JP7329486B2 - Rock ground condition learning device, rock ground condition determination device, rock ground condition learning method, and rock ground condition determination method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rock condition learning device, a rock condition determination device, a rock condition learning method, and a rock condition determination method.

In a mountain tunnel construction project, geological surveys such as geophysical surveys and boring surveys are conducted as preliminary surveys, and rock mass grades are determined, for example, by rock mass evaluation using rock mass classification tables. At the design stage, it is common to apply the standard support pattern corresponding to the ground grade determined in the preliminary survey. On the other hand, at the construction stage, the geological condition of the face is recorded as a face observation record, and an evaluation score is assigned based on the face observation record. dot method).

In this way, rock ground evaluation at the construction stage is mainly performed based on photographs observed or taken near the face by engineers during or immediately after excavation. There are many dangers in the vicinity of the face, and the evaluation by geological specialists is not always constant, and the rock mass evaluation may be revised later. For example, a technique described in Patent Literature 1 has been proposed as a technique for supporting ground evaluation in the construction stage.

The evaluation support device described in Patent Document 1 uses, as input data, learning divided images generated by dividing a photographed image of a face in underground observation by a reference number of pixels, and outputs an evaluation classification for an observation item of the face. a learning result storage unit for storing a model generated by machine learning; and a control unit connected to the input unit and the output unit. Then, the control unit divides the evaluation region of the face image to be evaluated obtained from the input unit by the reference number of pixels, applies the learning result to the divided evaluation target divided image, acquires the individual evaluation division, and obtains the individual evaluation. Based on the segment, predict the valuation segment of the valuation area.

In addition, as a technique for supporting rock mass evaluation in the preliminary survey stage, a rock mass evaluation method described in Non-Patent Document 1 is being researched. The rock mass evaluation method described in Non-Patent Document 1 predicts the rock mass class using a neural network (ANN: Artificial Neural Network). Specifically, "elastic wave velocity", "rock type", and "overburden" are used as input data, and "ground grade" is output as output data.

JP 2019-023392 A

Chitane Masuda, 3 others, "Evaluation results of rock mass in mountain tunnel classified into sedimentary rock and igneous rock using neural network", Proceedings of the 47th Rock Mechanics Symposium, Japan Society of Civil Engineers, January 2020, Lecture number 7, p.34-39

The technique according to Patent Document 1 needs to use a large number of high-quality captured images (high-resolution images with appropriate illumination at the time of shooting) for learning, so it takes a lot of time to prepare. There's a problem. In addition, since the captured image of the face indirectly expresses information related to the deformation and stability of the ground, which affects rock mass determination, the accuracy of rock mass determination based on the captured image of the face is not necessarily high. No, there were still cases where the rock mass evaluation was revised later. In addition, the technology according to Non-Patent Document 1 does not realize the rock mass determination accuracy required at the construction stage because the rock mass determination is performed only from the information obtained from the ground rock preliminary survey. In other words, there are constraints on the budget and time spent on preliminary surveys, and there are many sites where the amount of data obtained from preliminary surveys is insufficient due to technical limitations in geological surveys.

From this point of view, the present invention provides a natural ground condition learning device, a natural mountain condition determination device, a natural mountain condition learning method, and a natural mountain condition determination method that can realize rock mass determination with high accuracy in a short time. .

A rock condition learning device according to the present invention is a rock condition learning device that learns the condition of a tunnel face and includes a learning data acquisition unit and a learning processing unit.
The learning data acquiring unit acquires, as learning data, a set of pre-construction survey data at the position of the tunnel face, construction data obtained during construction, and ground grade.
The learning processing unit has a learning device, and machine-learns the learning device so as to output the ground grade by inputting the survey data and the data at the time of construction.

In the natural ground condition learning device according to the present invention, the learner performs machine learning using learning data including both information obtained in the investigation stage before construction and information obtained in the construction stage. Therefore, the effectiveness of the learning data is higher than in the conventional method, and learning can be performed in a short period of time. Further, by using a learning device trained by this method, it is possible to realize rock mass determination with high accuracy.

The ground grade is, for example, a graded classification of ground conditions, and is preferably the one finally adopted in actual construction (corresponding to the implemented support pattern).
Moreover, the survey data and the data at the time of construction may be, for example, information that is generally used in determining the rock mass grade, and is information that has some correlation with the rock mass grade.
The survey data preferably include any one of ground seismic velocity, overburden and rock type at the location of the tunnel face. The data at the time of construction preferably include any one of the amount of pushing displacement of the face and the frequency of falling rocks. Note that the elastic wave velocity and rock type may be those measured during construction (that is, the elastic wave velocity and rock type may be included in construction data).

The elastic wave velocity and the overburden may be reference values or statistical values for predetermined sections divided in the tunnel axial direction, and the rock type may be a representative rock type for the predetermined sections divided in the tunnel axial direction.
The rockfall frequency is the number of rockfalls that occur in the tunnel face within a predetermined time after the excavation work is completed, and the amount of pushing displacement of the face is determined in advance after the excavation work is completed. It may be a measured value obtained by measuring the tunnel face within a period of time.

A rock condition determination device according to the present invention is a rock condition determination device that determines the condition of a tunnel face and includes a determination data acquisition unit and an estimation processing unit.
The determination data acquisition unit acquires pre-construction survey data and construction-time data obtained during construction at the position of the tunnel face.
The estimation processing unit has a learned learner that has performed machine learning so as to output a rock mass grade by inputting the survey data and the data at the time of construction.

In the rocky ground condition determination device according to the present invention, a learner that is machine-learned using learning data including both information obtained in the investigation stage before construction and information obtained in the construction stage is used. It is possible to realize good rock mass determination.

A rock condition learning method according to the present invention is a rock condition learning method for learning conditions of a tunnel face. This ground condition learning method has a learning data acquisition step and a learning processing step.
In the learning data acquisition step, a set of pre-construction survey data at the tunnel face position, construction data obtained during construction, and ground grade is acquired as learning data.
In the learning processing step, the learning device is machine-learned so as to output the rock mass class by inputting the survey data and the data at the time of construction into the learning device.

In the rock ground condition learning method according to the present invention, the learner performs machine learning using learning data including both information obtained in the survey stage before construction and information obtained in the construction stage. Therefore, the effectiveness of the learning data is higher than in the conventional method, and learning can be performed in a short period of time. Further, by using a learning device trained by this method, it is possible to realize rock mass determination with high accuracy.

A rock condition determination method according to the present invention is a rock condition determination method for determining the condition of a tunnel face. This ground condition determination method has a determination data acquisition step and an estimation processing step.
In the determination data acquisition step, pre-construction survey data at the tunnel face position and construction-time data obtained during construction are acquired.
In the estimation processing step, the rock mass grade is estimated from the acquired data using a learning device that has undergone machine learning so as to output the rock mass grade by inputting the survey data and the construction data.

In the ground condition determination method according to the present invention, a learner that is subjected to machine learning using learning data including both information obtained in the investigation stage before construction and information obtained in the construction stage is used. It is possible to realize good rock mass determination.

According to the present invention, rock mass determination can be realized with high accuracy in a short period of time.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure for demonstrating the outline|summary of this invention, (a) is an image which shows the conventional standard determination method, (b) is an image which shows the determination method by this invention. 1 is a configuration diagram of a ground condition learning system according to an embodiment of the present invention; FIG. 1 is a configuration diagram of a ground condition determination system according to an embodiment of the present invention; FIG. It is an external view of a face extrusion displacement meter. 1 is an external view of a falling rock detection device; FIG. 1 is a functional configuration diagram of a natural ground condition learning device and a natural mountain condition determination device according to an embodiment of the present invention; FIG. FIG. 2 is an image diagram of a neural network;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments for carrying out the present invention will be described in detail with appropriate reference to the drawings. Each drawing is merely a schematic representation to the extent that the present invention can be fully understood. Accordingly, the present invention is not limited to the illustrated examples only. In addition, in each figure, the same code|symbol is attached|subjected about a common component and a similar component, and those overlapping description is abbreviate|omitted.

[Overview of the present invention]
An outline of the present invention will be described with reference to FIG. FIG. 1 is a diagram for explaining the outline of the present invention. As shown in Fig. 1(a), until now, the geological condition of the face was recorded as a face observation record, and an evaluation score was assigned based on the face observation record. selected the rock mass grade (new face evaluation point method). In addition, the results of measurements taken during construction were used as a reference for selecting the ground grade. On the other hand, as shown in FIG. 1(b), the inventors of the present invention have obtained survey data (e.g., elastic wave velocity, rock type, overburden) obtained in advance geological surveys, as well as data that can be obtained at the construction stage. We devised machine learning using construction data (for example, the amount of displacement of the face and the frequency of falling rocks) that is directly linked to the deformation and stability of the ground, which affects the determination of the ground. A detailed description will be given below.

<Regarding the configuration of the ground condition learning system according to the embodiment>
A natural ground condition learning system 1A according to the embodiment will be described with reference to FIG. FIG. 2 is a configuration diagram of the ground condition learning system 1A according to the embodiment. The natural ground condition learning system 1A is a system for learning the conditions of the tunnel face. The natural ground condition learning system 1A can be used for various types of tunnels without limiting the types of tunnels. In this embodiment, a mountain tunnel will be described as an example of a tunnel.

As shown in FIG. 2, the natural ground condition learning system 1A mainly includes management terminals 2, 2, . . . The management terminal 2 and the natural ground condition learning device 3 can communicate with each other via the network 4 . The natural mountain condition learning system 1A can also be configured as a cloud system, in which case the natural mountain condition learning device 3 corresponds to a cloud server. Note that the configuration of the natural ground condition learning system 1A shown in FIG. 2 is merely an example.

The management terminal 2 is installed, for example, in an office outside the tunnel, and is operated by a tunnel construction manager or the like. The management terminal 2 stores various information related to tunnel construction. The management terminal 2 stores the results of geological surveys such as geophysical surveys and boring surveys conducted as preliminary surveys, for example. The information stored in the management terminal 2 is, for example, the elastic wave velocity of the natural ground, soil cover, rock type, and the like. The management terminal 2 also stores information about the tunnel face obtained during construction of the tunnel. The information stored in the management terminal 2 includes, for example, the amount of extrusion displacement of the tunnel face, the frequency of falling rocks, and the ground grade. The ground grade is a stepwise classification of ground conditions, and is, for example, the one finally adopted in actual construction (corresponding to the implemented support pattern).

The rock condition learning device 3 plays a central role in the rock condition learning system 1A, and learns the condition of the tunnel face using information collected from each construction site. The rock condition learning device 3 receives, for example, the elastic wave velocity, overburden and rock type at the position of the tunnel face from the management terminal 2 used at the construction site of a tunnel under construction, as well as the extrusion displacement amount of the tunnel face, falling rocks, etc. Collect frequencies and rock grades (see open arrows in Figure 2). Then, the natural ground condition learning device 3 uses the collected information to cause the learner (model) to perform machine learning. Details of the learning method will be described later. The learned learner (learned model) learned by the rock condition learning device 3 is delivered to, for example, a construction site of a new tunnel, and used for ground rock determination of the tunnel (see the dotted arrow in FIG. 2). .

A device that receives a trained learner trained by the rock condition learning device 3 and is capable of determining the condition of the tunnel face using the learned learner is called a “rock condition determination device”. In the present embodiment, it is assumed that the learned learning device is reflected in the management terminal 2, so it is described as "rock condition determination device 2B". The natural ground condition learning device 3 can also cause the learner to perform machine learning using the information collected from the natural mountain condition determination device 2B (see the stripe pattern arrow in FIG. 2). Further, the rock mountain condition learning device 3 corresponding to the cloud server may be configured to determine the condition of the tunnel face using a learned learning device. ” (that is, the rock ground condition determination device may be on the cloud).

<Regarding the configuration of the ground condition determination system according to the embodiment>
A rock ground condition determination system 1B according to the embodiment will be described with reference to FIG. FIG. 3 is a configuration diagram of the rock ground condition determination system 1B according to the embodiment. The rock condition determination system 1B is a system for determining the condition of the tunnel face. The natural ground condition determination system 1B can be used for various types of tunnels without limiting the types of tunnels. In this embodiment, a mountain tunnel will be described as an example of a tunnel. It is desirable that the tunnel determined by the natural ground condition determination system 1B be similar to the tunnel for which learning was performed by the natural mountain condition learning system 1A. The similarity of tunnels is intended to include various similarities such as, for example, similarities in cross-sectional shape and cross-sectional area, and similarities in ground conditions.

As shown in FIG. 3, the rock condition determination system 1B includes a rock condition determination device 2B having a learned learning device, a three-point face extrusion displacement gauge 5, falling rock detectors 6, 6, and a A worker terminal 7 is provided. The ground state determination device 2B and the worker terminal 7 can communicate with each other via a network provided in the mine, for example. The face extrusion displacement gauge 5 and the falling rock detectors 6 and 6 may be configured to be able to communicate with the rock condition determination device 2B and the worker terminal 7 . Note that the configuration of the natural ground condition determination system 1B shown in FIG. 3 is merely an example.

The face extrusion displacement meter 5 is a device that measures the amount of extrusion displacement of the tunnel face (that is, the amount of displacement in the axial direction of the tunnel). The face extrusion displacement meter 5 is placed at a position a predetermined distance away from the tunnel face, and measures the amount of tunnel face extrusion displacement that occurs within a predetermined time (for example, 10 minutes) after the completion of the excavation work, for example. do. FIG. 4 is an external view of the face extrusion displacement meter 5. As shown in FIG. As shown in FIG. 4, the face extrusion displacement gauge 5 of this embodiment includes three laser displacement gauges 51 to measure the extrusion displacement amount at three points in the central portion of the tunnel face. The worker registers, for example, the average value of the displacement amounts measured at three points in the rock mass condition determination device 2B as the push displacement amount of the tunnel face. The amount of extrusion displacement of the tunnel face is an example of an index that indicates the deformability of the tunnel face. The index indicating the deformability of the tunnel face may be, for example, the inner space displacement amount and crest settlement amount (and their initial displacement) obtained by measurement work (underground measurement). For example, it is the amount of displacement for one day after setting the target, or the amount of displacement in a determined excavation length (for example, 3 m or 4 m, which is the length that can be excavated in one day).

In this embodiment, it is assumed that the amount of extrusion displacement is registered using the terminal 7 for workers. Registration to the determination device 2B may be performed. In addition, the face extrusion displacement meter 5 may measure the amount of displacement at a plurality of points other than one point or three points (that is, the number of measurement points is not limited), and information to be registered when measuring a plurality of points may be a representative value (ie any one measurement) or a statistical value. Moreover, although the face extrusion displacement meter 5 of the present embodiment is a self-supporting type provided with the legs 52, it may be attached to the ceiling or wall of the tunnel.

The falling rock detection device 6 is a device for detecting falling rocks on the tunnel face. A rockfall detector 6 is placed near the tunnel face and monitors rockfall on the tunnel face. In this embodiment, two falling rock detectors 6 are used to monitor the tunnel face. That is, the first falling rock detector 6a monitors the right half of the tunnel face, and the second falling rock detector 6b monitors the left half of the tunnel face. The falling rock detection device 6 is installed, for example, within a predetermined time after completion of the excavation removal work (for example, 10 minutes after the completion of the excavation removal work) as a monitoring period, and detects falling rocks that occur during the monitoring period. do.

FIG. 5 is an external view of the falling rock detection device 6. As shown in FIG. As shown in FIG. 5, the falling rock detection device 6 of the present embodiment includes a camera 61 and lighting fixtures 62, 62 provided above and below the camera 61, and uses images taken by the camera 61 to detect falling rocks. do. The lighting fixture 62 is, for example, LED (Light Emitting Diode) lighting or organic EL lighting (OLED), and includes at least a wavelength component corresponding to the sensitivity of the camera 61 (camera sensitivity). The camera 61 is, for example, a digital video camera, and shoots several tens of images (also called "frames") per second. A control unit (not shown) of the falling rock detection device 6 detects falling rocks using, for example, an inter-frame difference method. In the inter-frame subtraction method, an inter-frame subtraction image is generated between two consecutive frames, and rockfall is detected using the inter-frame subtraction image. Note that the technique for detecting falling rocks from an image is not limited to this.

For example, the worker uses the total number of times that the first falling rock detection device 6a and the second falling rock detection device 6b detect falling rocks within the monitoring period as the number of times that falling rocks have occurred at the tunnel face (the number of falling rocks), and judges the ground condition. Register with device 2B. The number of rock falls on the tunnel face is an example of an index that indicates the stability of the tunnel face.

In this embodiment, it is assumed that the number of rockfalls is registered using the terminal 7 for workers. 2B may be registered. In addition, one or three or more falling rock detectors 6 may be used to detect falling rocks on the tunnel face, or a worker may visually detect falling rocks on the tunnel face without using the falling rock detectors 6. . Information indicating the frequency of rockfalls other than the number of rockfalls per predetermined time (including the number of rockfalls per unit time) (for example, information obtained by classifying the number of rockfalls by level) may be registered in the rocky ground condition determination device 2B. good. Moreover, although the falling rock detection device 6 of the present embodiment is a self-supporting device having the legs 63, it may be attached to the ceiling or wall of a tunnel.

The rock condition determination device 2B is a device that plays a central role in the rock condition determination system 1B. The information is used to determine tunnel face conditions. The natural ground condition determination device 2B stores, for example, the elastic wave velocity of the natural ground at the position of the tunnel face, the overburden, and the rock type information. Further, the rock condition determination device 2B acquires the pushing displacement amount of the tunnel face from the face pushing displacement meter 5, for example, and also acquires the rockfall frequency of the tunnel face from the rockfall detector 6. FIG. Then, the rock mass condition determination device 2B obtains the rock mass class of the tunnel face by inputting this information into the learned learning device. Details of the determination method will be described later. The rock mass condition determination device 2B transmits information on the determined rock mass class to the worker terminal 7, for example.

The worker terminal 7 is, for example, a tablet terminal, and can browse information related to tunnel construction stored in the management terminal 2 . In addition, the worker inputs the pushing displacement amount measured by the face pushing displacement meter 5 and the falling rock frequency detected by the falling rock detector 6 into the worker terminal 7, and the worker instructs execution of tunnel face judgment. , the input information is transmitted to the rock mountain condition determination device 2B. In addition, the judgment result of the situation of the tunnel face (information on the grade of rock) is displayed on the worker terminal 7, and the worker can determine the support pattern to be implemented based on the judgment result (information on the grade of rock). ).

<Regarding the rock ground condition learning method according to the embodiment>
A ground condition learning method according to the embodiment will be described with reference to FIG. FIG. 6 is a functional configuration diagram of the rock mountain condition learning device 3 and the rock mountain condition determination device 2B according to the embodiment. When the natural ground condition learning device 3 itself determines the condition of the tunnel face, the natural mountain condition learning device 3 and the natural mountain condition determination device 2B can be configured as one device.

The natural ground condition learning device 3 includes a learning data acquisition unit 31 and a learning processing unit 32, and the learning processing unit 32 has a learning device. A learner is a learning system in machine learning that compares the results of classification, prediction, and judgment based on the given data with the actual results that are correct answers, and leads to good results by adjusting various parameters. becomes possible. The learning device is, for example, a neural network, and the present embodiment also assumes a neural network. A learning device is also called a “learning model”.

As is well known, a neural network is an information processing technique that was created for the purpose of making a computer imitate the movement of the human brain. Neural networks are good at complex nonlinear processing, and have the feature that they do not need to formulate the relationship between explanatory variables and target variables by using learning functions. As shown in FIG. 7, a neural network generally consists of an input layer, an intermediate layer, and an output layer. FIG. 7 is an image diagram of a neural network. An explanatory variable is input to the input layer, and an objective variable is output from the output layer. The middle layer plays a role of connecting the two. The number of intermediate layers and the number of neurons in each intermediate layer are not restricted and can be set arbitrarily.

The learning data acquisition unit 31 acquires learning data for making the learning device learn. The learning data is composed of a set of ground elastic wave velocity, rock type, overburden, extrusion displacement, rockfall frequency and ground grade at the tunnel face position. Although it is desirable that the correspondence of the learning data is strict, it is not necessarily strict, and there may be some deviation in the position of the tunnel face between each data. The learning data is preferably created based on, for example, construction results when similar tunnels were constructed.

The elastic wave velocity is a reference value or statistical value for a predetermined section divided in the tunnel axial direction. The elastic wave velocity may be a value obtained from a geological survey in advance, a measured value measured at intervals of 10 m in the tunnel axis direction, or a value calculated by statistically processing the measured value.
The rock type is the representative rock type of the predetermined section divided in the tunnel axis direction. The type of rock may be determined from a preliminary geological survey, or may be determined by observing the ground exposed in the tunnel or excavation muck. Numerical values are associated with rock types, and the associated numerical values are used as learning data. The correspondence between the rock type and the numerical value is, for example, "1" for "igneous rock" and "2" for "sedimentary rock".
The soil cover is a reference value or statistical value for a predetermined section divided in the tunnel axis direction. For soil cover, use values obtained from the results of preliminary geological surveys (longitudinal maps, etc.).

The amount of extrusion displacement is the amount of displacement per predetermined time that occurs in the tunnel face during tunnel construction. The pushing displacement amount is measured, for example, within a predetermined time (for example, 10 minutes) after completion of the excavation work.
The frequency of falling rocks is the number of falling rocks that occur around the tunnel face per predetermined time during tunnel construction. is the number of rockfalls that occurred at the tunnel face.
The rock mass class is the rock mass class actually adopted at the tunnel construction site, and is obtained, for example, for each predetermined section divided in the tunnel axial direction. Numerical values are associated with rock mass grades, and the associated numerical values are used as learning data. The correspondence relationship between the rock mass class and the numerical value is, for example, “D rock mass” is “1”, “CII rock mass” is “2”, “CI rock mass” is “3”, and “B rock mass” is “4”. and so on.

The learning processing unit 32 uses learning data configured as a set of elastic wave velocities of natural ground at the tunnel face position, rock type, overburden, amount of extrusion displacement, frequency of falling rocks, and natural ground grades, and uses the learning data to machine the learning device. let them learn The method of learning the learner is not particularly limited, and for example, the backpropagation method is used to learn the learner. Specifically, the five pieces of information in the training data (elastic wave velocity, rock type, overburden, face extrusion displacement, and rockfall frequency) are input to the input layer, and the results output from the output layer and the rock mass class of the training data are input. Adjust the intermediate layer based on the error with In other words, the learning processing unit 32 inputs the elastic wave velocity, the type of rock, the overburden, the amount of face extrusion displacement, and the frequency of falling rocks, and causes the learning device to perform machine learning so that an appropriate rock mass class is output. The number of pieces of learning data to be learned is not particularly limited, and learning ends when, for example, the expected accuracy (percentage of correct answers) is reached. The learned learning device is delivered to the natural ground condition determination device 2B.

<Regarding the rock ground condition determination method according to the embodiment>
A rock ground condition determination method according to the embodiment will be described with reference to FIG.
The rocky ground condition determination device 2B includes a determination data acquisition unit 21 and an estimation processing unit 22, and the estimation processing unit 22 has a learned learning device. The trained learner is a learner machine-learned by the rock condition learning device 3 . In other words, the trained learner is trained to output an appropriate ground grade by inputting elastic wave velocity, rock type, overburden, face extrusion displacement, and rockfall frequency.

The determination data acquisition unit 21 acquires determination data that serves as a basis for determining the rock mass class. The judgment data is composed of a set of ground elastic wave velocity, rock type, overburden, extrusion displacement amount and rock fall frequency at the tunnel face position. Although it is desirable that the correspondence relationship of the determination data is strict, it is not necessarily strict, and there may be some deviation in the position of the tunnel face between each data. The information of the elastic wave velocity, rock type, soil cover, extrusion displacement amount, and rockfall frequency that constitute the determination data may be measured by the same method as the learning data.

The estimation processing unit 22 estimates the state of the tunnel face using determination data configured as a set of seismic velocity of the ground at the position of the tunnel face, rock type, overburden, amount of displacement, and frequency of falling rocks. A learned learner is used for estimating tunnel face information. As shown in FIG. 7, five pieces of information are input to the input layer: elastic wave velocity, rock type, overburden, face extrusion displacement, and rockfall frequency. The output layer also outputs information on rock mass grades. The rock mass class information output from the output layer is, for example, output as a probability of each rock mass class (for example, rock mass D, rock mass CII, rock mass CI, rock mass B).

As described above, in the natural ground condition learning device 3 according to the present embodiment, learning data including both information (survey data) obtained in the survey stage before construction and information obtained in the construction stage (data during construction) is used to machine-learn the learner. Therefore, the effectiveness of the learning data is higher than in the conventional method, and learning can be performed in a short time. It is feasible.

Although the embodiment of the present invention has been described above, the present invention is not limited to this, and can be implemented within the scope of the claims.

In the embodiment, learning is performed using learning data including five items of elastic wave velocity, rock type, soil cover, extrusion displacement amount, and rock fall frequency. I was judging the situation. However, the configurations of learning data and determination data are not limited to this.
For example, learning can be performed by preparing a learner for each rock type and inputting learning data including elastic wave velocity, overburden, extrusion displacement, and falling rock frequency into the learner prepared for each rock type. It is also possible to In this case, by inputting judgment data including elastic wave velocity, overburden, extrusion displacement, and falling rock frequency into the learning device of the relevant rock type, the condition of the natural ground is judged.
In addition, machine learning is performed on the learner using learning data that includes either information on the deformation of the ground (the amount of extrusion displacement of the tunnel face) or information on the stability (rockfall frequency) that affects rock ground judgment. may It is desirable to make the learner perform machine learning using learning data containing information on stability (rockfall frequency).

1A rock condition learning system 1B rock condition determination system 2 terminal for management 2B rock condition determination device 3 rock condition learning device 5 face extrusion displacement meter 6 falling rock detection device 7 worker terminal 21 determination data acquisition unit 22 estimation processing Unit 31 Learning data acquisition unit 32 Learning processing unit

Claims (8)

A natural ground condition learning device for learning the condition of a tunnel face, comprising a learning data acquisition unit and a learning processing unit,
The learning data acquisition unit acquires, as learning data, a set of survey data before construction at the position of the tunnel face, data during construction obtained during construction, and a ground grade,
The learning processing unit has a learning device, and machine-learns the learning device so as to output the ground grade by inputting the survey data and the construction data.
A natural ground condition learning device characterized by: The survey data includes any one of the elastic wave velocity of the ground at the position of the tunnel face, the overburden and the rock type,
The natural ground condition learning device according to claim 1, characterized in that: The data at the time of construction includes any one of the amount of extrusion displacement of the face and the frequency of falling rocks,
3. The natural ground condition learning device according to claim 1 or 2, characterized in that: The elastic wave velocity and the overburden are reference values or statistical values of predetermined sections divided in the tunnel axial direction,
The rock type is a representative rock type of a predetermined section divided in the tunnel axis direction,
3. The natural ground condition learning device according to claim 2, characterized in that: The rockfall frequency is the number of rockfalls that occurred at the tunnel face within a predetermined time after the excavation work was completed,
The amount of extrusion displacement of the face is a measurement value obtained by measuring the tunnel face within a predetermined time after the completion of the excavation work,
4. The natural ground condition learning device according to claim 3, characterized in that: A rock condition determination device for determining the condition of a tunnel face, comprising a determination data acquisition unit and an estimation processing unit,
The determination data acquisition unit acquires survey data before construction at the position of the tunnel face and data during construction obtained during construction,
The estimation processing unit has a learned learner that has performed machine learning so as to output a rock mass grade by inputting the survey data and the construction data.
A rock ground condition determination device characterized by: A rock condition learning method for learning the condition of a tunnel face,
a learning data acquisition step of acquiring, as learning data, a set of pre-construction survey data at the position of the tunnel face, construction data obtained during construction, and a ground grade;
a learning processing step of machine-learning the learning device so as to output the rock mass grade by inputting the survey data and the construction data into the learning device;
A ground condition learning method characterized by: A rock ground condition determination method for determining the condition of a tunnel face,
a determination data acquisition step of acquiring pre-construction survey data at the position of the tunnel face and construction-time data obtained during construction;
an estimation processing step of estimating a rock mass grade from data acquired using a learner that has undergone machine learning so as to output a rock mass grade by inputting the survey data and the construction data; have
A rock ground condition determination method characterized by:

Images