A tunnel over-excavation intelligent detection and regulation method fusing multi-source perception and dynamic compensation

Abstract

A kind of tunnel over-under excavation intelligent detection and regulation method fusing multi-source perception and dynamic compensation.This application belongs to the field of tunnel engineering detection technology, and discloses a kind of tunnel over-under excavation intelligent detection and regulation method fusing multi-source perception and dynamic compensation, by integrating three-dimensional laser scanner, IMU, ultrasonic ranging and environmental sensor, a unified deployment, power supply and data transmission standard are formulated, sensor accurate positioning is realized by global coordinate calibration;At the same time, a special noise reduction scheme is designed for different raw data, point cloud and IMU data fusion is completed by Rodriguez formula and iterative closest point algorithm, the positioning deviation of tunnel inner wall fusion data is controlled within ±3mm, and the ranging deviation is ≤0.015m;Based on environmental sensor, IMU attitude data and geological survey data, the specific compensation of temperature and humidity, equipment attitude deviation and composite stratum characteristics is completed by quantitative formula respectively, and the improved voxel method is used to realize the millimeter-level calculation of over-under excavation volume.

Description

Technical Field

[0001] This invention belongs to the field of tunnel engineering detection technology, specifically a method for intelligent detection and control of tunnel over-excavation and under-excavation that integrates multi-source sensing and dynamic compensation. Background Technology

[0002] As a crucial infrastructure project in transportation, water conservancy, and municipal engineering, the construction quality and safety of tunnels directly determine the overall benefits of the project. Over-excavation significantly increases the consumption of support materials such as lining concrete, substantially raising the project cost. At the same time, the increased exposed area of ​​surrounding rock in over-excavated areas can easily lead to stress redistribution in the surrounding rock, increasing the risk of collapse. Under-excavation, on the other hand, results in insufficient tunnel clearance, failing to meet design requirements for passage and use. Furthermore, the insufficient thickness of the support structure in under-excavated areas reduces the overall load-bearing capacity of the tunnel, creating potential safety hazards for later operation.

[0003] Existing methods for detecting tunnel over- and under-excavation mostly rely on single sensors for data acquisition, such as 3D laser scanners or ultrasonic ranging sensors. These methods suffer from insufficient data accuracy and weak anti-interference capabilities. Furthermore, existing technologies have poor adaptability to construction environments. In curved tunnel construction, the complex tunnel alignment leads to significant sensor positioning deviations. In composite strata construction, the heterogeneity of geological properties causes data distortion, making it difficult to accurately obtain 3D spatial information about the tunnel wall. In addition, traditional detection methods are severely disconnected from construction control, resulting in delayed data processing and an inability to optimize construction parameters such as shotcrete in real time based on actual over- and under-excavation conditions. This makes it difficult to effectively control the amount of over- and under-excavation, leading to low construction quality and efficiency.

[0004] To address the aforementioned issues, while some existing technologies attempt to combine multi-sensor data for detection, they lack standardized sensor integration and deployment schemes and high-precision multi-source data fusion algorithms, making it impossible to effectively eliminate the impact of environmental interference and equipment errors. Furthermore, no dynamic compensation mechanism has been established for curved tunnels and complex strata, nor has an intelligent control model for construction parameters linked to the detection process been designed, making it difficult to achieve integrated and intelligent over-excavation and under-excavation detection and construction control. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent detection and control method for tunnel over-excavation and under-excavation that integrates multi-source sensing and dynamic compensation, in order to solve the problems of single sensing data, easy influence of environmental and equipment factors on detection results under complex working conditions, and insufficient connection between detection results and subsequent construction control in the existing tunnel over-excavation and under-excavation detection process, resulting in poor accuracy of over-excavation and under-excavation identification and construction control effect.

[0006] The process includes the following steps: Deploying sensor integration units along the tunneling direction in the tunnel construction section and completing global coordinate calibration; simultaneously collecting 3D laser point cloud data, inertial measurement unit attitude data, ultrasonic distance data, and environmental data; performing targeted noise reduction processing on the 3D laser point cloud data, inertial measurement unit attitude data, and ultrasonic distance data; fusing the 3D laser point cloud data with the inertial measurement unit attitude data; and fusing the 3D laser point cloud data with the ultrasonic distance data to obtain fused data of the tunnel inner wall. Based on the fused data of the tunnel inner wall, the tunnel median line is extracted and optimized. The tangent vectors of each point of the median line are calculated and a normal plane is constructed. The actual cross-sectional point cloud is extracted in each normal plane. The actual cross-sectional point cloud is fitted and smoothed with a non-uniform rational B-spline curve to obtain the actual cross-sectional model. Based on each actual cross-sectional model, a three-dimensional surface model of the tunnel is constructed. A design cross-section model is constructed based on the tunnel design documents. Over-excavation and under-excavation areas are identified based on the actual cross-section model and the design cross-section model. The original over-excavation and under-excavation volume is calculated using the improved voxel method. Based on the environmental data, inertial measurement unit attitude data, and geological survey data, environmental factor compensation, equipment error compensation, and stratum characteristic compensation are performed on the original over-excavation and under-excavation volume in sequence to obtain the final over-excavation and under-excavation volume. Based on the final over- and under-excavation volume, a shotcrete construction parameter control model is constructed. The optimal construction parameters are obtained by optimizing the shotcrete construction parameter control model and transmitting the optimal construction parameters to the shotcrete equipment control system for closed-loop feedback control of subsequent construction.

[0007] Preferably, the sensor integration units are arranged at intervals along the tunnel excavation direction, and the sensor integration units are installed on the mounting brackets on the tunnel sidewall, with the laser scanner emitting surface parallel to the tunnel axis; each sensor achieves synchronous transmission through an industrial bus interface, is powered by a power supply module, and transmits data through an industrial network.

[0008] Preferably, the global coordinate calibration includes: setting up a group of high reflectivity spherical targets distributed in a regular tetrahedral pattern every 8.5 meters in the curved tunnel section, with 4 targets in each group, the horizontal spacing between the targets being 1.2 meters and the vertical spacing being 0.8 meters; The three-dimensional coordinates of the tunnel entrance reference point are collected by a high-precision global navigation satellite system receiver as a global reference. The total station is used to measure each target using the trigonometric leveling method and the average value is taken to obtain the global coordinates of the target. The target is scanned by a laser scanner and the local coordinates of the target are extracted to establish the initial mapping relationship between the local coordinate system and the global coordinate system.

[0009] Preferably, in the data preprocessing and fusion process, the three-dimensional laser point cloud data is denoised using a combination of statistical filtering and bilateral filtering. Statistical filtering is used to remove noise points that deviate from the average distance by 1.8 times the standard deviation, while bilateral filtering is used to perform weighted averaging of the point cloud coordinates. The spatial domain standard deviation of the bilateral filtering is 0.08 meters, and the gray-scale domain standard deviation is 0.15. The inertial measurement unit attitude data is filtered using Kalman filtering, with the filtering iteration period consistent with the 50 Hz sampling period. The ultrasonic distance data is filtered using 5-point median filtering. When the fluctuation of three consecutive sets of data exceeds 0.03 meters, the moving average of the first 10 sets of data is taken as the correction result.

[0010] Preferably, the fusion process of the three-dimensional laser point cloud data and the attitude data of the inertial measurement unit includes: calculating the rotation matrices around the X-axis, Y-axis and Z-axis based on the pitch angle, roll angle and yaw angle output by the inertial measurement unit, and transforming the point cloud from the local coordinate system to the global coordinate system through the total rotation matrix to achieve preliminary alignment; then, using the global coordinates of the target as a reference, performing fine registration using an iterative nearest point registration algorithm, with an upper limit of 50 iterations and a convergence threshold of 0.003 meters.

[0011] Preferably, during the reconstruction of the tunnel 3D model, the tunnel median optimization adopts the weighted least squares method, and the weight coefficient is determined according to the local density of the point cloud. Specifically, when the local density of the point cloud is greater than 300 points per cubic meter, the weight coefficient is 0.8; when the local density of the point cloud is greater than 100 points per cubic meter and less than or equal to 300 points per cubic meter, the weight coefficient is 0.5; and when the local density of the point cloud is less than or equal to 100 points per cubic meter, the weight coefficient is 0.3. The tunnel cross-section is fitted with a 3rd-order non-uniform rational B-spline curve. When the root mean square error of the fitting is greater than 0.008 meters, control vertices are added, with a maximum of 30 control vertices. After fitting, the moving average method is used to smooth the curve twice.

[0012] Preferably, the process of calculating the original over-excavation and under-excavation volume using the improved voxel method includes: dividing the three-dimensional space of the tunnel into a voxel grid with a side length of 0.025 meters, where the volume of each voxel is 1.5625 x 10^-5 cubic meters; determining the region affiliation of the voxel center coordinates relative to the design cross-section model, and labeling over-excavation and under-excavation voxels; for edge voxels containing point cloud data, their effective volume is taken as 65% of the volume of a single voxel, and for non-edge voxels, their effective volume is calculated as the volume of a single voxel; and counting the total number of corrected over-excavation voxels and the total number of under-excavation voxels to calculate the original over-excavation volume and the original under-excavation volume.

[0013] Preferably, the dynamic compensation correction includes environmental factor compensation, equipment error compensation, and stratum characteristic compensation. Environmental factor compensation corrects the original over-excavation and under-excavation volumes based on changes in temperature and humidity relative to a reference value. Equipment error compensation corrects the compensated volume based on the absolute value of the difference between the filtered attitude and the calibrated attitude of the inertial measurement unit. Over-excavation volume is corrected by reduction, and under-excavation volume is corrected by increment, with the maximum correction not exceeding 3% of the compensated volume. Stratum characteristic compensation performs weighted correction based on the cross-sectional area ratio of different strata in the cross-section and the corresponding stratum correction coefficient. Specifically, the stratum correction coefficient for cohesive soil is 1.02, for sand and gravel it is 0.98, for silt it is 1.01, and for silty soil it is 1.03.

[0014] Preferably, the parameters to be optimized in the construction parameter control model include nozzle injection pressure, injection angle, nozzle moving speed, and adjacent path spacing. The nozzle injection pressure ranges from 0.3 MPa to 0.6 MPa, the injection angle ranges from 85 degrees to 95 degrees, the nozzle moving speed ranges from 0.02 m / s to 0.1 m / s, and the adjacent path spacing ranges from 0.1 m to 0.3 m. The optimization solution is obtained by using an improved whale optimization algorithm that incorporates adaptive weighting factors and Gaussian mutation.

[0015] Preferably, during real-time control and execution, the spraying equipment control system adjusts the spraying pressure via a proportional valve with an adjustment accuracy of ±0.01 MPa; adjusts the spraying angle via a robotic arm joint motor with a control accuracy of ±0.5 degrees; and adjusts the nozzle movement speed via a servo motor with an adjustment accuracy of ±0.001 meters per second. The optimized construction parameters are transmitted to the spraying equipment control system via an industrial Ethernet, with a transmission delay of less than or equal to 0.5 seconds. The target value of over- or under-excavation per meter for closed-loop feedback control is less than or equal to 0.05 cubic meters per meter in absolute value.

[0016] The beneficial effects of this invention are as follows: 1. This invention integrates a 3D laser scanner, IMU, ultrasonic ranging, and environmental sensors, establishing unified standards for deployment, power supply, and data transmission. Combined with global coordinate calibration, it achieves precise sensor positioning. Simultaneously, it designs a dedicated noise reduction scheme for different raw data, using the Rodriguez formula and ICP algorithm to fuse point cloud and IMU data. Ultrasonic measurement values ​​are used as local constraints to correct the point cloud, ensuring that the positioning deviation of the fused data from the tunnel inner wall is controlled within ±3mm and the ranging deviation is ≤0.015m, thereby improving the accuracy of tunnel morphology detection and the reliability of the data.

[0017] 2. Based on environmental sensors, IMU attitude data, and geological survey data, this invention uses quantitative formulas to perform targeted compensation for temperature and humidity, equipment attitude deviation, and composite strata characteristics. Combined with an improved voxel method, it achieves millimeter-level calculation of over- and under-excavation volume. Furthermore, a special target layout scheme is designed for curved tunnel sections to eliminate the influence of various errors on the detection results under complex working conditions. This makes the calculation results of over- and under-excavation volume more consistent with the actual shape of the tunnel and improves the adaptability of the detection method to complex construction environments.

[0018] 3. This invention constructs a control model with the goal of minimizing over- and under-excavation and concrete consumption. It solves for the optimal construction parameters by introducing an improved whale optimization algorithm with adaptive weights and Gaussian mutation. It can also transmit and adjust the parameters of the spraying equipment with high precision. The control process is repeated every 0.8m of tunneling to form a closed loop. This mechanism can control the over- and under-excavation of the tunnel to within 0.02m³ / m, while reducing concrete consumption by 12% and the rebound rate to 12%. This not only ensures the quality of tunnel support construction but also reduces material waste and improves the overall efficiency and intelligence level of tunnel construction. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a flowchart of the multi-source sensing data preprocessing and fusion process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figures 1 to 2 As shown, this embodiment uses a curved tunnel project in a complex stratum of an intercity railway as an application scenario. This tunnel is a single-bore, double-track curved tunnel with a curve radius of 800m, traversing a composite stratum of cohesive soil, gravel, and silt. The tunnel's designed cross-section is horseshoe-shaped, with an arch crown radius of 5.2m, an arch foot depth of 12.5m, and a bottom width of 10.8m. Tunnel excavation is divided into 50m monitoring and control sections. Construction employs drill-and-blast excavation combined with shotcrete initial support. This invention's method is applied to the detection of over- and under-excavation before the initial support of this tunnel and the control of shotcrete construction parameters. The specific implementation steps are as follows: Step 1: Multi-source sensing data acquisition This step forms the basis for over-excavation and under-excavation detection. Through standardized sensor integration and deployment, global target calibration, synchronous data acquisition, and strict quality control, multi-source raw sensing data of the tunnel inner wall is obtained. The acquisition time is selected when the tunnel blasting and muck removal are completed, the tunnel inner wall is unobstructed, and the initial support has not been carried out. The total acquisition time for a single 50m tunnel section is controlled within 9 minutes, meeting the technical requirement of ≤10 minutes.

[0022] 1.1 Sensor Selection and Integration Deployment Sixteen modular sensor integration units are employed, deployed every 3.2 meters along the excavation direction in a 50m tunnel section. Each unit integrates a phase-type 3D laser scanner, a six-axis IMU inertial measurement unit, four high-frequency ultrasonic ranging sensors, and environmental sensors for temperature, humidity, and dust concentration. All sensors achieve synchronous data transmission via an RS485 industrial-grade bus interface with a fixed baud rate of 115200bps, ensuring zero-delay and zero-packet-loss data transmission. All sensor integration units are fixed to pre-installed metal brackets on the tunnel sidewall using expansion bolts. The height of the brackets from the tunnel bottom is strictly controlled at 1.8m. After installation, a digital level is used for calibration to ensure that the laser scanner's emitting surface is parallel to the tunnel's design axis. The actual deviation after calibration is controlled within ±0.3°, less than the technical threshold of ±0.5°. All units are centrally powered via a 24V PoE switch, with a single unit power consumption of 28W≤30W. Data is transmitted to the field data acquisition terminal via industrial Ethernet and TCP / IP protocol, with an actual network latency of 6ms≤10ms.

[0023] 1.2 Target Placement and Global Coordinate Calibration Spherical targets with a diameter of 150mm and a diffuse reflection coating with a reflectivity of 88% ≥ 85% were selected. Magnetic adsorption bases were installed at the bottom of the targets. One set of targets was deployed every 8.5m in the curved tunnel section, for a total of 6 sets in the 50m section. Each set had 4 targets distributed in a regular tetrahedral pattern, with a horizontal spacing of 1.2m and a vertical spacing of 0.8m. On-site verification ensured that no three targets in any set were collinear.

[0024] A high-precision GNSS receiver was used to acquire the three-dimensional coordinates of the tunnel entrance reference point with a static positioning accuracy of ±3mm: X0=2568.321m, Y0=1895.678m, Z0=35.246m, serving as the global coordinate reference. A total station was then used, with an angle measurement accuracy of ±0.5″ and a distance measurement accuracy of ±1mm+2ppm×D). Trigonometric leveling was employed to measure the coordinates of each target group, with each target measured three times. The average value was taken as the final global coordinates of the target center. For example, the coordinates of target 1 were measured three times: 2570.156, 1896.325, 35.892; 2570.158, 1896.323, 35.891, 25. The values ​​were 70.157, 1896.324, and 35.893, with the final average being 2570.157, 1896.324, and 35.892. All measurement data were recorded in the format of target number, X, Y, Z, and measurement timestamp to the data acquisition terminal. After target calibration, the sensor integration unit's self-calibration program was initiated. The laser scanner performed a high-resolution specialized scan of the target area at 2000 points / ° to acquire point cloud data of the target surface. The coordinates of the target center in the scanner's local coordinate system were extracted using an ellipse fitting algorithm and correlated with the global coordinates measured by the total station. This established an initial mapping relationship between the local and global coordinate systems, laying the foundation for subsequent point cloud coordinate transformation.

[0025] 1.3 Multi-source data synchronous acquisition process The on-site data acquisition terminal sends a synchronous acquisition command, and the 16 integrated sensor units respond and enter the ready state. Acquisition is triggered uniformly by the main controller, and the measured synchronization accuracy is 0.8ms ≤ 1ms. The 3D laser scanner adopts a spiral scanning mode with a default scanning frequency of 10Hz and a scanning step of 0.12m along the tunnel axis. During the scanning process, after each 360° scan, the current pitch angle, roll angle, and yaw angle attitude data output by the IMU are automatically recorded and bound to the same timestamp with the point cloud data to ensure a one-to-one correspondence between the attitude data and the point cloud data. Four ultrasonic ranging sensors synchronously acquire local distance data of the tunnel wall, with each sensor outputting one distance value every 100ms, and continuously acquiring five data points from the same measurement point. After removing the maximum and minimum values, the average of the three middle data points is taken as the effective distance value. For example, if the five measured values ​​of a certain measuring point are 2.568m, 2.572m, 2.569m, 2.580m, and 2.567m, after removing the maximum value of 2.580m and the minimum value of 2.567m, the effective distance value is (2.568+2.572+2.569) / 3=2.5697m. The environmental sensor collects construction environment data in real time at a sampling frequency of 1Hz. The actual temperature T=28℃, humidity H=75%RH, and dust concentration of 2.3mg / m³ collected in this tunnel section are all linked with point cloud data, attitude data, and ultrasonic data through timestamps to facilitate subsequent error compensation analysis.

[0026] During the data acquisition process, the data acquisition terminal monitors the working status of each sensor in real time. In this embodiment, there are no sensor failures or data anomalies. After the acquisition is completed, all data is automatically stored on a 1TB SSD hard drive on site with a read / write speed of 550MB / s ≥ 500MB / s. At the same time, it is backed up to the cloud server in real time via a 5G module. The data format strictly follows the LAS1.4 standard and includes full-dimensional information such as point cloud coordinates, intensity values, color information, timestamps, and sensor attitude data.

[0027] 1.4 Data Acquisition Quality Control Before data acquisition, all sensors underwent preprocessing: the laser scanner warmed up for 12 minutes (≥10 minutes); the IMU underwent zero-bias calibration in a static state for 6 minutes (≥5 minutes); the measured zero-bias error after calibration was 0.008° / h (≤0.01° / h). During data acquisition, a dedicated person monitored the sensor's operating status. Due to the on-site dust concentration of 2.3 mg / m³ (<5 mg / m³), there was no need to activate the high-pressure air blowing device, and the scanning optical path remained unobstructed. After data acquisition, the raw data underwent comprehensive quality inspection. The inspection results are as follows: point cloud density ≥300 points / m² (≥350 points / m²), invalid points, and the percentage of points exceeding the 0.5m~10m range was 1.2% (≤2%); the standard deviation of the IMU attitude data was 0.08° (≤0.1°); the deviation between the ultrasonic distance data and the corresponding point distance measured by the laser scanner was 0.012m (≤0.015m); environmental data was complete and without any abnormalities. All indicators met the quality inspection standards, and re-acquisition was not required.

[0028] Step 2: Data Preprocessing and Fusion This step performs targeted noise reduction on the multi-source raw data collected in step 1, and then achieves high-precision fusion of the multi-source data through coordinate transformation, fine registration, and local constraint correction. This eliminates data noise and measurement deviations between sensors, and obtains real and unified fused data of the tunnel inner wall, providing a data foundation for subsequent 3D model reconstruction.

[0029] 2.1 Noise Reduction of Raw Data 2.1.1 Noise Reduction of 3D Laser Point Cloud Data A combined statistical filtering and bilateral filtering scheme was adopted: First, statistical filtering was performed, searching for 15 neighboring points within a 0.1m radius for each point cloud point, calculating the average distance and standard deviation between the point and its neighbors, and identifying and removing points whose distance deviated from the average distance by more than 1.8 times the standard deviation. In this embodiment, approximately 15,000 noise points were removed, effectively eliminating random noise caused by construction dust and vibration; then, bilateral filtering was performed, setting the spatial domain standard deviation. =0.08m, grayscale standard deviation =0.15, using the weighting formula Calculate the filtering weights and perform a weighted average on the filtered point cloud coordinates. This effectively preserves edge features such as tunnel arches and sidewalls while removing noise, avoiding feature loss caused by excessive smoothing of the point cloud.

[0030] 2.1.2 IMU Attitude Data Denoising Kalman filtering is used to reduce noise in IMU attitude data, and the state equation is defined. Observation equations The system matrix A is a 10×10 identity matrix, and the input matrix is... B Zero matrix, process noise W The covariance matrix is ([0.001, 0.001, 0.001, 0.01, 0.01, 0.001, 0.001, 0.001, 0.001]), observation matrix H The identity matrix is ​​given by the observation noise V covariance matrix. ([0.0005, 0.0005, 0.0005, 0.005, 0.005, 0.0005, 0.0005, 0.0005, 0.0005]), the filtering iteration period is the same as the IMU sampling period, both are 0.02s.

[0031] After Kalman filtering, the fluctuation range of pitch, roll, and yaw angle data output by the IMU is significantly reduced, and the stability of attitude data is improved by more than 80%, providing an accurate attitude reference for subsequent point cloud coordinate transformation.

[0032] 2.1.3 Noise Reduction of Ultrasonic Distance Data Five-point median filtering is used to reduce noise in ultrasonic data. After sorting the five consecutive distance data outputs from each sensor, the maximum and minimum values ​​are removed, and the average of the three middle data is taken as the effective distance value. In this embodiment, there is no case where the fluctuation range of three consecutive effective distance values ​​exceeds 0.03m, so there is no need to start the abnormal detection mechanism. The deviation of the ultrasonic data after noise reduction is controlled within ±0.001m, and the consistency with the laser ranging data is significantly improved.

[0033] 2.2 Multi-source data fusion 2.2.1 Fusion of Point Cloud and IMU Data First, based on the attitude data filtered by the IMU, the pitch angle... Roll angle Heading angle Calculate the rotation matrices about the X, Y, and Z axes using the Rodriguez formula. , , Then calculate the total rotation matrix. = By using a rotation matrix, the point cloud in the local coordinate system of the laser scanner is transformed to the global coordinate system, thus achieving preliminary alignment between the point cloud and the global coordinate system. Using the global coordinates of the target as the reference point set and the center coordinates of the target in the transformed point cloud as the point set to be registered, the ICP iterative nearest point algorithm is used for fine registration. The upper limit of the number of iterations is set to 50 and the convergence threshold is 0.003m. In this embodiment, after 32 iterations, the root mean square error reaches 0.0025m < 0.003m, which meets the convergence threshold, and the registration is stopped. After fine registration, the positioning deviation of the point cloud in the global coordinate system is controlled within ±3mm.

[0034] 2.2.2 Fusion of Point Cloud and Ultrasonic Data Using the effective ultrasonic distance as a local constraint, the point cloud of the corresponding region is corrected: For each ultrasonic sensor, a spherical region with a radius of 0.2m is selected in the point cloud centered on its installation location. The average distance from all point clouds within this region to the sensor's installation location is calculated using the formula... Calculate the deviation from the effective ultrasonic distance value; In this embodiment, measurements were taken in some areas. =0.018m>0.015m, the point cloud in this area is translated and corrected along the direction from the sensor pointing to the tunnel wall, with the correction amount being... ×0.7=0.0126m. After correction, the average distance is recalculated, and the deviation is reduced to 0.011m < 0.015m, meeting the technical requirements and requiring no further correction. For the remaining areas... With a resolution of ≤0.015m, no correction is required, ultimately achieving high-precision fusion of point cloud and ultrasonic data.

[0035] Step 3: Reconstruction of the 3D model of the tunnel Based on the fused high-precision data, this step constructs a three-dimensional surface model that is highly consistent with the actual shape of the tunnel through operations such as tunnel midline extraction, cross-section extraction and fitting, and surface splicing. The geometric accuracy of the model is controlled within ±5mm of the actual tunnel, providing an accurate three-dimensional geometric basis for subsequent over-excavation and under-excavation calculations.

[0036] 3.1 Extraction of tunnel midline 3.1.1 Tunnel Orientation and Feature Point Extraction The fused point cloud is projected onto the XOY plane, and the X and Y coordinates of all point clouds are traversed and recorded. =2568.123m =2618.456m =1895.321m、 =1905.678m, using these four values ​​to determine the minimum bounding rectangle of the point cloud, through calculation, The unit vector of , determines the approximate orientation of the curved tunnel section as 35° east of north; Iterate through the Z-coordinates of all point clouds and find the point with the largest Z-coordinate as the initial apex. (2593.256, 1900.589, 42.658), and then, using this point as the center, search for the point cloud within the cube range of x∈[2592.756, 2593.756], y∈[1900.089, 1901.089], z∈[42.158, 42.658]. Through the geometric center formula The final arch apex was calculated. (2593.257, 1900.588, 42.657); Take the minimum Z-coordinate of all point clouds =35.123m, define the bottom region of the tunnel as z∈[35.123, 36.901], then select the range y∈[1898.073, 1902.926] within this region, calculate the geometric center of the point cloud within this range, and obtain the bottom center. (2593.255, 1900.587, 35.896).

[0037] 3.1.2 Median Line Generation and Optimization Connect the apex Parch of the arch to the center point of the bottom. The line segment is obtained, and its length is given. =6.761m, sampling was conducted along the long axis of the tunnel at intervals of 0.35m, with a sampling number m= 6.761 / 0.35 +1=21, and the coordinates of each sampling point are calculated through linear interpolation to form an initial median line point set. ={ }; Import the design centerline data for this tunnel. The design centerline point set is: ={ For each point in the initial median point set In the design, find the nearest point in the set of line points. Calculate the deviation vector ; Calculate the local density of the point cloud Determine the weighting coefficients for weighted least squares. In this embodiment, the dome area =380 points / m³ > 300 points / m³ =0.8; Sidewall area =220 points / m³, 100< ≤300 points / m³ =0.5; bottom area =80 points / m³ < 100 points / m³ =0.3; by fitting the objective function min∑k=1m || ′ ||2 Solve for the optimized median coordinates to obtain the optimized tunnel median. ={ After optimization, the Euclidean distance deviation between the median and the design median is ≤0.004m, which meets the accuracy requirements.

[0038] 3.2 Tunnel Cross-Section Extraction and 3D Modeling 3.2.1 Tunnel Cross-Section Extraction and Fitting Based on the optimized median line, calculate each median line point. tangent vector : Midpoint (2≤ i Tangent vector ≤20) First point tangent vector Unmarked tangent vector

[0039] With each Origin As the normal vector, through the plane equation Construct a normal plane within the normal plane. Point clouds are extracted within a radius of 5m centered on the target area. In this embodiment, the number of point clouds extracted from all cross sections is 2800 > 500, so there is no need to expand the radius or interpolate to supplement them. The cross-sections were fitted using a 3rd-order NURBS non-uniform rational B-spline curve. Twenty points uniformly distributed in the original point cloud were selected as candidate control vertices. The control vertices and weighting factors were solved by the least squares method. In this embodiment, the root mean square error (RMSE) of all cross-section fittings was 0.006m < 0.008m, so there was no need to add control vertices and the fitting accuracy met the requirements. The fitted cross-sectional curve was smoothed using the moving average method. For each point on the curve, the average coordinates of its three adjacent points before and after it were taken as the new coordinates. The smoothing process was repeated twice. After smoothing, the cross-sectional curve was continuous and smooth, without inflection points or abrupt change points.

[0040] 3.2.2 Construction of the 3D Surface Model of the Tunnel All fitted cross sections were arranged sequentially according to their midline points. The cross section spacing was adjusted based on the tunnel curvature: the cross section spacing in the area with larger curvature of the curved tunnel section (curve radius 800m) was reduced to 0.2m, while the spacing in other areas remained at 0.35m. All cross sections were stitched together into a three-dimensional surface model of the tunnel using a surface stitching algorithm. During the stitching process, the transition between adjacent cross sections was ensured to be smooth, without steps or gaps. The final three-dimensional model highly matched the actual shape of the tunnel, with a model resolution of 0.005m, which could clearly restore the subtle features of the tunnel's inner wall.

[0041] Step 4: Accurate Calculation of Over- and Under-Excavation Amounts This step uses the NURBS model of the tunnel design cross-section as a benchmark. Through over-excavation and under-excavation area identification and improved voxel volume calculation, it achieves accurate calculation of the over-excavation and under-excavation volume. In this embodiment, the original over-excavation volume is... =1.256m³, original under-excavation volume =0.892m³, with a calculation accuracy down to the millimeter level.

[0042] 4.1 Design Section Model Construction Based on the tunnel design documents, the contour parameters of the horseshoe-shaped design cross-section were extracted: arch crown radius 5.2m, arch foot coordinates ±5.4m, 38.256m, sidewall slope 1:0.3, and bottom width 10.8m. Under the same global coordinate system as the actual cross-section, a 3rd-order NURBS curve consistent with the actual cross-section was used to construct the design cross-section model, denoted as [model name missing]. This ensures that the coordinate system and fitting standards of the design model and the actual model are consistent.

[0043] For each actual cross-section NURBS model 200 evaluation points are evenly selected within the cross-section and distributed using the polar coordinate method, with 10 equal parts radially and 20 equal parts circumferentially. The distance from each evaluation point to the design cross-section model is calculated. distance Determine region attributes based on threshold: This is an over-excavation area. This is an area that is under-excavated. The area is considered qualified. In this embodiment, the over-excavated area of ​​the 50m tunnel section is mainly distributed in the tunnel arch and part of the right side wall, while the under-excavated area is mainly distributed in the left side wall and part of the tunnel bottom, with the qualified area accounting for more than 85%.

[0044] 4.3 Calculation of Over-excavation and Under-excavation Volume The modified voxel method is used to calculate the over- and under-excavation volume. The specific operation is as follows: Voxel mesh generation: A uniform voxel mesh is generated within the 3D space of the tunnel, with voxel side lengths... Through formula Calculate the volume of the voxel ; Voxel attribution determination: Traverse all voxels and determine their center coordinates to the design model. Distance, distance Marked as a super-dug voxel, distance Voxels marked as under-excavated are considered acceptable. Edge voxel correction: Marked over-drilled and under-drilled voxels are edge-determined. If a voxel contains actual point cloud data, it is considered an edge voxel, and the effective volume is taken as... of Data without point cloud elements consists of non-edge voxels, and the effective volume is taken as... of In this embodiment, the proportion of voxels at the over-excavated edge is [not specified]. The proportion of under-excavated edge voxels All volumes have been corrected according to standards. Accumulated volume: Total number of overmined voxels after statistical correction =80384, total number of unmined voxels =57088, through the formula , Calculated original over-excavation volume Original under-excavation volume .

[0045] Step 5: Dynamic Compensation Correction This step addresses three error sources—environmental factors, equipment errors, and geological characteristics—by performing three-dimensional dynamic compensation on the original over- and under-excavation volumes calculated in step 4. This eliminates the influence of various errors on the calculation results of over- and under-excavation volumes, yielding the final, accurate over- and under-excavation volumes. In this embodiment, the final over-excavation volume Final under-excavation volume .

[0046] 5.1 Environmental Factor Compensation Based on the actual temperature collected by environmental sensors 28 ,humidity =75%RH, using the environmental compensation formula Compensation is performed, including a temperature compensation coefficient. Temperature reference value = Humidity compensation coefficient =0.001 / %RH, humidity baseline value RH.

[0047] Calculate the environmental compensation values ​​for over-excavation and under-excavation volumes separately: Environmental compensation for over-excavation volume: =1.256×[1+0.002×(28 25) + 0.001 × (75) 60)]=1.256×1.021=1.282

[0048] Environmental compensation for under-excavation volume: =0.892×[1+0.002×3+0.001×15]=0.892×1.021=0.911

[0049] 5.2 Equipment Error Compensation The attitude error is calculated based on the difference between the attitude data after IMU filtering and the calibrated attitude data. In this embodiment, the measured pitch angle error is 0.15° and the roll angle error is 0.12°, and the maximum absolute value is taken. =0.15° as the equipment attitude error; using the formula: Calculate the equipment error correction amount. 0.002618.

[0050] Calculate the equipment error correction amount and the compensated volume for over-excavation and under-excavation volumes respectively: Over-excavation volume correction amount: =1.282 × 0.002618 ≈ 0.00335 Over-excavation volume is represented by "-", after compensation. =1.282 0.00335 = 1.27865 (The correction amount of 0.26% is less than 3%, which meets the threshold requirement.) Under-excavation volume correction amount: =0.911 × 0.002618 ≈ 0.00238 The under-excavated volume is represented by a "+", and after compensation... =0.911 + 0.00238 = 0.91338 (The correction amount of 0.26% is less than 3%, which meets the threshold requirement.) 5.3 Formation characteristic compensation According to the geological survey data of the tunnel, this 50m section traverses a composite stratum of cohesive soil, gravel, and silt. The cross-sectional area percentage of each stratum in the tunnel section was calculated using cross-sectional analysis: cohesive soil... , sand and pebbles silt Individual correction factors for each stratum: cohesive soil =1.02, sand and pebbles silt .

[0051] Through formula Calculate the comprehensive stratigraphic correction factor: =0.4×1.02+0.35×0.98+0.25×1.01=0.408+0.343+0.2525=1.0035 Then through the formula Calculate the final over- and under-excavation volume Final over-excavation volume: =1.27865 × 1.0035 ≈ 1.298 Final under-excavation volume: =0.91338 × 1.0035 ≈ 0.915 The final over- or under-excavation amount is calculated per linear meter: 50≈0.02596 / m, / 50≈0.0183 / m all satisfy 0.05 The initial requirement for / m.

[0052] Step 6: Intelligent Control of Construction Parameters This step uses the final over- and under-excavation volume. Based on this, a sprayed concrete construction parameter control model was constructed. The construction parameters were optimized by the improved Whale Optimization Algorithm (IWOA) and real-time control and closed-loop feedback were achieved, which further enabled precise control of over-excavation and under-excavation, while reducing concrete consumption. In this embodiment, after the optimized construction parameters were applied, the over-excavation and under-excavation per meter of tunnel were controlled within 0.02 m³ / m, and concrete consumption was reduced by 12%.

[0053] 6.1 Construction of the Regulation Model To minimize over- and under-mining (target value | |≤0.05m³ / m, optimal target| With the dual optimization objectives of minimizing the consumption of sprayed concrete (≤0.02 m³ / m) and minimizing the consumption of sprayed concrete, a parameter control model for sprayed concrete construction is constructed to determine the parameters to be optimized and their value ranges. Nozzle injection pressure P: 0.3MPa~0.6MPa Spray angle φ: 85°~95° Nozzle moving speed v: 0.02m / s~0.1m / s The adjacent path spacing s: 0.1m~0.3m. The model's constraints are: shotcrete rebound rate ≤15%, initial support thickness deviation ≤±0.05m, which meets the requirements of tunnel construction specifications.

[0054] 6.2 Parameter Optimization Algorithm An improved whale optimization algorithm (IWOA) is used to solve the regulation model. Adaptive weighting factors and Gaussian mutation operations are introduced into the traditional whale optimization algorithm to improve the algorithm's global search capability and convergence accuracy. Adaptive weighting factor: through the formula Calculation, setting =0.9, =0.4, maximum number of iterations =100 iterations, which allows the algorithm to maintain a global search in the early stages of iteration and quickly converge to the optimal solution in the later stages; Gaussian mutation operation: Gaussian mutation is performed on the current optimal solution every 20 iterations, with a standard deviation of 0.05, to avoid the algorithm getting trapped in local optima.

[0055] In this embodiment, the improved whale optimization algorithm converges to the optimal solution after 92 iterations, yielding the optimized construction parameters: Nozzle injection pressure P = 0.45 MPa Spray angle φ=90° The nozzle moving speed is v = 0.06 m / s The distance between adjacent paths is s = 0.2m 6.3 Real-time control and execution The optimized construction parameters are transmitted to the control system of the tunnel shotcrete equipment via industrial Ethernet, with an actual transmission delay of 0.3s to 0.5s. The control system then precisely adjusts the nozzle's operating status based on the optimized parameters. The injection pressure is adjusted via a proportional valve with an adjustment accuracy of ±0.01MPa. It was actually adjusted to 0.45MPa with no deviation. The spray angle is controlled by the joint motor of the robotic arm, with an adjustment accuracy of ±0.5°. It is actually adjusted to 90° without deviation. The nozzle movement speed is adjusted by a servo motor with an adjustment accuracy of ±0.001m / s, and is actually adjusted to 0.06m / s with no deviation. The spacing between adjacent paths is updated in real time through the device path planning module, and the updated path spacing is accurate to 0.2m.

[0056] After adjustment, the spraying equipment performs sprayed concrete construction according to the optimized parameters. During the construction process, the equipment operation status is monitored in real time to ensure that the parameters do not drift or deviate.

[0057] 6.4 Closed-loop feedback adjustment According to the technical requirements of this invention, for every 0.8m of tunnel excavation, steps 1 to 6 above are repeated to re-perform multi-source sensing data acquisition, data preprocessing and fusion, three-dimensional model reconstruction, accurate calculation of over- and under-excavation volume, dynamic compensation correction and construction parameter optimization and control. In this embodiment, a second inspection and adjustment was carried out after the tunnel had been excavated for 0.8m. The measured over-excavation and under-excavation were further reduced, with the over-excavation per linear meter decreasing to 0.018m³ / m and the under-excavation per linear meter decreasing to 0.015m³ / m, both controlled within 0.02m³ / m, achieving the optimal target. At the same time, the rebound rate of shotcrete decreased to 12%≤15%, the initial support thickness deviation was ±0.03m≤±0.05m, and the unit consumption of concrete was reduced by 12% compared with before optimization. The construction quality and construction efficiency were significantly improved.

[0058] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent detection and control of tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation, characterized in that, Includes the following steps: S1. An integrated sensor unit consisting of a 3D laser scanner, an inertial measurement unit (IMU), an ultrasonic ranging sensor, and an environmental sensor is deployed along the tunnel excavation direction and completes global coordinate calibration. It simultaneously collects 3D laser point cloud data, IMU attitude data, ultrasonic distance data, and environmental data. S2. Perform targeted noise reduction processing on the three-dimensional laser point cloud data, IMU attitude data, and ultrasonic distance data, and fuse the three-dimensional laser point cloud data with the IMU attitude data and the ultrasonic distance data to obtain the tunnel inner wall fused data. S3. Based on the fused data of the tunnel inner wall, extract and optimize the tunnel median line, calculate the tangent vector of each point of the median line and construct the normal plane, extract the actual cross-sectional point cloud in each normal plane, perform NURBS curve fitting and smoothing on the actual cross-sectional point cloud to obtain the actual cross-sectional NURBS model, and construct the tunnel three-dimensional surface model by splicing the actual cross-sectional NURBS models. S4. Construct a design section NURBS model based on the tunnel design documents. Identify over-excavation and under-excavation areas based on the actual section NURBS model and the design section NURBS model, and calculate the original over-excavation and under-excavation volume using the improved voxel method. S5. Based on the environmental data, IMU attitude data, and geological survey data, the original over-excavation and under-excavation volume is sequentially compensated for environmental factors, equipment errors, and geological characteristics to obtain the final over-excavation and under-excavation volume. ; S6. Based on the final over- and under-excavation volume A sprayed concrete construction parameter control model is constructed, and the optimal construction parameters are obtained by optimizing the model. The optimal construction parameters are then transmitted to the spraying equipment control system for closed-loop feedback control of subsequent construction.

2. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 1, characterized in that, S1 includes: The sensor integration units are arranged at intervals along the tunnel excavation direction. The sensor integration units are mounted on the mounting brackets on the tunnel sidewall, and the laser scanner emitting surface is kept parallel to the tunnel axis. Each sensor transmits data synchronously via an industrial bus interface, is powered by a power supply module, and transmits data through an industrial network.

3. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 2, characterized in that, In S1, the global coordinate calibration includes: In the curved tunnel section, a group of four tetrahedral high-reflectivity spherical targets are deployed every 8.5m, with a horizontal spacing of 1.2m and a vertical spacing of 0.8m. The three-dimensional coordinates of the tunnel entrance reference point were collected by a high-precision GNSS receiver as a global reference. The total station was used to measure each target three times using the trigonometric leveling method and the average value was taken to obtain the global coordinates of the target. The laser scanner scans the target at a resolution of 2000 points / °, extracts the local coordinates of the target through an ellipse fitting algorithm, and establishes an initial mapping relationship between the local coordinate system and the global coordinate system.

4. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 3, characterized in that, The specific method for denoising the original data in the data preprocessing and fusion stage is as follows: The 3D laser point cloud is denoised using a combination of statistical filtering and bilateral filtering. Statistical filtering removes noise points that deviate from the average distance by 1.8 times the standard deviation, while bilateral filtering uses a weighted formula... Calculate the weighted average coordinates, where =0.08m =0.15; Kalman filtering was used for IMU attitude data, and the filtering iteration period was consistent with the IMU 50Hz sampling period; 5-point median filtering was used for ultrasonic distance data, and the moving average of the first 10 groups was taken when the fluctuation of 3 consecutive data sets exceeded 0.03m.

5. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 4, characterized in that, The specific process of fusing point cloud and IMU data in the data preprocessing and fusion stage is as follows: based on the pitch angle α and roll angle output by the IMU... β Heading angle γ Calculate the rotation matrices around the X, Y, and Z axes using the Rodriguez formula. , , The total rotation matrix R = The point cloud is transformed from the local coordinate system to the global coordinate system to achieve initial alignment; then, with the global coordinates of the target as a reference, the iterative nearest point algorithm is used for fine registration, with an upper limit of 50 iterations and a convergence threshold of 0.003m.

6. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 5, characterized in that, In the tunnel 3D model reconstruction stage, the tunnel median optimization adopts the weighted least squares method, and the fitted objective function is: The weighting coefficient Based on the local density of the point cloud Sure: >300 points / m³ =0.8, 100 points / m³ When ≤300 points / m³ =0.5, When ≤100 points / m³ =0.3; The tunnel cross-section was fitted using a third-order NURBS curve. When the root mean square error (RMSE) of the fit was greater than 0.008m, the number of control vertices was increased to a maximum of 30. After fitting, the moving average method was used to smooth the curve twice.

7. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 6, characterized in that, The specific steps for calculating the original over- and under-excavation volume using the improved voxel method in the precise calculation stage of over- and under-excavation volume are as follows: The three-dimensional space of the tunnel is divided into a voxel grid with a side length of 0.025m, and the voxel volume... =0.025m³=1.5625×10^-5m³; Determine the region affiliation of the voxel center coordinates relative to the design cross-section model, and mark over-excavated and under-excavated voxels; For edge voxels containing point cloud data, the effective volume is taken as... 65% of the voxels are over-dug and under-dug; non-marginal voxels are taken as 100%; the total number of over-dug and under-dug voxels after statistical correction. , through respectively = × , = × Calculate the original over-excavation and under-excavation volumes.

8. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 7, characterized in that, The specific formula for the three-dimensional dynamic compensation in the dynamic compensation and correction phase is as follows: Environmental factor compensation: ,in =0.002 / ℃, =25℃, =0.001 / %RH =60%RH; Equipment error compensation: , The absolute value of the difference between the IMU-filtered attitude and the calibrated attitude is taken as the over-excavation volume. Under-excavation volume The maximum correction amount does not exceed 3%; Formation characteristic compensation: , ,in The cross-sectional area percentage of the i-th stratum. The correction factors are: 1.02 for cohesive soil, 0.98 for sand and gravel, 1.01 for silt, and 1.03 for muddy soil.

9. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 8, characterized in that, The parameters to be optimized and their ranges in the intelligent control stage of construction parameters are as follows: Nozzle injection pressure P = 0.3MPa~0.6MPa, injection angle φ = 85°~95°, nozzle moving speed v = 0.02m / s~0.1m / s, and adjacent path spacing s = 0.1m~0.3m; The improved whale optimization algorithm introduces an adaptive weight factor. ,in =0.9, =0.4, =100 times, and Gaussian mutation is performed on the current optimal solution every 20 iterations, with a standard deviation of 0.

05.

10. The intelligent detection and control method for tunnel over-excavation and under-excavation integrating multi-source sensing and dynamic compensation according to claim 9, characterized in that, The accuracy requirements for real-time control during the intelligent control phase of the construction parameters are as follows: the spraying pressure is adjusted via a proportional valve with an adjustment accuracy of ±0.01MPa; the spraying angle is controlled via a robotic arm joint motor with a control accuracy of ±0.5°; and the nozzle movement speed is adjusted via a servo motor with an adjustment accuracy of ±0.001m / s. The optimized construction parameters are transmitted to the spraying equipment control system via an industrial Ethernet connection with a transmission delay ≤0.5s. The target value for over- or under-excavation per meter under closed-loop feedback control is [missing value]. |≤0.05m³ / m.

Images