A tunnel surrounding rock identification method and system based on multi-drilling space graph modeling

Abstract

The present application relates to the field of tunnel construction intelligent perception, and specifically discloses a tunnel surrounding rock identification method and system based on multi-drilling space graph modeling. The method comprises the following steps: after a drilling rig completes all drilling construction in a drilling cycle of a tunnel face, reading a plurality of drilling parameter sequences, drilling starting point space positions, drilling angles and hole depth records, wherein the drilling parameter sequence has a corresponding relationship between drilling depth and drilling parameters; determining drilling direction unit vectors of each drilling based on the drilling starting point space positions and drilling angles; and mapping drilling parameters at different depths of each drilling to space sampling points in a local three-dimensional coordinate system of the tunnel face according to the corresponding relationship between drilling depth and drilling parameters. The technical scheme of the present application can realize full-section surrounding rock distribution depiction, improve tunnel face surrounding rock identification accuracy, and realize three-dimensional reconstruction of structure surface space distribution and rock layer strike.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing in tunnel construction, and in particular to a method and system for identifying surrounding rock in tunnels using multi-bore spatial map modeling. Background Technology

[0002] In drill-and-blast tunnel construction, the surrounding rock conditions at the tunnel face directly affect the design of drilling and blasting parameters, the selection of support parameters, advanced geological forecasting, safety risk control, and construction efficiency. Accurately identifying the grade of the surrounding rock at the tunnel face, the development of structural planes, the location of weak interlayers, and the spatial orientation of the rock strata is a key issue in intelligent tunnel construction.

[0003] Existing methods for identifying the surrounding rock at the drilling face mainly include manual geological sketching, advanced drilling, geophysical exploration, core sampling and analysis, and intelligent identification methods based on drilling rig parameters. Among these, manual geological sketching relies on engineers' field experience and suffers from low efficiency, high subjectivity, and difficulty in quantification; advanced drilling and core sampling and analysis are costly and inefficient, and can only reflect local information from a small number of boreholes; geophysical exploration methods, such as seismic waves and electrical resistivity tomography, suffer from multiple interpretations and insufficient accuracy.

[0004] With the increasing automation and digitalization of rock drilling rigs, drilling parameters such as feed pressure, impact pressure, slewing pressure, feed speed, rotational speed, and hole depth can be recorded during drilling operations, providing a data foundation for intelligent identification of surrounding rock. Since the 1980s, researchers have confirmed a strong correlation between rock mechanics characteristics and drill string response information during drilling. The development of technologies such as measurement while drilling (MWD) and logging while drilling (LWD) has led to the gradual maturation of the acquisition and application of drilling parameters.

[0005] In the drilling rig's operation log, the drilling parameters usually already have a correspondence with the drilling depth, meaning that each or every drilling parameter can be mapped to a corresponding hole depth. Therefore, when identifying the surrounding rock at the tunnel face, it is not necessary to establish an additional time-depth conversion relationship. Instead, it is necessary to combine the borehole starting position and the drill pipe drilling angle to accurately map the drilling parameters at each depth to their corresponding positions in the three-dimensional space in front of the tunnel face.

[0006] Existing methods for identifying surrounding rock based on drilling parameters typically analyze individual boreholes. They extract statistical features, depth variation characteristics, or abrupt change characteristics of the drilling parameters for that borehole, and then establish machine learning or deep learning models to identify the type or lithology of the surrounding rock within a local area of ​​that borehole. In other words, based on a large number of single-bore drilling parameter samples collected in the field, a surrounding rock prediction dataset is constructed, and a machine learning model for intelligent surrounding rock grading is established to achieve surrounding rock grading based on single-bore drilling parameters. While this type of method can reflect changes in rock drillability at the drill bit's location, it still has the following limitations:

[0007] First, single-hole identification results can only reflect the surrounding rock conditions in a local area of ​​the tunnel face, making it difficult to characterize the spatial distribution features of the entire tunnel face. Because a single borehole can only cover a local area of ​​the tunnel face, it is impossible to depict the surrounding rock distribution across the entire cross-section, making it difficult for construction personnel to grasp the changes in the surrounding rock across the entire tunnel face.

[0008] Second, structural planes, joints, fractures, weak interlayers, and rock strata interfaces at the tunnel face typically exhibit spatial orientation and continuity. Analyzing a single borehole alone cannot effectively identify the extension relationship of structural planes across multiple boreholes. Existing single-borehole methods can only identify parameter abrupt changes within a single borehole, but cannot determine whether such a change is a continuous structural plane or a local fracture zone, resulting in insufficient accuracy in structural plane identification.

[0009] Third, existing methods typically treat different boreholes as independent samples, neglecting the spatial correlation of the drilling response between adjacent boreholes. This makes it difficult for the model to determine the stratigraphic interfaces, the dip of structural planes, and their spatial distribution. For example, when a stratigraphic interface obliquely crosses the tunnel face, the depth of the stratigraphic boundary varies in different boreholes. Existing independent analysis methods cannot utilize this spatial correlation information to fit the strike and dip angle of the stratigraphic layers.

[0010] Therefore, there is an urgent need for a tunnel surrounding rock identification method based on multi-bore spatial map modeling to improve the identification accuracy of surrounding rock at the tunnel face and to achieve three-dimensional reconstruction of the spatial distribution of structural surfaces and the orientation of rock strata. Summary of the Invention

[0011] This invention provides a tunnel surrounding rock identification method based on multi-bore spatial map modeling, which can realize the full-section surrounding rock distribution characterization, improve the identification accuracy of surrounding rock at the tunnel face, and realize the three-dimensional reconstruction of the spatial distribution of structural planes and the orientation of rock strata.

[0012] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0013] A method for identifying surrounding rock in tunnels using multi-bore spatial map modeling includes the following steps:

[0014] After the drilling rig completes all drilling operations within one drilling cycle at the tunnel face, the drilling parameter sequences, the spatial location of the borehole starting point, the drilling angle, and the borehole depth records for multiple boreholes are read. The drilling parameter sequences have a correspondence between drilling depth and drilling parameters.

[0015] Based on the spatial location of the borehole starting point and the borehole drilling angle, determine the unit vector of the drilling direction for each borehole;

[0016] Based on the correspondence between drilling depth and drilling parameters, the drilling parameters at different depths of each borehole are mapped to spatial sampling points in the local three-dimensional coordinate system of the face.

[0017] Using the spatial sampling points as graph nodes, the drilling parameters at the corresponding depth and their depth window statistical features are used as node features;

[0018] A multi-hole spatial map is constructed based on the three-dimensional spatial proximity relationship between multiple spatial sampling points, the depth continuity relationship within the same borehole, the cross-hole spatial proximity relationship between different boreholes, and the similarity of the drilling response.

[0019] The multi-hole spatial map is input into a graph neural network model, and the spatial association information of neighboring nodes is aggregated through graph convolution or graph attention mechanism to obtain a spatial feature representation of the surrounding rock at the working face.

[0020] Based on the spatial characteristics of the surrounding rock at the working face, the output includes the surrounding rock type, structural plane development probability, rock stratum boundary location, and rock stratum spatial strike corresponding to the spatial sampling point or spatial grid location.

[0021] Based on the surrounding rock type, the probability of structural plane development, the location of rock strata boundaries, and the spatial orientation of rock strata, a spatial distribution map of the surrounding rock at the tunnel face and a three-dimensional geological reconstruction result are generated.

[0022] The basic principles and beneficial effects of the scheme are as follows:

[0023] In the early stages of tunnel construction, the demand for surrounding rock identification was mainly limited to single-hole lithology assessment and general surrounding rock classification. The need for full-section three-dimensional geological reconstruction and identification of the spatial distribution of structural planes was not urgent.

[0024] Graph neural networks can process irregular spatial structure data and are suitable for expressing irregular spatial adjacency relationships between multiple boreholes. By constructing a spatial graph structure from multiple completed boreholes, using borehole depth sampling points as graph nodes and drilling parameters at that depth as node features, and aggregating information from adjacent boreholes through graph convolution or graph attention mechanisms, the spatial continuity and structural surface distribution characteristics of the rock mass at the tunnel face can be effectively captured.

[0025] Existing rock identification technologies based on drilling parameters have formed an inherent mindset of independent analysis of single boreholes. Researchers have long regarded each borehole as an independent sample, focusing only on the variation of drilling parameters within a single borehole, without realizing that the spatial correlation between multiple boreholes is the key to improving the accuracy of rock identification and realizing three-dimensional geological reconstruction. At the same time, the field of engineering geology has traditionally used methods such as grid interpolation to process spatial data, and the understanding and application of graph neural networks, a non-regular data modeling method, started relatively late.

[0026] Graph neural networks were initially applied to regular or semi-regular data domains such as images, text, and social networks. Adapting them to the multi-bore data scenario of tunnel construction requires solving a series of engineering challenges, including how to accurately map one-dimensional borehole depth data to three-dimensional space, how to define edge relationships and edge weights that conform to rock mechanics properties, and how to combine engineering geological features with the feature learning of graph neural networks. There are certain technical biases involved.

[0027] The core technical principle of this invention is to transform the drilling data of multiple boreholes in a drilling cycle of a rock drilling rig from an isolated one-dimensional borehole depth sequence into a three-dimensional spatial graph structure, and to use the powerful modeling ability of graph neural networks for irregular spatial data to learn the spatial continuity characteristics of the rock mass at the tunnel face, ultimately achieving a leap from local single-hole identification to full-section three-dimensional geological interpretation.

[0028] By leveraging the engineering characteristic that a single drilling cycle of a rock drilling rig naturally covers the entire face of the tunnel, including slotting holes, auxiliary holes, and peripheral holes, data can be read uniformly after all drilling is completed, avoiding the localized defects of analyzing each hole individually. At the same time, relying on the drilling depth and drilling parameters and their corresponding relationships provided by the drilling parameters, there is no need to establish an additional conversion relationship between time and depth, ensuring the accurate binding of data with the rock mass location.

[0029] By using the borehole starting point coordinates and the unit vector of the drilling direction, the one-dimensional drilling parameters distributed along the depth in each borehole are converted into discrete sampling points in the three-dimensional space in front of the tunnel face, establishing a one-to-one mapping relationship between drilling parameters, borehole depth position, and three-dimensional spatial coordinates, providing an accurate geometric benchmark for subsequent spatial modeling.

[0030] Each three-dimensional spatial sampling point is abstracted as a node in a graph, and the original drilling parameters and depth window statistical features are integrated. This not only preserves the real-time response information of the interaction between the drill bit and the rock mass, but also captures the continuous changing trend of rock mass properties along the drilling depth direction, thus solving the problem that single sampling point parameters are easily affected by drilling noise.

[0031] By employing four edge relationships—three-dimensional spatial proximity, continuous depth within the same borehole, spatial proximity across boreholes, and similarity in response during drilling—originally isolated borehole sampling points are connected into a complete spatial network. Among these, continuous edges within the same borehole preserve the variation pattern of the rock mass along the borehole depth, while spatial edges across boreholes and similarity in response express the spatial continuity and lithological similarity of the rock mass between different boreholes, thus fully restoring the three-dimensional spatial structural characteristics of the rock mass at the working face.

[0032] By utilizing graph convolution or graph attention mechanisms, feature transfer and aggregation are performed on irregular spatial graph structures, so that the features of each node not only include its own drilling response, but also incorporate information from spatially neighboring nodes, thereby learning the overall spatial correlation features of the surrounding rock at the tunnel face, breaking through the limitation of traditional methods that can only analyze single-hole data in isolation.

[0033] Based on the learned spatial features, the surrounding rock type, structural plane development probability, rock stratum boundary location, and rock stratum spatial orientation are output simultaneously. The feature correlation between different tasks is used to achieve mutual constraints. Then, the identification results of discrete sampling points are extended to the three-dimensional space in front of the entire tunnel face through spatial interpolation and other methods to generate a continuous surrounding rock spatial distribution map and a three-dimensional geological model.

[0034] This invention breaks through the bottleneck of existing single-hole methods that can only reflect local rock mass information. By using multi-hole data covering the entire cross-section in one cycle, it integrates scattered local information into overall information through spatial map modeling. It can accurately reflect the spatial distribution of surrounding rock grades from the center to the periphery and from shallow to deep parts of the tunnel face, and solves the problem that traditional methods cannot grasp the changes in surrounding rock across the entire cross-section.

[0035] By aggregating spatial correlation information from multiple boreholes, it is possible to effectively distinguish between local random noise and actual changes in rock mass properties. For example, a sudden change in parameters in a single borehole may be caused by stuck drill bit or local rock fragments falling during drilling, while synchronous parameter changes in multiple spatially adjacent boreholes are more likely to correspond to continuous structural planes or rock strata interfaces, thereby significantly reducing the misclassification rate and improving the accuracy of surrounding rock category identification.

[0036] By capturing the continuous distribution relationship of structural planes between different boreholes through multi-bore spatial mapping, the strike, dip, and dip angle of structural planes can be accurately identified, and their three-dimensional spatial morphology in front of the tunnel face can be restored. At the same time, by fitting the rock strata boundary points of multiple boreholes, the spatial strike and undulation of the rock strata can be accurately depicted, solving the problem that existing methods cannot determine the extension direction of structural planes and the attitude of rock strata.

[0037] After a drilling cycle is completed, all data is analyzed at once. By combining the parallel computing capabilities of graph neural networks, full-section surrounding rock identification and 3D reconstruction can be completed, which is much faster than manual geological sketching and traditional hole-by-hole machine learning analysis methods.

[0038] When individual boreholes experience sensor malfunctions, missing data, or abnormal drilling conditions, the graph neural network can supplement and correct the abnormal data through spatial correlation information of neighboring boreholes. This prevents the overall recognition results from failing due to data problems in a single borehole, significantly improving the robustness and reliability of the system.

[0039] It fully leverages the value of native drilling data generated during the drilling rig construction process, eliminating the need for additional exploration work such as advanced drilling and geophysical surveys, as well as any additional hardware equipment. High-precision surrounding rock identification can be achieved solely through data processing, significantly reducing geological exploration costs.

[0040] It can identify the spatial distribution and development level of unfavorable geological bodies such as weak interlayers, fault fracture zones, and water-rich zones in front of the tunnel face in advance, providing early warning for construction personnel to adjust blasting parameters, support plans, and cyclic advance, effectively reducing construction safety risks.

[0041] The spatial map-based modeling method does not depend on specific tunnel geological conditions and cross-sectional dimensions, and can adapt to tunnel scenarios with different lithologies (granite, sandstone, limestone, etc.) and different construction conditions (hard rock drilling, soft rock drilling). The model has stronger generalization ability and engineering applicability.

[0042] The output 3D geological reconstruction results can be directly integrated with the tunnel construction BIM system to achieve digital and visual management of surrounding rock information, providing data support for subsequent construction progress optimization, support quality assessment and as-built data archiving.

[0043] In summary, this invention achieves full-section depiction of surrounding rock distribution, improves the accuracy of surrounding rock identification at the tunnel face, and realizes three-dimensional reconstruction of the spatial distribution of structural planes and the orientation of rock strata.

[0044] Furthermore, the drilling parameters include one or more of the following: feed pressure, impact pressure, slewing pressure, slewing torque, feed speed, drilling speed, impact frequency, rotational speed, water pressure, water flow rate, and hole depth; the drilling angle includes one or more of the following: drilling azimuth angle, drilling inclination angle, drill arm attitude angle, drill rod attitude angle, drilling design angle, and actual drilling angle.

[0045] Furthermore, the coordinates of the spatial sampling points are obtained as follows: a local three-dimensional coordinate system is established using the center point of the tunnel face, the tunnel axis control point, or the positioning reference point of the drilling rig; the spatial coordinates of the starting point of each borehole are used as the initial coordinates; the unit vector of the drilling direction is determined according to the borehole azimuth, borehole inclination, or drill rod attitude; the three-dimensional spatial coordinates corresponding to the drilling parameters at the corresponding drilling depth are calculated along the unit vector of the drilling direction based on the corresponding drilling depth in the drilling record; wherein, the three-dimensional spatial coordinates of the sampling point at the k-th depth of the i-th borehole are expressed as:

[0046] P i,k =P i,0 +d i,k ·u i ,

[0047] Where P i,0Let u be the spatial coordinates of the starting point of the i-th borehole. i Let d be the unit vector of the drilling direction for the i-th borehole. i,k Let k be the drilling depth corresponding to the k-th record of the i-th borehole.

[0048] Furthermore, the node features include the original drilling parameter features at the current depth, statistical features within a preset depth window, variation features along the hole depth direction, and spatial auxiliary features; the statistical features include one or more of the following: mean, standard deviation, maximum value, minimum value, skewness, kurtosis, and rate of change; the spatial auxiliary features include one or more of the following: spatial sampling point coordinates, borehole number, borehole direction, drilling depth, and relative position within the hole.

[0049] Furthermore, the edge relationships of the multi-hole spatial graph are constructed through one or more of the following methods: establishing cross-hole spatial adjacency edges based on the three-dimensional Euclidean distance between spatial sampling points; establishing continuous edges within the same borehole based on adjacent depth sampling points within the same borehole; establishing directional similar edges based on the borehole direction angle; establishing response similar edges based on the drilling responses at similar spatial locations; determining edge connection relationships based on a preset radius neighborhood, K-nearest neighbors, or an adaptive threshold; the edge relationships have edge weights, which are jointly determined based on spatial distance, borehole depth interval, borehole direction angle, and drilling response similarity. Among these, the smaller the spatial distance, the smaller the borehole depth interval, the closer the borehole directions, and the higher the drilling response similarity, the larger the corresponding edge weight.

[0050] Furthermore, the graph neural network model includes an input layer, a borehole depth feature encoding layer, a graph convolutional layer, a spatial attention layer, and a multi-task output layer. The input layer is used to input node features, edge connections, and edge weights. The borehole depth feature encoding layer is used to extract continuous variation features along the depth direction of the same borehole. The graph convolutional layer is used to aggregate the drilling response information of adjacent spatial sampling points. The spatial attention layer is used to adaptively learn the contribution weights of different neighboring nodes to the current node identification result. The multi-task output layer is used to simultaneously output the surrounding rock category, the probability of structural surface development, the location of rock strata boundaries, and the spatial trend of rock strata. The graph neural network model is a graph convolutional network, a graph attention network, GraphSAGE, a heterogeneous graph neural network, a spatial graph neural network, or a combination thereof.

[0051] Furthermore, the probability of structural surface development is determined by the consistency of abrupt changes in drilling parameters at spatially adjacent locations of multiple adjacent boreholes, spatial continuity, graph neural network output probability, and edge weights.

[0052] Furthermore, the spatial orientation of the rock strata is obtained through the following steps: extracting spatial sampling points from multiple boreholes where the location of changes in the surrounding rock type or the probability of the development of structural surfaces is higher than a preset threshold; forming a set of candidate points for structural surfaces or a set of rock strata boundaries from the spatial sampling points; performing planar fitting, surface fitting, or spatial interpolation on the set of candidate points for structural surfaces or the set of rock strata boundaries; and calculating the orientation, dip, and dip angle of the structural surfaces or rock strata interfaces based on the fitting results.

[0053] Furthermore, the three-dimensional geological reconstruction results include one or more of the following: a three-dimensional voxel model of the surrounding rock grade at the working face, a spatial distribution model of structural planes, a rock layer interface model, and a risk zoning map of the surrounding rock at the working face.

[0054] A tunnel surrounding rock identification system based on multi-bore spatial map modeling includes a data reading module, a spatial coordinate mapping module, a node feature construction module, a spatial map construction module, a graph neural network identification module, a structural surface identification module, a 3D reconstruction module, and a result display module connected in sequence.

[0055] The data reading module is used to read the drilling parameter sequence, drilling start point spatial position, drilling angle and hole depth records of multiple holes after the rock drilling rig completes all drilling construction in one drilling cycle. The data reading module is connected to one or more of the following: the data acquisition terminal of the rock drilling rig, the construction database, the drill arm attitude sensor, the drill rod attitude sensor, the hole depth sensor, the pressure sensor, the speed sensor and the construction control system.

[0056] The spatial coordinate mapping module is used to map the drilling parameters at different depths into spatial sampling points based on the spatial position of the borehole starting point, the borehole drilling angle, and the drilling depth in the drilling record.

[0057] The node feature construction module is used to use spatial sampling points as graph nodes and construct node features from the drilling parameters and their statistical characteristics at the corresponding depth.

[0058] The spatial map construction module is used to construct a multi-hole spatial map based on three-dimensional spatial proximity relationships, borehole depth continuity relationships, cross-hole spatial proximity relationships, and drilling response similarity.

[0059] The graph neural network recognition module is used to perform graph convolution or graph attention operations on the multi-hole spatial map and output the identification result of the surrounding rock at the working face.

[0060] The structural surface recognition module is used to identify the development location, continuity, and spatial orientation of structural surfaces based on the output results of the graph neural network.

[0061] The three-dimensional reconstruction module is used to generate a three-dimensional geological reconstruction model of the working face based on the surrounding rock identification results and the structural surface identification results.

[0062] The results display module is used to display the distribution map of the surrounding rock grade at the working face, the spatial distribution map of the structural planes, the strike map of the rock strata, and the three-dimensional reconstruction results. The results display module is also used to overlay the identification results of the surrounding rock at the working face with the borehole layout map of the working face, and to mark areas with strong structural plane development, weak interlayer areas, and areas with adverse geological risks. Attached Figure Description

[0063] Figure 1 This is a flowchart of the tunnel surrounding rock identification method based on multi-hole spatial map modeling of the present invention.

[0064] Figure 2 This is a schematic diagram of the construction of the multi-hole spatial diagram structure in this invention.

[0065] Figure 3 This is a schematic diagram of the three-dimensional coordinate mapping of depth sampling points in this invention.

[0066] Figure 4 This is a schematic diagram of the neural network surrounding rock identification model in this invention.

[0067] Figure 5 This is a schematic diagram of structural plane identification and rock stratum strike fitting in this invention.

[0068] Figure 6 This is a system module structure block diagram of an embodiment of the present invention. Detailed Implementation

[0069] The following detailed description illustrates the specific implementation method:

[0070] A method for identifying surrounding rock in tunnels using multi-bore spatial map modeling (e.g.) Figure 1 , Figure 2 ,picture, Figure 4 , Figure 5 The above includes the following steps:

[0071] S1: After the drilling rig completes all drilling operations within one drilling cycle at the tunnel face, read the drilling parameter sequence, the spatial position of the drilling start point, the drilling angle and the hole depth records of multiple holes. The drilling parameter sequence has a correspondence between the drilling depth and the drilling parameters.

[0072] S2: Based on the spatial position of the borehole starting point and the borehole drilling angle, determine the unit vector of the drilling direction for each borehole;

[0073] S3: Based on the correspondence between drilling depth and drilling parameters, the drilling parameters at different depths of each borehole are mapped to spatial sampling points in the local three-dimensional coordinate system of the face.

[0074] S4: Using the spatial sampling points as graph nodes, the drilling parameters at the corresponding depth and their depth window statistical features are used as node features;

[0075] S5: Construct a multi-hole spatial map based on the three-dimensional spatial proximity relationship between multiple spatial sampling points, the depth continuity relationship within the same borehole, the cross-hole spatial proximity relationship between different boreholes, and the similarity of the drilling response.

[0076] S6: Input the multi-hole spatial map into a graph neural network model, and aggregate the spatial association information of neighboring nodes through graph convolution or graph attention mechanism to obtain a spatial feature representation of the surrounding rock at the working face;

[0077] S7: Based on the spatial characteristics of the surrounding rock at the working face, output the surrounding rock type, structural plane development probability, rock stratum boundary location, and rock stratum spatial orientation corresponding to the spatial sampling point or spatial grid location;

[0078] S8: Based on the surrounding rock type, structural plane development probability, rock stratum boundary location, and rock stratum spatial orientation, generate a spatial distribution map of the surrounding rock at the tunnel face and a three-dimensional geological reconstruction result.

[0079] In practical use: In S1, during the tunnel drilling and blasting method, the drilling rig drills holes at the working face according to a pre-set hole layout plan. A drilling cycle typically includes multiple holes such as slotting holes, auxiliary holes, peripheral holes, and bottom plate holes. Each hole has a defined opening position, drilling angle, and designed hole depth at the working face.

[0080] After all boreholes in the drilling cycle are completed, the drilling parameter sequence for each borehole is read from the drilling rig's construction database or data acquisition terminal. This drilling parameter sequence is recorded according to hole depth or sampling segment, with each record or segment corresponding to a drilling depth, thus directly determining the position of the drilling parameters within the hole. In this embodiment, the drilling parameters include feed pressure, impact pressure, slewing pressure, slewing torque, feed rate, drilling speed, impact frequency, rotational speed, water pressure, and water flow rate. These parameters reflect the drillability characteristics of the rock mass; different lithologies and rock masses with varying integrity will exhibit different parameter responses during drilling.

[0081] Simultaneously, the spatial geometric information of each borehole is read, including borehole number, borehole starting point coordinates, borehole azimuth, borehole inclination, drill rod attitude, designed hole depth, and actual hole depth. The spatial location of the borehole starting point can be obtained using a rock drilling rig positioning system, drill arm attitude sensor, laser scanner, total station, or face drilling design drawing, ensuring accurate acquisition of the initial position of each borehole on the face.

[0082] In S2-S4, a local three-dimensional coordinate system is established at the tunnel face, with the tunnel axis as one coordinate axis and the horizontal and vertical directions of the working face as the other two coordinate axes. In this coordinate system, the starting point of each borehole is represented as P. i,0 =(x i,0 , y i,0 , z i,0 ), where P i,0 This represents the spatial coordinates of the starting point of the i-th borehole.

[0083] Determine the unit vector u of the drilling direction for the i-th borehole based on the borehole azimuth, borehole inclination, or drill rod attitude. i For the k-th data entry in the drilling log, its corresponding drilling depth is d. i,k The spatial coordinates of the location where this drilling data occurred are represented as P. i,k =P i,0 +d i,k ·u i .

[0084] Using the above method, the drilling data at different depths in each borehole can be mapped to the three-dimensional space in front of the tunnel face, thus forming a multi-bore spatial sampling point set.

[0085] Each borehole depth sampling point is used as a graph node. Each node corresponds to a specific three-dimensional spatial coordinate and a set of drilling parameters.

[0086] The node features include the original drilling parameters at the current depth, the statistical features of the drilling parameters within the preset depth window, the rate of change along the hole depth direction, the abrupt change features of the drilling parameters, the mean, standard deviation, peak value and energy features under the multi-scale depth window, as well as spatial auxiliary features such as the coordinates of the spatial sampling point, the drilling direction, the borehole number and the relative depth in the hole.

[0087] The above node features can express both the local rock machine response at that depth and the continuous change of surrounding rock properties along the borehole depth direction.

[0088] In S5, since the boreholes at different locations on the working face are not distributed in a regular grid but are arranged at different positions according to the blasting design, this invention uses a graph structure to represent the spatial relationship between multiple borehole depth sampling points. In the graph structure, nodes represent spatial sampling points, and edges represent the spatial relationships between sampling points.

[0089] Specifically, edge relationships can be constructed based on spatial distance, K-nearest neighbor, contiguous hole, drilling direction, and similarity of drilling response. If the three-dimensional distance between two spatial sampling points is less than a preset threshold, a cross-hole spatial adjacency edge is established between the two nodes; for each node, the K nearest nodes in spatial distance can also be selected as neighbor nodes and edge connections can be established.

[0090] For adjacent depth sampling points within the same borehole, continuous edges are established within the borehole to preserve the continuous variation information of the drilling response along the borehole depth direction. If the included angle between the directions of two boreholes is small, it indicates that the rock mass regions traversed by the two boreholes have strong spatial comparability, and the weight of the corresponding edges can be increased. If similar drilling responses appear at spatially adjacent locations of different boreholes, it indicates that they may traverse the same structural plane or the same rock stratum, and similarity edges can be established or enhanced.

[0091] In S6, the spatial map of multiple boreholes is input into a graph neural network model. The graph neural network model includes an input layer, a borehole depth feature encoding layer, a graph convolutional layer, a spatial attention layer, and a multi-task output layer. The input layer receives node feature matrices, edge connection matrices, and edge weight matrices.

[0092] The borehole depth feature encoding layer is used to extract continuous variation features along the borehole depth direction within the same borehole. The graph convolutional layer aggregates information from neighboring spatial sampling points, enabling the identification of a node to rely not only on its own drilling parameters but also on the drilling responses of adjacent boreholes and nearby depth locations. The spatial attention layer adaptively calculates the influence weights of different neighboring nodes on the current node.

[0093] The multi-task output layer is used to simultaneously output the surrounding rock type, the probability of structural plane development, the location of rock strata boundaries, and the spatial orientation parameters of rock strata or structural planes.

[0094] In S7, the graph neural network model outputs the surrounding rock category corresponding to each spatial sampling point. The surrounding rock category can be set to Class I, Class II, Class III, Class IV, and Class V surrounding rock according to engineering needs, or it can be set to categories such as intact rock mass, relatively intact rock mass, fractured rock mass, weak interlayer, and fault fracture zone.

[0095] By mapping the identification results of multiple spatial sampling points back to the three-dimensional coordinate system of the working face and performing spatial interpolation on the unsampled areas, a spatial distribution map of the surrounding rock grade of the working face can be generated.

[0096] In S8, structural surfaces often cause abrupt changes in drilling parameters at spatially adjacent locations among multiple adjacent boreholes, such as sudden changes in advance rate, abnormal slewing pressure, changes in impact pressure, or changes in drilling energy. This invention uses a graph neural network to simultaneously analyze the drilling response of multiple boreholes. If multiple spatially adjacent nodes exhibit similar abrupt change characteristics, the model can determine the presence of a structural surface or stratum interface in the region.

[0097] Specifically, the probability of structural surface development can be determined by comprehensively considering the degree of abrupt change in the drilling parameters of a single borehole, the spatial continuity of abrupt change locations between adjacent boreholes, the geometric relationship between abrupt change points among multiple boreholes, the structural surface probability output by the graph neural network, and the edge weights and attention weights between nodes. When the probability of structural surface development is higher than a preset threshold, the corresponding node is marked as a candidate point for structural surface development.

[0098] Multiple candidate structural plane points or strata boundary points are mapped to the three-dimensional coordinate system of the tunnel face to form a candidate point set. Spatial fitting is then performed on the candidate point set. If the structural plane is approximately planar, a planar fitting method is used to obtain the structural plane equation; if the structural plane is curved, a surface fitting or spatial interpolation method is used to obtain the curved surface model of the structural plane.

[0099] Based on the fitted plane or curved surface, the strike, dip, and dip angle of the structural plane or rock stratum interface are calculated. Therefore, this invention can not only determine whether a borehole crosses a structural plane, but also further determine the spatial distribution direction of the structural plane throughout the entire working face.

[0100] Based on the surrounding rock category identification results, candidate structural planes, stratum boundaries, and spatial orientation of strata, a three-dimensional geological reconstruction is performed. The three-dimensional reconstruction can be achieved using the following methods: interpolating the node identification results into the three-dimensional mesh in front of the tunnel face to form a voxel model of the surrounding rock grade; performing planar or curved surface fitting on the candidate structural planes to form a three-dimensional structural plane model; performing spatial interpolation on the stratum boundaries to form a stratum interface model; and overlaying the surrounding rock grade model, structural plane model, and stratum interface model to generate the three-dimensional geological interpretation results of the tunnel face.

[0101] The three-dimensional reconstruction results can be used to guide the optimization of blasting parameters, the adjustment of support parameters, and the identification of adverse geological risks.

[0102] Based on the above method, the present invention also provides a tunnel surrounding rock identification system for multi-bore spatial map modeling (see [link]). Figure 6 The system is used to uniformly read, spatially map, graphically model, identify surrounding rock, and reconstruct three-dimensional records of drilling parameters, initial borehole position, drilling angle, and borehole depth for multiple boreholes after all boreholes have been drilled in a single drilling cycle.

[0103] The system includes a data reading module, a spatial coordinate mapping module, a node feature construction module, a spatial graph construction module, a graph neural network recognition module, a structural surface recognition module, a 3D reconstruction module, and a result display module. These modules are connected sequentially according to the data processing flow, forming a complete processing chain from the original drilling records to the spatial recognition results at the working face.

[0104] The data reading module is used to read the multi-hole drilling parameter sequence and its corresponding hole depth record after the construction is completed; the spatial coordinate mapping module is used to calculate the three-dimensional spatial coordinates corresponding to each drilling parameter based on the spatial position of the borehole starting point, the drilling angle and the depth value in the drilling record; the node feature construction module is used to convert the spatial sampling points and their drilling parameters into graph nodes and node features.

[0105] The spatial graph construction module is used to establish a multi-hole spatial graph based on the distance between spatial sampling points, the continuity relationship of the same hole, the proximity relationship across holes, the drilling direction relationship, and the similarity of the drilling response; the graph neural network recognition module is used to perform graph convolution or graph attention calculation on the multi-hole spatial graph and output the surrounding rock category and spatial feature expression.

[0106] The structural surface identification module is used to identify the development location of structural surfaces, the location of rock layer boundaries, and the spatial orientation of rock layers based on the identification results of multiple spatial sampling points, abrupt change characteristics, and spatial continuity. The three-dimensional reconstruction module is used to generate a three-dimensional distribution model of the surrounding rock at the working face, a spatial model of structural surfaces, and a model of rock layer boundaries. The result display module is used to output the identification map of the surrounding rock at the working face, the distribution map of structural surfaces, the orientation map of rock layers, and the three-dimensional reconstruction results.

[0107] The system described above enables fully automated processing from raw drilling parameters to 3D geological reconstruction results, requiring no manual intervention and taking only a few minutes to process, thus providing rapid geological information support for construction.

[0108] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for identifying surrounding rock in tunnels using multi-bore spatial map modeling, characterized in that, Includes the following steps: After the drilling rig completes all drilling operations within one drilling cycle at the tunnel face, the drilling parameter sequences, the spatial location of the borehole starting point, the drilling angle, and the borehole depth records for multiple boreholes are read. The drilling parameter sequences have a correspondence between drilling depth and drilling parameters. Based on the spatial location of the borehole starting point and the borehole drilling angle, determine the unit vector of the drilling direction for each borehole; Based on the correspondence between drilling depth and drilling parameters, the drilling parameters at different depths of each borehole are mapped to spatial sampling points in the local three-dimensional coordinate system of the face. Using the spatial sampling points as graph nodes, the drilling parameters at the corresponding depth and their depth window statistical features are used as node features; A multi-hole spatial map is constructed based on the three-dimensional spatial proximity relationship between multiple spatial sampling points, the depth continuity relationship within the same borehole, the cross-hole spatial proximity relationship between different boreholes, and the similarity of the drilling response. The multi-hole spatial map is input into a graph neural network model, and the spatial association information of neighboring nodes is aggregated through graph convolution or graph attention mechanism to obtain a spatial feature representation of the surrounding rock at the working face. Based on the spatial characteristics of the surrounding rock at the working face, the output includes the surrounding rock type, structural plane development probability, rock stratum boundary location, and rock stratum spatial strike corresponding to the spatial sampling point or spatial grid location. Based on the surrounding rock type, the probability of structural plane development, the location of rock strata boundaries, and the spatial orientation of rock strata, a spatial distribution map of the surrounding rock at the tunnel face and a three-dimensional geological reconstruction result are generated.

2. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 1, characterized in that, The drilling parameters include one or more of the following: feed pressure, impact pressure, slewing pressure, slewing torque, feed speed, drilling speed, impact frequency, rotational speed, water pressure, water flow rate, and hole depth; the drilling angles include one or more of the following: drilling azimuth angle, drilling inclination angle, drill arm attitude angle, drill rod attitude angle, drilling design angle, and actual drilling angle.

3. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 2, characterized in that, The coordinates of the spatial sampling points are obtained as follows: a local three-dimensional coordinate system is established using the center point of the tunnel face, the tunnel axis control point, or the positioning reference point of the drilling rig; the spatial coordinates of the starting point of each borehole are used as the initial coordinates; the unit vector of the drilling direction is determined according to the borehole azimuth, borehole inclination, or drill rod attitude; the three-dimensional spatial coordinates corresponding to the drilling parameters at the corresponding drilling depth are calculated along the unit vector of the drilling direction based on the corresponding drilling depth in the drilling record; wherein, the three-dimensional spatial coordinates of the sampling point at the k-th depth of the i-th borehole are expressed as: P i,k =P i,0 +d i,k ·u i , Where P i,0 Let u be the spatial coordinates of the starting point of the i-th borehole. i Let d be the unit vector of the drilling direction for the i-th borehole. i,k Let k be the drilling depth corresponding to the k-th record of the i-th borehole.

4. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 3, characterized in that, The node features include the original drilling parameters at the current depth, statistical features within a preset depth window, variation features along the hole depth direction, and spatial auxiliary features; the statistical features include one or more of the following: mean, standard deviation, maximum value, minimum value, skewness, kurtosis, and rate of change; the spatial auxiliary features include one or more of the following: spatial sampling point coordinates, borehole number, borehole direction, drilling depth, and relative position within the hole.

5. The tunnel surrounding rock identification method based on multi-bore spatial map modeling according to claim 4, characterized in that, The edge relationships of the multi-hole spatial graph are constructed through one or more of the following methods: establishing cross-hole spatial adjacency edges based on the three-dimensional Euclidean distance between spatial sampling points; establishing continuous edges within the same borehole based on adjacent depth sampling points within the same borehole; establishing directional similar edges based on the borehole direction angle; establishing response similar edges based on the drilling responses at similar spatial locations; determining edge connection relationships based on a preset radius neighborhood, K-nearest neighbors, or an adaptive threshold; the edge relationships have edge weights, which are jointly determined based on spatial distance, borehole depth interval, borehole direction angle, and drilling response similarity. Among these, the smaller the spatial distance, the smaller the borehole depth interval, the closer the borehole directions, and the higher the drilling response similarity, the larger the corresponding edge weight.

6. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 5, characterized in that, The graph neural network model includes an input layer, a borehole depth feature encoding layer, a graph convolutional layer, a spatial attention layer, and a multi-task output layer. The input layer is used to input node features, edge connections, and edge weights. The borehole depth feature encoding layer is used to extract continuous variation features along the depth direction of the same borehole. The graph convolutional layer is used to aggregate the drilling response information of adjacent spatial sampling points. The spatial attention layer is used to adaptively learn the contribution weights of different neighboring nodes to the identification result of the current node. The multi-task output layer is used to simultaneously output the surrounding rock category, the probability of structural surface development, the location of rock strata boundaries, and the spatial trend of rock strata. The graph neural network model is a graph convolutional network, a graph attention network, GraphSAGE, a heterogeneous graph neural network, a spatial graph neural network, or a combination thereof.

7. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 6, characterized in that, The probability of structural surface development is determined by the consistency of abrupt changes in drilling parameters at spatially adjacent locations of multiple adjacent boreholes, spatial continuity, graph neural network output probability, and edge weights.

8. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 7, characterized in that, The spatial orientation of the rock strata is obtained through the following steps: extracting spatial sampling points from multiple boreholes where the location of changes in surrounding rock type or the probability of development of structural surfaces is higher than a preset threshold; forming a set of candidate points for structural surfaces or a set of rock strata boundaries from the spatial sampling points; performing planar fitting, surface fitting, or spatial interpolation on the set of candidate points for structural surfaces or the set of rock strata boundaries; and calculating the orientation, dip, and dip angle of the structural surfaces or rock strata interfaces based on the fitting results.

9. The method for identifying surrounding rock in tunnels using multi-bore spatial map modeling according to claim 8, characterized in that, The three-dimensional geological reconstruction results include one or more of the following: a three-dimensional voxel model of the surrounding rock grade at the working face, a spatial distribution model of structural planes, a rock layer interface model, and a risk zoning map of the surrounding rock at the working face.

10. A tunnel surrounding rock identification system based on multi-bore spatial map modeling, characterized in that, It includes a data reading module, a spatial coordinate mapping module, a node feature construction module, a spatial graph construction module, a graph neural network recognition module, a structural surface recognition module, a 3D reconstruction module, and a result display module, which are connected in sequence. The data reading module is used to read the drilling parameter sequence, drilling start point spatial position, drilling angle and hole depth records of multiple holes after the rock drilling rig completes all drilling construction in one drilling cycle. The data reading module is connected to one or more of the following: the data acquisition terminal of the rock drilling rig, the construction database, the drill arm attitude sensor, the drill rod attitude sensor, the hole depth sensor, the pressure sensor, the speed sensor and the construction control system. The spatial coordinate mapping module is used to map the drilling parameters at different depths into spatial sampling points based on the spatial position of the borehole starting point, the borehole drilling angle, and the drilling depth in the drilling record. The node feature construction module is used to use spatial sampling points as graph nodes and construct node features from the drilling parameters and their statistical characteristics at the corresponding depth. The spatial map construction module is used to construct a multi-hole spatial map based on three-dimensional spatial proximity relationships, borehole depth continuity relationships, cross-hole spatial proximity relationships, and drilling response similarity. The graph neural network recognition module is used to perform graph convolution or graph attention operations on the multi-hole spatial map and output the identification result of the surrounding rock at the working face. The structural surface recognition module is used to identify the development location, continuity, and spatial orientation of structural surfaces based on the output results of the graph neural network. The three-dimensional reconstruction module is used to generate a three-dimensional geological reconstruction model of the working face based on the surrounding rock identification results and the structural surface identification results. The results display module is used to display the distribution map of the surrounding rock grade at the working face, the spatial distribution map of the structural planes, the strike map of the rock strata, and the three-dimensional reconstruction results. The results display module is also used to overlay the identification results of the surrounding rock at the working face with the borehole layout map of the working face, and to mark areas with strong structural plane development, weak interlayer areas, and areas with adverse geological risks.

Images