation of charge as simple as possible, but without losing the inherent PD characteristics that may enable PD source identification and location. To this end a number of approximation methods were tried to determine the most efficient analysis method. Three methods were used to evaluate the distorted PD pulse (in Figure12): a) The charge from the first peak waveform only is evaluated. Total charge obtained is designated: q=q1. b) The waveform parts with same the polarity are evaluated (e.g. the entire positive parts of pulse), total charge: q=q1+q2+q3. c) All parts of the waveforms are evaluated regardless of polarity, ie total charge: q=q1+q2+q3+q4+q5+q6. B. Measurement results of different location of sensors A number of measurements were taken during the experiments with the artificial PD defect. The results are shown in Figures 13 to 15. On the horizontal axis, L is the total length of cable between the PD site and the termination. 0% indicates the position of the artificial PD defect location. 100 % is the termination of the power cable. The first peak of the PD pulse has width of 0.02μs which indicates a frequency of about 50 MHz. The results of each method used (methods a, b and c) are shown in Figure 13 to 15. Figure 13 includes the pulse waveform within 0.05μs. Figure 14 includes all the waveform within 0.04μs and Figure 15 includes all the waveforms within 0.03μs. The results indicate that methods b and c have a high percentage (3-6 %) difference compared with method a. This is due to the result of Figure 13 and Figure 14 taking into account the reflected pulse from the termination and some distortions from background noise. In most cases, method a is the best method for measuring PD pulse characteristic. However, method b is also acceptable when dealing with the high frequency PD pulse. Furthermore, the results were confirmed by the fact that the apparent charge of the PD pulse, as a diagnostic parameter is not very sensitive when comparing other diagnostic parameters such as the shape of the PD pulse and its magnitude. The results of Figure 13 to Figure 15 can also used to evaluate the 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 4

attenuation of a PD pulse along the cable. The charge magnitude is decreased from 100% to 80% after travelling a distance of half of the 4 meter cable length. There is very considerable initial attenuation at the PD site after the PD pulse is generated at the defect point. This is due to the high electrical field between the insulation and semi-conducting layer at that site. From the travel along the 20-100% of the length, the attenuation is only 20%. C. Cable Model In order to attempt a prediction of the measurement results to validate a reasonable model for the cable PD propagation, the J-Marti single core cable model incorporated in ATPDraw was used to do the simulation [7]. The ATPDraw (Advanced Transient Program) is one of the EMTP (Electromagnetic Transient Program) forms and it is probably the most widely-used tool for the analysis of electromagnetic transients in power systems. The basic idea of cable modeling is to calculate the cable parameters from the geometrical data and material properties and then convert into new set of parameters for usage by available cable models. The general parameters used by cable models are the following: Z(ω) = R(ω) + jωl (ω) Eq (2) Y(ω) = G(ω) + jωc (ω) Eq (3) And the frequency dependent characteristic impedance of the cable can be defined as follows: Zc(ω) = Z(ω) Y(ω) Eq (4) Figure 13: PD magnitude by different methods (time width of 0.05μs). Where R, L, C, G are the series resistance, series inductance, shunt capacitance and shunt conductance per unit length of cable system. The ω is angular frequency. Z and Y are calculated by the geometrical data and material properties. Figure 14: PD magnitude by different methods (time width of 0.04μs). D. Simulated results The attenuation factors from the test measurements on the cable were compared to the attenuation from the cable model and the result is shown in Figure 16. The simulated result with the Marti model is close to the measurement with error of only ± 3%. Figure 17 shows a 3D plot of attenuation of PD pulse and its dependence on length of cable up to 100 metre and dependence on frequency of the PD pulse. The PD pulses in the measurements have a frequency band of about 50 M Hz, which give about 7dB attenuation at 100 metres of cable length. The calibration signal has a frequency about 10 MHz which gives lower attenuation in the measurements. The results indicate that the PD pulse is attenuated significantly by cable length and frequency of pulse. Figure 15: PD magnitude by different methods (time width of 0.03μs). Figure 16: Comparison of PD magnitude measurement and simulation by Marti model. 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 5

assistance and cooperation in performing the high voltage measurement tests on the cable. REFERENCES Figure 17: PD attenuation vs length of cable and frequency of PD pulse using simulated model VIII. CONCLUSION Partial discharge generates electrical signals with frequencies up to few hundred MHz. At these frequencies the propagation of the Pd signals along lossy power cables will cause significant attenuation and distortion of the PD signal. This can affect any PD characterisation techniques used to identify PD type and location. It is thus necessary to develop a valid cable model that will allow the measured signal to be convoluted into the true emitted signal so that characterisation and location is possible. This paper reports on the results of experiments used to test such a model. In order to have realistic tests in the laboratory, an artificial PD defect was applied on a short length of 11 kv single core XLPE cable. The PD defect generated an internal PD discharge and that was used to investigate the PD signal propagation along the XLPE cable. Capacitively-coupled external sensors were applied for the PD detection and the results show that external capacitive sensors have a number of advantages over the HF-CT most commonly used for PD pulse detection. Three methods were applied for evaluation of the PD waveforms received. Most researcher use method a to characterise the PD pulse. However, the results have shown that method b is acceptable for accurate characterisation of the PD waveform. The attenuation of the PD pulse propagation increases with the frequency of PD pulse and length of cable. However the results of the simulation using the J Marti high frequency model have shown that it can be used to evaluate the PD waveform characteristics as it propagates and thus can be used to provide adequate information about the emitted PD pulse from the measured pulse even when the measured pulse has been propagated some distance along the cable. [1] C. C. Barnes, Power cables: their design and installation, Chapman & Hall, London, 1953 pp. 45-46. [2] G. C. Stone and S. A. Boggs, Propagation of Partial Discharge Pulses in Shielded Power Cable, IEEE Conf. Electr. Insul. Dielectr. Phenomena, pp. 275-280, 1982. [3] Boggs, S., Pathak, A. and Walker, P., Partial discharge. XXII. High frequency attenuation in shielded solid dielectric power cable and implications thereof for PD location, Electrical Insulation Magazine, IEEE, Volume: 12, Issue: 1, Jan.-Feb. 1996, pp9 16. [4] T. Tanaka and A. Greenwood, Advanced power cable technology volume II present and future, CRC press, 1983, pp 4-7. [5] T.C. Yu and J. R. Marti, "A Robust Phase-Coordinate Frequency-Dependent Underground Cable Model(zCable) for the EMTP", IEEE Transactions on Power Delivery, Vol. PD-18, No. 1, January 2003, pp189-194. [6] Australian Standard 60270: High-voltage test techniques- Partial discharge measurements, pp33-34. [7] H.N. O, T.R. Blackburn, B.T. Phung, M. Vakilian, H. Zhang and M. S. Naderi The propagation characteristics of high frequency partial discharge pulses in XLPE cables. To be presented at Australasian Universities Power Engineering Conference (AUPEC 05), Tasmania, Australia, 2005. ACKNOWLEDGEMENTS The authors could like to thank Mr. Z. Liu of the School of Electrical Engineering, UNSW for his valuable and unstinting 2008 Australasian Universities Power Engineering Conference (AUPEC'08) Paper P-254 page 6