RTD as a Valuable Tool in Partial Discharge Measurements on Rotating Machines

RTD as a Valuable Tool in Partial Discharge Measurements on Rotating Machines Z. Berler, I. Blokhintsev, A. Golubev, G. Paoletti, A. Romashkov Cutler Hammer Predictive Diagnostics Abstract: This paper presents the authors practical experience in the on-line measurement of partial discharges in medium voltage motor and generator stator windings using the RTD as a partial discharge detector. Results of off-line calibration on several machines are also presented. Introduction On-line measurement of partial discharges (PD) has proved to be an effective tool in evaluating the condition of stator insulation in high and medium voltage electric motors and generators [1]. This method is widely used in addition to the traditional off-line insulation tests performed during scheduled outages. Most of PD technologies available now on the market for on-line measurements function within the radio-frequency band of PD signals. Such technologies have the common problem resulting from very rapid attenuation of the high frequency signal as it travels through the winding. Therefore, sensors commonly installed at winding terminals have a limited zone of sensitivity and provide valuable information for that zone only [4]. The evident solution to this problem is the use of PD sensors imbedded into the winding to get information on the winding itself. Some of the PD technology vendors suggest installing specially designed sensors into a winding, but this approach is relatively expensive and requires an extensive machine outage and invasion into the winding assembly. Alternatively, most of the HV machines already have RTD detectors embedded into the winding by the manufacturer and these detectors can be used for partial discharge measurements [2,4]. Cutler- Hammer has over two years of experience using RTDs as PD detectors. The special PD transducer (RFVS) was designed for connection to the RTD wire at the RTD terminal block located on the frame of the motor or generator. The transducer does not disturb temperature measurements and only passes high frequency PD signals to the PD instrument. Over 300 machines, primarily HV motors, were tested during the past two years with good results. RTDs were used for both the initial survey/evaluation and for ongoing periodic measurements and data trending. RTDs are currently very effective in trending of machine PD activity when used with an analyzer that can effectively reject noise and process PD data. With sensor calibration the use of RTDs can be further applied to allow comparisons between different machines. The issue of sensor calibration requires further evaluation to help advance the technology and use of RTDs in determining the machine s insulation condition. Several machines have been calibrated, but more field data is necessary for the establishment of good quantitative data. This paper proposes a calibration procedure and presents the results of off-line calibration on several machines. The problems and the vision of future improvements are also discussed. Why do We Need the RTD as a PD Detector? The traditional approach for PD detection in rotating machines uses sensors installed near machine line terminals. What is the value of data obtained from such sensors and what additional information is required to reliably assess winding insulation condition? Based on our experience, PD sensors located near machine line terminals provide valuable information for line terminals and, possibly, for a ring bus, but not for the winding. The example below presents the data obtained on-line from a 37,000kVA, 13.8kV generator and confirms this statement. The generator has an 80pF coupling capacitor installed on each line terminal and also 12 RTDs evenly distributed around the circumference of the stator core. Six RTDs are placed on the exciter and six on the turbine ends of the machine. Figure 1 presents three sets of oscillograms taken from 80 pf coupling capacitor (Plot #1) and from RTD#1 (Plot #2) and RTD#7 (Plot #3). All of them are on the same phase A. RTDs are also located in the same slot on the exciter and turbine ends respectively. The oscilloscope was triggered from the PD pulse originating near the line terminals on phase A and also from the pulse originating near each of the two RTDs. One can see that the coupling capacitor provides no response to the PD originating in the winding on either side of the generator. The opposite is true as well. The attenuation of PD signal along the slot is also very high and exceeds 10 times. Therefore, for a complete analysis, it is necessary to install additional sensors into a winding or to use RTDs to get information about the winding condition. 1 > 2 > 3 > 1) CC_A: 40 mvolt 200 ns 2) RTD01: 40 mvolt 200 ns 3) RTD07: 40 mvolt 200 ns Triggering from line terminal PD. 1 > 2 > 3 > 1 > 2 > 3 > 1) CC_A: 40 mvolt 200 ns 2) RTD01: 40 mvolt 200 ns 3) RTD07: 40 mvolt 200 ns Triggering from slot PD on the exciter end. 1) CC_A: 40 mvolt 200 ns 2) RTD01: 40 mvolt 200 ns 3) RTD07: 40 mvolt 200 ns Triggering from winding PD on the turbine end. Page 1 of 7

Figure 1. PD pulse attenuation in a winding The effects of signal attenuation discussed above may cause mistakes in evaluating stator winding insulation condition, if sensors located at machine line terminals were the only ones used for assessment. The example below at figure 2 presents PD test results of three 13.8kV motors of a similar design at the same facility. All describing motors have permanent RFCT sensors placed on surge capacitor grounding conductors. The test was also provided with temporal PD sensors connected to RTD terminals. Figure 2 (top) shows maximum PD magnitudes recorded from RFCT sensors. Based on these results, we can conclude that the motor1 is in a good state and the motors 2 and 3 have moderate level of discharges. The data from three RTD s showed highest reading for each motor are shown on figure 2 (bottom). Conclusions, based on this data, are the same as above for the motors 2 and 3. But the conclusion is different for the motor 1. It has high level of PD at the zone of RTD01 and is the first candidate for additional testing and internal inspection. RFCT Sensors Motor1 Motor2 Motor3 RTD Sensors Motor1 Motor2 Motor3 Figure 2. PD maximum magnitude by RFCT s and RTD s. Why do We Need Calibration? The real need to calibrate or normalize a PD measuring circuit on a rotating machine exists today. As the science of Partial Discharge measurement was making its first steps, it was agreed that calibration on rotating machine is a very difficult procedure. Therefore, it was decided to utilize the Partial Discharge Magnitude parameter measured in millivolts or Volts [1]. Based on that, the only valid procedure of using PD data is through relative comparison of PD data collected using the same vendor s technology over time on the same machine or between similar machines. This situation was bearable while the number of PD technology users was relatively small and most of sensors installation, data collection and interpretation were provided by a qualified expert. Now the situation is different. As PD technology is maturing, reallife cases reveal the need for a standardized PD measuring circuit calibration procedure. For instance, paper [3] reported a 19,000HP motor failure just because the 80pF couplers were installed about 4m away from the motor line terminals. That caused signal attenuation by a factor of 5 and resulted in misinterpretation of the PD data. As a result the authors of [3] have now normalized all their PD sensors with pulse generator and oscilloscope and are now using normalized data for relative comparison between monitored motors. This is an example of how uncalibrated sensors defeated the original predictive expectations of the on-line PD sensors. The reason of such a difference can be easily understood from simplified diagram below showing typical sensors connection in a motor terminal box. PD in a winding or near line terminals produces small surge traveling to the feeder. In general, pulse current induced by PD, which is really detecting by PD measuring instrument, is split into several branches. Part of the current goes through 80pF coupling capacitor, part of the current goes through surge capacitor circuit and the rest of it goes through the feeder. The current distribution through described branches and part of PD signal detected by any particular PD sensor depends upon impedance of each branch including inductance of every used wire. Therefore, the presence or absence of any element in this diagram and their wiring, how many cables is used per phase and a surge impedance of the single cables and many more reasons - all this will affect PD reading from sensors. PD Feeder Winding Surge Impedance Surge 80 pf of Cable 500,000pF 5-30 Ohms 50 Ohm RFCT RFCT Impedance Figure 3. Sensors layout in motor terminal box Authors, who are using universal PD analyzer, which is able to read PD data from PD sensors installed by different PD technology vendors, face the problem described above on everyday basis. The example presented below on figure 4 compares PD magnitudes from 80pF couplers and RFCT s placed on surge capacitor grounding conductor measured on-line on 13.8 kv motor at the same time. Both sensors were of the same manufacturer. The first three bars present data obtained from 80pF coupling capacitors on the phases A, B and C and three last bars are related to data obtained fron RFCT s. One can see two times difference in signal magnitudes. Where is the truth? Figure 4. PD magnitude from two different types of sensors on 13.8kV motor 1 The table below summarizes our experience with different sensors, we have used in both off-line and on-line PD tests on medium voltage motors. The better numbers, as a rule, can be reached at offline test, when all auxiliary equipment can be disconnected. In any case, the variations in sensitivity even for the same sensor can be very significant and confirm the necessity of a calibration. 1 The table does not cover all possible variations in measuring circuits and sensor layouts and can be somewhat different for PD instruments using different frequency band. Page 2 of 7

Sensor Type (1nF loaded 50Ohm) (80pF loaded 50Ohm) RFCT(5 Ohm) on surge capacitor ground Table 1 Sensitivity Comments Range [V/nC] 2.0 0.5 Depends upon surge impedance of auxiliary devices and feeder or bus. 1.0 0.2 Same as above and inductance of measuring circuit. 0.3 0.1 Depends upon surge impedance of auxiliary devices and feeder or bus. discharge value can be greater for PD near the sensor, but it can not be less. In spite of the approximate character of this approach, it is still more accurate than millivolts alone. It creates the opportunity to compare data taken from different sensors, taken from different machines and even for machines of different rated voltages. All of the above is true, to the same extent of approximation, for all quantities which can be derived from raw PD data. These quantities could be PD power or PD current and so on. The question left without an answer is the applicability of such approximation or, in other words where is the limit, beyond which a comparison looses any practical sense? The answer to this question is in the term described below and called the Zone of Sensor Sensitivity. RFCT(12.5 Ohm) on surge capacitor ground 0.8 0.2 Same as above. RTD 0.3 0.05 Depends upon machine design and less upon leads length. Another issue that further promotes the need for sensor normalization, or calibration, is the increasing flow of practical data collected by different vendors. For instance, the author of [5] reported the analysis of over 13,000 test samples. This data is not very useful for other users since normalization to conventional measurement units was not done. The above clearly indicates the need to establish a field calibration procedure to allow for the future advancement of the benefits of the varying PD technologies available today. Without such flexibility, the end user is limited to possibly outdated technology, and will not be able to benefit from advancements in the future. Calibration Procedure and Units The approach described below is offered as a possible and useful solution to the development of an acceptable field calibration procedure for various PD sensors. It is well known that the partial discharge transient wave, which is detected by a PD sensor, experiences very high attenuation and shape modification while travelling through the stator winding. This causes a difference in the response of a sensor to a signal originating in different points in the winding and therefore, becomes the main problem complicating sensor calibration. The ideal solution is in calibration of every sensor to all possible PD locations. This approach is impractical due to the extreme complexity and unknown PD source location during on-line testing. Two terms are proposed to establish a uniform calibration standard, and provide a basis for consistent calibration between various PD sensors and sensing technologies. Sensitivity to PD at Sensor Location (Sensitivity) - we can calibrate a sensor by injecting a known charge close to a sensor and determining its sensitivity in terms of nc/volt. Such sensitivity applies primarily for partial discharges originating close to a sensor. Signal attenuation is not taken into consideration in this factor. On the other hand, attenuation is a very important factor for PD signals distant to a sensor. Therefore, if sensitivity defined as it is described above is used, data obtained on-line from a sensor in terms of nano- Coulombs presents the lower limit estimation of an apparent charge for discharges originating close to a sensor. In other words, a Zone of Sensor Sensitivity - This term is more qualitative than quantitative. It limits the boundaries of a spatial zone that can be assessed using a particular sensor. We use 20dB attenuation of a signal as the criterion to determine the border of the Zone of Sensor Sensitivity. One can not evaluate PD data reliably beyond that zone of a particular sensor. From the example given above, we can evaluate the line terminal insulation condition based on the 80 pf capacitor readings, but we can not seriously discuss the winding condition due to the inability of the line terminal PD sensors to detect winding related PD signals. Any conclusions beyond the Zone of Sensor Sensitivity would be just a guess based on previous experience on similar machines with similar operating conditions, but not on the real data. The knowledge obtained based on the Zone of Sensor Sensitivity, for various PD sensor technologies, allows for better planning concerning the number and location of sensors for a particular application and provides a check on the reliability of the information obtained. The Zone of Sensor Sensitivity can be determined during off-line calibration. We calibrate the PD measuring circuit in terms of apparent charge using the procedure similar to that described in ASTM D1868 or IEC 270 Standards. Therefore, we inject a known charge through a differentiating (dosing) capacitor into a known point and record the response of all sensors in Volts. Consequently, sensitivity of a sensor in terms of nano-coulomb per Volt for a particular injection point can be calculated. Zone of Sensor Sensitivity can also be determined. Figure 5 presents the example of the calibrating circuit for a radio frequency current transformer placed on the surge capacitor-grounding conductor on a motor. The same circuit is used to calibrate any type of PD sensor. Aluminum foil is wrapped around the accessible part of the winding or the bus bar near the calibrated sensor. The foil capacitance to the HV conductor is commonly in the order of several hundreds to one thousand of pico- Farads. This exceeds the capacitance of the dosing capacitor by about 10 times. This capacitance is connected in series with the dosing capacitor. As a consequence, the dosing capacitor limits the injected charge. Therefore, an injected charge can be calculated as the product of pulse magnitude and the dosing capacitance. A small RFCT is additionally inserted into the charge injecting circuit and measures injected current. This is an additional method to obtain an injected charge. An injected charge is calculated as the area under the oscillogram of the injected current. The first peak of the oscillogram is used for injected charge calculation. In all cases, we have had within 20% agreement between the injected charges measured in both ways. This proves that either of the two methods can be used. Page 3 of 7

Response from all available sensors is measured for every point of pulse injection. Therefore, cross-coupling coefficients between different sensors can additionally be determined while obtaining the sensitivity of any particular sensor. In order to detect the Zone of Sensor Sensitivity, a pulse is injected into different points distant from a calibrated sensor and a distance resulting in 20dB attenuation is determined. Figures 6 a and b show photographs of the infield calibration on a 800MW 2-pole generator. Aluminum Foil RFCT and Dosing Figure 5. Calibration Circuit. Figure 6a. Pulse Injection into the Endwinding Area. Sample Calibration Results The results of calibration on small generator and two HV motors are presented below. 12.5 MW, 13.8 kv Generator This 42-slot generator is equipped with 12 RTDs distributed evenly around the circumference of the stator core. Two RTDs are placed in a slot, one on the exciter and another on the turbine end. RTDs 1-6 are placed on the exciter end and RTDs 7-12 are placed on the turbine one. Therefore, 6 slots are equipped with a RTD. The distance between the two nearest slots containing a RTD is 6 slots. Fourteen signals were recorded simultaneously for every injection point 12 RTDs were connected to the instrument through our specially designed RFVS sensors and PD analyzer s signal conditioning module; T1 line terminal was connected to the instrument through a 1,000pF, 20kV coupling capacitor sensor and our PD analyzer s signal conditioning module; RFCT measuring injected current was loaded with 50 Ohms at the oscilloscope end. Figure 7 shows the RTD response in terms of Volts per nano- Coulomb for pulse injection into four different points. Three of them are related to slots containing RTDs and one to the Slot 22, which is between RTD9 and RTD10. The other two related to RTDs were placed in the Slot 18, which is at RTD # 10, and Slot 25, which is at RTD # 9. All three RTDs showed approximately the same sensitivity. The response drops by about 10 times if the pulse is injected 3 slots away from the RTD. The attenuation of a signal along a slot is about 5 times. Figure 8 presents the coupling capacitor response to a PD injection into different slots on both the exciter and the turbine ends. The response drops very rapidly while moving the injection point away from the line Slot 18. At the same time, this sensor is insensitive to any pulse injected at the turbine end. Both Figures 7 and 8 indicate that a different level of criteria is required for evaluation of PD located internal to the winding versus at the line terminals. To detect PD near a winding RTD, using the RTD as the PD sensor, a sensitivity of ~ 0.06V/nC would be applied whereas PD occurring at the line terminals is detected with a sensitivity of ~ 1V/nC. Using only a line-terminal PD sensor would mask the low level of internal PD, which is detectable using the RTD. For example, equal partial discharges at the line terminals versus internal to the winding near an RTD would yield a measurement of 1 volt for the PD at the line terminal and less than 0.1V for the same magnitude of PD internal to the winding. Such a wide range of voltage measurements, at the line terminal, makes it almost impossible to detect partial discharges internal to the winding using only line-terminal PD sensors. With separate measurements obtained using RTD s, these internal partial discharges can now be detected using the sensitivity of ~ 06V/nC; therefore a measurement of 0.1 Volts at the RTD can be properly evaluated without the masking of higher voltage measurements associated with having only the line terminal type of PD sensor. Figure 6b. Pulse Injection into Line Terminals Area. Page 4 of 7

0.12 Sensors Response to Injected Charge steel shield. Sensitivity for such RTDs is commonly in the range from 0.015 to 0.02 Volt per nano-coulomb. Sensitivity [V/nCl] 0.1 0.08 0.06 0.04 0.02 0 RTD1 RTD2 RTD3 RTD4 RTD5 RTD6 RTD7 RTD8 RTD9 RTD10 RTD11 RTD12 Sensor Name Sl18 RTD10_Tr Sl22_Tr Sl25 RTD9_Tr Sl32 RTD8_Tr Figure 7. RTD Response to Injected Pulse The above results of RTD calibration confirm that RTDs can be used as PD detectors in PD technologies based on high frequency pulse recording. The difference in sensitivity between different machines may be high, therefore, a calibration is recommended for quantitative comparison between different machines or between sensors of different design. Note that relative comparison over time or between machines of the same design is not a problem without any calibration. Several examples of PD tests using RTDs presented below also confirm that RTDs are a very valuable tool in PD technology. Sensitivity [V/nC] 2.5 2 1.5 1 0.5 0 Coupling Response to Injected Charge Exciter End T1 Line Sl18 Sl20 Sl25 Sl29 Sl32 Sl18 Sl22 Sl25 Sl32 Injection Point Turbine End Figure 8. Coupling Response to Injected Pulse. 7550 HP 13.2 kv Synchronous Motor The motor has 72 slots and is equipped with 12 RTDs placed at the ring bus side of a slot. RTDs are distributed evenly along the winding, every six slots. The sensitivity of different RTDs varies from 0.05V/nC to 0.07V/nC with an average value of 0.06V/nC. RTDs located at a greater distance from the RTD terminals showed sligthly less sensitivity, possibly related to the RTD wire routing.. The signal attenuation from the opposite end was very stable for all RTDs and varies from 4.5 to 5.3 times. Off-line Test This off-line test was performed on a 12.5MW, 13.8 kv generator described above. Test voltage of 8 kv (phase to ground rated voltage) was applied to one phase at a time. The other two phases were grounded. PD data was collected in the form of traditional phase-resolved PD distribution with phase resolution of 2 degrees and magnitude resolution of 0.5dB by Cutler-Hammer Twins PD analyzer [6]. The sensitivities obtained from the calibration for RTD s and the 1,000pF coupling capacitor were used when processing PD data. The flat projection of the Phase-Resolved PD Distribution (PRPDD) on the phase-magnitude plane (top view) obtained during B phase test is presented on Figure 9. Figure 10 shows integral quantities calculated for data taken from all of the PD sensors in three subsequent tests. In spite of significant difference in signal magnitude (in terms of millivolts) obtained from sensors of different types (Fig. 9), one can see the reasonable scatter in integral quantities calculated from the coupling capacitor and RTD data using the sensitivity of the sensor. 8000 HP 13.2 kv Induction Motor This 4 pole motor has 96 slots and is equipped with 12 RTDs. Two RTDs are located in the same slot approximately in the center of the core. Six slots in total are equipped with RTDs. Wires from both RTDs placed in the same slot come out of the slot in opposite directions. This motor has a large diameter and a relatively short core of about 1.5 m. In spite of the short core and approximately centrally located RTD, they show 6 7 times better response to pulses injected from the side of the RTD wires.. A wire works as a RF antenna as well and therefore the effective length of antenna is greater for a pulse injected from a RTD wire side of the core. The motor also showed moderate scatter in RTD sensitivity for different RTDs. It varies from 0.2 to 0.28 Volt per nano-coulomb, which is about +15 to 20%. RTDs located closer to the RTD terminals at the motor showed higher sensitivity. Higher signal attenuation as a result of longer wires is the most probable reason for the observed scatter. The effect of signal attenuation from the RTD to the RTD terminals is most significant for RTD wires protected by a spiral Figure 9. Phase B PD data. Page 5 of 7

Phase 1 Phase 2 Phase 3 the same magnitude range and C-phase showed higher PD activity at the line terminals as well as inside the winding. This is evident by correlating the Maximum Apparent Charge for Phase C, shown in Figure 12, with the higher Maximum Apparent Charges also shown for RTD #3 and #6, both of which are installed at Phase C. This is also visually evident by reviewing the PRPDD of Figure 11 below. Phase C (CC_C) indicates more PD activity, and this is also evident for RTD 3 and RTD 6 below. Figure 12. Maximum Apparent Charge. Figure 10. Charts show PD Intensity, Maximum Apparent Charge and Pulse Repetition Rate Respectively. (All Calculated from PD data above 0.1 nc.) On-line Test This on-line test was performed on a 7500 HP 13.8 kv motor. The motor is equipped with three permanent radio-frequency current transformers (RFCT) placed on the surge capacitor grounding conductor in the motor terminal box and with 6 RTDs embedded into the winding. RTD 1 & 4, RTD 2 & 5 and RTD 3 & 6 are installed on the phases A, B and C respectively. RFVS sensors were used to obtain temporary connections to RTD terminals in the RTD connection box on the motor frame. The flat projection of PRPDD from all available sensors is presented on Figure 11, and maximum apparent charge is on Figure 12. It is very important to mention that data presented for each sensor is unique for a particular sensor. Any possible crosscoupling from sensor to sensor was rejected by the Twins analyzer. As one can see, both magnitudes from the RFCT and RTD are in approximately Conclusions 1. The attenuation of high frequency signals in rotating machine windings is the main factor which complicates PD measurements on such equipment. Sensors commonly placed near machine line terminals are insensitive, as a rule, to distant PD originated internal to the machine winding. Additional sensors placed in the winding are required to reliably detect partial discharges. Resistive Temperature Detectors (RTDs) already placed in a winding by the machine manufacturer can be used as high frequency antennas to collect partial discharge pulses from the depth of the winding. The use of RTDs allows PD data to be obtained without an outage to install invasive sensors into winding slots. 2. RTDs commonly have good sensitivity to PD originating nearby. Therefore, if used complimentary to conventional PD detectors, these provide better information on partial discharges in the entire stator winding and yield a more reliable winding insulation assessment. 3. Calibration is recommended to scale PD data taken from RTDs, and other types of PD sensors, of different machines, to the same base. This calibration correlates the characteristics of the various PD sensors which may have different characteristics for high frequency applications. Such calibration would also correlate different responses from similar RTD s, or other PD sensors, to the same discharge on machines of different designs. Note that relative comparison over time or between machines of the same design is not a problem without any calibration. In this case, the use of RTD s can be trended, similar to sensors which would require an outage for invasive installation. 4. The calibration procedure was designed with the aim to scale sensors of different design to the same basis. Two terms Sensitivity to PD at Sensor location and Zone of Sensor Sensitivity is suggested to perform sensor calibration in terms of apparent charge. Figure 11. Phase-Resolved PD data. 5. Over two years of practical experience confirms that RTDs can be used as a very valuable tool for on-line and off-line PD measurements on a rotating machine (with an adequate PD analyzer that can process data and efficiently reject all types of Page 6 of 7

noise). The key advantage is that the use of PD predictive technologies can be easily implemented with existing RTDs. References 1. Draft of the IEEE P1434 Guide to Measurement of Partial Discharges in Rotating Machinery, 1998. 2. K. Itoh, Y. Kaneda, S. Kitamura et al. New Noise Rejection Technique on Pulse-by-Pulse Basis for On-Line Partial Discharge Measurements of Turbine Generators, IEEE PES Paper # 96WM 154-5-EC 3. Osman M. Nassar, Thani S. Al-Anizi. Saudi Aramoco experience with partial discharge on-line motor monitoring equipment, IRIS Rotating Machine Technical Conference, March 10-13, 1998, Dallas, TX USA. 4. I. Blokhintsev, M. Golovkov, A. Golubev, C. Kane Field Experiences on the Measurement of Partial Discharges on Rotating Equipment, IEEE PES 98, February 1-5, Tampa, FL 5. V. Warren On-Line Partial Discharge Monitoring: Where do We Stand and What Next? EPRI Utility Generator and Predictive Maintenance & Refurbishment Conference, December 1-3, 1998, Phoenix, Arizona. 6. Z. Berler, A. Golubev, A. Romashkov, I. Blokhintsev A New Method of Partial Discharge Measurements, CEIDP-98 Conference, Atlanta, GA, October 25-28, 1998. Page 7 of 7