Trained model, estimation device, estimation method, and method for constructing a trained model

Abstract

This technology provides a method for easily determining the appropriate water content for soil materials during compaction. [Solution] The method for constructing a trained model (S1) includes a first acquisition step (S11) of acquiring image data of a soil material, a second acquisition step (S12) of acquiring parameters that define the shape of the compaction curve of the soil material obtained by analyzing the soil material, and a construction step (S13) of constructing a trained model by machine learning with image data as explanatory variables and parameters as the target variable. The trained model (M) constructed in this way outputs parameters that define the shape of the compaction curve of the soil material when image data of a soil material is input.

Description

[Technical Field]

[0001] This disclosure relates to a trained model, an estimation device, an estimation method, and a method for constructing a trained model. [Background technology]

[0002] Patent Document 1 describes a method for estimating the properties of soil, in which image data of a container obtained by performing a photography process on the soil to be estimated is input to a particle size distribution processing unit, and the particle size distribution processing unit outputs estimated data of the particle size distribution of the soil to be estimated using a particle size distribution estimation model. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2021-117124 [Overview of the Initiative] [Problems that the invention aims to solve]

[0004] To construct high-quality earth structures, it is crucial to appropriately set the water content of the soil material according to its particle size distribution. However, conventionally, there has been no established method for quickly estimating the appropriate water content for compaction on-site for the soil material being used. As a result, compaction of earth structures was sometimes carried out on-site without properly adjusting the water content of the soil material. Such construction methods can lead to a decrease in the quality of the earth structure.

[0005] One aspect of this disclosure aims to provide a technology that can easily determine the appropriate water content ratio for soil material during compaction. [Means for solving the problem]

[0006] To address the above issues, a trained model according to one aspect of this disclosure is constructed to output parameters that define the shape of the compaction curve of soil material when image data of soil material is input.

[0007] Furthermore, estimation devices according to other aspects of this disclosure include: an acquisition unit that acquires image data of a target soil material; a prediction unit that inputs the image data of the target soil material to the trained model and acquires the parameters output by the trained model as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material; and an output unit that outputs information based on the estimated values. The estimation device may be implemented by a processor, in which case a program for an estimation device that implements the estimation device by a processor by operating the processor as each part (software element) of the estimation device, and a processor-readable recording medium on which the program is recorded, also fall within the scope of this disclosure.

[0008] Furthermore, estimation methods relating to other aspects of this disclosure include an acquisition step of acquiring image data of the target soil material, a prediction step of inputting the image data of the target soil material into the trained model and acquiring the parameters output by the trained model as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material, and an output step of outputting information based on the estimated values.

[0009] Furthermore, a method for constructing a trained model according to another aspect of this disclosure includes: a first acquisition step of acquiring image data of a soil material; a second acquisition step of acquiring parameters that define the shape of the compaction curve of the soil material obtained by analyzing the soil material; and a construction step of constructing a trained model by machine learning using the image data and the parameters. [Effects of the Invention]

[0010] According to one aspect of this disclosure, the appropriate water content for soil material during compaction can be easily determined. [Brief explanation of the drawing]

[0011] [Figure 1] This is a block diagram illustrating an example of the configuration and input / output of a trained model according to an embodiment of one aspect of the present disclosure. [Figure 2] This is a diagram for explaining one of the effects of the trained model in Figure 1. [Figure 3] This is a block diagram illustrating an example of the configuration and input / output of a trained model according to a modified example of the same embodiment. [Figure 4] This is a block diagram illustrating an example of the configuration and input / output of a trained model according to a modified example of the same embodiment. [Figure 5] This is a flowchart illustrating an example of the flow of a model construction method. [Figure 6] This is a block diagram illustrating an example of the functional configuration of an estimation device according to an embodiment of another aspect of the present disclosure. [Figure 7] This is a block diagram illustrating an example of the functional configuration of an estimation device according to a modified example of the same embodiment. [Figure 8] This is a diagram for explaining one of the processes executed by the estimation device in Figure 6 or Figure 7. [Figure 9] This is a flowchart illustrating an example of the flow of an estimation method. [Mode for Carrying Out the Invention]

[0012] [Embodiment of Trained Model] Hereinafter, a trained model M according to an embodiment of the present disclosure will be described. Figure 1 is a block diagram illustrating an example of the configuration and input / output of the trained model M.

[0013] (Configuration of Trained Model M) As shown in Figure 1, the trained model M is constructed to output parameters that define the shape of the compaction curve of a soil material when image data of the soil material is input. The "soil materials" covered by this disclosure include natural soil, construction-generated soil, crushed stone, etc. Furthermore, the "soil materials" covered by this disclosure are those that are controlled by the degree of compaction (dry density of soil material / maximum dry density × 100 [%]), and the field density ratio D s or density ratio D c It shall be possible to manage it in this manner. Furthermore, for example, any material that satisfies at least one of the following conditions (1) to (3) may be considered the target "soil material". (1) Materials with a maximum particle size of less than 300 mm (2) Content of fine particles (particles with a particle size of 75 μm or less) F c materials with less than 20% (3) Materials with a slaking rate of 30% or less

[0014] The trained model M according to this embodiment outputs the coefficients α, β, and γ in the following equation (1), which represents the compaction curve (approximated by a function), as parameters. In equation (1), x represents the water content and y represents the dry density. y = -α*(x - β) 2 +γ··(1)

[0015] The trained model M according to this embodiment consists of a first model Ma and a second model Mb.

[0016] • First model Ma The first model Ma is constructed to output intermediate parameters that define a function approximating the particle size distribution of the soil material when image data is input. The first model Ma according to this embodiment outputs coefficients a, b, and c in the following equation (2), which approximates the particle size accumulation curve of the soil material. In the following equation (2), x represents the particle size of the soil and y represents the percentage of mass passing through. y = 0.5 * [tanh(b * Log 10 (a*x))+c]*100··(2)

[0017] It should be noted that the first model Ma may be a conventional model constructed to output only the coefficients a and b in the above formula (2). When the first model Ma is a conventional model, the coefficient c in the above formula (2) can be replaced with a constant (for example, 1).

[0018] Figure 2 is a graph showing the particle size distribution of a certain soil material. The solid line indicates the particle size distribution obtained by actually conducting a particle size test. The dashed line connecting the circle marks (○) indicates the particle size distribution approximated by a function using the outputs of the first model Ma as the coefficients a, b, and c. The dashed line connecting the square marks (□) indicates the particle size distribution approximated by a function in which the outputs of the conventional model are taken as the coefficients a and b, and the term of the coefficient c is replaced with a constant (1). For the graph defined only by the outputs (coefficients a and b) of the conventional model, the distribution of fine particles (particle size less than 75 μm, the region enclosed by the dashed line in Figure 2) deviates from the actual particle size test results. On the other hand, it can be seen that in the graph defined by the outputs (coefficients a, b, c) of the first model Ma, the distribution of fine particles (particle size less than 75 μm) also roughly matches the actual particle size test results. That is, according to the first model Ma, a more accurate particle size distribution can be obtained than with the conventional model.

[0019] · Second model Mb The second model Mb is constructed to output parameters (coefficients α, β, γ) when intermediate parameters (coefficients a, b, c) are input. It should be noted that when the first model outputs only the coefficients a and b in the above formula (2), the second model Mb may be constructed to output parameters when the coefficients a, b, and a constant are input. In addition, the second model Mb may be configured such that at least any one of the following indicators (a) to (d) is input together with the intermediate parameters. (a) Soil particle density ρ s (Test method for density of soil particles JIS A1202) (b) Liquid limit w of soil L , Plastic limit w p , Plasticity index I p (Test method for liquid limit and plastic limit of soil JIS A1205) (c) Particle size information (D 10 , D 30 , D 50 , D 60 , U c , U c ', or F c (e.g., Soil particle size distribution test method JIS A1204, or Soil fine particle content test method JIS A1223) (d) Compaction energy

[0020] (Variations of the trained model M) The above describes a trained model M that consists of a first model Ma and a second model Mb, and outputs parameters. However, the trained model M may also consist of a single model, as shown in Figure 3.

[0021] Furthermore, the trained model M may consist of three or more models. In that case, as shown in Figure 4, the trained model M may consist of, for example, the first model Ma described above, a third model Mc, and a fourth model Md. The third model Mc is a model that, when given intermediate parameters (coefficients a, b, c, or coefficients a, b and a constant), outputs one or two of the parameters (coefficients α, β, γ). The fourth model Md is a model that, when given intermediate parameters and the output of the third model Mc, outputs the remaining parameters (coefficients α, β, γ) that were not output by the third model Mc.

[0022] Furthermore, the trained model M may be configured to output a range of construction moisture content that satisfies the desired degree of compaction when image data is input.

[0023] (Effects of the trained model M) According to the trained model M described above, the compaction curve for the target soil material can be quickly and appropriately determined using the output parameters. Furthermore, by using this compaction curve, the appropriate water content during compaction for the soil material can be easily determined. As a result, the soil material can be properly compacted on-site by adjusting the water content according to the characteristics of the soil material.

[0024] [Embodiment of the method for constructing a pre-trained model] Next, a method S1 for constructing a trained model according to the embodiment of this disclosure (hereinafter referred to as "model construction method S1") will be described. Figure 5 is a flowchart showing an example of the flow of model construction method S1.

[0025] (Model construction method S1 flow) Model construction method S1 is a method for constructing the trained model M described above. As shown in Figure 5, model construction method S1 includes a first acquisition step S11, a second acquisition step S12, and a construction step S13.

[0026] In the initial first acquisition step S11, the acquisition unit acquires image data of the soil material. The acquisition unit includes, for example, a camera, a communication module, a drive, etc. If the acquisition unit consists of a camera, the acquisition unit photographs the target soil material and generates image data. If the acquisition unit consists of a communication module, the acquisition unit receives image data from other devices (not shown). If the acquisition unit consists of a drive, the acquisition unit reads image data from a medium (not shown) on which the image data is stored. Note that the acquisition step S11 may also be a step in which a processor used in the construction step S13, etc., described later, acquires data for calculation.

[0027] After the first acquisition step S11, the process moves to the second acquisition step S12. In the second acquisition step S12, the acquisition unit acquires parameters that define the shape of the compaction curve of the soil material obtained by analyzing the soil material. The acquisition unit includes, for example, a processor connected to an input interface (mouse, keyboard, etc.), a communication module, a drive, etc. If the acquisition unit consists of a processor, the acquisition unit accepts information input by operating the input interface as parameters. If the acquisition unit consists of a communication module, the acquisition unit receives parameters from other devices (not shown). If the acquisition unit consists of a drive, the acquisition unit reads parameters from a medium (not shown) on which the parameters are stored. The second acquisition step S12 may be performed before the first acquisition step S11, or it may be performed in parallel with the first acquisition step S11.

[0028] After the first acquisition step S11 and the second acquisition step S12, the process moves to the construction step S13. In the construction step S13, the processor constructs a trained model M using machine learning with image data and parameters. In the construction step S13 according to this embodiment, the processor first constructs a first model Ma using machine learning with image data as explanatory variables and intermediate parameters as the target variable. The processor then constructs a second model Mb using machine learning with intermediate parameters as explanatory variables and parameters as the target variable. Finally, the trained model M is constructed by merging the first model Ma and the second model Mb. In constructing the second model Mb, at least one of the indicators (a) to (d) above may be used as an explanatory variable along with the intermediate parameters.

[0029] (Variation of model building method S1) The above describes how to construct a pre-trained model M composed of the first model Ma and the second model Mb. However, in construction step S13, the processor may construct a single pre-trained model M by machine learning with image data as explanatory variables and parameters as the target variable.

[0030] Furthermore, in construction step S13, the processor may construct a trained model M consisting of three or more parts by multitasking learning. Specifically, for example, the processor may construct a first model Ma. The processor may then construct a third model Mc by machine learning with intermediate parameters as explanatory variables and one or two of the parameters (coefficients α, β, γ) as the target variable. The processor may then construct a fourth model Md by machine learning with the intermediate parameters and the output of the third model Mc as the target variables, and the remaining parameters (coefficients α, β, γ) that were not output by the third model Mc as the target variables.

[0031] Alternatively, in construction step S13, the processor may construct a trained model M that outputs the range of the construction moisture content when image data is input, using machine learning with image data as the explanatory variable and the range of the construction moisture content as the objective variable.

[0032] (Effects of Model Construction Method S1) According to the model construction method S1 described above, a trained model M can be obtained that outputs parameters for quickly and appropriately determining the compaction curve of the target soil material, or the range of construction moisture content derived from the compaction curve determined by said parameters, using the acquired image data.

[0033] [Embodiment of the Estimation Device] Next, an estimation device 1 according to an embodiment of this disclosure will be described. Figure 6 is a block diagram showing an example of the functional configuration of the estimation device 1.

[0034] (Configuration of Estimation Device 1) As shown in Figure 6, the estimation device 1 comprises an acquisition unit 11, a prediction unit 12, and an output unit 13. The estimation device 1 according to this embodiment further comprises a calculation unit 14. The estimation device 1 according to this embodiment further comprises a storage unit 15. The learned model M is stored in this storage unit 15. That is, in the estimation device 1 according to this embodiment, the learned model M is the configuration of the estimation device 1. Note that the learned model M does not have to be the configuration of the estimation device 1. For example, the estimation device 1 may further comprise a communication unit 16 as shown in Figure 7. The prediction unit 12 may be configured to communicate with another device 2 in which the learned model M is stored via the communication unit 16.

[0035] ·Acquisition part 11 The acquisition unit 11 acquires image data of the target soil material. The acquisition unit 11 consists of a camera, a communication module, a drive, etc. The acquisition unit 11 may be a separate, independent device from the housing that houses the prediction unit 12, which will be described later.

[0036] • Prediction unit 12 The prediction unit 12 inputs image data of the target soil material into the trained model M and obtains the parameters (coefficients α, β, γ) output by the trained model M as estimated values ​​for the parameters that determine the shape of the compaction curve of the target soil material.

[0037] • Calculation unit 14 The calculation unit 14 calculates the range of construction moisture content that satisfies the desired degree of compaction based on the estimated value. In this embodiment, the calculation unit 14 calculates the range of construction moisture content using a conventionally known method. Specifically, the calculation unit 14 determines a function (compactment curve) with the estimated value as a coefficient, as shown in Figure 8, and the value on the vertical axis of its peak (maximum dry density ρ) dmaxThe calculation unit 14 then calculates the minimum (X1) and maximum (X2) construction moisture content values ​​corresponding to the desired degree of compaction, based on the maximum dry density. Specifically, the calculation unit 14 calculates two values ​​on the horizontal axis of the function when the value on the vertical axis is A% of the maximum dry density (where A < 100). "A%" is the management standard value for dry density, which is set to approximately 97% if a higher degree of compaction is desired, and to approximately 92-93% if a lower degree of compaction is desired. These two values ​​are the minimum and maximum construction moisture content values, as well as the lower and upper limits of the construction moisture content range. Note that if the trained model M is configured to output a range of construction moisture content values, the estimation device 1 does not need to have the calculation unit 14.

[0038] Output section 13 The output unit 13 outputs information based on estimated values. In this embodiment, the output unit 13 outputs the range of the construction moisture content as information based on estimated values. The output unit 13 may also be configured to output the parameters (coefficients α, β, γ) themselves. The output unit 13 is also composed of a display device, a communication module, etc. If the output unit 13 is composed of a display device, the output unit 13 displays information based on estimated values. If the output unit 13 is composed of a communication module, the output unit 13 transmits information based on estimated values ​​in the form of data or signals to other devices not shown in the diagram. The output unit 13 may be a separate and independent device from the housing that houses the prediction unit 12.

[0039] (Example of implementation using software for estimation devices) The function of the estimation device 1 is a program that causes the processor to function as the estimation device 1, and this can be realized by a program that causes the processor to function as each of the control blocks 11 to 14 of the estimation device 1.

[0040] In this case, the estimation device 1 includes a processor having at least one control device (e.g., a processor) and at least one storage device (e.g., memory) as hardware for executing the program. By executing the program using this control device and storage device, each of the functions described in the above embodiment is realized.

[0041] The above program may be recorded on one or more processor-readable recording media, rather than temporary ones. This recording media may or may not be provided by the estimation device 1. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0042] Furthermore, some or all of the functions of each of the above control blocks can also be implemented by logic circuits. For example, an integrated circuit in which logic circuits functioning as each of the above control blocks are formed is also included in the scope of this disclosure. In addition, it is also possible to implement the functions of each of the above control blocks by, for example, a quantum processor.

[0043] (Effects of Estimation Device 1) According to the estimation device 1 described above, the appropriate water content for compaction can be quickly and accurately determined from the image data of the target soil material. As a result, the soil material can be properly compacted on-site by adjusting the water content according to the characteristics of the soil material.

[0044] [Embodiment of Estimation Method] Next, the estimation method S2 according to the embodiment of this disclosure will be described. Figure 9 is a flowchart showing an example of the flow of the estimation method S2.

[0045] (Flow of estimation method S2) As shown in Figure 9, the estimation method S2 includes an acquisition step S21, a prediction step S22, and an output step S23. The estimation method S2 according to this embodiment further includes a calculation step S24.

[0046] • Acquisition step S21 In the initial acquisition step S21, the acquisition unit acquires image data of the target soil material. The acquisition unit may be the acquisition unit 11 (camera, communication module, drive, etc.) provided in the estimation device 1, or it may be a different unit. Alternatively, the acquisition step S21 may be a step in which a processor used in the prediction step S22, etc., described later, acquires image data for calculation.

[0047] • Prediction step S22 In prediction step S22, the processor inputs image data of the target soil material into the trained model M. The processor then obtains the parameters output by the trained model M as estimated values ​​for the parameters that determine the shape of the compaction curve of the target soil material. The processor that inputs the image into the trained model M and obtains the estimated values ​​may be one provided by the estimation device 1, or it may be a different processor. If the trained model M is configured to output a range of construction moisture content, in prediction step S22, the processor may obtain the numerical range output by the trained model M as an estimated value for the range of construction moisture content.

[0048] • Calculation step S24 After the prediction step S22, the process moves to the calculation step S24. In the calculation step S24, the processor calculates the range of construction moisture content that satisfies the desired degree of compaction based on the estimated values. The processor that calculates the range of construction moisture content may be the same as the processor used in steps S21 and S22 above, or it may be a different processor. Note that if the trained model M is constructed to output a range of construction moisture content, the estimation method S2 does not need to include the calculation step S24.

[0049] Output step S23 After the prediction step S22 or calculation step S24, the process moves to the output step S23. In the output step S23, the output unit outputs information based on the estimated values. The output unit may be the output unit 13 (display device, communication module, etc.) provided in the estimation device 1, or it may be something different. Alternatively, the output step S23 may be a step in which the processor used in the prediction step S22 or calculation step S24 controls the output unit so that it outputs information based on the estimated values.

[0050] (Effects of estimation method S2) According to the estimation method S2 described above, similar to the estimation device 1, the appropriate water content for compaction of the target soil material can be quickly and accurately determined from the image data of the target soil material. As a result, the soil material can be properly compacted on-site by adjusting the water content according to the characteristics of the soil material.

[0051] [Variation] This disclosure is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of this disclosure.

[0052] 〔summary〕 The trained model according to aspect 1 of this disclosure is configured to output parameters that define the shape of the compaction curve of soil material when image data of soil material is input.

[0053] The trained model according to aspect 2 of this disclosure is, in aspect 1 above, the parameter of the equation representing the compaction curve y = -α*(x-β) 2 The configuration may also output the coefficients α, β, and γ in +γ.

[0054] The trained model according to aspect 3 of this disclosure may be configured as follows: a first model constructed to output intermediate parameters that define a function approximating the particle size distribution of the soil material when image data is input, and a second model constructed to output the parameters when the intermediate parameters are input.

[0055] The trained model according to aspect 4 of this disclosure is, in aspect 3 above, the first model approximates the particle size distribution curve of the soil material with the equation y = 0.5 * [tanh(b * Log 10 The configuration may also output the coefficients a, b, and c in (a*x))+c]*100.

[0056] The estimation device according to aspect 5 of the present disclosure may be configured to include, in any one of aspects 1 to 3 above, an acquisition unit that acquires image data of the target soil material, a prediction unit that inputs the image data of the target soil material to the trained model and acquires the parameters output by the trained model as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material, and an output unit that outputs information based on the estimated values.

[0057] The estimation device according to embodiment 6 of the present disclosure may further include a calculation unit that calculates a range of construction moisture content that satisfies a desired degree of compaction based on the estimated value, and the output unit may output the range.

[0058] The estimation method according to aspect 7 of the present disclosure may also include, in any one of aspects 1 to 6 above, an acquisition step of acquiring image data of the target soil material; a prediction step of inputting the image data of the target soil material into the trained model and acquiring the parameters output by the trained model as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material; and an output step of outputting information based on the estimated values.

[0059] A method for constructing a trained model according to aspect 8 of this disclosure includes a first acquisition step of acquiring image data of a soil material; a second acquisition step of acquiring parameters that define the shape of the compaction curve of the soil material obtained by analyzing the soil material; and a construction step of constructing a trained model by machine learning using the image data and the parameters.

[0060] The trained model according to aspect 9 of this disclosure is configured to output parameters that determine the shape of the compaction curve of a soil material when intermediate parameters that define a function approximating the particle size distribution of the soil material are input. [Explanation of Symbols]

[0061] M Trained Model Ma First Model Mb 2nd Model Mc's third model Md, the fourth model S1: How to build a pre-trained model S11 First Acquisition Step S12 Second Acquisition Step S13 Construction Steps 1 Estimation device 11 Acquisition Department 12 Prediction Section 13 Output section 14 Calculation Section 15 Storage section 16 Communications Department 2 Other devices S2 estimation method S21 Acquisition Steps S22 Prediction Step S23 Output Step S24 Calculation Step

Claims

1. A pre-trained model that, upon receiving image data of soil material, outputs parameters that define the shape of the compaction curve for that soil material.

2. The parameter is the following equation (1) which represents the compaction curve: y=-α*(+-β) ２ +c・・(1) Output the coefficients α, β, and γ in the following: The trained model described in claim 1.

3. A first model is constructed to output intermediate parameters that define a function approximating the particle size distribution of the soil material when the data of the aforementioned image is input, A second model constructed to output the aforementioned parameters when the aforementioned intermediate parameters are input, It consists of, The trained model described in claim 1.

4. The first model approximates the particle size distribution curve of the soil material using the following equation (2): y=0.5*[tanh(b*Log] １０ (a*x))+c]*100・・(2) Output the coefficients a, b, and c in the following: The trained model described in claim 3.

5. An acquisition unit that acquires image data of the target soil material, A prediction unit that inputs image data of the target soil material into a trained model according to any one of claims 1 to 3, and acquires the parameters output by the trained model as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material, An output unit that outputs information based on the estimated value, An estimation device equipped with the following features.

6. The system further includes a calculation unit that calculates a range of construction moisture content that satisfies the desired degree of compaction based on the aforementioned estimated values. The output unit outputs the range, The estimation device according to claim 5.

7. The acquisition step involves obtaining image data of the target soil material, A prediction step in which data of an image showing the target soil material is input to the trained model described in claim 1, and the parameters output by the trained model are obtained as estimated values ​​of parameters that determine the shape of the compaction curve of the target soil material, An output step that outputs information based on the estimated value, An estimation method that includes [this].

8. The first acquisition step involves obtaining image data of the soil material, A second acquisition step involves analyzing the soil material and obtaining parameters that define the shape of the compaction curve of the soil material, A construction step to construct a trained model by machine learning using the aforementioned image data and parameters, including How to build a pre-trained model.

9. A trained model constructed to output parameters that determine the shape of the compaction curve for a given soil material, when given intermediate parameters that define a function approximating the particle size distribution of that soil material.

Images