Construction period tunnel broken surrounding rock three-dimensional refined modeling method based on potential field interpolation

Abstract

The invention discloses a potential field interpolation-based three-dimensional refined modeling method for tunnel broken surrounding rock in a construction period. Comprising the following steps: automatically interpreting a fracture zone extension length of a geological radar image in front of a construction surface, and constructing an initial potential field boundary; automatically interpreting a continuous mileage section construction surface image set, and constructing a crack binary image set; carrying out skeleton line extraction on the binary image set one by one, discretizing to extract crack pixel points, and constructing a crack pixel point set; carrying out space coordinate conversion on the pixel point set to obtain a three-dimensional fracture discrete point set, and carrying out classification according to whether the three-dimensional fracture discrete point set belongs to the same fracture; constructing a directed bounding box of the discrete point set in combination with the size of the construction surface, and combining the directed bounding box with the initial potential field boundary to construct a complete potential field boundary; a fitting plane of each three-dimensional fracture point set is constructed by a least square method, a normal vector of each plane is calculated and processed into a unit polarization vector, and a two-tuple set of point coordinates and the polarization unit vector is established; inputting a two-tuple set and a potential field range, constructing a surrounding rock fracture three-dimensional refined model by using a potential field interpolation method, and outputting the model as a triangular mesh; and according to the mileage coordinates, adopting Boolean difference operation to remove triangular meshes in front of the construction surface, and outputting a surrounding rock fracture three-dimensional refined model in front of the construction surface.

Description

Technical Field

[0001] The present invention relates to the technical field of geographic space information systems, and in particular to a three-dimensional refined modeling method for broken surrounding rocks of a tunnel during construction based on potential field interpolation. Background Art

[0002] 3D geological modeling is a key technology that uses computer technology to intuitively represent the geometric mapping and topological relationships of geological bodies through the integration and assimilation of multi-source, heterogeneous data in a virtual 3D environment. 3D geological modeling at the tunnel construction surface scale provides a detailed representation of the tunnel's internal environment during construction and effectively predicts the structure and properties of undesirable geological bodies in the unexcavated section ahead of the construction surface, providing a crucial reference for dynamic optimization design and ensuring construction safety.

[0003] Due to the disturbances caused by tunnel construction, traversing fractured zones is prone to disasters such as sudden water and mud bursts, landslides, and roof collapses. Existing methods often rely on advanced geological prediction data (penetrating radar) to achieve two-dimensional exploration of the fractured zone ahead of the construction face. The penetrating radar method uses a radar antenna to transmit electromagnetic waves to the target. These waves are reflected by the electrical interface along the propagation path to the receiving antenna, where a computer-generated profile image is generated. Target attributes are interpreted based on features such as the image waveform frequency distribution, event continuity, and waveform amplitude. This method accurately reflects changes in the surrounding rock properties ahead of the construction face and identifies the fractured zone in tunnel engineering. However, due to the spatial limitations of two-dimensional exploration, penetrating radar data cannot express the horizontal and vertical spatial distribution characteristics and geometric structure of the fractured zone's surrounding rock. Therefore, achieving detailed three-dimensional modeling of the fractured structure ahead of the construction face, given the incomplete spatial attributes of penetrating radar data, remains an urgent challenge. Summary of the Invention

[0004] To address the problem of incomplete spatial attributes of geological radar data, making it difficult to accurately express the three-dimensional structural information of surrounding rock fractures ahead of the construction face, the present invention provides a method for fine-grained three-dimensional modeling of fractured surrounding rock during the construction period based on potential field interpolation. Taking into account the fact that the development pattern of fractures in the surrounding rock ahead of the construction face to be excavated conforms to geostatistics, the potential field boundary is determined by interpreting the extension length of the fracture zone ahead of the current construction face through geological radar image interpretation. Discrete fracture feature points and their corresponding polarization unit vectors are obtained from several consecutive mileage excavation face images preceding the current excavation face as input. The potential field interpolation method is then used to achieve fine-grained three-dimensional modeling of the surrounding rock fracture structure ahead of the current excavation face. This method effectively addresses the problem of difficult fine-grained three-dimensional modeling of the fracture structure ahead of the excavation face due to incomplete spatial attributes of geological radar data.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for three-dimensional refined modeling of broken surrounding rock in a tunnel during construction based on potential field interpolation, comprising the following steps:

[0006] Step 1: Automatically interpret the extension length of the broken zone in the geological radar image in front of the current construction surface, and determine the potential field boundary based on the current construction surface size;

[0007] Step 2: Automatically interpret and process the crack pixel points distributed in the current excavation surface and the previous 4-5 consecutive mileage construction surface image datasets, and output the excavation surface crack binary image dataset;

[0008] Step 3: Use a thinning algorithm to extract skeleton line pixels from the binary image dataset generated in step 2, and perform discrete sampling on the extracted crack feature pixels. Then, classify the pixel coordinates belonging to the same crack to obtain multiple sets of crack discrete feature pixel points.

[0009] Step 4: Obtain the mileage coordinates corresponding to the current excavation surface, perform spatial coordinate conversion on each set of pixel points generated in step 3 based on the construction surface size, and construct a three-dimensional crack feature point set that can reflect the coordinate information of the actual mileage segment;

[0010] Step 5: construct a plane by fitting the crack structure for each set of three-dimensional crack feature points generated in step 4, calculate the unit normal vector of each plane, process and output the polarization unit vector corresponding to each three-dimensional crack feature point, and output a two-tuple set of three-dimensional feature points and their polarization unit vectors;

[0011] Step 6: Input the two-tuple set of three-dimensional crack feature points and their corresponding polarization unit vectors constructed in step 5, use the potential field interpolation method to calculate the three-dimensional model of each crack structural surface, and finally output it as a triangular mesh. Then, based on the current excavation surface, the triangular mesh model of the excavated section is intercepted and eliminated to construct a detailed three-dimensional model of the surrounding rock cracks in front of the current construction surface.

[0012] Preferably, in step 1, the steps include:

[0013] Step 1.1: Install the open-source image annotation tool Labelme in the local Conda environment, access the geological radar image datasets of each mileage section, and manually annotate the dataset based on the significant image features of the broken zone, such as uneven waveform frequency distribution, poor regularity of waveform frequency changes, discontinuous phase axis, and cluttered waveforms. Use the rectangular annotation tool to select the broken zone and export the annotation information to a JSON file.

[0014] Step 1.2: Write a Python script to batch-process the json file output in step 1.1 that records the location of the fragmentation zones in each geological radar image, and output a set of binary image annotation files in png format;

[0015] Step 1.3: Build the training set and test set according to the VOC (Visual Objects Classes) dataset format. Create a VOC root directory folder on the computer disk. Create subdirectories of JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the geological radar image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set.

[0016] Step 1.4: Call the open source deep learning framework PyTorch in Python and use the object detection model YOLO-V5 to achieve automatic and precise positioning of the broken belt. Input the VOC format dataset built in step 1.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training.

[0017] Step 1.5: After the model training in step 1.4 is completed, the geological radar image corresponding to the current excavation surface is input into the model to automatically locate the broken zone. Then, the broken zone interpretation result is converted into spatial coordinates based on the current mileage coordinates to obtain the true extension length of the broken zone in front of the construction surface and its corresponding coordinates. An oriented bounding box is constructed based on the current construction surface size, and the coordinates of the vertices of the oriented bounding box are calculated and used as the potential field boundary.

[0018] Preferably, in step 2, the steps specifically include:

[0019] Step 2.1: Access the excavation surface image datasets for each mileage segment and write a script in Python using the open source computer vision library OpenCV. Using the methods encapsulated in the open source library, perform geometric transformations, random noise processing, and random light processing on the image datasets to enhance the existing datasets and effectively expand the construction surface image datasets.

[0020] Step 2.2: Install the open-source image annotation tool Labelme in the local Conda environment, access the construction surface image set after data enhancement in step 2.1, use the point annotation tool to select the crack pixels and output them as a json file;

[0021] Step 2.3: Write a Python script to batch-process the json file output in step 2.2 that records the pixel locations of cracks in each construction surface image, and output a set of binary image annotation files in png format;

[0022] Step 2.4: Build the training set and test set according to the VOC (Visual Objects Classes) dataset format. Create a VOC root directory folder on the computer disk. Create subdirectories of JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the construction surface image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set.

[0023] Step 2.5: Call the open-source deep learning framework PyTorch in Python and use the semantic segmentation model U-Net to automatically and accurately extract each fissure pixel. Input the VOC format dataset constructed in step 2.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training.

[0024] Step 2.6: Manually annotate the current construction surface and the previous 4-5 consecutive mileage excavation surface images, obtain the boundary contours corresponding to each excavation surface, and crop along the excavation surface contours. Then, call the open source computer vision library OpenCV in Python to resample the cropped images to obtain an excavation surface image dataset of the same size.

[0025] Step 2.7: After the model training in step 2.5 is completed, the excavation surface image dataset constructed in step 2.6 is input into the model, crack pixels are automatically identified and output as a mask file, the mask is processed, and a binary image dataset of the construction surface cracks is output.

[0026] Preferably, in step 3, the steps specifically include:

[0027] Step 3.1: Call the open source computer vision library OpenCV in Python and use the thinning algorithm encapsulated in the open source library to skeletonize the crack lines distributed in each binary crack image, and output a binary crack skeleton line image dataset;

[0028] Step 3.2: In Python, call the open-source scientific computing library Numpy and store the coordinates of the crack pixels in the binary crack skeleton image dataset output in Step 3.1 into Numpy arrays one by one. Set a fixed sampling rate, perform discrete sampling on each Numpy array, remove the coordinates of unsampled pixels, and output a Numpy array set storing the two-dimensional coordinates of the discrete crack pixels.

[0029] Step 3.3: Generate an array set of two-dimensional coordinates of discrete crack pixel points based on step 3.2. Based on the binary crack skeleton line image dataset output in step 3.1, establish an oriented bounding box for each crack skeleton line image by image, record the discrete pixel points within the bounding box and store them in the order of the structural surface to obtain multiple sets of crack discrete feature pixel points.

[0030] Preferably, in step 4, the steps include:

[0031] Obtain the mileage coordinates corresponding to the current construction surface input in step 2.4 and the excavated construction surfaces of several consecutive mileage segments before it. Establish a conversion equation between pixel coordinates and mileage coordinates. Convert the discrete pixel points of each crack output in step 3.3 into 3D discrete points. Construct an oriented bounding box (Oriented Bounding Box) for the 3D discrete point set based on the construction surface size. Combine this with the potential field boundary constructed in step 1.5 to express the spatial extent of the fracture zone in front of the construction surface to construct a complete potential field boundary. Write a Python script to organize the 3D discrete points belonging to the same crack in the order of their crack ID and spatial coordinates (x, y, z). This results in a 3D crack feature point set reflecting the actual mileage segment coordinate information and outputs it as a txt file.

[0032] Preferably, in step 5, the steps specifically include:

[0033] Step 5.1: Create a custom point class in C++ that contains the fracture ID and spatial coordinates (x, y, z). Call the standard library iostream input file stream to input the txt file constructed in step 4. Create a point object set based on the fracture and coordinate information of each point recorded in the txt file.

[0034] Step 5.2: In C++, call the open source linear calculation library Eigen, use the least squares method to perform plane fitting on the point object set of the same fault line output in step 5.1 and solve the plane equation. According to the plane equation, further solve the plane normal vector, normalize the normal vector to obtain the polarization unit vector (G x , G y , G z ), output the coordinates of each point (x, y, z) and its corresponding polarization unit vector (G x , G y , G z )’s set of two-tuples;

[0035] Step 5.3: In C++, call the standard library iostream output file stream to output the tuple set constructed in step 5.2 and export it as a csv file.

[0036] Preferably, in step 6, the steps specifically include:

[0037] Step 6.1: Call the open source geological modeling library Gempy in Python, input the constructed 3D fracture feature points and the constructed point set binary CSV file as initial data, set the potential field range and call the interpolation function to calculate and solve the 3D refined model of the surrounding rock fracture within the complete potential field range;

[0038] Step 6.2: In Python, call the visualization program ParaView, which is packaged in the open source geological modeling library Gempy, to visualize the 3D refined model of surrounding rock fractures generated in Step 6.1. Then, call the open source visualization tool library VTK (Visualization Toolkit) to output the 3D refined model of surrounding rock fractures as a triangular mesh model.

[0039] Step 6.3: Call the open source geometric modeling library OpenCascade in C++, input the surrounding rock fracture triangular mesh model constructed in step 6.2, and construct a geometric body by obtaining the oriented bounding box (Oriented Bounding Box) of the three-dimensional discrete point set of the excavated construction surface according to step 2.4. Perform a Boolean difference operation between the geometric body and the surrounding rock fracture triangular mesh model to eliminate the triangular mesh model of the excavated section, and obtain a refined three-dimensional model of the surrounding rock fracture in front of the current construction surface.

[0040] The beneficial effects of the present invention are as follows: the present invention is a method for solving technical bottlenecks such as the incomplete spatial attributes of geological radar data at the construction surface scale, which makes it impossible to establish a three-dimensional refined model of the surrounding rock cracks in front of the construction surface to achieve effective prediction of the horizontal and vertical spatial distribution characteristics and geometric structure information of the surrounding rock crack structure of the surrounding rock crack fracture zone in front of the current construction surface, and the inability to provide important references for the dynamic optimization design and construction safety of tunnel construction. The method provides a three-dimensional refined modeling method of the broken surrounding rock of the tunnel during the construction period based on potential field interpolation. According to the characteristic that the development law of the cracks on the construction surface conforms to geostatistics, the potential field is established by interpreting the crushing zone range with geological radar, and the potential field interpolation is performed on discrete crack points to construct a three-dimensional refined model of the surrounding rock cracks in front of the construction surface, effectively solving the problem that it is difficult to achieve fine modeling of the three-dimensional structure of the cracks in front of the excavation surface under the condition of incomplete spatial attributes of geological radar data. First, the broken zone in front of the current excavation face is automatically located using an open-source target detection network. Based on the conversion relationship between mileage coordinates and geological radar image coordinates, the mileage coordinates of the current construction face and the extension of the broken zone in front of the construction face are calculated. An oriented bounding box (OB) and its vertex coordinates are calculated based on the current construction face dimensions and output as the potential field boundary. Images of the current construction face and several consecutive mileage segments preceding it are obtained. These images are resampled and geometrically cropped to a uniform size. Based on an open-source semantic segmentation network, crack pixels distributed within the construction face image set are automatically extracted and output as a mask. The mask is then converted to a binary image for output. A thinning algorithm is used to extract skeleton lines from the output binary image. Discrete sampling is used to extract discrete crack pixels from each image, and pixels are classified according to whether they belong to the same crack. Based on the actual mileage coordinates and the actual dimensions of the construction surface, the spatial coordinates of the fracture pixels were transformed to obtain the 3D discrete feature points of the fracture. An oriented bounding box (ORB) was calculated and merged with the calculated potential field boundary in front of the construction surface to obtain the complete potential field range. The unit polarization vector of each 3D discrete fracture point was calculated. The open source geological modeling library Gempy was called in Python, and parameters such as the 3D discrete fracture point set, mesh resolution, and potential field range were input. A 3D refined model of the surrounding rock fractures was constructed based on the potential field interpolation method and output as a triangular mesh. A geometric body was constructed based on the current construction surface mileage coordinates. A Boolean difference operation was performed between the geometric body and the output triangular mesh to construct a 3D refined model of the surrounding rock fractures in front of the construction surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the overall steps of the method of the present invention;

[0042] Figure 2 A flow chart of the process of constructing the potential field boundary for the present invention;

[0043] Figure 3 A flowchart of the process of constructing a discrete pixel point set of cracks on the construction surface of a continuous mileage section according to the present invention;

[0044] Figure 4 This is a flow chart of the process of constructing a fine three-dimensional model of surrounding rock fissures in front of the construction face in the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] The embodiment of the present invention provides a three-dimensional refined modeling method for the broken surrounding rock of a tunnel during construction based on potential field interpolation. The three-dimensional structure of the surrounding rock cracks in front of the construction face is constructed by the potential field interpolation method, so as to effectively predict the spatial distribution and geometric structure information of the broken zone in front of the construction face under the condition of incomplete spatial attributes of geological radar. Figure 1 As shown, the following steps are included:

[0047] Step 1: Automatically interpret the extension length of the broken zone in the geological radar image in front of the current construction surface, and determine the potential field boundary based on the current construction surface size. Figure 2 The detailed sub-steps are as follows:

[0048] Step 1.1: Install the open-source image annotation tool Labelme in the local Conda environment, access the geological radar image datasets of each mileage section, and manually annotate the dataset based on the significant image features of the broken zone, such as uneven waveform frequency distribution, poor regularity of waveform frequency changes, discontinuous phase axis, cluttered waveforms, and strong waveform amplitudes. Use the rectangular annotation tool to select the broken zone and export the annotation information to a JSON file.

[0049] Step 1.2: Write a Python script to batch-process the json file output in step 1.1 that records the location of the fragmentation zones in each geological radar image, and output a set of binary image annotation files in png format;

[0050] Step 1.3: Build the training set and test set according to the VOC (Visual Objects Classes) dataset format. Create a VOC root directory folder on the computer disk. Create subdirectories such as JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the geological radar image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set;

[0051] Step 1.4: Call the open source deep learning framework PyTorch in Python and use the object detection model YOLO-V5 to achieve automatic and accurate positioning of the broken zone. Input the VOC format dataset built in step 1.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training.

[0052] Step 1.5: After the model training in step 1.4 is completed, the geological radar image corresponding to the current excavation surface is input into the model to automatically locate the broken zone. Then, the broken zone interpretation result is converted into spatial coordinates based on the current mileage coordinates to obtain the true extension length of the broken zone in front of the construction surface and its corresponding coordinates. An oriented bounding box is constructed based on the current construction surface size, and the coordinates of the vertices of the oriented bounding box are calculated to serve as the potential field boundary.

[0053] Step 2: Automatically interpret and process the crack pixel points distributed in the current excavation surface and the previous 4-5 consecutive mileage segment excavation surface image datasets, and output the excavation surface crack binary image dataset. Figure 3 The detailed sub-steps are as follows:

[0054] Step 2.1: Access the excavation surface image datasets for each mileage segment and write a script in Python using the open source computer vision library OpenCV. Using the methods encapsulated in the open source library, perform geometric transformations, random noise processing, and random light processing on the image datasets to enhance the existing datasets and effectively expand the construction surface image datasets.

[0055] Step 2.2: Install the open-source image annotation tool Labelme in the local Conda environment, access the construction surface image set after data enhancement in step 2.1, use the point annotation tool to select the crack pixels and output them as a json file;

[0056] Step 2.3: Write a Python script to batch-process the json file output in step 2.2 that records the pixel locations of cracks in each construction surface image, and output a set of binary image annotation files in png format;

[0057] Step 2.4: Build the training set and test set according to the VOC (Visual Objects Classes) dataset format. Create a VOC root directory folder on the computer disk. Create subdirectories such as JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the construction surface image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set;

[0058] Step 2.5: Call the open-source deep learning framework PyTorch in Python and use the semantic segmentation model U-Net to automatically and accurately extract each fissure pixel. Input the VOC format dataset constructed in step 2.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training.

[0059] Step 2.6: Manually annotate the current construction surface and the previous 4-5 consecutive mileage excavation surface images, obtain the boundary contours corresponding to each excavation surface, and crop along the excavation surface contours. Then, call the open source computer vision library OpenCV in Python to resample the cropped images to obtain an excavation surface image dataset of the same size.

[0060] Step 2.7: After the model training in step 2.5 is completed, the excavation surface image dataset constructed in step 2.6 is input into the model, crack pixels are automatically identified and output as a mask file, the mask is processed, and a binary image dataset of the construction surface cracks is output.

[0061] Step 3: Use the thinning algorithm to extract the skeleton line pixels of the binary image dataset generated in step 2, and perform discrete sampling on the extracted crack feature pixels. Then, classify the pixel coordinates belonging to the same crack to obtain multiple sets of crack discrete feature pixel points. Figure 3 The detailed sub-steps are as follows:

[0062] Step 3.1: Call the open source computer vision library OpenCV in Python and use the thinning algorithm encapsulated in the open source library to skeletonize the crack lines distributed in each binary crack image, and output a binary crack skeleton line image dataset;

[0063] Step 3.2: In Python, call the open-source scientific computing library Numpy and store the coordinates of the crack pixels in the binary crack skeleton image dataset output in Step 3.1 into Numpy arrays one by one. Set a fixed sampling rate, perform discrete sampling on each Numpy array, remove the coordinates of unsampled pixels, and output a Numpy array set storing the two-dimensional coordinates of the discrete crack pixels.

[0064] Step 3.3: Generate an array set of two-dimensional coordinates of discrete crack pixel points based on step 3.2. Based on the binary crack skeleton line image dataset output in step 3.1, establish an oriented bounding box for each crack skeleton line image by image, record the discrete pixel points within the bounding box and store them in the order of the structural surface to obtain multiple sets of crack discrete feature pixel points.

[0065] Step 4: Obtain the mileage coordinates corresponding to the current excavation surface, perform spatial coordinate conversion on each set of pixel points generated in step 3 based on the construction surface size, and construct a three-dimensional crack feature point set that can reflect the coordinate information of the actual mileage segment;

[0066] Obtain the mileage coordinates corresponding to the current construction surface input in step 2.4 and the excavated construction surfaces of several consecutive mileage segments before it. Establish a conversion equation between pixel coordinates and mileage coordinates. Convert the discrete pixel points of each crack output in step 3.3 into 3D discrete points. Construct an oriented bounding box (Oriented Bounding Box) for the 3D discrete point set based on the construction surface size. Combine this with the potential field boundary constructed in step 1.5 to express the spatial extent of the fracture zone in front of the construction surface to construct a complete potential field boundary. Write a Python script to organize the 3D discrete points belonging to the same crack in the order of their crack ID and spatial coordinates (x, y, z). This results in a 3D crack feature point set reflecting the actual mileage segment coordinate information and outputs it as a txt file.

[0067] Step 5: For each set of three-dimensional crack feature points generated in step 4, construct a plane by fitting the crack structure, calculate the unit normal vector of each plane and process and output it as the polarization unit vector corresponding to each three-dimensional crack feature point, and output a two-tuple set of three-dimensional feature points and their polarization unit vectors. Figure 4 The detailed sub-steps are as follows:

[0068] Step 5.1: Create a custom point class in C++ that contains the fracture ID and spatial coordinates (x, y, z). Call the standard library iostream input file stream to input the txt file constructed in step 4. Create a point object set based on the fracture and coordinate information of each point recorded in the txt file.

[0069] Step 5.2: In C++, call the open source linear calculation library Eigen, use the least squares method to perform plane fitting on the point object set of the same fault line output in step 5.1 and solve the plane equation. According to the plane equation, further solve the plane normal vector, normalize the normal vector to obtain the polarization unit vector (G x , G y , G z ), output the coordinates of each point (x, y, z) and its corresponding polarization unit vector (G x , G y , G z )’s set of two-tuples;

[0070] Step 5.3: In C++, call the standard library iostream output file stream to output the tuple set constructed in step 5.2 and export it as a csv file.

[0071] Step 6: Input the two-tuple set of three-dimensional crack feature points and their corresponding polarization unit vectors constructed in step 5, use the potential field interpolation method to calculate the three-dimensional model of each crack structure surface, and finally output it as a triangular mesh. Then, according to the current excavation surface, the triangular mesh model of the excavated section is intercepted and eliminated to construct a detailed three-dimensional model of the surrounding rock cracks in front of the current construction surface. Figure 4 The detailed sub-steps are as follows:

[0072] Step 6.1: Call the open source geological modeling library Gempy in Python, input the constructed 3D fracture feature points and the constructed point set binary CSV file as initial data, set the potential field range and call the interpolation function to calculate and solve the 3D refined model of the surrounding rock fracture within the complete potential field range;

[0073] Step 6.2: In Python, call the visualization program ParaView, which is packaged in the open source address modeling library Gempy, to visualize the 3D refined model of the surrounding rock fractures generated in Step 6.1, and call the open source visualization tool library VTK (Visualization Toolkit) to output the 3D refined model of the surrounding rock fractures as a triangular mesh model;

[0074] Step 6.3: Call the open source geometric modeling library OpenCascade in C++, input the surrounding rock fracture triangular mesh model constructed in step 6.2, and construct a geometric body by obtaining the oriented bounding box (Oriented Bounding Box) of the three-dimensional discrete point set of the excavated construction surface according to step 2.4. Perform a Boolean difference operation between the geometric body and the surrounding rock fracture triangular mesh model to eliminate the triangular mesh model of the excavated section, and obtain a refined three-dimensional model of the surrounding rock fracture in front of the current construction surface.

[0075] The three-dimensional refined modeling method of broken surrounding rock in tunnels during construction period based on potential field interpolation proposed in the present invention has the advantages of low cost and high model accuracy. It effectively solves the problem that the horizontal and vertical spatial distribution characteristics and geometric structure information of the fracture structure of the surrounding rock in the broken zone in front of the construction face cannot be expressed under the condition of incomplete spatial attributes of geological radar image data obtained at the tunnel construction face scale. It can realize three-dimensional refined modeling of the surrounding rock fractures in front of the construction face, providing an important reference for dynamic optimization design and construction safety assurance of tunnel construction.

[0076] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A three-dimensional refined modeling method for broken surrounding rock in tunnels during construction based on potential field interpolation, characterized by: The steps include: Step 1: Automatically interpret the extension length of the broken zone in the geological radar image in front of the current construction surface, and determine the potential field boundary based on the current construction surface size; Step 2: Automatically interpret and process the crack pixel points distributed in the current excavation surface and the previous 4-5 consecutive mileage segment excavation surface image datasets, and output the excavation surface crack binary image dataset; Step 3: Use a thinning algorithm to extract skeleton line pixels from the binary image dataset generated in step 2, and perform discrete sampling on the extracted crack feature pixels. Then, classify the pixel coordinates belonging to the same crack to obtain multiple sets of crack discrete feature pixel points. Step 4: Obtain the mileage coordinates corresponding to the current excavation surface, perform spatial coordinate transformation on each set of pixel points generated in step 3 based on the construction surface size, and construct a three-dimensional crack feature point set that reflects the coordinate information of the actual mileage segment; Step 5: construct a plane by fitting the crack structure for each set of three-dimensional crack feature points generated in step 4, calculate the unit normal vector of each plane, process and output the polarization unit vector corresponding to each three-dimensional crack feature point, and output a two-tuple set of three-dimensional feature points and their polarization unit vectors; Step 6: Input the two-tuple set of three-dimensional crack feature points and their corresponding polarization unit vectors constructed in step 5, use the potential field interpolation method to calculate the three-dimensional model of each crack structural surface, and finally output it as a triangular mesh. Then, based on the current excavation surface, the triangular mesh model of the excavated section is intercepted and eliminated to construct a detailed three-dimensional model of the surrounding rock cracks in front of the current construction surface.

2. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 1, the specific steps include: Step 1.1: Install the open-source image annotation tool Labelme in the local Conda environment, access the geological radar image datasets of each mileage section, and manually annotate the dataset based on the significant image features of the broken zone, such as uneven waveform frequency distribution, poor regularity of waveform frequency changes, discontinuous phase axis, and cluttered waveforms. Use the rectangular annotation tool to select the broken zone and export the annotation information to a JSON file. Step 1.2: Write a Python script to batch-process the json file output in step 1.1 that records the location of the fragmentation zones in each geological radar image, and output a set of binary image annotation files in png format; Step 1.3: Build the training set and test set according to the VOC dataset format, create a VOC root directory folder on the computer disk, and create subdirectories of JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the geological radar image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set; Step 1.4: Call the open source deep learning framework PyTorch in Python and use the object detection model YOLO-V5 to achieve automatic and precise positioning of the broken belt. Input the VOC format dataset built in step 1.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training. Step 1.5: After the model training in step 1.4 is completed, the geological radar image corresponding to the current excavation surface is input into the model to automatically locate the broken zone. Then, the broken zone interpretation result is converted into spatial coordinates based on the current mileage coordinates to obtain the true extension length of the broken zone in front of the construction surface and its corresponding coordinates. A directed bounding box is constructed based on the current construction surface size, and the coordinates of the vertices of the directed bounding box are calculated and used as the potential field boundary.

3. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 2, the specific steps include: Step 2.1: Access the excavation surface image datasets for each mileage segment and write a script in Python using the open source computer vision library OpenCV. Using the methods encapsulated by the open source library, perform geometric transformation, random noise, and random light processing on the image datasets to enhance the existing datasets and effectively expand the construction surface image datasets. Step 2.2: Install the open-source image annotation tool Labelme in the local Conda environment, access the construction surface image set after data enhancement in step 2.1, use the point annotation tool to select the crack pixels and output them as a json file; Step 2.3: Write a Python script to batch-process the json file output in step 2.2 that records the pixel locations of cracks in each construction surface image, and output a set of binary image annotation files in png format; Step 2.4: Build the training set and test set according to the VOC dataset format. Create a VOC root directory folder on the computer disk. Create subdirectories of JPEGImages, ImageSets, and SegmentationClass under the root directory folder: JPEGImages is used to store the construction surface image dataset, ImageSets is used to store the txt files after dividing the training set and test set, and SegmentationClass is used to store the image annotation file set; Step 2.5: Call the open-source deep learning framework PyTorch in Python and use the semantic segmentation model U-Net to automatically and accurately extract each fissure pixel. Input the VOC format dataset constructed in step 2.3, set parameters such as the learning rate, number of training rounds, and batch size, and perform model training. Step 2.6: Manually annotate the current construction surface and the previous 4-5 consecutive mileage excavation surface images, obtain the boundary contours corresponding to each excavation surface, and crop along the excavation surface contours. Then, call the open source computer vision library OpenCV in Python to resample the cropped images to obtain an excavation surface image dataset of the same size. Step 2.7: After the model training in step 2.5 is completed, the excavation surface image dataset constructed in step 2.6 is input into the model, crack pixels are automatically identified and output as a mask file, the mask is processed, and a binary image dataset of the construction surface cracks is output.

4. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 3, the specific steps include: Step 3.1: Call the open source computer vision library OpenCV in Python and use the thinning algorithm encapsulated in the open source library to skeletonize the crack lines distributed in each binary crack image, and output a binary crack skeleton line image dataset; Step 3.2: In Python, call the open-source scientific computing library Numpy and store the coordinates of the crack pixels in the binary crack skeleton image dataset output in Step 3.1 into Numpy arrays one by one. Set a fixed sampling rate, perform discrete sampling on each Numpy array, remove the coordinates of unsampled pixels, and output a Numpy array set storing the two-dimensional coordinates of the discrete crack pixels. Step 3.3: Generate an array set of two-dimensional coordinates of discrete crack pixel points based on step 3.

2. Based on the binary crack skeleton line image dataset output in step 3.1, establish a directed bounding box for each crack skeleton line image by image, record the discrete pixel points within the bounding box and store them in the order of the structural surface to obtain multiple sets of crack discrete feature pixel points.

5. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 4, the specific steps include: Obtain the mileage coordinates corresponding to the current construction surface input in step 2.4 and the images of the excavated construction surfaces of the previous several consecutive mileage segments. Establish a conversion equation between pixel coordinates and mileage coordinates. Convert the discrete pixel points of each crack output in step 3.3 into 3D discrete points. Construct a directed bounding box for the 3D discrete point set based on the construction surface size. Combine this with the potential field boundary constructed in step 1.5 to express the spatial extent of the fracture zone in front of the construction surface to construct a complete potential field boundary. Write a Python script to organize the 3D discrete points belonging to the same crack in the order of the crack ID and spatial coordinates (x, y, z). Obtain a 3D crack feature point set that reflects the actual mileage segment coordinate information and output it as a txt file.

6. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 5, the following steps are specifically included: Step 5.1: Create a custom point class in C++ that contains the fracture ID and spatial coordinates (x, y, z). Call the standard library iostream input file stream to input the txt file constructed in step 4. Create a point object set based on the fracture and coordinate information of each point recorded in the txt file. Step 5.2: In C++, call the open source linear calculation library Eigen, use the least squares method to perform plane fitting on the point object set of the same fault line output in step 5.1 and solve the plane equation. According to the plane equation, further solve the plane normal vector, normalize the normal vector to obtain the polarization unit vector (G x , G y , G z ), output the coordinates of each point (x, y, z) and its corresponding polarization unit vector (G x , G y , G z )’s set of two-tuples; Step 5.3: In C++, call the standard library iostream output file stream to output the tuple set constructed in step 5.2 and export it as a csv file.

7. The method for three-dimensional refined modeling of broken surrounding rock in tunnels during construction based on potential field interpolation according to claim 1 is characterized by: In step 6, specifically including: Step 6.1: Call the open source geological modeling library Gempy in Python, input the constructed 3D fracture feature points and the constructed point set binary CSV file as initial data, set the potential field range and call the interpolation function to calculate and solve the 3D refined model of the surrounding rock fracture within the complete potential field range; Step 6.2: In Python, call the visualization program ParaView, which is packaged in the open source geological modeling library Gempy, to visualize the 3D refined model of surrounding rock fractures generated in Step 6.

1. Then, call the open source visualization tool library VTK to output the 3D refined model of surrounding rock fractures as a triangular mesh model. Step 6.3: Call the open source geometric modeling library OpenCascade in C++, input the surrounding rock fracture triangular mesh model constructed in step 6.2, obtain the directed bounding box of the 3D discrete point set of the excavated construction surface according to step 2.4, construct a geometric body, perform a Boolean difference operation between the geometric body and the surrounding rock fracture triangular mesh model, eliminate the triangular mesh model of the excavated section, and obtain a refined 3D model of the surrounding rock fracture in front of the current construction surface.

Images