JP7401514B2 - Heat exchanger tube damage cause inference device and heat exchanger tube damage cause inference method - Google Patents

Description

The present disclosure relates to a heat exchanger tube damage cause inference device and a heat exchanger tube damage cause inference method.

Patent Document 1 discloses a method for determining a portion of a heat exchanger tube on a boiler furnace wall that has a high risk of overheating damage due to powder scale or the like. In this method, the temperature of multiple parts of the furnace wall is measured twice, a threshold is set through statistical processing based on the difference in temperature at each time, and parts where the difference deviates from the threshold are determined to be high-risk parts. .

JP2018-132240A

By the way, conventionally, when damage such as a blowout occurs to a heat transfer tube, workers are dispatched to the site to investigate the target area in detail, identify the cause, and then take specific measures (for example, replace parts). It was necessary to stop operation of the plant, including the heat transfer tubes, for a long period of time. Therefore, there is a need to quickly identify the cause of damage to heat exchanger tubes in order to increase plant availability.

In this regard, in the method described in Patent Document 1, although it is possible to suppress the occurrence of damage to heat exchanger tubes by monitoring areas determined to be high-risk areas and cleaning them as necessary, in practice When damage to heat exchanger tubes occurs, it is not possible to infer the cause and it is not possible to take prompt action against the damage to the heat exchanger tubes.

In view of the above circumstances, at least one embodiment of the present disclosure provides a heat exchanger tube damage cause inference device that makes it possible to quickly take action against damage to heat exchanger tubes by inferring the cause of damage to the heat exchanger tubes. The purpose of the present invention is to provide a method for inferring the cause of heat exchanger tube damage.

In order to achieve the above object, a heat exchanger tube damage cause inference device according to at least one embodiment of the present disclosure includes:
Machine learning using training data that includes a plurality of learning image data each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and a damage cause associated with each of the plurality of learning image data. a machine learning unit configured to generate a learning model for inferring the cause of damage to the heat exchanger tube from the image data;
a target image data acquisition unit configured to acquire target image data, which is image data indicating a damaged portion of the target heat exchanger tube, for the target heat exchanger tube from which the cause of damage is to be inferred;
Equipped with
The learning model is configured to output an inference result of a cause of damage to the target heat exchanger tube by receiving the target image data acquired by the target image data acquisition unit.

In order to achieve the above object, a heat exchanger tube damage cause inference method according to at least one embodiment of the present disclosure includes:
Machine learning using training data that includes a plurality of learning image data each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and a damage cause associated with each of the plurality of learning image data. a machine learning step of generating a learning model used to infer the cause of damage to the heat exchanger tube;
a target image data acquisition step of acquiring target image data that is image data indicating a damaged part of the target heat exchanger tube for the target heat exchanger tube from which the cause of the damage is to be inferred;
inputting the target image data acquired in the target image data acquisition step into the learning model, so that the learning model outputs an inference result of the cause of damage to the target heat exchanger tube;
Equipped with

According to at least one embodiment of the present disclosure, a heat exchanger tube damage cause inference device and a heat exchanger tube damage cause make it possible to take prompt action against damage to a heat exchanger tube by inferring the cause of damage to the heat exchanger tube. An inference method is provided.

It is a diagram showing an example of the hardware configuration of a heat exchanger tube blowout cause inference device 2 according to an embodiment of the present disclosure. FIG. 2 is a block diagram showing an example of the functional configuration of the heat exchanger tube blowout cause inference device 2 shown in FIG. 1. FIG. It is a figure which shows an example of the flow of inference and display of the cause of heat exchanger tube damage using the heat exchanger tube blowout cause inference device 2 mentioned above. 3 is a diagram illustrating an example of a flow of generating learning image data Dfl by the learning image data generation unit 12. FIG. It is a figure which shows an example of the image data Db which the learning image data generation part 12 acquires. It is a figure which shows the image part inferred by the learning model Md0. FIG. 3 is a diagram for explaining a method of generating learning image data. FIG. 6 is a diagram showing an example of a display screen E1 of the display unit 14d of the user's terminal 14 when inputting target image data Da and the like from the user's terminal 14 to the heat exchanger tube blowout cause inference device 2. 9 is a diagram showing an example of a display screen E2 outputted to the display unit 14d of the terminal 14 by pressing the inference button F6 in FIG. 8. FIG. 9 is a diagram showing another example of the display screen E2 outputted to the display section 14d of the terminal 14 by pressing the inference button F6 in FIG. 8. FIG.

Hereinafter, some embodiments of the present disclosure will be described with reference to the accompanying drawings. However, the dimensions, materials, shapes, relative arrangement, etc. of the components described as embodiments or shown in the drawings are not intended to limit the scope of the invention thereto, and are merely illustrative examples. .
For example, expressions expressing relative or absolute positioning such as "in a certain direction,""along a certain direction,""parallel,""orthogonal,""centered,""concentric," or "coaxial" are strictly In addition to representing such an arrangement, it also represents a state in which they are relatively displaced with a tolerance or an angle or distance that allows the same function to be obtained.
For example, expressions such as "same,""equal," and "homogeneous" that indicate that things are in an equal state do not only mean that things are exactly equal, but also have tolerances or differences in the degree to which the same function can be obtained. It also represents the existing state.
For example, expressions expressing shapes such as squares and cylinders do not only refer to shapes such as squares and cylinders in a strict geometric sense, but also include uneven parts and chamfers to the extent that the same effect can be obtained. Shapes including parts, etc. shall also be expressed.
On the other hand, the expressions "comprising,""comprising,""comprising,""containing," or "having" one component are not exclusive expressions that exclude the presence of other components.

(Configuration and functions of heat exchanger tube blowout cause inference device 2)
FIG. 1 is a diagram illustrating an example of the hardware configuration of a heat exchanger tube blowout cause inference device 2 according to an embodiment of the present disclosure. FIG. 2 is a block diagram showing an example of the functional configuration of the heat exchanger tube blowout cause inference device 2 shown in FIG. 1. As shown in FIG. The heat transfer tube blowout cause inference device 2 is a device for inferring the cause of a heat transfer tube blowout used in a heat exchanger of a boiler (not shown), etc., as described below.

As shown in FIG. 1, the heat exchanger tube blowout cause inference device 2 includes, for example, a processor 72, a RAM (Random Access Memory) 74, a ROM (Read Only Memory) 76, an HDD (Hard Disk Drive) 78, an input I/F 80, and an output I/F 82, which are configured using a computer connected to each other via a bus 84. Note that the hardware configuration of the heat exchanger tube blowout cause inference device 2 is not limited to the above, and may be configured by a combination of a control circuit and a storage device. The heat exchanger tube blowout cause inference device 2 is configured by a computer executing a program that implements each function of the heat exchanger tube blowout cause inference device 2. The functions of each part in the heat exchanger tube blowout cause inference device 2 described below include, for example, loading a program held in the ROM 76 into the RAM 74 and executing it in the processor 72, as well as reading and writing data in the RAM 74 and ROM 76. It is realized by

As shown in FIG. 2, the heat exchanger tube blowout cause inference device 2 includes a target image data acquisition unit 4, a machine learning unit 6, a reference information storage unit 8, a reference information search unit 9, a screen data output unit 10, and a learning An image data generation section 12 is provided.

The target image data acquisition unit 4 is configured to acquire target image data Da, which is image data indicating the blowout portion of the target heat exchanger tube, for the target heat exchanger tube whose cause of the blowout is to be inferred. The target image data acquisition unit 4 acquires target image data Da from the user's terminal 14 via the information communication network Nc, for example. The target image data acquisition unit 4 may acquire the target image data Da as well as location information Sa indicating the location of the heat exchanger tube indicated by the target image data Da in association with the target image data Da.

The machine learning unit 6 stores a plurality of learning image data Dfl each indicating a blowout part of a heat exchanger tube whose cause of blowout has been identified in the past, and a blowout cause Cd associated with each of the plurality of learning image data Dfl. By performing machine learning using teacher data Dt including , a learning model Md1 for inferring the cause of a heat exchanger tube blowout from image data is generated. Here, the teacher data Dt used by the machine learning unit 6 for machine learning may include location information St associated with each of the plurality of learning image data Dfl, and the location information St is different from the learning image data Dfl. This is information indicating the location of the heat exchanger tube shown in .

The learning model Md1 is based on the target image data Da acquired by the target image data acquisition unit 4 (if the target image data acquisition unit 4 acquires the location information Sa associated with the target image data Da, the target image data Da and the location). By inputting the information Sa), an inference result Ri of the cause of the blowout of the target heat exchanger tube is output. The above-mentioned inference result Ri outputted by the learning model Md1 includes a plurality of candidates for the cause of the blowout of the target heat exchanger tube and a probability of each of the plurality of candidates.

The reference information storage unit 8 stores reference information regarding past cases of multiple causes of heat exchanger tube blowouts (for example, accident reports of plants such as boilers including heat exchanger tubes).

The reference information search unit 9 stores reference information Ir regarding past cases regarding the eruption cause of the candidate with the highest probability among the plurality of candidates included in the inference result Ri output from the learning model Md1 into the reference information storage unit 8. Search and obtain reference information stored in . The reference information Ir acquired from the reference information storage unit 8 by the reference information search unit 9 includes not only the candidate with the highest probability among the plurality of candidates included in the inference result Ri, but also the candidate with the second or subsequent highest probability. may include reference information regarding past cases of causes of eruptions.

The screen data output unit 10 is configured to output screen data Ds of a display screen to be displayed on the user's terminal 14. The screen data Ds output by the screen data output unit 10 includes the inference result Ri output by the learning model Md1 and the reference information Ir acquired by the reference information search unit 9. The screen data Ds output by the screen data output section 10 is transmitted to the user's terminal 14 via the information communication network Nc, and displayed on the display section 14d of the terminal 14.

The learning image data generation unit 12 is configured to generate the above-mentioned learning image data Dfl from image data Db such as an accident report created in the past. This image data Db may be, for example, a PDF file generated by scanning an accident report, etc., and may include an image portion and a non-image portion (non-image portion such as a background). good. The learning image data generation unit 12 generates learning image data Dfl by inferring and extracting (cutting) an image portion from the image data Db using the learning model Md0.

Thereby, a lot of learning image data Dfl to be used as the above-mentioned teacher data Dt can be generated from a lot of image data Db such as accident reports created in the past. Note that the learning image data generation unit 12 generates the learning model Md0 by performing machine learning using teacher data D0 that indicates an image portion and a portion other than the image portion in the image data. Details of how the learning image data generation unit 12 generates the learning image data Dfl will be described later.

FIG. 3 is a diagram showing an example of a flow of inference and display of the cause of a heat exchanger tube blowout using the heat exchanger tube blowout cause inference device 2 described above.

As shown in FIG. 3, in S101, the target image data acquisition unit 4 obtains target image data Da, which is image data indicating the blowout part of the target heat exchanger tube, for the target heat exchanger tube from which the cause of the blowout is to be inferred. get. The target image data acquisition unit 4 acquires target image data Da from the user's terminal 14 via the information communication network Nc, for example.

In S102, the learning model Md1 of the machine learning unit 6 receives the target image data Da acquired by the target image data acquisition unit 4 and outputs an inference result Ri of the cause of the blowout of the target heat exchanger tube.

In S103, the reference information search unit 9 retrieves reference information regarding past cases of the most likely candidate for the cause of a blowout among the plurality of candidates for the cause of a blowout included in the inference result Ri output from the learning model Md1. Ir is retrieved from the reference information storage unit 8 and obtained.

In S104, the screen data output unit 10 outputs screen data Ds including the inference result Ri output from the learning model Md1 and the reference information Ir acquired by the reference information search unit 9. The screen data Ds output by the screen data output section 10 is transmitted to the user's terminal 14 via the network Nc, and displayed on the display section 14d (for example, a display) of the terminal 14.

FIG. 4 is a diagram illustrating an example of a flow in which the learning image data generation unit 12 generates the learning image data Dfl.

As shown in FIG. 4, in S201, the learning image data generation unit 12 acquires image data Db (see, for example, FIG. 5) such as an accident report created in the past. This image data Db includes an image portion and a non-image portion (non-image portion such as a background). In S202, the learning image data generation unit 12 uses the learning model Md0 to infer an image portion (feature portion) in which the feature amount in the image data Db acquired in S201 satisfies a predetermined condition. For example, the learning image data generation unit 12 may perform inference on the image portion for each pixel in the image data Db using the brightness and color of the target pixel that is the inference target and pixels around the target pixel as feature quantities. . In this case, if the target pixel and many surrounding pixels have the same color continuously, those pixels are judged to be background (that is, they are not part of the image), and many pixels around the target pixel and surrounding pixels are judged as background. If colors are included, those pixels may be determined as part of the image. For example, when the image data Db shown in FIG. 5 is acquired in S201, the learning model Md0 infers the image portion shown by the black part in FIG. 6 in S202. In S203, the learning image data generation unit 12 calculates the area of a continuous image portion among the image portions inferred in S202.

In S204, the learning image data generation unit 12 determines whether the area calculated in S203 is greater than or equal to a predetermined threshold in order to efficiently generate learning image data having a relatively large image area. do. In S204, if the area calculated in S203 is equal to or greater than the threshold, in S205, a portion of the set G of continuous image portions in the image data Db acquired in S201 is extracted by a rectangle (rectangle or square). (cutting out) to generate learning image data Dfl. For example, when the image portion (black portion) shown in FIG. 6 is inferred from the image data Db shown in FIG. 5 in S202, in S205, the minimum X coordinate Xmin and the maximum The inner part of the rectangle K defined by the coordinate Xmax, the minimum Y coordinate Ymin, and the maximum Y coordinate Ymax (for example, the part corresponding to the range of the lower left quadrangle K defined by Xmin, Xmax, Ymin, Ymax in FIG. 7) ) is extracted as an image to generate learning image data Dfl. In the example shown in FIG. 7, the horizontal axis is the X axis and the vertical axis is the Y axis, the right side of the figure shows the positive direction of the X axis, and the bottom side of the figure shows the positive direction of the Y axis.

If it is determined in S204 that the area calculated in S203 is not equal to or greater than the threshold, or if S205 is executed, it is determined in S206 whether or not there is another image part that was inferred in S202, and other inferences are made in S206. If there is an image portion, the process returns to S203 and S203 to S205 are executed for that image portion. In S206, if there is no other image portion inferred in S202, the generation of the learning image data Dfl using the image data Db is ended. In the example shown in FIG. 7, the learning image data generation unit 12 generates three pieces of learning image data Dfl by extracting three image parts corresponding to the ranges of three rectangles K from the image data Db. ing.

Next, FIG. 8 shows an example of the display screen E1 of the display unit 14d of the user's terminal 14 when the above-mentioned target image data Da etc. are input from the user's terminal 14 to the heat exchanger tube blowout cause inference device 2. I will explain using The display screen E1 shown in FIG. 8 is displayed on the display section 14d of the terminal 14 based on the output of the screen data output section 10.

As shown in FIG. 8, the display screen E1 displays a file name input field F1 for identifying the file of image data including the blown part of the target heat exchanger tube, and image data input in the file name input field F1. a frame F3 for cutting out a part of the image data displayed on the image data display section F2 and extracting it as target image data Da; A size adjustment button F4 for adjusting the size, a frame position adjustment button F5 for adjusting the vertical and horizontal positions of the frame F3, and the target image data Da cut out by the frame F3 as shown in the target image data Da. It includes an inference button F6 for inferring the cause of the blowout of the target heat exchanger tube, and an information update button F0 for updating information on the display screen E1.

The display screen E1 also includes a numerical value input field F7 for inputting numerical values indicating the X coordinate, Y coordinate, width, and height of the frame F3. Here, the X coordinate and Y coordinate of the frame F3 are the X coordinate and Y coordinate of a specific position of the frame F3, and for example, the minimum value of the X coordinate of the frame F3 and the minimum value of the Y coordinate of the frame F3 (for example, in FIG. (X and Y coordinates of the upper left corner of frame F3). Note that the upper left corner of frame F3 in FIG. 8 means the corner closest to the characters "search image" among the four corners of frame F3. The display screen E1 also includes a location input field F8 for inputting the location of the target heat exchanger tube that has blown out (for example, a boiler superheater or reheater, etc.), and identification information that identifies the target heat exchanger tube that has blown out (for example, the tube a tube input field F9 for inputting a number, etc.). The location and identification information of the target heat exchanger tube corresponds to the location information Sa described above.

FIG. 9 is a diagram showing an example of the display screen E2 that is output to the display section 14d of the terminal 14 when the inference button F6 in FIG. 8 is pressed.

As shown in FIG. 9, the display screen E2 includes the target image data Da cut out by the frame F3 on the display screen E1, the inference result Ri output from the learning model Md1, and the information obtained by the reference information search unit 9. Includes reference information Ir.

The inference result Ri is based on the target image data Da cut out by the frame F3 on the display screen E1, the location of the target heat exchanger tube inputted in the location input field F8, and the identification of the target heat exchanger tube inputted in the tube input field F9. The information is input to the learning model Md1, and is output from the learning model Md1.

In the illustrated example, the inference result Ri includes a plurality of candidates for the cause of the blowout of the target heat exchanger tube (erosion, creep fracture, sulfuric acid dew point corrosion, corrosion fatigue, and a total of others) and a probability of each of the plurality of candidates. In addition, the reference information Ir is used to access reference materials related to past cases of the most likely cause of eruption (erosion in the example shown) among the multiple candidates for the cause of eruption included in the inference result Ri. link F10 (for example, information indicating the location of a document such as an accident report). Note that the reference information Ir displayed on the display screen E2 is for accessing reference materials related to past cases regarding the cause of eruption of the candidate with the second or higher probability among the multiple candidates included in the inference result Ri. It may further include a link.

FIG. 10 is a diagram showing another example of the display screen E2 outputted to the display section 14d of the terminal 14 by pressing the inference button F6 in FIG.

The display screen E2 shown in FIG. 10 differs from the display screen E2 shown in FIG. 9 only in the content of reference information Ir, and is otherwise the same. In the display screen E2 shown in FIG. 10, the reference information Ir includes past cases of the most probable blowout cause (erosion in the illustrated example) among the plurality of blowout cause candidates included in the inference result Ri. includes an image F11 of a reference document regarding (for example, an image of a document such as an accident report). Note that the reference information Ir displayed on the display screen E2 is an image of a reference document regarding a past case regarding the cause of an eruption of the candidate with the second or higher probability among the plurality of candidates included in the inference result Ri. May contain.

(Effect of heat exchanger tube blowout cause inference device 2)
Next, the effects of the heat exchanger tube blowout cause inference device 2 will be explained.
According to the heat exchanger tube blowout cause inference device 2, each of the plurality of learning image data Dfl each indicating the blowout part of the heat exchanger tube whose cause of the blowout has been identified in the past, and the plurality of learning image data Dfl. By performing machine learning using training data Dt including associated blowout causes Cd, past knowledge about the relationship between image data of heat exchanger tube blowout parts and heat exchanger tube blowout causes can be accumulated. The machine learning unit 6 can generate a learning model Md1 that reflects the above. Therefore, by inputting the target image data Da into the generated learning model Md1, it is possible to accurately infer the cause of the blowout of the target heat exchanger tube indicated by the target image data Da. This makes it possible to quickly take measures against blowouts of heat exchanger tubes.

Further, the teacher data Dt includes location information St that is associated with each of the plurality of learning image data Dfl and indicates the location of the heat exchanger tube shown in the learning image data Dfl. For this purpose, the machine learning unit 6 generates a learning model Md1 that reflects the image data showing the blowout part of the heat exchanger tube and the accumulation of past knowledge about the relationship between the location of the heat exchanger tube and the cause of the blowout of the heat exchanger tube. I can do it. Thereby, it is possible to improve the accuracy of inferring the cause of the blowout of the target heat exchanger tube.

Further, the inference result Ri outputted by the learning model Md1 includes a plurality of candidates for the cause of the blowout of the target heat exchanger tube and a probability of each of the plurality of candidates. Therefore, it is possible to efficiently deal with the blowout of the heat transfer tube by considering the probability of the candidate for the cause of the blowout.

In addition, the screen data Ds output by the screen data output unit 10 includes the inference result Ri output by the learning model Md1 and the past information about the candidate with the highest probability among the plurality of candidates for the cause of the eruption included in the inference result Ri. Includes reference information Ir regarding the case. Therefore, it is possible to efficiently deal with the blowout of the heat exchanger tube by considering the reference information Ir regarding past cases regarding the cause of the blowout of the candidate with the highest probability among the plurality of candidates.

Furthermore, in the example shown in FIG. 9, the reference information Ir is for accessing reference materials related to past cases of the most probable cause of eruption among the multiple candidates for the cause of eruption included in the inference result Ri. The link F10 is included. Therefore, when displaying the screen data Ds including the inference result Ri and the reference information Ir output by the learning model Md1 on the user side terminal 14, the amount of information in the screen data Ds is suppressed from becoming excessively large. It becomes possible to easily grasp the information necessary to deal with a blowout of a heat exchanger tube.

Further, in the example shown in FIG. 10, the reference information Ir is a reference regarding past cases of the cause of eruption of the candidate with the highest probability among the plurality of candidates for the cause of eruption included in the inference result Ri. Since the image F11 of the document is included, when the screen data Ds including the inference result Ri and the reference information Ir output by the learning model Md1 is displayed on the user terminal 14, the content of the past case is displayed using the image F11 of the reference document. It becomes possible to understand easily.

In addition, as shown in FIG. 7 etc., among the image data Db, it is defined by the minimum X coordinate Xmin, the maximum X coordinate Xmax, the minimum Y coordinate Ymin, and the maximum Y coordinate Ymax in the range occupied by the set G. By extracting the inner part of the rectangle K (the part corresponding to the range of the rectangle K) as an image, the learning image data Dfl can be generated. For this reason, for example, when there are one or more image regions (multiple sets G of feature parts) in the image data Db, the parts corresponding to each image region can be appropriately extracted from the image data Db and used for learning. The image data Dfl can be efficiently generated, and the accuracy of inference of the cause of a blowout in the learning model Md1 can be improved.

The present disclosure is not limited to the embodiments described above, and also includes forms in which modifications are added to the embodiments described above, and forms in which these forms are appropriately combined.

For example, in the embodiment described above, the cause of heat exchanger tube blowout is estimated to be the cause of heat exchanger tube blowout (a state in which a through hole is formed in the heat exchanger tube that communicates the inside and outside of the heat exchanger tube), which is an example of damage to the heat exchanger tube. Although an inference device is shown, the present disclosure is applicable to a heat exchanger tube damage cause inference device that infers the cause of damage to a heat exchanger tube (for example, scratches on the surface of a heat exchanger tube, etc.), not just a blowout of a heat exchanger tube. The configuration, functions, effects, etc. of the heat exchanger tube damage cause inference device 2 can be understood by replacing “blowout” with “damage” in the description of the heat exchanger tube blowout cause inference device 2 described above.

The contents described in each of the above embodiments can be understood as follows, for example.

(1) A heat exchanger tube damage cause inference device (for example, the above-mentioned heat exchanger tube blowout cause inference device 2) according to at least one embodiment of the present disclosure,
A plurality of learning image data (for example, the plurality of learning image data Dfl described above) each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and damage associated with each of the plurality of learning image data. A learning model for inferring the cause of heat exchanger tube damage from image data is created by performing machine learning using the cause (for example, the blowout cause Cd described above) and training data (for example, the training data Dt described above) including the cause (for example, the blowout cause Cd described above). For example, a machine learning section (for example, the above-mentioned machine learning section 6) configured to generate the above-mentioned learning model Md1),
Target image data acquisition configured to acquire target image data (for example, the above-mentioned target image data Da) that is image data indicating a damaged part of the target heat exchanger tube for the target heat exchanger tube from which the cause of damage is to be inferred. unit (for example, the above-mentioned target image data acquisition unit 4),
Equipped with
The learning model is configured to output an inference result of the cause of damage to the target heat exchanger tube (for example, the above-mentioned inference result Ri) by receiving the target image data acquired by the target image data acquisition unit. Ru.

According to the heat exchanger tube damage cause inference device described in (1) above, each of the plurality of learning image data each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and the plurality of learning image data. A learning model that reflects the accumulation of past knowledge about the relationship between image data of damaged parts of heat exchanger tubes and causes of damage to heat exchanger tubes is created by performing machine learning using training data that includes associated damage causes. can be generated by the machine learning unit. Therefore, by inputting the target image data into the generated learning model, it is possible to accurately infer the cause of damage to the target heat exchanger tube indicated by the target image data. This makes it possible to quickly deal with damage to the heat exchanger tubes.

(2) In some embodiments, in the heat exchanger tube damage cause inference device described in (1) above,
The teacher data includes location information associated with each of the plurality of learning image data (for example, the location information St described above, which indicates the location of the heat exchanger tube shown in the learning image data).

According to the heat exchanger tube damage cause inference device described in (2) above, the machine learning unit creates a learning model that reflects the image data of the damaged part, the accumulation of past knowledge about the relationship between the location of the damaged part, and the cause of damage. can be generated. Thereby, it is possible to improve the accuracy of inferring the cause of damage to the target heat exchanger tube.

(3) In some embodiments, in the heat exchanger tube damage cause inference device described in (1) or (2) above,
The inference result output by the learning model includes a plurality of candidates for the cause of damage to the target heat exchanger tube and a probability of each of the plurality of candidates.

According to the heat exchanger tube damage cause inference device described in (3) above, it is possible to efficiently deal with damage to the heat exchanger tubes by considering the probability of damage cause candidates.

(4) In some embodiments, in the heat exchanger tube damage cause inference device described in (3) above,
comprising a screen data output unit (for example, the above-mentioned screen data output unit 10) that outputs screen data of a display screen to be displayed on a user-side display terminal;
The screen data output by the screen data output unit is related to the inference result output by the learning model and past cases regarding the cause of damage of the candidate with the highest probability among the plurality of candidates included in the inference result. Reference information (for example, the above-mentioned reference information Ir).

According to the heat exchanger tube damage cause inference device described in (4) above, the damage cause of heat exchanger tubes is determined by considering reference information regarding past cases regarding the damage cause of the candidate with the highest probability among multiple candidates. It is possible to deal with it efficiently.

(5) In some embodiments, in the heat exchanger tube damage cause inference device described in (4) above,
The reference information includes a link (for example, the above-mentioned link F10) for accessing reference material regarding the past case.

According to the heat exchanger tube damage cause inference device described in (5) above, when displaying the screen data including the inference results and reference information output by the learning model on the user's terminal, the amount of information in the screen data is excessive. This makes it possible to easily grasp the information necessary to deal with damage to heat exchanger tubes.

(6) In some embodiments, in the heat exchanger tube damage cause inference device described in (4) or (5) above,
The reference information includes an image of a reference document regarding the past case (for example, the above-mentioned image F11).

According to the heat exchanger tube damage cause inference device described in (6) above, when displaying screen data including the inference results and reference information output by the learning model on the user terminal, reference documents related to past cases are displayed. Images make it possible to easily understand the contents of past cases.

(7) In some embodiments, in the heat exchanger tube damage cause inference device according to any one of (1) to (6) above,
further comprising a learning image data generation unit (for example, the above-mentioned learning image data generation unit 12) for generating the learning image data from image data including an image portion and a portion other than the image portion,
The learning image data generation unit infers a set of feature portions of the image data whose feature amounts satisfy a predetermined condition (for example, the above-mentioned set G), and the The learning image data is generated by extracting as an image the inner part of a rectangle defined by the X coordinate, the maximum X coordinate, the minimum Y coordinate, and the maximum Y coordinate.

According to the heat exchanger tube damage cause inference device described in (7) above, for example, when there is one or more image regions in one original image data, the method corresponding to each image region from the original image data It is possible to efficiently generate training image data by appropriately extracting parts, and improve the accuracy of inferring the cause of damage in the learning model.

(8) A heat exchanger tube damage cause inference method according to an embodiment of the present disclosure includes:
A target image data acquisition step of acquiring target image data (for example, the above-mentioned target image data Da) that is image data indicating a damaged part of the target heat exchanger tube for the target heat exchanger tube whose damage cause is to be inferred;
A plurality of learning image data (for example, the plurality of learning image data Dfl described above) each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and damage associated with each of the plurality of learning image data. Regarding the learning model used to infer the cause of damage to heat exchanger tubes, which is generated by machine learning using training data (for example, training data Dt described above) including a cause (for example, the above-mentioned blowout cause Cd), the above-mentioned target image inputting the target image data acquired in the data acquisition step into the learning model, so that the learning model outputs an inference result (for example, the above-mentioned inference result Ri) of the cause of damage to the target heat exchanger tube;
Equipped with

According to the heat exchanger tube damage cause inference method described in (8) above, each of the plurality of learning image data each showing the damaged part of the heat exchanger tube whose cause of damage has been identified in the past, and the plurality of learning image data Accumulation of past knowledge about the relationship between image data of damaged parts of heat exchanger tubes and causes of damage to heat exchanger tubes using a learning model generated by machine learning using training data including associated damage causes. By inputting the target image data into a learning model that reflects the above, it is possible to accurately infer the cause of damage to the target heat exchanger tube indicated by the target image data. This makes it possible to quickly deal with damage to the heat exchanger tubes.

2 Heat exchanger tube blowout cause inference device 4 Target image data acquisition unit 6 Machine learning unit 8 Reference information storage unit 9 Reference information search unit 10 Screen data output unit 12 Learning image data generation unit 14 Terminal 14d Display unit 72 Processor 74 RAM
76 ROM
78 HDD
80 Input I/F
82 Output I/F
84 Bus Cd Blowout cause Da Target image data Db Image data Dfl Learning image data Ds Screen data Dt Teacher data E1, E2 Display screen F1 File name input field F2 Image data display section F3 Frame F4 Size adjustment button F5 Frame position adjustment button F6 Inference button F7 Numerical input field F8 Location input field F9 Tube input field F10 Link F11 Image G Set Ir Reference information K Rectangle Md0, Md1 Learning model Nc Information and communication network Ri Inference result Sa, St Location information

Claims (7)

A heat exchanger tube damage cause inference device for inferring the cause of damage to heat exchanger tubes used in a heat exchanger of a boiler including a superheater and a reheater,
Machine learning using training data that includes a plurality of learning image data each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and a damage cause associated with each of the plurality of learning image data. a machine learning unit configured to generate a learning model used to infer the cause of damage to the heat exchanger tube by performing the following steps;
a target image data acquisition unit configured to acquire target image data that is image data indicating a damaged portion of the target heat exchanger tube for the target heat exchanger tube from which the cause of the damage is to be inferred;
Equipped with
The learning model is configured to output an inference result of the cause of damage to the target heat exchanger tube by receiving the target image data acquired by the target image data acquisition unit,
The teacher data is location information associated with each of the plurality of learning image data , and includes location information indicating the type of the heat exchanger in which the heat exchanger tube shown in the learning image data is installed. ,
The target image data acquisition unit is configured to acquire target location information indicating a type of the heat exchanger in which the target heat exchanger tube indicated by the target image data is installed.
Heat exchanger tube damage cause inference device. The heat exchanger tube damage cause inference device according to claim 1, wherein the inference result output by the learning model includes a plurality of candidates for the cause of damage to the target heat exchanger tube and a probability of each of the plurality of candidates. The screen data output unit includes a screen data output unit that outputs screen data of a display screen to be displayed on a user-side terminal, and the screen data output by the screen data output unit is based on the inference result output by the learning model and the inference result. 3. The heat exchanger tube damage cause inference device according to claim 2, further comprising reference information regarding past cases regarding the damage cause of the candidate with the highest probability among the plurality of candidates included. The heat exchanger tube damage cause inference device according to claim 3, wherein the reference information includes a link for accessing reference materials regarding the past cases. The heat exchanger tube damage cause inference device according to claim 3 or 4, wherein the reference information includes an image of a reference document regarding the past case. further comprising a learning image data generation unit for generating the learning image data from image data including an image portion and a portion other than the image portion,
The learning image data generation unit infers a set of feature parts whose feature amounts satisfy a predetermined condition among the image data, and calculates a minimum X coordinate and a maximum X coordinate in a range occupied by the set of the image data. Any one of claims 1 to 5, wherein the learning image data is generated by extracting as an image an inner part of a rectangle defined by the coordinates, the minimum Y coordinate, and the maximum Y coordinate. The heat exchanger tube damage cause inference device according to item 1. A heat exchanger tube damage cause inference method for inferring the cause of damage to heat exchanger tubes used in a heat exchanger of a boiler including a superheater and a reheater, the method comprising:
a target image data acquisition step of acquiring target image data that is image data indicating a damaged part of the target heat exchanger tube for the target heat exchanger tube from which the cause of damage is to be inferred;
Machine learning using training data including a plurality of learning image data each showing a damaged part of a heat exchanger tube whose cause of damage has been identified in the past, and a cause of damage associated with each of the plurality of learning image data. By inputting the target image data acquired in the target image data acquisition step into the learning model, which is used to infer the cause of damage to the heat exchanger tube, the learning model generated by a step in which the learning model outputs a result;
Equipped with
The teacher data is location information associated with each of the plurality of learning image data , and includes location information indicating the type of the heat exchanger in which the heat exchanger tube shown in the learning image data is installed. ,
In the target image data acquisition step, acquiring target location information indicating the type of the heat exchanger in which the target heat exchanger tube indicated by the target image data is installed;
Method for inferring the cause of heat exchanger tube damage.

Images