PARTIAL DISCHARGE DETECTION AND LOCATION IN TRANSFORMERS BY PERFORMING PARTIAL DISCHARGE TESTS IN OIL USING UHF SENSORS by K.K.JEMBU KAILAS and W. ADITHYA KUMAR L&T CONSTRUCTION L&T CONSTRUCTION KKJK@LNTECC.COM AADI@LNTECC.COM
Objective Partial Discharge detection & location in Power transformers and to evaluate the most important property of UHF sensor i.e. SENSITIVITY through preliminary testing. To de-noise the captured signals and to extract the original PD signal after de-noising. To evaluate the approximate location of PD source
Partial Discharge (PD) PD - Intermittent nanosecond range pulses Causes of PD. Weak insulation - When electric field across the insulation > dielectric strength of the insulation = PD. Arcing due to poor contacts 3. Corona 4. Surface discharges Effects of PD. Insulation degradation over a period of time. Equipment failure, Effect on men & materials, Disruption of economic activities, Customer dissatisfaction, Penalty clauses 3
PD Detection using UHF Principle and aspects Electrons accelerated under high electric field and immediately brought to rest (nano sec range pulses) produce Ultra High Frequency (UHF) EM waves. UHF waves = 300 MHz -3000 MHz Non conventional method energy from EM waves can be captured using sensors and their outputs can be shown in V or mv PD pulses can be detected if wave length of UHF signals < tank dimensions Narrow pulse, more spectral energy & vice versa 4
UHF sensors & UHF PD Detector Module Internal & External sensor respectively, internal sensor placed inside the drain valve, while external sensor placed in specially designed dielectric windows on the transformer, both require modifications on transformer tank. 5
Conventional Method (IEC 6070) UHF Method Offline measurements Online measurements Measurements at factory Measurements at factory and site Considers the transformer as a single unit. Hence fault location is not possible Able to locate multiple faults as the system considers each fault location as a localized unit Basic method to analyze the Can be coupled with conventional method overall strength of insulation for diagnosis, while testing transformers in in the factory the factory. Highly sensitive to noise Less sensitive to noise, as this operates in UHF range. Even corona and other noises happen only below this range. Hence sensitivity is high 7
Types of Sensors Patch antenna Internal sensor External sensor 8
Preliminary testing of Sensors Before inserting the sensor directly in to the drain valve of Transformer it is necessary to evaluate the performance of antenna. More over the charge levels in a transformer will be in the order of some thousand pc. So the sensitivity of the sensor is evaluated by keeping the antenna at a relative distance from PD source. 9
Sensitivity verification of UHF Conical Monopole Antenna Experimental Set-up LDIC PD measuring system 0
Applied voltage (KV) v/s charge (pc) The charge (pc) was linearly varying with applied voltage i.e. it has a linear relationship. The center frequency of received signals at PD inception voltage was 963MHz. The minimum energy of the PD signal is 60μJ with sensitivity 90pC.
Experimental Set-up
Preliminary testing in Oil Internal UHF sensor Signal captured at 0.KV Signals captured by all the 4 sensors at KV 3
De-noised signal 4
Results Type of sensor Distance Max Peak to test (m) voltage peak (mv) voltage (mv) Frequency Energy Time Phase Power (MHz) (pj) period (degre (dbm) (ns) es) Time difference t s.4 47 376 960 49.03 3.8 55.8 46.53 s.4 400 970 38.6 3.5 83.75 4.7 s3.4 45 650 979 40.9 77. 5.47 s4.4 45 50 960 37 8.7 46.3 47 s.4 75 86 000 6 5.7 5.8 6.94 s. 95 6 979 306 4.8 50.7 59.9 s3.4 43 79 965 0 0.6 04.4 60.8 s4 0.75 6 3 980 45 8.9 5.48 64.98 Triggering Triggering voltage Channel (mv) t3 t4 4.3 43. 45. 50 C 43.9 49.5 4. 50 C in Air in Oil 5
Observations from Oil test The Internal sensor (conical monopole antenna) has detected the corona discharge at an inception of voltage of 0.KV where as all the other external antennas (Patch antennas) did not capture it. As the distance between internal sensor and the electrodes was 8cm where as the distance between external sensors and the electrodes is m, the UHF internal sensor detected the signal. After repeating the experiment keeping all the sensors at same positions and with voltage being increased, external sensors detected corona at KV, where as the UHF internal sensor detected at 0.6KV. The results show that the UHF internal sensor is more sensitive than the external sensors when placed very close to the PD source without actually knowing the level of PD. As the test was carried out under controlled conditions and without any ground noise de-noising the signal was done in oscilloscope. 6
Experiment on real time Transformer Internal Sensor Mounted in to one of the valves 7
UHF FFToutput 0.35 De-Noising Wavelet De-noising Principle Amplitude(volts) 0.3 FFT of original signal Maximum Amplitude = 0.669 volts at a frequency of 979 MHz 0.5 0. 0.5 0. 0.05 Original signaluhfcaptured by sensor output 0.5 0.06 0-0.05-0. -0.5-0. -.5 0.0 0-0.0-0.5 0 0.5 Time(seconds) -0.06.5 -.5-7 x 0 4 6 8 FFT offrequency(hz) Denoised signal 0 9 x 0 Maximum Amplitude is 0.53 at 000 MHz 0.08 0.06 0.04 0.0-0.04-0. Amplitude(volts) 0.05 0 0 0. 0.04 Amplitude(volts) Amplitude(volts) 0. De-noised signal using wavelets in MATLAB Denoised signal - -0.5 0 Time(seconds) 0.5 0 0.5-7 x 0 4 6 Frequency(Hz) 8 0 9 x 0 8
PD location using UHF Time of Arrival Approach There will be definite time difference between arrival of signal from a particular PD source at each sensor. The time difference governed by minimum time delay path traveled by the EM wave front to reach each sensor. Distance traveled (D) = EM Wave speed (v) x time (t) or more distinct but repeatable time period = multiple PD sources 9
Flow Chart of TDOA Enter the sensor coordinates in D/3D Calculate the sensor position values Call TDOA Read the time domain signal from each sensor Cross correlation TDOA values Solve the non-linear equations Newton Raphson Iterative method Non-Iterative method Source location 0
Evaluation of -Dimensional PD source Location using MATLAB Sl.no Applied Sensor Sensor Sensor 3 Time Computed Source Computed Error Error voltage Position Position Position Difference Of Velocity of position position in in (KV) (x,y) cm (x,y) cm (x3,y3) cm Arrival (TDOA) electromagneti (x,y) cm (xc,yc) cm X (%) Y (%) 50,00 50.069, 0.4-0.07 0.4-0.07 0.4-0.07 0.4-0.07 0.4-0.07 0.4-0.07 4.4 (50,00) (50,75) (50,0) T T3 c waves (ns) (ns) Cm/sec...3*0^0 99.93 4.6 (50,00) (50,75) (50,0)...3*0^0 50,00 50.069, 99.93 3 4.7 (50,00) (50,75) (50,0)...3*0^0 50,00 50.069, 99.93 4 4.8 (50,00) (50,75) (50,0)...3*0^0 50,00 50.069, 99.93 5 7 (50,00) (50,75) (50,0)...3*0^0 50,00 50.069, 99.93 6 3.3 (50,00) (50,75) (50,0)...3*0^0 50,00 50.069, 99.93 Comparison of actual source and evaluated source with sensor as reference
Evaluation of -Dimensional PD source Location using MATLAB Sl.no Applied Sensor Sensor Sensor 3 Time Difference Computed Source Computed Error in Error voltage Position Position Position Of Arrival Velocity of position position X (%) in (KV) (x,y) cm (x,y) cm (x3,y3) cm (TDOA) electromagnetic (x,y) cm (xc,yc) cm T3 T3 waves (ns) (ns) Cm/sec Y (%) 0.4 (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 5.85,03.4 5.7 3.4 4 (50,00) (50,75) (50,50).5 5 *0^0 50,00 53.55,0.6 6.9.6 3 4. (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 54.35,0.7 8.7.7 4 4. (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 54.4,0.4 8.4.4 8.4.4 8.4.4 8.4.4 5 4. (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 54.4,0.4 6.4 (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 54.4,0.4 7 8 (50,00) (50,75) (50,50) 3.4.7.3*0^0 50,00 54.4,0.4 Comparison of actual source and evaluated source with sensor 3 as reference
Results When the FFT of original signal was performed, the dominant frequency was 979MHz and the dominant frequency in denoised signal was found to be 000MHz. This shows that the noise which the sensor captured even between 900MHz-000MHz is eliminated The amplitude of denoised signal is less than that of the original signal amplitude. This gives clear information that noise levels are generally higher than PD signal amplitude. Wavelet de-noising using Daubechey s filter is the best method among different techniques -D source has been located in a steel tank with an error of 8% using iteration method (Newton-Raphson method). The accuracy of Dimensional source location has been improved to 96% when NewtonRaphson method is implemented on denoised signals 3
Conclusion UHF method has numerous applications and one of them is transformer. The sensor designed for inserting it in to the drain valve of transformer has very good sensitivity of 90pC under controlled conditions. So the UHF sensors can be installed at sites, factories without de-energizing the transformers and the approximate location of PD source can be traced such that the insulation strength of that particular area is increased and also the life of transformer is increased. -D PD source has been located in steel tank using iteration method (Newton-Raphson method) with a error of 8%, where as 3Dimensional source localization is done with non-iterative method like TDOA, PRPA. 3Dimensional PD source is also localized but with a certain error 4
FUTURE SCOPE This research is vast and can be extended to on-line monitoring of the transformer with a suitable hardware. The best way to implement this technique on-line is by using DAQ unit along with LABVIEW. This gives important information about condition monitoring and diagnostics of power transformers for a full day, week, month depending upon the size of data to be analyzed. So the best way to implement this technique on-line is to reduce the amount of data recorded in the hardware every second by erasing unnecessary data 5
6