Soil constant estimation device, program, and soil constant estimation method

Abstract

To estimate soil constant with higher accuracy by using data that is not generally used.SOLUTION: In a machine learning stage, a learning model is generated by machine learning with the data group obtained during a cone penetration test as an explanatory variable and a soil constant obtained from the boring survey as an objective variable. Next, as a soil constant estimation stage, this learning model is used to estimate the soil constant at the point where the cone penetration test was performed. The data group obtained during the cone penetration test includes three component data such as cone penetration resistance, circumferential friction resistance, and pore water pressure, and at least one of underground image data obtained by the image data generation device 21 during the cone penetration test, sound wave data measured by a sound wave transmitting / receiving device 22 provided in the cone, and soil data classified based on subsurface image data or subsurface sound wave data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for estimating soil constants of ground. [Background technology]

[0002] Conventional ground investigation methods generally involve boring surveys, and standard penetration tests and laboratory soil tests using sampled samples are conducted to obtain the soil constants used in design. A relatively simple ground investigation method is also the sounding test, typified by the electric cone penetration test (hereinafter also referred to as CPT). It is known that the data obtained from CPT correlates with soil constants, and various conversion formulas and systems for estimating soil constants have been proposed.

[0003] To obtain the N-value and Fc (fine grain content) from a boring survey, a standard penetration test or laboratory soil test is required, which increases the cost and time required for the survey. On the other hand, sounding tests, such as CPT, allow for relatively inexpensive ground surveys and can estimate soil constants, but the accuracy of the estimates can be significantly lower depending on the ground conditions, so they are ultimately positioned as a supplement to boring surveys.

[0004] For example, Patent Document 1 discloses a mechanism for estimating the N value and Fc at a location by performing machine learning based on boring data and drilling data to generate a learning model that estimates the N value and Fc, and then inputting drilling data at that location where the N value and Fc are unknown into this learning model. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2020-100949 Summary of the Invention [Problem to be solved by the invention]

[0006] However, it is desirable to estimate soil constants with high accuracy, and by using data that is not commonly used for soil property estimation, it is possible to estimate soil properties with higher accuracy. The object of the present invention is to estimate soil constants with higher accuracy by using data that is not commonly used. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, the present invention provides a system including: a generation unit that generates a learning model by machine learning using explanatory variables of a data group obtained during a cone penetration test, the data group being three-component data, and at least one of an image data group relating to an underground image, a sonic data group measured by a sonic wave transmitting / receiving device attached to a cone, and a soil data group classified based on the image data relating to an underground image or the sonic data measured underground, and soil constants obtained by boring as objective variables; and an estimation unit that inputs explanatory variables obtained during a cone penetration test conducted at a site where soil constants are to be estimated, corresponding to the explanatory variables in the learning model, into the generated learning model, and estimates the soil constants at the site where soil constants are to be estimated. The soil constants are strength characteristics, fine particle content, ground density, particle size composition, water content, degree of saturation, deformation characteristic value, or permeability coefficient. A soil constant estimation device is provided.

[0008] The image-related data may be visible image data.

[0009] The visible image data may be data captured by a camera attached to the cone.

[0010] The cone may have a protrusion on its outer periphery.

[0011] The image-related data may be non-visible image data.

[0012] The invisible image data may be invisible image data based on measurements by an electromagnetic wave transmitting / receiving device provided on the cone.

[0013] The electromagnetic waves transmitted and received by the electromagnetic wave transmitting and receiving device may be at least one of gamma rays, neutron rays, and infrared rays.

[0014] The present invention also provides a computer that functions as a generation unit that generates a learning model by machine learning using explanatory variables of a data group obtained during a cone penetration test, the data group being three-component data, an image data group relating to at least an underground image, a sonic data group measured by a sonic wave transmitting / receiving device attached to the cone, or a soil data group classified based on the image data relating to an underground image or the sonic data measured underground, and soil constants obtained by boring as objective variables; and an estimation unit that inputs explanatory variables obtained during a cone penetration test conducted at a soil constant estimation target point, which data group corresponds to the explanatory variables in the learning model, into the generated learning model, and estimates the soil constants at the estimation target point. The soil constants are strength characteristics, fine particle content, ground density, particle size composition, water content, degree of saturation, deformation characteristic value, or permeability coefficient. Provide programs.

[0015] The present invention also provides a soil constant estimation method executed by the soil constant estimation device. [Effects of the Invention]

[0016] According to the present invention, the soil constant estimation device generates a trained model by machine learning using three-component data obtained during a cone penetration test, at least one of image data relating to an underground image, sonic data measured by a sonic transmitter / receiver attached to the cone, or soil data classified based on image data relating to an underground image or sonic data measured underground, and soil constants obtained by boring. By using this trained model, soil constants can be easily estimated.

[0017] Furthermore, according to the present invention, the soil constant estimation device can estimate soil properties based on a target variable obtained by inputting, as explanatory variables, into a trained model, three-component data obtained during a cone penetration test and at least one of image data relating to an underground image, sonic data measured by a sonic transmitter / receiver attached to the cone, or soil data classified based on image data relating to an underground image or sonic data measured underground. This makes it possible to easily estimate changes in soil constants. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a block diagram showing an example of the overall configuration of a system according to a first embodiment of the present invention. [Figure 2] FIG. 1 is a block diagram showing an example of the hardware configuration of a soil constant estimation device according to a first embodiment. [Figure 3] FIG. 2 is a block diagram showing an example of the functional configuration of a first soil constant estimation device. [Figure 4] 4 is a flowchart showing an example of a method for generating a learning model in the first embodiment. [Figure 5] 4 is a flowchart showing an example of a method for estimating soil constants using a learning model in the first embodiment. [Figure 6] FIG. 10 is a block diagram showing an example of the overall configuration of a system according to a second embodiment of the present invention. [Figure 7] 10 is a table showing experimental results for explaining the effects of the third embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] An example of an embodiment of the present invention will be described. [First embodiment] [composition] FIG. 1 is a block diagram showing an example of the overall configuration of a system 1 according to a first embodiment of the present invention. The system 1 includes a boring system 10 that drills holes in the ground to conduct a boring survey; a cone penetration test system 20 that performs an electric cone penetration test on the ground; and a soil constant estimation device 30 that estimates the soil constants of the ground. An image data generation device 21 is provided inside the electric cone of the cone penetration test system 20. The image data generation device 21 is composed of, for example, a camera and a light for capturing visible images. A portion of the side of the tip of the electric cone is composed of a transparent material such as crystal glass. During the cone penetration test, the light constituting the image data generation device 21 irradiates light through this transparent material into the ground outside the cone, and the camera constituting the image data generation device 21 captures images of the ground. As a result, image data of the ground during the cone penetration test is generated by the image data generation device 21. This image data may be video data or a collection of multiple still image data that are consecutive in time. In this embodiment, however, video data is used.

[0020] The boring system 10 and the soil parameter estimation device 30, or the cone penetration test system 20 and the soil parameter estimation device 30, may or may not be electrically connected via wire or wirelessly. When the boring system 10 and the soil parameter estimation device 30, or the cone penetration test system 20 and the soil parameter estimation device 30, are electrically connected, data output from the boring system 10 or the cone penetration test system 20 is input to the soil parameter estimation device 30 via wire or wirelessly. When the boring system 10 and the soil parameter estimation device 30, or the cone penetration test system 20 and the soil parameter estimation device 30 are not electrically connected, data output from the boring system 10 or the cone penetration test system 20 is input to the soil parameter estimation device 30, for example, via a predetermined storage medium or manually by an operator.

[0021] The boring survey conducted by the boring system 10 measures soil constants of the ground, such as strength characteristics (hereinafter referred to as N-value) and fines content (hereinafter referred to as Fc). The N-value obtained from a standard penetration test conducted in conjunction with the boring survey can be used to estimate many design soil constants, and is therefore extremely useful. In addition, Fc obtained from indoor soil tests using sampled samples affects the permeability and liquefaction resistance of the ground. While boring surveys can directly measure these data and have high measurement accuracy, they are expensive and, depending on the soil quality, can be difficult to conduct.

[0022] The cone penetration test system 20 is a system that attaches an electric cone to the tip of a rod, penetrates the ground, and continuously measures three data components: cone penetration resistance, skin friction resistance, and pore water pressure. The measured data from the cone penetration test can be used to determine, for example, the classification of soil layers, confirm the effectiveness of ground improvement, or assess liquefaction. Furthermore, these three data components are empirically believed to correlate with soil constants obtained from boring surveys and other soil investigations, and many conversion formulas have been proposed.

[0023] In the system 1 according to the present embodiment, a learning model is generated through machine learning in a machine learning stage, where data sets obtained during a cone penetration test are used as explanatory variables and soil parameters obtained through a boring survey are used as objective variables. Then, in a soil parameter estimation stage, this learning model is used to estimate soil parameters at the site where the cone penetration test was conducted. The data obtained during the cone penetration test (hereinafter referred to as cone penetration test data) includes three-component data, such as cone penetration resistance, skin friction resistance, and pore water pressure, as well as subsurface image data obtained by an image data generating device 21 during the cone penetration test. Characteristics of the subsurface image data obtained during the cone penetration test are thought to be correlated with subsurface soil parameters, such as the presence of discontinuous surfaces (layer boundaries) and soil properties (such as clayey soil or sandy soil). Therefore, a learning model that includes such image data or a soil parameter set classified based on image data of the subsurface images as explanatory variables is thought to improve the accuracy of soil parameter estimation compared to a learning model that does not include such image data as explanatory variables.

[0024] FIG. 2 is a diagram showing the hardware configuration of the soil constant estimation apparatus 30. Physically, the soil constant estimation apparatus 30 is configured as a computer apparatus including a processor 3001, a memory 3002, a storage 3003, a communication device 3004, an input device 3005, an output device 3006, and a bus connecting these devices. Each of these devices operates using power supplied from a power source (not shown). In the following description, the term "apparatus" can be interpreted as a circuit, device, unit, etc. The hardware configuration of the soil constant estimation apparatus 30 may be configured to include one or more of the devices shown in FIG. 2, or may be configured without including some of the devices.

[0025] Each function of the soil constant estimation device 30 is realized by loading predetermined software (programs) onto hardware such as a processor 3001 and memory 3002, causing the processor 3001 to perform calculations, control communications via the communication device 3004, acquire data transmitted from other devices, and control at least one of reading and writing data in the memory 3002 and storage 3003.

[0026] The processor 3001 controls the entire computer by running, for example, an operating system. The processor 3001 may be configured as a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. Furthermore, for example, a baseband signal processing unit, a call processing unit, etc. may be realized by the processor 3001.

[0027] The processor 3001 reads programs (program codes), software modules, data, etc. from at least one of the storage 3003 and the communication device 3004 into the memory 3002 and executes various processes in accordance with these. The programs used are those that cause a computer to execute at least some of the operations described below. The functional blocks of the soil constant estimation device 30 may be implemented by a control program stored in the memory 3002 and running on the processor 3001. Various processes may be executed by one processor 3001, or may be executed simultaneously or sequentially by two or more processors 3001. The processor 3001 may be implemented on one or more chips. The programs may be transmitted to the soil constant estimation device 30 via a telecommunications line or installed in the memory 3002 or the storage 3003.

[0028] The memory 3002 is a computer-readable recording medium and may be configured by, for example, at least one of a ROM (Read Only Memory), an EPROM (Erasable Programmable ROM), an EEPROM (Electrically Erasable Programmable ROM), a RAM (Random Access Memory), etc. The memory 3002 may also be called a register, a cache, a main memory (primary storage device), etc. The memory 3002 can store an executable program (program code), a software module, etc. for implementing the method according to this embodiment.

[0029] Storage 3003 is a computer-readable recording medium, and may be composed of at least one of, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a solid-state drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, etc. Storage 3003 may also be referred to as an auxiliary storage device.

[0030] The communication device 3004 is hardware (transmission / reception device) for performing communication between computers via at least one of wired and wireless means, and is also called, for example, a network device, a network controller, a network card, or a communication module.

[0031] The input device 3005 is an input device (for example, a keyboard, a mouse, a microphone, a switch, a button, etc.) that receives input from the outside. The output device 3006 is an output device (for example, a display, a speaker, an LED lamp, etc.) that outputs to the outside. The input device 3005 and the output device 3006 may be integrated into one device (for example, a touch panel).

[0032] The soil constant estimation device 30 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 3001 may be implemented using at least one of these pieces of hardware.

[0033] 3 is a diagram showing an example of the functional configuration of the soil constant estimation device 30. Each function realized by the soil constant estimation device 30 is realized by loading predetermined software (programs) onto hardware such as a processor 3001 and a memory 3002, causing the processor 3001 to perform calculations, control communications via a communication device 3004, and control at least one of reading and writing of data from and to the memory 3002 and storage 3003.

[0034] 3, a data acquisition unit 31 acquires various data from outside the soil constant estimation device 30. This data includes the soil constants N value and Fc measured by a boring survey using the boring system 10, and cone penetration test data obtained during a cone penetration test performed by the cone penetration test system 20. As described above, the cone penetration test data includes three-component data, namely, cone penetration resistance value, skin friction resistance value, and pore water pressure value, as well as a group of underground image data obtained during the cone penetration test.

[0035] In the machine learning stage, the teacher data generation unit 32 generates teacher data used to generate a learning model using the data acquired by the data acquisition unit 31. More specifically, the teacher data generation unit 32 generates teacher data for a learning model in which the above-mentioned cone penetration test data (three-component data, such as cone penetration resistance, surface friction resistance, and pore water pressure, and image data) are used as explanatory variables, and the soil constants N-value and Fc are used as objective variables. In other words, the teacher data generation unit 32 considers three-component data detected at a certain point and depth in a cone penetration test and image data representing the image captured at that time (explanatory variables) as one set, and the N-value and Fc (objective variables) obtained by a boring survey at that point and depth as one set, and regards the collection of these sets as teacher data. This learning model is, for example, a model obtained by machine learning using a neural network.

[0036] In the machine learning stage, the model generation unit 33 performs machine learning using, for example, a neural network, with the cone penetration test data as explanatory variables and the N value and Fc as objective variables as training data to generate a learning model. Machine learning using a neural network can analyze the relationship between the explanatory variables and objective variables included in the training data to generate a learning model that outputs data corresponding to the objective variable when data corresponding to the explanatory variables is input.

[0037] The model storage unit 34 stores the learning model generated by the model generation unit 33 in the machine learning stage.

[0038] The verification unit 35 verifies the accuracy of the learning model in the machine learning stage. More specifically, the verification unit 35 applies cone penetration test data to the learning model stored in the model storage unit 34 to estimate the N value and Fc at points where the N value and Fc are known, compares the estimated value with the known N value and Fc, and evaluates the difference. If the accuracy of the learning model is insufficient, such as if the difference exceeds a threshold, the model generation unit 33 increases the amount of training data in the learning model, reviews the weighting, or reviews hyperparameters, etc., to generate a learning model with sufficient accuracy.

[0039] The estimation unit 36 ​​estimates and outputs the N value and Fc at a point where a cone penetration test was conducted using the learning model stored in the model storage unit 34 and the cone penetration test data (three-component data and image data group) acquired by the data acquisition unit 31. The design soil constants are determined based on the N value and Fc and are used for pre-construction design, etc.

[0040] [Operation] [Operations in the machine learning stage] First, the operation of this embodiment in the machine learning stage will be described. Fig. 4 is a flowchart showing an example of a method by which the soil constant estimation device 30 generates a learning model. First, a boring survey is carried out at a certain point by the boring system 10 (step S11). As a result, the N value and Fc of the ground at that point are measured.

[0041] Next, the cone penetration test system 20 performs a cone penetration test at the point where the boring survey was conducted (step S12). This allows for the acquisition of cone penetration test data (three-component data, such as cone penetration resistance, surface friction resistance, and pore water pressure, as well as image data) at the boring survey point. Note that either the boring survey in step S11 or the cone penetration test in step S12 can be performed first.

[0042] After such boring surveys and cone penetration tests have been conducted at a sufficient number of locations for machine learning, sets of N values, Fc, and cone penetration test data at each of these locations are input to the soil constant estimation device 30 (step S13). As a result, the data acquisition unit 31 of the soil constant estimation device 30 acquires the N values, Fc, and cone penetration test data for the multiple locations.

[0043] Next, the teacher data generation unit 32 of the soil constant estimation device 30 generates teacher data using the cone penetration test data acquired by the data acquisition unit 31 as explanatory variables and the N value and Fc acquired by the data acquisition unit 31 as objective variables (step S14).

[0044] Next, the model generation unit 33 performs machine learning using the training data (step S15), and the learning model thus generated is stored in the model storage unit 34 (step S16).

[0045] Next, the verification unit 35 verifies the accuracy of the learning model stored in the model storage unit 34 (step S17). Specifically, the verification unit 35 inputs the cone penetration test data at each point into the learning model as explanatory variables, compares the N value and Fc obtained as the objective variables with the N value and Fc obtained by boring survey at that point, and evaluates the difference.

[0046] If the accuracy of the learning model is insufficient (step S18; No), the model generation unit 33 reviews the hyperparameters etc. in the learning model and repeats the generation of the learning model until the accuracy becomes sufficient (steps S16 to S17). If it is determined that the accuracy of the learning model is sufficient (step S18; Yes), the process shown in FIG. 4 ends.

[0047] [Operation in the soil constant estimation stage] Next, the operation of this embodiment in the soil constant estimation stage will be described. Fig. 5 is a flowchart showing an example of a method for estimating soil constants using a learning model. First, a cone penetration test is conducted at a target point (hereinafter referred to as the estimation target point) for which soil constants are to be estimated, and the resulting cone penetration test data (three-component data, i.e., cone penetration resistance value, skin friction resistance value, and pore water pressure value, and image data group), i.e., a data group corresponding to the explanatory variables in the learning model, is input to the soil constant estimation device 30 (step S19).

[0048] Next, the estimation unit 36 ​​of the soil constant estimation device 30 inputs the cone penetration test data as explanatory variables into the learning model stored in the model storage unit 34 (step S20).

[0049] Then, the estimation unit 36 ​​acquires the N value and Fc obtained as the objective variables from the learning model (step S21). The N value and Fc are the N value and Fc estimated at the estimation target point.

[0050] According to the first embodiment described above, a learning model is generated using underground visible image data, which is thought to be correlated with soil constants, as explanatory variables, and by using this learning model, it is possible to estimate soil constants with high accuracy.

[0051] [Second embodiment] Next, a second embodiment of the present invention will be described. FIG. 6 is a block diagram showing an example of the overall configuration of a system 1a according to the second embodiment. The system 1a includes a boring system 10, a cone penetration test system 20, and a soil constant estimation device 30 similar to those of the first embodiment. However, an acoustic wave transmitting / receiving device 22 is provided inside the cone of the cone penetration test system 20 instead of, or in addition to, the image data generating device 21 of the first embodiment. The acoustic wave transmitting / receiving device 22 transmits acoustic waves of a predetermined frequency into the ground throughout the period during which the cone penetration test is being conducted, receives the acoustic waves reflected from the ground, and outputs the resulting measured acoustic wave data. The acoustic wave transmitting / receiving device 22 may transmit acoustic waves in the same direction as the cone's penetration, or at a predetermined angle relative to the penetration direction (e.g., 90 degrees relative to the penetration direction). The acoustic wave transmitting / receiving device 22 may be separated into an acoustic wave transmitting device and an acoustic wave receiving device, each of which is positioned so as to easily receive the acoustic waves reflected from the ground. The characteristics of the sonic data group measured in this way are thought to be influenced by the hardness or softness of the underground ground. In this second embodiment, instead of or in addition to the image data group as cone penetration test data used in the first embodiment, the above-mentioned sonic data group is used as one of the explanatory variables to generate a learning model and estimate soil constants. Furthermore, in this second embodiment, a learning model generated by further including a soil data group classified based on the sonic data as an explanatory variable may be used. According to this second embodiment, the learning model uses sonic data that is thought to be correlated with the soil constants as an explanatory variable, so it is possible to estimate soil constants with higher accuracy than when using a learning model that does not include sonic data as an explanatory variable.

[0052] The sonic data includes inaudible sound data and audible sound data. The audible sound data is the sound of contact between the cone and the target ground during cone penetration, and is collected by a sound-collecting microphone installed on the side of the tip of the cone as the sonic transmitter / receiver 22. When using audible sound data, it is desirable to use it in a state where it is linked to image data as cone penetration test data.

[0053] The image data and the audible sound data can be linked using the time of cone penetration and / or the depth of the penetrating cone. The time of cone penetration and / or the depth of the penetrating cone are also used to link the image data and the audible sound data with the invisible image described below.

[0054] [Third embodiment] Next, a third embodiment of the present invention will be described. The system according to the third embodiment comprises a boring system 10 similar to that of the first embodiment, a cone penetration test system 20, and a soil constant estimation device 30. At least the image data generation device 21 of the first embodiment is provided inside the cone of the cone penetration test system 20. In addition to this image data generation device 21, the sonic transmitting and receiving device 22 of the second embodiment may also be provided.

[0055] The explanatory variables in the third embodiment include at least three-component data obtained during a cone penetration test and image data relating to an underground image or a soil quality data group representing soil quality classified based on sonic data measured underground. In other words, the explanatory variables in the third embodiment only need to include the three-component data and the soil quality data group, and may also include the image data group described in the first embodiment or the sonic data group described in the second embodiment.

[0056] The soil constant estimation device 30 acquires image data relating to images of the underground using the same method as that described in the first embodiment, and classifies the soil type based on the acquired image data. Alternatively, the soil constant estimation device 30 may not itself perform the soil type classification process, but may instead classify the soil type based on the image data using the judgment of another device or a human, and input soil type data according to the classification results into the soil constant estimation device 30.

[0057] A possible method for classifying soil types based on image data of underground images is, for example, a machine learning method in which image data of underground images is used as an explanatory variable and the result of classifying the underground conditions using a known soil classification method is used as a target variable. However, this is only one example, and other methods may also be used. For example, a method for classifying soil types using the pore water pressure value of the three-component data in addition to the image data described above is also possible.

[0058] The soil constant estimation device 30 acquires sonic data underground using the same method as that described in the second embodiment, and classifies the soil type based on the acquired sonic data. Alternatively, the soil constant estimation device 30 may not itself perform the process of classifying the soil type, but may instead classify the soil type based on the sonic data using another device or human judgment, and input soil data according to the classification results into the soil constant estimation device 30.

[0059] A possible method for classifying soil types based on sonic data measured underground is, for example, a machine learning method in which the sonic data measured underground is used as an explanatory variable and the results of classifying the underground conditions using a known soil classification method are used as a target variable. However, this is only one example, and other methods may also be used. For example, a method for classifying soil types using the pore water pressure value of the three-component data in addition to the sonic data described above is also possible.

[0060] The effects of the third embodiment will now be described with reference to Fig. 7. The inventors of the present application compared soil constants estimated using different explanatory variables at different points A, B, and C with the actual soil constants obtained by boring surveys.

[0061] First, in Case 1 in the table of Figure 7, the three-component data and image data for each location and depth at points A, B, and C in the machine learning stage were combined into a set, along with the N-value and Fc (objective variable) obtained by boring surveys at those locations and depths. A learning model was generated using 75% of these sets as training data, and the remaining 25% of the sets were used to verify the accuracy of the learning model. That is, the cone penetration test data corresponding to the remaining 25% was applied to the learning model stored in the model storage unit 34 to estimate N-values ​​and Fc at locations where N-values ​​and Fc were known. The estimated values ​​were compared with the known N-values ​​and Fc, and the error (difference) was evaluated. In this Case 1, the error in Fc was 6.9%, and the error in N-value was 2.1.

[0062] Next, in Case 2 in the table of Figure 7, at points A and B, the three-component data and image data for each point and depth in the machine learning stage, and the N value and Fc (objective variable) obtained by boring surveys at those points and depths were treated as one set. A learning model was generated using the collection of these sets as training data, and the accuracy of the learning model was verified at point C, an area not yet subjected to machine learning. That is, the cone penetration test data corresponding to point C was applied to the learning model using the data from points A and B stored in the model storage unit 34 to estimate the N value and Fc at point C. The estimated values ​​were compared with the known N value and Fc at point C, and the error (difference) was evaluated. In this case 2, the error in Fc was 30.6%, and the error in the N value was 145.3%.

[0063] In Case 3 in the table of FIG. 7, the three-component data and soil data for each location and depth in the machine learning stage at Points A and B, and the N-value and Fc (objective variable) obtained by boring surveys at those locations and depths were treated as one set. A learning model was generated using the collection of these sets as training data, and the accuracy of the learning model was verified at Point C, an area not yet subjected to machine learning. That is, the cone penetration test data and soil data corresponding to Point C were applied to the learning model using the data from Points A and B stored in the model storage unit 34 to estimate the N-value and Fc at Point C. The estimated values ​​were compared with the known N-value and Fc at Point C, and the error (difference) was evaluated. In this Case 3, the error in Fc was 11.2%, and the error in the N-value was 7.6%.

[0064] From the above, when machine learning is performed using training data at the target location where soil constants are to be estimated (Case 1: data from a location that has already been trained), the accuracy of estimating soil constants is higher than when machine learning is performed using training data at a location other than the target location where soil constants are to be estimated (Case 2: data from a location that has not yet been trained).

[0065] Furthermore, even when machine learning is performed using training data from locations other than the location where the soil constants are to be estimated (Cases 2 and 3: untrained location data), when machine learning is performed using soil data classified based on image data of the location where the soil constants are to be estimated as explanatory variables (Case 3), the accuracy of estimating the soil constants is higher than when machine learning is performed without using soil data classified based on image data of the location where the soil constants are to be estimated as explanatory variables (Case 2).

[0066] In other words, according to the third embodiment, even when machine learning is performed using training data from locations other than the location where the soil constants are to be estimated, it is possible to expect a relatively high level of accuracy in estimating the soil constants.

[0067] [Variations] The above-described embodiment may be modified as follows: Furthermore, one or more of the following modifications may be implemented in combination with the above-described embodiment.

[0068] The estimated soil constants are not limited to the N value or Fc exemplified in the first to third embodiments. The estimated soil constants may be, for example, the density, particle size composition, water content, degree of saturation, deformation characteristic value, strength characteristic, permeability coefficient, etc. of the ground, but are not limited to these examples.

[0069] In the first embodiment, one or more protrusions may be provided on the outer periphery of the cone. This allows the state of deformation or localized damage of the ground when the cone penetrates to be captured in an image, which may result in image data of the underground with more prominent features that contribute to the estimation of soil constants.

[0070] Similarly, in the second embodiment, one or more protrusions may be provided on the outer periphery of the cone, which is thought to result in sonic data with more prominent features that contribute to the estimation of soil constants.

[0071] In the first embodiment, the data related to the image during the cone penetration test used as an explanatory variable for machine learning may be the image data itself as described above, or may be data representing the characteristics of the image data (for example, a histogram of the image). Similarly, in the second embodiment, the sound wave data measured by the sound wave transmitting / receiving device 22 or microphone attached to the cone used as an explanatory variable for machine learning may be the sound wave data itself as described above, or may be data representing the characteristics of the sound wave data (for example, the spectrum of the sound wave). The same applies when image data is used in the second embodiment. The same applies when image data or sound wave data is used in the soil data group used in the third embodiment.

[0072] In the first embodiment, the data related to the image obtained during the cone penetration test used as the explanatory variables for machine learning was visible image data, but invisible image data may also be used. Specifically, an electromagnetic wave transmitting / receiving device is provided inside the cone as the image data generating device 21, and invisible image data is generated based on the measured values ​​of the reflected waves of the electromagnetic waves transmitted by the electromagnetic wave transmitting / receiving device. The electromagnetic waves used here may be at least one of gamma rays, neutron rays, and infrared rays. Considering the characteristics of these wavelengths, soil constants related to the characteristics of invisible image data obtained by gamma rays include the density and water content of the ground. The characteristics of invisible image data obtained by neutron rays are thought to contribute to estimating the soil quality (e.g., clayey soil or sandy soil) and water content. The characteristics of invisible image data obtained by infrared rays are thought to contribute to estimating the presence or absence of buried objects or cavities underground. Note that in the first embodiment, both the visible image data and invisible image data may be used as explanatory variables for machine learning, or both the visible image data and / or invisible image data and sound wave data measured by the sound wave transmitting / receiving device 22 or a sound-collecting microphone may be used. The same applies when image data is used in the second embodiment, and the same applies when image data or sonic data is used in the soil data group used in the third embodiment.

[0073] The learning model used in the first embodiment is preferably a model trained using at least image data in addition to the three-component data of cone penetration resistance, circumferential frictional resistance, and pore water pressure as explanatory variables; the learning model used in the second embodiment is preferably a model trained using at least sonic data in addition to the three-component data of cone penetration resistance, circumferential frictional resistance, and pore water pressure as explanatory variables; and the learning model used in the third embodiment is preferably a model trained using at least soil data in addition to the three-component data of cone penetration resistance, circumferential frictional resistance, and pore water pressure as explanatory variables. The learning model is a machine-learned model using the three-component data of cone penetration resistance, circumferential frictional resistance, and pore water pressure as explanatory variables, as well as image data, sonic data, and soil data, and it is more preferable to use the three-component data, image data, sonic data, and soil data as explanatory variables when estimating soil constants. Furthermore, the learning model used in the first to third embodiments may use at least one of the three component data of the cone penetration resistance value, the peripheral friction resistance value, and the pore water pressure value.

[0074] The present invention may also be a program or a soil constant estimation method executed by the soil constant estimation device according to the first to third embodiments. [Explanation of symbols]

[0075] 1, 1a: System, 10: Boring system, 20: Cone penetration test system, 21: Image data generation device, 22: Ultrasonic wave transmitting and receiving device, 30: Soil constant estimation device, 31: Data acquisition unit, 32: Teacher data generation unit, 33: Model generation unit, 34: Model storage unit, 35: Verification unit, 36: Estimation unit, 3001: Processor, 3002: Memory, 3003: Storage, 3004: Communication device, 3005: Input device, 3006: Output device.

Claims

1. a generation unit that generates a learning model by machine learning using a data group obtained during a cone penetration test, the data group being three-component data, and at least one of an image data group relating to an underground image, an acoustic data group measured by an acoustic wave transmitting / receiving device attached to the cone, or a soil data group classified based on the image data relating to an underground image or the acoustic data measured underground, as explanatory variables, and a soil constant obtained by boring as a response variable; an estimation unit that inputs a group of data corresponding to the explanatory variables in the learning model, which are obtained during a cone penetration test conducted at a location where soil constants are to be estimated, into the generated learning model, and estimates the soil constants at the location where soil constants are to be estimated; Equipped with The soil constant estimation device, wherein the soil constant is strength characteristics, fine particle content, ground density, particle size composition, water content, degree of saturation, deformation characteristic value, or permeability coefficient.

2. The image data relating to the image is visible image data. The soil constant estimation device according to claim 1.

3. The visible image data is data captured by a camera attached to the cone. The soil constant estimation device according to claim 2.

4. A protrusion is provided on the outer periphery of the cone The soil constant estimation device according to claim 3.

5. The image data relating to the image is non-visible image data. The soil constant estimation device according to any one of claims 1 to 4.

6. The invisible image data is invisible image data based on the measurement values ​​of an electromagnetic wave transmitting and receiving device provided on the cone. The soil constant estimation device according to claim 5.

7. The electromagnetic waves transmitted and received by the electromagnetic wave transmitting and receiving device are at least one of gamma rays, neutron rays, and infrared rays. The soil constant estimation device according to claim 6.

8. Computer, The data set obtained during the cone penetration test includes three-component data and at least one subsurface image. Image data group related to the image, sonic data measured by the sonic wave transmitting and receiving device installed in the cone Based on a group of data, or image data relating to underground images or sonic data measured underground The soil data group classified based on the above is used as an explanatory variable, and A learning model is generated by machine learning using the soil constants obtained by the ring as the objective variable. a generation unit; The explanatory variables obtained during the cone penetration test conducted at the estimation target point of the soil constants are A data group corresponding to the explanatory variables in the learning model is used to generate the learning model. an estimation unit that inputs the data into the estimation unit and estimates the soil constants at the estimation target point; and make it work, The soil constants are strength characteristics, fine particle content, ground density, particle size composition, water content, degree of saturation, deformation characteristic values, or permeability coefficient.

9. A soil constant estimation method executed by the soil constant estimation device according to any one of claims 1 to 7.

Images