JP7176941B2 - Method for estimating N value and fine fraction content, soil improvement material and information processing device - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for estimating an N value and a fine fraction content, a ground improvement material, and an information processing apparatus.

When improving the ground as a countermeasure against liquefaction, it is common to conduct a ground survey in advance to determine the range and depth of ground improvement (called liquefaction judgment) and improve the ground that may liquefy. is. In order to improve the ground by injecting chemicals into reclaimed ground, etc., which has a complicated soil structure, it is necessary to select the injection parameters (injection amount and injection rate of chemicals) according to the soil quality. In order to determine the appropriate injection specifications that do not cause problems such as chemical agent leakage to the ground surface and upheaval of the ground surface, it is necessary to accurately determine the fine particle content (hereinafter referred to as Fc) of the ground. be. In addition to Fc, an accurate N-value of the ground is required for liquefaction determination.

Since conducting a detailed preliminary ground survey is a burden in terms of process and cost, we collect drilling data during drilling to inject chemicals, create training data, and let the neural network learn. A method for discriminating a stratum has been proposed (see Patent Document 1).

Further, in the penetration test, multiple test parameters are defined based on the load, rotational torque and penetration amount, and these are used as explanatory variables, and multiple regression analysis is performed with Fc as the objective variable to obtain a regression equation. A method of estimating Fc from the equation, and in the penetration test, define multiple test parameters based on the load, rotational torque and penetration amount, use them as explanatory variables, and perform multiple regression analysis with the objective variable as N value. A method has also been proposed in which a regression equation is obtained from the equation and the N value is estimated from this regression equation (see Patent Document 2).

Japanese Patent Application Laid-Open No. 2002-133391 JP 2014-134066 A

Considering that Fc is an important condition for determining the injection specifications, the method described in Patent Document 1 can divide the stratum, but cannot obtain an accurate Fc. Injection specifications cannot be determined. In addition, considering that accurate values of the N value and Fc are necessary to carry out the liquefaction determination, the method described in Patent Document 1 does not obtain the N value and Fc, so the liquefaction determination is performed. This cannot be done, and as a result, there is a risk of improving the ground, which has a low potential for liquefaction.

In addition, in the method described in Patent Document 2, regression equations are used to individually estimate the Fc and N values, but for highly accurate liquefaction countermeasures, more accurate Fc and N A method of estimating the value is required.

Accordingly, an object of the present invention is to obtain the N value and the fine grain fraction content of the ground more accurately.

In order to solve the above problems, the present invention provides a detection step for detecting drilling data including water supply pressure at the time of drilling for injecting a chemical used as a countermeasure against ground liquefaction, and a water supply pressure during drilling. Using a multi-task neural network learning model with drilling data containing as an input variable and the N value and fine grain content of the ground as objective variables, and the detected drilling data, liquefaction countermeasures and an estimating step of estimating the N value and the fine grain content of the target ground.
Further, the present invention includes a detection step of detecting drilling data including rotational torque during drilling for injecting a chemical used for ground liquefaction countermeasures, and drilling data including rotational torque during drilling. Using a learning model of a multi-task neural network with input variables and the N value and fine grain content of the ground as objective variables and the detected drilling data, the ground that is the target of liquefaction countermeasures and an estimating step of estimating the N value and the fine particle content.
Further, the present invention provides a detection step for detecting drilling data including a bit load during drilling for injecting a chemical used as a ground liquefaction countermeasure, and drilling data including the bit load during drilling. is used as an input variable and the N value and fine grain fraction content of the ground are used as objective variables. and an estimating step of estimating the N value and the fine particle content.
Further, the present invention includes a detection step of detecting drilling data including a drilling speed and a drilling data including a drilling speed during drilling, when drilling a hole for injecting a chemical used as a countermeasure against ground liquefaction. A multi-task neural network learning model with hole data as an input variable and the N value and fine grain content of the ground as objective variables, and the detected drilling data are used as targets for liquefaction countermeasures. and an estimating step of estimating the N-value and the fine grain content of the ground.

In the estimation step, the multi-task neural network learning model and one or more single-task learning models with the drilling data as input variables and the N value and fine grain fraction content of the ground as objective variables are used. , the estimation may be performed by weighting according to each of the learning models.

In the learning model, data weighted according to the distance from the boring position to the drilling position for measuring the N value and the fine grain content ratio, which are objective variables, are included as input variables. You may do so.

The learning model may include the variance or standard deviation of the drilling data as an input variable.

In the learning model, the history of the drilling data from the surface of the ground during drilling may be included as an input variable.

In addition, the present invention provides ground in which a chemical is injected based on the N value and the fine grain content estimated by any of the above methods for estimating the N value and the fine grain content and liquefaction countermeasures are taken. provide an improvement.

In addition, the present invention provides an information processing apparatus characterized by executing any of the methods for estimating the N value and the fine grain fraction content.

ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to obtain|require N-value and fine-grain fraction content rate of a ground more accurately.

1 is a block diagram showing an example of the configuration of the entire system according to one embodiment of the present invention; FIG. FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing apparatus according to the embodiment; FIG. 2 is a block diagram showing an example of the functional configuration of the information processing apparatus; 4 is a flowchart showing an example of a method of generating a learning model in the same embodiment; The flowchart which shows an example of the method of estimating an N value and a fine-grain fraction content rate using a learning model in the same embodiment.

An example of a mode for carrying out the present invention will be described.
[Constitution]
FIG. 1 is a block diagram showing an example of the overall configuration of a system 1 according to one embodiment of the invention. The system 1 includes a boring system 10 for drilling a cylindrical hole in the ground and boring, a drilling system 20 for drilling a hole for injecting a chemical agent used for countermeasures against liquefaction of the ground, and a ground N and an information processing device 30 for estimating the value and the fine particle content (hereinafter referred to as Fc). The drilling system 20 and the information processing device 30 are connected via a communication line. The N value and Fc of the ground are measured by boring performed by the boring system 10 (this is called a standard penetration test). The drilling system 20 has a function of collecting data during drilling (referred to as drilling data) in a drilling process before chemical injection. The drilling data includes, for example, water supply pressure, rotational torque, bit load, rotation speed, drilling speed and drilling energy during drilling.

In system 1, training data is created based on boring data and a large number of drilling data as preliminary work performed prior to the main construction, and machine learning is performed on this training data by a multitasking neural network. Then, a learning model is generated that estimates the N value and Fc. Then, in the main construction work on the ground to be improved, the drilling data of the ground is applied to this learning model, and the N value and Fc of the ground are simultaneously estimated.

FIG. 2 is a diagram showing the hardware configuration of the information processing device 30. As shown in FIG. The information processing device 30 is physically configured as a computer device including a processor 3001, a memory 3002, a storage 3003, a communication device 3004, an input device 3005, an output device 3006, and a bus connecting them. Each of these devices operates with power supplied from a battery (not shown). Note that in the following description, the term "apparatus" can be read as a circuit, device, unit, or the like. The hardware configuration of the information processing device 30 may be configured to include one or more of each device shown in the figure, or may be configured without including some of the devices.

Each function of the information processing device 30 is performed by causing the processor 3001 to perform calculations, controlling communication by the communication device 3004, and controlling the It is realized by controlling at least one of data reading and writing in 3002 and storage 3003 .

The processor 3001, for example, operates an operating system and controls the entire computer. The processor 3001 may be configured by a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic device, registers, and the like. Also, for example, a baseband signal processing unit, a call processing unit, and the like may be implemented by the processor 3001 .

The processor 3001 reads programs (program codes), software modules, data, etc. from at least one of the storage 3003 and the communication device 3004 to the memory 3002, and executes various processes according to them. As the program, a program that causes a computer to execute at least part of the operations described below is used. The functional blocks of information processing device 30 may be implemented by a control program stored in memory 3002 and running on processor 3001 . Various processes may be executed by one processor 3001, but may also be executed by two or more processors 3001 simultaneously or sequentially. Processor 3001 may be implemented by one or more chips. Note that the program may be transmitted from the network to the information processing device 30 via an electric communication line.

The memory 3002 is a computer-readable recording medium, and is composed of at least one of, for example, ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), and RAM (Random Access Memory). may be The memory 3002 may also be called a register, cache, main memory (main storage device), or the like. The memory 3002 can store executable programs (program code), software modules, etc. to perform the methods of the present invention.

The storage 3003 is a computer-readable recording medium, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (for example, a compact disk, a digital versatile disk, a Blu-ray disk), smart card, flash memory (eg, card, stick, key drive), floppy disk, magnetic strip, and/or the like. Storage 3003 may also be called an auxiliary storage device.

The communication device 3004 is hardware (transmitting/receiving device) for communicating between computers via at least one of a wired network and a wireless network, and is also called a network device, network controller, network card, communication module, or the like.

The input device 3005 is an input device (for example, keyboard, mouse, microphone, switch, button, sensor, joystick, ball controller, etc.) that receives input from the outside. The output device 3006 is an output device (for example, display, speaker, LED lamp, etc.) that outputs to the outside. Note that the input device 3005 and the output device 3006 may be integrated (for example, a touch panel).

The information processing device 30 includes hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array). A part or all of each functional block may be implemented by the hardware. For example, processor 3001 may be implemented using at least one of these pieces of hardware.

FIG. 3 is a diagram showing an example of the functional configuration of the information processing device 30. As shown in FIG. The data acquisition unit 31 acquires various data from the outside. This data includes ground N value and Fc measured during boring by the boring system 10 and drilling data detected during drilling by the drilling system 20 . The data acquired by the data acquisition unit 31 in the preliminary work are the N value and Fc of the ground targeted for the preliminary work and drilling data, and the data acquired by the data acquisition unit 31 in the main construction are the data of the main construction. This is the drilling data of the target ground.

The teacher data generation unit 32 uses the data acquired by the data acquisition unit 31 in the preliminary work to generate teacher data in the learning model. More specifically, the teacher data generator 32 generates teacher data in a multi-task neural network learning model (simultaneous regression model) with drilling data as input variables and ground N value and Fc as objective variables. do.

The model generation unit 33 divides the drilling data as input variables and the N value and Fc as objective variables into learning data and verification data at a ratio of about 8:2, for example, and uses the learning data as teacher data for a neural network. to generate a learning model.

The model storage unit 34 stores learning models generated by the model generation unit 33 .

The verification unit 35 verifies the accuracy of the learning model using the verification data described above. Specifically, the verification unit 35 applies the drilling data included in the verification data to the learning model stored in the model storage unit 34, estimates the N value and Fc, and verifies the estimated values. The difference is evaluated by comparing with the N value and Fc included in the data for Here, if the accuracy of the learning model is insufficient, the model generation unit 33 reviews hyperparameters and the like in the learning model and generates a learning model with sufficient accuracy.

The estimation unit 36 uses the learning model stored in the model storage unit 34 and the drilling data acquired by the data acquisition unit 31 in this construction to estimate the N value and Fc of the ground to be improved. Output. Based on this N value and Fc, liquefaction determination and injection specifications are determined, and ground improvement using chemicals is carried out.

[motion]
[Learning model generation operation]
FIG. 4 is a flow chart showing an example of a method for generating a learning model of N values and Fc. First, boring is performed by the boring system 10 as preliminary work on arbitrary ground (step S11). Thereby, the N value and Fc of the ground are measured.

Next, the drilling system 20 drills holes at a plurality of points in the vicinity of the boring location (step S12). For example, drilling is performed at about 5 points per point (for example, 32 points, 160 points in total) within a range of radius 7 m centered on the boring location. As a result, drilling data at each drilling point in the vicinity of the boring location can be obtained.

Next, these N value, Fc and drilling data are input to the information processing device 30 . The N value and Fc are input to the information processing device 30 via an arbitrary recording medium, for example (step S13). The drilling data is input from the drilling system 20 to the information processing device 30 via a communication line or an arbitrary recording medium, for example. Thereby, the data acquisition unit 31 of the information processing device 30 acquires the N value, Fc, and drilling data of the ground targeted for the preliminary work.

Next, the teacher data generation unit 32 of the information processing device 30 uses the drilling data acquired by the data acquisition unit 31 as an input variable, and the ground N value and Fc acquired by the data acquisition unit 31 as objective variables. Teacher data for a multitask neural network learning model (simultaneous regression model) is generated (step S14).

Next, the model generator 33 divides the drilling data as input variables and the N value and Fc as objective variables into learning data and verification data at a ratio of about 8:2, for example (step S15).

Next, the model generation unit 33 performs multi-task learning using a neural network using the learning data as teacher data (step S16), thereby generating a learning model and storing it in the model storage unit 34 (step S17).

Next, the verification unit 35 verifies the accuracy of the learning model stored in the model storage unit 34 using the verification data described above (step S18). Here, if the accuracy of the learning model is insufficient (step S19; No), as described above, the model generation unit 33 reviews the hyperparameters and the like in the learning model and generates a learning model with sufficient accuracy. (steps S16 and S17). If it is determined that the accuracy of the learning model is sufficient (step S19; Yes), the processing shown in FIG. 4 ends.

[Estimated operation of N value and Fc]
FIG. 5 is a flow chart showing an example of a method for estimating the N value and Fc. First, the drilling system 20 drills a plurality of holes in the ground targeted for the main construction (that is, the ground targeted for improvement) (step S21). As a result, drilling data at each drilling point is obtained.

Next, these drilling data are input from the drilling system 20 to the information processing device 30 via a communication line or via an arbitrary recording medium (step S22). Thereby, the data acquisition unit 31 of the information processing device 30 acquires the drilling data of the ground to be improved.

Next, the estimation unit 36 uses the learning model stored in the model storage unit 34 and the drilling data acquired by the data acquisition unit 31 to estimate the N value and Fc of the ground to be improved. Output (step S23).

Based on the N value and Fc, liquefaction determination and injection specifications are determined, and ground improvement (construction) using chemicals is carried out (step S24). As a result, a ground improvement body in which the ground has been improved with chemicals as a countermeasure against liquefaction is created.

The above is the description of the present embodiment. According to this embodiment, as machine learning for estimating the N value and Fc from drilling data, by performing multitask learning with a neural network, accurate estimation is possible, and appropriate setting of injection specifications and appropriate It is possible to set an appropriate improvement range and improvement depth. In addition, since drilling data obtained during drilling for injecting agents for countermeasures against liquefaction is used, the number of processes can be significantly reduced compared to, for example, the case of boring and measuring the N value and Fc. be.

Here, the effect of performing multitask learning using a neural network as machine learning for estimating the N value and Fc from drilling data is verified. The following is a table showing errors in the N value and Fc when estimating the N value and Fc from drilling data using a plurality of learning models.

The error at this time is a value obtained by multiplying the mean square error between (N value and Fc estimated from drilling data in each learning model) and (N value and Fc measured by boring) to the power of 1/2. However, the error evaluation method is not limited to this.

As shown in the table above, by performing multitask learning with a neural network as machine learning to estimate N value and Fc from drilling data, N value and Fc with much higher accuracy than when using other learning models It is possible to estimate Multi-task learning is a method of learning one model that performs a plurality of tasks, and is a method in which the accuracy of both is expected to be improved by sharing parameters to be learned for each task.

However, if there is no correlation between the shared parameters, it will not contribute to the accuracy improvement. In the present embodiment, by performing multitask learning that shares the N value and Fc, which were thought to have no correlation in the past, in fact, these N values and Fc are common to both in machine learning. It was found that there are useful features and properties. Because, by performing multitask learning with a neural network as machine learning for estimating the N value and Fc from drilling data, it is possible to estimate the N value and Fc with much higher accuracy than when using other learning models. Because it was possible. Therefore, when implemented in a single task that separately estimates the N value and Fc from the drilling data (when adopting the method described in JP 2014-134066 A described above), the method adopted in this embodiment It is not possible to obtain high accuracy compared to .

[Modification]
The invention is not limited to the embodiments described above. The embodiment described above may be modified as follows. Also, two or more of the following modified examples may be combined for implementation.

[Modification 1]
Since it is considered that the more teacher data is, the more the estimation accuracy is improved, the model generation unit 33 may generate the learning model by adding the data acquired in the main construction as teacher data.

[Modification 2]
For example, a random forest of a single-task model, a support vector machine, a neural network, etc. may be used for estimation, and the multi-task model used in the embodiment and each single-task model described above may be weighted according to their sophistication. That is, the estimation unit 36 uses a multitask neural network learning model and one or more single task learning models in which drilling data is used as an input variable and the N value and Fc of the ground are used as objective variables, and each learning You may make it estimate by performing the weighting according to the model. In the estimation step, using an estimated multitask neural network learning model and one or more single task learning models with drilling data as input variables and ground N value and Fc as objective variables, each learning model You may make it estimate by performing weighting according to it. For example, the estimating unit 36 assigns a weight of 0.5 to a multitasking neural network learning model, and assigns a weight of 0.5/N to N single-tasking learning models for estimation. You can do it.

[Modification 3]
In the learning model, even if drilling data weighted according to the distance from the boring location to the drilling point for measuring the N value and Fc, which are the objective variables, are included as input variables. good. For example, the model generating unit 33 may assign weights by weighted average of the distance from the boring location to each drilling point and the drilling data at that distance. Specifically, X1, X2, X3, . . . . when Dn is
(D1/X1+D2/X2+D3/X3...+Dn/Xn)/(1/X1+1/X2+1/X3...+1/Xn)
A weighted average calculation formula may be used. However, the weighting method is not limited to this.

[Modification 4]
In the learning model, in addition to the drilling data itself such as water supply pressure, rotational torque, bit load, rotation speed, drilling speed and drilling energy during drilling, for example, statistical analysis is performed for each 1 m of ground depth obtained by boring. The variance or standard deviation of the drilling data, which is calculated, may be included as an input variable. This makes it possible to generate a learning model that takes into consideration variations in drilling data at each depth.

[Modification 5]
In the learning model, in addition to the drilling data itself such as water supply pressure, rotating torque, bit load, rotation speed, drilling speed and drilling energy during drilling, the history of drilling data from the ground surface during drilling may be included as input variables. The water pressure, rotational torque, bit load, rotation speed, drilling speed and drilling energy measured during drilling are based on the tip of the member to be drilled (the underground location where drilling is performed). Since it is affected not only by the soil properties but also by the range from the ground surface to the point, drilling data from the ground surface at the time of drilling is used to create teacher data and estimate the N value and Fc. history as an input variable.

[Modification 6]
According to the present invention, a high-quality ground improvement body is constructed in which a chemical agent is injected based on the N value and Fc estimated by the above-described N value and Fc estimation method and liquefaction measures are taken. The present invention may be thought of as such a soil improvement body.

1: system, 10: boring system, 20: drilling system, 30: information processing device, 31: data acquisition unit, 32: teacher data generation unit, 33: model generation unit, 34: model storage unit, 35: verification unit , 36: estimation unit, 3001: processor, 3002: memory, 3003: storage, 3004: communication device, 3005: input device, 3006: output device.

Claims (10)

A detection step of detecting drilling data including water supply pressure during drilling for injecting a chemical used for ground liquefaction countermeasures;
Using a multi-task neural network learning model with drilling data including water pressure during drilling as an input variable and the N value and fine grain content of the ground as objective variables, and the detected drilling data and an estimating step of estimating the N value and the fine grain content of the ground targeted for liquefaction countermeasures. A detection step of detecting drilling data including rotational torque during drilling for injecting a chemical used as a ground liquefaction countermeasure;
Using a multi-task neural network learning model with drilling data including rotational torque during drilling as an input variable and the N value and fine grain content of the ground as objective variables, and the detected drilling data and an estimation step of estimating the N value and fine grain fraction content of the ground targeted for liquefaction countermeasures
A method for estimating the N value and the fine particle fraction content, comprising: a detection step of detecting drilling data including a bit load during drilling for injecting a chemical used as a ground liquefaction countermeasure;
Using a multi-task neural network learning model with drilling data including the bit load during drilling as an input variable and the N value and fine grain content of the ground as objective variables, and the detected drilling data and an estimation step of estimating the N value and fine grain fraction content of the ground targeted for liquefaction countermeasures
A method for estimating the N value and the fine particle fraction content, comprising: A detection step of detecting drilling data including drilling speed during drilling for injecting a chemical agent used for ground liquefaction countermeasures;
A multi-task neural network learning model with drilling data including the drilling speed during drilling as input variables and the N value and fine grain content of the ground as objective variables, and the detected drilling data. and an estimation step of estimating the N value and fine grain fraction content of the ground targeted for liquefaction countermeasures using
A method for estimating the N value and the fine particle fraction content, comprising: In the estimation step, the multi-task neural network learning model and one or more single-task learning models with the drilling data as input variables and the N value and fine grain fraction content of the ground as objective variables. 5. The method for estimating the N value and the fine grain fraction content rate according to any one of claims 1 to 4, wherein the estimation is performed by weighting according to each learning model. In the learning model, data weighted according to the distance from the boring position to the drilling position for measuring the N value and the fine grain content ratio, which are objective variables, are included as input variables. The method for estimating the N value and fine particle fraction content according to any one of claims 1 to 5 , characterized in that The N value and the fine grain fraction content according to any one of claims 1 to 6 , wherein the learning model includes the variance or standard deviation of the drilling data as an input variable. estimation method. The N value and N value according to any one of claims 1 to 7 , wherein the history of the drilling data from the surface of the ground during drilling is included as an input variable in the learning model. Method for estimating fine particle fraction content. A chemical is injected based on the N value and the fine particle content estimated by the method for estimating the N value and the fine particle content according to any one of claims 1 to 8 , and liquefaction countermeasures are taken. ground improvement body. An information processing apparatus that executes the method for estimating the N value and the fine grain fraction content rate according to any one of claims 1 to 8 .

Images