Practical Experience in On-Line Partial Discharge Measurements of MV Switchgear Systems Z. Berler, I. Blokhintsev, A. Golubev, G. Paoletti, V. Rashkes, A. Romashkov Cutler-Hammer Predictive Diagnostics 5421 Feltl Road, Suite 190, Minnetonka, MN 55343, USA Phone: 612-912-1358; Fax: 612-912-1355; Web site: www.partial-discharge.com; E-mail: goluba@ch.etn.com Abstract: On-line partial discharge (PD) monitoring has proved to be a fast, convenient, reliable and cost-effective tool in determining numerous areas of insulation deterioration within medium voltage switchgear lineups. The paper reflects the authors practical experience and concentrates on specific areas requiring further attention, such as sensor sensitivity, sensor calibration, PD recognition and discriminating various PD sources. Key words: Switchgear; Partial Discharges; Partial Discharge Discrimination; On-Line Monitoring; Sensor Calibration; Sensor Sensitivity. INTRODUCTION Medium voltage switchgear lineups and bus ducts usually receive less maintenance attention than the main electrical high voltage equipment. Their design is frequently less sophisticated, but the conditions of operation can be very unfavorable. Internal inspections are limited due to the necessary outage, therefore internal bus sections are sometimes not maintained for many years. At the same time the number of switchgear cubicles, especially in large power generation plants or in large industrial installations, can be very high, and the failure consequences can be extremely costly. For these reasons, there is a need to apply modern predictive technologies to switchgear systems. PD measurements were internationally recognized as an effective tool in evaluating the condition of insulation. They were accepted by IEEE in the Guide for PD measurements in switchgears [1]. Nevertheless, this technology was not immediately implemented into switchgear systems, primarily due to the low levels of PD observed. Progress in the development of measuring techniques [2] drastically increased the ability to reject noise during PD measurements and thus increased the sensitivity of these measurements. A new 8-channel analyzer (UPDA) is not limited to the measurement of single radio frequency pulses. It monitors the waveforms for several power frequency cycles and then identifies the PD pulses in the accumulated captured data using up to 5 independent systems for noise elimination. As a result, periodical on-line PD measurements in switchgear systems became technically feasible and provide maintenance personnel with a technology allowing for non-invasive, reliable, fast and cost-effective early detection of insulation problems. A comparison of the measurement results from different cubicles allows for identification of the positions of the defects. Other defects, such as corona or bad contacts, can also be identified. This allows for a planned approach to correct problems before the eventual unexpected failure. The first results of the application of this new technology were reported in [3]. This paper reports the further progress in switchgear monitoring with particular attention to specific areas that require further attention: PD recognition, discriminating PD sources, sensor sensitivity and calibration. TYPICAL PD SOURCES AND LEVELS Typical RF signal sources in switchgear and bus ducts are: Deterioration of bus supporting structures Bus insulation with dirty or moist surfaces Defects in cable terminations PD in potential and current transformers Locations where a ground is near a MV conductor Insulation structure of the circuit breaker Arcing contacts in the circuit breaker Arcing primary contacts (finger clusters and stabs) Arcing in loose high voltage and ground connections. We must note that common designs of bus insulation systems include several insulating components connected in series, for example, plastic insulation and small air gaps, or an insulation cover, small air gap and more insulation. Such designs are especially prone to developing discharges in the mentioned air gaps where the electric field concentrates. The discharges destroy the adjacent insulation and create the carbonized paths extending gradually to the grounded end of the insulation (Fig. 1). Unless the design is corrected, for example, by filling 1
Figure 1. A common problem in 15 kv switchgear: the PD damage and the pending failure occur at the point where the insulated bus passes through a bus support window leaving a small air gap. 2
the gaps with an insulating compound with a permittivity, which is close to the permittivity of the insulation, such defects will continue to grow until eventual failure, if not detected. PD in switchgear systems are commonly several times lower in magnitude as compared to PD in motors. Often the beginning of an insulation problem in switchgear or bus duct is accompanied with a PD signal of 15-50 mv. For this reason, the PD measuring device must be very sensitive. This becomes a difficult task, especially in electrically noisy environments. Noise cancellation is very important, and with the increase in electronic metering, noise conditions have worsened. Although it is common to reveal PD in 2.3-kV motors, our experience has shown that in only a few cases PD has been found at this voltage level in switchgear systems. It was quite common to find PD in 5 kv and 15 kv equipment [3,4]. Switchgears generally contain multiple T-connections (outgoing or incoming feeders or buses). This causes the PD signal to attenuate rapidly because the PD pulse energy splits at each T-connection. Such rapid signal attenuation can be effectively used for the purpose of PD site location. The cubicle producing PD can almost always be distinguished from all adjacent cubicles if a good signal-to-noise ratio exists. PD SENSORS USED IN SWITCHGEARS The installation of high voltage coupling capacitors connected to energized components inside switchgear was rejected due to the high cost and invasion into the initial switchgear design. There is no need for such sensors with the proper sensing equipment. The secondary circuitry provides several safe and convenient locations for connecting PD sensors. Particularly, the authors have over 5 years of experience in using current transformers circuits and the grounding of cable sheaths for that purpose. In the case of single-phase unshielded cables, we have developed flexible sensors which can be wrapped over the cable insulation. Generally, we apply two types of sensors in switchgear: Radio frequency voltage sensor (): a transducer with a high voltage ceramic capacitor inserted serially at the input and a radio frequency current transformer. This sensor has an extremely high input impedance at 60 Hz; Radio frequency current transformer (RFCT): the RFCT is a direction sensitive sensor allowing us to distinguish events occurring inside the cubicle from external pulse activity resulting from the incoming or outgoing feeder. The RFCT also allows us to get information concerning the connected cable splice or cable condition. Most MV breaker assemblies in switchgear have current transformers over the incoming bus and/or load cables to the breaker. Those current transformers have a stray capacitance between their primary (i.e. MV bus) and secondary winding, usually of about 100 pf. Low voltage secondary circuits can be conveniently used to connect a sensor, permanently or temporarily. The connection is made at the terminal block located in the front compartment of a breaker cubicle. This allows easy access and does not interrupt the normal operation of the cubicle. The sensor is connected to the neutral contact C 0 that is already grounded at some distance from the terminal strip. This assures a safe on-line connection and summarizes the signals from all three phases, but does result in a decreased sensitivity of approximately 50% when compared with the connection to individual phase wires C 1-C 3. Connecting to contacts C 1-C 3 also does not affect normal operation, but requires greater care since these points are connected to the breaker protective relays, and could result in an accidental trip. Contact C 0 is already at ground potential, therefore any connection risks are minimized When the sensor is connected to the neutral contact its sensitivity varies widely depending on the ground circuit inductance. It depends on the grounding wire length and is generally related to the design of the cubicle. It is desirable to have this ground wire inductance high, so at high frequencies the impedance of the grounding wire essentially exceeds the input impedance of the sensor (50 Ohm). That forces the majority of the high frequency signal through the, not through the grounding path. If the neutral point of the current transformer circuit is grounded directly at the terminal block, the inductance is very low, and it may be necessary to insert an additional inductance, by installing a clamp-on suppressor on the grounding wire. In 35 kv current transformers the design frequently includes a grounded metal shield positioned inside the molded insulation. This shield effectively excludes the capacitance between the primary bus and the secondary circuit, but provides its own capacitance to the primary bus, usually higher than of the secondary winding in a current transformer. If an RFCT sensor is placed over the shield grounding wire, then a very good PD recording is obtained. The installation of RFCT sensors over the cable shield ground connection in the rear compartments of cubicles will address both the insulation of the feeder terminations and the cable insulation. The cable shield must not be grounded until after it passes through the RFCT since this would provide an alternate PD path to ground. It is also common to find bus duct between a transformer and the main circuit breaker of a switchgear line-up. Because there is no cable, the sensor can be placed either in the current transformer circuit or on the bus duct enclosure. On the enclosure the best location for PD sensors are joints between the metal sheets. If adjacent sheets are interconnected with several wires, a RFCT can be installed on one of the wires, while the other wires can be equipped with suppressors, if 3
necessary. If these sheets are isolated a generator PD sensor can be connected between them. This sensor is similar in its circuitry to the sensor, and also contains a fuse for isolation and protection in the case of an internal short-circuit. The typical connection for PD sensors in switchgear cubicles is shown on Fig. 2, and pictures of the sensors installed are shown in Fig. 3. Cutler-Hammer offers new switchgears with pre-installed PD sensors. Older switchgear cubicles, of any manufacturer, can be field-retrofitted to include permanent PD sensors. procedure, which evaluates the sensitivity of the sensor for two cases: PD originating near the sensor (Sensitivity) and distant PD (Attenuation). The following terms define these two criteria of our calibration: Sensitivity to PD at the Sensor location (Sensitivity) - the sensor is calibrated by injecting a known charge close to a sensor to determine its sensitivity in terms of V/nC. Signal attenuation is not taken into consideration in this case. An average sensitivity for our designed sensors, when connected to current transformers neutrals, was found to be about 0.077 V/nC with variations of + 20%. For our RFCT sensors placed over cable shields grounding the sensitivity was evaluated as 0.15 V/nC + 17%. Cubicle #N RFCT Tie Breaker Cubicle #2 RFCT Main Incoming Breaker Figure 2. Typical connection of PD sensors in switchgear cubicles Figure 3. Sensors installed in the switchgear cubicles: Left- sensor connected to the neutral point of a current transformer secondary (in its terminal), in the front of the cubicle; Right- RFCT sensor placed around the cable shield connection to ground, in the rear of the cubicle. The bandwidth of our sensors is 1-20 MHz. The variety of bus configurations, current transformer designs and wire routings strongly affects the sensitivity of PD sensors installed in switchgear cubicles. This results in the need for an on-site calibration, which can be completed during the outage which was scheduled to install the permanent PD sensors. ON-SITE SENSOR CALIBRATING The PD transient wave detected by a PD sensor experiences a very high attenuation and shape modification while travelling through a line-up. This causes a difference in the responses of the sensor to signals originated at different points and, therefore, becomes a critical issue which complicates sensor calibration. Acceptable field calibrating procedures were discussed initially in [5]. We have developed a calibration Sensitivity to PD at a distant location (Attenuation) we inject a known charge close to a sensor and measure the response of other sensors located in the line-up. Thus we can obtain the ratio between magnitudes measured by different sensors along the line-up and evaluate the data necessary to adjust the parameters of an acceptancerejection algorithm of the recording system in the given switchgear. Particularly, for an adjacent cubicle the coefficient of attenuation, in average, was about 0.7 for sensors and 0.96 for RFCT sensors (the difference in attenuation is determined here mainly by the difference in sensor frequency response). Due to higher attenuation sensors are especially good in determining defect location. Attenuation data allows us to optimize the number of sensors installed in a line-up. Sensor placement can be determined by two methods: a preliminary review of the one-line and switchgear drawings, or following the results of field measurements using temporary sensors. In the latter case the sensors are temporarily placed in some location and, after initial measurements, are moved to another locations. When the location of the PD source is determined then the closest sensor can be used to evaluate the apparent PD charge. Note that if the distance from the sensor to the real PD source somewhat exceeds the distance to the point of the charge injection during calibrating, then the apparent PD charge will be somewhat underestimated, not overestimated. In spite of the approximation of this approach, it is still more accurate than millivolts alone. It creates the opportunity to compare data taken from different sensors, received on different apparatus and even on apparatus of different rated voltages. All of the above is true, with similar levels of approximation, for all quantities which can be derived from raw PD data, including PD power or PD current. We calibrate the PD measuring circuit in terms of apparent charge using the procedure similar to that which is prescribed in ASTM D1868 and IEC 270 Standards. Therefore, we inject a known charge from a small impulse generator through a 4
dosing capacitor into a known point and record the response of all sensors in Volts. Consequently, the sensitivity of a sensor for a particular injection point can be calculated. The same approach is used to calibrate any type of PD sensors in any line-up. We inject the known charge using aluminum foil, which is wrapped around an accessible part of the line-up s main bus or the bus/cable on the load side of a breaker, which is near a current transformer. Foil capacitance to the MV conductor is commonly in the order of several nano-farads; exceeding the dosing capacitor capacitance (100 pf) by 10 times or more. This capacitance is connected in series with the dosing capacitor. As the consequence, the dosing capacitor limits the injected charge. Therefore, the injected charge can be calculated as the product of the pulse magnitude and the dosing capacitance. A small RFCT is additionally installed in the charge-injecting circuit and measures injected current providing an additional method to measure the injected charge (the injected charge is calculated as an area under the oscillogram of the injected current. The first peak of the oscillogram is used for this calculation). In all cases, the difference in injected charge evaluation by the two methods varied no more than 20%. PD EVENT RECOGNITION The best algorithm for noise rejection and to detect only PD originating in cubicles, is by comparing the pulse magnitudes detected by different sensors along the switchgear line-up. If an event corresponds to a PD signal originated in one cubicle, then magnitudes of the signal detected by sensors in other cubicles should attenuate depending on the distance from the PD location. This is due to the fast attenuation of the high frequency signals while propagating along the line-up. Figure 4 presents a typical distribution of magnitudes along the lineup from the PD signal originated in cubicle # 1. External noise and the software acceptance curve are also shown. In addition, phase-resolved distributions, discharge patterns, etc. were used to discriminate between internal insulation discharges and corona, or between PD related to insulation versus arcing in the contact or finger cluster area. PRACTICAL RESULTS Initial surveys and periodical measurements have been completed during the last three years on over 200 line-ups Attenuation 1.2 1 0.8 0.6 0.4 0.2 0 1 3 PD signal attenuation 5 7 Cubicle ## 9 11 PD signal response External noise response Algorithm threshold Figure 4. PD signal attenuation along the line-up, external noise num magnitude distribution and software acceptance algorithm threshold. (1200 circuit-breaker cubicles) and dozens of bus ducts were tested for PD using the technology described above. Our methodology has allowed us to check a large number of cubicles in one day on-site, while equipment is energized. Numerous defects were revealed: primarily ranging from components with a moderate PD activity, to locations with severe insulation deterioration. Additionally, different defects were found, such as bad contacts, loose insulators, bad groundings, reduced distances between MV parts and LV secondary wiring, carbonization and tracking along insulation surfaces, etc. The most dangerous and most frequent defects were observed in industrial switchgear line-ups, where environmental conditions are especially harsh (contam-ination, chemicals) and the quality of design and maintenance is somewhat less than in traditional electric utilities. CONCLUSIONS On-line PD technology has shown to be a fast, convenient, cost effective and non-invasive tool for detecting areas of insulation deterioration in MV switchgear systems. In addition, other related defects such as loose insulators, reduced clearances to ground, and deteriorated contacts have been found. An engineering review of the switchgear and the calibration of sensors provide results with the greatest accuracy. On-line measurements, using temporary sensors can also be considered for base-line data. This technology has been successfully applied at numerous electric utilities and industrial facilities. REFERENCES 1. Guide for Partial Discharge Measurement in Power Switchgears. IEEE Std 1291-1993. 2. Berler, Z., Golubev, A., Romashkov, A., Blokhintsev, I., A New Method of Partial Discharge Measurements, CEIDP-98 Conference Annual Report, Atlanta, GA, October 25-28, 1998, pp.315-318. 5
3. Kane, C., Lease, B., Golubev, A., Blokhintsev, I., Practical Experience of On-Line Partial Discharge Measurements on a Variety of Medium Voltage Electrical Equipment IEEE-PCIC 98 Conference, November 1998, Indianapolis, IN. 4. G. Paoletti, G., Golubev, A. Partial Discharge Theory and Applications to Electrical Systems, IEEE-IAS Annual Pulp & Paper Industry Technical Conference Record, June 22-25, 1999, Seattle, WA,, pp.124-138. 5. Berler, Z., Blokhintsev, I., Golubev, A., Paoletti, G., Romashkov, A., RTD as the valuable tool in partial discharge measurements on rotating machines Proceedings: EIC/EMCW 99 Conference, Oct. 26-28, 1999, Cincinnati, OH, pp. 205-210. 6