Data structure, learning method, and estimation method

Abstract

To easily configure an estimation system of construction conditions of a pile by constructing a machine learning model based on information of a histogram stored in an existing database.SOLUTION: The data structure includes a record for each of a plurality of depths, the record including a set of information including position information indicating a position of a construction site of a pile, depth-dependent information that is information about a stratum that changes in accordance with a change in the depth at the construction site, depth-independent information that is information not dependent on the depth and that includes at least information about specifications of the pile, and depth information indicating a depth of the depth-dependent information.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a data structure, a learning method, and an estimation method. [Background technology]

[0002] BACKGROUND ART Conventionally, techniques have been disclosed for estimating the distribution of geological strata and geological properties required for foundation work on a building (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2022-146889 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the conventional technologies described above, specialized information such as images and sound waves is used for learning to estimate the distribution of strata and geological properties, and there was a problem in that the system could not be easily constructed.

[0005] The present invention has been made to solve the above problems, and its purpose is to easily construct a system for estimating pile construction conditions by building a machine learning model based on information on columnar diagrams and construction data stored in existing databases. [Means for solving the problem]

[0006] One embodiment of the present invention is a data structure in which a set of information consisting of location information indicating the location of the pile construction site, depth-dependent information which is information about the geological layer that changes depending on changes in depth at the construction site, depth-independent information which is information that is not dependent on the depth and includes at least information about the pile specifications, and depth information which indicates the depth for which the depth-dependent information is information is configured as a record for each of the depths, and which has the records for multiple of the depths.

[0007] One embodiment of the present invention is a learning method that uses any of the information contained in the above-mentioned data structure as a target variable, uses at least a portion of the other information as an explanatory variable, and generates a trained model trained using the data structure.

[0008] One embodiment of the present invention is an estimation method in which the soil quality, which is the objective variable output from the above-mentioned first trained model, is output as an estimated soil quality, and the estimated soil quality is given to the second trained model as an explanatory variable in addition to the depth information and the location information, thereby outputting the N value, which is the objective variable output from the second trained model, as an estimated N value. [Effects of the Invention]

[0009] According to this invention, a system for estimating pile construction conditions can be easily constructed by constructing a machine learning model based on information on columnar diagrams and construction data stored in an existing database. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 2 is a diagram illustrating an example of a functional configuration of a learning control device according to the present embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the flow of operations of the learning control device of the present embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of a data structure of past data according to the present embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of a data structure of a torque estimation model according to the present embodiment. [Figure 5] FIG. 4 is a diagram illustrating an example of a data structure of a rotation speed estimation model according to the present embodiment. [Figure 6] FIG. 2 is a diagram illustrating an example of a data structure of a speed estimation model according to the present embodiment. [Figure 7] FIG. 2 is a diagram illustrating an example of a functional configuration of the estimation device according to the present embodiment. [Figure 8] FIG. 2 is a diagram illustrating an example of a configuration of on-site data according to the present embodiment. [Figure 9] FIG. 2 is a diagram illustrating an example of the flow of operations of the estimation device according to the present embodiment. [Figure 10] FIG. 4 is a diagram illustrating an example of a data structure of a torque estimation model in an inference stage of the present embodiment. [Figure 11] FIG. 2 is a diagram illustrating an example of a data structure of a rotation speed estimation model in an inference stage of the present embodiment. [Figure 12] FIG. 2 is a diagram illustrating an example of a data structure of a speed estimation model in the inference stage of the present embodiment. [Figure 13] FIG. 2 is a diagram schematically illustrating a flow of estimation of a response variable according to the present embodiment. [Figure 14] FIG. 10 is a diagram schematically illustrating the flow of a model learning stage in a modified example of the present embodiment. [Figure 15] FIG. 10 is a diagram illustrating an example of a data structure of a torque estimation model in a modified example. [Figure 16] FIG. 10 is a diagram illustrating an example of a data structure of a rotation speed estimation model in a modified example. [Figure 17] FIG. 10 is a diagram illustrating an example of a data structure of a speed estimation model in a modified example. [Figure 18] FIG. 2 is a diagram illustrating an example of the flow of operations of the learning control device of the present embodiment. [Figure 19] FIG. 2 is a diagram illustrating an example of a data structure of a soil property estimation model according to the present embodiment. [Figure 20] FIG. 2 is a diagram illustrating an example of a data structure of an N-value estimation model according to the present embodiment. [Figure 21] FIG. 4 is a diagram illustrating an example of a data structure of a torque estimation model according to the present embodiment. [Figure 22] FIG. 4 is a diagram illustrating an example of a data structure of a rotation speed estimation model according to the present embodiment. [Figure 23] FIG. 2 is a diagram illustrating an example of a data structure of a speed estimation model according to the present embodiment. [Figure 24] FIG. 2 is a diagram illustrating an example of the flow of operations of the learning control device of the present embodiment. [Figure 25] FIG. 2 is a diagram illustrating an example of a data structure of a speed estimation model according to the present embodiment. [Figure 26] FIG. 1 is a diagram showing an example of a log obtained by a conventional boring test. DETAILED DESCRIPTION OF THE INVENTION

[0011] [Overview of the construction condition estimation system] Traditionally, when evaluating the ground at a proposed construction site for a building, boring tests have been conducted to create columnar diagrams showing the correspondence between depth and soil type and N-value for each depth. FIG. 26 is a diagram showing an example of a log obtained by a conventional boring test.

[0012] On the other hand, it is preferable to be able to predict the soil quality and N value at the planned building construction site (hereinafter referred to as the target location or the target position) before conducting such a boring test (or without conducting a boring test). Furthermore, it is also possible to use soil quality information related to soil quality, without being limited to this.

[0013] [First embodiment] The construction condition estimation system 1 of this embodiment is a system that predicts stratum data and construction conditions at a target location by providing location information of the target location, and columnar data and a desired depth as soil information. The soil information is not limited to columnar data.

[0014] In one example of this embodiment, it is also possible to estimate information regarding the operation of a construction device that performs pile construction. Information regarding the operation of the construction device refers, for example, to the torque that rotates the pile, the number of rotations per unit time, and the excavation speed, which is the speed at which the pile moves in the depth direction, if the pile to be constructed is a steel pipe pile with spiral blades at its tip that penetrates into the ground using the propulsion force of the blades caused by the rotation of the pile.

[0015] The construction condition estimation system 1 of this embodiment can be used at the design stage or material ordering stage of pile construction, that is, the pre-construction stage, or the on-site pile construction stage. In the following description, the pile to be constructed is a steel pipe pile having a spiral blade at its tip and penetrating into the ground by the propulsive force of the blade caused by the rotation of the pile, but the construction condition estimation system 1 can handle any type of pile, for example, a steel pipe pile or a concrete pile.

[0016] The construction condition estimation system 1 includes a learning control device 10 used in the learning stage of logarithm information and construction data, and an estimation device 40 used in the estimation stage of information related to soil properties, N values, and operation of construction equipment. First, the functional configuration of the learning control device 10 will be described with reference to FIG.

[0017] [Functional configuration of the learning control device 10] 1 is a diagram showing an example of the functional configuration of a learning control device 10 according to this embodiment. The learning control device 10 is a computer device that operates based on a program. The learning control device 10 includes a past data acquisition unit 110 and a learning control unit 120 as its functional units.

[0018] FIG. 2 is a diagram showing an example of the flow of operations of the learning control device 10 of this embodiment. (Step S10) The past data acquisition unit 110 acquires past data D1 from the past data storage device 21. The past data D1 is a collective term for various information such as boring tests conducted in various places in the past, various measured values ​​obtained by construction of piles, and design values ​​of driven piles.

[0019] The learning control unit 120 uses the past data D1 to acquire information on the position (e.g., latitude, longitude, and altitude) as depth-independent information. The learning control unit 120 acquires information on soil type for each excavation depth (e.g., every 10 cm) as this depth-dependent information. The learning control unit 120 may also acquire information on soil type for each excavation depth by extracting the soil type information from the log information included in the past data D1.

[0020] 3 is a diagram showing an example of the data structure of the past data D1 of this embodiment. As an example, the past data D1 includes the following items: depth, position, columnar diagram, pile specifications, and construction data. In the following diagrams, the symbol "-" indicates that a field is essentially blank (so-called a null value). Depth indicates the excavation depth at the pile construction site. Position indicates the two-dimensional or three-dimensional coordinates of the pile construction site. For example, position is information indicated by three-dimensional coordinates of latitude, longitude, and altitude using a GPS (global positioning system). As mentioned above, the log is information indicating the soil type and N value at each depth. Pile specifications are information that indicates the specifications (machine type) of the heavy equipment (hereinafter also referred to as construction machine) that constructs the piles, as well as the shape and dimensions of the constructed piles. Pile specifications have various items depending on the type of pile. For example, in the case of a pile with spiral blades, the pile specifications include the blade angle of attack, blade shape, pile tip shape (e.g., closed type or open type), blade pitch, number of blade threads, blade material, etc. The construction data are the operating specifications of the construction machine during pile construction, and may include, for example, torque during construction, number of revolutions per unit time, and excavation speed, depending on the type of pile.

[0021] The past data storage device 21 stores past data D1 having a data structure in which various data acquired in the past are divided into records for each depth. The unit depth is determined by the sampling unit used by the construction machine to acquire data during construction. The unit depth depends on the depth reached, but is, for example, a value in the range of 5 cm to 50 cm, and is typically 10 cm.

[0022] The past data D1 includes information based on the logarithm and information based on the construction data. Originally, the information based on the logarithm and the information based on the construction data are separate pieces of information that are not linked to each other's depths. The past data D1 is created by mapping the information based on the logarithm and the information based on the construction data to the above-mentioned unit depths, respectively.

[0023] Here, among the past data D1, the position and pile specifications are information that do not change with changes in excavation depth (i.e., information that does not depend on depth), whereas among the past data D1, the columnar diagram and construction data are information that change with changes in excavation depth (i.e., information that depends on depth). The past data D1 is configured as a record for each depth, with a set of information that is dependent on depth (depth-dependent information) and information that is not dependent on depth (depth-independent information).

[0024] In other words, the past data D1 is composed of a data structure having records for depth, in which sets of information are configured as records for each depth: location information indicating the location of the pile construction site; depth-dependent information, which is information about the geological strata that changes depending on changes in depth at the construction site; depth-independent information, which is information that is not dependent on depth and includes at least information on the pile specifications; and depth information, which indicates the depth for which the depth-dependent information is information. The past data D1 has the above-mentioned data structure for each construction site. That is, the past data D1 includes information about multiple sites. The order of the records is arbitrary, and they may be sorted by construction site or depth, and are not limited to a specific classification method.

[0025] In addition, in the past data D1, the depth-dependent information includes at least one of the soil type of the stratum at the depth indicated by the depth information included in the record, or the N value at that depth.

[0026] In addition, in the past data D1, the depth-dependent information includes at least one of the torque applied to the pile when constructing the pile at the depth indicated by the depth information included in the record, the number of rotations of the pile per unit time, or the excavation speed due to the rotation of the pile.

[0027] In the figure, the past data D1 may have an estimated value item (not shown). In this case, all the past data D1 have estimated value items, but the data structure of the past data D1 is not limited to this. The past data D1 does not have to have estimated value items.

[0028] The learning control device 10 of this embodiment performs learning using a data structure in which, like the above-mentioned past data D1, a set of location information, depth-dependent information, depth-independent information, and seismic intensity information indicating the depth at which the depth-dependent information is information is organized as a record for each depth. For example, the learning control device 10 uses any of the information included in the above-mentioned data structure as a target variable and at least some of the other information as explanatory variables to generate a trained model trained using this data structure.

[0029] Returning to FIG. 1, the past data acquisition unit 110 outputs the acquired past data D1 to the learning control unit 120. The learning control unit 120 executes model learning based on the past data D1 acquired in step S10 shown in FIG. 2. In this embodiment, the learning models to be learned include a torque estimation model M10, a rotation speed estimation model M11, and a speed estimation model M12. The learning control unit 120 sequentially learns each of these models based on the past data D1. The specific flow of the learning procedure will be described below.

[0030] In the following description, the symbol EV denotes the explanatory variable of the model, the symbol OV denotes the objective variable of the model, and the symbol IV denotes the estimation result (estimated value) by the trained model. The suffixes (a, b, c, d, ...) at the end of the symbols EVx, OVx, and IVx (x is an arbitrary number) are identifiers that identify the datasets. For example, the symbols EVxa and EVxb indicate that they are different datasets.

[0031] (Step S20) Returning to Fig. 2, the learning control unit 120 trains a torque estimation model M10. The torque estimation model M10 is a model in which the correspondence with the torque of the construction machine is learned using past data D1 for multiple sites and depths at the same site.

[0032] The learning control unit 120 acquires information on the location of the construction site (e.g., latitude, longitude, and altitude), machine type, blade diameter, and blade thickness as learning targets for the torque estimation model M10. The learning control unit 120 acquires information on the soil type, N value, and torque for each excavation depth (e.g., every 10 cm) as depth-dependent information for the construction site.

[0033] 4 is a diagram showing an example of the data structure of the torque estimation model M10 of this embodiment. In this example, the torque estimation model M10 has a data structure in which the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N value EV6a, machine type EV7a, blade diameter EV8a, and blade thickness EV9a are used as explanatory variables, and the torque OV10a is used as a response variable. In other words, the torque estimation model M10 is a model in which the correspondence between the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N value EV6a, machine type EV7a, blade diameter EV8a, and blade thickness EV9a and the torque OV10a are learned.

[0034] As shown in the inference stage of the same figure, when the excavation depth EV1b, latitude EV2b, longitude EV3b, altitude EV4b, soil type EV5b, N value EV6b, machine type EV7b, blade diameter EV8b, and blade thickness EV9b are given as input values ​​to the trained torque estimation model M10, it outputs a torque estimation value IV10b.

[0035] According to the torque estimation model M10 having the data structure configured in this manner, by specifying the depth, location, soil type, N value and pile specifications of a construction site, it is possible to estimate the torque value for each width (e.g., 10 cm) corresponding to the depth of the construction site, and by repeating the inference for each depth up to the desired final depth, the torque up to the final depth can be obtained.

[0036] It should be noted that the information indicating the position of the construction site (latitude EV2a, longitude EV3a, altitude EV4a) described above is not essential for learning the torque estimation model M10.

[0037] (Step S30) Returning to Fig. 2, the learning control unit 120 trains a rotation speed estimation model M11. The rotation speed estimation model M11 is a model in which a correspondence relationship with the rotation speed per unit time of the construction machine is learned using past data D1.

[0038] The learning control unit 120 acquires information on the location of the construction site (e.g., latitude, longitude, altitude), machine type, blade diameter, and blade thickness as depth-independent information of the construction site selected as the learning target for the rotation speed estimation model M11. The learning control unit 120 acquires information on the soil type, N value, torque, and rotation speed for each excavation depth (e.g., every 10 cm) as depth-dependent information of the construction site.

[0039] 5 is a diagram showing an example of the data structure of the rotation speed estimation model M11 of this embodiment. In this example, the rotation speed estimation model M11 has a data structure in which the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N value EV6a, machine type EV7a, blade diameter EV8a, blade thickness EV9a, and torque EV10a are used as explanatory variables, and the rotation speed OV11a is used as a response variable. In other words, the rotation speed estimation model M11 is a model in which the correspondence between the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N value EV6a, machine type EV7a, blade diameter EV8a, blade thickness EV9a, and torque EV10a and the rotation speed OV11a is learned.

[0040] As shown in the inference stage of the same figure, when the drilling depth EV1b, latitude EV2b, longitude EV3b, altitude EV4b, soil type EV5b, N value EV6b, machine type EV7b, blade diameter EV8b, blade thickness EV9b, and torque EV10b are given as input values ​​to the trained rotation speed estimation model M11, it outputs the rotation speed estimation value IV11b.

[0041] According to the rotation speed estimation model M11 having the data structure configured in this way, by specifying the depth, position, soil type, N value, pile specifications and torque of a certain construction site, it is possible to estimate the rotation speed value for each width (e.g., 10 cm) corresponding to the depth of the construction site, and by repeating this inference for each depth up to the desired final depth, the rotation speed up to the final depth can be obtained.

[0042] It should be noted that the information indicating the position of the construction site (latitude EV2a, longitude EV3a, altitude EV4a) is not essential for learning the rotation speed estimation model M11.

[0043] (Step S40) Returning to Fig. 2, the learning control unit 120 trains the speed estimation model M12. The speed estimation model M12 is a model in which the correspondence with the excavation speed of the construction machine is learned using the past data D1.

[0044] The learning control unit 120 acquires information on the location of the construction site (e.g., latitude, longitude, altitude), machine type, blade diameter, and blade thickness as depth-independent information on the construction site selected as the learning target for the speed estimation model M12. The learning control unit 120 acquires information on the soil type, N value, torque, rotation speed, and speed for each excavation depth (e.g., every 10 cm) as depth-dependent information on the construction site.

[0045] 6 is a diagram showing an example of the data structure of the speed estimation model M12 of this embodiment. In this example, the speed estimation model M12 has a data structure in which the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N-value EV6a, machine type EV7a, blade diameter EV8a, blade thickness EV9a, torque EV10a, and rotation speed EV11a are used as explanatory variables, and the speed OV12a is used as a response variable. In other words, the rotation speed estimation model M11 is a model in which the correspondence between the excavation depth EV1a, latitude EV2a, longitude EV3a, altitude EV4a, soil type EV5a, N-value EV6a, machine type EV7a, blade diameter EV8a, blade thickness EV9a, torque EV10a, and rotation speed EV11a and the speed OV12a is learned.

[0046] As shown in the inference stage of the same figure, when the excavation depth EV1b, latitude EV2b, longitude EV3b, altitude EV4b, soil type EV5b, N value EV6b, machine type EV7b, blade diameter EV8b, blade thickness EV9b, torque EV10b, and rotation speed EV11b are given as input values ​​to the trained speed estimation model M12, it outputs a speed estimation value IV12b.

[0047] According to the speed estimation model M12 having the data structure configured in this way, it is possible to estimate the speed value for each depth of a construction site by specifying the depth, position, soil type, N value, pile specifications, torque, and rotation speed of the construction site. Since it is possible to estimate the speed value for each width (for example, 10 cm) corresponding to the depth of the construction site, the speed up to the desired final depth can be obtained by repeating the inference for each depth.

[0048] It should be noted that the information indicating the position of the construction site (latitude EV2a, longitude EV3a, altitude EV4a) described above is not essential for learning the speed estimation model M12.

[0049] The above has described the learning procedures for the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12. Next, we will describe the configuration of the estimation device 40 that estimates the construction speed using these torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12.

[0050] [Functional configuration of the estimation device 40] 7 is a diagram showing an example of the functional configuration of the estimation device 40 of this embodiment. The estimation device 40 is a computer device that operates based on a program. The estimation device 40 includes a site information acquisition unit 410, an estimation unit 420, and an output unit 430 as its functional units. The estimation device 40 can be used at each stage, such as the design stage before the construction of a building, or the construction stage during the construction of a building.

[0051] As an example, when the estimating device 40 is used in the design stage of a building, the estimating device 40 acquires information such as information indicating the planned construction location of the building (for example, the construction site), planned construction location information including information on a logarithm at that location, and specifications of piles to be constructed from a device that stores design information of the building. In this example, the planned construction location of the building and specifications of piles to be constructed are also referred to as site data D2, and the device that stores the design information of the building is also referred to as site data supply device 22. That is, the estimation device 40 acquires the field data D2 from the field data supply device 22.

[0052] FIG. 8 is a diagram showing an example of the configuration of the site data D2 of this embodiment. As an example, the site data D2 includes the following items: depth, position, logarithm (soil quality, N value), and pile specifications. The above-mentioned past data D1 is data about multiple properties (e.g., property L1) that have been constructed in the past, whereas the site data D2 is data about multiple properties (e.g., property L2) that are scheduled to be constructed in the future (i.e., not yet constructed), which is a difference between the two. The details of each item are the same as those of the past data D1, and therefore will not be described here.

[0053] The estimation device 40 can be used in various situations, such as the planning stage before construction of a building, or the stage during construction of a building (construction stage). When the estimation device 40 is used in the construction stage, the estimation device 40 may be configured to acquire information on the construction position of the building, specifications of piles to be constructed or already constructed, etc., from a device that stores the operating status of a pile construction machine. In this example, the construction position of the building, specifications of piles to be constructed or already constructed, etc. are also referred to as site data D2, and the device that stores the operating status of the construction machine corresponds to the site data supply device 22.

[0054] FIG. 9 is a diagram showing an example of the flow of operations of the estimation device 40 of this embodiment. (Step S50) The site information acquisition unit 410 acquires the site data D2 from the site data supply device 22. The site information acquisition unit 410 outputs the acquired site data D2 to the estimation unit 420.

[0055] (Step S60) The estimation unit 420 reads the site data D2 acquired in step S50 and provides it to the trained model from the trained model storage unit 30. As described above, trained models such as the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12 are stored in the trained model storage unit 30. When information such as the depth, position, columnar diagram, pile specifications, and construction data contained in the site data D2 is provided as explanatory variables, these trained models output a target variable according to the trained model.

[0056] 10 is a diagram showing an example of the data structure of the torque estimation model M10 in the inference stage of this embodiment. The estimation unit 420 assigns to the torque estimation model M10 the following items of the site data D2: excavation depth EV1c, latitude EV2c, longitude EV3c, altitude EV4c, soil type EV5c, N value EV6c, machine type EV7c, blade diameter EV8c, and blade thickness EV9c.

[0057] In addition, if information indicating the location of the construction site (latitude, longitude, altitude) is not included in the explanatory variables in the learning of the torque estimation model M10, it is not necessary to provide the latitude EV2c, longitude EV3c, and altitude EV4c in the inference stage.

[0058] When the site data D2 is given, the torque estimation model M10 outputs a torque estimate IV10c.

[0059] 11 is a diagram showing an example of the data structure of the rotation speed estimation model M11 in the inference stage of this embodiment. The estimation unit 420 provides the rotation speed estimation model M11 with the following items of the job site data D2: excavation depth EV1c, latitude EV2c, longitude EV3c, altitude EV4c, soil type EV5c, N value EV6c, machine type EV7c, blade diameter EV8c, and blade thickness EV9c, as well as the torque estimation value IV10c output by the torque estimation model M10.

[0060] Note that if information indicating the location of the construction site (latitude, longitude, altitude) is not included in the explanatory variables in the learning of the rotation speed estimation model M11, it is not necessary to provide the latitude EV2c, longitude EV3c, and altitude EV4c in the inference stage.

[0061] When these field data D2 are given, the rotation speed estimation model M11 outputs the rotation speed estimated value IV11c.

[0062] 12 is a diagram showing an example of the data structure of the speed estimation model M12 in the inference stage of this embodiment. The estimation unit 420 provides the speed estimation model M12 with the following items of the job site data D2: excavation depth EV1c, latitude EV2c, longitude EV3c, altitude EV4c, soil type EV5c, N value EV6c, machine type EV7c, blade diameter EV8c, blade thickness EV9c, torque estimated value IV10c output by the torque estimation model M10, and rotation speed estimated value IV11c output by the rotation speed estimation model M11.

[0063] In addition, if information indicating the location of the construction site (latitude, longitude, altitude) is not included in the explanatory variables in the learning of the speed estimation model M12, it is not necessary to provide the latitude EV2c, longitude EV3c, and altitude EV4c in the inference stage.

[0064] Given these field data D2, the speed estimation model M12 outputs a speed estimate IV12c.

[0065] That is, the estimation unit 420 estimates the final objective variable (for example, the speed estimated value IV12c) using the various learned models described above. The output unit 430 acquires the objective variables output by the trained model and outputs them to the output device 50 (for example, a liquid crystal display or a printer device).

[0066] 13 is a diagram schematically illustrating the flow of estimation of a response variable according to this embodiment. The estimation unit 420 of this embodiment provides the on-site data D2 to the torque estimation model M10, thereby causing the torque estimation model M10 to output a torque estimation value IV10c. The estimation unit 420 provides the on-site data D2 and the torque estimation value IV10c output by the torque estimation model M10 to the rotation speed estimation model M11, thereby causing the rotation speed estimation model M11 to output a rotation speed estimation value IV11c. The estimation unit 420 provides the on-site data D2, the torque estimation value IV10c output by the torque estimation model M10, and the rotation speed estimation value IV11c output by the rotation speed estimation model M11 to the speed estimation model M12, thereby causing the speed estimation model M12 to output a speed estimation value IV12c.

[0067] As described above, the site data D2 is data about a property L2 that is scheduled for construction in the future (i.e., not yet constructed), and therefore does not include construction data such as torque, rotation speed, or speed (i.e., data obtained when construction is performed). According to the estimation device 40 configured as described above, it is possible to obtain a speed estimate from the site data D2 that does not include construction data. Because the speed estimate is a value that indicates the construction speed, obtaining the speed estimate makes it possible to calculate the time or number of days required for construction. Of course, it is also possible to obtain a torque estimate and a rotation speed estimate. That is, the estimation device 40 of this embodiment can calculate the time and number of days required for construction based on the site data D2 that does not include construction data. Therefore, the estimation device 40 of this embodiment can, for example, create a construction schedule at a pre-construction stage.

[0068] [Modification of the first embodiment] 14 is a diagram schematically illustrating the flow of the model learning stage in a modification of this embodiment. In the learning stage of the first embodiment described above, past data D1 is provided to the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12, and each model is individually trained. In this modification, in addition to the past data D1, the torque estimated value output by the torque estimation model M10a is provided to the rotation speed estimation model M11a, thereby training the rotation speed estimation model M11a. Furthermore, in addition to the past data D1, the torque estimated value output by the torque estimation model M10a and the rotation speed estimated value output by the rotation speed estimation model M11a are provided to the speed estimation model M12a, thereby training the speed estimation model M12a.

[0069] 15 is a diagram showing an example of the data structure of a torque estimation model M10a in a modified example. In this example, the torque estimation model M10a has a data structure in which excavation depth EV1d, latitude EV2d, longitude EV3d, altitude EV4d, soil type EV5d, N value EV6d, machine type EV7d, blade diameter EV8d, and blade thickness EV9d are used as explanatory variables, and torque OV10d is used as a response variable. In other words, the torque estimation model M10a is a model in which the correspondence between excavation depth EV1d, latitude EV2d, longitude EV3d, altitude EV4d, soil type EV5d, N value EV6d, machine type EV7d, blade diameter EV8d, and blade thickness EV9d and torque OV10d has been learned.

[0070] As shown in the learning inference stage in the same figure, when the excavation depth EV1e, latitude EV2e, longitude EV3e, altitude EV4e, soil type EV5e, N value EV6e, machine type EV7e, blade diameter EV8e, and blade thickness EV9e are given as input values ​​to the learned torque estimation model M10a, it outputs a torque estimation value IV10e.

[0071] The learning control unit 120 provides the rotation speed estimation model M11a with the torque estimation value IV10e output by the torque estimation model M10a as an estimated value of the construction data (torque).

[0072] 16 is a diagram showing an example of the data structure of a rotation speed estimation model M11a in a modified example. In this example, the rotation speed estimation model M11a has a data structure in which the excavation depth EV1e, latitude EV2e, longitude EV3e, altitude EV4e, soil type EV5e, N-value EV6e, machine type EV7e, blade diameter EV8e, and blade thickness EV9e obtained from past data D1, and the torque estimate IV10e output by the torque estimation model M10a are used as explanatory variables, and the rotation speed OV11e is used as a response variable. In other words, the rotation speed estimation model M11a is a model in which the correspondence relationships between the excavation depth EV1e, latitude EV2e, longitude EV3e, altitude EV4e, soil type EV5e, N-value EV6e, machine type EV7e, blade diameter EV8e, blade thickness EV9e, and the torque estimate IV10e are learned, and the rotation speed OV11e.

[0073] As shown in the inference stage for learning in the same figure, when the excavation depth EV1f, latitude EV2f, longitude EV3f, altitude EV4f, soil type EV5f, N value EV6f, machine type EV7f, blade diameter EV8f, blade thickness EV9f, and estimated torque value IV10f are given as input values ​​to the learned rotation speed estimation model M11a, it outputs estimated rotation speed IV11f.

[0074] The learning control unit 120 provides the rotation speed estimated value IV11f output by the rotation speed estimation model M11a to the speed estimation model M12a as an estimated value of the construction data (rotation speed).

[0075] 17 is a diagram showing an example of the data structure of a speed estimation model M12a in a modified example. In this example, the speed estimation model M12a has a data structure in which the excavation depth EV1f, latitude EV2f, longitude EV3f, altitude EV4f, soil type EV5f, N value EV6f, machine type EV7f, blade diameter EV8f, and blade thickness EV9f obtained from the past data D1, the torque estimated value IV10f output by the torque estimation model M10a, and the rotation speed estimated value IV11f output by the rotation speed estimation model M11a are used as explanatory variables, and the speed OV12f is used as a response variable. That is, the speed estimation model M12a is a model that has learned the correspondence between excavation depth EV1f, latitude EV2f, longitude EV3f, altitude EV4f, soil type EV5f, N value EV6f, machine type EV7f, blade diameter EV8f, blade thickness EV9f, torque estimate IV10f, rotation speed estimate IV11f, and speed OV12f.

[0076] As shown in the learning inference stage in the same figure, when the excavation depth EV1g, latitude EV2g, longitude EV3g, altitude EV4g, soil type EV5g, N value EV6g, machine type EV7g, blade diameter EV8g, blade thickness EV9g, torque estimate IV10g, and rotation speed estimate IV11g are given as input values ​​to the learned speed estimation model M12a, it outputs a speed estimate IV12g.

[0077] That is, in this modified example, the learning control unit 120 assigns a first explanatory variable to the torque estimation model M10a (first learned model) among the learned models, thereby making the first objective variable (e.g., torque estimated value) output from the torque estimation model M10a (first learned model) the second explanatory variable, and generates the rotation speed estimation model M11a (second learned model) by assigning the second explanatory variable and a second objective variable of a different type from the first objective variable. In addition, in this modified example, the learning control unit 120 assigns a first explanatory variable to the rotation speed estimation model M11a (first learned model) among the learned models, thereby using the first objective variable (e.g., rotation speed estimated value) output from the rotation speed estimation model M11a (first learned model) as a second explanatory variable, and generates the speed estimation model M12a (second learned model) by assigning the second explanatory variable and a second objective variable of a different type from the first objective variable.

[0078] As described above, the learning control device 10 learns, for multiple trained models, the output (i.e., the objective variable) of a first trained model at the inference stage as the input (i.e., the explanatory variable) of a second trained model at the learning stage. Among the multiple trained models described above, the torque estimation model M10a is an example of a first trained model, and the rotation speed estimation model M11a is an example of a second trained model when the torque estimation model M10a is the first trained model. Among the multiple trained models described above, the rotation speed estimation model M11a is an example of a first trained model, and the speed estimation model M12a is an example of a second trained model when the rotation speed estimation model M11a is the first trained model.

[0079] A first trained model (for example, soil quality estimation model M1) uses depth information and position information as explanatory variables, and soil quality, which is depth-dependent information, as a response variable. The second trained model (for example, torque estimation model M10, rotation speed estimation model M11, and speed estimation model M12) uses depth information, position information, and soil type as explanatory variables, and uses at least one of the depth-dependent information, namely, the torque applied to the pile during construction, the number of pile rotations per unit time, or the excavation speed due to pile rotation, as the objective variable.

[0080] According to the learning control device 10 configured in this manner, the learning models are trained individually, so that a learning method suitable for each learning model can be applied, thereby improving the efficiency and accuracy of learning. Furthermore, in the construction condition estimation system 1 of this embodiment, the output of the second trained model at the inference stage (i.e., the objective variable) is the operating condition of the construction machine when constructing piles. Therefore, according to the construction condition estimation system 1, the output of the second trained model at the inference stage is provided to the control computer of the construction machine, thereby improving the efficiency and accuracy of construction.

[0081] [Second embodiment] Next, a description will be given of a construction condition estimation system 1 according to a second embodiment. The construction condition estimation system 1 according to this embodiment differs from the first embodiment described above in that it uses a soil quality estimation model M1 and an N-value estimation model M2.

[0082] FIG. 18 is a diagram showing an example of the flow of operations of the learning control device 10 of this embodiment.

[0083] (Step S110) The past data acquisition unit 110 acquires past data D1 from the past data storage device 21. The past data D1 is the same as in the first embodiment described above, and therefore a description thereof will be omitted.

[0084] The past data acquisition unit 110 outputs the acquired past data D1 to the learning control unit 120. The learning control unit 120 executes model learning based on the past data D1 acquired in step S110. Here, the learning models to be learned include the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12 described in the first embodiment, as well as the soil quality estimation model M1 and the N-value estimation model M2.

[0085] (Step S120) The learning control unit 120 trains the soil quality estimation model M1. The soil quality estimation model M1 is a model in which the correspondence between the past data D1 and the soil quality is learned.

[0086] 19 is a diagram showing an example of the data structure of the soil type estimation model M1 of this embodiment. In this example, the soil type estimation model M1 is a model in which the correspondence relationships between the excavation depth EV1h, latitude EV2h, longitude EV3h, and altitude EV4h and the soil type OV5h are learned, with the excavation depth EV1h, latitude EV2h, longitude EV3h, and altitude EV4h as explanatory variables and the soil type OV5h as a response variable.

[0087] The learning control unit 120 trains the soil quality estimation model M1 using the acquired excavation depth EV1h, latitude EV2h, longitude EV3h, and altitude EV4h as explanatory variables and the soil quality OV5h as a response variable.

[0088] The learning control unit 120 uses past data D1 from a plurality of construction sites as a learning target to train the soil quality estimation model M1. The learning control unit 120 performs learning on the past data D1 of a plurality of construction sites collectively, so that the soil quality estimation model M1 becomes a learning model that can predict the soil quality for each excavation depth at various construction sites.

[0089] The soil quality estimation model M1 thus trained outputs a soil quality estimation value IV5i in the inference stage when given an excavation depth EV1i, a latitude EV2i, a longitude EV3i, and an altitude EV4i.

[0090] In other words, according to the soil quality estimation model M1 having the data structure configured in this way, even if the soil quality at a construction site is unknown, the soil quality at each depth at the construction site can be estimated by specifying the depth and location.

[0091] (Step S130) Returning to Fig. 18, the learning control unit 120 trains the N-value estimation model M2. The N-value estimation model M2 is a model in which the correspondence between the past data D1 and the N value is learned.

[0092] 20 is a diagram showing an example of the data structure of the N-value estimation model M2 of this embodiment. In this example, the N-value estimation model M2 is a model in which the correspondence relationships between the excavation depth EV1h, latitude EV2h, longitude EV3h, altitude EV4h, soil quality EV5h and the N-value OV6h are learned, with the excavation depth EV1h, latitude EV2h, longitude EV3h, altitude EV4h, soil quality EV5h as explanatory variables and the N-value OV6h as a response variable.

[0093] The learning control unit 120 acquires information on the location of the construction site (e.g., latitude, longitude, altitude). In addition, the learning control unit 120 acquires information on the soil type and N value for each excavation depth (e.g., every 10 cm) as depth-dependent information of the construction site.

[0094] The learning control unit 120 uses the acquired information on excavation depth, latitude, longitude, altitude, and soil type as explanatory variables (excavation depth EV1h, latitude EV2h, longitude EV3h, altitude EV4h, soil type EV5h), and the information on the N value as the objective variable (N value OV6h), and trains the N value estimation model M2.

[0095] The learning control unit 120 collectively learns the N-value estimation model M2 from the past data D1 of the selected construction site. The learning control unit 120 performs learning on the past data D1 of a plurality of construction sites collectively, and the N-value estimation model M2 becomes a learning model that can predict the N-value for each excavation depth at various construction sites.

[0096] The N-value estimation model M2 trained in this way outputs an N-value estimated value IV6i in the inference stage by providing the following information: excavation depth EV1i, latitude EV2i, longitude EV3i, altitude EV4i, and soil quality estimated value IV5i.

[0097] According to the N-value estimation model M2 having the data structure configured in this way, even if the N-value of a construction site is unknown, the N-value for each depth of the construction site can be estimated by specifying the depth, position, and soil type.

[0098] Furthermore, according to the soil quality estimation model M1 and the N-value estimation model M2 having the data structure configured as described above, even if the soil quality and N-value of a construction site are unknown, by specifying the depth and location, it is possible to use these two types of estimation models to estimate the soil quality and N-value for each depth of the construction site.

[0099] Furthermore, with the soil quality estimation model M1 and the N-value estimation model M2 having the data structure configured in this way, the soil quality estimation model M1 and the N-value estimation model M2 are trained separately, so that a learning method suitable for each learning model can be applied, thereby improving the efficiency and accuracy of learning.

[0100] (Step S140) Returning to Fig. 18, the learning control unit 120 trains the torque estimation model M10b. The torque estimation model M10b is a model in which the correspondence between the past data D1 and the torque of the construction machine is learned for each of a plurality of depths at a plurality of work sites.

[0101] 21 is a diagram showing an example of the data structure of the torque estimation model M10b of this embodiment. The configuration of the torque estimation model M10b is similar to that of the torque estimation model M10 described above, and therefore a description thereof will be omitted.

[0102] As shown in the inference stage of the same figure, when the excavation depth EV1i, latitude EV2i, longitude EV3i, altitude EV4i, estimated soil type IV5i, estimated N value IV6i, machine type EV7i, blade diameter EV8i, and blade thickness EV9i are given as input values ​​to the trained torque estimation model M10b, it outputs an estimated torque value IV10i.

[0103] (Step S150) Returning to Fig. 18, the learning control unit 120 trains a rotation speed estimation model M11b. The rotation speed estimation model M11b is a model in which the correspondence between the past data D1 and the rotation speed of the construction machine is learned for each of a plurality of depths at a plurality of work sites.

[0104] 22 is a diagram showing an example of the data structure of the rotation speed estimation model M11b of this embodiment. The configuration of the rotation speed estimation model M11b is similar to that of the rotation speed estimation model M11 described above, and therefore a description thereof will be omitted.

[0105] As shown in the inference stage of the same figure, when the excavation depth EV1i, latitude EV2i, longitude EV3i, altitude EV4i, estimated soil type IV5i, estimated N value IV6i, machine type EV7i, blade diameter EV8i, blade thickness EV9i, and estimated torque IV10i are given as input values ​​to the trained rotation speed estimation model M11b, it outputs estimated rotation speed IV11i.

[0106] (Step S160) Returning to Fig. 18, the learning control unit 120 trains a speed estimation model M12b. The speed estimation model M12b is a model in which the correspondence between the past data D1 and the speed of the construction machine is learned for each of a plurality of depths at a plurality of work sites.

[0107] 23 is a diagram showing an example of the data structure of the speed estimation model M12b of this embodiment. The configuration of the speed estimation model M12b is the same as that of the speed estimation model M12 described above, and therefore a description thereof will be omitted.

[0108] As shown in the inference stage of the same figure, when the excavation depth EV1i, latitude EV2i, longitude EV3i, altitude EV4i, estimated soil type IV5i, estimated N value IV6i, machine type EV7i, blade diameter EV8i, blade thickness EV9i, estimated torque IV10i, and estimated rotation speed IV11i are given as input values ​​to the trained speed estimation model M12b, it outputs estimated speed IV12i.

[0109] The process of the inference stage using each estimation model in this embodiment is the same as that in the first embodiment described above, except that it includes an inference process using the soil estimation model M1 and the N-value estimation model M2. Therefore, a detailed description of the process of the inference stage using each estimation model in this embodiment will be omitted.

[0110] According to the estimation model having the data structure shown in this embodiment, even if the site data D2 does not contain information on soil quality or N values, information equivalent to a columnar diagram, such as soil quality and N values, can be obtained using the soil quality estimation model M1 and the N value estimation model M2.

[0111] In other words, the estimation device 40 of this embodiment outputs a set of depth, estimated soil quality, and estimated N value as predicted columnar diagram information for any depth indicated by the depth information based on the estimated soil quality, which is the objective variable output from the soil quality estimation model M1 (first trained model), and the N value, which is the objective variable output from the N value estimation model M2 (second trained model).

[0112] In this embodiment, as described in the modified example of the first embodiment, multi-stage learning may also be adopted, in which the objective variables output from the first trained model are used as explanatory variables of the second trained model.

[0113] In this case, the estimation device 40 of this embodiment outputs the soil quality, which is the objective variable output from the soil quality estimation model M1 (first trained model), as the estimated soil quality, and provides the estimated soil quality as an explanatory variable to the N-value estimation model M2 (second trained model) in addition to the depth information and position information, thereby outputting the N-value, which is the objective variable output from the N-value estimation model M2 (second trained model), as the estimated N-value.

[0114] Furthermore, the information indicating the position of the construction site (latitude, longitude, and altitude) described above is not essential for learning the torque estimation model M10b, the rotation speed estimation model M11b, and the speed estimation model M12b. On the other hand, information indicating the location of the construction site (latitude, longitude, and altitude) can be required for the learning of each estimation model. In this case, each estimation model can produce different estimation results depending on the location of the construction site, allowing for more accurate estimation.

[0115] [Third embodiment] Next, a construction condition estimation system 1 according to a third embodiment will be described.

[0116] Here, the excavation speed is a value that directly affects the construction period of pile construction, and is therefore highly useful in planning the construction schedule and determining whether construction at the construction site is progressing according to the construction schedule.

[0117] 24 is a diagram showing an example of the operation flow of the learning control device 10 of this embodiment. The estimation device 40 of this embodiment integrates the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12, and uses a trained model, i.e., the speed estimation model M12c, that estimates the excavation speed directly from the site data D2 without estimating the torque or the rotation speed.

[0118] (Step S210) The past data acquiring unit 110 acquires past data D1 from the past data storage device 21. The past data D1 is similar to that in the first embodiment and the modified example of the first embodiment described above, and therefore a description thereof will be omitted.

[0119] The past data acquisition unit 110 outputs the acquired past data D1 to the learning control unit 120. The learning control unit 120 executes learning of the speed estimation model M12c based on the past data D1 acquired in step S110.

[0120] (Step S220) The learning control unit 120 trains a speed estimation model M12c. The speed estimation model M12c is a model that has learned the correspondence between past data D1 and speed for each of a plurality of depths at a plurality of sites.

[0121] 25 is a diagram showing an example of the data structure of the speed estimation model M12c of this embodiment. In this example, the speed estimation model M12c is a model in which the correspondence relationships between the excavation depth EV1j, latitude EV2j, longitude EV3j, altitude EV4j, soil type EV5j, N-value EV6j, machine type EV7j, blade diameter EV8j, blade thickness EV9j and the speed OV12j are learned using the excavation depth EV1j, position (latitude EV2j, longitude EV3j, altitude EV4j, soil type EV5j, N-value EV6j, machine type EV7j, blade diameter EV8j, blade thickness EV9j) as explanatory variables and the speed OV12j as a response variable.

[0122] As shown in the inference stage of the same figure, when the trained speed estimation model M12c is given the excavation depth EV1k, latitude EV2k, longitude EV3k, altitude EV4k, soil type EV5k, N value EV6k, machine type EV7k, blade diameter EV8k, and blade thickness EV9k as input values, it outputs a speed estimation value IV12k.

[0123] In this case, the estimation device 40 uses a group of information including at least depth information and depth-independent information such as position information and pile specification information as explanatory variables, and estimates the excavation speed due to pile rotation from a group of information including at least depth information and depth-independent information such as position information and pile specification information using a speed estimation model M12c (sixth trained model) trained with the excavation speed due to pile rotation as depth-dependent information as the objective variable.

[0124] According to the estimation device 40 configured in this manner, the excavation speed, which is useful for planning and managing the construction period of pile construction, can be directly estimated without going through multiple trained models, thereby reducing the computational load.

[0125] In other words, the estimation device 40 of this embodiment can calculate the time and number of days required for construction from the site data D2 that does not include construction data more easily than in the above-described embodiments.

[0126] It should be noted that the information indicating the position of the construction site (latitude, longitude, and altitude) described above is not essential for learning the speed estimation model M12c. On the other hand, in the learning of the speed estimation model M12c, information indicating the location of the construction site (latitude, longitude, and altitude) can be made essential. In this case, each estimation model can produce different estimation results depending on the location of the construction site, thereby enabling more accurate estimation.

[0127] As described above, the estimation device 40 of this embodiment can be said to estimate the operating conditions of the pile construction machine using the torque estimation model M10 (third trained model), the rotation speed estimation model M11 (fourth trained model), and the speed estimation model M12 (fifth trained model). As described above, the torque estimation model M10 (third trained model) is a trained model that uses a group of information including at least depth information and depth-independent information such as position information and pile specification information as explanatory variables, and uses depth-dependent information such as the torque applied to the pile during construction as the objective variable. The rotation speed estimation model M11 (fourth trained model) is a trained model that uses a group of information including at least depth information, depth-independent information such as location information and pile specification information, and depth-dependent information such as the torque applied to the pile during construction as explanatory variables, and uses the number of pile rotations per unit time as the objective variable. The speed estimation model M12 (speed estimation model M12 (fifth trained model)) is a trained model that uses as explanatory variables a group of information that includes at least the depth information, the depth-independent information of the position information and the pile specification information, and the depth-dependent information of the torque applied to the pile during construction and the number of pile rotations per unit time, and that uses as a target variable the excavation speed due to the rotation of the pile.

[0128] The estimation device 40 estimates the torque applied to the pile during construction, the number of rotations of the pile per unit time, and the excavation speed due to the rotation of the pile from a group of information including at least depth information, position information which is depth-independent information, and pile specification information, using a torque estimation model M10 (third trained model), a rotation speed estimation model M11 (fourth trained model), and a speed estimation model M12 (fifth trained model).

[0129] According to the estimation device 40 configured in this manner, individual trained models are combined and applied using learning methods appropriate for each learning model, thereby improving the accuracy of estimating the operating conditions of construction machines.

[0130] For example, in the learning stage of the speed estimation model M12 (sixth learned model), the learning control device 10 may use the excavation speed obtained by sequentially estimating using the soil estimation model M1, the N-value estimation model M2, the torque estimation model M10, the rotation speed estimation model M11, and the speed estimation model M12 as the target variable (i.e., training data) of the speed estimation model M12. According to the learning control device 10 configured in this manner, highly accurate inference results obtained by combining individual trained models using learning methods suitable for each learning model can be used as training data, thereby enabling the speed estimation model M12 (sixth trained model) to be trained efficiently and with high accuracy.

[0131] As described above, the construction condition estimation system 1 of this embodiment builds a machine learning model based on existing log information. The log information is readily available in an existing database. Therefore, the construction condition estimation system 1 of this embodiment can predict information about pile construction with a simple configuration.

[0132] Furthermore, even if information on a log corresponding to the location of a construction site is unavailable, the construction condition estimation system 1 of this embodiment can estimate information equivalent to the log using the soil quality estimation model M1 and the N-value estimation model M2. Therefore, for example, before a boring test is conducted at the construction site, it is possible to estimate the construction period from the operating conditions of the construction machine and create a construction schedule.

[0133] Although the embodiments of the present invention have been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment and can be appropriately modified without departing from the spirit of the present invention. The configurations described in the above-described embodiments may be combined.

[0134] Each unit included in each device in the above-described embodiments may be realized by dedicated hardware, or may be realized by a memory and a microprocessor.

[0135] In addition, each part of each device may be composed of a memory and a CPU (central processing unit), and the functions of each part of each device may be realized by loading a program into memory and executing it.

[0136] In addition, a program for realizing the functions of each unit of each device may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing by each unit of the control unit. Note that the term "computer system" here includes hardware such as an OS and peripheral devices.

[0137] Furthermore, if a WWW system is used, the "computer system" also includes the homepage provision environment (or display environment). "Computer-readable recording media" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording media" also includes devices that dynamically store programs for a short period of time, such as communication lines used when transmitting programs over networks like the Internet or over communication lines like telephone lines, and devices that store programs for a fixed period of time, such as volatile memory within computer systems that serve as servers or clients. Furthermore, the programs may be programs that implement some of the aforementioned functions, or may be programs that can realize the aforementioned functions in combination with programs already stored in the computer system. [Explanation of symbols]

[0138] 1...construction condition estimation system, 10...learning control device, 110...past data acquisition unit, 120...learning control unit, 21...past data storage device, 22...site data supply device, 30...learned model storage unit, 40...estimation device, 410...site information acquisition unit, 420...estimation unit, 430...output unit, 50...output device, D1...past data, D2...site data, M1...soil quality estimation model, M2...N value estimation model, M10...torque estimation model, M11...rotation speed estimation model, M12...speed estimation model

Claims

1. A data structure in which a set of information consisting of location information indicating the location of the pile construction site, depth-dependent information which is information about the geological layer that changes according to changes in depth at the construction site, depth-independent information which is information that is not dependent on the depth and includes at least information on the pile specifications, and depth information which indicates the depth for which the depth-dependent information is information is configured as a record for each depth, and has such records for multiple depths.

2. The depth-dependent information includes at least one of the soil type of the stratum at the depth indicated by the depth information included in the record, or the N value at the depth. The data structure of claim 1 .

3. The depth-dependent information includes at least one of the torque applied to the pile when constructing the pile at the depth indicated by the depth information included in the record, the number of rotations of the pile per unit time, or the excavation speed due to the rotation of the pile. The data structure of claim 1 .

4. A learning method for generating a trained model trained using the data structure described in claim 1, using any of the information contained in the data structure as a target variable and at least a portion of the other information as explanatory variables.

5. A first explanatory variable is given to a first trained model among the trained models, and a first objective variable output from the first trained model is added to a second explanatory variable, and the second explanatory variable and a second objective variable of a different type from the first objective variable are given to generate a second trained model. The learning method according to claim 4.

6. The first trained model uses the depth information and the position information as explanatory variables, and uses the soil quality added to the pile during construction of the pile, which is the depth-dependent information, as a target variable, The second trained model uses the depth information, the position information, and the soil type as explanatory variables, and uses at least one of the depth-dependent information, the number of rotations of the pile per unit time, or the excavation speed due to the rotation of the pile, as a target variable. The learning method according to claim 5 .

7. The soil quality, which is a response variable output from the first trained model according to claim 6, is output as an estimated soil quality, and the estimated soil quality is given to the second trained model as an explanatory variable in addition to the depth information and the position information, thereby outputting the N value, which is a response variable output from the second trained model, as an estimated N value. Estimation method.

8. For any depth indicated by the depth information, a set of the depth, the estimated soil quality, and the estimated N value is output as predicted log information based on the estimated soil quality, which is an objective variable output from the first trained model, and the N value, which is an objective variable output from the second trained model. The estimation method according to claim 7.

9. A third trained model trained using, as explanatory variables, a group of information including at least the depth information, the position information which is depth-independent information, and information on the specifications of the pile, among the information included in the data structure according to claim 1, and a torque applied to the pile during construction of the pile which is depth-dependent information, as a target variable; A fourth trained model trained using, as explanatory variables, a group of information including at least the depth information, the position information and the pile specification information, which are depth-independent information, and the torque applied to the pile during construction of the pile, which is depth-dependent information, among the information included in the data structure described in claim 1, and the number of rotations of the pile per unit time, as a target variable; A fifth trained model trained using, as explanatory variables, a group of information including at least the depth information, the position information and the pile specification information, which are depth-independent information, and the torque applied to the pile during construction of the pile and the number of rotations of the pile per unit time, among the information included in the data structure according to claim 1, and the excavation speed due to the rotation of the pile as a target variable; From a group of information including at least the depth information, the position information which is depth-independent information, and the pile specification information, the torque applied to the pile during construction of the pile, the number of rotations of the pile per unit time, and the excavation speed due to the rotation of the pile are estimated. Estimation method.

10. The sixth trained model is trained using, as explanatory variables, an information group including at least the depth information, the position information, and the pile specification information, which are depth-independent information, among the information included in the data structure of claim 1, and an excavation speed due to the rotation of the pile, which is depth-dependent information, as a target variable, and estimates the excavation speed due to the rotation of the pile from the information group including at least the depth information, the position information, and the pile specification information, which are depth-independent information. Estimation method.

Images