Rogowski coil sensor in the digitization process to detect partial discharge

This paper presents the construction of a Rogowski Coil sensor with an air core to detect partial discharge using a digital oscilloscope. Two types of sensors are used. The first is the primary sensor winding with back wire, and the second is without back wire, labeled BW and WBW, respectively. The numbers of primary-turn in the sensors are 5, 10, 20, and 40 turns. The performance of the sensors is tested using two types of tests. First, the wave response test with a fixed imitation partial discharge magnitude input is used to select the optimum sensitivity with the lower sampling rate, aims to select the peak or valley value as a magnitude partial discharge value. The second test is using an imitation partial discharge ramp to check the linearity of the sensors. The imitation of the partial discharge inputs is generated by a commercial charge calibrator. The wave response test results show an increase in the number of turns that corresponds to an increase of the sensor output for both sensors in a non-linear trend. In determining the sampling rate, the detection of magnitude in the valley is better than the peak. All sensors act linear toward the imitation partial discharge ramp either in BW or WBW conditions.


INTRODUCTION
Partial discharge is one of the main mechanisms of aging, which leads to failure of catastrophic insulation in continuous operation. The asymmetrical configuration of the electrodes or the different types of insulating material creates an inhomogeneous electric field. An inhomogeneous electric field causes partial discharge. One example of the cause of the unevenness of the insulating material is the occurrence of air bubbles in solid insulation. The partial discharges occur in air bubbles, which results in continuous destruction of the surface leading to one side of the conductor, which means the distance of the conductor separator becomes smaller than before. Even with an increase in the frequency of PD can cause excessive heat, which results in increased destruction activity that occurs in propagation that forms a tree (treeing) like a conductor containing an electric charge due to PD to one of the more intense conductors. Treeing increases the strength of the electric field significantly, failing insulation [1][2][3]. Catastrophic insulation must be overcome if the detection of PD periodically to be detected early symptoms of aging insulation. A PD TELKOMNIKA Telecommun Comput El Control  Rogowski coil sensor in the digitization process to detect partial discharge (Eka Putra Waldi) 1063 produces a signal that has a short duration and small magnitude compared to the actual source voltage, so it needs a hardware (sensor) to detect it. There are two conventional ways of measuring PD. The first way is to connect the sensor in series with the sample. The disadvantage of this circuit is not safe against overvoltages when sample failure occurs. The second way is to connect the sensor in series with the capacitor, then connect it in parallel with the sample. The advantage circuit is safe against overvoltages when sample failure occurs [4][5][6][7][8][9][10]. The capacitor functions as a voltage divider and as a High Pass filter so that the sensor is safe against high voltage and only accepts high-frequency signals (PD signals). To be read by the measuring instrument, an amplifier is used so that the small-signal can be read by the acquisition device, and the digitization process requires adjusting the pulse width using an integrator method.
Today's technological sophistication, the digitization process has reached the level of Mega Sampling per second (MS/s) even to Giga MS/s and the smallest scale of millivolts. It allows reading partial discharge signals that have high frequency and low magnitude without using an amplifier. Induction Sensor is one of the high-frequency signal detection [11][12][13][14]. In addition, manufacturing sensors are economical compared to commercial sensors. One induction sensor that is widely used is the Rogowski coil sensor. Rogowski coil sensors have Primary and Secondary parts. The Primary section has one turn and functions as a signal receiver in the form of instantaneous current. The Secondary Section has several turns and serves to change the instantaneous current signal in the form Primary magnetic flux section to the Secondary section into a transient Voltage. The ends of the cable are returned to their original positions so that both ends of the cable are at the same point. The wire that returns to its initial position is named by Return Wire, whereas in this paper, it is called Back Wire [15][16][17][18][19][20][21][22]. In a single-phase measurement, there is usually no disturbing external flux, as well as a three-phase circuit that is symmetrically mounted. Then there is no disturbing outside flux. Furthermore, this study does not use Back Wire.
Many kinds of research discuss the response to sensors that are made even to the point of researching the matching impedance problems needed when an acquisition device is used. Until now, the discussion about the effect of the number of primary turns of the sensor has not been too detailed [22][23][24]. Furthermore, data acquisition also requires a characteristic of the wave to be read so that the use of a sampling rate can be determined. This research makes an economical sensor by paying attention to the accuracy of the acquisition by varying the number of secondary winding that has back wire (BW) and which does not have back wire (NBW). It is essential to test the sensor's capabilities compared to commercial sensors.

RESEARCH METHOD
In general, Rogowski's coil (RC) sensors work according to the laws of electromagnetic induction, where the instantaneous current through the primary winding conductor produces an instantaneous magnetic field in the form of magnetic flux that is directly proportional to the surface area that covers it. This flux induces a secondary winding conductor. Following Faraday's law related to induction by flux changes, the secondary winding produces an opposing flux, which is characterized by a potential difference between the ends of the secondary winding conductors. The voltage on the secondary winding is the result of the instantaneous current response detected. Where the voltage is proportional to the function of current changes with time. Mathematically, the resulting induced voltage can be expressed by: where io is the instantaneous current of a PD pulse passing through the Primary part, which changes with time function, and M is the mutual inductance between the secondary part of the sensor loop and the Primary part.
N is the turn number of the Secondary part, and the picture shows in Figure 1 (a). The following parameters are the geometrical construction (core) of the Rogowski sensor, W is the height of the sensor, a is the inside diameter of the sensor, and d is the outside diameter of the sensor [9], the picture can show in Figure 1 (b). The nonmagnetic material core in the form of polyethylene. The values of the parameters display Table 1. Variation of the number of turns and using a back wire (BW) for RC variations as Sample conditions expression in Table 2. All researchers agree that the equivalent circuit is the same shape [23],  Figure 2. Sensitive sensors are carried out by linearity testing of input increases as shown in Figure 3. In this series, the function generator functions as a source and regulator of the amount of charge that is the input for the charge calibrator. Then the total charge is calculated by the formula: Q is the magnitude of the charge simulation with units of pico coulombs (pC), C is a capacitor with a unit of farads (F), and V is the magnitude of the voltage application. A wave magnitude that produces PD represents a magnitude PD that calibrates with a charge calibrator.  This study uses two types of samples with variations of the same number of turns. The first type of coil uses back wire and the second does not use back wire. The variation of the number of turns in each type of sensor is 5, 10, 20 and 40, as illustrated in Figure 4. The working principle of measuring the charge variation for the linearity test in Figure 3 is as follows. First, according to (4), the function generator (FG) can adjust the charge by determining the pulse voltage (V). The Output FG arrange for input to the charge calibrator (Haefely Type 9218 signal calibrator), the capacitor used is 141 pF.
Furthermore, determining the charge value is the multiplication voltage of the function generator with the capacitor used by the charge calibrator (CC) [25]. After that, the CC output gives the signal to the coupling capacitor (Haefely couling capacitor Type 9230). After that, the coupling capacitor in series with a commercial sensor (Haefly PD detector Type 9231) and RC sensor aims to compare the results of measurements on the same input. Response sensor measurements connect sequentially with Commercial and RC sensor output to the oscilloscope on channel-1 and channel-2. Labview program controls all measurement activity through a local area network (LAN) and storage of all digital data. There is no effect on the number of secondary turns of the RC sensor on the PD significance read by the commercial sensor, nor is there any influence on the PD magnitude reading on sensors that have Back wire or do not have back wire. These results indicate that the measurement using the RC sensor does not interfere with the electrical activity that is taking place. Figure 6 (a) is a reading of commercial sensors and RC sensors that are given a back wire and are not given a back wire with a total of 40 turns. The result of the RC sensor response can be read directly by the oscilloscope well without using an amplifier, even though the magnitude that is read is much smaller than the commercial sensor response, it is shown in the picture in a dotted rectangular box. The waveforms produced by RC sensors are very different from commercial sensors, where the waveforms oscillate (underdamped). The measurement results of commercial sensors have larger than RC sensors. Thus, it is difficult to compare. Consequently, comparisons require a specific method.

Figure 5. Effect of Commercial sensor output on RC sensors installed in series
Comparison of the characteristics of two waves using wave normalization by dividing the wave with its maximum value. Normalization shows that two waves have the same magnitude. The results show in Figure 6 (b). RC sensors respond to input signals earlier than commercial sensors, and RC wavelengths are shorter than commercial sensor waves. The output wave characteristics of the RC sensor are different from those of Commercial sensors. The results state that the RC sensor output oscillates concerning the time function while the commercial sensor does not oscillate, even though the oscilloscope uses an input impedance of 50 ohms (pure resistive) for which there is no capacitance value. The oscillation of the data informs that the RC sensor not only contains inductor and resistor but also contains capacitance [6], and there is no significant difference between RC sensors that are given BW or without BW, where the wave normalization in the two waves is the same shape.  Figure 7 is the result of measurements of each RC sensor with a different number of secondary turns. In Figure 7 (a) is the result of measuring RC sensors without using BW, and Figure 7 (b) is the result of measuring RC sensors using BW. The determination of the PD magnitude represented by the oscillating maximum wave amplitude has been defined at the dominant amplitude [15]. The results of the RC sensor measurements in Figure 7 illustrate that the positive amplitude is slightly different from the absolute value of the negative amplitude, as consideration for selecting that value as the peak value. Furthermore, in each RC sensor response, the amplitude of the positive and negative wave in Figure 7 (a) and Figure 7 (b), an absolute magnitude function of the number of turns can be developed into Figure 8. The information in the figure, where BWN (max) and WBWN (max) is the positive amplitude of each wave type, while BWN (min) and WBWN (min) is the absolute value of the negative amplitude of each wave type. The results show that generally, the amplitude increases when the number of turns is increased from 5 to 20 turns and remains even though the number of turns is increased to 40 turns. The BWN sensor has a greater magnitude compared to the WBWN sensor.
The absolute magnitude of the negative wave uses it as an alternative amplitude of PD magnitude. Thus it is necessary to calculate how much the percentage of the value decreases to the value of the positive magnitude wave. The calculation uses the data in Figure 8, which is a function of the number of secondary turns of the RC sensor. Comparisons made are WBWN (absMin) against WBWN (Max) and BWN (absMin) with BWN (Max). Sine or cosine waves are symmetrical waves, sampling the waves for the uncomplicated digitization process. The sampling rate of this wave is lower than the impulse wave. Impulse waves need little attention because the waves are of short duration with a duration above 90% peak located at no more than 10% of their period. That results in a sampling rate of peak detection of at least half the time of the wave peak period (Nyquist theorem). The magnitude of the sampling rate determines the accuracy of the peak value reading. The higher the sampling rate, the more accurate the peak value reading with the consequence requires a large amount of data [26]. Expectations require a technique for adjusting or changing the characteristics of the RC sensor in this study by varying the number of primary turns to change the sensor characteristics.
The input calculation from Figure 7 (a) and 7 (b) produces the output in Figure 9, and the calculation is a function of the number of turns, where the calculation use reference 26. The main vertical axis places the output on the left side with a left direction sign. BWN (Max) and WBWN (Max), in the number of turns 5 and 10, both have the same sampling rate with the value of 65 MS/s and a decrease to a value of 50 MS/s on the number of turns 20 and 40. BWN (absMin) sampling rate decreased from 40 to 33 MS/s when the number of turns increased from 5 to 10 turns. After that, it remained at 33 MS/s even though the number of turns increased to 40 turns. The WBWN (absMin) sampling rate decreases from 40 to 33 when the number of turns increases from 5 to 10 turns and continues to decrease up to 29 MS/s when the turns become 20 turns, after that, it remains even though the number of turns is increased to 40 turns. WBWN (DecMin) is the percentage of reduction in WBWN (absMin) to WBWN (Max), and BWN (DecMin) is the percentage of reduction in BWN (absMin) to BWN (Max) which is the output function of the number of turns. The graph puts the output on the secondary vertical axis and gives directions to the right in Figure 9. In general, the two data states an increase occurs if the number of turns increases. Reviewing the overall data are two essential things. There are two oval shapes added to graph 9 which are named Oval-1 and Oval-2. Oval-1 demonstrates the lowest sampling rate at WBWN (absMin) and BWN (absMin) although there is a slight difference in the value of the lowest sampling rate between WBWN (absMin) and BWN (absMin). Whereas the highest sampling rate is at WBWN (Max) and BWN (Max) with 40 turns. Oval-2 demonstrates the same sampling rate as Oval-1 but Oval-2 has a smaller number of turns than Oval-1, which is 20 turns and WBWN (DecMin) has a decrement of magnitude smaller than BWN (DecMin). Figure 10 is the result of the sensor linearity test for the partial discharge imitation input. Figure 10 (a) is a sensor that uses a back wire while Figure 10(b) is a sensor that does not use a back wire. Both images illustrate that the sensor is linear with respect to the input of either positive or negative waves. These results state that the two types of sensors made qualify as a partial discharge measurement tool. Furthermore, the selection of the method to be used is in accordance with reference 26 where the acquisition data is able to read a peak value of 90% and is followed by the sensitivity of the equipment. When linked to Figure 9 the recommended reading is the smallest sampling value with a reading error of positive value of no more than 30%.

CONCLUSION
The creation of the Rogowski sensor for detecting PD has been well-reviewed. Under the specific conditions described in this article with the following conclusions: (a) the wave response test results show an increase in the number of turns that corresponds to an increase in the sensor output for both sensors in a non-linear trend, but sensors act linear toward the imitation PD ramp; (b) PD reading technique with a sampling rate method is very effective reading on negative waves, where the smallest sampling rate is found on these waves.