KR20220089666A - Device for the rock mass classification in tunnel design using AI and the rock mass classification in excavation - Google Patents

Abstract

The present invention discloses a rock classification device for tunnel design using artificial intelligence technology and rock classification results during excavation. The bedrock classification apparatus of the present invention collects the ground information collected through the ground survey for tunnel design and the learning data collection unit that collects the actual rock classification results of the makjang that appeared during the excavation of the tunnel, and the information collected from the learning data collection unit. A learning model generator that generates a learning model related to bedrock classification by learning, and when a new tunnel design is in progress, the results of the ground investigation for the relevant area are input to the learning model, and the final rock classification results are obtained for each section of the tunnel route. It includes a data analysis unit to analyze each.

Description

Device for the rock mass classification in tunnel design using AI and the rock mass classification in excavation

The present invention relates to a rock classification technology, and more particularly, the accuracy of rock classification when designing other tunnels and underground spaces by analyzing the ground investigation results at the time of designing an already excavated tunnel and the rock classification results at the time of excavation with artificial intelligence technology It is related to an artificial intelligence technology to improve the performance and to a rock classification device when designing a tunnel using the results of rock classification during excavation.

In general, it is very important in terms of safety and economics to understand the condition of bedrock in the design of underground spaces such as tunnels. Therefore, an engineering method that can objectively evaluate the condition of bedrock is required, and this method is called bedrock classification.

Among the rock classification methods, the most commonly used was Z.T. These are the Rock Mass Rating (RMR) proposed by Bieniawski and the Q value proposed by Nick Barton in 1975. The RMR is calculated by adding up the points assigned to each of the five items, including the strength of the new rock, the Rock Quality Designation (RQD), the joint (discontinuous) interval, the joint (discontinuous) state, and the groundwater state. The final score is calculated by making corrections. The Q value is calculated as a combination of the Rock Quality Designation (RQD), the number of joints, the roughness of the joint, the degree of deterioration of the joint, the groundwater reduction factor in the joint, and the stress reduction coefficient.

When designing a tunnel, ground surveys such as drilling surveys, electrical resistivity surveys, seismic surveys, topography surveys, and rock tests are conducted for tunnel sections, and through these, electrical resistivity measurements, seismic wave velocity, rock types (rock types), and each point of the tunnel are conducted. Rock characteristics such as the depth from the star surface to the tunnel and the strength of the rock are identified.

As shown in FIG. 1 , the electrical resistivity survey was conducted after deriving a correlation expression by analyzing the results of the electrical resistivity survey conducted for all or some sections of the tunnel and the results of the drilling survey conducted only for some points of the tunnel. The electrical resistivity values measured for other sections in the same tunnel where rock classification could not be performed because no investigation was conducted were input into the above-mentioned correlation equation to derive the rock classification results (RMR or Q value) for each section on the tunnel route.

In this case, the rock classification results for most of the tunnel sections where no drilling has been performed are determined only by the electrical resistivity survey results, which is very different from the actual rock quality that appears during tunnel excavation. There is a disadvantage that shows a considerable difference from this reality.

Korean Patent Application Laid-Open No. 10-2021-0105307 (2021.08.26.)

The technical task to be achieved by the present invention is an artificial intelligence technology that improves the accuracy of rock classification when designing other tunnels and underground spaces by analyzing the ground investigation results at the time of designing an already excavated tunnel and the rock classification results at the time of excavation with artificial intelligence technology. The purpose of this study is to provide a rock classification device when designing a tunnel using the rock classification results during excavation and excavation.

In order to achieve the above object, in the tunnel design using the artificial intelligence technology and the rock classification results during excavation according to the present invention, the rock classification device includes the ground information collected through the ground survey for tunnel design and the A learning data collection unit that collects actual rock classification results, a learning model generation unit that learns the information collected from the learning data collection unit and generates a learning model related to rock classification, and when a new tunnel design is in progress, A data analysis unit for inputting the investigated ground investigation results into the learning model and analyzing the final rock classification results for each predetermined section of the tunnel route, wherein the learning data collection unit includes: The results of the ground investigation including at least one of exploration, drilling, seismic exploration, the type of rock, and the depth from the surface to the top of the tunnel, and the actual rock classification results of the makjang shown during excavation of the tunnel are collected, and the collected information A preliminary rock classification result is derived based on the results of the electrical resistivity survey conducted for the entire section or part of the tunnel and the drilling investigation conducted for only some points, and the learning model generation unit is configured to include the preliminary rock classification result and the preliminary rock classification result At least one learning model is generated by taking other ground investigation results as input data, and performing learning using the actual rock classification results of the makjang as output results, and the data analysis unit, when designing a new tunnel, Among the ground survey results, the preliminary rock classification results derived for each section on the tunnel route based on the electrical resistivity survey results and the drilling survey results and other ground survey results other than the preliminary rock classification results are input to the learned learning model and It is characterized in that the final rock classification results are analyzed for each specific section.

In addition, the learning data collection unit, the preliminary rock classification results for each section on the tunnel route derived based on the results of the electrical resistivity survey and the drilling survey results, the seismic velocity analyzed in the seismic survey, the type of rock, and the depth from the surface to the top of the tunnel It is characterized by deriving data corresponding to the rock classification result of each makjang that appears during excavation of the corresponding tunnel for the same section from at least one result.

In addition, the learning model generation unit includes preliminary rock classification results, seismic wave velocity by seismic wave exploration in the tunnel section, depth from the surface to the tunnel at each point of the tunnel route, types of rocks, and actual rock classification results of the rock formations displayed during excavation of the tunnel. It is characterized in that the learning model is generated by learning based on at least one piece of information.

In addition, the learning model generation unit learns a plurality of predictive models using a plurality of different data samples extracted from a plurality of training data for which labels are set, and generates a learning model using the learned plurality of predictive models. characterized.

In addition, the learning model generation unit trains a prediction model corresponding to a plurality of data samples extracted from a plurality of training data with labels set, and inputs the plurality of training data to the learned prediction model. Generating the next training data by multiplying the data by a preset weight, repeating the process of generating the next training data to generate a plurality of predictive models, and generating a learning model using the generated plurality of predictive models characterized.

In addition, the learning model generating unit is characterized in that it learns a plurality of different predictive models by using a plurality of training data with labels set in common, and generates a learning model by using the learned plurality of predictive models.

In addition, the learning model generator, Artificial Neural Networks (ANN), bagging (Bootstrap Aggregating, Bagging), boosting (Boosting), voting (Voting) and SVM (Support Vector Machine) learning using at least one technique It is characterized by creating a model.

In addition, when designing a new tunnel, the data analysis unit derives preliminary rock classification results for the new tunnel, and includes the preliminary rock classification results and seismic velocity derived through seismic exploration, from the surface at each point of the tunnel route to the tunnel. It is characterized in that the final rock classification result in the new tunnel is analyzed by inputting the ground investigation result including at least one of the depth and the type of rock to the learning model.

When designing a tunnel using the artificial intelligence technology according to the present invention and the rock classification result at the time of excavation, the rock classification device uses the artificial intelligence technology to convert the ground investigation result at the time of designing the already excavated tunnel and the rock classification result (RMR or Q value) at the time of excavation. By analyzing the RMR or Q value for each section of the tunnel route by applying it to , it is possible to derive more accurate evaluation results than the conventional rock classification results, which were derived only from the electrical resistivity survey results and drilling survey results collected at the time of design.

Through this, the present invention can reduce the error in the tunnel construction cost and period estimated at the time of design.

1 is a view for explaining a conventional rock classification process.
2 is a block diagram for explaining an apparatus for classifying rock according to an embodiment of the present invention.
3 is a view for explaining a rock classification process according to an embodiment of the present invention.
4 is a view for explaining a learning data collection unit according to an embodiment of the present invention.
5 is a diagram for explaining a process of generating a learning model using an artificial neural network according to an embodiment of the present invention.
6 is a diagram for explaining a process of generating a learning model using a bagging technique according to another embodiment of the present invention.
7 is a view for explaining a process of generating a learning model using a boosting technique according to another embodiment of the present invention.
8 is a diagram for explaining a process of generating a learning model using a voting technique according to another embodiment of the present invention.
9 is a flowchart for explaining a rock classification method according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. First, in adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known configuration or function is obvious to those skilled in the art or may obscure the gist of the present invention, the detailed description thereof will be omitted.

2 is a block diagram for explaining a rock classification apparatus according to an embodiment of the present invention, FIG. 3 is a diagram for explaining a rock classification process according to an embodiment of the present invention, and FIG. 4 is an embodiment of the present invention FIG. 5 is a diagram for explaining a process of generating a learning model using an artificial neural network according to an embodiment of the present invention, and FIG. 6 is another embodiment of the present invention. It is a view for explaining a process of generating a learning model using the bagging technique according to 8 is a diagram for explaining a process of generating a learning model using a voting technique according to another embodiment of the present invention.

2 to 8 , the rock classification apparatus 100 analyzes the ground investigation results at the time of designing the already excavated tunnel and the rock classification results at the time of excavation with artificial intelligence technology to classify the rock when designing other tunnels and underground spaces. improve the accuracy of The rock classification apparatus 100 includes an input unit 10 , a control unit 30 , an output unit 50 , and a storage unit 70 .

The input unit 10 receives the ground information related to the ground survey collected for designing a new tunnel. Here, the ground information is information indicating the characteristics of the bedrock, and may be information including, for example, electrical resistivity survey, drilling survey, seismic survey, type of rock, depth from the surface to the top of the tunnel, and the like. In addition, the input unit 10 derives preliminary rock classification results based on the results of the electrical resistivity survey conducted for all or some sections of the tunnel among the collected information and the results of the drilling survey conducted only for some points.

The control unit 30 performs overall control of the rock classification apparatus 100 . The control unit 30 collects the ground information investigated when designing the tunnel in which excavation has been completed in advance and the result of rock classification during actual excavation, and performs data pre-processing of the collected information. Data preprocessing may be a process of sorting, changing the format, etc. so that the collected information is suitable for analysis performed later. The control unit 30 generates a learning model related to bedrock classification by learning the preprocessed information. The control unit 30 may learn preprocessed information using various artificial intelligence technologies, and preferably, artificial neural networks (ANN), bootstrap aggregating (Bagging), boosting (Boosting), voting (Voting). ) and SVM (Support Vector Machine) may be used to generate a learning model using at least one technique. When a new tunnel design is in progress, the control unit 30 inputs the results of the ground investigation for the corresponding area into the generated learning model and analyzes the final rock classification results (RMR or Q value) for each predetermined section of the tunnel route, respectively. .

In detail, the control unit 30 includes a learning data collection unit 31 , a learning model generation unit 33 , and a data analysis unit 35 .

The learning data collection unit 31 collects ground data information for learning and pre-processes it. In detail, the learning data collection unit 31 includes at least one of electrical resistivity surveys, drilling surveys, seismic surveys, types of rocks, and depth from the surface to the top of the tunnel for a number of tunnel designs performed in the past. Collect the results and actual rock classification results of the Makjang during excavation of the corresponding tunnel. The learning data collection unit 31 derives preliminary rock classification results based on the results of the electrical resistivity survey conducted for all or some sections of the tunnel among the collected information and the results of the drilling survey conducted only for some points. At this time, the learning data collection unit 31 selects the result of at least one of the preliminary rock classification results for each predetermined section on the derived tunnel route and the seismic wave velocity analyzed in the seismic survey, the type of rock, and the depth from the surface to the top of the tunnel for the same section. The data corresponding to the rock classification results of each makjang that appeared during the excavation of the corresponding tunnel are derived. That is, the learning data collection unit 31 may pre-process the collected and derived information to fit the analysis to be performed later, as shown in FIG. 4 , to arrange the format of each information for each tunnel section.

The learning model generation unit 33 generates at least one learning model by using the preliminary rock classification result and the preliminary rock classification result as input data and other ground investigation results as input data, and performing learning using the actual rock classification result of the makjang as an output result. do. Here, the learning model is a model generated based on artificial intelligence technology, and preferably, at least one of artificial neural network, bagging, boosting, voting, and SVM may be used, but is not limited thereto.

For example, as shown in FIG. 5 , an artificial intelligence model may be generated using an artificial neural network. In this case, the artificial neural network includes a plurality of layers. The plurality of layers includes an input layer (IL), hidden layers (HL1 to HLk), and an output layer (OL).

In addition, each of the plurality of layers (IL, HL, OL) includes one or more nodes. For example, as illustrated, the input layer IL may include n input nodes i1 to in, and the output layer OL may include one output node v. In addition, among the hidden layers HL, the first hidden layer HL1 includes a nodes h11 to h1a, the second hidden layer HL2 includes b nodes h21 to h2b, and the kth hidden layer HL2 includes a number of nodes h21 to h2b. The layer HLk may include c nodes hk1 to hkc.

Then, a method of generating a learning model from the prototype of an artificial neural network will be described. The learning model generating unit 33 prepares as learning data at least one of the preliminary rock classification result, the seismic wave velocity by seismic exploration in the tunnel section, the depth from the surface to the tunnel at each point of the tunnel route, and the type of rock . When the learning data is prepared, the learning model generating unit 33 sets the actual rock classification result of the makjang that appears when the tunnel is excavated in response to the input data of the learning data as a label. Then, the learning model generating unit 33 converts the input data into an input feature vector, and inputs it to the prototype of the learning model, that is, the prototype of the artificial neural network. Then, the plurality of nodes of the plurality of layers apply the parameters of the artificial neural network including weights and thresholds to the calculation results corresponding to the input feature vectors, perform the calculation input to the calculation of the next layer, and finally calculate the output value. That is, the plurality of nodes of the plurality of layers of the artificial neural network perform a plurality of operations to which parameters of the artificial neural network including weights and thresholds are applied to the input feature vector, and finally calculate an output value. These output values are estimates of the actual rock classification results of Makjang.

The learning model generating unit 33 generates a learning model by repeatedly performing the optimization as described above using a plurality of different learning data until the prototype of the learning model reaches a predetermined accuracy through the evaluation index.

Next, a method for generating a learning model according to another embodiment of the present invention will be described. 6 is a diagram for explaining a method of generating a learning model through a bagging technique.

The learning model generating unit 33 provides a plurality of learning data fd. At this time, each of the learning data fd means a label corresponding to the input data, that is, the actual bedrock classification result data of the makjang displayed during excavation of the corresponding tunnel.

Then, the learning model generating unit 33 randomly selects a plurality of initially input training data fd (eg, a sample of size 4 such as {A, B, D, E}) and a plurality of data samples configured (eg, sd1, sd1, sd3) are provided.

Next, the learning model generating unit 33 prepares prototypes of a plurality of predictive models m1, m2, and m3 corresponding to each of the plurality of data samples. As illustrated, first to third prediction models m1 , m2 , and m3 are prepared corresponding to the first to third data samples sd1 , sd2 , and sd3 , respectively. The predictive model may be exemplified by an artificial neural network (ANN), a decision tree, or the like. However, the present invention is not limited thereto, and all types of models to which learning can be applied may be the predictive models of the present invention.

Next, the learning model generating unit 33 obtains a predicted value from each of the plurality of predictive models by inputting the input to the predictive model corresponding to each of a plurality of different data samples. That is, the learning model generator 33 inputs each of the first to third samples sd1, sd2, and sd3 into the prototype of the corresponding prediction models m1, m2, m3, and the first to third prediction models m1 , m2, m3) to obtain first to third predicted values P1, P2, and P3, respectively.

Next, the learning model generating unit 33 sets the prototype of the first to third predictive models m1, m2, and m3 such that the difference between each of the first to third predicted values P1, P2, and P3 and the corresponding label is minimized. Learning to optimize each parameter is performed to complete the first to third predictive models. Accordingly, the learning model generating unit 33 generates a learning model including a plurality of predictive models.

Next, a method for generating a learning model according to another embodiment of the present invention will be described. 7 is a view for explaining a method of generating a learning model through a boosting technique according to another embodiment of the present invention.

First, the learning model generating unit 33 prepares a plurality of first learning data fd1 that are initially input. At this time, each of the first learning data fd1 means a label corresponding to the input data, that is, the actual bedrock classification result data of the makjang displayed during the excavation of the corresponding tunnel.

Next, referring to FIG. 7 , the learning model generating unit 33 performs sampling by extracting a sample of a predetermined size from the plurality of first training data fd1 to obtain a plurality of first data samples sd1 {B D E I} to extract Next, the learning model generator 33 prepares a prototype of one first predictive model m1. According to the illustrated example, the prototype of the first predictive model m1 is prepared. As described above, the predictive model may exemplify an artificial neural network (ANN), a decision tree, and the like, and all kinds of algorithms to which learning is applied may be the predictive model.

Then, the learning model generating unit 33 inputs the plurality of first data samples sd1 to the prototype of the first prediction model m1, and the difference between the first predicted value P1 and the corresponding label is the minimum. The first predictive model m1 is completed by performing learning to optimize the circular parameters of the first predictive model m1 so that .

Next, the learning model generator 33 inputs all of the plurality of first learning data fd1 into the first predictive model m1 in order to verify the validity of the learned first predictive model m1, and the calculated prediction value ( Compare P1) with the corresponding label. Then, the learning model generating unit 33 excludes the correct answer learning data {B D E I} when the calculated predicted value P1 and the label are the same, and the incorrect answer learning data {A C F G H} different from the label is sampled again in the future model. After multiplying the training data for the incorrect answer by a predetermined weight so as to increase it, the corrected training data {A'C'F'G'H'} is replaced with the corresponding data {A C F G H} in the first learning data fd1 The second learning data fd2 is generated.

Then, the learning model generator 33 performs sampling by extracting a sample of a predetermined size from the plurality of second training data fd2 in the same manner as in the first prediction model m1, and a plurality of second data samples Extract (sd2){A'E F'H'}. Next, the learning model generator 33 prepares a prototype of one second predictive model m2. According to the illustrated example, a prototype of the second predictive model m2 is prepared. As described above, the predictive model may exemplify an artificial neural network (ANN), a decision tree, and the like, and all kinds of algorithms to which learning is applied may be the predictive model.

Then, the learning model generating unit 33 inputs the plurality of second data samples sd2 to the prototype of the second prediction model m2, and the difference between the second predicted value P2 and the corresponding label is the minimum. The second predictive model m2 is completed by performing training to optimize the circular parameters of the second predictive model m2 as much as possible.

Next, the learning model generator 33 inputs all of the plurality of second learning data fd2 into the second predictive model m2 to verify the validity of the learned second predictive model m2 and calculates Compare the predicted value (P2) with the corresponding label. Then, the learning model generation unit 33 excludes the correct answer learning data {A'B D G'I} when the calculated predicted value P2 is the same as the label, but the incorrect answer learning data {C'E F' In order to increase the likelihood that H'} will be re-sampled in the future model, the wrong answer training data is multiplied by a predetermined weight to make the corrected training data {C“ E' F”H“}, and then to the second training data fd2 The third training data fd3 is generated by replacing it with the corresponding data {C'E F'H'}.

Then, the learning model generator 33 sequentially performs the same series of processes as the data sample extraction, training, and validation processes performed in the first predictive model m1 and the second predictive model m2. repeated many times with

Next, when new data is input to the learning model generating unit 33 established through this process, the learning model generating unit 33 predicts the predicted model through each of the prediction models m1, m2, m3, ... The final result is obtained by multiplying the predicted values (P1, P2, P3,...) by a predetermined model weight (w1, w2, w3,...) corresponding to each prediction model (m1, m2, m3,...) derive the result

As described above, the learning model generator 33 completes the predictive model by learning the prototype of the predictive model using a plurality of data samples, and verifies the completed predictive model using a plurality of learning data. , the data samples with errors are weighted to increase the likelihood of being sampled in the future model, replacing the existing data with the existing data to create new training data, and then creating a plurality of sampled data and learning and verifying sequentially. repeat many times. In this way, when new data is input to the completed model, the final result is derived by multiplying the predicted value derived from each prediction model by a predetermined model weight set for each prediction model. Accordingly, a learning model including a plurality of predictive models is generated.

Next, a method for generating a learning model according to another embodiment of the present invention will be described. 8 is a diagram for explaining a method of generating a learning model through a voting technique according to another embodiment of the present invention.

Referring to FIG. 8 , the learning model generating unit 33 first prepares a plurality of learning data fd. At this time, each of the learning data fd means a label corresponding to the input data, that is, the actual bedrock classification result data of the makjang displayed during excavation of the corresponding tunnel.

Next, prototypes of a plurality of different prediction models (ma, mb, mc) are prepared. According to the illustrated example, prototypes of the predictive model A, the predictive model B, and the predictive model C (ma, mb, mc) may be provided. Here, a plurality of different prediction models (ma, mb, mc) refer to, for example, different types of models such as artificial neural networks and decision trees, or different types of algorithms such as CNNs and RNNs, for example, layers , means a model with a different algorithm structure, such as a node.

Then, the learning model generating unit 33 inputs the same training data fd to each of the prototypes of the plurality of prediction models ma, mb, and mc, and from each of the prototypes of the plurality of prediction models ma, mb, and mc. get the predicted value. That is, the learning model generating unit 33 inputs the training data fd into the prototypes of the predictive model A, the predictive model B, and the predictive model C (ma, mb, mc), and the predictive model A, the predictive model B and the predictive model C First to third predicted values (P1, P2, P3) are obtained from each of the prototypes of (ma, mb, mc).

Next, the learning model generation unit 33 is configured to minimize the difference between each of the first to third predicted values P1, P2, and P3 and the corresponding label. mc), learning to optimize each parameter is performed to complete predictive model A, predictive model B, and predictive model C (ma, mb, mc). Accordingly, a learning model including a plurality of predictive models is generated.

When designing a new tunnel, the data analysis unit 35 derives a final rock classification result for the new tunnel. That is, in the data analysis unit 35, the preliminary rock classification result in the new tunnel input from the input unit 10, the seismic wave velocity derived through seismic exploration, the depth from the surface to the tunnel at each point of the tunnel route, and the type of rock The final rock classification results are analyzed for each section on the tunnel route by inputting the ground survey results including at least one piece of information into the learned learning model. Here, the learning model means a learning model generated and learned by the learning model generating unit 33 . The data analysis unit 35 controls the final rock classification result, which is the analysis result, to be output through the output unit 50 .

The output unit 50 outputs the information input from the input unit 10 and outputs the analyzed final rock classification result from the control unit 30 .

The storage unit 70 stores a program or algorithm for driving the rock classification apparatus 100 . The storage unit 70 stores information input from the input unit 10 , and stores the final rock classification result analyzed from the control unit 30 . In addition, the storage unit 70 stores the learning data learned from the control unit 30 . The storage unit 70 includes a flash memory type, a hard disk type, a media card micro type, a card type memory (eg, SD or XD memory, etc.), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), Magnetic Memory, It may include at least one storage medium of a magnetic disk and an optical disk.

9 is a flowchart for explaining a method for classifying rocks according to an embodiment of the present invention.

2 and 9, the rock classification method applies the ground investigation result at the time of designing the already excavated tunnel and the rock classification result (RMR or Q value) at the time of excavation to AI technology for each predetermined section of the tunnel route. By analyzing the RMR or Q value, it is possible to derive more accurate evaluation results than the conventional rock classification results, which were derived only from the results of the electrical resistivity survey and the drilling survey results collected at the time of design. Through this, the rock classification method can reduce the error in the tunnel construction cost and period estimated at the time of design.

In step S110, the rock classification apparatus 100 collects the ground information collected through the ground survey for tunnel design and the actual rock classification result of the makjang displayed during excavation of the corresponding tunnel. At this time, the rock classification apparatus 100 includes at least one of electrical resistivity survey, drilling survey, seismic wave survey, rock type, and depth from the surface to the top of the tunnel for multiple tunnel designs performed in the past. Collect the actual rock classification results of the makjang that appeared during tunnel excavation, and derive preliminary rock classification results based on the results of the electrical resistivity survey conducted for the entire section or part of the tunnel and the results of the drilling investigation conducted only for some points among the collected information do.

In step S120, the bedrock classification apparatus 100 generates a learning model related to bedrock classification by learning the collected information. The rock classification apparatus 100 uses the preliminary rock classification result and the preliminary rock classification result as input data, and performs learning using the actual rock classification result of the makjang as an output result to generate at least one learning model. . At this time, the bedrock classification apparatus 100 may generate a learning model by using at least one of artificial neural networks, bagging, boosting, voting, and SVM.

In step S130, the rock classification apparatus 100 determines whether to proceed with the design of a new tunnel. The rock classification apparatus 100 performs step S140 when a new tunnel design is in progress, and terminates the system if not in progress.

In step S140, the rock classification apparatus 100 collects and inputs ground survey data in a new tunnel. The rock classification apparatus 100 includes at least one of information on preliminary rock classification results in the new tunnel, the seismic wave velocity derived through seismic surveying, the depth from the surface to the tunnel at each point of the tunnel route, and the type of rock. The results are input to the trained learning model.

In step S150, the bedrock classification apparatus 100 inputs the results of the ground investigation for a new tunnel-related area into the learning model and analyzes the final rock classification results for each predetermined section of the tunnel route, respectively. The rock classification apparatus 100 is a ground survey result other than the preliminary rock classification result and preliminary rock classification result derived for each section on the tunnel route based on the electric resistivity investigation result and the drilling investigation result among the previously conducted ground investigation results. By input to the learning model, the final rock classification results can be analyzed for each section on the tunnel route. The rock classification apparatus 100 outputs the final rock classification result, which is the analyzed result, to support the user to check the corresponding result.

The method according to an embodiment of the present invention may be provided in the form of a computer-readable medium suitable for storing computer program instructions and data. Such a computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination, and includes all types of recording devices in which data readable by a computer system is stored. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, compact disk read only memory (CD-ROM), and optical recording media such as DVD (Digital Video Disk). Stores program instructions such as Magneto-Optical Media, ROM (Read Only Memory), RAM (Random Access Memory), Flash memory, etc. and hardware devices specially configured to perform In addition, the computer-readable recording medium is distributed in a computer system connected through a network, so that the computer-readable code can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the technical field to which the present invention pertains.

Although the above has been described and illustrated in relation to the preferred embodiment for illustrating the technical idea of the present invention, the present invention is not limited to the configuration and operation as shown and described as such, and without departing from the scope of the technical idea. It will be apparent to those skilled in the art that many changes and modifications to the present invention are possible. Accordingly, all such suitable alterations and modifications and equivalents are to be considered as being within the scope of the present invention.

10: input
30: control unit
31: learning data collection unit
33: learning model generation unit
35: data analysis unit
50: output unit
70: storage
100: rock classification device

Claims (8)

a learning data collection unit that collects the ground information collected through the ground survey for tunnel design and the actual rock classification results of the makjang displayed during excavation of the tunnel;
a learning model generation unit for learning the information collected from the learning data collection unit and generating a learning model related to bedrock classification; and
When a new tunnel design is in progress, a data analysis unit that inputs the results of the ground investigation for the corresponding area into the learning model and analyzes the final rock classification results for each predetermined section of the tunnel route, respectively;
The learning data collection unit,
Ground survey results including at least one of electrical resistivity surveys, drilling surveys, seismic surveys, types of rocks, and the depth from the surface to the top of the tunnel for multiple tunnel designs conducted in the past, and Collect the actual rock classification results, and derive preliminary rock classification results based on the results of the electrical resistivity survey conducted for all or some sections of the tunnel and the results of the drilling investigation conducted for only some points among the collected information,
The learning model generation unit,
At least one learning model is generated by using the preliminary rock classification result and the preliminary rock classification result as input data, and performing learning using the actual rock classification result of the makjang as an output result,
The data analysis unit,
When designing a new tunnel, the preliminary rock classification results derived for each section on the tunnel route based on the electrical resistivity investigation results and the drilling investigation results among the previously conducted ground investigation results and the ground investigation results other than the preliminary rock classification results are learned above. A rock classification device when designing a tunnel using artificial intelligence technology and the rock classification results during excavation, characterized in that it analyzes the final rock classification results for each section on the tunnel route by inputting it into the trained learning model. The method of claim 1,
The learning data collection unit,
The results of at least one of the preliminary rock classification results for each section on the tunnel route derived based on the electrical resistivity survey results and the drilling survey results and the seismic wave velocity analyzed in the seismic survey, the type of rock, and the depth from the surface to the top of the tunnel are the same. A rock classification device when designing a tunnel using artificial intelligence technology and the rock classification result during excavation, characterized in that it derives data corresponding to the rock classification results of each block during excavation of the corresponding tunnel for each section. The method of claim 1,
The learning model generation unit,
Based on at least one of the preliminary rock classification results, the seismic velocity from seismic surveying in the tunnel section, the depth from the surface to the tunnel at each point of the tunnel route, the type of rock, and the actual rock classification results of the makjang shown during excavation of the tunnel A rock classification device when designing a tunnel using artificial intelligence technology to create a learning model by learning with The method of claim 1,
The learning model generation unit,
Artificial intelligence technology, characterized in that a plurality of predictive models are trained using a plurality of different data samples extracted from a plurality of training data with labels set, and a learning model is generated using the learned plurality of predictive models; Rock classification device for tunnel design using the results of rock classification during excavation. The method of claim 1,
The learning model generation unit,
A predictive model corresponding to a plurality of data samples extracted from a plurality of training data with labels is trained, and a preset weight is multiplied by a preset weight among the predicted values calculated by inputting the plurality of training data to the learned predictive model. Artificial intelligence technology, characterized in that generating next learning data, repeating the process of generating the next learning data to generate a plurality of predictive models, and generating a learning model using the generated plurality of predictive models; Rock classification device for tunnel design using the results of rock classification during excavation. The method of claim 1,
The learning model generation unit,
Artificial intelligence technology and rock classification results during excavation, characterized in that a plurality of different predictive models are trained using a plurality of learning data with labels set in common, and a learning model is generated using the learned plurality of predictive models Rock classification device when designing tunnels using The method of claim 1,
The learning model generation unit,
It is characterized by generating a learning model using at least one technique of Artificial Neural Networks (ANN), Bootstrap Aggregating, Bagging, Boosting, Voting, and SVM (Support Vector Machine). Rock classification device for tunnel design using artificial intelligence technology and the results of rock classification during excavation. The method of claim 1,
The data analysis unit,
When designing a new tunnel, the preliminary rock classification results for the new tunnel are derived, and the seismic velocity derived from the preliminary rock classification results and seismic surveying, the depth from the surface to the tunnel at each point of the tunnel route, and the type of rock Rock classification device for tunnel design using artificial intelligence technology and rock classification results during excavation, characterized in that the final rock classification results in the new tunnel are analyzed by inputting the ground investigation results including at least one piece of information into the learning model .

Images