KR102576001B1 - Automation device and method for evaluating Rock Mass Rating(RMR) using deep learning models - Google Patents

Abstract

Disclosed is an automated device and method for evaluating rock mass grade using a deep learning model. The apparatus of the present invention is a score for strength of fresh rock, rock quality designation (RQD), joint interval, joint state, and groundwater state by using an input unit for receiving input information on a rock mass for evaluating the rock mass grade and the input information. and a control unit that evaluates the rock mass grade by RMR (Rock Mass Rating) using the calculated scores, wherein the control unit determines the position of the joint formed on the rock surface, the joint path, the weathered area, the type of rock, and the quality of the rock. The rock surface image included in the input information is input as an input value to the first deep learning model that has pre-learned at least one of the colors to calculate the score for the rock quality index and the joint interval, and the filling material formed in the joint, the joint interval, and the joint surface A score for the joint state is calculated by inputting the joint image included in the input information as an input value to the second deep learning model that has previously learned at least one of the roughness of the 2 It is characterized by calculating the score for the groundwater condition by inputting the groundwater condition image included in the input information to the deep learning model as an input value.

Description

Automation device and method for evaluating Rock Mass Rating (RMR) using deep learning models}

The present invention relates to a rock mass grading technology, and more particularly, to an automated device and method for estimating the rock mass grading using a deep learning model for automating the stability evaluation of tunnels, slopes, mines, etc. based on the deep learning model. .

In general, it is very important in terms of safety and economy to identify and excavate the structural characteristics of rock mass in tunnel construction, rock slopes, and mines. Furthermore, ground surveys and drilling surveys have a problem in that it is difficult to grasp the structural characteristics of the ground 100% due to reasons such as limitations, economic feasibility, and time.

Therefore, for the safe design and construction of structures during tunnel design or excavation, an engineering method that can objectively evaluate the state of rock mass is required, and this method is called rock mass classification.

The most commonly used of these rock mass classification methods was found in 1974 by Z.T. This is the Rock Mass Rating (RMR) proposed by Bieniawski. As shown in FIG. 1, the rock mass grade is divided into five categories depending on the state of a total of five items, including strength of fresh rock, rock quality designation (RQD), joint (discontinuous surface) interval, joint (discontinuous surface) state, and groundwater state, , points are assigned for each.

That is, in order to evaluate the rock grade, after summing up the points for the five items described above for the rock face in tunnels, slopes, and mines, the sixth item, the direction of the joint (discontinuous surface) for the excavation direction of the tunnel (strike and The value according to the slope) is corrected to calculate the final rock grade value. At this time, as the calculated value is higher, the bedrock is judged to be in a relatively good condition.

However, the conventional method for evaluating the rock mass level has a problem in that a lot of effort and time are invested as it is performed by manpower.

Korean Registered Patent Publication No. 10-2204442 (2021.01.12.)

The technical problem to be achieved by the present invention is to evaluate the rock quality index (RQD), joint (discontinuous surface) interval, joint (discontinuous surface) state, and groundwater condition among rock grade items for stability evaluation of tunnels, slopes, mines, etc. through deep learning models. An object of the present invention is to provide an automated device and method for evaluating rock mass grades using a deep learning model that automatically evaluates items.

In order to achieve the above object, an automated device for evaluating the rock mass grade using a deep learning model according to the present invention is an input unit for receiving input information on the rock mass for evaluating the rock mass grade, and the strength of fresh rock and the rock quality index using the input information. (Rock Quality Designation, RQD), a control unit that calculates scores for each joint interval, joint state, and groundwater state, and evaluates the rock mass grade by RMR (Rock Mass Rating) using the calculated scores, The control unit inputs the rock surface image included in the input information as an input value to the first deep learning model that has previously learned at least one of the color of the joint position, joint path, weathered area, carcinoma, and rock quality formed on the rock surface. The joint image included in the input information is an input value to the second deep learning model that calculates scores for the rock quality index and the joint interval, and has previously learned at least one of a filler formed in the joint, a joint interval, and a joint surface roughness. and calculates the score for the joint state, and inputs the groundwater state image included in the input information as an input value to the second deep learning model that has previously learned at least one of the groundwater runoff amount and the groundwater trace to determine the groundwater state. It is characterized by calculating a score for.

In addition, when the collection information for learning the first deep learning model is input, the control unit sets a plurality of rock surface images included in the collection information in a structure in which a U-curve convolutional layer is stacked, Characterized in that the set rock surface image is learned.

In addition, the control unit converts the color according to the location of the joint, the joint path, the weathered portion, the type of cancer and the quality of the rock in the rock face image, and learns the first deep learning model based on the data.

In addition, when the collection information for learning the second deep learning model is input, the control unit converts the cutting image and the groundwater state image included in the collection information into a convolution layer, a pooling layer, and a fully connected connected) hierarchical structure, and the set joint image and groundwater state image are individually learned.

In addition, the control unit classifies the joint image into five grades according to the joint state classification criteria in the rock mass classification table by RMR, learns the second deep learning model based on the classified grades, and the groundwater state. It is characterized in that the image is classified into five grades according to groundwater condition classification criteria in the rock mass classification table by RMR, and the second deep learning model is learned based on the classified grades.

In addition, the control unit is characterized in that the grade of the rock mass is evaluated by adding the calculated scores and then correcting a value according to the directionality of the joint with respect to the excavation direction.

Also, the first deep learning model is a U-Net model, and the second deep learning model is a VGGNet model.

An automated method for evaluating a rock mass grade using a deep learning model according to the present invention includes the step of receiving input information about a rock mass for which a rock mass grade evaluation device evaluates, the rock mass grade evaluation device using the input information Calculating scores for strength, rock quality designation (RQD), joint interval, joint state, and groundwater state, respectively, and the rock mass rating evaluation device using the calculated scores to determine RMR (Rock Mass Rating) Evaluating the rock mass grade by a step of evaluating the rock mass grade, wherein the calculating step includes a first deep learning model that has previously learned at least one of a joint position formed on the rock surface, a joint path, a weathered area, and a color according to the type and quality of the rock mass. Calculating a score for the rock quality index and the joint interval by inputting a rock surface image included in the input information as an input value, and a second step in which at least one of the filler material formed in the joint, the joint interval, and the roughness of the joint surface is pre-learned. Calculating a score for the joint state by inputting a joint image included in the input information to a deep learning model as an input value, and inputting the input information to a second deep learning model that has previously learned at least one of groundwater runoff and groundwater traces. and calculating a score for the groundwater condition by inputting the groundwater condition image included in as an input value.

The automated apparatus and method for evaluating the rock mass grade using the deep learning model of the present invention are based on the rock quality index (RQD), the joint interval, the joint state, and the groundwater state among the evaluation items for evaluating the rock mass grade by RMR using the deep learning model. By automatically evaluating the items, it is possible to make a fast, non-subjective, and objective evaluation.

In addition, evaluation effort and time can be reduced compared to the conventional manpower-based evaluation method.

1 is a diagram for explaining RMR evaluation items and scores for each item for evaluating rock mass grade.
2 is a diagram for explaining a rock mass grade evaluation apparatus according to an embodiment of the present invention.
3 is a diagram illustrating the structure of a U-Net model, which is a deep learning model according to an embodiment of the present invention.
4 is a diagram for explaining the VGGNet model structure, which is a deep learning model according to an embodiment of the present invention.
5 is a diagram comparing the conventional rock mass grade evaluation and the rock mass grade evaluation according to an embodiment of the present invention.
6 is a flowchart for explaining a rock mass grade evaluation method according to an embodiment of the present invention.
7 is a flowchart illustrating learning of a U-Net model as a first deep learning model according to an embodiment of the present invention.
8 is a flowchart for explaining learning of a second deep learning model, the VGGNet model, 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 have the same numerals as much as possible even if they are displayed 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 will be omitted.

2 is a diagram for explaining a rock mass rating evaluation device according to an embodiment of the present invention, FIG. 3 is a diagram for explaining the U-Net model structure, which is a deep learning model according to an embodiment of the present invention, and FIG. It is a diagram explaining the VGGNet model structure, which is a deep learning model according to an embodiment of the present invention, and FIG. 5 is a diagram comparing conventional rock mass evaluation and rock mass evaluation according to an embodiment of the present invention.

Referring to FIGS. 2 to 5 , the rock mass evaluation apparatus 100 may include an input unit 10 and a control unit 30, and may further include an output unit 50 and a storage unit 70.

The input unit 10 receives input information about the rock mass for evaluating the rock mass grade. The input unit 10 receives collection information for learning a deep learning model. Here, the input information and collection information may be at least one piece of image information captured from an external device (camera).

The control unit 30 performs overall control of the rock grade evaluation device 100 . The control unit 30 calculates scores for strength of fresh rock, rock quality designation (RQD), joint interval, joint state, and groundwater state, respectively, using the input information. At this time, the control unit 30 calculates the score for the rock quality index and the joint interval using the U-Net model, which is the first deep learning model, and the VGGNet model, the second deep learning model, for the joint state and the groundwater state. Calculate points for each.

In detail, the control unit 30 estimates the strength of the rock mass using drilling information of the rock mass included in the input information and the actual strength of the rock mass to evaluate the rock mass grade, and calculates a score for the strength of fresh rock using the estimated rock mass strength. do. That is, the control unit 30 may estimate the rock mass strength by predicting the correlation between the drilling information and the actual rock mass strength. Here, drilling information is information recorded inside the equipment when drilling with a jumbo drill used in drilling work performed for tunnel blasting, and includes drilling speed, striking force, rotational force, and the like.

The control unit 30 evaluates the rock grade included in the input information as an input value of the first deep learning model that has previously learned at least one of the joint position, joint path, weathered area, and color according to the type and quality of the rock formed on the rock surface. After calculating the length, position and direction of the joint by inputting the bedrock image of the rock mass, calculating the joint interval based on the calculated result (location of the joint), and calculating the rock quality index based on the calculated joint interval, the equation [ Calculate the rock quality index using Equation 1] and [Equation 2]. The control unit 30 may calculate scores for the calculated cancer quality index and joint interval according to the RMR evaluation table of FIG. 1 .

For example, the control unit 30 analyzes a photo of a tunnel face (rock surface image) taken with a high-resolution camera with a U-Net model, which is a first deep learning model, to determine the size, direction, spacing, etc., of joints formed on the surface of the tunnel face. After extracting the information, scores for the rock quality index and joint spacing can be finally calculated.

Here, the U-Net model is a type of convolutional neural networks (CNN), and is a network structure for segmentation in a structure in which U-curve convolutional layers (CN) are stacked. Segmentation here refers to the process of dividing an image into a plurality of pixel sets in order to simplify or transform the expression of a digital image into something more meaningful and easy to interpret, and is especially used to find objects and boundaries (lines, curves) in an image. do. The structure of the U-Net model is largely divided into two, the first is a downsampling process, and the second is an upsampling process. The role of the downsampling process is to obtain the context of the image, i.e., to see the overall composition in the image and information about what objects are in the image. The role of the upsampling process is localization, that is, finding the location of an object. Overall, the U-Net model has a structure in which the resolution is reduced through pooling in the downsampling process and the resolution is increased again in the upsampling process.

The control unit 30 inputs a rock joint image for evaluating the rock mass grade included in the input information as an input value of the second deep learning model in which at least one of the filling material formed in the joint, the joint interval, and the roughness of the joint surface has been previously learned, Calculate points for joint condition.

In addition, the control unit 30 calculates a score for the groundwater condition by inputting a groundwater condition image for evaluating the bedrock grade included in the input information as an input value of the second deep learning model in which at least one of the groundwater runoff amount and the groundwater trace has been previously learned. do.

Here, the VGGNet model is used to classify various types of images into up to 1000 types according to their characteristics. It consists of a convolution layer, a pooling layer, and a fully connected layer, and VGG16 and VGG19 have been developed according to the number of layers. For example, VGG16 consists of 13 convolutional layers, 5 pooling layers and 3 fully connected layers.

Meanwhile, the control unit 30 may perform a learning process for the first deep learning model and the second deep learning model before calculating scores for rock class evaluation using the first deep learning model and the second deep learning model. there is.

In detail, when the collection information for learning the first deep learning model is input, the control unit 30 sets a plurality of rock surface images included in the collection information in a structure in which U-curve convolution layers are stacked, and the set rock surface images are set. learn video At this time, the control unit 30 may convert the joint location, joint path, weathered portion, and color according to the type and quality of the rock in the rock face image into data, and learn the first deep learning model based on the data data. Depending on the configuration method of the model, the corresponding process can be omitted. Here, the bedrock surface image may be a bedrock field image for various rock types and rock quality in a tunnel, slope, mine, and the like.

When the collection information is input to learn the second deep learning model, the control unit 30 sets the cutting image and the groundwater state image included in the collection information to a convolutional layer, a pooling layer, and a fully connected hierarchical structure, respectively, and sets the cutting image. and groundwater condition images are individually learned. At this time, the control unit 30 converts the filling material formed in the joint in the joint image, the joint spacing, and the roughness of the joint surface into data, and classifies the data data into 5 grades according to the joint state classification criteria in the rock class classification table by RMR, , The second deep learning model can be learned based on the classified grade, but the corresponding process can be omitted depending on the construction method of the second deep learning model. In addition, the controller 30 converts the amount of groundwater runoff and traces of groundwater in the groundwater state image into data, classifies the dataized data into five grades according to the groundwater condition classification standard in the rock mass classification table by RMR, and based on the classified grades The second deep learning model can be learned, but the corresponding process can be omitted depending on the configuration method of the second deep learning model. Here, when learning the joint (discontinuous surface) state, the control unit 30 learns using a joint image for the state of the joint existing on the last scene instead of an image taken of the entire scene surface, and learns the state of groundwater. , Learning can be done using images taken of the entire scene or part of the scene. As such, the control unit 30 may learn a model suitable for evaluating the joint state and a model suitable for evaluating the groundwater state using the same second deep learning model.

The control unit 30 adds up the scores for the strength of the sacred rock, the rock quality index, the joint interval, the joint state, and the groundwater state, and then corrects the value according to the direction (strike and inclination) of the joint with respect to the excavation direction, and finally the rock mass. rate the grade Here, the relationship between the excavation direction and the joint direction is calculated according to the excavation direction of the tunnel and the angle formed by the strike of the main joint group formed on the face and the slope of the joint group. At this time, the excavation direction of the tunnel can be known by receiving the measurement value measured in advance, and the strike direction on the surface of the face can be known by receiving the measurement value measured through the clinometer or lidar device.

The output unit 50 outputs input information and collection information input through the input unit 10 . The output unit 50 outputs the grade of the rock mass evaluated by the control unit 30 . The output unit 50 may output visual effects and auditory effects.

The storage unit 70 stores a program or algorithm for the rock mass grading apparatus 100 to operate. The storage unit 70 stores input information and collection information input through the input unit 10 . The storage unit 70 stores the grade of rock mass evaluated by the control unit 30 . The storage unit 70 is a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (eg SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, It may include at least one storage medium of a magnetic disk and an optical disk.

As described above, the rock mass grading apparatus 100 of the present invention, unlike the conventional rock mass grading method that evaluates based on manpower, uses a deep learning model for rock mass index, joint interval, joint state, and groundwater state. By calculating the grade score and automatically evaluating it, it is possible to perform quick and objective evaluation.

6 is a flowchart for explaining a rock mass grade evaluation method according to an embodiment of the present invention.

Referring to FIGS. 2 and 6, the rock grade evaluation method automatically evaluates the items for the rock quality index, joint interval, joint state, and groundwater condition among the rock grade items using a deep learning model, thereby providing fast, non-subjective, and objective can evaluate In addition, the rock mass evaluation method can save effort and time compared to the conventional manpower-based evaluation method.

In step S110, the rock mass grade evaluation device 100 receives input information for evaluation. Here, the input information is information for calculating the score for the strength of fresh rock, rock quality index (RQD), joint interval, joint state, and groundwater state, such as drilling information, actual rock strength, bedrock image, joint image, groundwater state image, etc. can include

In step S120, the rock mass evaluation apparatus 100 evaluates the strength of the new rock. The rock mass rating evaluation apparatus 100 estimates the strength of the rock mass using drilling information and the actual strength of the rock mass, and calculates a score for the strength of fresh rock using the estimated strength of the rock mass. That is, the rock mass grade evaluation apparatus 100 may estimate the rock mass strength by predicting the correlation between the drilling information and the actual rock mass strength.

In step S130, the rock grade evaluation device 100 evaluates the rock quality index (RQD) / joint interval. The rock mass grading apparatus 100 is a first deep learning model obtained by pre-learning at least one of the color of the joint position, joint path, weathered area, carcinoma type, and rock quality formed on the rock mass surface image of the rock mass for evaluating the rock mass grade. It is input as an input value to calculate the score for the rock quality index and the joint interval. Here, the first deep learning model may be a U-Net model.

In step S140, the rock mass rating evaluation device 100 individually evaluates the joint state and the groundwater state. That is, the rock mass grade evaluation apparatus 100 inputs the joint image of the rock mass for evaluating the rock mass grade as an input value of the second deep learning model in which at least one of the filler formed in the joint, the joint interval, and the roughness of the joint surface has been previously learned, and Calculate points for condition. The rock mass grade evaluation apparatus 100 calculates a score for the ground water state by inputting a ground water state image for evaluating the rock mass grade as an input value of the second deep learning model obtained by pre-learning at least one of ground water runoff and ground water traces. Here, the rock mass rating evaluation device 100 may simultaneously calculate scores for the joint state and the groundwater state or in a predetermined order, and the second deep learning model may be a VGGNet model.

In step S150, the rock grade evaluation apparatus 100 evaluates the relationship between the excavation direction and the joint direction. The rock mass grading apparatus 100 sums the scores for the strength of the new rock, the rock quality index, the joint interval, the joint state, and the groundwater state, and then corrects the value according to the direction of the joint with respect to the excavation direction to finally determine the rock mass grade. Evaluate.

7 is a flowchart illustrating learning of a U-Net model as a first deep learning model according to an embodiment of the present invention.

Referring to FIGS. 2 and 7 , the rock mass grade evaluation apparatus 100 learns a first deep learning model, the U-Net model, before calculating scores for rock mass grade evaluation.

In step S210, the rock mass grade evaluation apparatus 100 receives a rock surface image. Here, the rock surface image may be at least one bedrock surface image of various types and types of rocks in a tunnel, a slope, a mine, and the like.

In step S220, the rock mass evaluation apparatus 100 classifies the rock mass surface image. The rock mass rating evaluation apparatus 100 transforms the rock surface image into data according to the location of the joint, the joint path, the weathered region, the type of carcinoma, and the quality of the rock, and classifies the data.

In step S230, the rock mass evaluation device 100 sets and learns the U-Net model. The rock mass grade evaluation apparatus 100 sets a plurality of rock surface images in a structure in which U-curve convolution layers are stacked based on the classified data, and learns the set rock surface images. That is, the rock mass evaluation apparatus 100 performs learning after pre-processing a plurality of rock surface images to be optimized for the U-Net model.

8 is a flowchart for explaining learning of a second deep learning model, the VGGNet model, according to an embodiment of the present invention.

Referring to FIGS. 2 and 8 , the rock mass grade evaluation apparatus 100 learns a second deep learning model, the VGGNet model, before calculating scores for rock mass grade evaluation.

In step S310, the rock mass rating evaluation apparatus 100 receives an image of a joint and an image of a groundwater state. Here, the joint image may be at least one image obtained by photographing the entire scene, or at least one image of the state of a joint existing on the scene, and the groundwater state image may be at least one image obtained by photographing the entire scene, or , may be at least one image of the state of groundwater on the surface of the surface.

In step S320, the rock mass grading apparatus 100 classifies the joint image and the groundwater state image. The rock mass rating apparatus 100 classifies the joint image into five grades according to the joint state classification criteria in the rock mass classification table based on RMR. In addition, the rock mass rating evaluation apparatus 100 classifies the groundwater state image into five grades according to the groundwater state classification criteria in the rock mass classification table based on RMR.

In step S330, the rock mass rating apparatus 100 sets and learns a VGGNet model suitable for joint state analysis. The rock mass rating evaluation apparatus 100 sets the joint image into a convolutional layer, a pooling layer, and a fully connected hierarchical structure based on the classified grades, and learns the set joint image. That is, the rock mass rating apparatus 100 preprocesses a plurality of joint images to be optimized for the VGGNet model, and then performs learning. In addition, the rock mass evaluation device 100 sets and learns a VGGNet model suitable for groundwater condition analysis. The rock mass rating evaluation apparatus 100 sets the groundwater state image into a convolutional layer, a pooling layer, and a fully connected hierarchical structure based on the classified grades, and learns the set groundwater state image. That is, the rock mass rating apparatus 100 preprocesses a plurality of groundwater state images to be optimized for the VGGNet model, and then performs learning.

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 commands, data files, data structures, etc. alone or in combination, and includes all types of recording devices storing data that can be read by a computer system. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs (Compact Disk Read Only Memory) and DVDs (Digital Video Disks). Optical media), magneto-optical media such as floptical disks, and program instructions such as ROM (Read Only Memory), RAM (RAM, Random Access Memory), flash memory, etc. and a hardware device specially configured to do so. In addition, the computer-readable recording medium is distributed in computer systems connected through a network, so that computer-readable codes 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 belongs.

Although the above has been described and illustrated in relation to preferred embodiments for illustrating the technical idea of the present invention, the present invention is not limited to the configuration and operation as shown and described in this way, without departing from the scope of the technical idea. It will be readily apparent to those skilled in the art that many changes and modifications can be made to the present invention. Accordingly, all such appropriate changes and modifications and equivalents should be regarded as falling within the scope of the present invention.

10: input unit
30: control unit
50: output unit
70: storage unit
100: rock grade evaluation device

Claims (8)

an input unit for receiving input information about a rock mass for evaluating a rock mass grade; and
Using the input information, scores for the strength of fresh rock, rock quality designation (RQD), joint interval, joint state, and groundwater state are calculated, respectively, and RMR (Rock Mass Rating) is calculated using the calculated scores. Including; a control unit for evaluating the rock mass grade by
The control unit,
The rock quality index is obtained by inputting the rock surface image included in the input information as an input value to the first deep learning model that has already learned at least one of the joint location, joint path, weathered area, carcinoma type, and color according to the rock quality formed on the rock surface. And calculating a score for the joint interval,
A score for the joint state is calculated by inputting the joint image included in the input information as an input value to the second deep learning model that has previously learned at least one of the filling material formed in the joint, the joint spacing, and the roughness of the joint surface,
A deep learning model characterized in that a score for the groundwater state is calculated by inputting the groundwater state image included in the input information as an input value to a second deep learning model that has previously learned at least one of the groundwater runoff amount and the groundwater trace. An automated device that evaluates the grade of rock mass used. According to claim 1,
The control unit,
When collection information for learning the first deep learning model is input, a plurality of rock surface images included in the collection information are set to a structure in which U-curve convolutional layers are stacked, and the set rock surface image An automated device for evaluating the rock mass grade using a deep learning model, characterized in that for learning. According to claim 2,
The control unit,
A rock mass using a deep learning model characterized in that the color according to the joint position, joint path, weathered area, carcinoma and rock quality in the rock face image is dataized, and the first deep learning model is learned based on the dataized data. An automated device that evaluates grades. According to claim 1,
The control unit,
When the collection information for learning the second deep learning model is input, the cutting image and the groundwater state image included in the collection information are converted into a convolution layer, a pooling layer, and a fully connected layer structure. An automated device for evaluating the rock mass grade using a deep learning model, characterized in that each is set and the set joint image and groundwater state image are individually learned. According to claim 4,
The control unit,
The joint image is classified into five grades according to the joint state classification criteria in the rock mass classification table by RMR, and the second deep learning model is learned based on the classified grades,
A deep learning model characterized in that the groundwater state image is classified into five grades according to groundwater state classification criteria in the rock mass classification table by RMR, and the second deep learning model is learned based on the classified grades. An automated device that evaluates the grade of rock mass used. According to claim 1,
The control unit,
An automated device for evaluating the rock grade using a deep learning model, characterized in that for evaluating the rock grade by adding the calculated scores and then correcting the value according to the joint direction with respect to the excavation direction. According to claim 1,
The first deep learning model is a U-Net model, and the second deep learning model is an automated device for evaluating rock grade using a deep learning model, characterized in that the VGGNet model. receiving input information about a rock mass for which a rock mass grade is evaluated by a rock mass grade evaluation device;
Calculating, by the rock mass rating apparatus, scores for strength of fresh rock, rock quality designation (RQD), joint interval, joint state, and groundwater state, respectively, using the input information; and
Evaluating, by the rock mass rating apparatus, a rock mass rating by RMR (Rock Mass Rating) using the calculated score;
The calculating step is
The rock quality index is obtained by inputting the rock surface image included in the input information as an input value to the first deep learning model that has previously learned at least one of the joint position, joint path, weathering region, carcinoma type, and color according to the rock quality formed on the rock surface. and calculating a score for the cutting interval;
Calculating a score for the joint state by inputting the joint image included in the input information as an input value to a second deep learning model that has previously learned at least one of a filling material formed in the joint, a joint gap, and a joint surface roughness; and
calculating a score for the groundwater condition by inputting the groundwater condition image included in the input information as an input value to a second deep learning model obtained by pre-learning at least one of the groundwater runoff amount and the groundwater trace;
An automated method for evaluating rock mass using a deep learning model, comprising:

Images