Tunnel full-space deformation extraction method based on mobile laser scanning

Abstract

The present invention discloses a method for extracting tunnel full-space deformation based on mobile laser scanning, comprising: using a tunnel profiler to detect the cross-sectional deformation value S_ of the tunnel at each sampling location. D And record the tunnel cross-section position L_ D and the angle value G_ of the cross-section survey line D The entire tunnel space was scanned using a vehicle-mounted mobile laser scanner to obtain the cross-sectional deformation value S_ at various locations within the tunnel space. Y The cross-sectional deformation value S_ obtained by the tunnel cross-section instrument at the sampling location. D The output is the cross-sectional deformation value S_ obtained by moving the laser scanner at the same location. Y As input, a sample set is constructed; the XGBoost machine learning model is trained using the sample set. After training, the cross-sectional deformation values ​​S_ at various locations in the tunnel space are calculated. Y Input the trained XGBoost machine learning model and output the deformed value S_ D This serves as the cross-sectional deformation value at the corresponding location. By integrating tunnel cross-section data and full-domain moving 3D laser scanning data, the correlation mechanism between the deformation at the sampling location and the overall deformation of the tunnel is revealed, enabling rapid extraction of the deformation amount throughout the tunnel space.

Description

Technical Field

[0001] This invention belongs to the field of tunnel monitoring technology, and more specifically, this invention relates to a method for extracting full-space deformation of tunnels based on mobile laser scanning. Background Technology

[0002] With the continuous advancement of the "Transportation Powerhouse" strategy, the mileage and number of tunnels in my country have increased rapidly. While appreciating these fruitful results, we must be soberly aware that many tunnels built in the early years have exposed more and more problems. Tunnel development has shifted from a construction-oriented stage to a stage that emphasizes both construction and maintenance. Health status analysis and assessment of tunnel structures are becoming increasingly important, and tunnel cross-sectional deformation is an important parameter for achieving health status analysis and assessment of tunnel structures.

[0003] Currently, the detection of convergence deformation in mountain tunnels largely relies on traditional methods, where technicians use tunnel profilers to detect deformation at monitoring points. These traditional methods have the following problems:

[0004] (1) The full-space deformation of the tunnel cannot be obtained in a timely and comprehensive manner;

[0005] (2) Manual monitoring methods require a lot of manpower and resources and are inefficient. Summary of the Invention

[0006] This invention provides a method for extracting full-space deformation of tunnels based on mobile laser scanning, aiming to improve the above-mentioned problems.

[0007] This invention is implemented as follows: a method for extracting full-space deformation of tunnels based on mobile laser scanning, the method specifically including the following steps:

[0008] S1. Use a tunnel cross-section analyzer to detect the cross-sectional deformation value S_ of the tunnel at each sampling location. D And record the tunnel cross-section position L_ D and the angle value G_ of the cross-section survey line D ;

[0009] S2. Use the mobile laser scanner mounted on the tunnel inspection vehicle to scan the entire tunnel space and obtain the cross-sectional deformation value S_ at various locations within the tunnel space. Y ;

[0010] S3, the cross-sectional deformation value S_ obtained by the tunnel cross-section instrument at the sampling location. D The output is the cross-sectional deformation value S_ obtained by moving the laser scanner at the same location. Y Construct a sample set as input;

[0011] S4. Train the XGBoost machine learning model using the sample set. After training, calculate the cross-sectional deformation values ​​S_ at various locations in the tunnel space. Y Input the trained XGBoost machine learning model and output the deformed value S_ D This serves as the cross-sectional deformation value at the corresponding location.

[0012] Furthermore, the vehicle-mounted mobile laser scanner will be placed at the cross-sectional location L_ Y Within the pre- and post-set distances, the angle value G_ of the cross-section survey line. D The deformation data in the vicinity are averaged to obtain the cross-sectional position L_ Y Cross-sectional deformation value S_ Y .

[0013] Furthermore, the cross-sectional deformation value S_ Y The specific process for obtaining it is as follows:

[0014] Preprocessing is performed on the laser point cloud obtained from the vehicle-mounted mobile laser scanner.

[0015] Acquire the cross-sectional position L_ of the vehicle-mounted mobile laser scanner Y Within the pre- and post-set distances, the angle value G_ of the cross-section survey line. D Nearby sector point cloud data;

[0016] The fan-shaped point cloud data is projected along the longitudinal direction of the tunnel. Using a set distance as the calculation unit, the inner and outer boundary point clouds formed after the fan-shaped point cloud projection are obtained. The inner and outer boundary point clouds are fitted to obtain the inner and outer boundary contour fitting lines. The cross-sectional deformation value S_ is calculated from the average coordinates of the inner and outer boundary contour fitting lines. Y .

[0017] Further preprocessing includes removing discrete point clouds and point clouds of interior decorations within the tunnel.

[0018] Furthermore, the specific method for obtaining the sampling location is as follows:

[0019] The tunnel is segmented into straight tunnels and curved tunnels with different radii of curvature. The intersections of straight tunnels and curved tunnels, as well as the intersections of curved tunnels with different radii of curvature, are set as sampling locations.

[0020] For straight tunnels, a sampling point is taken at predetermined intervals;

[0021] For curved tunnels, the sampling interval is set based on the radius of curvature of the curved tunnel, and the sampling position is selected based on the sampling interval.

[0022] Furthermore, if the tunnel cross-section is a single-slope tunnel, the value measured by the tunnel cross-section gauge is the tunnel cross-section deformation value S_. DIf the tunnel cross-section is a double-slope, the tunnel cross-section deformation value S_ is calculated using trigonometric functions based on the geometric relationship between the slope gradient and the height of the tunnel cross-section gauge. D .

[0023] Furthermore, the tunnel cross-section location L_ is identified by the tunnel mileage marker. D and tunnel cross-section location L_ Y .

[0024] Furthermore, the tunnel is a mountain tunnel.

[0025] Furthermore, the sample set was randomly divided into 5 parts. The range of values ​​for the hyperparameters n_tree and n_depth of the XGBoost machine learning model was set. Four parts of the data were selected as the model training set for model training. The remaining part was used as the validation set for prediction accuracy verification. This process was repeated 5 times. The prediction error of the model was evaluated by the root mean square error. The smaller the root mean square error of the test set, the higher the model accuracy. The model with high accuracy was selected as the trained XGBoost machine learning model.

[0026] Guided by the principles of "discrete sampling location focus detection" and "full-domain tunnel laser scanning," this invention integrates tunnel cross-section data and full-domain moving 3D laser scanning data to reveal the correlation mechanism between deformation at sampling locations and full-domain deformation of the tunnel point cloud. This enables rapid extraction of full-space deformation in mountain tunnels. This method is adaptable to different mountain tunnel cross-sectional forms, unaffected by the cross-sectional structure of mountain tunnels, and avoids the process of extracting the central axis from circular cross-sections. It is more suitable for mountain tunnels with special cross-sectional shapes such as horseshoe shapes. Furthermore, the XGboost algorithm has few hyperparameters, making hyperparameter optimization easy to implement. It also exhibits strong model transferability under small sample conditions, making it suitable for establishing deformation extraction models for mountain tunnels. Attached Figure Description

[0027] Figure 1 A flowchart of a tunnel full-space deformation extraction method based on mobile laser scanning provided in an embodiment of the present invention. Detailed Implementation

[0028] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0029] This invention, relying on a tunnel inspection vehicle, fully explores the potential of mobile laser scanners and combines them with traditional tunnel cross-section instruments to address the challenge of digital intelligent perception of tunnel deformation across the entire area. Based on the XGBoost algorithm, it proposes a method for extracting full-space deformation of mountain tunnels using mobile laser scanning. Guided by the concept of "discrete sampling location key detection" + "tunnel full-area laser scanning," it integrates tunnel cross-section instrument data at sampling locations with full-area mobile 3D laser scanning data. Through machine learning algorithms, it reveals the correlation mechanism between cross-sectional deformation information at sampling locations and point cloud full-area deformation, achieving rapid extraction of tunnel full-space deformation and providing a data foundation for the maintenance of mountain tunnels.

[0030] Figure 1 The flowchart of the tunnel full-space deformation extraction method based on mobile laser scanning provided in the embodiments of the present invention includes the following steps:

[0031] S1. The cross-sectional deformation value S_ of the mountain tunnel at various sampling locations was detected using a tunnel profiler. D And record the tunnel cross-section position L_ D and the angle value G_ of the cross-section survey line D ;

[0032] The mountain tunnel is divided into straight tunnels and curved tunnels with different radii of curvature. The intersections of the straight tunnels and the curved tunnels, as well as the intersections of the curved tunnels with different radii of curvature, are set as sampling locations.

[0033] For straight tunnels, a sampling point is taken at predetermined intervals;

[0034] For curved tunnels, the sampling interval is set based on the radius of curvature of the curved tunnel, and the sampling position is taken based on the sampling interval. If the radius of curvature is large, the sampling interval is large, and vice versa.

[0035] In this embodiment of the invention, cross-sectional data at various sampling locations in the mountain tunnel are collected on-site using a calibrated tunnel profiler. The sampling locations can be reasonably selected based on the total length of the tunnel, with the cross-sectional spacing (i.e., sampling interval) forming a tunnel point deformation database Convergence_set. Since the data in the tunnel point deformation database Convergence_set comes from the tunnel profiler and is a test method recognized in the handover and acceptance specifications, the cross-sectional data in the tunnel point deformation database Convergence_set is regarded as the true value of the tunnel cross-sectional deformation.

[0036] The tunnel point deformation database Convergence_set contains cross-sectional data for each sampling location. The cross-sectional data consists of: tunnel cross-sectional location L_ D (Based on the mileage markers in the design drawings), cross-sectional deformation value S_ DAngle value G_ of cross-section survey line D Among them, the cross-sectional survey angle value G_ D Starting from the plane where the road surface is located, facing the direction of traffic in the tunnel, the angle formed by rotating the measuring line counterclockwise is generally measured at intervals of 5° or 10°, with a measurement range of about 330° (usually the measurement range of the tunnel cross-section instrument does not include the road surface).

[0037] In this embodiment of the invention, when the tunnel cross-section is a single-slope, the value measured by the tunnel cross-section instrument is the tunnel cross-section deformation value S_. D For tunnels with a double-slope cross-section, the tunnel cross-section deformation value S can be calculated using trigonometric functions based on the geometric relationship between the slope (generally 1.5-2%) and the height of the tunnel cross-section measuring instrument (measured by the instrument itself). D .

[0038] S2. Use the mobile laser scanner on the tunnel inspection vehicle to scan the entire tunnel space and obtain the cross-sectional deformation values ​​at various locations in the tunnel space.

[0039] A vehicle-mounted mobile laser scanner was used to scan the entire tunnel space, using tunnel mileage markers as identifiers to unify the positional relationship between data collected by a tunnel profiler and the point cloud data scanned by the vehicle-mounted mobile laser scanner. Since the mileage markers obtained by the tunnel profiler are more accurate, the tunnel cross-section position L_ from the Convergence_set tunnel point deformation database was used as the reference. D Using the vehicle-mounted mobile laser scanner as a reference, the cross-sectional position L_ is obtained. Y The cross-sectional deformation data within 25cm before and after the tunnel profile were averaged, and the angle value G_ of the cross-sectional survey line obtained from the tunnel profile instrument was taken into account. D Consistent with this, the cross-sectional deformation value S_ at the same cross-sectional location and survey line angle as in the tunnel point deformation database Convergence_set was obtained. Y .

[0040] In obtaining the deformation value S_ of the cross-section of the moving laser scanning surface Y In this process, the first step is to reduce data noise. The specific noise reduction methods are as follows:

[0041] Step 1: Use the built-in automatic noise reduction function of laser point cloud processing software (such as COPRE software) to remove discrete points; Step 2: Manually identify and remove relevant point clouds of interior decorations such as pipelines, decorative panels, and flue panels inside the tunnel; Step 3: Measure the cross-section angle value G_ D Location of cross-section L_ Y Point cloud data within 25cm before and after the cross-section are averaged, taking into account the angle value G_ of the cross-section survey line. DPoint cloud data within a 0.5° range before and after the tunnel are collected, forming a point cloud data area spanning 25cm longitudinally and 1° across the tunnel cross-section. This fan-shaped point cloud data is projected along the tunnel's longitudinal direction, using 1cm as the calculation unit. The inner and outer boundary point clouds of this projected area are calculated, yielding 25 inner and outer boundary points each. The Random Sample Consensus Algorithm (RANSAC) is then used to fit these inner and outer boundary point clouds, obtaining fitted contour lines. The cross-sectional deformation value S_ is calculated from the average coordinates of these fitted contour lines. Y .

[0042] S3, the cross-sectional deformation value S_ obtained by the tunnel cross-section instrument at the sampling location. D The output is the cross-sectional deformation value S_ obtained by moving laser scanning within the same cross-sectional area. Y Construct a sample set as input;

[0043] S4. Train the XGBoost machine learning model using the sample set. After training, calculate the cross-sectional deformation values ​​S_ at various locations in the tunnel space. Y Input the trained XGBoost machine learning model and output the deformed value S_ D This serves as the cross-sectional deformation value at the corresponding location.

[0044] In this embodiment of the invention, the cross-sectional deformation value S_ obtained by a vehicle-mounted mobile laser scanner Y As input, the cross-sectional deformation value S_ obtained from the tunnel profiler. D To output the model, an XGBoost machine learning model is built, with the goal of establishing S_ through machine learning. Y and S_ D The relationship is shown in the following formula:

[0045] S_ D =f(S_ Y )

[0046] The input and output data of the XGBoost machine learning model are preprocessed, and normalization is used to unify the input and output data. The sample set is randomly divided into 5 parts for 5-fold cross-validation. The range of values ​​for the hyperparameters related to the XGBoost machine learning model, namely the number of decision trees n_tree and the depth n_depth, is set. Four parts of the data are selected as the model training set for model training, and the remaining part is used as the validation set to verify the model's prediction accuracy. This process is repeated 5 times. The root mean square error (RMSE) is used to evaluate the model's prediction error. The smaller the RMS error of the test set, the higher the model accuracy. The model with the highest prediction accuracy is selected as the trained XGBoost machine learning model. When the models with similar prediction accuracies are compared, the training time of each candidate model is compared, and the model with the shortest training time is selected as the final prediction model, i.e., the trained XGBoost machine learning model.

[0047] By processing cross-sectional data collected at various locations using a vehicle-mounted mobile laser scanner using a trained XGBoost machine learning model, tunnel convergence deformation over the entire tunnel mileage can be obtained, enabling rapid and intelligent perception of tunnel deformation across the entire space. Utilizing a 5-fold cross-validation algorithm, considering factors such as the number and depth of decision trees in the XGBoost machine learning model, the accuracy of various models is evaluated using root mean square error. Taking into account model computation time, a suitable model is selected to obtain the correlation between deformation at sampling locations and global point cloud deformation, achieving accurate perception of the deformation across the entire space of different types of mountain tunnels.

[0048] The present invention has been described by way of example. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A method for extracting full-space deformation of tunnels based on mobile laser scanning, characterized in that, The method specifically includes the following steps: S1. Use a tunnel cross-section analyzer to detect the cross-sectional deformation values ​​of the tunnel at various sampling locations. And record the location of the tunnel cross section. and the angle value of the cross-section survey line ; S2. Use the mobile laser scanner mounted on the tunnel inspection vehicle to scan the entire tunnel space and obtain the cross-sectional deformation values ​​at various locations within the tunnel space. ; S3. Cross-sectional deformation value obtained by the tunnel cross-section instrument at the sampling location. The output is the cross-sectional deformation value obtained by moving the laser scanner at the same location. Construct a sample set as input; S4. Train the XGBoost machine learning model using the sample set. After training, collect the cross-sectional deformation values ​​at various locations in the tunnel space. Input the trained XGBoost machine learning model and output deformed values. As the cross-sectional deformation value at the corresponding location; Cross-sectional deformation value The specific process for obtaining it is as follows: Preprocessing is performed on the laser point cloud obtained from the vehicle-mounted mobile laser scanner. Acquire the position of the vehicle-mounted mobile laser scanner at the cross-section Within the pre- and post-set distances, the angle value of the cross-section survey line. Nearby sector point cloud data; The fan-shaped point cloud data is projected along the longitudinal direction of the tunnel. Using a set distance as the calculation unit, the inner and outer boundary point clouds formed by the projection of the fan-shaped point cloud are obtained. The inner and outer boundary point clouds are fitted to obtain the inner and outer boundary contour fitting lines. The cross-sectional deformation value is calculated by the average value of the coordinates of the inner and outer boundary contour fitting lines. .

2. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, The vehicle-mounted mobile laser scanner is positioned at the cross-section. Within the pre- and post-set distances, the angle value of the cross-section survey line. The deformation data in the vicinity are averaged to obtain the cross-sectional location. Cross-sectional deformation value .

3. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, Preprocessing includes removing discrete point clouds and point clouds of interior decorations within the tunnel.

4. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, The specific method for obtaining the sampling location is as follows: The tunnel is segmented into straight tunnels and curved tunnels with different radii of curvature. The intersections of straight tunnels and curved tunnels, as well as the intersections of curved tunnels with different radii of curvature, are set as sampling locations. For straight tunnels, a sampling point is taken at predetermined intervals; For curved tunnels, the sampling interval is set based on the radius of curvature of the curved tunnel, and the sampling position is selected based on the sampling interval.

5. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, If the tunnel cross-section is a single slope, the value measured by the tunnel cross-section gauge is the tunnel cross-section deformation value. ; If the tunnel cross-section is a double-slope, the tunnel cross-section deformation value can be calculated using trigonometric functions based on the geometric relationship between the slope gradient and the height of the tunnel cross-section gauge. .

6. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, The location of the tunnel section is marked by the tunnel mileage markers. and tunnel cross-section location .

7. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, The tunnel is a mountain tunnel.

8. The method for extracting tunnel full-space deformation based on mobile laser scanning as described in claim 1, characterized in that, The sample set was randomly divided into 5 parts. The range of values ​​for the hyperparameters n_tree and n_depth of the XGBoost machine learning model was set. Four parts of the data were selected as the model training set for model training. The remaining part was used as the validation set for prediction accuracy verification. This process was repeated 5 times. The prediction error of the model was evaluated by the root mean square error. The smaller the root mean square error of the test set, the higher the model accuracy. The model with high accuracy was selected as the trained XGBoost machine learning model.

Images