Machine learning device, machine learning method, and device for determining the presence or absence of anomalies in wells

Abstract

To provide a machine learning device, a machine learning method, and a well abnormality determination device to be used for diagnosis of a state of deterioration in a well conducted in such a way that results are not user dependent.SOLUTION: A machine learning device 1 is provided, comprising an image acquisition unity 4 for acquiring images 3 of the inside of a well, an evaluation value acquisition unit 6 for acquiring evaluation values 5 regarding presence / absence of an abnormality in the images 3 of the inside of the well, and a learning unit 7 configured to build a learning model for determining the presence or absence of an abnormality inside the well by performing supervised learning using pairs of the images 3 of the inside of the well acquired by the image acquisition unit 4 and the evaluation values 5 acquired by the evaluation value acquisition unit 6 as teacher data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a machine learning device, a machine learning method, and a well abnormality determination device for use in diagnosing the deterioration status and the like inside a well.

Background Art

[0002] Conventionally, for example, the inside of a well has been photographed using a well camera device as disclosed in Patent Document 1, and the deterioration status and the like have been diagnosed to propose a repair method.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, since the conventional diagnosis largely relies on the determination based on the experience of a skilled technician who is the user, differences in the quality of the product are likely to occur depending on the determiner, and problems such as the training of young technicians have arisen.

[0005] The present invention has been made in view of such circumstances, and provides a machine learning device, a machine learning method, and a well abnormality determination device for use in diagnosing the deterioration status and the like inside a well so that no differences occur due to the user.

Means for Solving the Problems

[0006] In order to solve such problems, the present invention provides a machine learning device including an image acquisition unit that acquires an image inside a well, an evaluation value acquisition unit that acquires an evaluation value regarding the presence or absence of an abnormality in the image inside the well, and a learning unit that constructs a learning model for determining the presence or absence of an abnormality inside the well by performing supervised learning using a set of the image inside the well acquired by the image acquisition unit and the evaluation value acquired by the evaluation value acquisition unit as teacher data.

[0007] In the machine learning device, the image inside the well may be classified into an image of a screen portion including a water intake hole and an image of a casing portion not including the water intake hole, and the evaluation value may be classified into normal, blocked, and damaged for the image of the screen portion, and classified into normal, adhered, and damaged for the image of the casing portion.

[0008] In the machine learning device, the evaluation value may be determined based on a user's visual judgment.

[0009] Further, the present invention provides a machine learning method performed by a machine learning device, including an image acquisition step of acquiring an image inside a well, an evaluation value acquisition step of acquiring an evaluation value regarding the presence or absence of an abnormality in the image inside the well, and a learning step of constructing a learning model for determining the presence or absence of an abnormality inside the well by performing supervised learning using a set of the image inside the well acquired in the image acquisition step and the evaluation value acquired in the evaluation value acquisition step as teacher data.

[0010] In the machine learning method, the image inside the well may be classified into an image of a screen portion including a water intake hole and an image of a casing portion not including the water intake hole, and the evaluation value may be classified into normal, blocked, and damaged for the image of the screen portion, and classified into normal, adhered, and damaged for the image of the casing portion.

[0011] In the machine learning method, the evaluation value may be determined based on a user's visual judgment.

[0012] The present invention also provides an abnormality determination device for a well, which includes an imaging unit that captures an image inside the well, and determines the presence or absence of an abnormality inside the well by using a learning model constructed by the machine learning device for the image inside the well captured by the imaging unit.

Effects of the Invention

[0013] According to the machine learning device, the machine learning method, and the abnormality determination device for a well of the present invention, it is possible to propose a more appropriate well maintenance management method by detecting early signs of abnormalities inside the well that cannot be confirmed by the user's visual inspection, and to extend the life of the well.

[0014] In addition, it becomes possible to create a work report at the site, eliminating the need to reproduce the video at the office again, and reducing the labor and time required for creating the work report. This leads to a reduction in the burden on the user and a reform of the working style, and also makes it possible to transfer the skills of skilled technicians to novice technicians.

[0015] Furthermore, appropriate maintenance management of wells is also significant in aiming for SDGs (such as "3. Ensure healthy lives and promote well-being for all at all ages", "6. Ensure availability and sustainable management of water and sanitation for all", "11. Make cities and human settlements inclusive, safe, resilient and sustainable", "12. Ensure sustainable consumption and production patterns", etc.).

Brief Description of the Drawings

[0016]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Best Mode for Carrying Out the Invention

[0017] Hereinafter, with reference to the drawings, preferred embodiments of the machine learning device, machine learning method, and well abnormality determination device of the present invention will be described. However, the present invention is not limited to the following description, and various modifications and changes can be made by those skilled in the art based on the gist of the invention described in the claims or disclosed in the best mode for carrying out the invention. Such modifications and changes are also included in the scope of the present invention.

[0018] FIG. 1 is a block diagram showing a schematic configuration of a machine learning device 1 and a well abnormality determination device 2 according to a preferred embodiment of the present invention. The machine learning device 1 includes an image acquisition unit 4 that acquires an image 3 inside the well, an evaluation value acquisition unit 6 that acquires an evaluation value 5 regarding the presence or absence of an abnormality in the image 3 inside the well, and a learning unit 7 that constructs a learning model for determining the presence or absence of an abnormality inside the well by performing supervised learning using a set of the image 3 inside the well acquired by the image acquisition unit 4 and the evaluation value 5 acquired by the evaluation value acquisition unit 6 as teacher data.

[0019] The learning model constructed by the learning unit 7 is stored in the learning model storage unit 8. When new teacher data is acquired after constructing the learning model, supervised learning may be additionally performed on the learning model stored in the learning model storage unit 8, and the once-constructed learning model may be updated to store a new learning model in the learning model storage unit 8.

[0020] The well abnormality determination device 2 includes a photographing unit 9 that photographs an image 3 inside the well, and determines the presence or absence of an abnormality inside the well using the learning model constructed by the machine learning device 1 for the image 3 inside the well photographed by the photographing unit 9. As the photographing unit 9, for example, a well camera device as disclosed in Patent Document 1 can be used.

[0021] The well abnormality determination device 2 also includes an image conversion unit 10, a depth determination unit 11, an image determination unit 12, an abnormality determination unit 13, a storage unit 14, an end determination unit 15, and a display unit 16.

[0022] The image conversion unit 10 of the well abnormality determination device 2 reads video frames from the video data captured by the imaging unit 9 and converts them into images 3 of the inside of the well as still images. Alternatively, the image conversion unit 10 acquires a still image of the image 3 inside the well.

[0023] The depth determination unit 11 of the well abnormality determination device 2 determines whether the depth is the target depth for image determination based on the depth information measured by depth measurement means (not shown).

[0024] The image determination unit 12 of the well abnormality determination device 2 determines the classification of the image 3 inside the well.

[0025] The abnormality determination unit 13 of the well abnormality determination device 2 determines the presence or absence of an abnormality in the image 3 inside the well based on the constructed learning model.

[0026] The storage unit 14 of the well abnormality determination device 2 stores and saves the determination result of the presence or absence of an abnormality in the well.

[0027] The end determination unit 15 of the well abnormality determination device 2 determines whether to end the determination.

[0028] The display unit 16 of the well abnormality determination device 2 is, for example, a display or the like, and is for displaying the determination result.

[0029] FIG. 2 is a partial perspective view showing a configuration example of the tubular body 17 of the well. The tubular body 17 has a plurality of water intake holes 18 for taking in groundwater. On the inner wall surface of the tubular body 17, there are a screen portion which is a portion including the water intake holes 18 and a casing portion which is a portion not including the water intake holes 18. In the screen portion where the water intake holes 18 are present, problems such as blockage or damage due to clogging of the water intake holes 18 particularly become issues, and in the casing portion where the water intake holes 18 are not present, problems such as the presence or absence of scale adhesion to the tubular body 17 and damage particularly become issues. Therefore, when determining the presence or absence of abnormalities by an image, it is necessary to focus on different feature points depending on whether the water intake holes 18 are present or not, and the same judgment criteria cannot be used. Thus, the image 3 in the well is classified, for example, into an image of the screen portion including the water intake holes 18 and an image of the casing portion not including the water intake holes 18.

[0030] The evaluation value 5 used for machine learning is determined based on the visual judgment of a user such as an engineer. The evaluation value 5 is classified, for example, as normal, blocked, and damaged for the image of the screen portion, and as normal, adhered, and damaged for the image of the casing portion. As an example, FIG. 3 shows a specific example of the classification of the evaluation value 5.

[0031] For example, for the image of the screen portion, when there is no blockage or damage in the water intake holes, it is classified as "normal", when the water intake holes are clogged and prevent water intake, it is classified as "blocked", and when the water intake holes are damaged and there is concern about the intrusion of filled gravel or formation particles, it is classified as "damaged".

[0032] For example, for the image of the casing portion, when there is no scale adhesion or damage to the tubular body, it is classified as "normal", when scale is adhered to the tubular body, it is classified as "adhered", and when the tubular body is damaged and there is concern about the intrusion of filled gravel or formation particles, it is classified as "damaged".

[0033] As described above, there is a correlation useful as teacher data for supervised learning between the image 3 inside the well and the evaluation value 5. Based on such a correlation, by determining which evaluation value 5 to classify the image 3 inside the well into, it is possible to determine the presence or absence of abnormalities inside the well. Therefore, by performing machine learning on such a correlation using a large amount of data with the machine learning apparatus 1 and the machine learning method of the present embodiment, it is possible to realize the automation of accurate determination by the well abnormality determination apparatus 2.

[0034] Note that the tubular body may be a slit-type screen or a wound wire-type screen in addition to the round hole wound wire-type screen shown in FIGS. 2 and 3, and the evaluation value 5 can be similarly classified for each of the portion including the groundwater intake portion and the portion not including it.

[0035] FIG. 4 is a flowchart showing the flow of construction of a learning model according to a preferred embodiment of the present invention.

[0036] First, in step S11, the image acquisition unit 4 acquires the image 3 inside the well. The image acquisition unit 4 outputs the image 3 inside the well to the learning unit 7.

[0037] Next, in step S12, the evaluation value acquisition unit 6 acquires the evaluation value 5 associated with the image 3 inside each well. The evaluation value acquisition unit 6 outputs the evaluation value 5 in association with the image 3 inside each well to the learning unit 7. Note that the two steps of step S11 and step S12 may be performed with step S12 first or simultaneously in parallel.

[0038] Next, in step S13, the learning unit 7 generates teacher data with each data of the image 3 inside the well and the corresponding evaluation value 5 as a set. The image 3 inside the well is classified, for example, into an image of the screen portion and an image of the casing portion, and teacher data is generated.

[0039] Next, in step S14, the learning unit 7 performs machine learning based on the teacher data generated in step S13. This machine learning is supervised learning.

[0040] Next, in step S15, the learning unit 7 determines whether to end the machine learning. The end determination condition can be, for example, that supervised learning has been performed a predetermined number of times or the like.

[0041] If the condition for ending the machine learning is not satisfied, the process returns to step S11 and the machine learning is repeated. If the condition for ending the machine learning is satisfied, the process proceeds to step S16, and the constructed learning model is stored and saved in the learning model storage unit 8.

[0042] FIG. 5 is a flowchart showing the operation of the well abnormality determination device 2 according to a preferred embodiment of the present invention. In the present embodiment, the case where the imaging unit 9 of the well abnormality determination device 2 captures a moving image will be described, but it may capture a still image. In this case, instead of step S21 shown below, a step in which the image conversion unit 10 acquires a still image of the image 3 inside the well is performed.

[0043] First, in step S21, the image conversion unit 10 reads a video frame from the video data captured by the imaging unit 9 and converts it into a still image of the image 3 inside the well.

[0044] Next, in step S22, the depth determination unit 11 determines whether it is the target depth for image determination based on the depth information measured by a depth measurement means (not shown). The target depth can be arbitrarily set, and for example, it can be set to perform determination every 0.5 m. If it is the target depth, the process proceeds to the next step S23.

[0045] Next, in step S23, the image determination unit 12 determines whether the image 3 inside the well is an image of the screen part or an image of the casing part.

[0046] If it is determined in step S23 that it is an image of the screen part, in step S241, the abnormality determination unit 13 performs determination of normal, blocked, and damaged. The determination is performed based on the learning model constructed by the above method.

[0047] When it is determined in step S23 that the image is of the casing part, in step S242, the abnormality determination unit 13 determines whether it is normal, adhered, or damaged. The determination is made based on the learning model constructed by the above method.

[0048] Next, in step S25, the determination result in step S241 or S242 is stored and saved in the storage unit 14 of the well abnormality determination device 2.

[0049] Finally, in step S26, the end determination unit 15 determines whether to end the determination. If the end condition is not satisfied, it returns to step S21. If the end condition is satisfied, the determination ends. The end condition can be, for example, when all the analyses from the well opening to the well bottom are completed.

[0050] Machine learning can use a convolutional neural network (CNN: Convolutional Neural Network), which is a neural network suitable for learning targetting image data. In a CNN, it creates its own feature amounts based on data, so instead of the user designing the feature amounts, the machine learning device 1 itself acquires high-order feature amounts and classifies images based on them.

[0051] As described above, by using the machine learning device 1 and the machine learning method of this embodiment in which the learning model is constructed by supervised learning with the evaluation value 5 for the image 3 in each well as the teacher data, for any image 3 in the well, it becomes possible to automatically determine by the well abnormality determination device 2 into which of the evaluation values 5 it is classified.

[0052] The machine learning device 1 can be realized by installing an application software program in, for example, a personal computer or a server device. That is, the machine learning device 1 includes at least an arithmetic processing device such as a CPU (Central Processing Unit) and a storage device that stores an application software program. Since the amount of calculation increases in supervised learning, the machine learning device 1 may further include, for example, a GPU (Graphics Processing Unit).

[0053] The image conversion unit 10, depth determination unit 11, image determination unit 12, abnormality presence / absence determination unit 13, storage unit 14, and end determination unit 15 of the well abnormality presence / absence determination device 2 can be realized by installing an application software program in, for example, a personal computer or a server device. That is, the well abnormality presence / absence determination device 2 includes at least an arithmetic processing device such as a CPU (Central Processing Unit) and a storage device that stores an application software program. The well abnormality presence / absence determination device 2 determines the presence or absence of an abnormality in the well by the calculation of an application software program that takes as inputs the image 3 inside the well photographed by the photographing unit 9 and the learning model constructed by the machine learning device 1, and outputs and displays the determination result on the display unit 16 of the well abnormality presence / absence determination device 2. The user can confirm the determination result of the presence or absence of an abnormality in the well almost in real time while photographing the video inside the well, without the need for the conventional visual determination.

Explanation of Signs

[0054] 1 Machine learning device 2 Well abnormality presence / absence determination device 3 Image inside the well 4 Image acquisition unit 5 Evaluation value 6 Evaluation value acquisition unit 7 Learning unit 8 Learning model storage unit 9 Photographing unit 10 Image conversion unit 11 Depth determination unit 12 Image determination unit 13 Anomaly presence / absence determination unit 14 Memory unit 15 End determination unit 16 Display unit 17 Pipe body 18 Water intake hole

Claims

1. an image acquisition unit for acquiring an image of the inside of a well; an evaluation value acquisition unit that acquires an evaluation value regarding the presence or absence of an abnormality in the image of the well; a learning unit that performs supervised learning using a set of the image of the inside of the well acquired by the image acquisition unit and the evaluation value acquired by the evaluation value acquisition unit as training data to construct a learning model for determining the presence or absence of an abnormality in the well; Equipped with The images of the inside of the well are classified into an image of a screen portion including a water intake hole and an image of a casing portion not including a water intake hole; A machine learning device that classifies the evaluation values ​​into normal, blocked, and broken for the image of the screen portion, and into normal, stuck, and broken for the image of the casing portion.

2. The machine learning device according to claim 1 , wherein the evaluation value is determined based on a visual judgment of a user.

3. A machine learning method performed by a machine learning device, comprising: An image acquisition step of acquiring an image inside the well; An evaluation value acquisition step of acquiring an evaluation value regarding the presence or absence of an abnormality in the image of the well; A learning step of constructing a learning model for determining the presence or absence of an abnormality in the well by performing supervised learning using the image of the well acquired in the image acquisition step and the set of the evaluation value acquired in the evaluation value acquisition step as training data; Including, The images of the inside of the well are classified into an image of a screen portion including a water intake hole and an image of a casing portion not including a water intake hole; A machine learning method in which the evaluation values ​​are classified into normal, blocked and broken for the images of the screen portion, and normal, stuck and broken for the images of the casing portion.

4. The machine learning method according to claim 3 , wherein the evaluation value is determined based on a visual judgment of a user.

5. A photographing unit is provided for photographing the inside of the well, A well anomaly determination device that determines the presence or absence of anomalies in a well using a learning model constructed by the machine learning device described in claim 1 or 2 for images of the inside of the well captured by the imaging unit.

Images