THE PROPAGATION OF PARTIAL DISCHARGE PULSES IN A HIGH VOLTAGE CABLE

Transcription

1 THE PROPAGATION OF PARTIAL DISCHARGE PULSES IN A HIGH VOLTAGE CABLE Z.Liu, B.T.Phung, T.R.Blackburn and R.E.James School of Electrical Engineering and Telecommuniications University of New South Wales Abstract Partial discharge within the dielectric of HV solid dielectric power cable greatly shortens the cable life. It is desirable to detect and locate the PD in the cable before breakdown occurs. Understanding the phenomena of propagation of the PD pulse in a power cable is fundamental to the detection and location of PDs. In this paper a power cable model represented by a coaxial transmission lossless line was analyzed theoretically. Based on this model PD pulse propagation in a power cable was simulated by using EMTP ( Electro-Magnetic Transients Program ). A capacitive sensor has also been developed with a high frequency response to detect PD propagation. Pulse transmission properties of PDs were measured on a short length of 22 kv XLPE cable in the laboratory. The experimental results show a good agreement with the theoretical analysis. The paper describes the test results and the theoretical modelling performed. It also discusses the pulse attenuation in time and frequency domain and discusses this with reference to practical PD detection in the field. 1. INTRODUCTION High voltage power cable failure can cause a long interruption, costly repairs and loss of revenue. Insulation failure is a major cause of cable outages. Early detection and accurate diagnosis of insulation problems can prevent such faults. Partial discharge ( PD ) activity in high voltage cables is caused by various defects, such as voids, shield protrusions, contaminants, advanced stage of water trees etc. PDs will gradually degrade and erode the dielectric materials, eventually leading to final breakdown. Fig 1 is a typical 22 KV paper cable in which insulation breakdown occurred at the cable terminal. The measurement and location of such discharges are of significant importance for power system reliability. signals and their propagation characteristics in the power cables. This paper gives a theoretical analysis of PD signal and a mathematical model for PD pulse propagation along a power cable. Experimental results performed in the laboratory are also presented and compared with the calculations. 2. THEORETICAL ANALYSIS 2.1 Generation of PDs in a High Voltage Cable Gas-filled cavities or voids are formed in solid insulation during manufacture, installation or operation. When the electric stress in the void exceeds the breakdown strength of gas within the void, partial discharges will occur. The electric stress in the void is the critical factor for the generation of partial discharges at known temperature and pressure. XLPE sheath conductor void Fig 1 22 kv Paper Insulated Cable with Fault In order to understand and predict the aging of cable insulation subject to these discharges, it is necessary to investigate and understand the occurrence of PD Fig 2 Equipotential lines in a cable and the effect of a gas-filled 1 mm diameter void ( Equipotential spacing is 5 % ) A typical electric potential distribution for a 22kV XLPE cable is shown in Fig 2. Diameter of the conductor is 1 mm and the thickness of the XLPE insulation is 8 mm. The gas void, with diameter of 1 mm, is located near the conductor. It is clearly seen

2 from the distribution of equipotentials that the electric field distribution is distorted by the void. ( kv/mm ) Stress Distribution ( no Void ) ( mm ) Fig 3 Electric Stress Distribution With No Void ( 12.7 kv rms on the cable conductor ) ( kv/mm ) Stress Distribution ( with Void ) ( mm ) Fig 4 Electric Stress Distribution With a Void ( 12.7 kv rms on the cable conductor ) The quantitative comparison of electric stress in the XLPE insulation without and with the void can be clearly seen in Fig 3 and Fig 4. In the XLPE with no void, the maximum stress appears at the outer surface of the conductor. This is the normally permitted maximum stress in the insulation. But within the void, the stress may be even higher than this maximum designed stress. The maximum stress in the void is about 2.8kV/mm (rms) which is near to the breakdown stress of 3 kv/mm (rms) for air. Under hot or overvoltage conditions discharges may occur in this small void and generate the PD pulses which can then travel along the cable. Investigation of the transmission of PD pulses in a high voltage cable is thus essential for detection and location of such a fault. 2.2 Propagation Model of PDs in a High Voltage cable Partial discharge detection and location is a powerful and useful tool for the maintenance and condition monitoring of electric power cables and has attracted much research and development effort since the 198 s. A complicated mathematical model of a PD signal was described in [1]. A PD signal model using a narrow Gaussian pulse and quadratic approximation for the cable attenuation was proposed. But the parameters for propagation in the quadratic model must be carefully selected to provide a reasonable approximation because of the variation of the propagation function according to the type of cable and degree of aging. To simplify the model, a medium voltage cable can be represented by a coaxial transmission line, which is described by the propagation constant γ[2]. This constant is, according to basic transmission line theory, fully determined by the primary parameters R, G, L, and C which depend more or less on the frequency. But discussion of this model is better focused on the frequency domain rather than time domain for the purposes of PD location. In the practical situation, compared with overhead lines, power cables are much shorter, typically less than 2 km. Fast transient simulation using a lossless transmission line model based on Bergeron s method is adequate for reasonable accuracy in such a short length[3]. It was shown that travelling wave solutions were much faster and better than for traditional cascaded π-networks. In our investigation the length of the test cable is only a few metres. A lossless transmission line model was applied for numerical calculation of propagation of PD signals in this cable. 3. NUMERICAL CALCULATION AND EXPERIMENTAL RESULTS The power cable used for the investigation was a 22 kv XLPE insulated cable with 1 mm diameter of Al stranded conductor and 1 mm thickness of XLPE insulation. The sheath includes a thin layer of semiconductor tape and a number of copper strands spiralling around the cable. The total length of the test cable is 3.84 m. An artificial fault ( point C,see Fig 5 ) was located 1.45 m from one end ( point B ). Fig 5 : a 22 kv XLPE cable with an artificial fault at C In the calculation the widely used EMTP program was employed in the Window98 environment. A narrow pulse with normalised magnitude and 2ns width was selected to simulate a fast PD pulse [4] and was injected at the selected points. The monitoring points were at the cable terminals, i.e. points A and B. The required parameters in the program such as

3 propagation speed and surge impedance can be easily derived from the relevant equations by using the measured travel time of the pulse, the physical length and the capacitance of the cable. In the experiment, a fast pulse with a risetime of about 1 ns was injected at the selected points through a short 5 Ω coaxial cable. The travelling signals were measured at the cable terminals by using a capacitive sensor which has a 1 GHz bandwidth. The detected signals from this sensor were transmitted by a 5 Ω cable to a digitizing oscilloscope ( LeCroy 9362C ) with 1.5 GHz bandwidth and 1 GS/s sample rate. Digital signal processing was possible by utilizing internal Math Function such as FFT in this CRO. arbitrary unit injection at C, measure at B ( ns ) Fig 7 Numerical Calculation : Injection at Point C, Measured at Point B The injected pulse is transmitted into the power cable and totally reflected at the open end. The reflected time location Fig 6 Lattice Network Pattern of travelling Pulse in the 22 kv XLPE Cable pulse then returns to the connection point. Due to the different surge impedance of the two cables this discontinuity point produces another reflection pulse back in to the power cable. A lattice network pattern is given in Fig 6 with injection point at C. The first pulse will arrive at point B at time p1. The second pulse will travel to point A firstly and then reflect back to point B at time p2. The computer simulation and measurement results are shown in Fig 7 and Fig 8 respectively. There is good agreement between them. After about three reflections the signal attenuates to a negligible level. Measurement of the first two or three pulses will thus be important for the location of PDs. To increase the sensitivity of the small output signal a wide bandwidth amplifier with high cutoff frequency of 3 MHz was used to enhance the signal in this case. This causes little distortion of the signals, as can be observed in Fig 8. The time series of these pulses gives the basic details of the propagation of PD pulses in a power cable. It also shows that the mathematical model of the PD transmission is applicable. The simulation and measurement results at point A are shown in Fig 9 and Fig 1. Fig 8 Experimental Measurement : Injection at Point C, Measured at Point B arbitrary unit injection at C, measure at A ( ns ) Fig 9 Numerical Calculation : Injection at Point C, Measured at Point A Fig 1 Experimental Measurement : Injection at Point C, Measured at Point A 4. DISCUSSION AND INTERPRETATION

4 4.1 Location of PDs in Time Domain PD location in cables based on time-domain reflectometry is the most popular method. The PD pulse travels from C simultaneously towards A and B ( see Fig 5 ). At each of these points, it experiences total/partial reflection and travels in the opposite direction until the signal is totally attenuated by the cable. The position of the fault is determined by measuring the relative time of arrival of multiple reflections of the pulses. Assuming the measurement point is at B, the following equations for the time intervals (t 1, t 2, and t 3 in Fig 6)can be easily written : t 1 = l cb / v, ( 1 ) t 2 = l ac / v = ( l ab l cb ) / v ( 2 ) t 3 = l ab / v ( 3 ) Where v is the pulse propagation speed in the cable and the l s are the appropriate section lengths of cable. Thus the PD will be located by following formula: l cb = l ab.5 v t ( 4 ) Where l ab is the total length of the cable and t is the time interval between the 1 st and 2 nd pulses t = t 3 + t 2 t 1 ( 5 ) error decreases in proportion to the square root of the number of measurements. Reduction of detection bandwidth will result in decrease of the location resolution, however. For a fault in a long cable, a low bandwidth system may have enough resolution to identify the first and the second pulses. But if the fault occurs in a short cable, the time interval between the first and second pulses will be very narrow. In this situation pulse overlapping may cause errors in both location and magnitude detection. Therefore, selection of the bandwidth of the system will affect the accuracy and resolution of PD location in power cables. arbitrary unit measure at A, injection at B ( ns ) Fig 11 Numerical Calculation : Injection at Point B, Measured at Point A The location limit and error will depend on the accuracy of measurement of the total cable length, speed and the time interval. The cable length can be accurately decided prior to installation. The travelling speed of PD pulses can be measured in a short cable with reasonable accuracy. But the time interval will be far more complicated and affected by many factors such as bandwidth of the measurement system, sampling rate, waveform oscillation and overlapping, attenuation of high frequency components, background noise etc. Ideally the time interval should be measured from the base of the first pulse to the base of the second pulse. But this method will suffer difficulties in actual measurement due to waveform oscillation. A typical simple method is to measure the time interval between the first peak and the second peak. But in a power cable, PD pulses spread out as they propagate away from the PD source because of high frequency attenuation and dispersion in the cable [5]. This will reduce the accuracy of PD location. It is expected that longer cables will have higher error because the attenuation will get worse with a longer cable. A practical method to reduce this error is to reduce the bandwidth of the measurement system to extend the wavefront of the first pulse. The accuracy also can be improved by accumulating and analyzing a large number of independent measurements because the Fig 12 Experimental Measurement : Injection at Point B, Measured at Point A In our experiment the cable first used was 3.84 m and a pulse generating site was placed 1.45 m from point B ( Fig 5 ) with uncertainty of.1 m. In such a short cable the loss and attenuation of high frequency for the first 2 pulses can be neglected with the 3 MHz cutoff frequency of measurement system. To measure the travelling speed of pulses along this cable a narrow pulse was injected at the Point B. Fig 11 and Fig 12 are the waveforms of simulation and measurement. From measurement of the time interval of the first two pulses the calculated travelling speed is.146 m/ns. Based on the formula ( 4 ) and Fig 8 the fault location was apparently 1.58 m from point B. This is an error of about.13 m compared with actual measurement,

5 i.e. less than 1 ns in the time domain. The maximum locatable distance from the fault to the Point B can be calculated from equation (4) where t must be the minimum time interval between the first and the second pulses, i.e. the width of the first pulse. This maximum distance is about 3.11 m based on the pulse width of 1 ns. If the fault is beyond this distance, the detected pulses will be superimposed and cause difficulty in discrimination of the first two pulses. 4.2 High Frequency Attenuation of PDs in Frequency Domain The high frequency attenuation of PD pulses in a shielded power cable is simply the sum of : ( i ) the power loss in the dielectric as a result of its DDF at the frequency of interest and ( ii ) dissipation in the semi-conducting layers as a result of the radial capacitive displacement current that flows through the resistance of the semi-conducting layer. From the pulse distribution in the time series discussed above, the attenuation of the frequency components is associated with not only the pulse itself but also its propagation characteristics. To investigate the attenuation of high frequency components of the PD pulse itself a 5 m length of 11 kv cable was used with same terminal marking as in Fig 5. There was no artificial fault used. The measured first and second pulses waveforms are shown in Fig 13 and Fig 15. Their corresponding spectra are displayed in Fig 14 and Fig 16 respectively. The first pulse is a steep and short pulse with risetime of 1 ns and width of 4 ns. After this pulse travels 1 m ( go and return ), the second pulse ( reflection pulse ) becomes a pulse with risetime of 7 ns and width of 25 ns. The high frequency reduction can be clearly seen from their spectra analysis in Fig 14 and Fig 16. The frequency of the peak point of the first pulse is at 1 MHz compared with 2~3 MHz for the second pulse. The highest frequency component of the first pulse is about 1GHz. But at 2 MHz the spectrum of second pulse already attenuates to zero. The location error caused by pulse distortion is about 6ns, corresponding to.96 m cable length. The minimum detectable distance to the point A is about.64m. If the record length is long enough to include more reflection pulses, this multiple reflection pattern also contributes to the distribution of the high frequency components, which will vary with the fault location. This is illustrated in Fig 17 and Fig 18 for the 3.84 m length of 22 XLPE cable of Fig 5. Fig 17 shows the waveform measured at point B and its spectrum when a fast pulse was injected at the same end of the 22 kv XLPE cable. When moving the injection point to Point C ( Fig 5 ) the spectrum of the waveform shown in Fig 8 includes more high frequency components ( Fig 18 ) due to multiple reflections. Fig 13 1 st Pulse : ( 11 kv cable ) Injection at Point B, Measured at Point B Fig 14 Normalized Spectrum of Fig 13 Fig 15 2 nd Pulse : ( 11 kv cable ) Injection at Point B, Measured at Point B Fig 16 Normalized Spectrum of Fig 15 The relationship between high frequency components due to reflection and fault location in, say, an 8 m long power cable was simulated and the result is shown in Fig 19. If the fault occurs near either cable terminal the highest frequency due to reflection can reach 8 MHz. But this type of high frequency will attenuate to less than 1 MHz within 1 m.

6 properties. The experimental results show good agreement with the theoretical analysis. Fig 17 Waveform and Spectrum : Injection at Point B, Measured at Point B PD location based on time-domain reflectometry of PD pulses can be successfully used if the measurement system is selected properly. Decrease of the lower cutoff frequency of measurement system can reduce the resolution of location. In the experiment the location accuracy of.13 m can be obtained with maximum locatable length of 81% of the total cable. The fraction of locatable length will be much higher for longer cables in operating situations The error of location is mainly caused by the high frequency attenuation which associated with not only the pulse itself but also its propagation characteristics. The experimental and analytical results also demonstrate that if the faults occur near the measurement point, more high frequency components can be detected. Using properly designed high frequency couplers to detect this type of high frequency may be another way to monitor PDs in power cables in the field. 6. ACKNOWLEDGEMENT Fig 18 Spectrum of Fig 8 : Injection at Point C, Measured at Point B frequency ( MHz ) High Frequency distribution due to reflection with single fault fault location ( m ) Fig 19 High Frequency Distribution due to Reflection Measurement of high frequency characteristics at cable terminals may give another opportunity to detect and locate the faults nearby. Past experience indicated that more than 9% of cable faults took place at cable terminals or joints. So application of this technique could be attractive to researchers and utilities. 5. CONCLUSIONS PDs in power cables may be generated due to overstress in voids. In order to understand the phenomena of propagation of the PD pulse in a power cable a lossless line model was analyzed theoretically. Based on this model PD pulse propagation in a short length of 22 kv XLPE cable was simulated. A capacitive sensor with a high frequency response was employed to measure the pulse transmission The authors wish to express their thanks to Integral Energy for their support of this project. 7. REFERENCES [1] Z.Du, P.K.Willett and M.S.Mashikian, Performance Limits of PD Location Based on Timedomain Reflectometry, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 4 No 2,April 1997 pp [2] R.Villefrance, J.T.Holboll and M.Henriksen, Estimation of Medium Voltage Cable Parameters for PD-Detection, Conference Proceeding of the 1998 IEEE International Symposium on Electrical Insulation, Arlington, Virginia, USA, Vol. 1, pp19 pp112 [3] EMTP Theory Book 1994, pp4-57 [4] Z. Liu, B.T. Phung, R. Morrow and T.R. Blackburn, "Measurement of fast partial discharge current waveforms in a void under ac electric stress", Proc. AUPEC'98, Univ. of Tasmania, Sep.27-3, 1998, pp [5] S.Boggs, A. Pathak and P.Walter, Partial Discharge XXII : High Frequency Attenuation in Shielded Solid Dielectric Power Cable and Implications Thereof for PD Location, Electrical Insulation Magazine, Vol. 12, No1, Jan/Feb 1996, pp.9-16.