1 Introduction On high-voltage transmission lines, the aging of insulators directly threatens the safe operation of the power system. Aging insulators, especially zero-value insulators, are prone to overheating and expansion, leading to cracking of the insulator head and causing line breakage accidents. Furthermore, their low flashover voltage reduces the insulation level of high-voltage lines. To achieve online detection of faulty insulators, many methods have been adopted, including the spark fork method, photoelectric method, electrostatic probe method, and self-propelled faulty insulator resistance detection method [1-3]. However, the devices developed based on these methods require operators to climb the towers and inspect all insulators on the tower one by one, which is very labor-intensive. In addition, the spark fork, at the moment of generating a spark, is equivalent to short-circuiting the insulator being tested, posing a risk of flashover of the insulator string. Therefore, there is an urgent need for a method that can be performed from the ground. In recent years, some new methods have emerged based on the partial discharge, infrared radiation, electromagnetic waves, ultrasonic waves, or abnormal mechanical vibrations generated by defective insulators, such as ultrasonic detection [2], infrared detection, Doppler effect method [4], and pulse current method [5]. These are all methods that can be used for detection on the ground, but they still have their limitations and are not currently suitable for field use. Since the detection method based on pulse current is low-cost and easy to implement, many people have been trying to use the pulse current method to achieve accurate detection of defective insulators in recent years [5-9]. From the current research status, there are two different detection schemes. The first scheme is to study the current waveform within one power frequency cycle and try to find the difference between the current waveform with defective insulators and the current waveform without defective insulators [8, 9], but this scheme is still only at the theoretical discussion level. The second scheme is to study the number of corona pulses and identify insulator strings containing defective insulators by measuring the number of pulses within multiple power frequency cycles [5, 6]. However, this scheme has serious problems in the detection method and its misjudgment rate is high. The use of corona pulse current to detect defective insulators was first proposed by the Japanese in the 1980s [5]. Over the years, many people have tried to study it, but the results have been unsatisfactory. Based on in-depth research on the corona discharge fingerprint (N-φ spectrum) of insulators and insulator strings, we propose a method to detect defective insulator strings using the corona discharge fingerprint of insulator strings, and apply pattern recognition technology to identify defective insulator strings. [b]2 Method for detecting defective insulator strings using corona discharge fingerprint 2.1 Principle of detecting defective insulator strings using pulse current[/b] Under normal working conditions, each insulator in the line bears a certain working voltage. When a defective insulator is present in a three-phase insulator string, the voltage distribution on that phase insulator string will change. The voltage shared by the defective insulator decreases, and the voltage shared by the good insulator increases. When the voltage shared by some good insulators at the conductor end of the line increases, the number and intensity of their corona discharge will also increase, making the number and distribution of corona pulses corresponding to the phase containing the defective insulator different from the other two phases. Therefore, by using sensors to extract current pulses from the tower grounding wire, the presence of defective insulator strings on the tower can be determined based on the different numbers and distributions of these three-phase corona pulses [5, 6]. In the laboratory, we applied a voltage of 66.5 kV (simulating a single-phase 110 kV line) to an insulator string composed of 7 good insulators and used a Rokowski high-frequency current sensor (1 kHz to 15 MHz) to take signals from the grounding wire. No corona pulse signal was measured. When the string contained one zero-value insulator, we detected a corona pulse signal; when it contained two zero-value insulators, more corona pulses could be detected. [b]2.2 Disadvantages of the Three-Phase Unbalance Method[/b] Under laboratory conditions, 110 kV good insulator strings and hardware conductors generally do not experience measurable corona discharge. However, in actual transmission lines, corona discharge can occur in three-phase insulator strings, hardware, and conductors. In this case, the presence or absence of corona pulse signals cannot be used to determine the existence of zero-value insulator strings. Furthermore, the corona discharge characteristics of insulators from different manufacturers, at different production times, of different types, on different lines, and in different geographical environments may vary. Therefore, the presence of defective insulator strings cannot be determined based on the absolute number of pulses detected over a period of time. In view of these two issues, previous corona current methods employed a method of comparing the three-phase corona pulses corresponding to the three-phase voltages [5, 6], i.e., using a three-phase imbalance method. This transforms the absolute judgment of signal parameters into a relative judgment, making the pulse current detection method more practical. The three-phase imbalance method assumes that the probability of all three phase insulator strings on a tower simultaneously containing defective insulators is very small. When a string contains a defective insulator, the number and intensity of corona discharges in that phase will increase. Therefore, the corona pulse signal corresponding to the phase containing the defective insulator will differ from the other two phases. Thus, by judging the balance of the number of three-phase pulses, it can be determined whether the tower contains a defective insulator string. Traditional methods for determining three-phase imbalance use a field-coupled signal as a reference signal to identify a zero point in the power frequency three-phase voltage. Using this zero point as a reference, the pulses on the 360° phase of the power frequency are divided into three equal segments, each representing a phase of the corona pulse (positive or negative) within one cycle. The number of pulses in each phase is then compared. However, there is a phase difference between the field-coupled reference signal and the zero point of the power frequency voltage. This phase difference is not fixed and varies depending on factors such as line structure and coupling method. Therefore, it is difficult to determine the zero point of the power frequency voltage in the field. This deviation in the zero point leads to inaccurate phase division, reducing the accuracy and reliability of the detection. Furthermore, even in towers without defective insulators, the three-phase corona pulses will differ due to variations in the corona discharge difficulty of the insulators on each string. Therefore, to accurately determine whether the three-phase insulator strings are balanced, it is not sufficient to simply compare the number of corona pulses corresponding to each string. Instead, it is necessary to utilize information such as the number, phase, and amplitude of pulses, as well as subtle differences, to identify defective insulator strings. [b]2.3 Pattern Recognition Method for Corona Discharge Fingerprint[/b] According to our research results [7], the negative corona pulses of single-phase insulators are concentrated in the phase range of 210° to 270° of the applied voltage, while the positive corona pulses are mainly distributed in the phase range of 30° to 90° of the applied voltage. Therefore, in one power frequency cycle, the corona pulse signals (positive or negative) of the three-phase insulator string will form three sets of distributions with a phase interval of 120°, which represent the corona discharge fingerprint of the three-phase insulator. This spectrum generally contains three clusters of positive pulses and three clusters of negative pulses, with each cluster of pulses concentrated in a phase width of about 60°, and there is a gap of about tens of degrees between two clusters of positive pulses or two clusters of negative pulses, as shown in Figure 1. Figure 2 shows the corona pulse signals in one cycle when each of the two-phase insulator strings contains a low-value insulator in our laboratory. The two-phase pulses in the figure can be clearly distinguished, where the sine waveform represents the phase of the applied A-phase voltage. Experimental results show that the corona discharge fingerprints of three-phase insulator strings exhibit different patterns. The corona discharge fingerprint pattern of a normal three-phase insulator string differs from that of an insulator string containing zero-value insulators. Based on these results, during detection, it is unnecessary to locate the zero point of the power frequency voltage, divide the 360° phase into three equal parts, or explicitly distinguish the three-phase corona pulse clusters. Instead, the statistical distribution (fingerprint) of corona pulses within one power frequency cycle can be considered as a pattern. The feature vector of this pattern can be extracted, and pattern recognition methods can be used to determine the presence of zero-value insulator strings. [img=250,131]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t56-1.gif[/img][align=center] Fig.1 Sketch of phase distribution of three phases of corona pulses[/align][img=250,159]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t56-2.gif[/img][align=center] Fig.2 The corona pulses in one periodic when each of the two insulator strings contains a low-value insulator[/align][align=center] In the laboratory, the author conducted extensive statistical analysis (500 cycles) on corona pulses of three-phase insulator strings exceeding a certain level, plotting the relationship between the number of pulses and phase (N-φ spectrum). This was then convolved with a window function to obtain a curve that clearly reflects the corona mode. When there are no defective insulators, no pulses were detected, and the curve is a straight line with zero amplitude. When a phase contains one zero-value insulator, the curve has two maxima, as shown in Figure 3. When two phases contain one zero-value insulator and one low-value insulator, respectively, the curve has four maxima, as shown in Figure 4. These results demonstrate that using pattern recognition to distinguish the fingerprints of corona discharge and thus determine the presence of zero-value insulator strings is feasible. [img=245,152]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t56-3.gif[/img][align=center] Fig.3 One kind of corona pattern[/align][img=250,151]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t56-4.gif[/img][align=left] Fig.4 Another kind of corona pattern [b]3 Study on the sensitivity of the pulse current method[/b] In the experiment, we found five factors affecting the detection sensitivity of the pulse current method: ① The resistance of the defective insulator. The lower the resistance of the defective insulator, the higher the detection sensitivity. ② The position of the defective insulator in the insulator string. The closer the defective insulator is to the conductor side, the higher the voltage applied to the normal insulator, the stronger the insulator corona discharge, and the higher the detection sensitivity. ③ The effect of the insulator string length on the detection sensitivity. The more insulator discs in the string, the greater the impedance of the discharge detection circuit, and the lower the detection sensitivity. ④ The effect of the corona characteristics of the normal insulator on the detection sensitivity. The lower the corona initiation voltage of the normal insulator, the higher the detection sensitivity. ⑤ The effect of air humidity. Our experimental research shows that air humidity has a certain impact on the corona characteristics of the normal insulator. When the humidity increases, the number of insulator corona pulses decreases, thus reducing the detection sensitivity. [b]4 Analysis of Field Interference 4.1 Sources and Analysis of Field Interference[/b] Generally speaking, the main sources of field interference are: ① power frequency leakage current and harmonic current flowing through the grounding wire of the tower; ② power system carrier communication current and radio wave induced current; ③ corona discharge from fittings and conductors; ④ pulse interference from nearby towers and substations. For power frequency leakage current and harmonic current, because their frequencies are very low, they can be filtered out using filters. For corona discharge from fittings and conductors, since fittings and conductors are designed to be corona-free in clear weather, corona pulses are few, and fittings and conductors on the same tower generally have the same environment. Even if corona discharge occurs, they will basically maintain a balanced state and will not affect the test results. For pulse interference from nearby towers and substations, due to the attenuation effect of the line, its amplitude is very small and will not have a significant impact on the test results. Only the carrier signal, with its large amplitude and dispersed frequency, is the biggest source of interference. [align=left][b]4.2 Selection of Signal Frequency Band 4.2.1 Spectrum of Corona Pulse[/b] The rising edge of the corona pulse waveform measured in the laboratory is about 50 ns, and the width is about 250-400 ns. Some pulse waveforms show non-periodic decay, with slow rising and decay rates and small amplitude; some waveforms show oscillating decay, with fast rising and decay rates and large amplitude. From the spectrum of the measured corona pulse signal, it can be seen that the main energy distribution of the corona pulse is within 20 MHz. [b]4.2.2 Spectrum of Line Carrier[/b] We used a clamp current sensor combined with a DL1540L digital oscilloscope (bandwidth 150 MHz, maximum sampling frequency 200 MHz) to measure multiple iron towers and cement towers of the Wangfeng, Qingji and Qingshaqi branches near the Qinghe substation in Haidian District, Beijing. The typical carrier signals and their spectra are shown in Figures 5 and 6. [img=260,155]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t57-1.gif[/img][align=center] Fig.5 The carrier signal on the ground line of a 110 kV cement tower[/align][img=260,149]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t57-2.gif[/img][align=left] Fig.6 The frequency spectrum of the carrier signal in Fig.5 As can be seen from the figure, the carrier signal is composed of multiple sine waves with different frequencies and amplitudes superimposed, which is a huge interference to the corona pulse current signal. Moreover, its frequency band is wide, and the center frequency of each band varies with time, line, and region. Some existing devices for detecting partial discharge often select a very narrow frequency band outside the interference band to detect the signal. Due to the uncertainty of the carrier frequency, this method is difficult to apply to the detection of faulty insulators on the line. In order to extract the pulse signal components as much as possible, we selected a frequency band above 2 MHz as the detection frequency band. [b]5 Application of Pulse Current Method in Actual Detection 5.1 Composition and Performance of Faulty Insulator Detector[/b] Based on the above analysis of the pulse current method, we developed a faulty insulator detector, the principle block diagram of which is shown in Figure 7. The corona pulse current signal of the insulator flowing through the grounding wire of the tower is acquired by a broadband sensor, filtered by a high-pass filter to remove the carrier interference signal, amplified by a broadband amplifier, and then converted into a digital quantity by an analog-to-digital converter. After statistical analysis, a discharge mode is formed, and after pattern recognition, a conclusion is drawn. Finally, the detection result is displayed. [/align][img=330,157]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t57-3.gif[/img][align=left] Fig.7 The sketch of the instrument This instrument is small in size and lightweight, powered by batteries, and is perfectly suitable for portable use when tracing lines in the field. It uses a clamp-on current sensor to obtain signals from the grounding wire of the tower, and can quickly complete the detection of all insulators on the tower without climbing the tower. It is easy to operate. [b]5.2 Field Detection Results[/b] On August 20, 1998, the author's instrument was used to detect the No. 65 cement tower of the Yuefan Line (110 kV) of Puyang Power Supply Bureau in Henan Province. No defective insulators were detected, and the curve of the number of corona pulses and the phase relationship was a straight line. No faulty insulators were found when the tower was inspected using a spark fork. Then, the first insulator on the line side was shorted using the spark fork, and then tested with a detector. The corona pattern was similar to that shown in Figure 3, indicating the presence of a faulty insulator. On May 7, 1999, the author's detector was used to inspect some lines of the Lanzhou Power Supply Bureau, and faulty insulators were found on some towers of the Kaihe No. 2 line. For example, the corona pattern of tower No. 9 was a corona pattern indicating a faulty insulator (see Figure 8). Subsequent inspection using a spark fork revealed a zero-value insulator on this tower, preliminarily proving the feasibility of the fingerprint detection method. Currently, the author's research results are only applicable to 110 kV lines; future research needs to target lines with higher voltage levels. [/align][img=240,148]http://zszl.cepee.com/cepee_kjlw_pic/files/wx/zgdjgcxb/zgdj2000/0004/image4/t58-1.gif[/img][align=left] Fig.8 The corona pattern of tower No.9 [b]6 Conclusion[/b] (1) By studying the characteristics of corona discharge of insulator strings, a method for detecting defective insulator strings using the discharge fingerprint (N-φ spectrum) of corona pulses was proposed. (2) The pattern recognition method can eliminate the error caused by inaccurate phase separation and reduce the false judgment rate. (3) The detection sensitivity of the pulse current method is related to five factors. (4) Through the analysis of on-site interference such as carrier waves and anti-interference measures, an intelligent portable testing prototype was developed and achieved good results in the field trial on a 110 kV line, proving the feasibility of the fingerprint detection method and promoting the development of research on pulse current method for detecting defective insulator strings. References: [1] Bai Kehan. Live-line work [M]. Beijing: China Water Resources and Electric Power Press, 1988. [2] Wada, Yuji. Electrical equipment diagnostic technology and its automation [M]. Beijing: Machinery Industry Press, 1990. [3] Mao Fenglin. GYF series fiber optic voice reporting distributed voltage tester [J]. High Voltage Engineering, 1995, 21(3): 75. [4] Takehiko. Development of a remote sensing system for faulty suspension insulators [J]. Proceedings of 1994 International Joint Conference: 26th Symposium on Electrical Insulating Material, JC2-8: 467. [5] Takeda Masatoshi. Pulse discharge current type transmission line faulty insulator detector and its application method [J]. Electrical Computing, 1980: 64. [6] Jin Hongzhang. Research on pulse method for detecting faulty insulators in ultra-high voltage lines [J]. High Voltage Engineering, 1984, 10(4): 42. [7] Shi Qian. Experimental study on detecting zero-value insulators using pulse current method [D]. Beijing: North China Electric Power University, 1998. [8] Li Tianyun. A new method for diagnosing insulators using modern time series analysis theory [J]. High Voltage Engineering, 1992, 18(2): 33-35. [9] Li Weidong. A new method for diagnosing faulty insulators based on artificial neural networks [J]. High Voltage Engineering, 1994, 20(3): 24-27.