Overexcavation amount estimation device, overexcavation amount estimation program, overexcavation amount estimation method, and trained overexcavation amount inference model used in overexcavation amount estimation device

Abstract

To accurately estimate the amount of overexcavation when drilling and blasting is performed at a tunnel face based on a drilling and blasting plan.SOLUTION: An overexcavation amount estimation device 1 includes a drilling and blasting plan input accepting unit 2 that accepts input of a drilling and blasting plan, including positions of the outermost charging holes, which are charging holes at the outermost periphery of a tunnel face, a drilling data acquisition unit 4 that acquires drilling data obtained by a rock drill, and an overexcavation amount estimation unit 6 that accepts input of learning plan input data generated based on a drilling and blasting plan from drilling and blasting operations in past and learning drilling data related to the drilling and blasting plan as learning input data, and estimates the overexcavation amount when drilling and blasting is performed based on the drilling and blasting plan using a learned overexcavation amount inference model 11 that is trained to infer the overexcavation amount after the drilling and blasting is performed, with the plan input data generated based on the drilling and blasting plan and the drilling data related to the drilling and blasting plan as inputs.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to an overexcavation amount estimation device, an overexcavation amount estimation program, an overexcavation amount estimation method, and a learned overexcavation amount inference model used in the overexcavation amount estimation device. [Background technology]

[0002] In mountain tunnel construction, the following process is repeated: drilling (boring) charge holes at the tunnel face with a rock drill, charging the charge holes, then blasting, removing the resulting debris outside the tunnel, and spraying concrete onto the wall. At this time, a drilling and blasting plan, including the drilling pattern such as the position of the charge holes relative to the tunnel face, is determined in advance or as needed, and drilling and blasting are carried out based on this drilling and blasting plan.

[0003] If the drilling and blasting plan is not appropriate, the tunnel wall may be excavated deeper in the radial direction of the tunnel than the position planned at the time of design, resulting in overexcavation. If overexcavation occurs, more concrete is required to spray the concrete onto the wall, resulting in costs greater than expected at the time of design. Furthermore, if the drilling and blasting plan is not appropriate, the tunnel wall may not be excavated to the position planned at the time of design, resulting in a hit state. In this case, it will be necessary to chip the wall at least to the position planned at the time of design, which may extend the construction period.

[0004] In this way, it is important to properly plan drilling and blasting in order to proceed with mountain tunnel construction as scheduled in terms of both budget and construction period. In this regard, Patent Document 1 discloses a blasting information processing device comprising: a first input data acquisition unit that acquires, as first input data for each drilling hole, the drilling energy consumed when a first drilling hole, which is located on the outer peripheral edge of the drilling face among multiple drilling holes located on the drilling face, and the amount of explosives loaded in the first drilling hole; a first output data acquisition unit that acquires, as first output data for each drilling hole, the size of a cross section perpendicular to the axial direction of a first area formed by the explosion of the explosives loaded in the first drilling hole, which is set from the cross section after blasting; and a model generation unit that performs machine learning based on the first input data and the first output data to generate a trained model that shows the correlation between the drilling energy, the amount of explosives loaded, and the size of the area formed by the explosion. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2023-43638 Summary of the Invention [Problem to be solved by the invention]

[0006] When formulating a drilling and blasting plan, if it were possible to more accurately estimate the amount of overexcavation, which is the difference between the position of the wall surface in the radial direction of the tunnel after drilling and blasting based on the drilling and blasting plan and the position planned at the time of design, it would be possible to review and revise the drilling and blasting plan based on the estimated amount of overexcavation so as to reduce overexcavation and hits. This could potentially allow mountain tunnel construction to proceed as planned and prevent increases in costs and longer construction periods. Thus, when drilling and blasting a tunnel face, it is desirable to accurately estimate the amount of overexcavation that will occur when drilling and blasting is performed based on a drilling and blasting plan.

[0007] The problem that the present invention aims to solve is to provide an overexcavation amount estimation device, an overexcavation amount estimation program, an overexcavation amount estimation method, and a learned overexcavation amount inference model for use in the overexcavation amount estimation device, which accurately estimate the amount of overexcavation that will occur when drilling and blasting is performed on a tunnel face based on a drilling and blasting plan. [Means for solving the problem]

[0008] In order to solve the above problems, the present invention employs the following means. That is, the present invention is an overexcavation amount estimation device that estimates the overexcavation amount, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at the tunnel face based on a drilling and blasting plan and the design position planned at the time of design, wherein the drilling and blasting plan includes a drilling pattern, and the drilling pattern includes the positions of outermost charge holes which are charge holes at the outermost periphery of the tunnel face, and the device includes a drilling and blasting plan input receiving unit that receives input of the drilling and blasting plan, a drilling data acquisition unit that acquires drilling data by a rock drill machine, and a previous drilling and blasting data acquisition unit that acquires data on drilling by a rock drill machine and previous drilling and blasting data acquired in the past. The present invention provides an overexcavation amount estimation device, which is equipped with an overexcavation amount estimation unit that inputs, as learning input data, plan input data generated based on the drilling and blasting plan and learning drilling data related to the drilling and blasting plan, and estimates the overexcavation amount when drilling and blasting is performed based on the drilling and blasting plan, using a learned overexcavation amount inference model that is trained to input learning input data and infer the overexcavation amount after the drilling and blasting is performed, and inputs plan input data generated based on the drilling and blasting plan that has been accepted as input as the object for estimating the overexcavation amount, and the drilling data related to the drilling and blasting plan.

[0009] When drilling and blasting a tunnel face, the area crushed by the explosives loaded in one charge hole is basically limited to the area around the charge hole. Therefore, the more the outermost charge hole, which is the charge hole at the outermost periphery of the tunnel face, is located radially inward, the more of the crushed area is included inside the design position of the wall surface planned at the time of design, and the smaller the area excavated radially outward from the design position. Conversely, the more the outermost charge hole is located radially outward from the tunnel, the more of the crushed area extends outside the design position, and the larger the area excavated radially outward from the design position. Thus, when drilling and blasting is performed, there is a correlation between the overexcavation amount, which is the difference between the position of the wall surface in the radial direction of the tunnel and the design position planned at the time of design, and the position of the outermost charge hole. In contrast, with the above-described configuration, the overexcavation amount estimation device is equipped with a trained overexcavation amount inference model that is trained to infer the overexcavation amount after drilling and blasting is performed by inputting, as part of the training input data, training-time plan input data generated based on drilling and blasting plans from previous drilling and blasting operations. Here, the drilling and blasting plan on which the training-time plan input data input during training is based includes a drilling pattern including the position of the outermost charge hole. In this way, the trained overexcavation amount inference model is configured to be able to grasp the position of the outermost charge hole and is trained to infer the overexcavation amount based on this. By inputting plan input data including the position of the outermost charge hole, similar to the training-time plan input data generated based on the drilling and blasting plan accepted by the drilling and blasting plan input receiving unit as the target for overexcavation amount estimation, into this trained overexcavation amount inference model, the overexcavation amount when drilling and blasting is performed based on the input drilling and blasting plan can be estimated. The size and shape of the area crushed by the explosives loaded in one charge hole will vary depending on the properties of the rock mass. Therefore, there is a correlation between the properties of the rock mass and the amount of over-excavation. In contrast, with the above-described configuration, during learning, the learned overexcavation amount inference model receives, as learning plan input data, the learning-time plan input data generated based on the drilling blasting plan of a previous drilling blasting operation, as described above, along with the learning-time drilling data related to the drilling blasting plan. Because the drilling data is acquired by a rock drill during drilling, the learning-time drilling data accurately and clearly reflects the properties of the rock mass at the time of drilling. In this way, the learned overexcavation amount inference model is configured to grasp the properties of the rock mass and is trained to infer the overexcavation amount based on this. By inputting the drilling data related to the drilling blasting plan, along with the plan input data generated based on the drilling blasting plan accepted by the drilling blasting plan input receiving unit as the input for overexcavation amount estimation, into the learned overexcavation amount inference model, the overexcavation amount when drilling blasting is performed based on the input drilling blasting plan can be more accurately estimated, taking into account the properties of the rock mass. In this way, an overexcavation amount estimation device can be provided that, when drilling and blasting a tunnel face, accurately estimates the amount of overexcavation that will occur when drilling and blasting is performed based on a drilling and blasting plan.

[0010] In one aspect of the present invention, the learning planning input data and the planning input data include the distance of the outermost charging hole from the design position calculated based on the position of the outermost charging hole included in the drilling pattern.

[0011] As explained above, there is a correlation between the position of the outermost charge hole and the amount of overexcavation. Here, since the amount of overexcavation is the difference between the position of the wall surface in the radial direction of the tunnel after drilling and blasting and the design position, it is considered that what is correlated with the amount of overexcavation is particularly the relative positional relationship between the outermost charge hole and the design position, i.e., the distance of the outermost charge hole from the design position. In contrast, according to the above configuration, the learning-time planning input data includes the distance from the design position of the outermost charge hole, calculated based on the position of the outermost charge hole included in the drilling pattern. In this way, the learned over-excavation amount inference model is configured to be able to grasp the distance from the design position of the outermost charge hole, and is trained to infer the over-excavation amount based on this. Furthermore, the planned input data that the over-excavation amount estimation unit inputs to the learned over-excavation amount inference model when estimating the over-excavation amount also includes, like the learning planned input data, the distance from the design position of the outermost charge hole calculated based on the position of the outermost charge hole included in the drilling pattern. Therefore, as described above, data consistent with the learning planned input data input during learning is input to the learned over-excavation amount inference model. In this way, when drilling and blasting the tunnel face, the amount of overexcavation that will occur when drilling and blasting is performed can be estimated more accurately based on the drilling and blasting plan.

[0012] In another aspect of the present invention, the learning time plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan in the drilling and blasting performed in the past, and the learned overexcavation amount inference model is configured to input the learning time plan input data and the learning time drilling data related to the pair of outermost charge holes as the learning time input data, and to infer the overexcavation amount at the midpoint position between the pair of outermost charge holes after the drilling and blasting is performed. The system is trained as follows: the plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan that has been received as an input for estimating the overexcavation amount; the overexcavation amount estimation unit inputs the plan input data and the drilling data related to the pair of outermost charge holes into the learned overexcavation amount inference model, and estimates the overexcavation amount at the midpoint position between the pair of outermost charge holes when the drilling and blasting is carried out based on the drilling and blasting plan.

[0013] As already explained, when drilling holes in a tunnel face and blasting, the area that is crushed by the explosives loaded in one charge hole is basically limited to the area around the charge hole. When loading the outermost charge hole and blasting, it is possible for workers to predict the amount of overexcavation that will occur in the radially outer part of the tunnel than the outermost charge hole. However, it is not easy for workers to predict the amount of overexcavation at a position between two outermost charge holes, because they need to consider the relationship between the two outermost charge holes. In contrast, with the above-described configuration, the learning-time plan input data is generated for a pair of adjacent outermost charge holes in a drilling blasting plan for a previously performed drilling blasting operation, and the learned overexcavation amount inference model is trained to input the learning-time plan input data as learning-time input data and infer the overexcavation amount at the midpoint between the pair of outermost charge holes after the drilling blasting operation is performed. Furthermore, the plan input data is generated for a pair of adjacent outermost charge holes in a drilling blasting plan for which input is received as an overexcavation amount estimation target, and the overexcavation amount estimation unit inputs the plan input data into the learned overexcavation amount inference model to estimate the overexcavation amount at the midpoint between the pair of outermost charge holes when drilling blasting is performed based on the drilling blasting plan. With this configuration, it is possible to estimate the overexcavation amount at the position between the two outermost charge holes, which is difficult for workers to predict.

[0014] In another aspect of the present invention, the learned over-excavation amount inference model is learned by inputting, as the learning-time drilling data, the drilling data of the outermost charge holes that were drilled at positions corresponding to each of the pair of outermost charge holes for which the learning-time plan input data was generated in drilling performed immediately before the drilling blasting performed in the past, and the over-excavation amount estimation unit inputs, as the drilling data, the drilling data of the outermost charge holes that were drilled at positions corresponding to each of the pair of outermost charge holes for which the plan input data was generated in drilling performed immediately before the drilling blasting performed based on the drilling blasting plan that was accepted as input for estimating the over-excavation amount into the learned over-excavation amount inference model.

[0015] The overexcavation amount estimation unit inputs plan input data generated based on a drilling and blasting plan for the tunnel face where drilling and blasting are about to be performed into the learned overexcavation amount inference model, and estimates the overexcavation amount when drilling and blasting is performed based on the drilling and blasting plan. An overexcavation amount estimation device equipped with such an overexcavation amount estimation unit is used, for example, to verify the validity of a drilling and blasting plan. Here, drilling data is input into the learned overexcavation amount inference model along with the plan input data. However, because drilling and blasting based on the drilling and blasting plan accepted as input for overexcavation amount estimation is about to be performed, drilling data corresponding to the drilling and blasting plan itself on which the plan input data is based cannot exist in the first place. Here, the tunnel face about to be drilled and blasted was revealed by the previous (actually the last) drilling and blasting, and the rock mass of the previous (previous) tunnel face where this previous drilling and blasting was performed is considered to have similar properties to the rock mass of the tunnel face about to be drilled and blasted. Therefore, in the above configuration, the drilling data acquired from the previous drilling is used as the drilling data. In particular, as described above, the trained overexcavation amount inference model inputs the planned input data generated for a pair of adjacent outermost charge holes and estimates the overexcavation amount at the midpoint between the pair of outermost charge holes. Therefore, in the above configuration, to address this, the drilling data is input to the learned overexcavation amount inference model as the drilling data for the outermost charge holes drilled at positions corresponding to the pair of outermost charge holes that are considered to have rock mass properties particularly similar to those of the pair of outermost charge holes for which the plan input data was generated in the immediately preceding drilling, for example, so that their relative positions from the center of the tunnel in a plane perpendicular to the tunnel axis are approximately the same. This makes it possible to accurately estimate the overexcavation amount even if there is no drilling data corresponding to the drilling and blasting plan. Furthermore, in the above-described configuration, the learned overbreak amount inference model is trained by inputting, as training drilling data, drilling data for the outermost charge holes drilled at positions corresponding to each of the pair of outermost charge holes for which the training plan input data was generated in the drilling performed immediately before (the previous drilling blasting) a previous drilling blasting. In this way, when the learned overbreak amount inference model is trained, the relationship between the training plan data and the training drilling data can be made the same as the relationship between the plan data and the drilling data input when estimating the overbreak amount. In this way, by aligning the data input to the learned overbreak amount inference model during training and inference, it is possible to further improve the accuracy of estimating the overbreak amount.

[0016] In another aspect of the present invention, the plan input data generated for the pair of outermost charge holes and the drilling data related to the pair of outermost charge holes are input into the learned over-excavation amount inference model, and the over-excavation amount at the midpoint position between the pair of outermost charge holes is estimated sequentially for all pairs of adjacent outermost charge holes in the drilling and blasting plan for which input has been accepted as the target for over-excavation amount estimation.

[0017] According to the above-mentioned configuration, the overexcavation amount estimation unit estimates the overexcavation amount at the midpoint between a pair of outermost charge holes sequentially for all pairs of adjacent outermost charge holes in the drilling and blasting plan that has been received as the target for overexcavation amount estimation. This makes it possible to estimate the overexcavation amount around the entire circumference of the tunnel face that is about to be drilled and blasted.

[0018] In another aspect of the present invention, the learned overexcavation amount inference model is trained using the overexcavation amount after the drilling and blasting has actually been performed as training data so that the inferred overexcavation amount is close to the training data, and the overexcavation amount used as the training data is calculated based on the results measured by 3D scanning the excavated surface after the drilling and blasting.

[0019] According to the above configuration, the amount of overexcavation in actual drilling and blasting, which is used as training data, can be accurately obtained.

[0020] In another aspect of the present invention, the learning drilling data and the drilling data are any one or any combination of the depth of the charge hole, drilling speed, impact pressure, feed pressure, damper pressure, rotational speed, rotational pressure, water volume, and water pressure.

[0021] According to the above configuration, the properties of the rock mass can be appropriately reflected in the learning drilling data and the drilling data.

[0022] In another aspect of the invention, the drilling and blasting plan includes a charge amount for the outermost charge hole.

[0023] In the outermost charging hole, if the charge amount is large, the crushed area becomes large, and therefore the area excavated radially outward from the design position becomes large. On the other hand, if the charge amount is small, the crushed area becomes small, and therefore the area excavated radially outward from the design position becomes small. Thus, in the outermost charging hole, there is a correlation between the charge amount and the amount of over-excavation. In contrast, with the above-described configuration, the drilling and blasting plan includes the charge amount for the outermost charge hole. The learned overexcavation amount inference model is configured to be able to grasp the charge amount for the outermost charge hole because the learning-time plan input data generated based on the drilling and blasting plans in past drilling and blasting operations is input as (part of) the learning-time input data, and is trained to infer the overexcavation amount based on this. By inputting plan input data including the charge amount similar to the learning-time plan input data generated based on the drilling and blasting plan accepted by the drilling and blasting plan input receiving unit as the target for overexcavation amount estimation into this learned overexcavation amount inference model, the overexcavation amount when drilling and blasting is performed based on the input drilling and blasting plan can be more accurately estimated taking into account the charge amount.

[0024] The present invention also provides an overexcavation volume estimation program for estimating an overexcavation volume, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at the tunnel face based on a drilling and blasting plan and the design position planned at the time of design, wherein the drilling and blasting plan includes a drilling pattern, and the drilling pattern includes the positions of outermost charge holes which are charge holes at the outermost periphery of the tunnel face, and the computer has a drilling and blasting plan input receiving function for receiving input of the drilling and blasting plan, a drilling data acquisition function for acquiring drilling data by a rock drill, and a function for acquiring data on drilling and blasting performed in the past. The present invention provides an overexcavation amount estimation program for realizing an overexcavation amount estimation function that inputs, as learning input data, plan input data generated based on the drilling and blasting plan and drilling data related to the drilling and blasting plan, and estimates the overexcavation amount when drilling and blasting is performed based on the drilling and blasting plan, using a learned overexcavation amount inference model that has been trained to infer the overexcavation amount after the drilling and blasting is performed, by inputting plan input data generated based on the drilling and blasting plan that has been accepted as input for estimating the overexcavation amount and the drilling data related to the drilling and blasting plan.

[0025] According to the above configuration, similar to the above description regarding the overexcavation amount estimation device, when drilling and blasting a tunnel face, the amount of overexcavation that will occur when drilling and blasting is performed can be accurately estimated based on the drilling and blasting plan.

[0026] The present invention also provides a method for estimating an overexcavation amount, which is a difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at a tunnel face based on a drilling and blasting plan and a design position planned at the time of design, wherein the drilling and blasting plan includes a drilling pattern, and the drilling pattern includes positions of outermost charge holes which are charge holes at the outermost periphery of the tunnel face, and the method includes a drilling data acquisition step for acquiring drilling data by a rock drill machine, a drilling and blasting plan input acceptance step for accepting input of the drilling and blasting plan, and a method for estimating an overexcavation amount, which is a difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at a tunnel face based on a drilling and blasting plan, the method comprising: The present invention provides an overexcavation amount estimation method, characterized by including an overexcavation amount estimation step in which learning-time plan input data generated based on a drilling blasting plan and learning-time drilling data related to the drilling blasting plan are input as learning-time input data, and plan input data generated based on the drilling blasting plan that has been accepted as input as a target for overexcavation amount estimation and the drilling data related to the drilling blasting plan are input using a learned overexcavation amount inference model that has been trained to infer the overexcavation amount after the drilling blasting is performed, to estimate the overexcavation amount when drilling blasting is performed based on the drilling blasting plan.

[0027] According to the above configuration, similar to the above description regarding the overexcavation amount estimation device, when drilling and blasting a tunnel face, the amount of overexcavation that will occur when drilling and blasting is performed can be accurately estimated based on the drilling and blasting plan.

[0028] In one aspect of the present invention, the learning time plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan in the drilling and blasting performed in the past, and the learned overexcavation amount inference model is learned by inputting the learning time plan input data and the learning time drilling data related to the pair of outermost charge holes as the learning time input data, and estimating the overexcavation amount at the midpoint position between the pair of outermost charge holes after the drilling and blasting is performed. The plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan that has been accepted as the target for estimating the overexcavation amount, and in the overexcavation amount estimation step, the plan input data and the drilling data related to the pair of outermost charge holes are input into the learned overexcavation amount inference model to estimate the overexcavation amount at the midpoint position between the pair of outermost charge holes when the drilling and blasting is carried out based on the drilling and blasting plan.

[0029] According to the above configuration, similar to the above description regarding the over-excavation amount estimation device, it is possible to estimate the amount of over-excavation at the position between the two outermost charging holes, which is difficult for workers to predict.

[0030] In another aspect of the present invention, the learned over-excavation amount inference model is learned by inputting, as the learning-time drilling data, the drilling data of the outermost charge holes that were drilled at positions corresponding to each of the pair of outermost charge holes for which the learning-time plan input data was generated in a drilling performed immediately before the drilling blasting performed in the past, and in the over-excavation amount estimation step, the drilling data of the outermost charge holes that were drilled at positions corresponding to each of the pair of outermost charge holes for which the plan input data was generated in a drilling performed immediately before the drilling blasting performed based on the drilling blasting plan that was accepted as input for estimating the over-excavation amount is input to the learned over-excavation amount inference model.

[0031] According to the above configuration, as with the above description of the overexcavation amount estimation device, even if there is no drilling data corresponding to the drilling and blasting plan, it is possible to accurately estimate the overexcavation amount. Furthermore, by aligning the data input during learning and during inference of the trained overexcavation amount inference model, it is possible to further improve the accuracy of the overexcavation amount estimation.

[0032] The present invention also provides a trained overexcavation amount inference model used to estimate the amount of overexcavation, which is the difference between the position of the wall surface in the radial direction of the tunnel and the design position planned at the time of design when drilling and blasting at a tunnel face based on a drilling and blasting plan, wherein the drilling and blasting plan includes a drilling pattern, which includes the position of the outermost charge hole, which is a charge hole at the outermost periphery of the tunnel face, and the trained overexcavation amount inference model is trained to infer the amount of overexcavation after the drilling and blasting is performed by inputting training plan input data generated based on the drilling and blasting plan from a drilling and blasting performed in the past and training drilling data related to the drilling and blasting plan as training input data.

[0033] According to the above configuration, similar to the above description regarding the overexcavation amount estimation device, when drilling and blasting a tunnel face, the amount of overexcavation that will occur when drilling and blasting is performed can be accurately estimated based on the drilling and blasting plan. [Effects of the Invention]

[0034] According to the present invention, it is possible to provide an overexcavation amount estimation device, an overexcavation amount estimation program, an overexcavation amount estimation method, and a learned overexcavation amount inference model for use in the overexcavation amount estimation device, which accurately estimate the amount of overexcavation that will occur when drilling and blasting is performed at a tunnel face based on a drilling and blasting plan. [Brief explanation of the drawings]

[0035] [Figure 1] FIG. 1 is a schematic explanatory diagram of a tunnel construction procedure. [Figure 2] An explanatory diagram of a tunnel face and a charge hole drilled in the tunnel face. [Figure 3] FIG. 2 is an enlarged view of the portion viewed from the arrow A in FIG. [Figure 4] FIG. 1 is a close-up view of an example of a tunnel wall after blasting. [Figure 5] FIG. 10 is an explanatory diagram regarding the amount of overexcavation. [Figure 6] 1 is a block diagram of an overexcavation amount estimation device according to an embodiment of the present invention. [Figure 7] FIG. 1 is an explanatory diagram showing the process of drilling a hole using a rock drill. [Figure 8] FIG. 1 is an explanatory diagram showing an image of the operation of the rock drill's drifter and drilling rod. [Figure 9] FIG. 10 is an explanatory diagram of punching data. [Figure 10] FIG. 10 is an explanatory diagram of a perforation pattern. [Figure 11] This is an explanatory diagram showing the flow of data when learning an over-excavation amount inference model and when inferring the amount of over-excavation using a learned over-excavation amount inference model that has completed learning. [Figure 12] This is an explanatory diagram showing the results of a 3D scan of the excavated surface after drilling blasting. [Figure 13] 10 is a flowchart showing the processing flow when learning the overexcavation amount inference model. [Figure 14] 10 is a flowchart showing the processing flow when inferring the amount of over-excavation using a learned over-excavation amount inference model. [Figure 15] FIG. 10 is an explanatory diagram showing verification results obtained by the overexcavation amount estimation device of the embodiment. [Figure 16] FIG. 10 is a block diagram of an overexcavation amount estimation device in a second modified example of the embodiment. [Figure 17] This is an explanatory diagram showing the flow of data when learning the over-excavation amount inference model and the over-excavation amount adjustment model. [Figure 18] This is an explanatory diagram showing the flow of data when inferring the amount of over-excavation using a learned over-excavation amount inference model that has completed learning and a learned over-excavation amount adjustment model. [Figure 19] 10 is a table showing verification results obtained by the overexcavation amount estimation device of the second modified example. DETAILED DESCRIPTION OF THE INVENTION

[0036] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Figure 1 is a schematic explanatory diagram of the tunnel construction procedure. Figure 2 is an explanatory diagram of the tunnel face and the charge holes drilled in the tunnel face. Figure 3 is an enlarged view of the portion indicated by the arrow A in Figure 1. Figure 3 is also a cross-sectional view of the portion indicated by the arrow II in Figure 2. During the construction of tunnel T, drilling (boring) of a charge hole H at tunnel face M, charging, and blasting are repeated. More specifically, for example, for tunnel face M1 in the situation shown in FIG. 1(a), a drilling and blasting plan P1 is prepared and determined, including the location of charge hole H. Based on this drilling and blasting plan P1, as shown in FIGS. 1(b) and 2, a charge hole H is drilled (bored) with a rock drill, explosives are loaded into the charge hole H, and blasting is performed. The resulting debris is then removed to the outside of the tunnel, and tunnel T is excavated, as shown in FIG. 1(c), and a new tunnel face M2 appears. Similarly, a drilling and blasting plan P2 is prepared and determined for this tunnel face M2. Based on this drilling and blasting plan P2, a charge hole H is drilled (bored) with a rock drill, explosives are loaded into the charge hole H, and blasting is performed, as shown in FIG. 1(d). The drilling and blasting plans P1, P2 planned for different tunnel faces M may be the same or different.

[0037] As shown in Figure 2, when the tunnel face M is viewed from the front, the line SL that crosses the widest part of the tunnel T in the horizontal direction is the X axis, and the line CL that crosses the highest part of the tunnel T in the vertical direction is the Y axis. Here, when the origin where these X axis and Y axis intersect is the center O of the tunnel, the outermost charging hole H in the radial direction of the tunnel T, that is, the charging hole H provided at the outermost periphery of the tunnel face M, is hereinafter referred to as the outermost charging hole OH. In Figure 2, the outermost charging hole OH is illustrated as a black circle, unlike the other charging holes H that are located further inward in the radial direction.

[0038] As shown in Fig. 3, when blasting is performed and a new tunnel face M2 appears, supports 150 that support the wall of the tunnel T from the inside are installed around it, and concrete 160 is sprayed on it. In this way, in order to secure space for installing the supports 150, the outermost charging holes OH are drilled at an angle with respect to the axial direction AC of the tunnel T so as to be inclined radially outward from the position where drilling starts.

[0039] Figure 4 shows an enlarged view of an example of a tunnel wall after blasting. Before drilling and blasting, the drilling and blasting plan P as already explained is determined in advance or as needed. Drilling and blasting are carried out based on this drilling and blasting plan P. If the contents of the drilling and blasting plan P are inappropriate, as shown in the B arrow portion in FIG. 4, after blasting the tunnel T, the wall surface may be excavated in the radial direction of the tunnel T to an excessively deep depth than the design position PL planned at the time of design, resulting in overexcavation. Also, as shown in the C arrow portion in FIG. 4, the wall surface may not be excavated to the design position PL, resulting in a hit. In the overexcavation amount estimation device of this embodiment, when drilling and blasting the tunnel face M, the overexcavation amount when drilling and blasting is performed is estimated based on the drilling and blasting plan P. By modifying the drilling and blasting plan P based on the estimated overexcavation amount, overexcavation and hits can be reduced.

[0040] 5 is an explanatory diagram relating to the amount of overexcavation, and is also a cross-sectional view of the portion viewed from the arrow D in FIG. In this embodiment, the overexcavation amount d when the tunnel T is actually excavated by blasting means the difference in the radial direction of the tunnel T between the wall position AL as the actual value actually excavated by blasting and the design position PL, i.e., the relative positional relationship of the wall position AL with respect to the design position PL. In particular, in this embodiment, the overexcavation amount d is the value obtained by subtracting the design position PL from the wall position AL as the actual value when the outside of the tunnel T is viewed as the positive direction. In this embodiment, the wall position AL as the actual value is the midpoint in the radial direction between the innermost and outermost parts of the tunnel T after excavation by blasting. Furthermore, as will be explained later, the overexcavation amount estimation device estimates the overexcavation amount d when drilling and blasting are performed based on the drilling and blasting plan P. In this case, the overexcavation amount d means the difference between the position AL of the wall surface estimated to appear when drilling and blasting are performed based on the drilling and blasting plan P in the radial direction of the tunnel T and the design position PL, i.e., the relative positional relationship of the wall position AL to the design position PL. In this case, as with the above, the overexcavation amount d can be said to be the value obtained by subtracting the design position PL from the position AL of the wall surface estimated to appear when drilling and blasting are performed based on the drilling and blasting plan P, when the outside of the tunnel T is viewed as the positive direction.

[0041] Thus, in this embodiment, the overexcavation amount d is a value obtained by subtracting the design position PL from the actual or estimated wall position AL when the outside of the tunnel T is considered to be the positive direction. Therefore, if it is estimated that the wall of the tunnel T will be excavated too much by blasting and overexcavation will occur, the estimated overexcavation amount d will be a positive value. On the other hand, if the wall of the tunnel T will not be excavated sufficiently by blasting and it is estimated that a hit will occur, the estimated overexcavation amount d will be a negative value.

[0042] FIG. 6 is a block diagram of an overexcavation amount estimation device according to an embodiment of the present invention. The overexcavation amount estimation device 1 includes a drilling and blasting plan input receiving unit 2, a learning unit 3, a drilling data acquisition unit 4, a preprocessing unit 5, an overexcavation amount estimation unit 6, and a result output unit 7. The overexcavation amount estimation unit 6 includes a learned overexcavation amount inference model 11.

[0043] The overexcavation amount estimation device 1 is a computer equipped with hardware such as a processor and memory, and by executing the overexcavation amount estimation program according to this embodiment, it is functionally equipped with the above-described configuration. That is, by executing the overexcavation amount estimation program, the overexcavation amount estimation device 1 causes the computer to realize the drilling and blasting plan input acceptance function, learning function, drilling data acquisition function, preprocessing function, overexcavation amount estimation function, and result output function, which correspond to the drilling and blasting plan input acceptance unit 2, learning unit 3, drilling data acquisition unit 4, preprocessing unit 5, overexcavation amount estimation unit 6, and result output unit 7, respectively.

[0044] As will be explained later, the overexcavation amount estimation unit 6 infers the overexcavation amount d. To perform this inference effectively, the overexcavation amount d is provided with a learned overexcavation amount inference model 11, which is generated by learning the (pre-learning) overexcavation amount inference model using the learning unit 3. For this reason, the overexcavation amount estimation device 1 performs two main operations: learning the overexcavation amount d and estimating the overexcavation amount d. For simplicity's sake, the following will first explain the drilling and blasting plan P and drilling data that form the input basis for the overexcavation amount inference model and the learned overexcavation amount inference model 11 when both learning the overexcavation amount d and estimating the overexcavation amount d. Then, the behavior of each component of the overexcavation amount estimation device 1 when learning the overexcavation amount d and the behavior of each component when estimating the overexcavation amount d will be explained in sequence.

[0045] The drilling and blasting plan P is an execution plan for drilling and blasting the charge holes H that will be performed on the tunnel face M. In this embodiment, the drilling and blasting plan P includes a drilling pattern. Specifically, the drilling pattern may include the drilling start position, drilling end position, and angle φ of each charge hole H with respect to the axial direction AC of the tunnel T. In this way, the drilling pattern includes the positions of the charge holes H. In particular, the drilling pattern includes the positions (drilling start positions) of the outermost charge holes OH on the tunnel face M. In the drilling pattern, the drilling start position and drilling end position can be expressed as coordinates on a three-dimensional coordinate system formed by the X-axis, Y-axis as shown in Figure 2, and the Z-axis formed by the depth direction of the paper in Figure 2, i.e., the axial direction AC of the tunnel T.

[0046] Fig. 7 is an explanatory diagram showing the drilling (boring) process using a rock drill. Fig. 8 is an explanatory diagram showing the operation of the rock drill's drifter and drilling rod. In mountain tunnel construction, a charge hole H for loading explosives used for blasting is drilled by a rock drill 200 such as a computer jumbo.

[0047] The rock drill 200 is equipped with multiple booms 203. Each boom 203 is equipped with a guide shell 204, a drifter 205, and a drilling rod 206. The boom 203 is provided so as to be rotatable in any direction and to be extendable and retractable. The guide shell 204 is attached to the tip of the boom 203. The drifter 205 is a hydraulic impact device that is arranged at the rear of the guide shell 204 and is configured to move forward and backward while being guided by the guide shell 204. The drilling rod 206 is a rod-shaped member for drilling a hole in the rock mass G and is attached to the drifter 205 so as to move on the guide shell 204. A bit 207 for excavating the rock mass G is attached to the tip of the drilling rod 206.

[0048] The drifter 205 includes a hydraulically driven piston 205a. When the piston 205a strikes the rear end of the drill rod 206, an elastic stress wave is generated. This stress wave propagates through the drill rod 206 and reaches the bit 207, crushing the rock mass G. A feed pressure F is applied to the drifter 205 so that the front of the bit 207 abuts against the rock mass G and so that the energy of the stress wave is efficiently transmitted to the rock mass G. When the feed pressure F is applied in this manner, the force from the rock mass G acts on the drill rod 206 as a reaction force, but this reaction force is absorbed by the damper 208. Furthermore, rotation is applied to the drill rod 206 via the spline portion 206a to maintain high crushing efficiency. During drilling, water W is pumped into the drill rod 206 and the bit 207 to discharge cuttings from the charge hole H.

[0049] The drilling data is any one or a combination of the depth of the charge hole H, the drilling speed, the impact pressure, the feed pressure, the damper pressure, the rotation speed, the rotation pressure, the water volume, and the water pressure. FIG. 9 is an explanatory diagram of the drilling data. In this embodiment, the drilling data includes the drilling speed during drilling, the impact pressure N of the drifter 205, the feed pressure F, the damper pressure (the pressure acting on the damper 208), the rotation speed, the rotation pressure, the water volume, and the water pressure of the drilling rod 206. The drilling data is acquired for each charge hole H. The drilling data can be acquired when the bit 207 reaches a certain depth from the surface of the rock mass G. In the example of FIG. 9, the drilling data is acquired at each position (depth) at approximately 10 cm intervals from the surface of the rock mass G.

[0050] In this embodiment, as will be described later, for each charging hole H, a plurality of data obtained according to the depth are averaged by depth, so the depth of the charging hole H is not used to infer the evaluation category. From this point of view, in this embodiment, the drilling data is composed of the drilling speed, impact pressure, feed pressure, damper pressure, rotation speed, rotation pressure, water volume, and water pressure.

[0051] The drilling data expresses what kind of action the rock drill 200 had to exert on the rock mass when drilling or boring a hole with the rock drill 200, and the degree of that action is thought to vary depending on the properties of the rock mass. From this perspective, the drilling data can be said to indicate the properties of the rock mass at the tunnel face.

[0052] Next, the behavior of each component of the overexcavation amount estimation device 1 when learning the overexcavation amount d will be described. The drilling and blasting plan input receiving unit 2 receives input of a drilling and blasting plan P that was carried out in the past for the purpose of learning the overexcavation amount d. The drilling and blasting plan input receiving unit 2 generates learning-time plan input data for learning an overexcavation amount inference model, which will be described later, based on the drilling and blasting plan P for drilling and blasting performed in the past, and generating a learned overexcavation amount inference model 11. When inferring the overexcavation amount d, the plan input data is input to the learned overexcavation amount inference model 11. The learning-time plan input data is essentially the same as this plan input data, and to make it clear that it is used during learning, it is distinguished from the plan input data and is given a different name.

[0053] FIG. 10 is an explanatory diagram of the perforation pattern. The drilling and blasting plan input receiving unit 2 generates one learning time plan input data for each of a pair of adjacent outermost charge holes OH of the drilling and blasting plan P received as input. The learning plan input data includes distances l1 and l2 between the (center of) the outermost charge hole OH and the design position PL on the line connecting the center O of the tunnel shown in Figure 2 and the center (of the drilling start position) of the outermost charge hole OH for each of the pair of outermost charge holes OH. The drilling and blasting plan input receiving unit 2 extracts the drilling start positions of each of the outermost charge holes OH from the drilling pattern of the drilling and blasting plan P for the pair of outermost charge holes OH, and calculates the distances l1 and l2 between the outermost charge hole OH and the design position PL by calculating the distances between each of these drilling start positions and the design position PL.

[0054] The learning plan input data also includes the distance L between (the drilling start positions of) a pair of outermost charge holes OH. The drilling and blasting plan input receiving unit 2 extracts the drilling start positions of each of the outermost charge holes OH from the drilling pattern of the drilling and blasting plan P for the pair of outermost charge holes OH, and calculates the distance between the drilling start positions to thereby calculate the distance L between the outermost charge holes OH.

[0055] Furthermore, the learning plan input data includes an average angle, which is the average value of the angle φ (see Figure 2) of each of the pair of outermost charge holes OH in the radial direction of the tunnel T with respect to the axial direction AC of the tunnel T. The drilling and blasting plan input receiving unit 2 extracts the angles corresponding to each of the pair of outermost charge holes OH from the drilling pattern of the drilling and blasting plan P and calculates the average of these to derive the average angle.

[0056] The drilling and blasting plan input receiving unit 2 generates a set of learning plan input data for a pair of outermost charging holes OH, including the distances l1, l2 between the outermost charging holes OH and the design position PL, the distance L between the outermost charging holes OH, and the average angle calculated and obtained in this manner. As shown in Fig. 2, there are multiple combinations of two adjacent outermost charge holes OH in the drilling and blasting plan P received as input by the drilling and blasting plan input receiving unit 2. The drilling and blasting plan input receiving unit 2 generates the above-mentioned learning time plan input data by treating each of all such combinations of adjacent outermost charge holes OH as a pair of outermost charge holes OH. Therefore, the drilling and blasting plan input receiving unit 2 generates learning time plan input data from the drilling and blasting plan P for the number of combinations of adjacent outermost charge holes OH.

[0057] In this way, the learning plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling blasting plan P based on the drilling blasting plan P from a drilling blasting operation carried out in the past. In addition, the learning planning input data includes distances l1 and l2 of the outermost charge hole OH from the design position PL, calculated based on the position (start position of drilling) of the outermost charge hole OH included in the drilling pattern.

[0058] The drilling data acquisition unit 4 acquires drilling data for the purpose of learning the overexcavation amount d. The drilling data acquired here relates to the drilling blasting plan P in the drilling blasting carried out in the past, which was input to the drilling blasting plan input receiving unit 2 for the purpose of learning the overexcavation amount d and served as the basis for generating the learning-time plan input data. More specifically, when learning the overexcavation amount d, the drilling data acquisition unit 4 acquires, as the drilling data related to the drilling blasting plan P in the drilling blasting carried out in the past, data acquired by the rock drill 200 in the drilling carried out immediately before this drilling blasting carried out in the past.

[0059] For example, consider a case where the drilling and blasting plan P for a previous drilling and blasting operation input for the purpose of learning the overexcavation amount d is the drilling and blasting plan P2 for the tunnel face M2 shown in FIG. 1(c), and the learning plan input data is generated based on this. This tunnel face M2 appears as a result of drilling and blasting based on the drilling and blasting plan P1 that was designed for the tunnel face M1 shown in FIG. 1(a). In such a case, the drilling data acquisition unit 4 acquires drilling data acquired by the rock drill 200 during drilling performed on the tunnel face M1 based on the drilling and blasting plan P1 immediately before (i.e., one time before) the previous drilling and blasting performed based on the drilling and blasting plan P2.

[0060] The drilling data acquisition unit 4 extracts, from the drilling data of the drilling performed immediately before the drilling blasting performed in the past, acquired as described above, drilling data related to the pair of outermost charge holes OH for which the learning time plan input data was generated, for each of the learning time plan input data generated by the drilling blasting plan input receiving unit 2. More specifically, the drilling data acquisition unit 4 first acquires the drilling start positions of each of the set of outermost charge holes OH for which the learning-time plan input data is generated from the drilling and blasting plan P2. Next, the drilling data acquisition unit 4 extracts, for each of the set of outermost charge holes OH, the outermost charge hole OH in the drilling pattern of the drilling and blasting plan P1 that has a drilling start position closest to the drilling start position of each of the set of outermost charge holes OH when the tunnel face M is viewed from the front as shown in FIG. 2 (i.e., when ignoring the coordinate value of the Z axis, which is the depth direction). The outermost charge hole OH of the drilling and blasting plan P1 extracted in this way is close to the outermost charge hole OH of the drilling and blasting plan P2 for which the learning-time plan input data is generated, among the outermost charge holes OH of the drilling and blasting plan P1, and is considered to have similar rock mass properties.

[0061] In this way, the drilling data acquisition unit 4 acquires the drilling data of the outermost charge holes OH drilled at positions corresponding to each of the set of outermost charge holes OH for which learning-time plan input data was generated in the drilling performed immediately before the previous drilling blasting, as learning-time drilling data, which is drilling data for learning the overexcavation amount inference model described later and generating the learned overexcavation amount inference model 11. When inferring the overexcavation amount d, the drilling data is input into the learned overexcavation amount inference model 11. The learning-time drilling data is substantially the same as this drilling data, and to make it clear that it is used during learning, it is distinguished from the drilling data and is given a different name. As explained as the drilling data, the training drilling data also includes any one or a combination of the depth of the charge hole H, drilling speed, impact pressure, feed pressure, damper pressure, rotational speed, rotational pressure, water volume, and water pressure.

[0062] In addition, the above-mentioned learning drilling data is actually adjusted by the pre-processing unit 5 before being used in the overcut amount inference model. The preprocessing unit 5 executes any one or a combination of inappropriate data deletion processing, missing value processing, item selection processing, dimension reduction processing, and averaging processing.

[0063] Immediately after the bit 207 is placed against the surface of the rock mass G and drilling is started, drilling is performed at low pressure and low speed to prevent the bit 207 from vibrating and shifting from the planned drilling position. To reduce the influence of such exceptional behavior immediately after drilling is started, the pre-processing unit 5 performs inappropriate data deletion processing to remove drilling data at positions shallower than a predetermined depth, such as 10 cm, from the surface of the rock mass G.

[0064] Furthermore, if an abnormality occurs in the sensor that acquires each piece of drilling data during drilling, the acquired values ​​may not be appropriate. Alternatively, if drilling is stopped midway through, the charge hole H will not be used for blasting, so it is not desirable to use the data related to the charge hole H as drilling data. As a missing value processing, the preprocessing unit 5 treats the drilling data acquired in the above cases as error data and excludes the error data from the drilling data.

[0065] Furthermore, for example, when inferring the overexcavation amount estimation unit 6, there are cases where better inference results are obtained by using only some of the items in the drilling data and not using other items. In such cases, the preprocessing unit 5 performs an item selection process in which it selects some items from the multiple items included in the drilling data and deletes the other items.

[0066] Furthermore, the pre-processing unit 5 reduces the number of items by extracting features from multiple items through statistical analysis such as principal component analysis to create new items as a dimensionality reduction process. In this way, the features of the drilling data can be more clearly expressed in fewer items, which may allow for efficient learning of the over-excavation amount inference model and inference by the learned over-excavation amount inference model 11.

[0067] Furthermore, the pre-processing unit 5 averages the multiple data obtained according to the depth of the charging hole H by depth as an averaging process. In the state shown in FIG. 9, for example, data is acquired for each of eight items, from drilling speed to water pressure, at each of 14 different depths, resulting in a total of 112 pieces of data for one charging hole H. However, by averaging each item by the depth of the charging hole H, the number of data corresponding to one charging hole H can be reduced to eight, the same number as the number of items. In this way, by determining a representative value for each charging hole H for each item of drilling data, the total number of drilling data is reduced, and learning of the over-excavation amount inference model and inference by the learned over-excavation amount inference model 11 may be performed efficiently. Note that the representative value for each charging hole H is not limited to the average value, and the maximum value, minimum value, variance, etc. of each item may also be used as the representative value.

[0068] Finally, the pre-processing unit 5 calculates one value for each item of the drilling data by calculating the average between the charging holes H for each item of the drilling data. For example, if eight items, namely drilling speed, impact pressure, feed pressure, damper pressure, rotation speed, rotation pressure, water volume, and water pressure, are used as the drilling data and the dimension reduction process and the item selection process are not performed, the drilling data will have eight values ​​as a whole. In this way, through the processing by the pre-processing unit 5, learning-time punching data is generated. In particular, in this embodiment, the pre-processing unit 5 performs inappropriate data deletion processing, missing value processing, and averaging processing.

[0069] FIG. 11 is an explanatory diagram showing the flow of data when training an over-excavation amount inference model and when inferring the over-excavation amount using a trained over-excavation amount inference model that has completed training. The learning unit 3 uses the learning time plan input data generated by the drilling and blasting plan input receiving unit 2 based on the drilling and blasting plan P for drilling and blasting performed in the past as described above, and the learning time drilling data related to the drilling and blasting plan P acquired by the drilling data acquisition unit 4 as learning time input data, to learn the overexcavation amount inference model 11L so as to infer the overexcavation amount d after the drilling and blasting is performed, and generates a learned overexcavation amount inference model 11.

[0070] In this embodiment, the learned over-excavation amount inference model 11 (over-excavation amount inference model 11L) is realized by a fully connected neural network. Alternatively, the learned over-excavation amount inference model 11 (over-excavation amount inference model 11L) may be realized by other methods such as SVM (support vector machine) or RF (random forest).

[0071] 11, the overexcavation amount inference model 11L includes an input layer 11i, multiple intermediate layers 11m, and an output layer 11o. The overexcavation amount inference model 11L is configured such that multiple layers are arranged in series, and the nodes constituting each layer are fully connected to each other between adjacent layers.

[0072] As already explained, the learning time planning input data and the learning time drilling data are each generated and acquired for a pair of outermost charge holes OH. Therefore, the over-excavation amount inference model 11L is trained so that the learning time planning input data and the learning time drilling data for a pair of outermost charge holes OH are input and the over-excavation amount d related thereto is inferred. For example, in this embodiment, the learning-time plan input data has four values: distances l1 and l2 between the outermost charging holes OH and the design position PL, distance L between the outermost charging holes OH, and average angle. Also, in this embodiment, the learning-time drilling data has a total of eight values ​​as described above. For example, these 12 values ​​are input to the input layer 11i. Therefore, the input layer 11i has 12 input nodes 11p corresponding to each of these 12 values.

[0073] When the learning-time planned input data and learning-time drilling data are input to the input layer 11i, calculations are performed sequentially based on the weights set for the connecting lines between the nodes, tracing each layer in order from the input layer 11i to the intermediate layer 11m and the output layer 11o, and the final calculation results are stored in each of the output nodes 11n of the output layer 11o. The output nodes 11n are set according to the value of the inferred overcut amount d. For example, in the case where the overcut amount d can have a value in the range of -20 cm or more and 20 cm or less, this range can be divided into eight ranges at predetermined intervals such as 5 cm: -20 cm or more and less than -15 cm, -15 cm or more and less than -10 cm, -10 cm or more and less than -5 cm, -5 cm or more and less than 0 cm, 0 cm or more and less than 5 cm, 5 cm or more and less than 10 cm, 10 cm or more and less than 15 cm, and 15 cm or more and less than 20 cm, and eight output nodes 11n can be provided to correspond to each of these eight ranges.

[0074] Here, the overexcavation amount d inferred by the overexcavation amount inference model 11L and the learned overexcavation amount inference model 11 generated by learning the overexcavation amount inference model 11L is at the midpoint position CP between a pair of outermost charge holes OH corresponding to the input learning-time plan input data and learning-time drilling data, as shown in Figure 10. More specifically, the midpoint position CP is set on a line connecting the centers of the pair of outermost charge holes OH at a position equidistant from each of the centers of the pair of outermost charge holes OH. The overexcavation amount d on a line LC that passes through this midpoint position CP so as to extend in the radial direction of the tunnel T when drilling and blasting is performed based on the drilling and blasting plan P corresponding to the learning-time plan input data is inferred by the overexcavation amount inference model 11L and the learned overexcavation amount inference model 11.

[0075] The inference result during learning is expressed as a softmax function, which indicates, for example, which of the above eight ranges the overexcavation amount inference model 11L infers the overexcavation amount d for the input learning-time planned input data and learning-time drilling data falls into. This inference result during learning is compared with the training data. The training data in this embodiment is the actual measured value of the overexcavation amount d at the midpoint position CP of the pair of outermost charge holes OH corresponding to the training plan input data when drilling and blasting is performed based on the drilling and blasting plan P (the drilling and blasting plan P2 in FIG. 1) from which the training plan input data was generated. Since the drilling and blasting plan P from which the training plan input data was generated has actually been executed in the past, the overexcavation amount d at this time can be calculated based on the actually measured value.

[0076] FIG. 12 is an explanatory diagram showing the results of a 3D scan of the unexcavated surface after drilling and blasting. The amount of overexcavation d after actual drilling and blasting, which is used as training data as described above, can be obtained by 3D scanning the excavated surface after drilling and blasting, measuring the position of the wall surface of tunnel T, and then calculating the difference between the measured value and the design position PL. This training data is the correct value of the overexcavation amount d prepared in correspondence with the training-time planned input data and training-time drilling data input to the overexcavation amount inference model 11L. More specifically, the training data has a value corresponding to each of the output nodes 11n. For example, if the overexcavation amount d is 3 cm, in the above example, the training data for the output node 11n corresponding to the range of 0 cm or more and less than 5 cm will be, for example, a value of 1, and the training data for the other output nodes 11n will be, for example, a value of 0.

[0077] As described above, the inference results during learning are compared with the teacher data, and the values ​​of the weights and the like of the overexcavation amount inference model 11L are adjusted by the backpropagation method or the gradient descent method so that the inference results during learning have values ​​close to the teacher data, thereby performing machine learning on the overexcavation amount inference model 11L. As a result, when learning-time planning input data and learning-time drilling data are input as learning-time input data, the overexcavation amount inference model 11L is trained to output an overexcavation amount d close to the teacher data corresponding to the learning-time input data.

[0078] In the manner described above, the learning unit 3 generates a learned over-digging amount inference model 11 as a learned model in which learning parameters such as weights have been learned and which is used as a program module that is part of artificial intelligence software.

[0079] In this way, the learned overexcavation amount inference model 11 is trained to infer the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH after the drilling blasting is performed, based on the drilling blasting plan P from a drilling blasting operation performed in the past, by inputting the learning plan input data generated for the pair of adjacent outermost charge holes OH in the drilling blasting plan and the learning drilling data related to the pair of outermost charge holes OH as learning input data. In addition, the learned overexcavation amount inference model 11 is trained using the overexcavation amount d after actual drilling and blasting as training data so that the inferred overexcavation amount d is close to the training data. In addition, in this embodiment, the learned overexcavation amount inference model 11 is learned by inputting, as learning drilling data, drilling data for the outermost charge holes OH that were drilled at positions corresponding to each of the pair of outermost charge holes OH for which learning plan input data was generated, in drilling performed immediately before a drilling blasting performed in the past.

[0080] Next, the behavior of each component when estimating the overexcavation amount d will be described. The drilling and blasting plan input receiving unit 2 receives input of a drilling and blasting plan P as a target for estimating the overexcavation amount d at the stage when drilling and blasting is about to be carried out for the tunnel face M based on the drilling and blasting plan P. The drilling and blasting plan input receiving unit 2 generates plan input data to be input to the learned overexcavation amount inference model 11 based on the drilling and blasting plan P input as the target for estimating the overexcavation amount d.

[0081] The drilling and blasting plan input receiving unit 2 generates one plan input data for each of a pair of adjacent outermost charging holes OH of the drilling and blasting plan P that has been received as an input for estimating the overexcavation amount d, just as when generating plan input data during learning. The drilling and blasting plan input receiving unit 2 generates a set of plan input data for a pair of adjacent outermost charging holes OH, including the distances l1 and l2 between the outermost charging holes OH and the design position PL, the distance L between the outermost charging holes OH, and the average angle. The drilling and blasting plan input receiving unit 2 generates plan input data from the drilling and blasting plan P, the number of which is equal to the number of combinations of adjacent outermost charging holes OH.

[0082] In this way, the plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling and blasting plan P, based on the drilling and blasting plan P that has been received as an input for estimating the overexcavation amount d. The planning input data also includes distances l1 and l2 of the outermost charging hole OH from the design position PL, calculated based on the position (drilling start position) of the outermost charging hole OH included in the drilling pattern.

[0083] The drilling data acquisition unit 4 acquires drilling data from the rock drill machine 200. The drilling data acquired here relates to the drilling and blasting plan P of the drilling and blasting that is about to be actually executed, which was input to the drilling and blasting plan input acceptance unit 2 and was the basis for generating the plan input data. From the viewpoint of accurately estimating the overexcavation amount d, it is desirable that the drilling data to be input to the learned overexcavation amount inference model 11 together with the drilling and blasting plan P be acquired by actually drilling the tunnel face M that is to be drilled and blasted based on the drilling and blasting plan P using the rock drill 200. However, since the tunnel face M is about to be drilled and blasted, no drilling data for the tunnel face M can exist. For this reason, in this embodiment, the drilling data related to the drilling and blasting plan P for the drilling and blasting that is about to be executed and input to the learned overexcavation amount inference model 11 together with the drilling and blasting plan P is acquired from the drilling data acquired by the rock drill 200 in the drilling that was performed immediately before (i.e., the previous time, the last time), as in the case of learning.

[0084] For example, in a situation where drilling and blasting is about to be performed on the tunnel face M2 shown in Figure 1(c) based on the drilling and blasting plan P2, there is no drilling data for the tunnel face M2. Therefore, the drilling data acquisition unit 4 acquires the drilling data acquired by the rock drill 200 when drilling was performed on the tunnel face M1 based on the drilling and blasting plan P1 just before the tunnel face M2.

[0085] The drilling data acquisition unit 4 extracts drilling data related to the pair of outermost charge holes OH for which the plan input data was generated, for each of the plan input data generated by the drilling blasting plan input receiving unit 2, from the drilling data of the drilling performed immediately before the drilling blasting performed based on the drilling blasting plan P that was received as input as the target for estimating the over-excavation amount d, as in the case of learning. More specifically, the drilling data acquisition unit 4 first acquires the drilling start position of each of the set of outermost charge holes OH, for which the plan input data is generated, from the drilling and blasting plan P2. Next, the drilling data acquisition unit 4 extracts, for each of the set of outermost charge holes OH, the outermost charge hole OH in the drilling pattern of the drilling and blasting plan P1, which has a drilling start position closest to the drilling start position of each of the set of outermost charge holes OH when the tunnel face M is viewed from the front as in Figure 2 (i.e., when ignoring the coordinate value of the Z axis, which is the depth direction).

[0086] In this way, the drilling data acquisition unit 4 acquires the drilling data of the outermost charge holes OH drilled at positions corresponding to each of the set of outermost charge holes OH for which the plan input data is generated, in the drilling performed immediately before the drilling blasting performed based on the drilling blasting plan P that has been received as input as the target for estimating the overexcavation amount d, as drilling data to be input into the learned overexcavation amount inference model 11. The drilling data as described above is actually adjusted by the pre-processing unit 5 before being used in the over-excavation amount inference model. When inferring the over-excavation amount d, the pre-processing unit 5 performs inappropriate data deletion processing, missing value processing, and averaging processing, just as during learning.

[0087] The overexcavation amount estimation unit 6 inputs the plan input data generated by the drilling and blasting plan input receiving unit 2 based on the drilling and blasting plan P for the drilling and blasting that is about to be executed as described above, and the drilling data related to the drilling and blasting plan P acquired by the drilling data acquisition unit 4, into the learned overexcavation amount inference model 11 as input data. The over-excavation amount estimation unit 6 inputs the plan input data generated for a pair of adjacent outermost charge holes OH among the outermost charge holes OH in the drilling and blasting plan P that has been received as an input for estimating the over-excavation amount d, and the drilling data related to the pair of outermost charge holes OH, into the learned over-excavation amount inference model 11.

[0088] The learned over-excavation amount inference model 11 has a configuration similar to that of the over-excavation amount inference model 11L. When input data is input to the over-excavation amount inference model 11L, calculations are performed sequentially within the learned over-excavation amount inference model 11 in the same manner as described during learning, and inference results are output to each of the output nodes 11n of the output layer 11o. The inference result, like the inference result during learning, is expressed as a softmax function, which indicates which of, for example, eight ranges the learned overexcavation amount inference model 11 has inferred the overexcavation amount d for the input planned input data and drilling data falls into. The overexcavation amount estimation unit 6 estimates that the range corresponding to the output node 11n with the largest value is the overexcavation amount d inferred by the learned overexcavation amount inference model 11.

[0089] As in the explanation during learning, the over-digging amount d estimated in this embodiment is at the midpoint position CP between a pair of outermost charging holes OH. The over-excavation amount estimation unit 6 inputs the plan input data generated for a pair of outermost charging holes OH and the drilling data related to the pair of outermost charging holes OH into the learned over-excavation amount inference model 11, and estimates the over-excavation amount d at the midpoint position CP between the pair of outermost charging holes OH, sequentially executing this for all pairs of adjacent outermost charging holes OH in the drilling and blasting plan P that has been accepted as input for estimating the over-excavation amount d.

[0090] The result output unit 7 displays the estimation result of the overexcavation amount d on a display device (not shown) such as a display. The result output unit 7 displays, for example, the tunnel face M and the position of the charge hole H as part of the drilling and blasting plan P, superimposed as shown in Figure 2. Then, the result output unit 7 displays the design position PL, and plots and displays the overexcavation amount d as a position relative to the design position PL.

[0091] Next, an overexcavation amount estimation method using the above-mentioned overexcavation amount estimation device 1 will be described with reference to Figures 1 to 12, 13 and 14. Figure 13 is a flowchart showing the process flow when learning the overexcavation amount inference model. Figure 14 is a flowchart showing the process flow when inferring the overexcavation amount using the learned overexcavation amount inference model.

[0092] First, the flow of processing when learning the overcut amount d will be described. When learning the overcut amount d, learning data is prepared (step S1). The drilling and blasting plan input receiving unit 2 generates learning-time plan input data for learning the overexcavation amount d based on the drilling and blasting plan P in drilling and blasting performed in the past, to learn the overexcavation amount inference model 11L and generate the learned overexcavation amount inference model 11. The drilling and blasting plan input receiving unit 2 generates one learning-time plan input data for a pair of outermost charge holes OH of the drilling and blasting plan P received as input for the purpose of learning the overexcavation amount d.

[0093] Next, the drilling data acquisition unit 4 acquires drilling data related to the drilling blasting plan P in a drilling blasting operation carried out in the past, which was input to the drilling blasting plan input receiving unit 2 and served as the basis for generating the learning-time plan input data. The drilling data acquisition unit 4 acquires, as learning-time drilling data, which is drilling data for training the overexcavation amount inference model 11L to generate the learned overexcavation amount inference model 11, the drilling data for the outermost charge holes OH that were drilled at positions corresponding to each of the set of outermost charge holes OH for which the learning-time plan input data was generated, in the drilling carried out immediately before the drilling blasting carried out in the past.

[0094] Furthermore, when drilling and blasting is carried out based on the drilling and blasting plan P from which the learning-time plan input data was generated, the actual measured value of the over-excavation amount d at the midpoint position CP of the pair of outermost charging holes OH corresponding to the learning-time plan input data is prepared as the correct value, i.e., teacher data. The combination of the learning-time planning input data, learning-time drilling data, and teacher data generated for a pair of outermost charging holes OH constitutes a set of learning data.

[0095] The learning unit 3 uses the learning time plan input data generated by the drilling and blasting plan input receiving unit 2 based on the drilling and blasting plan P for a drilling and blasting operation performed in the past, and the learning time drilling data related to the drilling and blasting plan P acquired by the drilling data acquisition unit 4 as learning time input data, and learns the overexcavation amount inference model 11L to infer the overexcavation amount d after the drilling and blasting operation is performed, thereby generating a learned overexcavation amount inference model 11 (step S2).

[0096] Next, the flow of processing when estimating the overexcavation amount d will be described. First, at the stage where drilling and blasting is about to be carried out for a tunnel face M based on a certain drilling and blasting plan P, the drilling data acquired by the rock drill 200 in the drilling and blasting carried out immediately before the drilling and blasting that is about to be carried out, i.e., the drilling data acquired in the previous drilling and blasting for the tunnel face M (e.g., tunnel face M1 in Figure 1) that resulted in the appearance of the above-mentioned tunnel face M (e.g., tunnel face M2 in Figure 1), is acquired (step S11: drilling data acquisition step).

[0097] Next, an operator formulates a drilling and blasting plan P for the tunnel face M. This drilling and blasting plan P is input to the overexcavation amount estimation device 1. The drilling and blasting plan input receiving unit 2 receives the input of the drilling and blasting plan P (step S12: drilling and blasting plan input receiving step). The drilling and blasting plan input receiving unit 2 generates plan input data to be input to the learned overexcavation amount inference model 11 based on the input drilling and blasting plan P. The drilling and blasting plan input receiving unit 2 generates one plan input data for a pair of outermost charging holes OH of the drilling and blasting plan P received as input, in the same manner as generating plan input data during learning when learning the overexcavation amount inference model 11L. The drilling and blasting plan input receiving unit 2 generates plan input data from the drilling and blasting plan P that has been input and received as an object for estimating the overexcavation amount d, the number of which is equal to the number of combinations of adjacent outermost charging holes OH. The drilling data acquisition unit 4 extracts drilling data related to the pair of outermost charging holes OH for which the planned input data was generated from the drilling data acquired in step S11, so as to correspond to each of the planned input data generated as described above.

[0098] The overexcavation amount estimation unit 6 inputs the plan input data generated by the drilling and blasting plan input receiving unit 2 based on the drilling and blasting plan P for the drilling and blasting that is about to be executed as described above, and the drilling data related to the drilling and blasting plan P acquired by the drilling data acquisition unit 4 into the learned overexcavation amount inference model 11 (step S13: overexcavation amount estimation step). The over-excavation amount estimation unit 6 inputs the plan input data generated for a pair of outermost charging holes OH and the drilling data related to the pair of outermost charging holes OH into the learned over-excavation amount inference model 11, and estimates the over-excavation amount d at the midpoint position CP between the pair of outermost charging holes OH, sequentially executing this for all pairs of adjacent outermost charging holes OH in the drilling and blasting plan P that has been accepted as input for estimating the over-excavation amount d.

[0099] The result output unit 7 displays the estimation result of the overexcavation amount d on a display device (not shown) such as a display (step S14). Based on the estimation results displayed in this way, the operator judges whether the drilling and blasting plan P is appropriate (step S15). This judgment is made, for example, by confirming that the overexcavation amount d is sufficiently close to the design position PL. If it is determined that the drilling and blasting plan P is valid (Yes in step S15), drilling and blasting is carried out based on the drilling and blasting plan P (step S16). If the drilling and blasting plan P is determined to be inappropriate (No in step S15), the process returns to step S12, and the drilling and blasting plan P is re-drafted (or the drilling and blasting plan P previously input into the overexcavation amount estimation device 1 is adjusted) and the processes from estimation of the overexcavation amount d onwards are repeated until the drilling and blasting plan P is determined to be appropriate in step S15.

[0100] The overexcavation amount estimation device 1 as described above is an overexcavation amount estimation device 1 that estimates an overexcavation amount d, which is the difference between the position AL of the wall surface in the radial direction of the tunnel T and the design position PL planned at the time of design when drilling and blasting at the tunnel face M based on a drilling and blasting plan P, when drilling and blasting is performed, and the drilling and blasting plan P includes a drilling pattern, which includes the position of the outermost charge hole OH, which is the charge hole H at the outermost periphery of the tunnel face M, and includes a drilling and blasting plan input receiving unit 2 that receives input of the drilling and blasting plan P, and a drilling data acquisition unit 4 that acquires drilling data by the rock drill 200. The system is equipped with an overexcavation amount estimation unit 6 that inputs, as learning input data, plan input data generated based on a drilling blasting plan P in a drilling blasting operation performed in the past, and drilling data related to the drilling blasting plan P at the time of learning, and is trained to infer the overexcavation amount d after the drilling blasting is performed, and inputs plan input data generated based on the drilling blasting plan P that has been accepted as an input for estimating the overexcavation amount d, and drilling data related to the drilling blasting plan P, and estimates the overexcavation amount d when drilling blasting is performed based on the drilling blasting plan P.

[0101] When drilling and blasting the tunnel face M, the area crushed by the explosives loaded in one charge hole H is basically a limited area around the charge hole H. Therefore, the more the outermost charge hole OH, which is the charge hole H at the outermost periphery of the tunnel face M, is located radially inward of the tunnel T, the more of the crushed area is included inside the design position PL of the wall surface planned at the time of design, and the smaller the area excavated radially outward from the design position PL. Conversely, the more the outermost charge hole OH is located radially outward of the tunnel T, the more of the crushed area protrudes outside the design position PL, and the larger the area excavated radially outward from the design position PL. Thus, when drilling and blasting is performed, there is a correlation between the overexcavation amount d, which is the difference between the wall position AL in the radial direction of the tunnel T and the design position PL planned at the time of design, and the position of the outermost charge hole OH. In contrast, according to the above-described configuration, the overexcavation amount estimation device 1 is equipped with a learned overexcavation amount inference model 11 that is trained to infer the overexcavation amount d after drilling and blasting is performed by inputting, as part of the training input data, training-time plan input data generated based on the drilling and blasting plan P of a previous drilling and blasting operation. Here, the drilling and blasting plan P, which is the basis for generating the training-time plan input data input during learning, includes a drilling pattern including the position of the outermost charge hole OH. In this way, the trained overexcavation amount inference model 11 is configured to be able to grasp the position of the outermost charge hole OH and is trained to infer the overexcavation amount d based on this. By inputting plan input data including the position of the outermost charge hole OH, as well as the training-time plan input data generated based on the drilling and blasting plan P accepted by the drilling and blasting plan input receiving unit 2 as the input for estimating the overexcavation amount d, into the trained overexcavation amount inference model 11, the overexcavation amount d when drilling and blasting is performed based on the input drilling and blasting plan P can be estimated. Here, the size and shape of the area crushed by the explosives loaded in one charging hole H will vary depending on the properties of the rock mass. Therefore, there is a correlation between the properties of the rock mass and the over-excavation amount d. In contrast, with the above-described configuration, during learning, the learned overexcavation amount inference model 11 receives, as learning-time plan input data, learning-time plan input data generated based on the drilling blasting plan P for drilling blasting performed in the past, as described above, as well as learning-time drilling data related to the drilling blasting plan P. Because the drilling data is acquired by the rock drill 200 during drilling, the learning-time drilling data accurately and clearly reflects the properties of the rock mass at the time of drilling. In this way, the learned overexcavation amount inference model 11 is configured to be able to grasp the properties of the rock mass, and is trained to infer the overexcavation amount d based on this. By inputting drilling data related to the drilling blasting plan P, along with plan input data generated based on the drilling blasting plan P that the drilling blasting plan input receiving unit 2 has received as input for estimating the overexcavation amount d, into such a learned overexcavation amount inference model 11, the overexcavation amount d when drilling blasting is performed based on the input drilling blasting plan P can be more accurately estimated, taking into account the properties of the bedrock. In this way, an overexcavation amount estimation device 1 can be provided that, when drilling and blasting the tunnel face M, accurately estimates the overexcavation amount d that will be produced when drilling and blasting is performed based on the drilling and blasting plan P.

[0102] In addition, the learning planning input data and the planning input data include the distance of the outermost charging hole OH from the design position PL, calculated based on the position of the outermost charging hole OH included in the drilling pattern.

[0103] In the above, it has been explained that there is a correlation between the position of the outermost charge hole OH and the overexcavation amount d. Here, since the overexcavation amount d is the difference between the wall position AL and the design position PL in the radial direction of the tunnel T after drilling and blasting, it is considered that what is correlated with the overexcavation amount d is particularly the relative positional relationship between the outermost charge hole OH and the design position PL, that is, the distance of the outermost charge hole OH from the design position PL. In contrast, according to the above configuration, the learning-time plan input data includes the distance of the outermost charging hole OH from the design position PL, calculated based on the position of the outermost charging hole OH included in the drilling pattern. In this way, the learned over-excavation amount inference model 11 is configured to be able to grasp the distance of the outermost charging hole OH from the design position PL, and is learned to infer the over-excavation amount d based on this. Furthermore, the planned input data that the over-excavation amount estimation unit 6 inputs to the learned over-excavation amount inference model 11 when estimating the over-excavation amount d also includes, like the planned input data at the time of learning, the distance of the outermost charge hole OH from the design position PL, calculated based on the position of the outermost charge hole OH included in the drilling pattern. Therefore, as described above, data consistent with the planned input data at the time of learning input during learning is input to the learned over-excavation amount inference model 11. In this way, when drilling and blasting the tunnel face M, the amount of overexcavation d that will be achieved when drilling and blasting is performed can be estimated more accurately based on the drilling and blasting plan P.

[0104] Further, the learning time plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling blasting plan P based on the drilling blasting plan P in the drilling blasting performed in the past, and the learned over-excavation amount inference model 11 is trained to infer the over-excavation amount d at the midpoint position CP between the pair of outermost charge holes OH after the drilling blasting is performed by inputting the learning time plan input data and the learning time drilling data related to the pair of outermost charge holes OH as learning time input data. The plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling and blasting plan P, based on the drilling and blasting plan P that has been received as an input for estimating the overexcavation amount d, and the overexcavation amount estimation unit 6 inputs the plan input data and drilling data related to the pair of outermost charge holes OH into the learned overexcavation amount inference model 11, and estimates the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH when drilling and blasting is performed based on the drilling and blasting plan P.

[0105] As already explained, when drilling and blasting the tunnel face M, the area that is crushed by the explosives loaded in one charge hole H is basically a limited area around the charge hole H. When loading and blasting the outermost charge hole OH, it is possible for workers to predict the amount of overexcavation d in the radially outer part of the tunnel T than the outermost charge hole OH. However, it is not easy for workers to predict the amount of overexcavation d at a position between two outermost charge holes OH, for example, because the relationship between the two outermost charge holes OH must be taken into consideration. In contrast, with the above-described configuration, learning plan input data is generated for a pair of adjacent outermost charge holes OH in a drilling blasting plan P in a previous drilling blasting operation, and the learned overexcavation amount inference model 11 is trained to infer the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH after the drilling blasting operation is performed using the learning plan input data as learning input data. Furthermore, plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling blasting plan P for which input is accepted as an estimation target for the overexcavation amount d, and the overexcavation amount estimation unit 6 inputs the plan input data into the learned overexcavation amount inference model 11 to estimate the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH when drilling blasting is performed based on the drilling blasting plan P. With this configuration, it is possible to estimate the overexcavation amount d at the position between the two outermost charge holes OH, which is difficult for operators to predict.

[0106] In particular, the midpoint position CP between a pair of adjacent outermost charge holes OH is a portion where no charge holes H are provided. Therefore, at the radially outer positions, the rock mass is not sufficiently excavated by blasting, increasing the possibility that a strike will occur rather than an overexcavation. In this embodiment, the amount of overexcavation d in such portions where strikes are particularly likely to occur is estimated. This allows workers to confirm the amount of overexcavation d in portions where strikes are likely to occur, and allows them to plan a drilling and blasting plan P that reduces strikes. Therefore, the extension of the construction period due to chipping off strikes is efficiently prevented.

[0107] In addition, the learned overexcavation amount inference model 11 learns by inputting, as learning drilling data, the drilling data of the outermost charge holes OH that were drilled at positions corresponding to each of the pair of outermost charge holes OH for which the learning plan input data was generated in drilling performed immediately before a drilling blasting performed in the past, and the overexcavation amount estimation unit 6 inputs, as drilling data, the drilling data of the outermost charge holes OH that were drilled at positions corresponding to each of the pair of outermost charge holes OH for which the plan input data was generated in drilling performed immediately before a drilling blasting performed based on the drilling blasting plan P that was accepted as input as the target for estimating the overexcavation amount d into the learned overexcavation amount inference model 11.

[0108] The overexcavation amount estimation unit 6 inputs plan input data generated based on the drilling and blasting plan P for the tunnel face M where drilling and blasting are about to be performed into the learned overexcavation amount inference model 11, and estimates the overexcavation amount d when drilling and blasting is performed based on the drilling and blasting plan P. An overexcavation amount estimation device 1 equipped with such an overexcavation amount estimation unit 6 is used, for example, to verify the validity of the drilling and blasting plan P. Here, drilling data is input to the learned overexcavation amount inference model 11 along with the plan input data. However, because drilling and blasting based on the drilling and blasting plan P that has been accepted as an input for estimating the overexcavation amount d is about to be performed, drilling data corresponding to the drilling and blasting plan P itself, on which the plan input data is based, cannot exist in the first place. Here, the tunnel face M about to be drilled and blasted has emerged from the immediately preceding (actually the last) drilling and blasting, and the rock mass of the immediately preceding (previous) tunnel face M where this immediately preceding drilling and blasting was performed is considered to have properties similar to those of the rock mass of the tunnel face M about to be drilled and blasted. Therefore, in the above configuration, the drilling data acquired from the immediately preceding drilling is used as the drilling data. In particular, as described above, the learned overexcavation amount inference model 11 receives the planned input data generated for a pair of adjacent outermost charge holes OH and estimates the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH. Therefore, in order to deal with this, in the above configuration, as the drilling data, in the drilling performed immediately before, the drilling data of the outermost charge hole OH drilled at a position corresponding to the pair of outermost charge holes OH that is considered to have rock properties particularly similar to those of each of the pair of outermost charge holes OH for which the plan input data was generated, for example, so that the relative positions from the center of the tunnel in a plane perpendicular to the axial direction AC of the tunnel T are approximately the same, is input to the learned overexcavation amount inference model 11. As a result, even if there is no drilling data corresponding to the drilling blasting plan P, the overexcavation amount d can be accurately estimated. Furthermore, in the above-described configuration, the learned overbreak amount inference model 11 is trained by inputting, as training drilling data, drilling data for the outermost charge holes OH drilled at positions corresponding to each of the pair of outermost charge holes OH for which the training plan input data was generated in the drilling performed immediately before (the previous drilling blasting) the previous drilling blasting. In this way, when the learned overbreak amount inference model 11 is trained, the relationship between the training plan data and the training drilling data can be made the same as the relationship between the plan data and the drilling data input when estimating the overbreak amount d. In this way, by aligning the data input to the learned overbreak amount inference model 11 during training and inference, it is possible to further improve the estimation accuracy of the overbreak amount d.

[0109] In addition, the over-excavation amount estimation unit 6 inputs the plan input data generated for a pair of outermost charging holes OH and the drilling data related to the pair of outermost charging holes OH into the learned over-excavation amount inference model 11, and estimates the over-excavation amount d at the midpoint position CP between the pair of outermost charging holes OH, sequentially executing this for all pairs of adjacent outermost charging holes OH in the drilling and blasting plan P that has been accepted as input for estimating the over-excavation amount d.

[0110] According to the above configuration, the overexcavation amount estimation unit 6 estimates the overexcavation amount d at the midpoint position CP between a pair of outermost outermost charge holes OH, sequentially for all pairs of adjacent outermost outermost charge holes OH in the drilling and blasting plan P that has been input as an estimation target for the overexcavation amount d. This makes it possible to estimate the overexcavation amount d over the entire circumference of the tunnel face M that is about to be drilled and blasted.

[0111] In addition, the learned overexcavation amount inference model 11 uses the overexcavation amount d after actual drilling and blasting as training data, and is trained so that the inferred overexcavation amount d is close to the training data, and the overexcavation amount d used as training data is calculated based on the results measured by 3D scanning the excavated surface after drilling and blasting.

[0112] According to the above configuration, the overexcavation amount d in the actual drilling and blasting can be accurately obtained and used as training data.

[0113] In addition, the learning drilling data and the drilling data are any one of the depth of the charge hole H, drilling speed, impact pressure, feed pressure, damper pressure, rotation speed, rotation pressure, water volume, and water pressure, or any combination thereof.

[0114] According to the above configuration, the properties of the rock mass can be appropriately reflected in the learning drilling data and the drilling data.

[0115] The overexcavation amount estimation program as described above is an overexcavation amount estimation program that estimates an overexcavation amount d, which is the difference between the wall position AL in the radial direction of the tunnel T and the design position PL planned at the time of design when drilling and blasting at the tunnel face M based on a drilling and blasting plan P, when drilling and blasting is performed. The drilling and blasting plan P includes a drilling pattern, and the drilling pattern includes the position of the outermost charge hole OH, which is the charge hole H at the outermost periphery of the tunnel face M. The computer has a drilling and blasting plan input receiving function that receives input of the drilling and blasting plan P, and a drilling and blasting function that acquires drilling data by the rock drill 200. The system is provided with a data acquisition function, and an overexcavation amount estimation function that inputs plan input data generated based on a drilling blasting plan P from a previous drilling blasting operation and drilling data related to the drilling blasting plan P as learning input data, and uses a learned overexcavation amount inference model 11 that is trained to infer the overexcavation amount d after the drilling blasting is performed, and inputs plan input data generated based on the drilling blasting plan P that has been accepted as an input for estimating the overexcavation amount d and drilling data related to the drilling blasting plan P, and estimates the overexcavation amount d when drilling blasting is performed based on the drilling blasting plan P. Furthermore, the above-described overexcavation amount estimation method is an overexcavation amount estimation method for estimating an overexcavation amount d, which is the difference between the wall position AL in the radial direction of the tunnel T and the design position PL planned at the time of design when drilling and blasting at the tunnel face M based on a drilling and blasting plan P, in which the drilling and blasting plan P includes a drilling pattern, and the drilling pattern includes the position of the outermost charge hole OH, which is the charge hole H at the outermost periphery of the tunnel face M, and includes a drilling data acquisition step (step S11) for acquiring drilling data by the rock drill machine 200, and a drilling and blasting plan input acceptance step (step S20) for accepting input of the drilling and blasting plan P. The method includes an overexcavation amount estimation step (step S13) in which learning plan input data generated based on a drilling blasting plan P in a drilling blasting operation performed in the past and learning drilling data related to the drilling blasting plan P are input as learning input data, and the learned overexcavation amount estimation model 11, which has been trained to infer the overexcavation amount d after the drilling blasting operation is performed, inputs plan input data generated based on the drilling blasting plan P that has been accepted as an input for estimating the overexcavation amount d and drilling data related to the drilling blasting plan P, and estimates the overexcavation amount d when drilling blasting is performed based on the drilling blasting plan P. In addition, the above-mentioned learned overexcavation amount inference model 11 is a learned overexcavation amount inference model 11 that is used to estimate the overexcavation amount d, which is the difference between the wall position AL in the radial direction of the tunnel T and the design position PL planned at the time of design when drilling and blasting the tunnel face M based on the drilling and blasting plan P, where the drilling and blasting plan P includes a drilling pattern, which includes the position of the outermost charge hole OH, which is the charge hole H at the outermost periphery of the tunnel face M, and the learning plan input data generated based on the drilling and blasting plan P from a drilling and blasting performed in the past and the learning drilling data related to the drilling and blasting plan P are input as learning input data, and the model is trained to infer the overexcavation amount d after the drilling and blasting is performed.

[0116] According to the above configuration, similar to the above description regarding the overexcavation amount estimation device 1, when drilling and blasting the tunnel face M, the overexcavation amount d that will be produced when drilling and blasting is performed can be accurately estimated based on the drilling and blasting plan P.

[0117] In addition, in the above-described overexcavation amount estimation method, the learning-time plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling blasting plan P based on the drilling blasting plan P in a drilling blasting operation performed in the past, and the learned overexcavation amount inference model 11 is trained to infer the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH after the drilling blasting operation is performed by inputting the learning-time plan input data and the learning-time drilling data related to the pair of outermost charge holes OH as learning-time input data. The learned overexcavation amount inference model 11 estimates the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH when drilling and blasting is performed based on the drilling and blasting plan P, and plan input data is generated for a pair of adjacent outermost charge holes OH in the drilling and blasting plan P, and in the overexcavation amount estimation step (step S13), the plan input data and drilling data related to the pair of outermost charge holes OH are input into the learned overexcavation amount inference model 11, and the overexcavation amount d at the midpoint position CP between the pair of outermost charge holes OH when drilling and blasting is performed based on the drilling and blasting plan P is estimated.

[0118] According to the above configuration, similarly to the above description regarding the over-excavation amount estimation device 1, it is possible to estimate the over-excavation amount d at the position between the two outermost charging holes OH, which is difficult for an operator to predict.

[0119] In addition, in the above-mentioned overexcavation amount estimation method, the learned overexcavation amount inference model 11 learns the drilling data input from the drilling performed immediately before the drilling blasting performed in the past, at positions corresponding to each of the pair of outermost charge holes OH for which the learning-time plan input data was generated, and in the overexcavation amount estimation step (step S13), the drilling data input from the learned overexcavation amount inference model 11 is the drilling data input from the drilling performed immediately before the drilling blasting performed based on the drilling blasting plan P for which the overexcavation amount d was estimated, at positions corresponding to each of the pair of outermost charge holes OH for which the plan input data was generated.

[0120] According to the above configuration, similar to the above description regarding the overexcavation amount estimation device 1, it is possible to accurately estimate the overexcavation amount d even if there is no drilling data corresponding to the drilling and blasting plan P. Furthermore, by aligning the data input into the learned overexcavation amount inference model 11 at the time of learning and at the time of inference, it is possible to further improve the estimation accuracy of the overexcavation amount d.

[0121] (Verification example) Next, a verification example using the overexcavation amount estimation device 1 of the above embodiment will be described. FIG. 15 is an explanatory diagram showing the verification results using the overexcavation amount estimation device of the above embodiment. In FIG. 15, the actual measured values ​​of the overexcavation amount d are plotted as black circles, and the estimated values ​​of the overexcavation amount d by the overexcavation amount estimation device 1 are plotted as triangles. FIG. 15 shows that the overexcavation amount d is accurately estimated by the overexcavation amount estimation device 1.

[0122] (First Modification of the Embodiment) The drilling and blasting plan P is a plan for drilling and blasting. Therefore, the drilling and blasting plan P may be configured to include the amount of explosives to be loaded in each charge hole H, particularly the outermost charge hole OH. In this case, in addition to the input node 11p described as the above embodiment, the over-excavation amount inference model 11L and the learned over-excavation amount inference model 11 are provided with a new input node 11p to which the charge amount loaded in each of the pair of outermost charging holes OH is input. In this case, one input node 11p corresponding to the charge amount may be provided, and the average value of each of the charge amounts of the pair of outermost charging holes OH may be input to this input node 11p. Alternatively, two input nodes 11p corresponding to the charge amount may be provided, and each of the charge amounts of the pair of outermost charging holes OH may be input to these two input nodes 11p.

[0123] Thus, in this modified example, the drilling and blasting plan P includes the amount of charge for the outermost charge hole OH.

[0124] In the outermost charging hole OH, if the amount of charge is large, the crushed area becomes large, so the area excavated radially outward from the design position PL becomes large. Also, if the amount of charge is small, the crushed area becomes small, so the area excavated radially outward from the design position PL becomes small. Thus, in the outermost charging hole OH, there is a correlation between the amount of charge and the over-excavation amount d. In contrast, according to the above-described configuration, the drilling blasting plan P includes the charge amount for the outermost charge hole OH. The learned overexcavation amount inference model 11 is configured to be able to grasp the charge amount for the outermost charge hole OH because the learning-time plan input data generated based on the drilling blasting plan P in a drilling blasting operation performed in the past is input as (part of) the learning-time input data, and is trained to infer the overexcavation amount d based on this. By inputting plan input data including the charge amount similar to the learning-time plan input data generated based on the drilling blasting plan P that the drilling blasting plan input receiving unit 2 received as an input for estimating the overexcavation amount d into this learned overexcavation amount inference model 11, the overexcavation amount d when drilling blasting is performed based on the input drilling blasting plan P can be more accurately estimated taking into account the charge amount.

[0125] (Second Modification of the Embodiment) Next, a second modified example of the above embodiment will be described. Fig. 16 is a block diagram of an overexcavation amount estimation device in this second modified example. This modified example differs from the above embodiment in that the overexcavation amount estimation device 1A is equipped with a learned overexcavation amount adjustment model 12. By configuring the overexcavation amount estimation device 1 as in the above embodiment, it is basically possible to estimate the overexcavation amount with high accuracy. However, for example, as a result of training the trained overexcavation amount inference model 11 to follow the trends of the data at the time of training, the overexcavation amount is output with high accuracy for the data at the time of training, but the accuracy decreases for unknown data, which is called overtraining, and in some cases, the accuracy of the overexcavation amount estimation may not be high due to such factors as overtraining. The overexcavation amount estimation device 1A of this modified example treats the overexcavation amount estimated by the learned overexcavation amount inference model 11 as a provisional overexcavation amount, inputs this provisional overexcavation amount to the learned overexcavation amount adjustment model 12 to estimate an adjusted overexcavation amount, and outputs the adjusted overexcavation amount as the overexcavation amount finally estimated by the overexcavation amount estimation device 1A. This improves the accuracy of estimating the overexcavation amount.

[0126] First, the behavior of each component of the overexcavation amount estimation device 1A when learning the overexcavation amount will be described. The drilling and blasting plan input receiving unit 2 receives input of drilling and blasting plans that have been carried out in the past for the purpose of learning the amount of overexcavation. The drilling and blasting plan input receiving unit 2 generates learning plan input data based on drilling and blasting plans for drilling and blasting performed in the past, in the same way as in the above embodiment. The learning input data in this modified example is a combination of the distances l1 and l2 between the outermost charge holes OH and the design position PL, the distance L between the outermost charge holes OH, and the average angle, as in the above embodiment, but may be configured in other ways.

[0127] Next, the drilling data acquisition unit 4 acquires, as training drilling data, the drilling data of the outermost charge holes OH drilled at positions corresponding to each of the set of outermost charge holes OH for which learning-time plan input data was generated in the drilling performed immediately before the previous drilling and blasting. The learning drilling data in this modified example is, as in the above embodiment, any one or any combination of the depth of the charge holes H, the drilling speed, the impact pressure, the feed pressure, the damper pressure, the rotation speed, the rotation pressure, the water volume, and the water pressure. The learning-time drilling data is adjusted as appropriate by the pre-processing unit 5, similarly to the above embodiment.

[0128] In this way, the learning input data generated based on the drilling and blasting plan of the drilling and blasting performed in the past and the learning drilling data related to the drilling and blasting plan are collected as the learning input data. In addition, as explained in the first modified example, the learning input data may include the amount of charge. As in the above embodiment, the learning unit 3 learns the overexcavation amount inference model 11L based on the learning input data and the actual value of the overexcavation amount as teacher data corresponding to the learning input data. More specifically, when learning input data is input, the learning unit 3 learns the overexcavation amount inference model 11L so as to infer the actual value of the overexcavation amount corresponding to the learning input data, and generates the learned overexcavation amount inference model 11. In this modified example, the overexcavation amount inference model 11L and the learned overexcavation amount inference model 11 are RF (random forest), but they may also be realized by other methods such as neural networks or SVM (support vector machines).

[0129] FIG. 17 is an explanatory diagram showing the flow of data when learning the over-digging amount inference model and the over-digging amount adjustment model. Once the learning of the learned over-excavation amount inference model 11 is completed, the learning unit 3 learns the over-excavation amount adjustment model 12L. Specifically, the learning unit 3 uses as input data the learning input data used when learning the overexcavation amount inference model 11L and the provisional overexcavation amount inferred by inputting the learning input data into the learned overexcavation amount inference model 11, and uses as training data the actual value of the overexcavation amount corresponding to the learning input data that is the same as that used when generating the learned overexcavation amount inference model 11, to learn the overexcavation amount adjustment model 12L. In this way, the learning unit 3 trains the overexcavation amount adjustment model 12L so that when the learning input data and the provisional overexcavation amount are input, the overexcavation amount adjustment model 12L infers an overexcavation amount adjusted so that the provisional overexcavation amount is a more accurate value, and generates the learned overexcavation amount adjustment model 12. The learning unit 3 generates a learned over-excavation amount adjustment model 12 as a learned model that is used as a program module that is part of artificial intelligence software, with learning parameters such as weights learned, similar to the learned over-excavation amount inference model 11. In this modified example, the overexcavation amount adjustment model 12L and the learned overexcavation amount adjustment model 12 are fully connected neural networks, but they may also be realized using other methods such as RF (random forest) or SVM (support vector machine). As will be explained later, when actually estimating the overexcavation amount, the adjusted overexcavation amount output by the learned overexcavation amount adjustment model 12 learned as described above becomes the estimated value of the overexcavation amount that is ultimately output as the estimation result by the overexcavation amount estimation device 1A.

[0130] Next, the accuracy of overexcavation amount prediction by the learned overexcavation amount inference model 11 and the learned overexcavation amount adjustment model 12 is verified. For example, when learning input data and actual values ​​of over-excavation amounts as teacher data corresponding to the learning input data are collected, a portion of this data is separated as verification data and actual values ​​of over-excavation amounts corresponding to the verification data. These verification data and actual values ​​of over-excavation amounts corresponding to the verification data are not used for training the learned over-excavation amount inference model 11 and the learned over-excavation amount adjustment model 12, but are used for the purpose of verifying the prediction accuracy of the learned over-excavation amount inference model 11 and the learned over-excavation amount adjustment model 12.

[0131] Specifically, the learning unit 3 inputs the verification data into the learned overexcavation amount inference model 11 and obtains a provisional overexcavation amount. Next, the learning unit 3 inputs the verification data and the provisional overexcavation amount obtained by inputting the verification data into the learned overexcavation amount inference model 11 into the learned overexcavation amount adjustment model 12 and obtains an adjusted overexcavation amount. The learning unit 3 verifies the prediction accuracy of the overexcavation amount by comparing the actual value of the overexcavation amount corresponding to the verification data with the adjusted overexcavation amount obtained from the learned overexcavation amount adjustment model 12 for the verification data. It is desirable that the learning unit 3 repeats learning of the learned over-excavation amount inference model 11 and the learned over-excavation amount adjustment model 12 until the prediction accuracy becomes sufficient.

[0132] Next, the behavior of each component when estimating the overexcavation amount will be described. First, the drilling and blasting plan input receiving unit 2 generates plan input data in the same manner as in the above embodiment. Further, the punching data acquisition unit 4 acquires punching data in the same manner as in the above embodiment. The punching data is adjusted as appropriate by the pre-processing unit 5 in the same manner as in the above embodiment.

[0133] FIG. 18 is an explanatory diagram showing the flow of data when inferring the amount of excess digging using a learned excess digging amount inference model and a learned excess digging amount adjustment model for which learning has been completed. Thereafter, the overexcavation amount estimation unit 6A inputs the planned input data and the drilling data into the learned overexcavation amount inference model 11. The overexcavation amount estimation unit 6A obtains a provisional overexcavation amount as the output of the learned overexcavation amount inference model 11. Then, the overexcavation amount estimation unit 6A inputs the planned input data and drilling data input to the learned overexcavation amount inference model 11, and the provisional overexcavation amount estimated by the learned overexcavation amount inference model 11 for the planned input data and the drilling data, to the learned overexcavation amount adjustment model 12. The overexcavation amount estimation unit 6A acquires, as the output of the learned overexcavation amount adjustment model 12, an adjusted overexcavation amount that has been adjusted to have higher accuracy from the provisional overexcavation amount input to the learned overexcavation amount adjustment model 12. The overexcavation amount estimation unit 6A outputs the adjusted overexcavation amount as an estimated overexcavation amount.

[0134] In this way, in this modified example, the overexcavation amount estimation unit 6A uses the overexcavation amount estimated by the learned overexcavation amount inference model 11 as the provisional overexcavation amount, and inputs the learning input data and the provisional overexcavation amount inferred by inputting the learning input data into the learned overexcavation amount inference model 11, and inputs the planned input data, drilling data, and the provisional overexcavation amount estimated by the learned overexcavation amount inference model 11 for the planned input data and the drilling data into the learned overexcavation amount adjustment model 12, which has been trained to infer the adjusted overexcavation amount, to estimate the adjusted overexcavation amount, and outputs the adjusted overexcavation amount as the estimated overexcavation amount. With this configuration, even if the accuracy of the over-excavation amount output by the learned over-excavation amount inference model 11 is not high due to over-learning or the like, this over-excavation amount is input as a provisional over-excavation amount to the learned over-excavation amount adjustment model 12, and the adjusted over-excavation amount is output, thereby adjusting the value of the over-excavation amount. This makes it possible to improve the accuracy of the over-excavation amount.

[0135] (Verification example of the second modified example) Next, a verification example of the overexcavation amount estimation device 1A described above as the second modified example will be described. In this verification, the distance L between the outermost charge holes OH and the average angle were used as planned input data. In addition, in order to reduce the number of input data and the calculation load, principal component analysis was performed on the maximum, minimum, and average values ​​of the depth of the charge hole H, drilling speed, impact pressure, feed pressure, damper pressure, rotation speed, and rotation pressure, and the extracted first to sixth principal components and correlation coefficients were used as drilling data. Examples 1 and 2 were prepared, which were configured to use the above-mentioned planning input data and drilling data for learning and estimation. In Example 1, only the first to sixth principal components were used as the drilling data. In Example 2, both the first to sixth principal components and the correlation coefficient were used as the drilling data. For each of Examples 1 and 2, the learning model was trained and the over-excavation amount was estimated in two cases: when only the learned over-excavation amount inference model 11, which corresponds to the above embodiment, was used, and when both the learned over-excavation amount inference model 11 and the learned over-excavation amount adjustment model 12, which corresponds to the above second variant example, were used.

[0136] Fig. 19 is a table showing the verification results. In Fig. 19, "one-stage AI" indicates the case where only the learned over-excavation amount inference model 11 is used, which corresponds to the above embodiment, and "two-stage AI" indicates the case where the learned over-excavation amount inference model 11 and the learned over-excavation amount adjustment model 12 are used, which corresponds to the above second modified example. As shown in Figure 19, for each case, the root mean square error (RMSE), the coefficient of determination (R 2 ), and the coefficient of determination in the validation data were evaluated. In this verification, overfitting occurred during learning of the learned over-digging amount inference model 11, and the coefficient of determination in the verification data for the "single-stage AI" was low in both Examples 1 and 2. In contrast, in the "two-stage AI" that introduced the learned over-digging amount adjustment model 12, the coefficient of determination in the verification data was improved to a sufficiently high value of 0.72 in both Examples 1 and 2.

[0137] (Other Modifications of the Embodiment) In addition, the over-excavation amount estimation device, over-excavation amount estimation program, over-excavation amount estimation method, and learned over-excavation amount inference model used in the over-excavation amount estimation device of the present invention are not limited to the above-mentioned embodiments and modified examples described with reference to the drawings, and various other modified examples are possible within their technical scope.

[0138] For example, in the above embodiment, the learned over-excavation amount inference model 11 has multiple output nodes 11n corresponding to multiple ranges of the value of the over-excavation amount d, and the range corresponding to the output node 11n having the largest value is estimated to be the over-excavation amount d inferred by the learned over-excavation amount inference model 11, but this is not limited to this. For example, the learned over-excavation amount inference model may be configured to have one output node, and the value of the over-excavation amount inferred by the learned over-excavation amount inference model may be directly stored in this one output node.

[0139] Furthermore, in the above embodiment, when the overexcavation amount inference model 11L is trained, the training plan input data is generated based on the drilling blasting plan P of the drilling blasting performed in the past, and the drilling data acquired by the rock drill 200 in the drilling performed immediately before the past drilling blasting is used as the training drilling data, but this is not limited to this. Regarding the drilling blasting plan P on which the training plan input data is generated, drilling blasting has already been performed based on this, so corresponding drilling data should also exist, and this drilling data may be used as the training drilling data. However, even in this case, as already explained, when inferring the overexcavation amount d, drilling and blasting are about to be carried out for the tunnel face M, so drilling data corresponding to the drilling and blasting plan P itself, which is the basis of the plan input data, cannot exist in the first place. Therefore, at the time of inference, as in the above embodiment, it is necessary to input into the learned overexcavation amount inference model 11 the drilling data acquired from the drilling carried out immediately before the drilling and blasting carried out based on the drilling and blasting plan P that has been accepted as input as the target for estimating the overexcavation amount d.

[0140] Furthermore, in the above embodiment, the average angle of a pair of outermost charging holes OH was input into the learned over-excavation amount inference model 11, but instead, the angles of a pair of outermost charging holes OH may be input individually.

[0141] In addition to this, it is possible to select and discard the configurations given in the above embodiment and modified examples, or to change them to other configurations as appropriate. [Explanation of symbols]

[0142] 1. 1A Overexcavation volume estimation device 2. Drilling and blasting plan input reception section 4. Drilling data acquisition section 6, 6A Overexcavation volume estimation section 11 Learned over-mining amount inference model 12 Learned overexcavation adjustment model 200 Jackhammer AL Wall position CP midpoint position H Charge hole M Tunnel Face OH outermost charging hole P Drilling and blasting plan PL design position T-Tunnel d Overexcavation volume

Claims

1. An overexcavation amount estimation device that estimates the amount of overexcavation, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at the tunnel face based on a drilling and blasting plan, and the design position planned at the time of design, The drilling and blasting plan includes a drilling pattern, and the drilling pattern includes positions of outermost charge holes, which are charge holes at the outermost periphery of the tunnel face; A drilling and blasting plan input receiving unit that receives input of the drilling and blasting plan; a drilling data acquisition unit that acquires drilling data by the rock drill; an overexcavation amount estimation unit that inputs, as learning input data, plan input data generated based on the drilling blasting plan in a drilling blasting performed in the past and drilling data related to the drilling blasting plan, and estimates the overexcavation amount when drilling blasting is performed based on the drilling blasting plan by inputting, using a learned overexcavation amount inference model that has been trained to infer the overexcavation amount after the drilling blasting is performed, plan input data generated based on the drilling blasting plan that has been accepted as an input for estimating the overexcavation amount and the drilling data related to the drilling blasting plan; An overexcavation amount estimation device comprising:

2. The learning-time planning input data and the planning input data include a distance of the outermost charging hole from the design position calculated based on the position of the outermost charging hole included in the drilling pattern.

2. The overexcavation amount estimation device according to claim 1.

3. The learning plan input data is generated for a pair of adjacent outermost charge holes in the drilling blasting plan based on the drilling blasting plan in the drilling blasting performed in the past, and the learned overexcavation amount inference model is trained by inputting the learning plan input data and the learning drilling data related to the pair of outermost charge holes as the learning input data, and by inferring the overexcavation amount at the midpoint position between the pair of outermost charge holes after the drilling blasting is performed, The plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan that has been received as an input for estimating the overexcavation amount, and the overexcavation amount estimation unit inputs the plan input data and the drilling data related to the pair of outermost charge holes into the learned overexcavation amount inference model, and estimates the overexcavation amount at the midpoint position between the pair of outermost charge holes when the drilling and blasting is performed based on the drilling and blasting plan.

2. The overexcavation amount estimation device according to claim 1.

4. The learned overexcavation amount inference model is learned by inputting, as the learning time drilling data, the drilling data of the outermost charge holes drilled at positions corresponding to each of the pair of outermost charge holes for which the learning time plan input data is generated in the drilling performed immediately before the drilling blasting performed in the past, The overexcavation amount estimation unit inputs, as the drilling data, the drilling data of the outermost charge holes drilled at positions corresponding to each of the pair of outermost charge holes for which the plan input data is generated in the drilling performed immediately before the drilling blasting performed based on the drilling blasting plan that has been received as input as the overexcavation amount estimation target, into the learned overexcavation amount inference model.

4. The overexcavation amount estimation device according to claim 3.

5. The over-excavation amount estimation unit inputs the plan input data generated for the pair of outermost outer charge holes and the drilling data related to the pair of outermost outer charge holes into the learned over-excavation amount inference model, and estimates the over-excavation amount at the midpoint position between the pair of outermost outer charge holes. This is executed sequentially for all pairs of adjacent outermost outer charge holes in the drilling and blasting plan for which input is accepted as the over-excavation amount estimation target.

5. The overexcavation amount estimation device according to claim 3 or 4.

6. The learned overexcavation amount inference model uses the overexcavation amount after the drilling blasting is actually performed as training data, and is trained so that the inferred overexcavation amount is close to the training data, The overexcavation amount used as the training data is calculated based on the results measured by 3D scanning the unexcavated surface after the drilling blasting.

2. The overexcavation amount estimation device according to claim 1.

7. The learning drilling data and the drilling data are any one of the depth of the charge hole, drilling speed, impact pressure, feed pressure, damper pressure, rotation speed, rotation pressure, water volume, and water pressure, or any combination thereof.

2. The overexcavation amount estimation device according to claim 1.

8. The overexcavation amount estimation device according to claim 1 , wherein the drilling and blasting plan includes a charge amount for the outermost charge hole.

9. A program for estimating the amount of overexcavation, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting at the tunnel face is performed based on a drilling and blasting plan and the design position planned at the time of design, The drilling and blasting plan includes a drilling pattern, and the drilling pattern includes positions of outermost charge holes, which are charge holes at the outermost periphery of the tunnel face; On the computer, A drilling and blasting plan input receiving function for receiving input of the drilling and blasting plan; A drilling data acquisition function that acquires drilling data from a rock drill, an overexcavation amount estimation function that inputs, as learning input data, plan input data generated based on the drilling blasting plan in a drilling blasting performed in the past and drilling data related to the drilling blasting plan, and estimates the overexcavation amount when drilling blasting is performed based on the drilling blasting plan by inputting, using a learned overexcavation amount inference model that has been trained to infer the overexcavation amount after the drilling blasting is performed, plan input data generated based on the drilling blasting plan that has been accepted as an input for estimating the overexcavation amount and the drilling data related to the drilling blasting plan; A program to estimate the amount of overexcavation required to achieve this.

10. A method for estimating the amount of overexcavation, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at the tunnel face based on a drilling and blasting plan and the design position planned at the time of design, The drilling and blasting plan includes a drilling pattern, and the drilling pattern includes positions of outermost charge holes, which are charge holes at the outermost periphery of the tunnel face; a drilling data acquisition step of acquiring drilling data by a rock drill; A drilling and blasting plan input receiving step for receiving input of the drilling and blasting plan; an overexcavation amount estimation step in which learning-time plan input data generated based on the drilling blasting plan in a drilling blasting performed in the past and learning-time drilling data related to the drilling blasting plan are input as learning-time input data, and using a learned overexcavation amount inference model trained to infer the overexcavation amount after the drilling blasting is performed, plan input data generated based on the drilling blasting plan that has been accepted as an input for estimating the overexcavation amount and the drilling data related to the drilling blasting plan are input, and the overexcavation amount is estimated when drilling blasting is performed based on the drilling blasting plan; A method for estimating an amount of overexcavation, comprising:

11. The learning plan input data is generated for a pair of adjacent outermost charge holes in the drilling blasting plan based on the drilling blasting plan in the drilling blasting performed in the past, and the learned overexcavation amount inference model is trained by inputting the learning plan input data and the learning drilling data related to the pair of outermost charge holes as the learning input data, and by inferring the overexcavation amount at the midpoint position between the pair of outermost charge holes after the drilling blasting is performed, The plan input data is generated for a pair of adjacent outermost charge holes in the drilling and blasting plan based on the drilling and blasting plan that has been received as an input for estimating the overexcavation amount, and in the overexcavation amount estimation step, the plan input data and the drilling data related to the pair of outermost charge holes are input to the learned overexcavation amount inference model to estimate the overexcavation amount at the midpoint position between the pair of outermost charge holes when the drilling and blasting is performed based on the drilling and blasting plan. The method for estimating the amount of overexcavation according to claim 10.

12. The learned overexcavation amount inference model is learned by inputting, as the learning time drilling data, the drilling data of the outermost charge holes drilled at positions corresponding to each of the pair of outermost charge holes for which the learning time plan input data is generated in the drilling performed immediately before the drilling blasting performed in the past, In the overexcavation amount estimation step, the drilling data of the outermost charge holes drilled at positions corresponding to each of the pair of outermost charge holes for which the plan input data is generated in the drilling performed immediately before the drilling blasting performed based on the drilling blasting plan that has been received as input as the overexcavation amount estimation target is input to the learned overexcavation amount inference model. The method for estimating the amount of overexcavation according to claim 11 .

13. A trained overexcavation amount inference model used to estimate the overexcavation amount, which is the difference between the position of the wall surface in the radial direction of the tunnel when drilling and blasting is performed at the tunnel face based on a drilling and blasting plan and the design position planned at the time of design, The drilling and blasting plan includes a drilling pattern, and the drilling pattern includes positions of outermost charge holes, which are charge holes at the outermost periphery of the tunnel face; Learning plan input data generated based on the drilling blasting plan in the past drilling blasting and learning drilling data related to the drilling blasting plan are input as learning input data, and the system is trained to infer the overexcavation volume after the drilling blasting is performed. A trained over-mining inference model characterized by:

Images