Partial discharge diagnostic device, partial discharge diagnostic method, and partial discharge diagnostic system

Abstract

To provide a partial discharge diagnostic system capable of suppressing deterioration in determination accuracy of partial discharge even when pseudo partial discharge occurs, a partial discharge diagnostic device, and a partial discharge diagnostic method.SOLUTION: A partial discharge diagnostic device for an apparatus executes factor determination of a partial discharge signal of an insulator, and includes a processing unit and a learning model generation unit. The processing unit generates and processes data of a predetermined format in which a pseudo partial discharge signal in an electric signal varying according to a phase is suppressed. The learning model generation unit generates a learning model for determining at least any of a factor of the partial discharge and presence or absence of the partial discharge based on the data.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] Embodiments of the present invention relate to a partial discharge diagnosis device, a partial discharge diagnosis method, and a partial discharge diagnosis system.

Background Art

[0002] Power equipment such as generators, motors, inverter devices, switchgears, and cables is provided as plant equipment. The power equipment has an insulator on the surface of the conductor, and the insulation performance of the insulator deteriorates over time. When repeated thermal expansion and contraction due to temperature changes occur, as in the case of generator coils of a generator, the insulator deteriorates. Such deterioration of the insulator becomes a cause of failure of power equipment due to insulation breakdown.

[0003] It is known that partial discharge occurs from power equipment when the insulator deteriorates. Therefore, based on the occurrence phenomenon of partial discharge, the insulation state is determined.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, pseudo partial discharge similar to partial discharge may occur. In such a case, there is a risk that the determination accuracy of the insulation state will decrease.

[0006] The problem to be solved by the present invention is to provide a partial discharge diagnosis system, a partial discharge diagnosis device, and a partial discharge diagnosis method capable of suppressing a decrease in the determination accuracy of partial discharge even when pseudo partial discharge occurs.

Means for Solving the Problems

[0007] A partial discharge diagnostic device according to an embodiment of the present invention is a partial discharge diagnostic device that performs factor determination of a partial discharge signal of an insulator, comprising a processing unit and a learning model generation unit, The processing unit generates data in a predetermined format that suppresses pseudo-partial discharge signals in electrical signals that fluctuate according to the phase. The learning model generation unit generates a learning model based on the data that determines at least one of the factors causing the partial discharge and whether or not the partial discharge is present. [Effects of the Invention]

[0008] According to the present invention, the accuracy of determining the cause of partial discharge in power equipment can be improved. [Brief explanation of the drawing]

[0009] [Figure 1] This diagram illustrates the differences in the appearance of the φ-q-n characteristic curves for each partial discharge factor. [Figure 2] A diagram showing examples of partial discharge signals and pseudo-discharge signals. [Figure 3] This figure shows an example where partial discharge signals generated by the U and W phases are superimposed on the V phase. [Figure 4] A block diagram showing an example configuration of a partial discharge diagnostic system. [Figure 5] A block diagram showing an example configuration of the data processing unit. [Figure 6] This figure shows an example of how the calculation unit generates a φ-q characteristic curve. [Figure 7] This figure shows an example of how the calculation unit generates a φ-q-n characteristic curve. [Figure 8] This figure illustrates an example of performing a Fast Fourier Transform on a time interval of time-series data of electrical signals. [Figure 9] A diagram showing an example of spectrogram generation by the calculation unit. [Figure 10] A diagram illustrating an example of normalization for time-series data of electrical signals. [Figure 11] This figure shows an example of applying a logarithmic transformation to a two-dimensional matrix of the φ-q-n characteristic diagram. [Figure 12] A diagram for explaining an increase processing method using averaging. [Figure 13] A block diagram showing a configuration example of a feature extraction unit. [Figure 14] A diagram showing an example of a frequency spectrum generated by performing a fast Fourier transform on a simulated electrical signal. [Figure 15] A diagram showing an example of setting a frequency range from the spectrum values of a simulated electrical signal. [Figure 16] A diagram showing an example of selecting a frequency range from the magnitude and occurrence frequency of a spectrum. [Figure 17] A diagram showing an example of selecting a frequency range using a spectrogram. [Figure 18] A diagram illustrating a method of selecting a phase range from a φ-q-n characteristic diagram. [Figure 19] A diagram showing an example of generating a charge amount histogram for the charge amount. [Figure 20] A diagram illustrating a method of selecting a phase range from the peak position of data points in a φ-q characteristic diagram. [Figure 21] A φ-q characteristic diagram showing pseudo partial discharges with arrows. [Figure 22] A diagram showing an example of performing an inclusion process on data points on a φ-q characteristic diagram. [Figure 23] A diagram illustrating a method of selecting a phase range from a spectrogram. [Figure 24] A flowchart showing a processing example of a partial discharge diagnosis system. [Figure 25] A diagram showing an example of an imaging process by an image generation unit. [Figure 26] A diagram showing an example of a masking process by a filtering unit. [Figure 27] A diagram schematically showing an arrangement example of a water turbine generator coil and a sensor. [Figure 28] A diagram showing an example of a reference point adjustment process of an applied voltage by a filtering unit. [Figure 29] A diagram showing an example of a masking process by a filtering unit.

Embodiments for Carrying Out the Invention

[0010] Hereinafter, a partial discharge diagnostic device, a partial discharge diagnostic method, and a partial discharge diagnostic system according to embodiments of the present invention will be described in detail with reference to the drawings. Note that the embodiments shown below are examples of embodiments of the present invention, and the present invention is not limited to these embodiments. Furthermore, in the drawings referenced in these embodiments, the same or similar reference numerals are used for identical parts or parts having similar functions, and repeated descriptions may be omitted. Also, the dimensional ratios in the drawings may differ from the actual ratios for explanatory purposes, and some components may be omitted from the drawings. Hereinafter, partial discharge and pseudo-partial discharge will be described using Figures 1 to 3.

[0011] [Aspects of partial discharge in insulators] The characteristics of partial discharge in an insulator will be explained using Figure 1. The φ-q characteristic diagram shows the relationship between the applied voltage phase and the value of the electrical signal. The electrical signal is a signal corresponding to the phase of the voltage applied to power equipment, etc. For example, an electrical signal is at least one of a charge quantity, current, and voltage that fluctuates according to the phase. Therefore, in this embodiment, when we refer to an electrical signal, it is assumed that it includes at least one of a charge quantity signal, current signal, and voltage signal that fluctuates according to the phase.

[0012] Furthermore, although this embodiment uses a charge quantity signal as the electrical signal for explanation, it is not limited to this, and it is also possible to use a current signal, a voltage signal, or the like. The φ-q-n characteristic diagram is a characteristic diagram that shows the correlation between the partial discharge charge amount, the number of occurrences, and the applied voltage phase, obtained by accumulating the φ-q characteristics over a certain period of time.

[0013] Figure 1 shows the φ-q characteristic diagram for each partial discharge factor. The φ-q characteristic diagram in Figure 1 shows the relationship between the phase φ of the applied voltage L10 and the charge amount q. The vertical axis represents the charge amount, and the horizontal axis represents the phase (time). There are multiple insulation degradation factors that cause a partial discharge signal to occur during one cycle of the applied voltage L10. Therefore, the appearance of the φ-q characteristic diagram differs for each insulation degradation factor. For example, Figure 1(a) shows that a partial discharge signal occurs between 180° and 270°, and Figure 1(b) shows that partial discharge signals occur between 0° and 90°, and between 90° and 270°. As will be described later, the partial discharge signal is measured as a charge amount larger than, for example, the average charge amount.

[0014] Figure 1(c) shows partial discharge signals occurring on the negative side from 0° to 90° and on the positive side from 90° to 270°. It is possible to associate the cause of discharge with each of these patterns. In this embodiment, it is possible to use training data associated with the cause of discharge when performing machine learning to determine the insulation state of power equipment. Note that Figure 1 is an example of the φ-q characteristic diagram for each partial discharge cause, and the patterns of partial discharge causes are not limited to these.

[0015] [Pseudo partial discharge signal] In this implementation, electrical signals that occur independently of the insulating material being identified are sometimes referred to as pseudo-partial discharge signals. It is believed that factors contributing to pseudo-partial discharge include, for example, the propagation of discharge signals generated outside the measurement target to the sensor, and their subsequent measurement by the sensor. Figure 2 is a φ-q characteristic diagram showing an example of a pseudo-partial discharge signal. The vertical axis represents the amount of charge, and the horizontal axis represents the phase (time). For example, Figure 2(a) shows an example where a partial discharge signal occurs in the ranges from 0° to 90° and from 90° to 270°. In contrast, Figure 2(b) shows an example where a pseudo-partial discharge signal occurs near 90° and near 270°.

[0016] For example, in the φ-q characteristic diagram, partial discharge signals occur frequently around at least one range of the applied voltage phase, from 0° to 90° and from 180° to 270°, as shown in Figures 1 and 2(a). In contrast, as shown in Figure 2(b), the pseudo-partial discharge signals tend to statistically deviate from the generation phase of the actual partial discharge signals, as has been observed from the applicant's experimental results.

[0017] [Pseudo-partial discharge signal in 3-phase AC] Furthermore, experimental results from the applicant have shown that signal superposition can occur between the U, V, and W phases in generators and motors. Figure 3 shows an example where partial discharge signals generated by the U and W phases are superimposed on the V phase. The vertical axis represents the amount of charge, and the horizontal axis represents the phase (time). Figure 3(a) is the φ-q characteristic diagram of the U phase, Figure 3(b) is the φ-q characteristic diagram of the V phase, and Figure 3(c) is the φ-q characteristic diagram of the W phase. In this example, partial discharge signals of the U and W phases are superimposed on the V phase partial discharge signal S10 as pseudo-partial discharge signals PU and PW.

[0018] [Configuration of the partial discharge diagnostic system] Next, the system configuration of the partial discharge diagnostic system 1 will be described. Figure 4 is a block diagram showing an example configuration of the partial discharge diagnostic system 1. As shown in Figure 4, the partial discharge diagnostic system 1 is a system that can reduce the influence of pseudo-partial discharge signals and can determine at least one of the following: the presence or absence of partial discharge and the cause of partial discharge. The partial discharge diagnostic system 1 comprises a measuring instrument 10, a display device 20, an operating device 30, and a partial discharge diagnostic processing device 40.

[0019] The measuring instrument 10 supplies time-series data of electrical signals measured from sensors attached to power equipment to the electrical diagnostic processing unit 40. For example, the sensors may be high-frequency current sensors (current-based) or electromagnetic antennas (electromagnetic wave-based). As described above, electrical signals include at least one of charge signals, current signals, and voltage signals.

[0020] The display device 20 is, for example, a monitor. This display device 20 displays image data supplied from the partial discharge diagnostic processing device 40.

[0021] The operating device 30 is comprised of input devices such as a keyboard and a mouse. This operating device 30 inputs signals corresponding to the operator's actions to the partial discharge diagnostic processing device 40.

[0022] As shown in Figure 4, the partial discharge diagnostic processing device 40 is a device capable of partial discharge diagnosis while suppressing the effects of pseudo-partial discharge. This partial discharge diagnostic processing device 40 comprises a data acquisition unit 100, an electrical signal generation unit 102, a feature extraction unit 104, a processing unit 105, a learning model generation unit 110, a partial discharge determination unit 112, and a storage unit 116. The processing unit 105 has a data processing unit 106 and an image generation unit 108.

[0023] The partial discharge diagnostic processing device 40 has a CPU (Central Processing Unit), which is, for example, a computer. By executing a program stored in the memory unit 116, the partial discharge diagnostic processing device 40 can configure a data acquisition unit 100, an electrical signal generation unit 102, a feature extraction unit 104, a processing unit 105, a learning model generation unit 110, and a partial discharge determination unit 112. It is also possible to configure the data acquisition unit 100, the electrical signal generation unit 102, the feature extraction unit 104, the data processing unit 106, the image generation unit 108, the learning model generation unit 110, and the partial discharge determination unit 112 using electronic circuits.

[0024] The data acquisition unit 100 acquires time-series data of electrical signals measured from sensors attached to power equipment from the measuring instrument 10. The measuring instrument 10 has multiple sensors. The data acquisition unit 100 can acquire time-series data of multiple different electrical signals from the measuring instrument 10. Note that the measuring instrument 10 is not limited to one, and it is possible to use multiple measuring instruments 10.

[0025] The electrical signal generation unit 102 acquires or generates training electrical signals (simulated electrical signals) to be used in machine learning. The electrical signal generation unit 102 generates, for example, a simulated electrical signal having partial discharge signal data. The electrical signal generation unit 102 generates a simulated electrical signal having a partial discharge signal, for example, as shown in the φ-q characteristic diagram in Figure 1. The electrical signal generation unit 102 can also generate a simulated electrical signal that does not have a partial discharge signal.

[0026] The electrical signal generation unit 102 acquires simulated electrical signals via the data acquisition unit 100, for example, in a test device that reproduces the entire power equipment or a part thereof, simulating the factors that cause partial discharge. The electrical signal generation unit 102 can also generate signals by adding pseudo-partial discharge signal data to data that already contains partial discharge signal data. These data are stored in the storage unit 116 in association with the cause of discharge.

[0027] Furthermore, the electrical signal generation unit 102 may use simulation as a method for generating simulated electrical signals. Alternatively, the electrical signal generation unit 102 may combine simulated electrical signals acquired by the test device with data generated by simulation. In this way, the amount of data regarding partial discharge signals for each factor can be increased, and the accuracy of judgment by machine learning can be further improved. The simulated electrical signals generated by the electrical signal generation unit 102 are signals that indicate charge amount, voltage, current, etc., with respect to phase (time), similar to the electrical signals acquired by the data acquisition unit 100.

[0028] The feature extraction unit 104 extracts features indicating the occurrence of a partial discharge signal from the simulated electrical signal generated by the electrical signal generation unit 102, for each cause of partial discharge signal generation. For example, the feature extraction unit 104 extracts features indicating the occurrence of a partial discharge signal from the magnitude and frequency of the electrical signal, and generates feature quantities containing range information where the partial discharge signal occurs, such as the phase range, frequency range, and magnitude range of the partial discharge signal. In other words, these feature quantities represent range information where the partial discharge signal occurs, such as the phase range, frequency range, and magnitude range of the partial discharge signal. By using a simulated electrical signal, the influence of noise such as pseudo-partial discharge signals is reduced, and feature quantities related to the partial discharge signal can be effectively extracted for each cause of generation. Further details of the feature extraction unit 104 will be described later.

[0029] The data processing unit 106 performs preprocessing on the electrical signals acquired by the data acquisition unit 100 and the simulated electrical signals generated by the electrical signal generation unit 102. Furthermore, the data processing unit 106 performs suppression processing of pseudo-partial discharge signals based on information about the partial discharge signals generated by the feature extraction unit 104. For example, the data processing unit 106 can suppress electrical signals in ranges where there is a high probability that no partial discharge signals exist. Suppression in this embodiment also includes deletion. Details of the data processing unit 106 will be described later.

[0030] The image generation unit 108 can generate data in a predetermined format using data after processing by the data processing unit 106, or data before processing by the data processing unit 106. For example, the image generation unit 108 generates data in a predetermined format as image data having numerical values ​​in a two-dimensional array. In this way, the processing unit 105, which includes the data processing unit 106 and the image generation unit 108, processes data in a predetermined format that suppresses pseudo-partial discharge signals in electrical signals that fluctuate according to phase.

[0031] More specifically, the image generation unit 108 can generate images in grayscale or color. In this embodiment, the data used for machine learning or judgment processing may be arranged in a two-dimensional matrix, but is not limited to this. Such two-dimensional matrix-arranged training data and judgment data may be referred to as image data. In other words, image-capable training data and judgment data arranged in such a two-dimensional matrix may be referred to as image data. By displaying such image data on the display device 20, the characteristics of the training data and judgment data can be visually grasped.

[0032] The learning model generation unit 110 generates a learning model (discriminator) using the data processed by the data processing unit 106 as training data. The learning model generation unit 110 can also generate a learning model using image-capable two-dimensional matrix-like image data generated by the image generation unit 108 as training data. This learning model generation unit 110 associates the data processed by the data processing unit 106 with training signals such as the cause of partial discharge and the presence or absence of partial discharge, using these as training signals. The learning model generation unit 110 can use neural networks and other discriminative learning algorithms for machine learning, and the learning algorithm is not limited. In this way, the learning model generation unit 110 generates a learning model using training data that associates the data generated by the processing unit 105 with at least one of the causes of partial discharge and the presence or absence of partial discharge.

[0033] The partial discharge determination unit 112 uses, for example, the learning model generated by the learning model generation unit 110 to determine the presence or absence of a partial discharge signal in the electrical signal to be diagnosed (determination data) and its cause, based on the processing results of the data processing unit 106. The display control unit 114 can display the two-dimensional data generated by the image generation unit 108 on the display device 20.

[0034] The storage unit 116 is composed of, for example, an HDD (hard disk drive) or an SSD (solid state drive). This storage unit 116 stores various data and programs used in the partial discharge diagnostic processing device 40. For example, the storage unit 116 stores data acquired by the data acquisition unit 100, training data, judgment data, and data related to trained learning models.

[0035] Here, the details of the data processing unit 106 will be explained. Figure 5 is a block diagram showing an example configuration of the data processing unit 106. As shown in Figure 5, the data processing unit 106 includes a calculation unit 200, a scale adjustment unit 202, a data expansion unit 204, a threshold processing unit 206, and a filtering unit 208. The calculation unit 200 performs calculations on the electrical signals acquired by the data acquisition unit 100 and the simulated electrical signals generated by the electrical signal generation unit 102, such as statistical processing, nonlinear transformations such as fast Fourier transform, wavelet transform, linear transform, logarithmic transform, normalization, canonicalization, and averaging. Alternatively, the calculation unit 200 can perform calculations on a single one of these processes or a combination of multiple processes.

[0036] Furthermore, the calculation unit 200 can output, for example, a φ-q characteristic curve, a φ-q-n characteristic curve, a spectrogram, and a scalogram as calculation results. Here, an example of generating a φ-q-n characteristic curve by the calculation unit 200 will be explained using Figures 6 and 7.

[0037] Figure 6 shows an example of how the calculation unit 200 generates a φ-q characteristic diagram. Figure 6(a) shows the time-series electrical signal acquired by the data acquisition unit 100. The vertical axis represents the amount of charge, and the horizontal axis represents the phase (time) from 0 degrees to 360 degrees of phase in one period interval of the applied voltage. Figure 6(b) shows the data from Figure 6(a) divided into multiple intervals, with discrete values ​​for each interval. The time-series data may be for one period or multiple periods. In other words, the number of times the time-series data is superimposed can be set to any number. The calculation unit 200 is capable of generating a φ-q characteristic diagram in which the amount of charge is plotted against the phase of the applied voltage in this manner.

[0038] Figure 7 shows an example of how the calculation unit 200 generates a φ-q-n characteristic diagram. The vertical axis represents the amount of charge, and the horizontal axis represents the phase (time) from 0 degrees to 360 degrees in one period interval of the applied voltage. The φ-q-n characteristic diagram is obtained by dividing the φ-q characteristic diagram in Figure 6 into arbitrary intervals N10 with respect to the amount of charge on the vertical axis and the phase on the horizontal axis, and accumulating and summing the number of times the charge occurs in each interval N10 to represent the occurrence frequency as A to F, etc., in a two-dimensional histogram. The accumulation of the number of times the charge occurs may be performed for one period of electrical signal or for multiple periods of electrical signal.

[0039] Here, we will explain an example of spectrogram generation by the arithmetic unit 200 using Figures 8 and 9. Figure 8 shows an example of performing a Fast Fourier Transform (FFT) on time-series data of an electrical signal over an arbitrary time interval. Figure 8(a) shows an example of converting time-series data of an electrical signal into a spectrum with respect to frequency. The vertical axis represents the amount of charge, and the horizontal axis represents the phase (time) from 0 to 360 degrees in one period interval of the applied voltage. Figure 8(b) shows the spectrum over one interval TSn. The vertical axis represents the spectrum, and the horizontal axis represents the frequency.

[0040] As shown in Figure 8(a), the calculation unit 200 divides the phase range into n time segments TS1 to TSn, where n can be set to any number. The calculation unit 200 performs a Fast Fourier Transform on the time-series data of the electrical signal for each time segment TS1 to TSn corresponding to the phase. In other words, the spectrum corresponding to the charge amount in Figure 8(b) shows the frequency-specific component values ​​obtained by the Fast Fourier Transform on the time-series data of the electrical signal in one interval.

[0041] Figure 9 shows an example of spectrogram generation by the arithmetic unit 200. Figure 9(a) schematically shows the spectral diagrams for each interval TS1 to TSn corresponding to the phase in Figure 8(b). Figure 9(b) divides the phase in Figure 9(a) into multiple intervals and represents the spectral values ​​for each interval S10 as a two-dimensional histogram. The horizontal axis is phase, and the vertical axis is frequency. The numerical values ​​a to f within each section are the spectral values ​​for each interval S10. Thus, a spectrogram is obtained by performing a Fast Fourier Transform on time-series data of an electrical signal over an arbitrary time interval, and representing the frequency on the vertical axis, the phase with respect to the applied voltage on the horizontal axis, and the spectrum corresponding to the amount of charge as a two-dimensional matrix.

[0042] In the generation of the scalogram by the calculation unit 200, a wavelet transform is performed on the time-series data of the electrical signal instead of a fast Fourier transform. In other words, the frequency of occurrence of frequency components resulting from the wavelet transform is represented as a two-dimensional histogram. Thus, the scalogram is a representation of the time-series data of an electrical signal in which a wavelet transform has been performed on an arbitrary time interval, with frequency on the vertical axis, phase with respect to the applied voltage on the horizontal axis, and frequency components corresponding to the amount of charge represented as a two-dimensional matrix.

[0043] The scale adjustment unit 202 can perform scale transformation of time-series data of electrical signals as a result of calculation processing. Figure 10 shows an example of normalization of time-series data of electrical signals. The horizontal axis represents time corresponding to the phase, and the vertical axis represents the amount of charge. Figures 10(a) and (c) show the data before normalization, and Figures 10(b) and (d) show the normalized data.

[0044] As shown in Figure 10, the scale adjustment unit 202 converts the values ​​of the time-series data of the electrical signal so that the reference value measured over an arbitrary time range becomes a predetermined value. For example, the predetermined value is set to 1. This normalization process by the scale adjustment unit 202 makes it possible to unify the signal strength of the electrical signal of the actual device acquired by the data acquisition unit and the simulated electrical signal generated by the electrical signal generation unit 102, even if their scales (ranges of signal strength) are different. In this way, the influence of cases where the maximum and minimum values ​​are not determined or where statistically sudden outliers exist can be suppressed. As a result, even when the signal strength of the simulated electrical signal generated by the electrical signal generation unit 102 differs from that of the electrical signal acquired by the actual device, a decrease in the judgment accuracy of the learning model can be prevented.

[0045] The reference value for normalization may be any value. For example, using the maximum value in one period of the applied voltage or the maximum value over multiple periods as the reference value will make the scaling more stable. In Figure 10, normalization is performed on the charge amount of time-series data as an example, but it is not limited to this. The scale adjustment unit 202 may perform normalization on other data. Alternatively, instead of normalization, the scale adjustment unit 202 may use normalization so that the range between the minimum and maximum values ​​is within a predetermined range.

[0046] The scale adjustment unit 202 can also perform scale transformations on two-dimensional matrix data such as φ-q-n characteristic diagrams, spectrograms, and scalograms. The scale adjustment unit 202 can perform scale transformations such as linear transformation, logarithmic transformation, sigmoid transformation, binarization, normalization, and canonicalization.

[0047] Figure 11 shows an example of applying a logarithmic transformation as a scale transformation to a two-dimensional matrix of the φ-q-n characteristic curve. Figure 11(a) is before the transformation, and Figure 11(b) is after the transformation. The horizontal axis represents the phase, and the vertical axis represents the amount of charge. In this way, by applying a logarithmic transformation to the scale, it is possible to emphasize the electrical signal from partial discharge, where the frequency of charge appearance is low compared to the normal discharge waveform. Note that the scale adjustment unit 202 uses 10 as the base of the logarithm in the logarithmic transformation in Figure 11, but is not limited to this. For example, the scale adjustment unit 202 may use other bases such as Napier's number.

[0048] The data expansion unit 204 performs data augmentation processing. For example, in the acquired data, there may be a bias in the amount of data for each factor of partial discharge. In such cases, the learning model generation unit 110 tends to prioritize factors with a large amount of data when building the learning model, which may lead to a decrease in judgment accuracy for infrequent events.

[0049] Therefore, the data augmentation unit 204 performs data augmentation processing on such data to suppress statistical bias in the training data. This makes it possible to generate statistically unbiased training data when building a learning model, thereby improving the accuracy of factor determination. Similarly, by performing augmentation processing on simulated electrical signals, the data augmentation unit 204 can reduce the time and effort required to acquire data through testing and simulation. In this way, the data augmentation unit 204 can equalize the number of training data for each cause.

[0050] In other words, the data augmentation unit 204 can perform data augmentation processing by, for example, adding random noise to at least one of the electrical signals (simulated electrical signals) generated by the electrical signal generation unit and the electrical signals acquired from a simulator or the like via the data acquisition unit, or by converting the data scale. The data augmentation unit 204 can also perform data augmentation processing by combining these data. For example, the data augmentation unit 204 can perform an augmentation processing method using averaging (or addition) on at least one of the generated electrical signals and the acquired electrical signals.

[0051] The data augmentation method using averaging by the data augmentation unit 204 will be explained in more detail using Figure 12. Figure 12 is a diagram illustrating the data augmentation method using averaging. As shown in Figure 12, the data augmentation unit 204 acquires 50 data points from an electrical signal (which can also include simulated electrical signals) for one partial discharge factor, and extracts 8 data points from it. The data augmentation unit 204 then generates an electrical signal with a different charge amount from the original data by taking the average of the charge amounts of the 8 extracted data points at each time point. In this way, the data can be increased by the number of combinations for which averaging is taken (50C8).

[0052] Note that Figure 12 illustrates combinations of selecting 8 data points from 50 data points, but is not limited to this. For example, the number of source data points and the number of data points to extract can be set arbitrarily. For example, n could be set so as to maximize the number of combinations of extracting n data points from m data points (mCn), or the number of n could be set considering the computational cost. In addition, the augmentation process is sometimes referred to as augmentation.

[0053] Furthermore, data may be randomly extracted from the increased mCn data points. This makes it possible to equalize the number of data points for each partial discharge factor. This allows for adjustment of the training time while maintaining the uniformity of the data points when used as training data.

[0054] Furthermore, as an augmentation method, data obtained directly from the power equipment to be diagnosed may be added to the electrical signals used for training. In particular, data is generated by combining electrical signals obtained when the power equipment to be diagnosed is stopped. When power equipment is stopped, the random noise generated by the power equipment is reduced, making it possible to generate training data with reduced noise while suppressing statistical bias in the training data. Such data augmentation may be performed not only on time-series data but also after converting it into two-dimensional matrix data such as φ-q-n characteristic diagrams, spectrograms, and scalograms.

[0055] The threshold processing unit 206 performs the setting of a threshold. The threshold processing unit 206 can also perform preliminary determination processing using the threshold. For example, the threshold processing unit 206 can determine the presence or absence of a partial discharge signal from a threshold relative to the maximum discharge charge amount. More specifically, the threshold processing unit 206 determines that a partial discharge signal is present if the number of times per second the discharge charge amount exceeds the threshold exceeds a predetermined value. For example, if the threshold processing unit 206 determines that a partial discharge signal is present, it may perform partial discharge factor classification using a machine learning learning model (discriminator). In this way, the partial discharge determination unit 112 can also perform partial discharge presence / absence determination and factor determination using the threshold processing unit 206 and a learning model.

[0056] The filtering unit 208 uses information about the partial discharge signal extracted by the feature extraction unit 104 to generate characteristic ranges from the data used for learning or judgment processing as learning data or judgment data. For example, the filtering unit 208 can perform filtering processes such as bandpass filtering, window functions, and masking on electrical signals. The filtering unit 208 can also perform adjustment processing to align the phase reference point of electrical signals.

[0057] Here, we will explain the details of the feature extraction unit 104. Figure 13 is a block diagram showing an example of the configuration of the feature extraction unit 104. As shown in Figure 13, the feature extraction unit 104 includes a frequency range selection unit 302 and a phase range selection unit 304.

[0058] The frequency range selection unit 302 extracts frequency characteristics of the partial discharge signal based on electrical signals (which may include simulated electrical signals) generated or acquired by, for example, the electrical signal generation unit 102. The frequency range selection unit 302 uses electrical signals for each generation cause to extract frequency characteristics of the partial discharge signal for each generation cause.

[0059] Figure 14 shows an example of a frequency spectrum generated by the Fast Fourier Transform for an electrical signal. The horizontal axis represents frequency, and the vertical axis represents spectrum. Figure 14(a) is a simulated electrical signal with partial discharge, and Figure 14(b) is an electrical signal without partial discharge.

[0060] Comparing Figures 14(a) and (b), it can be seen that there is a range in which the spectral values ​​of the simulated electrical signal with partial discharge are larger than those of the simulated electrical signal without partial discharge. The frequency range selection unit 302 generates a frequency range that exceeds a predetermined threshold as a feature, for example. The frequency range selection unit 302 automatically sets the frequency range according to the magnitude and frequency of the spectrum, for example. Alternatively, the frequency range selection unit 302 may generate a frequency range as a feature determined by a person through the operating device 30. Furthermore, the frequency range may be one region or multiple regions. In addition, the Fast Fourier Transform may be performed on an electrical signal for one period or on an electrical signal for multiple periods. Alternatively, the Fast Fourier Transform may be performed on regions obtained by dividing the electrical signal into arbitrary lengths.

[0061] Figure 15 shows an example of setting a frequency range from the spectral values ​​of electrical signals with multiple partial discharges. The x-axis represents frequency, and the y-axis represents spectrum. Figures 15(a) to (c) are simulated electrical signals with partial discharges. The frequency range selection unit 302 can also select a frequency range characteristic of the partial discharge signal from multiple frequency spectra, as shown in Figure 15, for example. In this case, the frequency range selection unit 302 selects the widest range among the frequency ranges determined for each frequency spectrum as the frequency range of the partial discharge signal. In this way, the influence of noise that occurs sporadically during electrical signal acquisition in the test can be suppressed.

[0062] In this case, the frequency range selection unit 302 may select representative data from among multiple electrical signals without partial discharge as the electrical signal without partial discharge to be compared. Alternatively, the frequency range selection unit 302 may use data obtained by averaging the frequency spectra of multiple electrical signals without partial discharge as the electrical signal without partial discharge to be compared.

[0063] Figure 16 shows an example of how the frequency range selection unit 302 selects a frequency range using a two-dimensional histogram. The horizontal axis represents frequency, and the vertical axis represents spectrum. The numbers within each sub-section represent the frequency of occurrence of spectra that satisfy the conditions within that section. In other words, Figure 16 is a two-dimensional histogram that accumulates the number of occurrences for each frequency and spectrum of the data in Figure 15. The frequency range selection unit 302 uses this two-dimensional histogram to select a frequency range based on the magnitude and number of occurrences of the spectrum. For example, the frequency range selection unit 302 selects the frequency of a section where the spectrum exceeds a predetermined value and the frequency of occurrence also exceeds a predetermined value as the frequency range S20. On the other hand, the frequency range selection unit 302 does not select the frequency range N20 because, although the frequency of occurrence exceeds a predetermined value, the spectrum does not exceed a predetermined value.

[0064] The frequency range selection unit 302 may extract frequency features of the partial discharge signal from the spectrogram and / or scalogram, or a combination thereof. For example, when these are combined, the frequency range selection unit 302 selects the widest range among the frequency ranges determined in each two-dimensional data as the frequency range of the partial discharge signal.

[0065] Figure 17 shows an example of how the frequency range selection unit 302 selects a frequency range using a spectrogram. Figure 17(a) shows an electrical signal with partial discharge. The horizontal axis represents time, and the vertical axis represents the electrical signal. Figure 17(b) shows a spectrogram for the electrical signal with partial discharge. The horizontal axis represents time, and the vertical axis represents frequency. Note that in Figure 17, due to the display density of the drawing, a lower density indicates a stronger spectral intensity.

[0066] As shown in Figure 17(b), there is a frequency range in which the spectral intensity of the electrical signal with partial discharge is greater than that of the electrical signal without partial discharge. The frequency range selection unit 302 can extract frequency ranges exceeding a predetermined value as feature quantities of the partial discharge signal. Feature quantities can also be determined from multiple spectrograms. In this case, the frequency range selection unit 302 selects the widest range among the frequency ranges determined by each spectrogram as the frequency range of the partial discharge signal. Similarly, frequency-related feature quantities are extracted from the spectrogram.

[0067] In this case, the frequency range selection unit 302 may select representative data from among multiple spectrograms without partial discharge as the spectrogram without partial discharge to be compared. Alternatively, the frequency range selection unit 302 may use data obtained by averaging multiple spectrograms without partial discharge as the spectrogram without partial discharge to be compared. Although the above explanation is based on electrical signals (which may include simulated electrical signals) generated or acquired by the electrical signal generation unit 102, it is not limited to this, and processing can be done using any electrical signal. That is, the data used by the frequency range selection unit 302 can be simulated electrical signals, electrical signals acquired from the power equipment to be diagnosed, etc. Furthermore, the operator can set the frequency range selection to any range in advance.

[0068] The phase range selection unit 304 extracts features related to the phase of the partial discharge signal based on the electrical signals (which may include simulated electrical signals) generated or acquired by the electrical signal generation unit 102. The phase range selection unit 304 uses the electrical signals for each generating factor to extract features related to the phase range of the partial discharge signal for each generating factor.

[0069] Figure 18 illustrates a method for selecting a phase range from a φ-q-n characteristic curve. It shows φ-q ​​characteristic curves for multiple electrical signals, where the cumulative number of charge occurrences is calculated over the entire period. Figures 18(a), (b), and (c) are φ-q characteristic curves for different simulated electrical signals. The horizontal axis represents phase (time), and the vertical axis represents charge.

[0070] As shown in Figures 18(a), (b), and (c), the phase range selection unit 304 selects the phase range in which a partial discharge signal is generated for each partial discharge factor's φ-q ​​characteristic diagram, and selects the widest range among the phase ranges of each partial discharge factor as the phase range of the partial discharge signal. The phase range selection unit 304 can automatically set the phase range according to the magnitude and frequency of occurrence of the partial discharge signal. Alternatively, the phase range selection unit 304 may determine the phase range based on the input signal from the operator's control device 30.

[0071] Furthermore, in Figure 18, the phase range selection unit 304 sets the phase ranges S30 and S32 to two regions. This is because it is known that the partial discharge signal occurs near at least one range of the applied voltage phase, between 0° and 90° and 180° and 270°, although it may be set to any number of regions.

[0072] More specifically, the method for automatically setting the phase range of the partial discharge signal by the phase range selection unit 304 will be explained using Figure 19. Figure 19 is a diagram showing an example of generating a charge histogram for the amount of charge in a φ-q characteristic diagram. Figure 19(a) is a φ-q characteristic diagram for an electrical signal with a partial discharge signal. The horizontal axis represents the phase, and the vertical axis represents the amount of charge. Figure 19(b) is a charge histogram of the φ-q characteristic diagram in Figure 19(a). The horizontal axis represents the frequency of charge generation, and the vertical axis represents the amount of charge.

[0073] As shown in Figure 19(b), the phase range selection unit 304 sets data exceeding a predetermined range as a partial discharge signal, centering on the charge amount with the maximum frequency in the charge amount histogram, and extracts the phase range in which the partial discharge signal occurs as feature quantities S40 and S42.

[0074] Figure 20 illustrates a method for selecting the phase range from the peak position of data points in the φ-q characteristic curve. Figures 20(a), (b), and (c) are φ-q characteristic curves for different electrical signals. The horizontal axis represents phase (time), and the vertical axis represents charge. In Figures 20(a), (b), and (c), the peaks of the simulated electrical signals are indicated by arrows.

[0075] As shown in Figures 20(a), (b), and (c), the phase range selection unit 304 can also select the phase range of the partial discharge signal based on the peak position of the simulated electrical signal in the φ-q characteristic diagram of each partial discharge factor. The peak position may be selected as the data point where the charge amount is maximal (minimal), or it may be selected from the results of comprehensive data processing as shown in Figure 22, which will be described later. Note that the peak position can also be determined for signal data for one period. In this case, the phase range is set as a range with statistical variability.

[0076] Figure 21 is a φ-q characteristic diagram showing pseudo-partial discharges indicated by arrows. The horizontal axis represents phase (time), and the vertical axis represents the amount of charge. As described above, when the phase range selection unit 304 sets the phase ranges S34 and S36 of the partial discharge signal using Figure 19, it becomes possible to statistically distinguish it from the partial discharge signal even when pseudo-partial discharges occur over a wide range of phases, as shown in Figure 21.

[0077] Figure 22 shows an example of comprehensive processing of data points on the φ-q characteristic diagram. The horizontal axis represents phase (time), and the vertical axis represents charge quantity. The phase range selection unit 304 can also select the peak position from the results of comprehensive data processing as shown in Figure 22.

[0078] Furthermore, the phase range selection unit 304 can also select a phase range from the centroid position of the data points of the partial discharge signal, similar to the peak position, as a method for selecting feature quantities related to the phase of the partial discharge signal.

[0079] Furthermore, the phase range selection unit 304 may extract features related to the phase of the partial discharge signal from the spectrogram and / or scalogram, or a combination thereof. For example, when these are combined, the phase range selection unit 304 selects the widest range among the phase ranges determined in each two-dimensional data as the phase range of the partial discharge signal.

[0080] Figure 23 shows an example of the phase range selection unit 304 selecting a phase range using a spectrogram. Figure 23(a) shows a simulated electrical signal with partial discharge. The horizontal axis represents time, and the vertical axis represents the electrical signal. Figure 23(b) shows the spectrogram for the simulated electrical signal with partial discharge. The horizontal axis represents time, and the vertical axis represents frequency. Note that in Figure 23, due to the display density of the drawing, a lower density indicates a stronger spectral intensity.

[0081] As shown in Figure 23(b), there is a phase range in which the spectral intensity of the electrical signal with partial discharge is greater than that of the electrical signal without partial discharge. The phase range selection unit 304 can extract the phase ranges S38 and S40 in which the spectral intensity exceeds a predetermined value as feature quantities of the partial discharge signal. Feature quantities can also be determined from multiple spectrograms. In this case, the frequency range selection unit 302 selects the widest range among the frequency ranges determined by each spectrogram as the frequency range of the partial discharge signal. Similarly, in the case of spectrograms, frequency-related feature quantities are extracted.

[0082] In this case, the frequency range selection unit 302 may select representative data from among multiple spectrograms without partial discharge as the spectrogram without partial discharge to be compared. Alternatively, the frequency range selection unit 302 may use data obtained by averaging multiple spectrograms without partial discharge as the spectrogram without partial discharge to be compared. Although the data used in the phase range selection unit 304 has been described based on electrical signals (which may include simulated electrical signals) generated or acquired by the electrical signal generation unit 102, it is not limited to this, and processing can be done using any electrical signal. That is, the data used in the frequency range selection unit 302 can be simulated electrical signals, electrical signals acquired from the power equipment to be diagnosed, etc. Furthermore, the operator can set the phase range selection to any range in advance. In this way, the feature extraction unit 104 can select the frequency range and phase range in which the characteristics of the partial discharge signal appear.

[0083] [Example of processing by a partial discharge diagnostic system] Referring to Figures 25 to 29, an example of the processing of the partial discharge diagnostic system 1 will be explained using Figure 24. Figure 24 is a flowchart of an example of the processing of the partial discharge diagnostic system 1. Here, the cause determination of a partial discharge signal occurring in a turbine generator coil will be used as the electrical signal to be diagnosed. Charge amount data measured by a sensor will be used as the electrical signal to be diagnosed. Note that the electrical signal to be diagnosed includes not only intermittent data obtained from power equipment at all times, but also intermittent data measured periodically. In the following explanation, charge amount will be used as an example of an electrical signal, but it is not limited to this, and signals that fluctuate according to phase can be used. For example, signals other than charge amount, such as electrical signals and voltage signals, may also be used.

[0084] [The process of generating a learning model] First, the electrical signal generation unit 102 generates electrical signals to be used for machine learning. The electrical signal generation unit 102 conducts a simulation test on the insulation degradation patterns occurring in the insulating material of the generator coil and generates electrical signals for each partial discharge factor (step S100). The patterns that constitute partial discharge factors are obtained in advance from field equipment. For example, the electrical signal generation unit 102 simulates the patterns that constitute partial discharge factors as seven patterns A to G and generates electrical signals for each partial discharge factor. In addition, one of the seven patterns A to G generates an electrical signal without partial discharge.

[0085] Next, the calculation unit 200 of the data processing unit 106 generates a spectrogram for each period based on the generated electrical signal (charge quantity signal). Then, the frequency range selection unit 302 of the feature extraction unit 104 selects a frequency range for each partial discharge factor based on the spectrogram for each period (step S102a), and the phase range selection unit 304 selects a phase range (step S102b).

[0086] Next, the data expansion unit 204 of the data processing unit 106 performs data augmentation processing by averaging on the generated charge quantity signal (step S104). In this case, the processing is performed so that the occurrence frequency of the seven patterns that cause partial discharge is made uniform. As shown in Figure 12, averaging processing may also be performed on the measured time series data. In this way, by including electrical signals without partial discharge, it becomes possible to make a comprehensive judgment even for data in which no partial signal occurred.

[0087] Next, the filtering unit 208 of the data processing unit 106 performs bandpass filtering on each charge quantity signal after the augmentation process based on the selected frequency range and phase range (step S106a).

[0088] Next, the scale adjustment unit 202 of the data processing unit 106 performs charge quantity normalization on the charge quantity signal after bandpass filtering. In this case, it normalizes to the value that has the maximum amplitude within one period of the applied voltage (step S108).

[0089] Next, the calculation unit 200 of the data processing unit 106 generates a φ-q-n characteristic diagram based on the normalized charge quantity signal (step S110).

[0090] Next, the scale adjustment unit 202 of the data processing unit 106 performs a logarithmic transformation as a scale transformation with respect to frequency in the φ-q-n characteristic diagram of the charge quantity signal (step S112).

[0091] Next, the image generation unit 108 visualizes the φ-q-n characteristic diagram of the charge quantity signal as a two-dimensional grayscale image, generating a φ-q-n characteristic diagram image (step S114a). Figure 25 shows an example of the imaging process by the image generation unit 108. Figure 25(a) shows an example of the φ-q-n characteristic diagram of the charge quantity signal. Figure 25(b) shows an example of the image being visualized as a two-dimensional grayscale image. The display control unit 116 displays the two-dimensional grayscale image of the φ-q-n characteristic diagram on the display device 20. This allows the operator to visually inspect the characteristics of the φ-q-n characteristic diagram.

[0092] Next, the filtering unit 208 of the data processing unit 106 performs masking on the φ-q-n characteristic diagram image of the charge quantity signal, based on the selected phase range, for phase regions outside the phase range characteristic of the partial discharge signal (step S114b).

[0093] Figure 26 shows an example of masking processing by the filtering unit 208. Figure 26(a) shows an example of imaging into a two-dimensional grayscale image. Figure 26(b) shows an example of masking the two-dimensional grayscale image. Note that in Figure 26, data is deleted by masking, but this is not the only example. For example, the filtering unit 208 masks so that the data values ​​in the phase region outside the phase range become smaller. That is, the filtering unit 208 suppresses the data values ​​in the phase region outside the phase range.

[0094] Next, the learning model generation unit 110 generates a learning model using the φ-q-n characteristic diagram image of the masked charge quantity signal as training data (step S116). Here, one of the seven patterns A to G that cause partial discharge is associated with the φ-q-n characteristic diagram image as a training signal. As a result, this learning model outputs one of the seven patterns A to G that cause partial discharge in response to a two-dimensional array of data input equivalent to the φ-q-n characteristic diagram image.

[0095] [Decision Process] Next, we will explain the process of generating and determining a φ-q-n characteristic diagram image for the charge quantity signal obtained from the turbine generator coil to be diagnosed, similar to the process of generating a learning model.

[0096] The data acquisition unit 100 acquires a time-series electrical signal (charge quantity signal) from the measuring instrument 10 placed on the turbine generator coil, which is the object of measurement (step S118).

[0097] Figure 27 schematically shows an example of the arrangement of a turbine generator coil and sensors. Sensors 402 are placed around each of the U-phase, V-phase, and W-phase conductors 400. Sensors 402 include partial discharge sensors that measure charge amount data. By installing these sensors around the lead conductors of the turbine generator, charge amount data can be easily acquired even in existing plants.

[0098] As shown in Figure 27, sensors may be installed adjacent to the lead conductors of the U, V, and W phases. For example, interphase superposition occurs as shown in the φ-q-n characteristic diagram in Figure 3. Interphase superposition means that the partial discharge signal propagated through the lead conductor reaches the three sensors, resulting in the detection of signals from phases other than the target phase.

[0099] The filtering unit 208 of the data processing unit 106 selects the base point of the applied voltage for each phase and modifies the phase information for the charge quantity signal acquired from the turbine generator coil. At this time, the filtering unit 208 modifies the phase information so that the phases of each charge quantity signal are aligned.

[0100] Figure 28 shows an example of the base point adjustment process for applied voltage by the filtering unit 208. The horizontal axis represents time, and the vertical axis represents the amount of charge. Figures 28(a), (b), and (c) are the charge amount signals acquired from sensors around the lead conductors of the U-phase, V-phase, and W-phase, respectively. Figures 28(d), (e), and (f) are the charge amount signals after selecting a base point and changing the phase information.

[0101] Next, the filtering unit 208 of the data processing unit 106 performs bandpass filtering on each charge quantity signal after the base point adjustment process based on the selected frequency range and phase range (step S106b).

[0102] Next, the scale adjustment unit 202 of the data processing unit 106 performs charge quantity normalization on the charge quantity signal after bandpass filtering. In this case, it normalizes to the value that has the maximum amplitude within one period of the applied voltage (step S120).

[0103] Next, the calculation unit 200 of the data processing unit 106 generates a φ-q-n characteristic diagram based on the normalized charge quantity signal (step S122).

[0104] Next, the scale adjustment unit 202 of the data processing unit 106 performs a logarithmic transformation as a scale transformation with respect to frequency in the φ-q-n characteristic diagram of the charge quantity signal (step S1124).

[0105] Next, the image generation unit 108 visualizes the φ-q-n characteristic diagram of the charge quantity signal after the base point adjustment process as a two-dimensional grayscale image, and generates a φ-q-n characteristic diagram image (step S114c).

[0106] Next, the filtering unit 208 of the data processing unit 106 performs masking on the φ-q-n characteristic diagram image of the charge quantity signal after base point adjustment processing, based on the selected phase range, for phase regions outside the phase range characteristic of the partial discharge signal (step S114d).

[0107] Figure 29 shows an example of masking processing by the filtering unit 208. Figure 29(a) shows an example of imaging the φ-q-n characteristic diagram of the charge quantity signal after base point adjustment processing into a two-dimensional grayscale image. Figure 29(b) shows an example of masking a two-dimensional grayscale image. Note that in Figure 29, data is deleted by masking, but this is not the only example. For example, the filtering unit 208 retains the partial discharge signals S44 and S48 in the phase range, which are feature quantities, and masks the data values ​​of the charge quantity signals S42 and S46 in the phase region outside the phase range to make them smaller. In other words, the filtering unit 208 suppresses the data values ​​of the phase region outside the phase range.

[0108] Next, the partial discharge determination unit 112 uses the learned model to perform a discrimination process using the φ-q-n characteristic diagram image of the masked charge quantity signal as discrimination data (step S116). The partial discharge determination unit 112 outputs one of the seven patterns A to G that cause partial discharge as the discrimination result. In this way, the partial discharge determination unit 112 determines the presence or absence of partial discharge or the cause of its occurrence based on a two-dimensional array of data input equivalent to the φ-q-n characteristic diagram image.

[0109] According to the embodiments described above, data in a predetermined format is generated by suppressing pseudo-partial discharge signals in electrical signals that fluctuate according to phase, and a learning model is generated based on this data to determine at least one of the factors causing partial discharge and whether or not partial discharge is present. In this way, by generating a learning model using machine learning based on data with suppressed pseudo-partial discharge signals, the accuracy of determining the factors causing partial discharge in power equipment can be improved.

[0110] While embodiments of the present invention have been described, they are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of symbols]

[0111] 1: Partial discharge diagnostic system, 10: Measuring instrument, 20: Display device, 40: Partial discharge diagnostic processing device, 100: Data acquisition unit, 102: Electrical signal generation unit, 104: Feature extraction unit, 105: Processing unit, 106: Data processing unit, 108: Image generation unit, 110: Learning model generation unit, 112: Partial discharge determination unit, 200: Calculation unit, 202: Scale adjustment unit, 204: Data expansion unit, 206: Threshold processing unit, 208: Filtering unit, 302: Frequency range selection unit, 304: Phase range selection unit.

Claims

[Claim 1] A partial discharge diagnostic device capable of determining the cause of partial discharge of an insulator, A data expansion unit generates a second electrical signal by performing a uniformization process in which, for each of the factors causing partial discharge, there is a bias in the amount of data, and for factors with low frequency, a predetermined number of first electrical signals are obtained from the first electrical signals that fluctuate according to the phase, a predetermined number of first electrical signals are extracted from the obtained first electrical signals in order to combine them, the charge amount, current, and voltage of the extracted predetermined number of first electrical signals are averaged or added at each time, the number of first electrical signals is increased by the number of combinations for which the average or addition is taken from the obtained predetermined number of first electrical signals, and further, a uniformization process is performed in which first electrical signals are randomly extracted from the increased first electrical signals to generate a second electrical signal, thereby increasing the number of second electrical signals compared to the number of first electrical signals before the uniformization process. A scale adjustment unit that generates a third electrical signal by normalizing the magnitude of the second electrical signal, A calculation unit that generates and processes φ-q-n characteristic image data based on the third electrical signal, A learning model generation unit generates a learning model for determining the low-frequency factors by using the φ-q-n characteristic image data as training data. A feature extraction unit extracts from the first electrical signal the phase range information in which the partial discharge of the first electrical signal occurs as a feature quantity, A filtering unit generates φ-q-n characteristic image data in which electrical signals outside the range of the phase range information are suppressed as pseudo-partial discharge signals from the φ-q-n characteristic image data based on the aforementioned feature quantities. Equipped with, The learning model generation unit uses the suppressed φ-q-n characteristic image data as learning data in the partial discharge diagnostic device.

Images