Method for non-contact measurement of gap between open-type TBM shield tail and surrounding rock

Abstract

The invention belongs to the field of TBM shield tail gap measurement, and particularly relates to a method for non-contact measurement of a gap between an open TBM shield tail and a surrounding rock, and the method combines Blender software to establish a virtual scene comprising the surrounding rock, a TBM shield, a reflective sticker, a binocular camera and an LED spotlight. The textures of the TBM shield, the surrounding rock and the reflective sticker in the virtual scene can be set according to a real image, a large number of binocular camera images can be rendered through changes of the textures and the surrounding rock surface, meanwhile, a gap between the tail of the TBM shield and the surrounding rock under a corresponding binocular graph is obtained through a written blender script, and a data set is generated. The data set is used for building a neural network model on a server, training is carried out, and a training weight is obtained; and then deploying the model and the weight into a measuring device, namely shooting an image of the binocular camera through the measuring device, and measuring the gap between the tail part of the shield and the surrounding rock.

Description

Technical Field

[0001] The present application relates to the technical field of TBM shield tail gap measurement, and in particular to a method for non-contact measurement of the gap between an open TBM shield tail and surrounding rock. Background Art

[0002] An open tunnel boring machine (TBM) is a heavy machine used to build tunnels. This type of machine is mainly used to process soil and soft rock with relatively stable geological conditions, and is suitable for the construction of urban underground transportation, subways, tunnels and other projects.

[0003] During the excavation of an open tunnel boring machine, a certain gap is usually reserved between the shield and the surrounding rock to ensure the smooth progress of the equipment. However, since the surrounding rock may be affected by factors such as stress changes, convergence or settlement during the excavation process, the surrounding rock gradually deforms and approaches the shield, and may eventually come into contact with the shield. When the deformation of the surrounding rock continues to increase and the extrusion pressure is too large, it may cause the TBM shield to become stuck, seriously affecting the construction efficiency and progress. Summary of the invention

[0004] The embodiment of the present application provides a method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock, and monitors the change of the gap between the surrounding rock and the shield in real time through visual measurement technology.

[0005] To achieve the above purpose, the present application provides a method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock. A training data set is generated by blender software, and after deep learning model training, the deep learning model and the trained weights are deployed on a measuring device. The measuring device can be used to obtain the distance between the tail of the shield and the gap. The method includes the following steps:

[0006] Step 1.1: Use a camera or the Internet to collect multiple images of the TBM shield, surrounding rock, and rock surface;

[0007] Step 1.2: Add a cylindrical geometry in the blender software, edit the cylindrical aggregate and modify the details according to the actual size of the TBM shield to obtain the TBM shield model, add a material to the TBM shield model and adjust its properties, use the Image Texture node in the material node editor to load the texture image, the texture image is the different TBM shield images in step 1.1, change the surface texture of the TBM shield, and thus obtain TBM shield images in various different situations for the training of the deep learning model to enhance the robustness of the model;

[0008] Step 1.3: Create a reflective sticker model in the blender software and stick the reflective sticker on the rear edge of the TBM shield; create nodes, connect relationships, and adjust node parameters so that the texture of the reflective sticker has different degrees of occlusion noise;

[0009] Step 1.4: Outside the TBM shield, build a surrounding rock model surrounding the TBM shield; add materials to the surrounding rock model and adjust its properties. Use the Image Texture node in the material node editor to load texture images. The texture images are images of different surrounding rocks and rock surfaces in step 1.1. The surrounding rock surface can be concave, convex, and convergent within the set range. In this way, surrounding rock images under various conditions are obtained for training the deep learning model to enhance the robustness of the model.

[0010] Step 2.1: In the blender software, make sure that the LED spotlight can cover the TBM shield, reflective tape and surrounding rock, and randomly adjust the LED's irradiation angle, position and light intensity;

[0011] Step 2.2: In the blender software, make sure that the binocular camera can capture the TBM shield, reflective stickers and surrounding rocks, and then randomly adjust the shooting angle and position of the binocular camera;

[0012] Step 2.3: Render the image of the binocular camera to obtain the left and right images;

[0013] Step 2.4: Through the script in the blender software, within the image range of the binocular camera, find the intersection line between the point on the tail edge of the TBM shield and the center point of the tail of the TBM shield, extend the intersection line to intersect with the surrounding rock surface, and obtain the distance from the point on the tail edge of the TBM shield to the surrounding rock. The minimum value of these distance values ​​is the gap between the tail of the TBM shield and the surrounding rock.

[0014] Step 3.1: Repeat steps 1.1 to 2.4 and change the parameters therein to obtain a training data set. The input of the training data set is the left and right images, and the output is the gap data set between the tail of the TBM shield and the surrounding rock. On the server, a neural network model is built to train the training data set and obtain the training weights.

[0015] Step 4.1: Deploy the neural network model on the measuring device and apply it in practice. The measuring device includes a binocular camera module, an LED spotlight, a computer module, a display and a power module. The binocular camera module, the LED spotlight and the display are all electrically connected to the computer module. The power module is used to power the binocular camera module, the LED spotlight, the computer module and the display.

[0016] Optionally, in step 4.1, each measurement can be selected whether to perform manual measurement verification. If manual measurement verification is performed, the manually measured gap between the TBM shield and the surrounding rock is collected, and the gap and the left and right images taken by the measuring equipment are saved as a correction training set. The correction training set is then added to the original training data set and trained using a neural network model to optimize the model parameters and improve measurement accuracy.

[0017] Optionally, steps 1.2 to 2.4 are based on a virtual scene built by blender software in the server, and the virtual scene includes a TBM shield, surrounding rocks, reflective stickers, a binocular camera and an LED spotlight.

[0018] Optionally, in step 4.1, in actual application, a reflective sticker is attached to the tail of the TBM shield, the LED spotlight is used to illuminate the tail of the TBM shield, and the computer module collects images of the TBM shield, the reflective sticker and the surrounding rock through the binocular camera module, analyzes and processes, and obtains the gap between the tail of the TBM shield and the surrounding rock.

[0019] Optionally, step 2.4 specifically includes the following steps:

[0020] Step 2.4.1: Obtain the position of the tail edge and center point of the TBM shield: First, use Blender Python API to obtain the geometric data of the tail of the TBM shield. The points on the tail edge of the TBM shield are extracted through the vertex information of the TBM shield model. The tail center point of the TBM shield is determined by directly selecting the center point of the TBM shield model.

[0021] Step 2.4.2: Calculate the direction of the intersection line: For each point on the tail edge of the TBM shield, calculate the vector direction from the tail center point of the TBM shield to the point, and generate the straight line equation based on these direction vectors;

[0022] Step 2.4.3: Extend the intersection line with the surrounding rock surface: Use the ray projection function of Blender software to extend the intersection line from the center point of the tail of the TBM shield to the edge point in the TBM shield model until it collides with the surface of the surrounding rock. The result of ray projection can give the intersection point of the ray and the surrounding rock;

[0023] Step 2.4.4: Calculate the distance between each point on the tail edge of the TBM shield and the intersection with the surrounding rock surface using the vector distance formula, store all these distance values, and find the minimum value among them. The minimum value is the gap between the tail of the TBM shield and the surrounding rock.

[0024] Optionally, step 3.1 specifically includes the following steps:

[0025] Step 3.1.1: Preprocessing and correction of input images: The left image and right image are obtained by Blender rendering, which represent the scene captured by the binocular camera respectively. After image correction processing, ensure that the left and right images are in the same horizontal plane;

[0026] Step 3.1.2: The corrected image enters the convolution layer, and a small convolution kernel is used to capture the edge and high-frequency features of the reflective sticker, and identify the characteristics of the smooth area and the reflective area; a large convolution kernel is used to capture the complex texture and rock structure characteristics of the surrounding rock;

[0027] Step 3.1.3: The feature blocks output by all convolutional layers are further convolved to compress the spatial information of each input channel into a scalar;

[0028] Step 3.1.4: Use the multi-head attention mechanism to perform preliminary matching of the features of the left and right images: The multi-head attention mechanism captures the association between features in parallel on different channels to ensure that the reflective stickers and surrounding rock features in the left and right images can accurately correspond;

[0029] Step 3.1.5: Use the multi-head attention mechanism to match feature points and obtain disparity features;

[0030] Step 3.1.6: After the disparity feature is regressed, it enters the fully connected layer and finally outputs the predicted gap value;

[0031] Step 3.1.7: The neural network model compares the predicted gap value with the true gap value, calculates the error through the loss function, and adjusts the model parameters using the optimization algorithm.

[0032] The method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock provided by the present application has the following beneficial effects:

[0033] The visual non-contact measurement technology is used to avoid the interference and limitations brought by the traditional contact measurement method. Secondly, by sticking reflective marks on the target position, the system can work stably even in environments with poor lighting conditions, significantly improving the adaptability and reliability of the measurement.

[0034] In addition, a training data set was generated using Blender software combined with actual on-site pictures, and the pictures were processed and analyzed using a neural network model to achieve accurate prediction of the gap, which makes the measurement method more robust in complex environments.

[0035] Finally, the solution also supports manual supplementation of data, which makes it possible to continuously improve and optimize the measurement model in practical applications and gradually improve the accuracy and reliability of the measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] in:

[0038] Figure 1 It is a schematic diagram of the relationship between the shield, surrounding rock, reflective stickers, binocular camera and LED spotlight in a method for non-contact measurement of the gap between the tail of an open TBM shield and surrounding rock shown in an embodiment of the present application;

[0039] Figure 2 It is a schematic diagram of the calculation principle of the gap in the method of non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock shown in one embodiment of the present application;

[0040] Figure 3 It is a framework diagram of a neural network model in a method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock shown in an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to facilitate the understanding of the present application, the present application will be described more comprehensively with reference to the relevant drawings below. The preferred embodiments of the present application are provided in the drawings. However, the present application can be implemented in many other different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive.

[0042] It should be noted that when an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or indirectly on the other element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element.

[0043] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In the description of the present application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0045] The embodiment of the present application provides a method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock. By using visual measurement technology to monitor the gap changes between the surrounding rock and the shield in real time, an early warning can be issued when the surrounding rock gradually deforms and approaches the shield. This monitoring method based on visual measurement can provide timely warnings before the shield jams, buying time for the construction team to take corresponding measures, thereby effectively reducing the construction risks caused by shield jams and improving the safety and efficiency of tunnel excavation.

[0046] In an embodiment of the present application, the method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock is to generate a training data set by blender software, and after training with a deep learning model, the deep learning model and the trained weights are deployed on a measuring device, and the measuring device can be used to obtain the distance between the tail of the shield and the gap.

[0047] It should be noted that Blender is a free and open-source 3D graphics software that provides a series of animation short film production solutions from modeling, animation, materials, rendering, to audio processing, video editing, etc. Blender has a variety of user interfaces that are convenient for use in different tasks, and built-in advanced film and television solutions such as green screen keying, camera reverse tracking, mask processing, and post-node synthesis. Blender has built-in Cycles renderer and real-time rendering engine EEVEE. It also supports a variety of third-party renderers.

[0048] This method combines Blender software to establish a virtual scene including surrounding rock, TBM shield, reflective stickers, binocular camera, and LED spotlight. The textures of TBM shield, surrounding rock, and reflective stickers in the virtual scene can be set according to the real image, the angles of binocular camera and LED spotlight can be set freely, and the surrounding rock surface can be deformed freely. A large number of binocular camera images can be rendered by the changes in texture and surrounding rock surface. At the same time, the gap between the tail of TBM shield and surrounding rock under the corresponding binocular graphics can be obtained by the written blender script, that is, a data set is generated. The neural network model is built on the server using the data set, and training is performed to obtain the training weights. Then the model and weights are deployed to the measuring device, and the binocular camera image can be taken by the measuring device to measure the gap between the tail of the shield and the surrounding rock.

[0049] Specifically, combined Figure 1-Figure 2 As shown, the method comprises the following steps:

[0050] Step 1.1: Use a camera or the Internet to collect a large number of images of the TBM shield, surrounding rock, and rock surface;

[0051] Step 1.2: Add a cylindrical geometry in the blender software, edit the cylindrical aggregate and modify the details according to the actual size of the TBM shield to obtain the TBM shield model, add a material to the TBM shield model and adjust its properties, and use the "Image Texture" node in the material node editor to load the texture image. The texture image can be the different TBM shield images in step 1.1. Change the surface texture of the TBM shield, thereby obtaining a large number of shield images in different situations for the training of the deep learning model to enhance the robustness of the model;

[0052] Step 1.3: Create a reflective sticker model in the blender software and stick the reflective sticker on the rear edge of the TBM shield; create nodes, connect relationships, and adjust node parameters so that the texture of the reflective sticker has different degrees of occlusion noise;

[0053] Step 1.4: Outside the TBM shield, build a surrounding rock model that surrounds the TBM shield; add materials to the surrounding rock model and adjust its properties. Use the "Image Texture" node in the material node editor to load texture images. The texture images can be images of different surrounding rocks and rock surfaces in step 1.1, and the surrounding rock surface can be concave, convex, and convergent within the set range. The purpose is to obtain a large number of surrounding rock images under different conditions for training deep learning models to enhance the robustness of the model;

[0054] Step 2.1: After ensuring that the LED spotlight can cover the TBM shield, reflective stickers and surrounding rocks, randomly adjust the angle, position and intensity of the LED;

[0055] Step 2.2: After ensuring that the binocular camera can capture the TBM shield, reflective stickers and surrounding rocks, randomly adjust the angle and position of the binocular camera;

[0056] Step 2.3: Render the image of the binocular camera to obtain the left and right images;

[0057] Step 2.4: Through the script in the blender software, within the binocular image range, find the intersection line between the point on the tail edge of the TBM shield and the center point of the tail of the TBM shield, extend the intersection line to intersect with the surrounding rock surface, and obtain the distance from the point on the tail edge of the TBM shield to the surrounding rock. The minimum value of these distance values ​​is the gap between the top of the TBM shield and the surrounding rock.

[0058] Step 3.1: Repeat steps 1.1 to 2.4 and change the parameters therein to obtain a training data set. The input of the training data set is the left and right images, and the output is a training data set of the gap between the tail of the shield and the surrounding rock. On the server, build a neural network model, train the training data set and obtain the training weights;

[0059] Step 4.1: Deploy the neural network model on the measuring equipment and apply it in practice. The measuring equipment includes a binocular camera module, an LED spotlight, a computer module, a display and a power module. The binocular camera module, the LED spotlight and the display are all electrically connected to the computer module. The power module is used to power the binocular camera module, the LED spotlight, the computer module and the display.

[0060] In actual application, the reflective sticker is attached to the tail of the TBM shield, and the LED spotlight is used to illuminate the tail of the TBM shield. The computer module collects images of the TBM shield, reflective sticker and surrounding rock through the binocular camera module, analyzes and processes them, and obtains the gap between the tail of the TBM shield and the surrounding rock.

[0061] In one embodiment, in step 4.1, each measurement can be selected whether to perform manual measurement verification. If manual measurement verification is performed, the manually measured gap between the TBM shield and the surrounding rock is collected, and the gap and the left and right images taken by the measuring equipment are saved as a correction training set. The correction training set is then added to the original training data set and trained using a neural network model to optimize the model parameters and improve measurement accuracy.

[0062] This setting supports manual supplementation of data, which makes it possible to continuously improve and optimize the measurement model in practical applications and gradually improve the accuracy and reliability of the measurement.

[0063] In one embodiment, step 2.4 specifically includes the following steps:

[0064] Step 2.4.1: Get the position of the tail edge and center point of the TBM shield: First, use Blender Python API to obtain the geometric data of the tail of the TBM shield. The point on the tail edge of the TBM shield (such as Figure 2 Point A in the figure can be extracted through the vertex information of the TBM shield model, and the tail center point of the TBM shield (such as Figure 2 Point O) in the figure can be determined by directly selecting the center point of the TBM shield model;

[0065] Step 2.4.2: Calculate the direction of the intersection line: For each point on the tail edge of the TBM shield, calculate the vector direction from the tail center point of the TBM shield to the point, and generate the straight line equation based on these direction vectors;

[0066] Step 2.4.3: Extend the intersection line with the surrounding rock surface: Use the raycasting function of Blender software to extend the intersection line from the tail center point of the TBM shield to the edge point of the TBM shield in the TBM shield model until it collides with the surface of the surrounding rock. The result of raycasting can give the intersection point of the ray and the surrounding rock (such as Figure 2 Point B in the

[0067] Step 2.4.4: Calculate the distance between each point on the tail edge of the TBM shield and the intersection point on the surrounding rock surface using the vector distance formula ( Figure 2 The distance between point A and point B), store all these distance values, and find the minimum value among them, which is the gap between the tail of the shield and the surrounding rock.

[0068] In one embodiment, in combination Figure 3 As shown, step 3.1 specifically includes the following steps:

[0069] Step 3.1.1: Preprocessing and correction of input images: The left and right images are obtained by Blender rendering, representing the scene captured by the binocular camera, and are processed by image correction and camera distortion correction (eliminating the distortion caused by the lens); binocular image registration is performed to ensure that the left and right images are in the same horizontal plane to avoid alignment errors;

[0070] Step 3.1.2: The corrected image enters the convolutional layer, and a small convolution kernel (such as 3x3) is used to focus on capturing the edges and high-frequency features of the reflective stickers, identifying the characteristics of smooth areas and reflective areas, and a larger convolution kernel (such as 7x7) is used to capture the complex texture and rock structure characteristics of the surrounding rock;

[0071] Step 3.1.3: The feature blocks output by all convolutional layers are further convolved to compress the spatial information of each input channel into a scalar. This step can use 1x1 convolution to achieve cross-channel fusion, effectively reducing redundant features and reducing the number of output channels.

[0072] Step 3.1.4: Use the multi-head attention mechanism to perform preliminary matching of the features of the left and right images: The multi-head attention mechanism captures the association between features in parallel on different channels to ensure that the reflective stickers and surrounding rock features in the left and right images can accurately correspond. Through this mechanism, the model can more accurately identify the positional relationship of the same object in the left and right images, thereby improving the matching accuracy.

[0073] Step 3.1.5: Use the multi-head attention mechanism to match feature points and obtain disparity features;

[0074] Step 3.1.6: After the disparity feature is regressed, it enters the fully connected layer and finally outputs the predicted gap value;

[0075] Step 3.1.7: The neural network model compares the predicted gap value with the actual gap value, calculates the error through the loss function, and uses the optimization algorithm to adjust the model parameters to continuously improve its performance and accuracy.

[0076] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for non-contact measurement of the gap between the tail of an open TBM shield and the surrounding rock, characterized in that: The following steps are involved: Step 1.1: Use a camera or the Internet to collect multiple images of the TBM shield, surrounding rock, and rock surface; Step 1.2: Add a cylindrical geometry in the blender software, edit the cylindrical aggregate and modify the details according to the actual size of the TBM shield to obtain the TBM shield model, add a material to the TBM shield model and adjust its properties, use the Image Texture node in the material node editor to load the texture image, the texture image is the different TBM shield images in step 1.1, change the surface texture of the TBM shield, and thus obtain TBM shield images in various different situations for the training of the deep learning model to enhance the robustness of the model; Step 1.3: Create a reflective sticker model in the blender software and stick the reflective sticker on the rear edge of the TBM shield; create nodes, connect relationships, and adjust node parameters so that the texture of the reflective sticker has different degrees of occlusion noise; Step 1.4: Outside the TBM shield, build a surrounding rock model surrounding the TBM shield; add materials to the surrounding rock model and adjust its properties. Use the Image Texture node in the material node editor to load texture images. The texture images are the images of different surrounding rocks and rock surfaces in step 1.

1. Obtain surrounding rock images under various conditions for training deep learning models to enhance the robustness of the model. Step 2.1: In the blender software, make sure that the LED spotlight can cover the TBM shield, reflective tape and surrounding rock, and randomly adjust the LED's irradiation angle, position and light intensity; Step 2.2: In the blender software, make sure that the binocular camera can capture the TBM shield, reflective stickers and surrounding rocks, and then randomly adjust the shooting angle and position of the binocular camera; Step 2.3: Render the image of the binocular camera to obtain the left and right images; Step 2.4: Through the script in the blender software, within the image range of the binocular camera, find the intersection line between the point on the tail edge of the TBM shield and the center point of the tail of the TBM shield, extend the intersection line to intersect with the surrounding rock surface, and obtain the distance from the point on the tail edge of the TBM shield to the surrounding rock. The minimum value of these distance values ​​is the gap between the tail of the TBM shield and the surrounding rock. Step 3.1: Repeat steps 1.1 to 2.4 and change the parameters therein to obtain a training data set. The input of the training data set is the left and right images, and the output is the gap data set between the tail of the TBM shield and the surrounding rock. On the server, a neural network model is built to train the training data set and obtain the training weights. Step 4.1: Deploy the neural network model on the measuring device and apply it in practice. The measuring device includes a binocular camera module, an LED spotlight, a computer module, a display and a power module. The binocular camera module, the LED spotlight and the display are all electrically connected to the computer module. The power module is used to power the binocular camera module, the LED spotlight, the computer module and the display.

2. The method according to claim 1, characterized in that In step 4.1, you can choose whether to perform manual measurement verification each time you measure. If manual measurement verification is performed, collect the manually measured gap between the TBM shield and the surrounding rock, save the gap and the left and right images taken by the measuring equipment as a correction training set, and then add the correction training set to the original training data set and use the neural network model for training to optimize the model parameters and improve the measurement accuracy.

3. The method according to claim 1, characterized in that Steps 1.2 to 2.4 are based on a virtual scene built with the blender software in the server, which includes a TBM shield, surrounding rocks, reflective stickers, a binocular camera, and LED spotlights.

4. The method according to claim 1, characterized in that: In step 4.1, in actual application, the reflective sticker is attached to the tail of the TBM shield, the LED spotlight is used to illuminate the tail of the TBM shield, and the computer module collects images of the TBM shield, the reflective sticker and the surrounding rock through the binocular camera module, analyzes and processes, and obtains the gap between the tail of the TBM shield and the surrounding rock.

5. The method according to claim 1, characterized in that Step 2.4 specifically includes the following steps: Step 2.4.1: Obtain the position of the tail edge and center point of the TBM shield: First, use Blender Python API to obtain the geometric data of the tail of the TBM shield. The points on the tail edge of the TBM shield are extracted through the vertex information of the TBM shield model. The tail center point of the TBM shield is determined by directly selecting the center point of the TBM shield model. Step 2.4.2: Calculate the direction of the intersection line: For each point on the tail edge of the TBM shield, calculate the vector direction from the tail center point of the TBM shield to the point, and generate the straight line equation based on these direction vectors; Step 2.4.3: Extend the intersection line with the surrounding rock surface: Use the ray projection function of Blender software to extend the intersection line from the center point of the tail of the TBM shield to the edge point in the TBM shield model until it collides with the surface of the surrounding rock. The result of ray projection can give the intersection point of the ray and the surrounding rock; Step 2.4.4: Calculate the distance between each point on the tail edge of the TBM shield and the intersection with the surrounding rock surface using the vector distance formula, store all these distance values, and find the minimum value among them. The minimum value is the gap between the tail of the TBM shield and the surrounding rock.

6. The method according to claim 1, characterized in that Step 3.1 specifically includes the following steps: Step 3.1.1: Preprocessing and correction of input images: The left image and right image are obtained by Blender rendering, which represent the scene captured by the binocular camera respectively. After image correction processing, ensure that the left and right images are in the same horizontal plane; Step 3.1.2: The corrected image enters the convolution layer, and a small convolution kernel is used to capture the edge and high-frequency features of the reflective sticker, and identify the characteristics of the smooth area and the reflective area; a large convolution kernel is used to capture the complex texture and rock structure characteristics of the surrounding rock; Step 3.1.3: The feature blocks output by all convolutional layers are further convolved to compress the spatial information of each input channel into a scalar; Step 3.1.4: Use the multi-head attention mechanism to perform preliminary matching of the features of the left and right images: The multi-head attention mechanism captures the association between features in parallel on different channels to ensure that the reflective stickers and surrounding rock features in the left and right images can accurately correspond; Step 3.1.5: Use the multi-head attention mechanism to match feature points and obtain disparity features; Step 3.1.6: After the disparity feature is regressed, it enters the fully connected layer and finally outputs the predicted gap value; Step 3.1.7: The neural network model compares the predicted gap value with the true gap value, calculates the error through the loss function, and adjusts the model parameters using the optimization algorithm.

Images