JP2021143853A - Method for generating estimation model for perpendicular magnetism survey analysis, perpendicular magnetism survey analysis system, perpendicular magnetism survey analysis device, and computer-readable program - Google Patents

Abstract

To provide a method for generating an estimation model for perpendicular magnetism survey analysis, a perpendicular magnetism survey analysis system, a perpendicular magnetism survey analysis device, and a computer-readable program.SOLUTION: A method for generating an estimation model for perpendicular magnetism survey analysis for generating an estimation model (28) for estimating a buried state of a magnetic body under the ground from signal waveform of a detector for surveying perpendicular magnetism inserted into an observation hole includes an acquisition step of acquiring correlation between a buried state (r, M, θ) of a magnetic body under the ground, and an image (Im) of signal waveform generated by a detector for surveying perpendicular magnetism, and a learning step for allowing a convolution neural network using acquired teacher data to learn an estimation model and for storing a learned estimation model (28).SELECTED DRAWING: Figure 4

Description

The present invention relates to a vertical magnetic exploration in which a boring hole (observation hole) is excavated vertically downward from the ground surface of a land in which an unexploded bullet or the like is buried, and the peripheral magnetic field is explored by scanning the observation hole with a detector. More specifically, the present invention relates to a method for generating an estimation model for vertical magnetic exploration analysis used for vertical magnetic exploration analysis, a vertical magnetic exploration analysis system, a vertical magnetic exploration analysis device, and a computer-readable program.

In vertical magnetic exploration, a detector is inserted into a boring hole (observation hole) to detect magnetic anomalies in the surrounding area, and magnetic materials such as bombs, iron pipes, and sheet piles buried in the ground (hereinafter, simply "magnetic materials"). This is a method for investigating the distance and the amount of magnetism (see, for example, Patent Document 1 and the like). This method is effective when the burial depth of the magnetic material is large and cannot be detected by horizontal exploration, or when it is desired to accurately grasp the burial position of the magnetic material.

In this vertical magnetic exploration, a columnar detector attached to the tip of the cable (see, for example, Patent Document 2 and the like) is inserted into the observation hole, and the signal waveform output from the detector is recorded. Since this signal waveform can be recorded in real time on the measuring paper by, for example, a pen recorder, the technician at the site advances the excavation of the observation hole while confirming the safety based on the signal waveform, and goes to a deep position. Can be explored.

In recent years, it has become possible to record the signal waveform output from the detector as digital data on a mobile PC (PC: Personal Computer), display the waveform with dedicated software, and save it as a CAD diagram. became. Scene technician, after brought back mobile PC to the office, the dimensions of the characteristic waveform on the diagram CAD (maximum amplitude [Delta] H _max, period 2x ₀₎ reads, by fitting the predetermined formula, observation holes It is possible to accurately calculate the separation distance r from the magnetic material to the magnetic material and the magnetic charge M of the magnetic material.

Japanese Unexamined Patent Publication No. 11-183633 Japanese Unexamined Patent Publication No. 2014-126440

However, in geophysical exploration such as this magnetic exploration, analysis based on the judgment of engineers is the mainstream, and the credibility and accuracy of analysis results are currently left to the abilities of engineers. In the above-described vertical magnetic survey, although capable of storing signal waveforms as CAD drawings, technician dimension (maximum amplitude [Delta] H _max, periodic 2x _0, etc.) of the signal waveform on the screen of the PC since it takes a manual such measure, The credibility and accuracy of the analysis results were greatly influenced by the accuracy of manual work.

Therefore, the present invention provides a method for generating an estimation model for vertical magnetic exploration analysis, a vertical magnetic exploration analysis system, a vertical magnetic exploration analysis device, which can always obtain highly accurate analysis results without being influenced by the ability of engineers. And to provide computer readable programs.

In order to solve the above problem, the invention according to the present invention according to claim 1 is for estimating the buried state of the magnetic material based on the signal waveform of the detector for vertical magnetic exploration inserted in the observation hole in the ground. This is a method of generating an estimation model for vertical magnetic exploration analysis, which generates an estimation model for vertical magnetic exploration. For vertical magnetic exploration analysis, which includes an acquisition procedure to be acquired as teacher data and a learning procedure to train an estimated model by a convolutional neural network using the acquired teacher data and save the learned estimated model. A method for generating an estimation model is provided.

In order to solve the above-mentioned problems, the present invention according to claim 2 is the method for generating an estimation model for vertical magnetic exploration analysis according to claim 1. It is characterized in that the separation distance r to the body is included.

In order to solve the above problems, the present invention according to claim 3 further has the magnetic material in the embedded state in the method for generating the estimation model for vertical magnetic exploration analysis according to claim 2. It is characterized in that at least one of the magnetic charge M and the direction θ of the magnetic poles of the magnetic material as seen from the observation hole is included.

In order to solve the above problems, the present invention according to claim 4 changes the embedded state of the magnetic sample in the acquisition procedure in the method for generating an estimation model for vertical magnetic exploration analysis according to claim 3. It is characterized in that the teacher data is expanded by acquiring images of a plurality of signal waveforms and performing at least one of inversion processing, scaling processing, and noise processing on the images of the plurality of signal waveforms. ..

In order to solve the above problems, the present invention according to claim 5 is the method for generating an estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 4, wherein the detector is uniaxially differential. It is characterized by being a fluxgate type magnetic sensor.

In order to solve the above problem, in the present invention according to claim 6, the magnetic material is unexploded in the method for generating an estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 5. It is characterized by that.

In order to solve the above problems, the present invention according to claim 7 manages the estimation model generated by the method for generating the estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 6. By inputting the means and the image of the signal waveform generated by the detector for vertical magnetic exploration into the estimation model, the estimation means for estimating the buried state of the magnetic material in the ground and the buried state of the magnetic material to the user. Provided is a vertical magnetic tomography analysis system characterized by providing a presentation means for presentation.

In order to solve the above problems, the present invention according to claim 8 is the vertical magnetic tomography analysis system according to claim 7, wherein the estimation means transfers the signal waveform to a predetermined calculation formula that does not use the estimation model. The buried state of the magnetic material is calculated by fitting the dimensions, and the presenting means presents the estimated buried state and the calculated buried state to the user in a state in which they can be compared with each other.

In order to solve the above problems, the present invention according to claim 9 is characterized in that the vertical magnetic exploration analysis system according to claim 7 or 8 further includes a detector for vertical magnetic exploration.

The present invention according to claim 10 in order to solve the above problems is a vertical magnetic exploration analysis device applied to the vertical magnetic exploration analysis system according to any one of claims 7 to 9, and is the estimation. Provided is a vertical magnetic tomography analysis apparatus including the means and the presentation means.

To solve the above problems, the present invention according to claim 11 provides a computer-readable program, which comprises operating a computer as the vertical magnetic tomography analysis apparatus according to claim 10.

According to the method of generating the estimation model for vertical magnetic exploration analysis according to the present invention, the correspondence between the buried state of the magnetic material in the ground and the image of the signal waveform generated by the detector for vertical magnetic exploration is used as the teacher data. convolutional neural networks: since (CNN convolution neural network) by train the estimation model, a user (technician) dimension (amplitude [Delta] H, periodic 2x _0, etc. change) of the signal waveform on the screen of the PC manual work that measure I don't need it.

Here, the embedded state of the magnetic material is, for example, the separation distance r from the observation hole to the magnetic material, the magnetic amount M of the magnetic material, and the direction θ (elevation / depression angle) of the magnetic pole of the magnetic material as seen from the observation hole. It is a combination with.

Then, there is a strong correlation between the shape (pattern) of the signal waveform and the embedded state of the magnetic material. For example, the period 2x ₀ of changes in separation distance r is shorter signal waveform of the magnetic material becomes shorter, the magnetic charge M is too large signal waveform separation distance r is short magnetic of the magnetic amplitude ΔH also increased the It is already known (see formulas (1) and (2) described later). Furthermore, according to experiments by the inventors of the present application, the types of signal waveform shapes (patterns) are MV type, M type, V type, S type, and inverted S according to the direction θ of the magnetic poles of the magnetic material seen from the observation hole. It has also been found to vary between type, inverted V type, inverted M type, inverted MV type and the like.

Moreover, since the convolutional neural network is good at capturing the feature points of the image, it recognizes the characteristics of the shape (pattern) of the signal waveform and has a strong correlation with the shape (pattern) of the signal waveform. It is suitable for estimating the buried state.

Therefore, according to the method of generating the estimation model for vertical magnetic exploration analysis according to the present invention, high-performance estimation for estimating the embedded state of the magnetic material (for example, the combination of the separation distance r, the magnetic charge M, and the direction θ of the magnetic poles). The effect is that the model can be generated reliably.

Further, according to the vertical magnetic exploration analysis system, the vertical magnetic exploration analysis device, or the computer-readable program according to the present invention, the estimation model generated by the generation method of the estimation model for the vertical magnetic exploration analysis according to the present invention is used. Since the buried state of the magnetic material is estimated, it is possible to obtain an effect that highly accurate analysis results can always be obtained without being influenced by the ability of the user (engineer).

FIG. 1 is a schematic configuration diagram illustrating a vertical magnetic exploration system to which an example of the vertical magnetic exploration analysis system according to the present invention is applied. FIG. 2 is a flowchart illustrating the procedure of the vertical magnetic exploration method. FIG. 3 is an explanatory diagram illustrating input data and output data of the estimation model for vertical magnetic exploration analysis according to the present invention. FIG. 4 is an explanatory diagram illustrating a method of generating an estimation model for vertical magnetic exploration analysis according to the present invention. FIG. 5 is a flowchart illustrating a procedure of analysis by the vertical magnetic tomography analysis apparatus according to the present invention according to the present invention. FIG. 6 is an explanatory diagram illustrating a display screen of the vertical magnetic tomography analysis apparatus according to the present invention according to the present invention. FIG. 7 is an explanatory diagram illustrating a display area of the difference signal waveform. FIG. 8 is an explanatory diagram illustrating a display area of the analysis result. FIG. 9 is a graph showing the relationship between the separation distance estimated by the estimation model for vertical magnetic exploration analysis according to the present invention and the true value (actual measurement value). FIG. 10 is a graph showing the relationship between the separation distance estimated by the estimation model for vertical magnetic exploration analysis according to the present invention and the separation distance calculated by the calculation formula. FIG. 11 is a configuration diagram illustrating a system using a desktop PC instead of a server. FIG. 12 is a configuration diagram illustrating a system in which AI is mounted on a mobile PC. FIG. 13 is a flowchart illustrating the procedure of the vertical magnetic exploration method using the AI determination in the excavation process. FIG. 14 is an explanatory diagram for explaining the AI determination in the excavation process.

Hereinafter, an embodiment of a method for generating an estimation model for vertical magnetic exploration analysis, a vertical magnetic exploration analysis system, a vertical magnetic exploration analysis device, and a vertical magnetic exploration system to which a computer-readable program is applied will be described. ..

1. 1. Vertical magnetic exploration system FIG. 1 is a schematic configuration diagram illustrating a vertical magnetic exploration system to which an example of the vertical magnetic exploration analysis system according to the present invention is applied.

As shown in FIG. 1, the vertical magnetic exploration system 1 inserts a boring machine 10 that excavates the ground vertically downward to drill a columnar observation hole 11 and a columnar detector 21 inserted into the observation hole 11. It is equipped with a vertical magnetic exploration analysis system 2 that performs magnetic exploration.

The vertical magnetic tomography analysis system 2 includes a detector 21 inserted into the observation hole 11, a cable 22 connected to the detector 21, a control device 23 connected to the detector 21 via the cable 22, and a control device. The AD converter 24 connected to the output terminal of the 23, the mobile PC 25 (an example of the vertical magnetic tomography analysis device) connected to the control device 23 via the AD converter 24, and the network 26 such as the Internet were connected. It includes a server 27.

The detector 21 is a uniaxial differential fluxgate type magnetic sensor in which a pair of magnetic detection coils are coaxially arranged at intervals of, for example, about 0.5 to 2 m, and the differential signals of the pair of magnetic detection coils are It is output from the detector 21. In this differential signal, the noise of the signals output individually by the pair of magnetic coils (noise caused by environmental factors such as the earth's magnetic field and the vibration of the detector 21) cancel each other out. In addition to the differential signal, the detector 21 can also output a geomagnetic signal that the coil on the tip side independently outputs. Further, the distance between the pair of magnetic detection coils in the detector 21 is appropriately set according to the required search range and the required resolution.

The control device 23 can output the differential signal output from the detector 21 scanning the observation hole 21 and the depth of the detector 21 to the AD converter 24 in association with each other. The AD converter 24 AD-converts the differential signal with depth output from the detector 21 and then outputs the differential signal to the mobile PC 25, and the mobile PC has a depth according to a predetermined program stored in its internal memory. The differential signal is stored in the internal memory in a predetermined format (CSV format).

The control device 23 has a switch for the user (engineer) to turn on / off the power supplied from the battery to the detector 21, and a differential signal output from the detector 21 (and output from the tip side coil). A display unit that displays the level of the geomagnetic signal), an adjustment unit for the user (engineer) to switch the sensitivity of the differential signal (and the geomagnetic signal output from the tip side coil) output from the detector 21, and an adjustment unit. It is provided with an adjusting unit for the user (engineer) to adjust the offset of the differential signal (and the geomagnetic signal output from the tip side coil) output from the detector 21. The mobile PC 25 operates according to a predetermined program, monitors the differential signal given from the control device 23 side (or the geomagnetic signal output from the tip side coil), and a change of a certain level or more appears in the level. In this case (when a "magnetic anomaly" appears, which indicates that a magnetic material may be buried directly under the detector), an alarm is issued.

2. Vertical Magnetic Tomography Method FIG. 2 is a flowchart illustrating a procedure of the vertical magnetic tomography method. As shown in FIG. 2, in the vertical magnetic exploration method using the vertical magnetic exploration analysis system 1, the user (engineer) first installs the boring machine 10 at the survey point (S11) and observes vertically downward at a depth of 1 m. The hole 11 is excavated (S12). Then, the user (engineer) inserts the detector 21 into the observation hole 11 and performs magnetic exploration about 1 m below the bottom surface of the observation hole 11 (S13). At this time, the user (engineer) confirms the presence or absence of a magnetic anomaly based on the differential signal or the geomagnetic signal (S14), and if there is a magnetic anomaly (S14YES), the process proceeds to the continuous measurement procedure (S16). If there is no magnetic anomaly (S14NO), further 1 m is dug for magnetic exploration and confirmation (S12 to S14). Then, the series of procedures (S12 to S14) from excavation to confirmation are repeated until there is a magnetic anomaly (S14YES) or the bottom surface of the observation hole 11 reaches the planned depth (for example, 11 m) (S15YES).

After that, continuous measurement is performed by scanning the detector 21 from the bottom surface (lower end) of the observation hole 11 to the hole opening (upper end) at a constant speed, and the signal waveform of the differential signal output by the detector 21 during that period ( And the signal waveform of the geomagnetic signal) is saved in the internal memory of the mobile PC 25 (S16). After that, an analysis (described later) on the signal waveform of the differential signal is performed (S17).

If the danger of contact with the magnetic material 30 (magnetic anomaly) is observed during excavation (S14YES), the exploration is continued by taking measures such as shifting the position of the boring hole (position of the observation hole 11). You may.

Further, with respect to the depth at which the magnetic anomaly is recognized, the detector 21 may be scanned a plurality of times in the vertical direction to confirm the reproducibility of the data.

3. 3. Server The server 27 shown in FIG. 1 has a storage unit for storing programs and data, a processor for executing programs, a communication interface for receiving requests from a mobile PC 25 via a network 26, and a user interface for receiving input from a server administrator. It also has a trained estimation model 28 trained by a predetermined algorithm (here, a convolutional neural network). The trained estimation model 28 is a program corresponding to a recognizer of a convolutional neural network, and stores the coupling state of the recognizer in advance as a result of learning. In the present embodiment, the learned estimation model 28 is used when estimating the buried state of the magnetic body 30 in the ground. In the following, the trained estimation model 28 will be simply referred to as “estimation model 28” or “AI” as necessary.

The processor of the server 27 manages the estimation model 28, and when it receives the differential signal waveform data (image Im) from the mobile PC 25 via the network 26 and the communication interface, it inputs the image Im into the estimation model 28. The buried state of the magnetic body 30 in the ground is estimated.

Then, the processor of the server 27 provides the estimated embedded state of the magnetic material 30 to the mobile PC 25 via the network 26. The embedded state of the magnetic material 30 provided by the server 27 is presented to the user (engineer) via the screen of the mobile PC 25 or the like.

4. The estimation model FIG. 3 is an explanatory diagram for explaining the input data and the output data of the estimation model for vertical magnetic exploration analysis according to the present invention. As shown in FIG. 3, the estimation model 28 of the present embodiment is a trained estimation model learned by the convolutional neural network, and is a differential signal waveform of the detector 21 for vertical magnetic exploration inserted into the observation hole 11. This is an estimation model (estimation model for vertical magnetic exploration analysis) that estimates the buried state of the magnetic material 30 in the ground from the data (image Im). Hereinafter, as the buried state of the magnetic body 30, the separation distance r from the observation hole 11 to the magnetic body 30, the magnetic charge M of the magnetic body 30, and the direction (elevation / depression angle) of the magnetic poles of the magnetic body 30 as seen from the observation hole 11 ) Assuming a combination with θ (see FIG. 1).

5. Method of Generating Estimated Model FIG. 4 is an explanatory diagram illustrating a method of generating an estimated model for vertical magnetic exploration analysis according to the present invention. The method of generating this estimation model is mainly executed by the server 27.

The processor of the server 27 teaches the correspondence between the buried state of the magnetic body 30 (combination of r, M, and θ) in the ground and the image Im of the differential signal waveform generated by the detector 21 for vertical magnetic exploration. The acquisition procedure to be acquired as data and the learning procedure to train the estimation model 28 by the convolutional neural network using the acquired teacher data and save the learned estimation model 28 are executed.

As shown in FIG. 4, the teacher data is an image Im of a plurality of differential signal waveforms having different buried states (combinations of r, M, and θ) of the magnetic material 30. Images of these differential signal waveforms Im can be acquired using the above-mentioned vertical magnetic exploration system 1 or a system having the same specifications. For example, an observation hole 11 having a depth of 11 m is drilled by a boring machine 10, a sample of a magnetic material 30 (simulated bomb, iron pipe, etc.) is embedded at a position horizontally distant from the observation hole 11, and a hole mouth is formed from the bottom of the hole. By scanning the detector 21 at a predetermined speed toward the image, data (image Im) of the differential signal waveform is acquired. Further, about 400 images Im are acquired while changing the embedded state (combination of r, M, and θ) of the sample of the magnetic material 30 in about 400 ways. Here, according to the experiments of the inventors of the present application, these 400 images Im are roughly classified into four types according to the pattern type of the differential signal waveform (S type, V type, M type, and MW type). I was able to do it. Further, when comparing the images Im belonging to the same pattern type, it was found that the images were inverted vertically or horizontally, the size of the vertically or horizontally was different, or the noise frequency or the amount of noise was different.
Therefore, in the present embodiment, about 400 images Im that can be classified into four types are flipped vertically and horizontally, and the image Im is scaled vertically and horizontally, and noise is artificially added. By applying noise processing to be added, the number of teacher data was increased to about 60,000. This makes it possible to increase (expand) the number of teacher data without actual measurement. As a result, it is considered that the estimation model 28 can be easily and surely improved in performance.

6. Vertical Magnetic Tomography Analysis Device FIG. 5 is a flowchart illustrating a procedure of analysis by the vertical magnetic tomography analysis device according to the present invention according to the present invention. This analysis procedure is executed by the mobile PC 25 according to a program stored in the internal memory of the mobile PC 25.

As shown in FIG. 5, the processor of the mobile PC 25 internally contains CSV data (CSV file) of the differential signal waveform specified by the user (engineer) and CSV data (CSV file) of the corresponding geomagnetic signal waveform. It is read from the memory (S171), and the differential signal waveform and the geomagnetic signal waveform are displayed on the screen as graphs (S172, FIGS. 6A2 and A3).

Next, the processor of the mobile PC 25 causes the user (engineer) to specify each point (FIG. 7) of the waveform appearing in the graph (reference numerals A2 and A3 in FIG. 6) (S173, FIG. 7). This specification, a primary depth d, the analysis range AA, period 2x ₀ of the differential signal waveform, such as the maximum amplitude [Delta] H _max is specified. The method of designation by the user (engineer) will be described later.

_{Next, the processor of the mobile PC 25 reads dimensions such as a period 2 x 0} and a maximum amplitude ΔH _{max of} the differential signal waveform based on each point specified by the user (engineer), and transmits the dimensions to the server 27 via the network 26. The dimensions (period 2 x ₀ , maximum amplitude ΔH _max ) are transmitted. _{The server 27 applies the values of the period 2 x 0} and the maximum amplitude ΔH _{max to} the following calculation formulas (1) and (2) without using the estimation model 28, so that the separation distance r'of the magnetic body 30 and the magnetism of the magnetic body 30 are applied. The quantity M'is calculated, and the calculated values of the separation distance r'and the magnetic charge M'are transmitted to the mobile PC 25 via the network 25. Then, the mobile PC 25 receives the separation distance r'and the magnetic charge M'via the network 25 (S174).

_{^{r '2 = {(x 0}} +0.5) (x 0 -0.5)} 2/3 × {(x 0 +0.5) 2/3 + (x 0 -0.5) 2/3} ... (1)

^{_{M '= ΔH max (0.5 2}} + r' 2) 3/2 × 10 ... (2)

Next, the processor of the mobile PC 25 displays the separation distance r'and the magnetic charge M'received from the server 27 on the screen (reference numerals A4 and 8 in S175 and FIG. 6).

On the other hand, the processor of the mobile PC 25 generates the data of the differential signal belonging to the analysis range AA (FIG. 7) as the image Im, and transmits the generated image Im to the server 27 via the network 26. The server 27 that has received the image Im estimates the separation distance r, the magnetic charge M, and the direction θ of the magnetic poles of the magnetic body 30 by inputting the image Im into the estimation model 28, and the separation distance r, via the network 26, The magnetic charge M and the direction θ of the magnetic poles are transmitted to the mobile PC 25. As a result, the mobile PC 25 can acquire the separation distance r of the magnetic body 30, the magnetic charge M, and the direction θ of the magnetic poles (S176).

Next, the processor of the mobile PC 25 displays the separation distance r, the magnetic charge M, and the direction θ of the magnetic poles received from the server 27 on the screen (S177, reference numeral A4 in FIG. 8).
As a result of the above, the user (engineer) can compare the separation distance r'and the magnetic quantity M'calculated by the calculation formula with the separation distance r, the magnetic quantity M, and the magnetic pole direction θ estimated by the estimation model 28. It is presented to (Fig. 8). At this time, it is desirable that the depth d is also displayed.

7. Display screen FIG. 6 is an explanatory diagram illustrating a display screen of the vertical magnetic tomography analysis apparatus according to the present invention according to the present invention. This display screen is displayed on the screen of the mobile PC by the processor of the mobile PC 25 according to the program stored in the internal memory of the mobile PC.

At the upper part of the display screen of FIG. 6, an area A1 for a user (engineer) to specify a file of differential signal waveform data is provided.

On the left side of the center of the display screen of FIG. 6, the area A2 in which the geomagnetic signal waveform generated independently by the coil on the tip side of the detector 21 is displayed as a graph and the differential signal waveform generated by the detector 21 are displayed as a graph. Area A3 (see FIG. 7) is provided side by side.

On the right side of the center of the display screen of FIG. 6, the separation distance r, the magnetic charge M, and the magnetic charge direction θ estimated by the estimation model 28, and the separation distance r'and the magnetic charge M'calculated by the calculation formula can be listed. An area A4 (see FIG. 8) for displaying the state is provided.
At the bottom of the display screen of FIG. 6, an area A5 is provided in which execution logs related to the analysis performed via the server 27 are displayed in chronological order.

8. Display area of the differential signal waveform FIG. 7 is an explanatory diagram illustrating a display area A3 of the differential signal waveform. The user (engineer) can specify the upper end depth and the lower end depth _{of the half cycle x 0} of the differential signal waveform by pointing with the mouse or inputting a numerical value while visually observing the display area A3 of the difference signal waveform. can. Thus, the mobile PC25 can identify the period 2x ₀ of the differential signal waveform.

Further, the user (engineer) can specify the tangent line of the peak having the largest amplitude in the differential signal waveform by performing pointing or numerical input with the mouse while visually observing the display area A3 of the difference signal waveform. can. Thereby, the mobile PC 25 can specify _{the maximum amplitude ΔH max of the differential signal waveform.}

Further, the user (engineer) can specify the reference depth d by pointing with the mouse or inputting a numerical value while visually observing the display area A3 of the difference signal waveform. The reference depth d may be specified by the user (engineer) based on the graph of the geomagnetic signal waveform (reference numeral A2 in FIG. 6).

Further, the user (engineer) can specify the upper end depth and the lower end depth of the analysis range AA by performing pointing or numerical input with the mouse while visually observing the display area A3 of the difference signal waveform. The mobile PC 25 may automatically set a predetermined range centered on the depth d as the analysis range AA.

6. Analysis result display area FIG. 8 is an explanatory diagram illustrating an analysis result display area A4. As shown in FIG. 8, in the analysis result display area A4, an area A41 for displaying the direction θ of the magnetic pole, an area A42 for displaying the separation distance r, an area A43 for displaying the depth d, and a magnetic charge M are displayed. Area A44, _{area A51 displaying the maximum amplitude ΔH max} , area A52 displaying the period 2x0, area A53 displaying the separation distance r', area A54 displaying the magnetic charge M', and magnetic charge M. An area A61 for displaying "" is provided. The magnetic charge M "is obtained by applying the separation distance r estimated by the estimation model 28 and the above-mentioned maximum amplitude ΔH _max to the calculation formula (2). This is the magnetic charge calculated by the processor of the server 27.

9. Evaluation of Estimated Model FIG. 9 is a graph showing the relationship between the separation distance estimated by the estimated model for vertical magnetic exploration analysis according to the present invention and the true value (measured value). As shown in FIG. 9, the separation distance r estimated by the estimation model 28 has a slight error as compared with the true value of the separation distance, but this error is within a range of about ± 0.5 m.

FIG. 10 is a graph showing the relationship between the separation distance estimated by the estimation model for vertical magnetic exploration analysis according to the present invention and the separation distance calculated by the calculation formula. As shown in FIG. 10, the separation distance r estimated by the estimation model 28 tends to show a slightly larger value than the separation distance r'calculated by the calculation formula. In addition, the separation distance r shows a characteristic distribution depending on the combination of the separation distance and the weight (that is, the magnetic charge). For example, the accuracy at the separation distance of 2 m or less is high for the 150 mm shell size, and 3 m or more for the 500 kg bomb size. The results showed that a slightly smaller value was likely to be obtained at the separation distance.

Although only the estimation error of the separation distance r is verified here, it is considered that the estimation model 28 can also estimate the magnetic charge M and the direction θ of the magnetic poles with a certain accuracy. _{This is because it is well known that there is a correlation between the maximum amplitude ΔH max} of the differential signal waveform and the magnetic amount M of the magnetic material 30, and the pattern types of the differential signal waveform (MV type, M type, V type, This is because it has been found from experiments by the inventors of the present application and the like that there is a correlation between the S type, the inverted S type, the V type, the inverted M type, and the inverted MV type) and the magnetic pole direction θ. (See FIG. 4).

10. Effect of Embodiment According to this embodiment, the correspondence between the buried state of the magnetic body 30 (combination of r, M, and θ) and the image Im of the differential signal waveform generated by the detector 21 is convolved as teacher data. neural network: since (CNN Convolution neural network) by train the estimation model 28, the user (technician) is a differential signal waveform of dimensions (amplitude [Delta] H, periodic 2x ₀ such changes) on the screen manually that measure I don't need it.

Then, there is a strong correlation between the shape (pattern) of the differential signal waveform and the embedded state (combination of r, M, θ) of the magnetic body 30. For example, the period 2x ₀ of changes in separation distance r is short enough differential signal waveform of the magnetic body 30 becomes shorter, the differential signal waveform as distance r of the magnetic member 30 is large magnetic charge M of shorter magnet 30 It is already known that the maximum amplitude ΔH _max also increases (see formulas (1) and (2)). Further, according to an experiment by the inventors of the present application, the pattern types of the differential signal waveforms are MV type, M type, V type, S type, and inverted S according to the direction θ of the magnetic poles of the magnetic body 30 seen from the observation hole 11. It has also been found to change between types, inverted V types, inverted M types, inverted MV types, and the like (Fig. 4).

Moreover, since the convolutional neural network is good at capturing the feature points of the image, it recognizes the feature of the shape (pattern) of the differential signal waveform as an image Im and strongly correlates with the shape (pattern) of the differential signal waveform. It is suitable for estimating the buried state (combination of r, M, and θ) of the magnetic material 30 having the above.

Therefore, according to the present embodiment using the trained estimation model 28 by the convolutional neural network, the embedded state (combination of r, M, and θ) of the magnetic body 30 can be estimated with high accuracy.

11. Others As described above, the preferred embodiment of the present invention has been described in detail, but the present invention is not limited to the specific embodiment, and is within the scope of the gist of the present invention described in the claims. Needless to say, various modifications and changes are possible.

11-1. Additional learning of AI For example, in the system described above, the estimation model 28 may be updated by additionally learning the estimation model 28 with new teacher data. By repeating such an update, the performance of the estimation model 28 can be improved over time.

11-2. Allocation of functions For example, the allocation of functions in the above-mentioned system is not limited to the above-mentioned ones, and some or all of the functions installed on the server 27 side may be installed on the mobile PC 25 side. However, some of the functions mounted on the mobile PC 25 side may be mounted on the server 27 side. Further, some of the functions installed in the mobile PC 25 (function of displaying the estimation result, function of accessing the server, etc.) may be installed in a desktop PC (not shown). Further, a part of the functions of the control device 23 and the AD converter 24 may be mounted on the side of the mobile PC 25, or a part or all of the functions of the AD converter 24 and the mobile PC 25 may be mounted on the side of the control device 23. You may. It is also possible to use a PC instead of the server 27, and for example, a desktop PC 127 installed in an office or the like may be used instead of the server 27 (see FIG. 11).

11-3. AI-equipped mobile PC
Further, if the estimation model 28 is mounted (installed) on the mobile PC 25 in advance as shown in FIG. 12, the estimation model 28 can be used at the site where the mobile PC 25 cannot connect to the network 26. Further, in that case, the process of learning the estimation model 28 by the convolutional neural network using the teacher data (the process of generating the estimation model 28) may be executed by the server 27 (FIG. 1), or in an office or the like. The installed desktop PC 127 (FIG. 11) may be executed. This is because the trained estimation model 28 can be installed on the mobile PC 25 via the network 26. When the server 27 (FIG. 1) or the desktop PC 127 (FIG. 11) updates the estimation model 28 by additional learning, the updated estimation model 28 is re-installed on the mobile PC 25 via the network 26 (program is updated). )do it. Further, the server 27 (FIG. 1) or the desktop PC 127 (FIG. 11) may collect additional teacher data necessary for additional learning via the network 26.

11-4. Direct judgment using AI Further, direct judgment using AI may be performed in the excavation process at the site. FIG. 13 is a flowchart illustrating the procedure of the vertical magnetic exploration method using the AI determination in the excavation process. The difference between the flow of FIG. 13 and the flow of FIG. 2 is that steps S101, S102, and S103 are inserted between steps S14 to S16.

In the flow of FIG. 13, when a waveform that seems to be the initial stage of the abnormal waveform is observed during the observation every 1 m during excavation (S14YES), it is described above whether or not the magnetic material 30 exists directly under the observation hole 11. Estimate with the estimation model 28 (AI) (S101). Then, when the estimation model 28 estimates that the magnetic material 30 exists directly under the observation hole 11 (S101YES), the excavation is immediately completed (S103) and the process proceeds to step S16 (S103). If not (S101NO), the process proceeds to step S15 for further excavation.

In the learning of such an estimation model 28, as the buried state of the magnetic body 30 in the ground, the classification of the area to which the separation distance r and the depth d belong (separate classifications 1 to 9 in FIG. 14) and the vertical magnetic exploration are performed. The association of the differential signal waveform generated by the detector 21 with the image Im may be used as the teacher data. In that case, the trained estimation model 28 can estimate which of the categories 1 to 9 the area where the magnetic body 30 exists. Therefore, in step S102 described above, for example, when the estimation model 28 estimates whether or not the classification of the area to which the magnetic material 30 belongs is "1", and the estimation result that the classification is "1" is obtained. It may be determined that the magnetic body 30 exists directly under the observation hole 11 (step S102YES), and if not, it may be determined that the magnetic body 30 does not exist directly under the observation hole 11 (step S102NO). ..

11-4. System update For example, unexploded ordnance exploration measures the presence or absence of unexploded ordnance at a depth of 1 m while boring, so if the estimation accuracy by the estimation model is low, there is a risk of a major accident. Therefore, it can be said that the ability to quickly estimate without human error using the estimation model is of great value in terms of preventing accidents. The system of the present embodiment can be variously modified with respect to the user interface. Further, in the present embodiment, it is possible to devise to enhance the versatility as a tool that can compare the result using the calculation formula and the result using the estimation model. Furthermore, the estimated model is sequentially learned (additional learning) with new teacher data prepared based on the comparison (the estimated model is updated), and the estimated model is further learned with the nearest teacher data with a separation distance of 50 cm or less. By making it (update the estimation model), it is possible to improve the estimation accuracy and expand the applicable range. It is also possible to build a system so that new teacher data can be learned each time analysis is performed.

1 Vertical magnetic exploration system 10 Boring machine 2 Vertical magnetic exploration analysis system 21 Detector 22 Cable 23 Control device 24 AD converter 25 Mobile PC
27 Server 28 Estimated model 20 Estimated model 26 Network 30 Magnetic material r Separation distance M Magnetic charge θ Direction of magnetic poles

Claims (11)

It is a method of generating an estimation model for vertical magnetic exploration analysis that generates an estimation model for estimating the buried state of a magnetic material based on the signal waveform of a detector for vertical magnetic exploration inserted in an observation hole in the ground.
An acquisition procedure for acquiring the correspondence between the buried state of the magnetic material in the ground and the image of the signal waveform generated by the detector for vertical magnetic exploration as teacher data, and
A learning procedure that trains an estimation model by a convolutional neural network using the acquired teacher data and saves the trained estimation model.
A method for generating an estimation model for vertical magnetic exploration analysis, which comprises. In the method for generating an estimation model for vertical magnetic exploration analysis according to claim 1.
In the embedded state of the magnetic material,
A method for generating an estimation model for vertical magnetic exploration analysis, characterized in that the separation distance r from the observation hole to the magnetic material is included. In the method for generating an estimation model for vertical magnetic exploration analysis according to claim 2.
Further, in the embedded state of the magnetic material,
The magnetic charge M of the magnetic material and
A method for generating an estimation model for vertical magnetic exploration analysis, which comprises at least one of the directions θ of the magnetic poles of the magnetic material as seen from the observation hole. In the method for generating an estimation model for vertical magnetic exploration analysis according to claim 3.
In the acquisition procedure,
By acquiring images of a plurality of signal waveforms while changing the embedded state of the magnetic sample, and performing at least one of inversion processing, scaling processing, and noise processing on the images of the plurality of signal waveforms. A method for generating an estimation model for vertical magnetic exploration analysis, which is characterized by expanding teacher data. In the method for generating an estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 4.
The detector is
A method for generating an estimation model for vertical magnetic exploration analysis, which is characterized by being a uniaxial differential fluxgate type magnetic sensor. In the method for generating an estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 5.
A method for generating an estimation model for vertical magnetic exploration analysis, characterized in that the magnetic material is unexploded. A management means for managing the estimation model generated by the method for generating the estimation model for vertical magnetic exploration analysis according to any one of claims 1 to 6.
An estimation means for estimating the buried state of a magnetic material in the ground by inputting an image of a signal waveform generated by a detector for vertical magnetic exploration into the estimation model, and an estimation means.
A vertical magnetic exploration analysis system characterized by being provided with a presentation means for presenting the buried state of a magnetic material to a user. In the vertical magnetic exploration analysis system according to claim 7.
The estimation means
The embedded state of the magnetic material is calculated by applying the dimensions of the signal waveform to a predetermined calculation formula that does not use the estimation model.
The presentation means
A vertical magnetic exploration analysis system characterized in that the estimated buried state and the calculated buried state are presented to the user in a state in which they can be compared with each other. In the vertical magnetic exploration analysis system according to claim 7 or 8.
A vertical magnetic exploration analysis system characterized by further equipped with a detector for vertical magnetic exploration. A vertical magnetic exploration analysis device applied to the vertical magnetic exploration analysis system according to any one of claims 7 to 9.
A vertical magnetic tomography analysis apparatus including the estimation means and the presentation means. A computer-readable program according to claim 10, wherein the computer functions as the vertical magnetic exploration and analysis apparatus.

Images