WO2024080415A1

WO2024080415A1 - Method for determining type of partial discharge of live-line high-voltage stator winding by applying support vector machine technique to prpd pattern image

Info

Publication number: WO2024080415A1
Application number: PCT/KR2022/015580
Authority: WO
Inventors: 이상규; 김용주; 이진
Original assignee: (주)오앤엠 코리아
Priority date: 2022-10-12
Filing date: 2022-10-14
Publication date: 2024-04-18
Also published as: KR102518013B1

Abstract

The present invention relates to a method for determining the type of partial discharge of a live-line high-voltage stator winding by applying a support vector machine technique to a PRPD pattern image, in which a partial discharge pattern generated in a motor of a power plant, etc. is automatically analyzed to enable determination of the type of partial discharge. In order to solve the above problem, the method for determining the type of partial discharge of a live-line high-voltage stator winding by applying a support vector machine technique to a PRPD pattern image, according to the present invention, comprises: a partial discharge signal detection step; a signal processing step; a partial discharge pattern generation step; a data processing step; a multi-class support vector machine application step; a risk diagnosis step; and an output step.

Description

Method for determining live-line high-voltage stator winding partial discharge type by applying support vector machine technique to PRPD pattern image

The present invention relates to a method for determining the type of partial discharge in a live high-voltage stator winding by applying the support vector machine technique to a PRPD pattern image. More specifically, it relates to internal discharge, surface discharge, single-wire discharge, phase-to-phase discharge and discharge occurring in high-voltage rotating machines such as power plants. This relates to a method of automatically identifying various types of partial discharges, such as slot discharges, through Support Vector Machine (SVM) analysis techniques.

In general, a live-line partial discharge (PD) diagnostic system is applied to power plants, etc. to prevent early failures due to insulation deterioration of high-voltage stator windings, etc. As a prior art for this purpose, the partial discharge diagnosis system of Registered Patent Publication No. 2037103 is used. An analysis device and method (hereinafter referred to as 'patent document') is disclosed.

The patent document receives a first signal and a second signal for two of the three phases from a first sensor and a second sensor located around the cable junction box, and a third signal that is a differential signal between the first signal and the second signal. A signal acquisition unit that acquires a signal; a partial discharge determination unit that determines the occurrence and location of partial discharge in the junction box based on the obtained partial discharge occurrence pattern and the analyzed waveform targeting the first signal, the second signal, and the differential signal; a pattern acquisition unit that obtains a partial discharge generation pattern by performing PRPD (Phase Resolved Partial Discharge) analysis on the first, second, and third signals; and a waveform analysis unit that analyzes the waveforms of the first signal, the second signal, and the third signal, and the partial discharge determination unit determines whether and where the partial discharge occurs based on the waveform analyzed by the waveform analysis unit. It includes a waveform determination unit that determines, wherein the waveform determination unit compares the polarity of the waveform peaks of the first signal and the second signal to detect a partial discharge due to an external signal or a partial discharge due to noise, The shorter the time interval of the waveform peak between the first signal and the second signal, the closer to the location of the sensors it is determined that partial discharge occurred, and the waveform peaks of the first signal and the second signal are generated by linking them. It is made to determine whether partial discharge has occurred based on the outline shape of the waveform.

However, there are various patterns of partial discharge, and depending on these patterns, there are differences in the risk of partial discharge and the type of partial discharge. However, in the case of the conventional partial discharge diagnosis system, even if a partial discharge is diagnosed through the system, the operator must determine the partial discharge pattern. Since the type of partial discharge must be additionally determined empirically by checking separately, there are many difficulties in quickly and accurately determining the type of partial discharge.

Therefore, there is a need for the development of an improved partial discharge diagnosis method that can detect and diagnose the type of partial discharge by analyzing the partial discharge pattern.

The present invention was created to solve the problems of the conventional partial discharge diagnosis method as described above. The problem that the present invention aims to solve is to automatically analyze partial discharge patterns occurring in high-voltage rotors of power plants, etc. to determine the type of partial discharge. It provides a method for determining the type of partial discharge in a live high-voltage stator winding by applying the support vector machine technique to a discernible PRPD pattern image.

The method for determining the partial discharge type of a live stator winding by applying the support vector machine technique to the PRPD pattern image according to the present invention to solve the above problems includes a partial discharge signal detection step of detecting a partial discharge signal from a partial discharge sensor; A signal processing step of removing noise from the signal detected through the partial discharge signal detection step and synchronizing the pulse signal; A partial discharge pattern generating step of generating a partial discharge pattern by accumulating signals synchronized through the signal processing step; A data processing step of dividing the partial discharge pattern generated through the partial discharge pattern generating step into regions by size and phase and processing the data; A multi-class support vector machine application step of applying the data for each image area obtained through the data processing step to a support vector machine (SVM) in which a plurality of partial discharge types are machine-learned; A risk diagnosis step of diagnosing the risk according to the partial discharge type determined through the multi-class support vector machine application step; and an output step that guides the determined partial discharge type and risk through a display.

In the present invention, the data processing step divides the partial discharge pattern into equal parts 12 to 36 on the Another characteristic is that it is converted into data as an overall pattern.

In addition, the present invention is to determine the partial discharge type by inputting the data extracted in the data processing step to a support vector machine in which a plurality of partial discharge types are machine-learned as each pattern in the multi-class support vector machine application step. It has other features.

In addition, the present invention provides partial discharge patterns of internal discharge, mica tape peeling discharge, peeling discharge between conductors and insulators, slot discharge, terminal winding corona discharge, surface discharge, and phase-to-phase discharge in the application step of the multi-class support vector machine. Another characteristic is that it is learned and classified.

In addition, in the present invention, the greater the number of partial discharge signals at higher positions relative to the Y-axis compared to the total number of partial discharge signals on the image divided in the risk diagnosis step, the higher the risk, and the greater the number of partial discharge signals at low positions, the lower the risk, Another characteristic of risk is that it is divided into five levels: very dangerous, dangerous, average, good, and very good, or four levels: dangerous, bad, good, and very good.

According to the present invention, when a partial discharge occurs, the partial discharge pattern is divided into a plurality of regions according to phase and size, the number of occurrences of the partial discharge signal is counted and converted into data, and this partial discharge data is sent to a multi-class support vector machine machine-learned by type. Because it is applied, the type of partial discharge can be accurately determined using artificial intelligence, and at the same time, it has the advantage of diagnosing and guiding the risk of partial discharge based on the size and number of occurrences of the partial discharge signal.

1 is a flowchart showing an example of a method for determining a live high-voltage stator winding partial discharge type by applying the support vector machine technique to a PRPD pattern image according to the present invention.

Figure 2 is a configuration diagram showing an example of a method for determining the type of partial discharge of a live high-voltage stator winding by applying the support vector machine technique to the PRPD pattern image according to the present invention.

Figure 3 is a diagram showing an example of data processing steps according to the present invention.

Figure 4 is a diagram showing an example of an internal discharge pattern.

Figure 5 shows an example of a mica tape peeling discharge pattern.

Figure 6 shows an example of a discharge pattern resulting from delamination between a conductor and an insulating material.

Figure 7 is a diagram showing an example of a slot discharge pattern.

Figure 8 is a diagram showing an example of a terminal winding corona discharge pattern.

Figure 9 is a diagram showing an example of a surface discharge pattern.

Figure 10 is a diagram showing an example of a phase-to-phase discharge pattern.

Figure 11 is a diagram showing an example of a multi-class support vector machine application step according to the present invention.

[Explanation of symbols]

1: Partial discharge sensor 2: Partial discharge identification device

10: signal processing unit 11: noise filter

12: synchronous pulse generation circuit 20: partial discharge pattern generation unit

30: Data processing unit 40: Multi-class SVM application unit

50: Risk diagnosis unit 60: Output unit

Hereinafter, the present invention will be described in detail according to the accompanying drawings showing preferred embodiments.

The present invention aims to provide a method for determining the type of partial discharge in a live high-voltage stator winding by applying the support vector machine technique to a PRPD pattern image that can automatically analyze the partial discharge pattern occurring in a motor of a power plant, etc. to determine the type of partial discharge. As shown in FIG. 1, the present invention includes a partial discharge signal detection step (S10), a signal processing step (S20), a partial discharge pattern generation step (S30), a data processing step (S40), and a multi-class support vector machine application step. (S50), a risk diagnosis step (S60), and an output step (S70).

In addition, as shown in FIG. 2, the device 2 for determining the partial discharge type of the present invention includes a signal processing unit 10, a partial discharge pattern generation unit 20, a data processing unit 30, and a multi-class SVM application unit ( 40), and includes a risk diagnosis unit 50 and an output unit 60.

(1) Partial discharge signal detection step (S10)

This step is to detect a partial discharge signal from a motor such as a power plant using a partial discharge sensor. The present invention is configured to detect (acquire) a partial discharge signal using a partial discharge sensor 1 of various known structures. You can.

This example may consist of a partial discharge sensor 1 that is directly mounted on the stator winding slot of the motor to detect partial discharge signals more quickly and accurately.

(2) Signal processing step (S20)

In this step, the signal detected from the partial discharge sensor 1 is received through the signal processing unit 10, the noise included in the signal is removed through the noise filter 11, and the noise-removed analog signal is converted to A/ It is converted into a digital signal through a D converter (not shown), and the phase of the signal converted into a digital signal is synchronized through the synchronization pulse generation circuit 12 to easily generate a partial discharge pattern, which will be described later.

At this time, in order to sufficiently obtain a partial discharge pattern, the present invention acquires a signal of 6,000 cycles, which is 100 seconds (1 minute 40 seconds), and synchronizes it to 60 Hz through the synchronization pulse generation circuit 12.

(3) Partial discharge pattern generation step (S30)

This step generates a partial discharge pattern through the partial discharge pattern generator 20 from a signal synchronized to 60 Hz through the signal processing step (S20) above. For this purpose, signals are accumulated on a 360° phase basis (1 cycle). A pattern is formed.

(4) Data processing step (S40)

In this step, the partial discharge pattern generated through the above partial discharge pattern generation step (S30) is divided into regions of a predetermined size and phase through the data processing unit 30, and then the partial discharge signal located in the divided region is generated as a whole. The data is processed according to the pattern.

At this time, the It can be divided into equal parts to form a total of 432 spaces.

Alternatively, each space divided on the

As described above, with multiple cycles of partial discharge signals accumulated, an overall partial discharge pattern is formed by detecting the presence or absence of a partial discharge signal in an area divided by phase and size, and the partial discharge pattern formed in this way is described later as a multi-discharge pattern. The partial discharge type is determined through the class support vector machine application step (S50).

(5) Multi-class support vector machine application step (S50)

In this step, after one partial discharge pattern is generated in the data processing step (S40), the partial discharge type is determined through the multi-class SVM application unit (40) of the partial discharge pattern thus generated.

At this time, the multi-class support vector machine machine learns the partial discharge type and determines a partial discharge pattern with high similarity between the input partial discharge pattern and the previously learned partial discharge pattern. An example of the partial discharge pattern learned for this purpose is Figure 4. An internal discharge pattern as shown in, a mica tape peeling discharge pattern as shown in FIG. 5, a discharge pattern due to peeling between a conductor and an insulating material as shown in FIG. 6, and a discharge pattern as shown in FIG. 7. It may include a slot discharge pattern, a terminal winding corona discharge pattern as shown in FIG. 8, a surface discharge pattern as shown in FIG. 9, and a phase-to-phase discharge pattern as shown in FIG. 10, and the generated From the partial discharge pattern, internal discharge, mica tape peeling discharge, peeling discharge between conductor and insulating material, slot discharge, terminal winding corona discharge, surface discharge, and interphase discharge are determined.

As shown above, in the multi-class support vector machine, as shown in FIG. 11, multiple types are input as class data and learned repeatedly, and two or more types are compared and classified.

(6) Risk diagnosis step (S60)

In this step, after the partial discharge type is determined through the above multi-class support vector machine application step (S50), the risk according to the size of the partial discharge pattern is diagnosed through the risk diagnosis unit 50.

At this time, the risk level is determined by detecting the size of the partial discharge pattern divided into size and phase, that is, the number of partial discharge signals present based on the Y axis, and guiding the risk level through this.

For this example, as shown in FIG. 3, the greater the number of partial discharge signals in area D (high position) based on the Y axis, the higher the risk, the greater the number of partial discharge signals in area A (low position), the lower the risk, and the overall area ( The greater the number of partial discharge signals in the area (A, B, C, D), the higher the risk, and the fewer signals in the entire area (A, B, C, D), the lower the risk can be diagnosed, and in addition, partial discharge signals can be detected. If not, the risk may be diagnosed as lowest.

Also, if there are many partial discharge signals in area D, it is very dangerous, if there are many partial discharge signals in area C, it is dangerous, if there are many partial discharge signals in area B, it is normal, and if there are many partial discharge signals in area A, it is good. , if no partial discharge signal is detected, the diagnosis can be divided into five levels, such as very good.

Alternatively, it can be diagnosed in four levels, such as dangerous, bad, good, and very good.

(7) Output stage (S70)

In this step, the partial discharge type and risk determined through the multi-class support vector machine application step (S50) and the risk diagnosis step (S60) are guided to the worker through the output unit (60, display).

The worker checks the type and risk of partial discharge displayed on the display and inspects the motor, etc., as necessary.

As described above, in the present invention, when a partial discharge occurs, the partial discharge pattern is divided into a plurality of regions according to phase and size, the number of occurrences of the partial discharge signal is counted and converted into data, and this partial discharge data is machine-learned by type into a multi-class support vector. Since it is applied to machines, it is possible to accurately determine the type of partial discharge using artificial intelligence.

In addition, the risk of partial discharge is diagnosed and guided based on the size and number of occurrences of the partial discharge signal.

Above, for convenience of explanation, reference numerals and names have been given to the drawings showing preferred embodiments and the configurations shown in the drawings. However, this is an embodiment according to the present invention, and the shapes shown in the drawings and the names given are The scope of the rights should not be construed as limited, and changes to various shapes predictable from the description of the invention and simple substitution with a configuration that performs the same function are within the scope of changes that can be easily carried out by a person skilled in the art. You will see this as extremely self-evident.

Claims

A partial discharge signal detection step (S10) of detecting a partial discharge signal from a partial discharge sensor;

A signal processing step (S20) of removing noise from the signal detected through the partial discharge signal detection step (S10) and synchronizing the pulse signal;

A partial discharge pattern generation step (S30) of generating a partial discharge pattern by accumulating signals synchronized through the signal processing step (S20);

A data processing step (S40) of dividing the partial discharge pattern generated through the partial discharge pattern generating step (S30) into regions by size and phase and processing the data;

A multi-class support vector machine application step (S50) of applying the data for each image area obtained through the data processing step (S40) to a support vector machine (SVM) in which a plurality of partial discharge types are machine-learned;

A risk diagnosis step (S60) of diagnosing the risk according to the partial discharge type determined through the multi-class support vector machine application step (S50); and

An output step (S70) that guides the identified partial discharge type and risk through a display;

A method for determining the type of partial discharge in a live high-voltage stator winding by applying a support vector machine technique to a PRPD pattern image, characterized in that it consists of:
In claim 1,

The data processing step (S40) is,

The partial discharge pattern is divided into 12~36 on the A method for determining the type of partial discharge in a live high-voltage stator winding by applying the support vector machine technique to the characteristic PRPD pattern image.
In claim 2,

In the multi-class support vector machine application step (S50),

The data extracted in the data processing step (S40) is input to a support vector machine in which a plurality of partial discharge types are machine-learned as respective patterns, and the partial discharge type is determined by applying the support vector machine technique to the PRPD pattern image. Method for determining the type of partial discharge in the applied live high-voltage stator winding.
In claim 3,

In the multi-class support vector machine application step (S50),

Support vector for PRPD pattern image, characterized in that partial discharge patterns of internal discharge, mica tape peeling discharge, peeling discharge between conductor and insulating material, slot discharge, terminal winding corona discharge, surface discharge, and phase-to-phase discharge are classified through machine learning. A method for determining the type of partial discharge in live high-voltage stator windings using machine techniques.
In claim 2,

In the risk diagnosis step (S60),

Compared to the total number of partial discharge signals in the divided image, the more partial discharge signals there are at higher positions relative to the Y-axis, the higher the risk, and the more partial discharge signals there are at low positions, the lower the risk. The risk levels are very dangerous, dangerous, normal, A method for determining the type of partial discharge in a live high-voltage stator winding by applying the support vector machine technique to a PRPD pattern image characterized by 5 levels of good, very good, or 4 levels of dangerous, bad, good, and very good.