Intelligent detection system for highway tunnel lining structure disease based on laser point cloud

Abstract

This invention discloses an intelligent detection system for defects in highway tunnel lining structures based on laser point clouds, belonging to the field of tunnel engineering maintenance and intelligent detection technology. The system includes: a multimodal data acquisition module that simultaneously acquires three-dimensional laser point cloud data and internal stress wave signals; a seepage environment point cloud correction module that corrects the original point cloud under seepage conditions using a dedicated formula; a cross-modal visualization fusion module that achieves dual-source data registration and fusion and defect identification; a dynamic early warning module that dynamically updates thresholds and provides tiered early warnings; and a terminal output module that maps to a digital twin model and generates a visual report. This invention corrects point cloud data using dual formulas for seepage environments, fuses stress wave and point cloud data to identify defects, dynamically updates thresholds, and, combined with modular design, tiered early warning, and visual output, improves the accuracy of defect identification in complex environments, simplifies maintenance processes, and reduces data processing and decision-making costs.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering maintenance and intelligent detection technology, specifically to an intelligent detection system for defects in highway tunnel lining structures based on laser point clouds. Background Technology

[0002] Highway tunnels, as a crucial component of transportation infrastructure, play a vital role in connecting regional traffic and ensuring smooth passage. The lining structure, as the core protective and load-bearing component of the tunnel, directly impacts its long-term safe operation. With increasing traffic volume and extended service life, tunnel linings face increasingly complex service environments, placing higher demands on lining structure inspection technologies. Currently, laser point cloud technology, with its convenient acquisition of three-dimensional spatial information, is widely used in lining inspection, while stress wave technology, with its internal detection capabilities, serves as a supplementary method. The application of both technologies provides data support for tunnel maintenance; however, the industry still urgently needs more efficient and collaborative inspection solutions to adapt to complex operational scenarios.

[0003] Traditional tunnel lining inspection methods have significant limitations. Laser point cloud technology is susceptible to external factors in specific environments, leading to a decrease in the accuracy of the collected data. Stress wave technology, on the other hand, lacks an intuitive presentation of its results. These two technologies are often used independently, making it difficult to effectively correlate surface and internal inspection information. Furthermore, existing early warning mechanisms rely on fixed standards for judgment, failing to flexibly adjust according to actual working conditions during the inspection process. This results in discrepancies between early warning results and actual needs, increasing the workload for maintenance personnel and failing to meet the practical requirements for precise and intelligent tunnel maintenance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent detection system for defects in highway tunnel lining structures based on laser point clouds. This system simultaneously acquires three-dimensional laser point cloud data of the tunnel lining and internal stress wave signals through a multimodal data acquisition module. It eliminates data deviations caused by seepage by using a reflection coupling correction formula for seepage environment and a normal vector linkage compensation formula. Then, it achieves visual fusion of dual-source data through a stress wave amplitude point cloud visual trajectory feature mapping formula. Combined with an attention mechanism and dual-modal collaborative rules, it accurately identifies defect types. It also adapts to different detection conditions by using a speed linkage dynamic threshold update formula and a graded early warning mechanism. Finally, the system maps the results to a digital twin model and generates a visual report through a terminal output module.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent detection system for defects in highway tunnel lining structures based on laser point clouds, the system comprising: Multimodal data acquisition module: used to simultaneously acquire three-dimensional laser point cloud data and internal stress wave signals of highway tunnel lining structure to provide dual-source detection data support; Point cloud correction module for seepage environment: Corrects the original laser point cloud data under seepage environment by using the reflectivity coupling correction formula of seepage environment and the point cloud normal vector linkage compensation formula of seepage environment. Cross-modal visualization fusion module: By adopting the stress wave amplitude point cloud visual trajectory feature mapping formula, the stress wave signal is transformed into three-dimensional visual trajectory features and registered and fused with the corrected laser point cloud data. The attention mechanism is used to focus on suspected disease areas and the dual-modal collaborative rules are used to identify disease types. Dynamic early warning module: By adopting a speed-linked dynamic threshold update formula, it dynamically updates the visual feature threshold and generates hierarchical early warning information; Terminal output module: used to map the fused detection results and early warning information to the tunnel digital twin model to generate a visual detection report.

[0006] Furthermore, the multimodal data acquisition module includes a laser point cloud acquisition unit and a stress wave detection unit. The laser point cloud acquisition unit uses a vehicle-mounted laser scanner with an adjustable scanning frequency of 50~100Hz. It calibrates its own position through a GPS and IMU combined positioning module and acquires the three-dimensional coordinates, reflectivity, and initial value of the normal vector of the lining structure in real time. The stress wave detection unit uses a combination of a piezoelectric stress wave generator and an array receiver. The generator emits stress wave signals by attaching to the lining surface with silicone coupling agent. The array receiver receives stress wave signals in a one-to-many mode and simultaneously acquires the propagation time, amplitude, and attenuation coefficient of the stress waves inside the lining.

[0007] Furthermore, the reflectivity coupling correction formula for the infiltration environment is as follows: ,in, To correct the reflectance of the point cloud, The original point cloud reflectance was obtained directly from the laser point cloud acquisition unit. The material adaptive correction coefficient is 0.02~0.15, calibrated through experiments on C30 to C50 concrete samples. S represents the water permeability strength grade, determined by calculating the variance of point cloud reflectance, and is assigned a value of 1~3. To detect the actual humidity in the area, humidity sensors deployed inside the tunnel collect data in real time. The baseline humidity is 60% of the industry standard dry ambient humidity.

[0008] Furthermore, the formula for the linkage compensation of the point cloud normal vector in the seepage environment is as follows: ,in, For the compensated normal vector, The original normal vector is calculated after data is collected by the laser point cloud acquisition unit. The normal vector reflectivity correlation coefficient was calibrated using point cloud samples from over 1000 different seepage conditions, with values ​​ranging from 0.8 to 1.2. The absolute value of the reflectance correction difference is calculated using the reflectance coupling correction formula for infiltration environments. With the original The absolute value of the difference is obtained.

[0009] Furthermore, the formula for mapping the visual trajectory features of the stress wave amplitude point cloud is: ,in, The grayscale value of the visual trajectory color. β is an adjustable parameter, and its values ​​were set to 50 and 1 through multiple sets of tunnel detection experiments. The stress wave amplitude is obtained by analyzing the stress wave signal acquired through an array receiver. The point cloud density for the corresponding region is calculated using the scanning density of the laser point cloud acquisition unit. The stress wave attenuation coefficient is calculated by the stress wave propagation time and amplitude change.

[0010] Furthermore, the cross-modal visualization fusion module employs an elastic registration algorithm based on the lining reference profile to register the stress wave visual trajectory with the corrected laser point cloud data. The registration steps involve fitting the lining reference profile using the RANSAC algorithm, calculating the coordinate deviation between the stress wave visual trajectory and the corrected point cloud, and using a thin-plate spline interpolation algorithm for elastic adjustment to ensure precise alignment. The registration accuracy is ≤ ±2mm. The attention mechanism identifies suspected defect areas by analyzing the synergistic characteristics of point cloud normal vector mutations and stress wave trajectory distortions. The dual-modal feature weight of this area is increased to 70%, while the feature weight of non-critical areas is reduced to 30% to suppress interference signals. The dual-modal synergistic rule is that surface cracks correspond to laser point cloud normal vector mutations ≥ 0.3rad and stress wave trajectories that are continuous with slight amplitude attenuation, while surface spalling corresponds to laser point cloud curvature anomalies ≥ 0.05mm. -1 Furthermore, the stress wave trajectory showed no obvious distortion, the internal voids corresponded to no significant abnormality in the curvature of the laser point cloud, and the stress wave trajectory was broken with an amplitude of <2V. The grouting was not dense, which corresponded to poor uniformity of the laser point cloud reflectivity and discontinuous stress wave trajectory with an amplitude fluctuating between 2 and 3V.

[0011] Furthermore, the interference signals suppressed by the attention mechanism include noise points generated by dust, water accumulation, and pipeline obstruction in the tunnel. By strengthening the weight of the bimodal features in the suspected defect area, the influence of interference signals on defect identification is filtered.

[0012] Furthermore, the speed-linked dynamic threshold update formula is as follows: ,in, For real-time dynamic thresholds, The initial threshold was set based on detection data of similar tunnel defects from the past three years. The adjustment coefficient was calibrated to a value of 0.3 through multiple sets of speed condition experiments. The actual scanning speed is obtained in real time through the speed measurement module of the inspection vehicle. The baseline scanning speed is set at 3 km / h. The variance of the current sliding window features was calculated using 50 sets of continuously collected data. The sliding window size is linked to and matched with the detection distance. The historical average variance was calculated using statistical data from similar tunnel inspections over the past three years.

[0013] Furthermore, the dynamic early warning module's hierarchical early warning mechanism includes a first-level early warning and a second-level early warning. A first-level early warning occurs when a single visual feature exceeds the dynamic threshold, and is only pushed to the backend operation and maintenance system and generates a suspected defect marker. A second-level early warning occurs when two or more visual features exceed the dynamic threshold simultaneously, and simultaneously triggers an audible and visual alarm on the vehicle terminal, a pop-up notification on the backend system, and information push notifications to the operation and maintenance personnel's mobile devices. The threshold update cycle of the dynamic early warning module is linked to the laser scanning speed. When the scanning speed is ≥5km / h, the update cycle is set to 10s, and when the scanning speed is <5km / h, the update cycle is set to 5s.

[0014] Furthermore, the visual inspection report generated by the terminal output module includes a three-dimensional morphological diagram of the defect, risk level markings, and early warning and handling suggestions. The accuracy of the tunnel digital twin model is ≤ ±5mm from the actual tunnel. It supports three-dimensional rotation, scaling, and cross-sectional viewing. The terminal output module uses a combination of local storage and cloud encrypted storage to save data. Local storage data is retained for no less than one year, while cloud storage supports data retrieval by tunnel station number, inspection time, and defect type. It also supports data export and remote push.

[0015] Compared with existing technologies, this intelligent detection system for highway tunnel lining structure defects based on laser point clouds has the following advantages: I. This invention, through the combined use of a reflectivity coupling correction formula and a normal vector linkage compensation formula in a seepage environment, accurately eliminates reflectivity distortion and normal vector deviation in laser point clouds under seepage conditions, making the point cloud data more closely match the actual structural state of the lining. Combined with a stress wave amplitude point cloud visual trajectory feature mapping formula, the stress wave signal is converted into a three-dimensional visual trajectory, which is then deeply registered and fused with the corrected point cloud data. An attention mechanism is then used to focus on suspected defect areas. Based on dual-modal collaborative rules, various defect types such as surface cracks, internal voids, and incomplete grouting can be accurately identified. The speed-linked dynamic threshold update formula can flexibly adapt to different scanning speeds and tunnel section feature differences, making defect judgment more consistent with actual detection scenarios and effectively improving the accuracy of defect identification and positioning in complex environments.

[0016] Second, this invention, through modular integration and a collaborative design of tiered early warning and visualized output, makes the flow of detection data, early warning response, and result presentation smoother, eliminating the need for maintenance personnel to repeatedly switch between multiple devices and data sets. Tiered early warning can quickly distinguish the urgency of defects, while visualized reports and digital twin mapping can intuitively present key information about defects and directions for handling. The dual-storage architecture facilitates the tracing of historical detection data and the tracking of defect trends, reducing data processing and decision-making costs during maintenance and making tunnel maintenance work more efficient.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a diagram showing the overall module structure and data flow of the system of the present invention; Figure 2 This is a flowchart of the point cloud data correction process under water seepage conditions in this invention; Figure 3 This is a flowchart of the cross-modal visualization fusion and intelligent disease identification process in this invention. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example This embodiment discloses a specific implementation of an intelligent detection system for defects in highway tunnel lining structures based on laser point clouds, such as... Figure 1 As shown, the system simultaneously acquires 3D laser point cloud data and internal stress wave signals of the lining structure through a multimodal data acquisition module. A seepage environment point cloud correction module eliminates data deviations caused by environmental interference. A cross-modal visualization fusion module then achieves deep fusion of the dual-source data and identification of disease features. Combined with the flexible threshold updates and tiered early warning mechanism of the dynamic early warning module, the system finally presents visualized detection results and stored data through a terminal output module. The entire system achieves collaborative detection of surface and internal defects, adapts to complex detection environments and working conditions, ensures the accuracy of defect identification and the efficiency of operation and maintenance response, and provides a complete solution for health monitoring of tunnel lining structures.

[0022] The specific implementation process is as follows: Operation process of multimodal data acquisition module The core function of the multimodal data acquisition module is to simultaneously acquire the surface three-dimensional information and internal stress wave information of the tunnel lining structure, providing dual-source data support for subsequent detection and analysis. This module includes a laser point cloud acquisition unit and a stress wave detection unit. Both maintain a consistent acquisition frequency through a time synchronization protocol, ensuring that each set of point cloud data corresponds to stress wave data at the same time and location. This guarantees a one-to-one correspondence between the two source data in both spatial and temporal dimensions, laying the foundation for subsequent fusion analysis.

[0023] The laser point cloud acquisition unit uses a vehicle-mounted laser scanner, which uses a GPS and IMU combined positioning module to calibrate its spatial position in the tunnel in real time. Based on the principle of laser ranging, it continuously emits laser beams to the lining surface. After receiving the reflected beams, it calculates the three-dimensional coordinates and reflectivity data of the lining structure based on the laser propagation time and angle. At the same time, it derives the initial value of the normal vector based on the spatial coordinate relationship of adjacent point clouds. The scanning frequency can be flexibly adjusted according to the actual detection needs, achieving a balance between detection efficiency and data accuracy.

[0024] The stress wave detection unit consists of a piezoelectric stress wave generator and an array of receivers. The generator is tightly bonded to the lining surface using silicone coupling agent to eliminate the influence of air gaps at the contact surface on wave propagation, ensuring that the stress wave can effectively penetrate into the lining. This unit employs a one-transmitter-multiple-receiver operating mode. The generator emits stress waves at a preset frequency, and the array of receivers is evenly distributed within a certain range around the generator, synchronously receiving the stress wave signals after propagation through the lining. By analyzing the arrival time and energy changes of the signals, the propagation time, amplitude, and attenuation coefficient of the stress wave are obtained, thus indirectly reflecting the structural integrity of the lining.

[0025] Operation process of the point cloud correction module for water seepage environment The seepage environment in tunnels alters the physical properties of the lining surface, causing distortion in the reflectivity of the laser point cloud and disrupting the spatial distribution consistency of the point cloud data, thus affecting the accuracy of normal vector calculation. Therefore, a seepage environment point cloud correction module is needed to specifically correct the original point cloud data. For example... Figure 2 As shown, this module eliminates seepage interference and restores data authenticity by synergistically applying the reflectivity coupling correction formula and the point cloud normal vector linkage compensation formula for the seepage environment, from the two core dimensions of reflectivity and normal vector.

[0026] The formula for the reflectivity coupling correction in a seepage environment is: ,in, To correct the reflectance of the point cloud, The original point cloud reflectance was directly acquired by the laser point cloud acquisition unit. However, water seepage can adhere to the lining surface, altering the surface reflection efficiency, leading to… The reflection characteristics of the lining itself cannot be accurately reflected, so multiple correction parameters need to be introduced for compensation. k is the material adaptive correction coefficient. The reflection characteristics of concrete linings of different strength grades are different. It is obtained by conducting a large number of comparative experiments on concrete samples of the corresponding strength range and is used to adapt to the reflection correction requirements of different lining materials. S is the seepage strength level, which is determined by calculating the variance of the point cloud reflectance in the same detection area. The larger the variance of reflectance, the more uneven the seepage distribution in the area and the more significant the impact, corresponding to a higher seepage strength level. To detect the actual humidity in the area, humidity sensors deployed inside the tunnel collect data in real time, directly reflecting the current humidity conditions related to water seepage in the environment. The industry standard dry ambient humidity serves as a benchmark reference value for reflectance correction, representing the ideal ambient humidity conditions without water seepage interference. The original reflectance is then calculated using this formula. The point cloud reflectance is obtained by coupling calculations with various correction parameters. It effectively eliminates reflectivity distortion caused by water seepage.

[0027] The formula for the linkage compensation of the normal vector of the point cloud in seepage environment is as follows: ,in For the compensated normal vector, The original normal vector is calculated after data is collected by the laser point cloud acquisition unit. The presence of distortion will also lead to deviations in the spatial distribution analysis results of the point cloud, thus resulting in... It cannot accurately reflect the geometry of the lining surface. The normal vector reflectance correlation coefficient is obtained by training and calibration using a large number of point cloud samples under different seepage conditions. It is used to quantify the degree of influence of reflectance deviation on the normal vector calculation result. The absolute value of the reflectance correction difference is calculated using the reflectance coupling correction formula for infiltration environments. With the original The absolute value of the difference directly quantifies the degree of reflectivity distortion. This formula correlates the reflectivity correction result with the normal vector calculation. Perform targeted compensation to obtain the compensated normal vector. This ensures that the normal vector accurately reflects the geometry of the lining surface.

[0028] Operation process of cross-modal visualization fusion module The core of the cross-modal visualization fusion module is to transform abstract stress wave signals into visualized three-dimensional features, achieve deep fusion with corrected point cloud data, and realize accurate identification of defects through complementary verification of dual-source data. For example... Figure 3 As shown, its operation process is mainly divided into four steps: signal conversion, data registration, feature focusing and disease identification. Each step is progressive to ensure the fusion effect and identification accuracy.

[0029] The signal conversion stage uses the stress wave amplitude point cloud visual trajectory feature mapping formula: This transforms stress wave signals into three-dimensional visual trajectory features that spatially match point cloud data. The amplitude of the stress wave is obtained by analyzing the stress wave signal after it is collected by an array receiver. It reflects the energy change of the stress wave after it propagates inside the lining. When there are defects inside the lining, the stress wave energy will be lost, which will cause the amplitude to change. The point cloud density of the corresponding area is calculated from the scanning density of the laser point cloud acquisition unit, reflecting the density of the point cloud data in that area. The higher the point cloud density, the richer the spatial information, which is more conducive to disease location. The stress wave attenuation coefficient is calculated by the change in the propagation time and amplitude of the stress wave. It reflects the influence of the internal material of the lining on the attenuation of the stress wave. Internal defects can change the propagation path of the stress wave, resulting in an abnormal attenuation coefficient. β is an adjustable parameter, calibrated through multiple tunnel detection experiments, used to convert the numerical range of stress wave-related parameters into grayscale values ​​that meet visual presentation requirements. This leads to the formation of a three-dimensional visual trajectory that corresponds one-to-one with the spatial location of the point cloud data, transforming abstract stress wave information into intuitive visual features.

[0030] In the data registration stage, an elastic registration algorithm based on the lining reference contour is adopted to ensure that the stress wave visual trajectory and the corrected point cloud data are accurately aligned in spatial position. First, the corrected point cloud data is processed using the RANSAC algorithm, which can effectively remove abnormal noise points in the point cloud. The reference contour of the lining is obtained by fitting the remaining valid point cloud, ensuring the accuracy of the contour. Then, the deviation between the stress wave visual trajectory and the corrected point cloud data in three-dimensional spatial coordinates is calculated to clarify the degree of mismatch between the two in spatial position. Finally, a thin plate spline interpolation algorithm is used to elastically adjust the visual trajectory. This algorithm can make local adaptive adjustments according to the distribution of coordinate deviation, so that the stress wave visual trajectory and the point cloud data are accurately matched in spatial position, ensuring the spatial consistency of dual-source data in the subsequent defect identification process.

[0031] The feature focusing stage utilizes an attention mechanism to accurately locate suspected defect areas. This mechanism simultaneously analyzes the abrupt changes in the normal vector of the corrected point cloud and the distortion characteristics of the stress wave visual trajectory, identifying regions where these two types of features change synergistically. These regions, exhibiting both surface geometric anomalies and internal stress wave propagation anomalies, become suspected defect areas. By strengthening the dual-modal feature weights of these areas while reducing the feature weights of non-critical areas, the characteristic signals of suspected defect areas are highlighted, and the influence of noise points caused by dust, water accumulation, and pipeline obstruction within the tunnel is weakened, reducing interference for subsequent defect identification.

[0032] The defect identification stage is carried out based on the dual-modal collaborative rule. Surface cracks correspond to abrupt changes in the point cloud normal vector and continuous stress wave trajectories with slight amplitude attenuation. The abrupt change in normal vector reflects abrupt changes in the geometry of the lining surface, while the continuous stress wave trajectory indicates the absence of obvious internal voids. The slight amplitude attenuation reflects the small loss of stress wave energy caused by the cracks. Surface spalling corresponds to abnormal point cloud curvature and no obvious distortion in the stress wave trajectory. Abnormal curvature reflects geometric changes caused by the shedding of surface material, while the absence of distortion in the stress wave trajectory indicates that the internal structure is not affected. Internal voids correspond to no significant abnormalities in point cloud curvature and broken stress wave trajectories with reduced amplitude. Normal point cloud curvature indicates no obvious surface defects, while broken stress wave trajectories and reduced amplitudes reflect the interruption of stress wave propagation paths and significant energy loss caused by internal voids. Inadequate grouting corresponds to poor uniformity of point cloud reflectivity and discontinuous stress wave trajectories with fluctuating amplitudes. Poor reflectivity reflects the uneven surface material caused by inadequate grouting, while discontinuous stress wave trajectories and fluctuating amplitudes correspond to the unevenness of the internal structure.

[0033] The operation process of the dynamic early warning module The dynamic early warning module takes the disease features identified by the cross-modal visual fusion module as input, and realizes flexible adjustment of the threshold through the speed-linked dynamic threshold update formula. Based on the hierarchical early warning mechanism, it outputs targeted early warning information to ensure that the early warning results are consistent with the actual detection conditions.

[0034] The formula for updating the speed-linked dynamic threshold is: ,in The initial threshold is set based on the detection data of similar tunnel defects in the past three years, which comprehensively reflects the benchmark level of defect characteristics of similar tunnels and serves as the basis for threshold updates. The adjustment coefficient was obtained through experiments with multiple sets of different speed conditions. It is used to balance the influence of scanning speed and feature variance on the threshold and avoid threshold deviation caused by a single factor. The actual scanning speed is collected in real time by the speed measurement module of the detection vehicle. Different scanning speeds will lead to differences in data acquisition density and feature extraction accuracy, which need to be adjusted and adapted through threshold. The baseline scanning speed serves as a reference standard for scanning speed and represents the scanning speed under normal testing conditions. The characteristic variance of the current sliding window is calculated by collecting a set number of continuously collected data. The size of the sliding window is linked to the detection distance. The characteristic variance reflects the degree of fluctuation of the disease characteristics in the current detection section. The greater the fluctuation, the worse the structural consistency within the section. The historical average variance, calculated statistically from inspection data of similar tunnels over the past three years, reflects the average fluctuation level of defects in similar tunnels. The real-time dynamic threshold is calculated using this formula. This allows the threshold to be flexibly adjusted according to changes in scanning speed and fluctuations in the characteristics of the current segment, avoiding false alarms or missed alarms caused by a fixed threshold.

[0035] The tiered early warning mechanism is based on real-time dynamic thresholds. When a single visual feature exceeds the real-time dynamic threshold, a Level 1 warning is triggered. At this point, the risk of the defect is low or there are suspected features. The system only pushes the warning information to the backend maintenance system and generates a suspected defect marker for maintenance personnel to conduct subsequent unified investigations. When two or more visual features simultaneously exceed the real-time dynamic threshold, it indicates that the defect features are clear and the risk is high, triggering a Level 2 warning. The system simultaneously triggers an audible and visual alarm on the vehicle terminal, a pop-up notification in the backend system, and a push notification to the maintenance personnel's mobile device, ensuring that maintenance personnel receive the warning information immediately and respond quickly. Meanwhile, the threshold update cycle of the dynamic warning module is linked to the laser scanning speed. When the scanning speed is faster, the data update frequency is higher, and the threshold update cycle is correspondingly shorter. When the scanning speed is slower, the data update frequency is lower, and the threshold update cycle is appropriately extended, ensuring real-time matching between the threshold and the detection data.

[0036] Operation process of terminal output module The terminal output module receives early warning information from the dynamic early warning module and the disease identification results from the cross-modal visualization fusion module. Combined with the spatial positioning information from the multimodal data acquisition module, it completes data processing, visualization, and secure storage, providing direct support for operation and maintenance decisions.

[0037] This module first constructs a digital twin model of the tunnel based on the overall spatial structure data of the tunnel and the collected lining 3D information. It accurately maps the location, shape, type and warning level of defects obtained from the fusion detection into the model, enabling maintenance personnel to intuitively view the spatial distribution and detailed features of defects from different angles through 3D rotation, scaling and sectioning operations. The model can realistically restore the actual state of the tunnel lining, providing an intuitive basis for defect assessment.

[0038] Subsequently, the module generates a visualized inspection report. This report integrates 3D morphological diagrams of the defects, risk level annotations, and early warning and handling suggestions, transforming complex inspection data and analysis results into clear and easy-to-understand text and image information, directly providing maintenance personnel with specific handling directions. Simultaneously, the terminal output module uses a combination of local storage and encrypted cloud storage to save data. Local storage ensures access to inspection data even without a network connection, while encrypted cloud storage enables secure data backup and multi-terminal sharing. Stored data can be retrieved by tunnel station number, inspection time, and defect type, facilitating maintenance personnel to trace historical inspection data and track defect trends.

[0039] In summary, the multimodal data acquisition module, through the collaborative operation of two units, achieves simultaneous acquisition of detection data from the lining surface and interior, providing a data basis for subsequent analysis; the seepage environment point cloud correction module, through a combination of proprietary formulas, effectively eliminates the interference of the seepage environment on the data, ensuring data accuracy; the cross-modal visualization fusion module realizes the visualization transformation of stress wave signals and the deep fusion of dual-source data, combining attention mechanisms and dual-modal collaborative rules to ensure accurate identification of different types of defects; the dynamic early warning module, through flexible threshold updates and hierarchical early warning mechanisms, adapts to different detection conditions, achieving accurate assessment and timely early warning of defect risks; the terminal output module's visual presentation and dual-storage architecture make the detection results intuitive and easy to understand, and the data secure and traceable. The entire system's modules are logically interconnected and functionally complementary, achieving efficient and accurate detection of defects in tunnel lining structures.

[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent detection system for defects in highway tunnel lining structures based on laser point clouds, characterized in that, The system includes: Multimodal data acquisition module: used to simultaneously acquire three-dimensional laser point cloud data and internal stress wave signals of highway tunnel lining structure to provide dual-source detection data support; Point cloud correction module for seepage environment: Corrects the original laser point cloud data under seepage environment by using the reflectivity coupling correction formula of seepage environment and the point cloud normal vector linkage compensation formula of seepage environment. Cross-modal visualization fusion module: By adopting the stress wave amplitude point cloud visual trajectory feature mapping formula, the stress wave signal is transformed into three-dimensional visual trajectory features and registered and fused with the corrected laser point cloud data. The attention mechanism is used to focus on suspected disease areas and the dual-modal collaborative rules are used to identify disease types. Dynamic early warning module: By adopting a speed-linked dynamic threshold update formula, it dynamically updates the visual feature threshold and generates hierarchical early warning information; Terminal output module: used to map the fused detection results and early warning information to the tunnel digital twin model to generate a visual detection report.

2. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The multimodal data acquisition module includes a laser point cloud acquisition unit and a stress wave detection unit. The laser point cloud acquisition unit uses a vehicle-mounted laser scanner with an adjustable scanning frequency of 50~100Hz. It calibrates its own position through a GPS and IMU combined positioning module and acquires the three-dimensional coordinates, reflectivity, and initial value of the normal vector of the lining structure in real time. The stress wave detection unit uses a combination of a piezoelectric stress wave generator and an array receiver. The generator emits stress wave signals by attaching to the lining surface with silicone coupling agent. The array receiver receives stress wave signals in a one-to-many mode and simultaneously acquires the propagation time, amplitude, and attenuation coefficient of the stress waves inside the lining.

3. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The formula for coupling correction of reflectivity in the infiltration environment is: ,in, To correct the reflectance of the point cloud, The original point cloud reflectance, Here, S is the material adaptive correction coefficient, and S is the water permeability strength grade. To detect the actual humidity in the area, The reference humidity is used.

4. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The formula for the linkage compensation of the normal vector of the infiltration environment point cloud is as follows: ,in, For the compensated normal vector, The original normal vector, The correlation coefficient between the normal vector and reflectivity. This represents the absolute value of the reflectance correction difference.

5. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The formula for mapping the visual trajectory features of the stress wave amplitude point cloud is: ,in, The grayscale value of the visual trajectory color. β is an adjustment parameter. The amplitude of the stress wave. This represents the point cloud density for the corresponding region. This is the stress wave attenuation coefficient.

6. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The cross-modal visualization fusion module employs an elastic registration algorithm based on the lining reference profile to register the stress wave visual trajectory with the corrected laser point cloud data. The registration process involves fitting the lining reference profile using the RANSAC algorithm, calculating the coordinate deviation between the stress wave visual trajectory and the corrected point cloud, and then using a thin-plate spline interpolation algorithm for elastic adjustment to ensure precise alignment. The registration accuracy is ≤ ±2mm. An attention mechanism identifies suspected defect areas by analyzing the synergistic characteristics of point cloud normal vector mutations and stress wave trajectory distortions. The dual-modal feature weights for these areas are increased to 70%, while the feature weights for non-critical areas are reduced to 30% to suppress interference signals. The dual-modal synergistic rules are: surface cracks correspond to laser point cloud normal vector mutations ≥ 0.3 rad and continuous stress wave trajectories with slight amplitude attenuation; surface spalling corresponds to laser point cloud curvature anomalies ≥ 0.05mm. -1 Furthermore, the stress wave trajectory showed no obvious distortion, the internal voids corresponded to no significant abnormality in the curvature of the laser point cloud, and the stress wave trajectory was broken with an amplitude of <2V. The grouting was not dense, which corresponded to poor uniformity of the laser point cloud reflectivity and discontinuous stress wave trajectory with an amplitude fluctuating between 2 and 3V.

7. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 6, characterized in that, The attention mechanism suppresses interference signals including noise points generated by dust, water accumulation, and pipeline obstruction in the tunnel. By strengthening the weight of bimodal features in suspected defect areas, the influence of interference signals on defect identification is filtered.

8. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The formula for updating the speed-linked dynamic threshold is: ,in, For real-time dynamic thresholds, As the initial threshold, For adjustment coefficients, This refers to the actual scanning speed. As the baseline scan speed, The variance of the current sliding window features. This represents the historical average variance.

9. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The dynamic early warning module's hierarchical early warning mechanism includes a Level 1 early warning and a Level 2 early warning. A Level 1 early warning occurs when a single visual feature exceeds the dynamic threshold, and is only pushed to the backend operation and maintenance system, generating a suspected defect marker. A Level 2 early warning occurs when two or more visual features simultaneously exceed the dynamic threshold, simultaneously triggering an audible and visual alarm on the vehicle terminal, a pop-up notification in the backend system, and information push notifications to the operation and maintenance personnel's mobile devices. The threshold update cycle of the dynamic early warning module is linked to the laser scanning speed. When the scanning speed is ≥5km / h, the update cycle is set to 10s, and when the scanning speed is <5km / h, the update cycle is set to 5s.

10. The intelligent detection system for highway tunnel lining structure defects based on laser point clouds according to claim 1, characterized in that, The visual inspection report generated by the terminal output module includes a 3D morphological diagram of the defect, risk level markings, and early warning and handling suggestions. The accuracy of the tunnel digital twin model is ≤ ±5mm from the actual tunnel. It supports 3D rotation, scaling, and cross-sectional viewing. The terminal output module uses a combination of local storage and cloud encrypted storage to save data. Local storage data is retained for no less than 1 year, while cloud storage supports data retrieval by tunnel station number, inspection time, and defect type. It also supports data export and remote push.

Images