JP2024129552A - Maintenance and Inspection Equipment - Google Patents

Abstract

To provide a maintenance and inspection device that enables a maintenance and inspection worker to detect any abnormality or a sign of an abnormality without visiting the site, and that can realize labor saving.
[Solution] The maintenance and inspection device 10 has a memory means 12 that stores power information of a transformer that is to be inspected for leakage current and power information of one or more other transformers adjacent to the transformer, a learning means 13 that performs machine learning based on the information stored in the memory means 12, the learning means 13 generating a generative model by performing machine learning based on normal data of the power information stored in the memory means 12, and a judgment means 14 that uses the generative model to judge abnormalities in electrical equipment based on measurement data of the power information stored in the memory means 12.
[Selected figure] Figure 2

Description

The present invention relates to a maintenance and inspection device used for the maintenance and inspection of cubicles (cubicle-type high-voltage power receiving equipment).

In recent years, there has been a demand for automation of systems used for cubicle maintenance and inspection.
For example, Patent Document 1 describes a security system that aims to automate cubicle security inspections that have been performed by humans and perform inspections 24 hours a day. This security system is composed of multiple cubicle devices each having at least two units, and an alarm output device that acquires various information from the cubicle devices via a network.

JP 2021-64384 A

However, even in the case of cubicle maintenance and inspection, it is necessary to balance resolving the current labor (human resource) shortage with improving the quality of maintenance and inspection work. Therefore, in addition to reducing the number of people required for maintenance and inspection work, there is a demand for technology that allows maintenance and inspection workers to detect abnormalities and signs of abnormalities without having to go to the site (inspection location).

Therefore, the security system described in Patent Document 1 outputs an alarm based on a combination of two or more types of information, such as sound information, odor information, and temperature information acquired by a sound collection unit that collects sound and ultrasound, an odor detection unit that detects odors, and a temperature measurement unit that measures temperature (paragraph 0088), and therefore cannot detect abnormalities based on information that is not acquired by so-called sensors such as a sound collection unit or odor detection unit.

The present invention aims to provide a maintenance and inspection device that can detect any abnormalities or signs of abnormalities without the need for a maintenance and inspection worker to go to the site, and can reduce the number of personnel required.

The maintenance and inspection device of the present invention has a storage means for storing power information of a transformer to be inspected for leakage current and power information of one or more other transformers adjacent to the transformer, a learning means for performing machine learning based on the information stored in the storage means, the learning means generating a generative model by performing machine learning based on normal data of the power information stored in the storage means, and a determination means for determining an abnormality in electrical equipment based on measurement data of the power information stored in the storage means using the generative model.

This allows machine learning to be performed using power information from the cubicle transformer being inspected and one or more other adjacent transformers, generating a generative model, which is then used to determine abnormalities in electrical equipment based on the measured power information data.

In addition, it is preferable that the storage means stores information regarding the temperature and/or humidity around the transformer to be inspected, the learning means generates a generative model by performing machine learning based on normal data of the information stored in the storage means, and the determination means uses the generative model to determine an abnormality in the electrical equipment based on the measurement data of the information stored in the storage means.

This allows machine learning to be performed using information other than that of the cubicle transformer being inspected, to generate a generative model.

In particular, it is preferable that the power information stored in the storage means is one or more of the voltage of each phase, the current of each phase, active power, reactive power, apparent power, frequency, and power factor, the learning means generates a generative model of the autoencoder by performing machine learning based on normal data of the information stored in the storage means, and the determination means uses the generative model to output output data using measurement data of the information stored in the storage means as input data, obtains a standard deviation based on the deviation between the input data and the output data, and performs an abnormality determination to determine that an abnormality exists if the standard deviation is outside a certain range.

Furthermore, it is preferable that the storage means stores time series data of each piece of information for a certain period of time, the learning means generates a generative model by performing machine learning based on the time series data of normal data for the certain period stored in the storage means, and the determination means uses the generative model to determine an abnormality in the electrical equipment based on the time series data of measurement data of the information stored in the storage means for the certain period of time.

This allows machine learning to be performed using comprehensive information over a certain period of time (for example, one minute to one hour) rather than instantaneous information, to generate a generative model.

The maintenance and inspection device of the present invention, with such a configuration, performs machine learning using the power information of the cubicle transformer to be inspected and one or more other adjacent transformers, generates a generative model, and uses the generative model to determine anomalies in the electrical equipment based on the measurement data of the power information, so that workers can become aware of any abnormalities (signs of abnormalities) that would not be noticed by visual inspection or inspection using a measuring instrument. Therefore, when workers arrive at the site to perform maintenance and inspection work, they can first (focus on) inspecting areas where there are abnormalities or signs of abnormalities, based on the abnormalities or signs of abnormalities that they were able to detect in advance.

In addition, the storage means stores information about the temperature and/or humidity around the transformer to be inspected, and the learning means generates a generative model by performing machine learning based on normal data of the information stored in the storage means, or the storage means stores time series data of each piece of information for a certain period of time, and the learning means generates a generative model by performing machine learning based on the time series data of normal data for the certain period of time stored in the storage means.In addition, machine learning is performed using information other than the cubicle transformer to be inspected, and a generative model is generated, so that any abnormalities (signs of abnormalities) that an operator would not notice through visual inspection or inspection using a measuring instrument can be detected with greater accuracy.

1 is a schematic configuration diagram of a maintenance and inspection system according to an embodiment of the present invention. 1 is a schematic configuration diagram of a maintenance and inspection device according to an embodiment of the present invention. FIG. 13 is a diagram for explaining a determination model used for abnormality determination. 1 is a schematic configuration diagram of a sensor group according to an embodiment of the present invention; FIG. 2 illustrates an example of a sensor. FIG. 13 is a diagram showing an example of measurement data displayed on a terminal. FIG. 4 is a diagram showing an example of an operation screen of the camera.

The following describes in detail an embodiment of the present invention. However, the following description of the components is merely an example (representative example) of an embodiment of the present invention, and the present invention is not limited to the following content as long as it does not deviate from the gist of the present invention.

[Maintenance and inspection system]
FIG. 1 is a schematic diagram of a maintenance and inspection system according to an embodiment of the present invention.
1, the maintenance and inspection system 1 includes a maintenance and inspection device 10, a sensor group 20, and a collection device 30. These are connected to each other via a communication network N.

The sensor group 20 consists of multiple sensors that are attached to the cubicles at the inspection sites on-site. The sensor group 20 is attached to each of the cubicles at the inspection sites (inspection site A, inspection site B). The sensor group 20 acquires various information about the cubicles at the inspection sites, and details of this will be described later.

The collection device 30 also collects various information acquired by the sensor group 20 and transmits it to the maintenance and inspection device 10 via the network communication network N.

The maintenance and inspection device 10 then determines whether there is an abnormality in the electrical equipment (cubicle) based on the information transmitted from the collection device 30. This allows a worker (e.g., a chief electrical engineer) to understand the condition of the cubicle (e.g., whether there is an abnormality such as leakage current) before going to the site to perform maintenance and inspection of the cubicle.

Furthermore, the maintenance and inspection system 1 is connected to terminals of workers 40, terminals of the management company 41, terminals of consumers 42, an IoT center 50, and the like, via a network communication network N. A terminal is, for example, a device equipped with communication and display functions, such as a smartphone, tablet, or PC.

As described above, the worker 40 is a person who visits the site to perform maintenance and inspection of the cubicle, such as a chief electrical engineer. The management company 41 is a company that employs the worker 40 and provides the service of maintaining and inspecting the cubicle.
On the other hand, the consumer 42 is a company or an individual who has installed a cubicle within their company, factory, facility, or the like.

The IoT center 50 is a company that handles fee collection on behalf of the customers. For example, when the IoT center 50 receives a request from the management company 41 to collect fees for maintenance and inspection (maintenance and inspection costs), the IoT center 50 requests the collection agency service company 51 to issue an invoice to the customer 42. When the customer 42 receives the invoice and pays the maintenance and inspection costs, the management company 41 can collect the maintenance and inspection costs from each customer 42.

[Maintenance and inspection equipment]
FIG. 2 is a schematic diagram of a maintenance and inspection device according to an embodiment of the present invention.
As shown in FIG. 2, the maintenance and inspection device 10 includes a communication means 11, a storage means 12, a learning means 13, a determination means 14, and a control means 15.

The communication means 11 communicates with other devices, such as the collection device 30, the terminal of the worker 40, and the terminal of the management company 41 (see Figure 1), via the network communication network N.

The storage means 12 stores various information acquired by the sensor group 20 and collected and transmitted by the collection device 30. The storage means 12 also stores other information required by the maintenance and inspection device when performing processing. Note that the storage means 12 does not have to be provided inside the maintenance and inspection device 10, and may be, for example, cloud storage.

The information stored in the storage means 12 includes power information of the cubicle transformer (hereinafter simply referred to as "transformer") that is the object of leakage current inspection. This power information includes normal data and measurement data, and the normal data is the power information when the transformer is operating normally (when there is no abnormality in the transformer).
On the other hand, the measurement data is power information of a transformer that is the subject of maintenance and inspection (the subject of determining whether or not there is an abnormality), and is data acquired in real time.

The learning means 13 performs machine learning based on the information stored in the storage means 12. For example, the learning means 13 uses an autoencoder to perform machine learning based on the information stored in the storage means 12 and generate a generative model.

More specifically, the learning means 13 uses an autoencoder to perform machine learning based on the normal data of power information stored in the storage means 12 to generate a generative model.

Then, the judgment means 14 uses the generative model generated by the learning means 13 to judge anomalies in the electrical equipment based on the measurement data of the power information stored in the storage means 12. Since the generative model generated by the learning means 13 is used to judge anomalies in the electrical equipment in this way, hereinafter, the generative model is referred to as a judgment model.

The control means 15 controls processes other than those performed by the communication means 11 to the determination means 14. For example, the control means 15 displays the measurement data stored in the storage means 12 and the state of the transformer determined by the determination means 14 on the terminal of the worker 40 (smartphone screen) and the terminal of the management company 41 (personal computer screen).

[Decision model]
FIG. 3 is a diagram for explaining a determination model for performing an abnormality determination.
As described above, the learning means 13 generates a decision model using an autoencoder. Here, the autoencoder has a network structure such as an input (=feature) [X], an encoder [Encoder], a compressed (dimensionally reduced) feature [Z], a decoder [Decoder], and an output (=feature) [X'], as shown in the upper diagram of Fig. 3.

The goal of autoencoder training is to output data from the output layer that matches the input data input from the input layer. Therefore, during the autoencoder training process, the weights of each edge are adjusted so that the input and output match.

In other words, during the learning process, the autoencoder constructs a neural network so that input [X] = output [X']. Therefore, when normal data is input to a judgment model generated by machine learning based on normal data, the deviation between the input [X] and the output [X'] will be small. On the other hand, when abnormal data (data different from normal data) is input to the judgment model, the deviation between the input [X] and the output [X'] will be large. In short, since the judgment model attempts to restore the input data and output the same data, the more the data cannot be restored, the more it is clear that the input data is abnormal.

The learning means 13 of the maintenance and inspection device 10 generates a judgment model by performing machine learning based on the normal data of the power information stored in the storage means 12. Therefore, when the measurement data is input to this generated judgment model,
(1) When the measurement data is normal, the input measurement data and its features are almost identical to the restored output data and its features, so the deviation between the input and output is small.
(2) When the measurement data is abnormal, the input measurement data and its features will differ from the restored output data and its features, resulting in a large deviation between the input and output.

In this way, the judgment means 14 uses the judgment model generated by the learning means 13 to judge an abnormality in the electrical equipment from the deviation between the input and output based on the measured data of the power information stored in the storage means 12. Here, the judgment means 14 can perform judgment processing based on the standard deviation by taking the standard deviation of the deviation.

For example, as shown in the lower diagram of FIG. 3, the judgment means 14 judges (1) that the standard deviation is "-1σ to 1σ (-1σ or more and 1σ or less)" as normal, (2) that the standard deviation is "-3σ to -1σ (-3σ or more and less than -1σ)" or "1σ to 3σ (more than 1σ and less than 3σ)", and (3) that the standard deviation is "less than -3σ" or "more than 3σ", and judges (warns)" as abnormal.
It can be determined that:
That is, the determining means 14 can determine that some kind of abnormality has occurred when the standard deviation is outside a certain range (-3σ to 3σ).

As described above, when a judgment model is generated by the learning means 13 and the judgment model is used by the judgment means 14 to judge an abnormality, information other than the power information of the cubicle transformer to be inspected can be used.

First, the power information of a transformer includes, for example, the voltage of each phase (in a single-phase three-wire system or a three-phase three-wire system), the current of each phase, active power, reactive power, apparent power, frequency, power factor, etc.

On the other hand, information other than power information is, for example, information about the temperature and/or humidity around the transformer being inspected.

In addition, power information from the transformer in the cubicle being inspected and one or more other adjacent transformers can also be used. That is, the voltages of each phase, the currents of each phase, the active power, the reactive power, the apparent power, the frequency, the power factor, etc. of the adjacent transformers can also be used.

In addition, such information can be obtained in a time series, and the time series data can be used to generate a judgment model, etc. Indices that can be derived from time series data include, for example, effective value, peak value, crest factor, kurtosis, skewness, entropy, and autocorrelation. In short, time series data is data obtained when measuring leakage current, etc., and the aforementioned effective value and peak value are indices (representative values) that are calculated from that time series data. For example, when time series data is obtained at one-second intervals, one effective value can be obtained from 3,600 pieces of time series data (one hour).

Specifically, this time series data can be acquired at one-second intervals (information on earth leakage current) or at one-minute intervals (information other than earth leakage current, etc.).
Furthermore, when performing machine learning or anomaly determination, the learning means 13 and the determination means 14 can use such time series data as it is (as raw data) or derive the index as described above and use it. Furthermore, the learning means 13 and the determination means 14 can use, for machine learning or anomaly determination, information acquired by the sensor group 20 and stored in the storage means 12 after noise has been removed using a filter.

The learning means 13 and the judgment means 14 can generate a judgment model based on such various information and perform an abnormality judgment using the judgment model. Then, by performing an abnormality judgment based on such various information, the maintenance inspection device 10 can become aware of any abnormality (signs of abnormality) that an operator would not notice through visual inspection or inspection using a measuring instrument.

[Sensor group]
FIG. 4 is a schematic diagram of a sensor group according to an embodiment of the present invention.
As shown in FIG. 4 , the sensor group 20 includes a leakage current measuring device 21 , a current multi-sensor 22 , a thermometer 23 , a camera 24 , and a thermo camera 25 .

The leakage current measuring device 21 is a device that measures (gauges) the leakage current of a transformer. The current multi-sensor 22 is a sensor that can measure the voltage of each phase, the current of each phase, active power, reactive power, etc., as described above.

The thermometer 23 is an instrument that measures the temperature of the transformer, and can also measure the temperature around the transformer being inspected. For example, multiple thermometers 23 can be installed in different locations to measure the respective temperatures. Although not shown, the sensor group 20 can also include a hygrometer that measures humidity.

The camera 24 is for photographing each location and each part of the inspection target (cubicle). The object (range) photographed by the camera 24 can be changed by remotely operating the camera 24 from the maintenance and inspection device 10, etc.
In addition, images and videos taken by the camera 24 are collected by the collection device 30 and can be viewed (monitored) in real time from the terminals of the workers 40 and the management company 41 via the maintenance and inspection device 10. The thermo camera 25 acquires images and videos (for example, thermography images of a transformer) showing the temperature distribution of each location and each part of the inspection target.

Various information measured by these sensors (sensor group 20) is collected by the collection device 30 and transmitted to the maintenance and inspection device 10 via the network communication network N, or transmitted via the maintenance and inspection device 10 to the terminal of a worker 40 or the terminal of a management company 41, etc.
Therefore, after workers or the like know the results of the abnormality judgment made by the judgment model, they can visually check using the camera 24 the locations (parts) at the inspection site where there is an abnormality or where there is likely to be an abnormality, based on that result.

Various types of cameras 24 can be used, such as panning cameras and 360-degree cameras. Furthermore, multiple cameras 24 can be installed at the inspection site. Then, image data of areas where abnormalities or suspected abnormalities are found at the inspection site photographed by the cameras 24 can be transferred to the terminal of the worker 40, and used (images can be attached) in the inspection report R created by the worker 40 and sent to the customer 42 (see Figure 1).

Furthermore, FIG. 7 is a diagram showing an example of a camera operation screen, and the camera 24 can be remotely operated from the operation screen shown in FIG.
For example, in the "Camera" menu, you can select which of the multiple installed cameras will display the image in the "Preview". In addition, in the "Preset" menu, you can switch the viewpoint of the pre-set camera. In addition, in the "Pan Tilt" menu, you can control the left/right direction (pan) and up/down direction (tilt) of the camera.

FIG. 5 is a diagram showing an example of a sensor.
As shown in FIG. 5, a current transformer (CT) for measuring Ior is installed around the earth wire of a transformer as a leakage current measuring device 21, and the leakage current of the transformer can be measured.

[display]
FIG. 6 is a diagram showing an example of measurement data displayed on a terminal.
As described above, the control means 15 causes the measurement data acquired by the sensor group 20, collected and transmitted by the collection device 30, and stored in the memory means 12 to be displayed on the terminal of the worker 40 or the management company 41.

Therefore, the worker 40 or the management company 41 can know the information of the cubicle to be inspected, no matter when or where they are. Furthermore, the worker 40 or the management company 41 can know the judgment result of the judgment means 14 no matter when or where they are, so that the worker can know any abnormality or signs of abnormality without going to the site.

Because workers can find out about any abnormalities or signs of abnormalities without going to the site, when they arrive at the site to perform maintenance and inspection work, they can first (and primarily) inspect the areas where there are abnormalities or signs of abnormalities, based on the abnormalities or signs of abnormalities that they were able to find out about in advance.

Conversely, if no abnormality or signs of abnormality are detected, the normality of the inspection target is guaranteed to a certain extent, so the burden on specific workers can be reduced by sending less experienced workers to the inspection site, thereby potentially reducing the number of workers required.

This invention is a maintenance and inspection device that can detect any abnormalities or signs of abnormalities without the need for a maintenance and inspection worker to go to the site, and can reduce the number of people required. It can also be used for statutory safety obligations such as future safety inspections and daily inspections, making it industrially useful.

REFERENCE SIGNS LIST 1 Maintenance and inspection system 10 Maintenance and inspection device 11 Communication means 12 Storage means 13 Learning means 14 Determination means 15 Control means 20 Sensor group 21 Earth leakage current measuring device 22 Current multi-sensor 23 Thermometer 24 Camera 25 Thermo camera 30 Collection device 40 Worker 41 Management company 42 Consumer 50 IoT center 51 Collection agency service company R Inspection report N Network communication network

Claims (4)

A storage means for storing power information of a transformer to be inspected for leakage current and power information of one or more other transformers adjacent to the transformer;
a learning means for performing machine learning based on the information stored in the storage means, the learning means generating a generative model by performing machine learning based on normal data of the power information stored in the storage means;
A maintenance and inspection device having a determination means for determining an abnormality in electrical equipment based on the measurement data of power information stored in the storage means, using the generation model. The storage means stores information regarding the temperature and/or humidity around the transformer to be inspected,
The learning means generates a generative model by performing machine learning based on normal data of the information stored in the storage means;
2. The maintenance and inspection device according to claim 1, wherein the determining means uses the generative model to determine an abnormality in the electrical equipment based on the measurement data of the information stored in the storage means. The power information stored in the storage means is one or more of information on a voltage of each phase, a current of each phase, an active power, a reactive power, an apparent power, a frequency, and a power factor,
The learning means generates a generative model of the autoencoder by performing machine learning based on normal data of the information stored in the storage means;
The maintenance and inspection device according to claim 2, wherein the judgment means uses the generation model to input measurement data of the information stored in the memory means as input data and output data, obtains a standard deviation based on the deviation between the input data and the output data, and performs an abnormality judgment in which an abnormality is judged to exist if the standard deviation is outside a certain range. The storage means stores time series data of each piece of information for a certain period of time,
The learning means generates a generative model by performing machine learning based on the time-series data of the normal data for the certain period stored in the storage means;
The maintenance and inspection device according to claim 3 , wherein the determination means uses the generation model to determine an abnormality in electrical equipment based on time-series data of the measurement data of the information stored in the storage means for the certain period.