Subway shield tunnel vibration detection method based on hilbert curve point cloud serialization

Abstract

The application discloses a subway shield tunnel vibration detection method based on Hilbert curve point cloud serialization, relates to the technical field of tunnel vibration, and comprises a tunnel vibration detection system, which is used for collecting three-dimensional point clouds and synchronous image data of the inner wall of a tunnel. The tunnel data acquisition module and the data processing terminal are arranged, a perfect tunnel vibration detection system is constructed, the tunnel vibration detection system is used for collecting three-dimensional point clouds and synchronous image data of the inner wall of a tunnel, denoising, downsampling, coordinate normalization and splicing are performed on original point clouds, a parallel acceleration mode is adopted to perform Hilbert encoding and serialization on large-scale point cloud data, each link is monitored and processed in real time, tunnel vibration detection data and corresponding analysis results are managed, visualized and stored, and tunnel vibration management is realized through internet cloud management and control, so that the intelligent level of tunnel vibration management is improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel vibration technology, specifically a method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization. Background Technology

[0002] During long-term operation, subway shield tunnels are susceptible to structural vibration anomalies and deformation defects such as segment misalignment, track bed voids, and circumferential joint opening due to factors such as train vibration, ground settlement, and changes in groundwater. Timely and accurate detection of tunnel vibration is of great significance for ensuring operational safety.

[0003] Currently, tunnel vibration detection mainly relies on methods such as manual inspection, total station measurement, accelerometer placement, and 3D laser scanning. Manual inspection is highly subjective, inefficient, and difficult to quantify; point sensors can only acquire vibration information at limited locations and cannot form continuous cross-sectional detection; 3D laser scanning can acquire high-precision point clouds of the tunnel inner wall, but the original point cloud data volume is enormous (often reaching millions to tens of millions of points in a single interval), resulting in high computational costs for direct processing. Furthermore, 3D point clouds lack an inherent sequential structure, making it difficult to directly input them into conventional sequence models (such as recurrent neural networks and one-dimensional convolutional neural networks) for vibration feature extraction. Existing point cloud-based tunnel deformation analysis methods mostly employ cross-section fitting or triangular mesh comparison methods, projecting the point cloud onto a two-dimensional cross-section for analysis. However, these methods disrupt the spatial continuity of the point cloud, ignore local neighborhood relationships in the circumferential and longitudinal directions, are insensitive to minor vibrations and early anomalies, and cannot output intuitive vibration probability maps and anomaly masks.

[0004] To address the aforementioned issues, some studies have attempted to serialize point clouds using space-filling curves (such as Z-order curves and Hilbert curves) to preserve spatial proximity. However, most existing serialization methods employ fixed order and uniform grid division, making it difficult to adapt to different tunnel diameters, point cloud densities, and detection accuracy requirements. When encountering sparse point cloud regions, simple void discarding can lead to discontinuous sequences, affecting model recognition performance. Furthermore, the high computational latency of the serialization process for large-scale point cloud data cannot meet the needs of rapid on-site detection, resulting in numerous inconveniences.

[0005] Therefore, this invention requires the design of a vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization to solve the aforementioned problems. Summary of the Invention The purpose of this invention is to provide a method for efficiently and adaptively serializing tunnel point cloud data and integrating deep learning models to achieve precise vibration anomaly detection in order to solve the above-mentioned problems. It also supports parallel acceleration and dynamic parameter optimization to improve the automation, accuracy and real-time performance of vibration detection in subway shield tunnels.

[0006] To address the above problems, the present invention provides a technical solution: The vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization includes the following specific steps: S1. Construct a tunnel vibration detection system. The tunnel vibration detection system is used to collect three-dimensional point cloud and synchronous image data of the tunnel inner wall, and to denoise, downsample, normalize coordinates and stitch the original point cloud to generate a standardized tunnel cross-section point cloud. At the same time, the time, location, equipment parameters and environmental information of each collection are recorded to form a traceable data metadata tag. S2. Based on the Hilbert curve, the standardized point cloud data is spatially filled and serialized to convert the three-dimensional point cloud into one-dimensional sequence data, retaining local neighborhood relationships and spatial correlations, and adding a preset single or multiple combination algorithms to generate a vibration detection model; the vibration detection model is used as input to the one-dimensional point cloud sequence after Hilbert serialization, and output to the vibration amplitude, frequency and anomaly classification label at each location in the tunnel. S3. Input the serialized one-dimensional data into the vibration detection model, extract the deformation and vibration characteristics of the tunnel cross section, identify the vibration anomaly area, output the vibration probability map, anomaly area mask and quantification index, and trigger a graded early warning; based on the detection result feedback, dynamically adjust the Hilbert curve order or coordinate mapping strategy, and perform the Hilbert encoding and serialization process in parallel acceleration mode for large-scale point cloud data.

[0007] In a preferred embodiment of the present invention, the specific process of point cloud serialization based on Hilbert curves in step S2 includes: S201. Divide the tunnel point cloud space into a uniform grid, with the grid size corresponding to the minimum resolution of the Hilbert curve; S202. Calculate the centroid coordinates of the point cloud in each grid and use them as the representative point of that grid. If there is no point cloud data in a certain grid, fill in the virtual points using the bilinear interpolation method to ensure the continuity and integrity of the sequence. S203. Traverse all grids according to the spatial filling path of the Hilbert curve, and output the coordinates of representative points and their corresponding original point cloud sets in sequence to form a one-dimensional sequence.

[0008] In a preferred embodiment of the present invention, the order of the Hilbert curve in step S2 is adaptively determined according to the tunnel diameter, point cloud density and detection task requirements. Specifically, the overall bounding box size of the point cloud and the desired serialization resolution are calculated, and the optimal order that prevents the space-filling curve from aliasing is selected. The order range is from 4 to 10, and dynamic adjustment is supported during the detection process.

[0009] In a preferred embodiment of the present invention, step S3 uses a parallel acceleration method to perform the Hilbert encoding and serialization process on large-scale point cloud data. Specifically, a GPU or distributed computing framework is used to divide the point cloud space into multiple sub-regions, calculate the Hilbert index in parallel for each sub-region, and then merge and sort the results to form a complete one-dimensional sequence, thereby reducing serialization latency.

[0010] As a preferred embodiment of the present invention, the method for dynamically adjusting the order of the Hilbert curve or the coordinate mapping strategy based on the detection result feedback in step S3 is as follows: the abnormal region density, confidence level and calculation time output by the vibration detection model are used as feedback indicators, and the order and mapping parameters are automatically adjusted by reinforcement learning or Bayesian optimization strategy to balance the local spatial preservation after serialization and the overall detection efficiency.

[0011] In a preferred embodiment of the present invention, the tunnel vibration detection system includes a tunnel data acquisition module, a vibration detection model, and a data processing terminal. The tunnel data acquisition module is used to collect the original point cloud and image data, the vibration detection model is used to perform feature extraction and anomaly detection, and the data processing terminal is used to display results and issue control commands.

[0012] In a preferred embodiment of the present invention, the tunnel data acquisition module includes a tunnel data acquisition unit, an image data preprocessing unit, and a data source information unit. The output end of the tunnel data acquisition unit is communicatively connected to the input end of the image data preprocessing unit, and the data source information unit is integrated inside the tunnel data acquisition unit. The tunnel data acquisition unit is used to acquire three-dimensional point cloud and synchronous image data of the tunnel inner wall through a laser scanner, structured light camera or depth camera; The image data preprocessing unit is used to denoise, downsample, normalize coordinates and stitch together the original point cloud to generate a standardized tunnel cross-section point cloud. The data source information unit is used to record the time, location, equipment parameters, and environmental information of each collection, forming a traceable data metadata tag.

[0013] In a preferred embodiment of the present invention, the vibration detection model includes a model data center, a model optimization unit, and a model analysis unit. The output end of the model optimization unit is communicatively connected to the input end of the model data center, and the model optimization unit and the model analysis unit are bidirectionally communicatively connected. The vibration detection model is a deep learning model of recurrent neural network. The input is a one-dimensional point cloud sequence after Hilbert serialization, and the output is the vibration amplitude, frequency and anomaly classification label at each location in the tunnel. The model data center is used to store the serialized point cloud sequence, historical vibration detection results, and corresponding label data. The model optimization unit is used to train and update the parameters of the deep learning model based on contrastive learning or self-supervised learning strategies to improve the vibration recognition accuracy. The model analysis unit is used to receive real-time serialized data, run the trained model, and output vibration probability maps, abnormal region masks, and quantification indicators.

[0014] In a preferred embodiment of the present invention, the data processing terminal includes a data processing center, a mobile control terminal, and a mobile backup database. The data processing center and the mobile backup database are bidirectionally connected, and the mobile control terminal and the mobile backup database are also bidirectionally connected. The data processing center is used to integrate the detection results of multiple sections, generate an overall vibration distribution map of the tunnel, and trigger graded early warnings; The mobile control terminal is used by on-site personnel to view the test results in real time, control the start and stop of the acquisition equipment, and adjust the Hilbert serialization parameters. The mobile backup database is used to locally store all point cloud sequences, intermediate model outputs, and early warning records in a network-free environment, and supports data synchronization after network recovery.

[0015] In a preferred embodiment of the present invention, the tunnel vibration detection system further includes an algorithm supplementation module, which incorporates the following algorithms: Hilbert curve order adaptive selection algorithm: Based on the tunnel diameter, point cloud density and detection task requirements, it automatically selects the optimal order that prevents aliasing of the space-filling curve by calculating the overall bounding box size of the point cloud and the desired serialization resolution, and supports dynamic adjustment. Dynamic optimization algorithm: Based on vibration detection feedback, automatically adjust the order of Hilbert curves or coordinate mapping strategy to balance the preservation of local space after serialization and computational efficiency; Parallelization acceleration algorithm: For large-scale point cloud data, GPUs or distributed computing frameworks are used to process the Hilbert encoding and serialization process in parallel to reduce serialization latency.

[0016] The beneficial effects of this invention are as follows: This invention constructs a complete tunnel vibration detection system by setting up a tunnel data acquisition module, a vibration detection model, and a data processing terminal. The tunnel vibration detection system is used to collect three-dimensional point cloud data and synchronous image data of the tunnel inner wall, and to perform denoising, downsampling, coordinate normalization, and stitching on the original point cloud to generate a standardized tunnel cross-section point cloud. Simultaneously, it records the time, location, equipment parameters, and environmental information of each acquisition, forming traceable data metadata tags. Based on the Hilbert curve, the standardized point cloud data is spatially filled and serialized, converting the three-dimensional point cloud into one-dimensional sequence data, preserving local neighborhood relationships and spatial correlations, and incorporating preset single or multiple combination algorithms to generate a vibration detection model. The vibration detection model is used as input to Hilbert curves. The one-dimensional point cloud sequence after Hilbert serialization outputs vibration amplitude, frequency, and anomaly classification labels at various locations in the tunnel. The serialized one-dimensional data is then input into the vibration detection model to extract deformation and vibration characteristics of the tunnel cross-section, identify abnormal vibration areas, and output a vibration probability map, anomaly area mask, and quantification indicators, triggering tiered early warnings. Based on the detection results, the order of the Hilbert curve or coordinate mapping strategy is dynamically adjusted, and the Hilbert encoding and serialization process is executed in parallel to accelerate the processing of large-scale point cloud data. This facilitates real-time monitoring and processing of each step, enabling the management, visualization, and storage of tunnel vibration detection data and corresponding analysis results. This facilitates tunnel vibration management through internet cloud control, improving the level of intelligence in tunnel vibration management. Attached Figure Description For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0017] Figure 1 This is a flowchart of the overall process of the vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization of the present invention. Figure 2 This is a flowchart of step S2 in the vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization of the present invention. Figure 3 This is a topological diagram of the overall tunnel vibration detection system structure of the subway shield tunnel vibration detection method based on Hilbert curve point cloud serialization according to the present invention. Detailed Implementation like Figure 1 , Figure 2 and Figure 3 As shown, the specific implementation adopts the following technical solution: The vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization includes the following specific steps: S1. Construct a tunnel vibration detection system. The tunnel vibration detection system is used to collect three-dimensional point cloud and synchronous image data of the tunnel inner wall, and to denoise, downsample, normalize coordinates and stitch the original point cloud to generate a standardized tunnel cross-section point cloud. At the same time, the time, location, equipment parameters and environmental information of each collection are recorded to form a traceable data metadata tag. S2. Based on the Hilbert curve, the standardized point cloud data is spatially filled and serialized to convert the three-dimensional point cloud into one-dimensional sequence data, retaining local neighborhood relationships and spatial correlations, and adding a preset single or multiple combination algorithms to generate a vibration detection model; the vibration detection model is used as input to the one-dimensional point cloud sequence after Hilbert serialization, and output to the vibration amplitude, frequency and anomaly classification label at each location in the tunnel. The specific process of point cloud serialization based on Hilbert curves includes: S201. Divide the tunnel point cloud space into a uniform grid, with the grid size corresponding to the minimum resolution of the Hilbert curve; S202. Calculate the centroid coordinates of the point cloud in each grid and use them as the representative point of that grid. If there is no point cloud data in a certain grid, fill in the virtual points using the bilinear interpolation method to ensure the continuity and integrity of the sequence. S203. Traverse all grids according to the spatial filling path of the Hilbert curve, and output the coordinates of the representative points and their corresponding original point cloud sets in sequence to form a one-dimensional sequence. The order of the Hilbert curve is adaptively determined based on the tunnel diameter, point cloud density, and detection task requirements. Specifically, the overall bounding box size of the point cloud and the desired serialization resolution are calculated, and the optimal order that prevents the space-filling curve from aliasing is selected. The order range is from 4 to 10, and dynamic adjustment is supported during the detection process. S3. Input the serialized one-dimensional data into the vibration detection model, extract the deformation and vibration characteristics of the tunnel cross section, identify the vibration anomaly area, output the vibration probability map, anomaly area mask and quantification index, and trigger a graded early warning; based on the detection result feedback, dynamically adjust the Hilbert curve order or coordinate mapping strategy, and perform the Hilbert encoding and serialization process in parallel acceleration mode for large-scale point cloud data. For large-scale point cloud data, a parallel acceleration method is used to perform the Hilbert encoding and serialization process. Specifically, GPUs or distributed computing frameworks are used to divide the point cloud space into multiple sub-regions, calculate the Hilbert index in parallel for each sub-region, and then merge and sort the results to form a complete one-dimensional sequence, thereby reducing serialization latency. The method for dynamically adjusting the order of the Hilbert curve or the coordinate mapping strategy based on the feedback of the detection results is as follows: the abnormal area density, confidence level and computation time output by the vibration detection model are used as feedback indicators, and the order and mapping parameters are automatically adjusted by reinforcement learning or Bayesian optimization strategies to balance the local spatial preservation after serialization and the overall detection efficiency.

[0018] The tunnel vibration detection system includes a tunnel data acquisition module, a vibration detection model, and a data processing terminal. The tunnel data acquisition module is used to collect the original point cloud and image data, the vibration detection model is used to perform feature extraction and anomaly detection, and the data processing terminal is used to display results and issue control commands.

[0019] The tunnel data acquisition module includes a tunnel data acquisition unit, an image data preprocessing unit, and a data source information unit. The output of the tunnel data acquisition unit is communicatively connected to the input of the image data preprocessing unit, and the data source information unit is integrated within the tunnel data acquisition unit. The tunnel data acquisition unit is used to acquire three-dimensional point clouds and synchronous image data of the tunnel inner wall using a laser scanner, structured light camera, or depth camera. The image data preprocessing unit is used to denoise, downsample, normalize coordinates, and stitch the original point cloud to generate a standardized tunnel cross-section point cloud. The data source information unit is used to record the time, location, equipment parameters, and environmental information of each acquisition, forming a traceable data metadata tag.

[0020] The vibration detection model includes a model data center, a model optimization unit, and a model analysis unit. The output of the model optimization unit is communicatively connected to the input of the model data center, and the model optimization unit and the model analysis unit are bidirectionally connected. The vibration detection model is a deep learning model based on a recurrent neural network. The input is a one-dimensional point cloud sequence serialized by Hilbert, and the output is the vibration amplitude, frequency, and anomaly classification labels at various locations in the tunnel. The model data center is used to store the serialized point cloud sequence, historical vibration detection results, and corresponding label data. The model optimization unit is used to train and update the parameters of the deep learning model based on contrastive learning or self-supervised learning strategies to improve the vibration recognition accuracy. The model analysis unit is used to receive real-time serialized data, run the trained model, and output a vibration probability map, anomaly area mask, and quantification indicators.

[0021] The data processing terminal includes a data processing center, a mobile control terminal, and a mobile backup database. The data processing center and the mobile backup database are bidirectionally connected. The mobile control terminal and the mobile backup database are also bidirectionally connected. The data processing center is used to fuse multi-segment detection results, generate an overall vibration distribution map of the tunnel, and trigger graded early warnings. The mobile control terminal is used by on-site personnel to view detection results in real time, control the start and stop of acquisition equipment, and adjust Hilbert serialization parameters. The mobile backup database is used to locally store all point cloud sequences, intermediate model outputs, and early warning records in a network-free environment, and supports data synchronization after network recovery.

[0022] The tunnel vibration detection system also includes an algorithm supplement module, which incorporates the following algorithms: an adaptive Hilbert curve order selection algorithm: based on the tunnel diameter, point cloud density, and detection task requirements, it automatically selects the optimal order that prevents aliasing of the space-filling curve by calculating the overall bounding box size of the point cloud and the desired serialization resolution, and supports dynamic adjustment; a dynamic optimization algorithm: based on vibration detection result feedback, it automatically adjusts the Hilbert curve order or coordinate mapping strategy to balance the preservation of local space after serialization and computational efficiency; and a parallel acceleration algorithm: for large-scale point cloud data, it uses GPUs or distributed computing frameworks to process the Hilbert encoding and serialization process in parallel, reducing serialization latency.

[0023] Example 1: Medium-density point cloud and standard tunnel detection (fixed Hilbert order) 1. Construction and Data Acquisition of Tunnel Vibration Detection System Tunnel data acquisition unit: Using a Z+FPROFILER9012 laser scanner and an industrial panoramic camera, the unit acquired 3D point cloud and synchronous panoramic images of the inner wall of a subway shield tunnel section (inner diameter 5.8m, inspection length 120m, segment ring width 1.2m, total 100 rings). The scanning frequency was 200Hz, and the original point cloud contained approximately 2.85 million points.

[0024] Image data preprocessing unit: Performs the following steps sequentially on the raw point cloud: Noise reduction: Statistical filtering (k=50, standard deviation factor=1.2), removing approximately 120,000 outliers; Downsampling: Voxel filtering (voxel side length = 0.02m), retaining approximately 1.46 million points; Coordinate normalization: After fitting the tunnel axis, transform it to cylindrical coordinate system (mileage s, circumferential angle θ, radius r), and normalize s∈[0,120], θ∈[0,2π), r∈[5.75,5.85] to the interval [0,1]. Stitching: Generate a standardized tunnel cross-section point cloud.

[0025] Data source information unit: Automatically records the collection time (2024-05-17 22:34:12), start and end mileage (K10+223-K10+343), device parameters (firmware version v2.1.4), and environmental information (temperature 18℃, humidity 72%), forming a traceable data metadata tag.

[0026] 2. Serialization and vibration detection model based on Hilbert curves Model Data Center: Stores historical tunnel point cloud sequences and corresponding labels. The original coordinates and preprocessing parameters before this serialization are temporarily stored here.

[0027] Hilbert curve serialization (executed by the parallelized accelerated algorithm in the algorithm supplementation module, called by the model analysis unit): Mesh generation: Based on the normalized bounding box (1×1×1), the mesh side length is set to 0.025 (corresponding to an actual mileage of 3m, a circumferential angle of 9°, and a radial angle of 0.003m), and the total number of meshes is 40×40×40=64000.

[0028] Empty grid processing: A total of 4183 empty grids (accounting for 6.5%) were detected. Virtual points were filled using bilinear interpolation (in the s-θ plane), and the radius of the virtual points was the weighted average of the radii of adjacent non-empty grids.

[0029] Hilbert order selection: Based on the tunnel diameter (5.8m), point cloud density (1.46 million points) and detection task requirements, order=6 (2^6=64≥40) is adaptively determined according to the formula 2^order≥max(grid_dim), with an order range of 4-10.

[0030] Serialization output: One-dimensional sequence length 64000, each element is a grid representative point or interpolation point.

[0031] Model optimization unit: Since this embodiment uses a standard tunnel section, a contrastive learning pre-trained model is adopted, and online updates are not triggered in this instance; existing model parameters are used directly for inference.

[0032] 3. Detection, early warning and dynamic feedback Model Analysis Unit: Receives real-time serialized data (1D sequence length 64000, feature dimension 4), runs a 1D-CNN+Transformer model, and outputs: Vibration probability map: Mileage K10+280-K10+295, circumferential angle 90°-150° region, maximum vibration amplitude 3.2mm (threshold 2.0mm), dominant frequency 8.2Hz (normal range 0.5-5Hz). Anomaly classification label: Moderate segment misalignment + track bed void; Abnormal region mask: 47 abnormal grids (0.73%), stored in JSON format.

[0033] Data Processing Center: Integrates the detection results of multiple sections to generate an overall vibration distribution map of the tunnel and triggers a Level 3 yellow alert (re-inspection is required within 48 hours).

[0034] Mobile control terminal: On-site personnel can view the test results and control the start and stop of the data acquisition equipment in real time via tablet. Hilbert parameters were not adjusted in this test.

[0035] Mobile backup database: In the absence of network conditions (weak 4G signal at the tunnel exit), all point cloud sequences, intermediate model outputs (feature maps), and early warning records are automatically stored on a local 1TB solid-state drive and subsequently transmitted back to the ground center.

[0036] Parallelization acceleration algorithm: Using NVIDIA A10 GPU, the point cloud space is divided into 8 sub-regions (each region is 15m long) according to claim 4. The Hilbert index of each sub-region is calculated in parallel and then merged and sorted. The total serialization time is reduced from 124 seconds on a single CPU to 4.7 seconds.

[0037] Dynamic optimization algorithm: In this embodiment, the order adjustment was not triggered because the detection confidence was higher than 85% and the density of abnormal regions was only 0.73%. However, the system still recorded feedback indicators (abnormal region density, confidence, and computation time) to accumulate experience for subsequent reinforcement learning strategies.

[0038] Example 2: High-density point cloud and complex deformation section tunnel detection (dynamic order adjustment and feedback optimization) 1. Construction and Data Acquisition of Tunnel Vibration Detection System Tunnel data acquisition unit: Using a FAROFocus S350 ground laser scanner (with IMU positioning), with a resolution of 1 / 4 (point spacing 3.1mm@10m), 3D point cloud and synchronous images of a shield tunnel section in a soft stratum (inner diameter 6.2m, detection length 50m, with obvious settlement section K12+005-K12+030) were acquired.

[0039] Image data preprocessing unit: The original point cloud contains approximately 12.6 million points, which are filtered by voxel (side length 0.005m) to retain 4.1 million points, and the coordinates are normalized to [0,1]³.

[0040] Data source information unit: Record the collection time (2024-08-22 01:15:07), start and end mileage (K12+000-K12+050), groundwater level (-2.3m), grouting pressure (0.24MPa), and form metadata tags.

[0041] 2. Dynamic Hilbert curve serialization and vibration detection model Model Data Center: Stores the historical point cloud sequence of this segment and the label of the previous period (including normal segment data from K11+900 to K12+000).

[0042] Model analysis unit: For the first run, order=8 was used (automatically calculated, 2^8 = 256 grids / dimensional, total grid 16.7M). After serialization, the input model was evaluated, and the feedback was as follows: Anomaly density: 21% Confidence level: 62% (below 70%) Calculation time: 187 seconds (over 120 seconds) Model optimization unit: Writes the above feedback metrics into the model data center and calls the dynamic optimization algorithm (reinforcement learning strategy) in the algorithm supplement module to automatically adjust the Hilbert parameters.

[0043] Dynamic optimization algorithm execution: Bayesian optimization search space: order (5-9), coordinate mapping strategy (uniform / circular encryption / radial encryption).

[0044] Recommended parameters: order=7 (2^7=128 grids / dimensional, total grid 2.1M), which increases the sampling rate by 2 times within a circumferential angle of ±10°.

[0045] Reserialization (the parallelization acceleration algorithm in the algorithm supplement module also participates): Grid division: 128×128×128=2,097,152 grids.

[0046] Empty grid processing: The percentage of empty grids increased from 3.2% to 8.7%, and was filled using radial basis function interpolation (RBF) with a smoothing factor of 0.01.

[0047] Example of serialized output (near K12+018 of the settlement section).

[0048] 3. Optimized detection, early warning, and data storage Second run of the model analysis unit: Output vibration amplitude: maximum 4.7mm (K12+020, circumferential angle 95°), frequency 18.4Hz (high frequency abnormality, determined to be segment circumferential gap opening and impact vibration).

[0049] Anomaly Classification: Severity Level I, triggering a Level II Orange Alert (immediate shutdown and handling required).

[0050] Improved quantitative indicators: Compared with the measured displacement gauge, the recall rate in abnormal areas increased from 68% (order=8) to 93% (order=7 + optimized mapping), and the false alarm rate decreased from 15% to 6%.

[0051] Data Processing Center: Integrates the detection results of neighboring sections, overlays the vibration probability map on the tunnel BIM model, and sends an emergency shutdown suggestion to the subway dispatch center.

[0052] Mobile control terminal: On-site engineers receive the orange alert via an explosion-proof mobile phone. After clicking "Confirm", the data collection device is automatically paused and the new parameters recommended by the algorithm (order=7, circumferential encryption) are saved as the default configuration for the future.

[0053] Mobile backup database: There is absolutely no 4G / 5G signal inside the tunnel. This module will: Original point cloud sequence (2.1GB after compression) Intermediate output of the model (approximately 420MB of feature maps per layer) Warning records and mask JSON (87KB) All data was stored on a portable solid-state drive (1TB, ExFAT format). After network recovery, the data was automatically resumed from where it left off, with 100% data integrity.

[0054] The algorithm is accelerated by parallelization: using a distributed Spark framework (4 worker nodes), the total time for Hilbert encoding and serialization is reduced from 203 seconds on a single machine to 19 seconds, meeting the requirements of GPU or distributed computing frameworks.

[0055] Specifically, in practical applications, multiple distributed acquisition devices are used in conjunction with a PLC controller, cloud management terminal, monitoring and data center. These distributed acquisition devices are located in different geographical locations. This invention constructs a complete tunnel vibration detection system by setting up a tunnel data acquisition module, a vibration detection model, and a data processing terminal. The tunnel vibration detection system is used to acquire three-dimensional point cloud and synchronous image data of the tunnel inner wall, and to denoise, downsample, normalize coordinates, and stitch the original point cloud to generate a standardized tunnel cross-section point cloud. At the same time, it records the time, location, equipment parameters, and environmental information of each acquisition, forming a traceable data metadata tag. Based on the Hilbert curve, the standardized point cloud data is spatially filled and serialized, converting the three-dimensional point cloud into one-dimensional sequence data, preserving local neighborhood relationships and spatial correlations. The system incorporates one or more preset algorithms to generate a vibration detection model. The model takes a one-dimensional point cloud sequence serialized by Hilbert as input and outputs vibration amplitude, frequency, and anomaly classification labels for each location within the tunnel. The serialized one-dimensional data is input into the vibration detection model to extract deformation and vibration characteristics of the tunnel cross-section, identify abnormal vibration areas, output a vibration probability map, anomaly area mask, and quantification indicators, and trigger tiered early warnings. Based on the detection results, the Hilbert curve order or coordinate mapping strategy is dynamically adjusted, and the Hilbert encoding and serialization process is executed in parallel to accelerate the processing of large-scale point cloud data. This allows for the management, visualization, and storage of tunnel vibration detection data and corresponding analysis results, facilitating tunnel vibration management through internet cloud control and improving the intelligence level of tunnel vibration management.

[0056] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, equipment, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] In the embodiments provided in this application, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or equipment, and may be electrical, mechanical, or other forms.

[0059] The modules serving as tunnel data acquisition, vibration detection model, algorithm supplementation, and data processing terminal may or may not be physically separate. The components displayed as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0060] Furthermore, it should be noted that the combination of the various technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0061] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

[0062] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A vibration detection method for subway shield tunnels based on Hilbert curve point cloud serialization, characterized in that, The specific steps include the following: S1. Construct a tunnel vibration detection system. The tunnel vibration detection system is used to collect three-dimensional point cloud and synchronous image data of the tunnel inner wall, and to denoise, downsample, normalize coordinates and stitch the original point cloud to generate a standardized tunnel cross-section point cloud. At the same time, the time, location, equipment parameters and environmental information of each collection are recorded to form a traceable data metadata tag. S2. Based on the Hilbert curve, the standardized point cloud data is spatially filled and serialized to convert the three-dimensional point cloud into one-dimensional sequence data, retaining local neighborhood relationships and spatial correlations, and adding a preset single or multiple combination algorithms to generate a vibration detection model; the vibration detection model is used as input to the one-dimensional point cloud sequence after Hilbert serialization, and output to the vibration amplitude, frequency and anomaly classification label at each location in the tunnel. S3. Input the serialized one-dimensional data into the vibration detection model, extract the deformation and vibration characteristics of the tunnel cross section, identify the vibration anomaly area, output the vibration probability map, anomaly area mask and quantification index, and trigger a graded early warning; based on the detection result feedback, dynamically adjust the Hilbert curve order or coordinate mapping strategy, and perform the Hilbert encoding and serialization process in parallel acceleration mode for large-scale point cloud data.

2. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 1, characterized in that: The specific process of point cloud serialization based on Hilbert curves in step S2 includes: S201. Divide the tunnel point cloud space into a uniform grid, with the grid size corresponding to the minimum resolution of the Hilbert curve; S202. Calculate the centroid coordinates of the point cloud in each grid and use them as the representative point of that grid. If there is no point cloud data in a certain grid, fill in the virtual points using the bilinear interpolation method to ensure the continuity and integrity of the sequence. S203. Traverse all grids according to the spatial filling path of the Hilbert curve, and output the coordinates of representative points and their corresponding original point cloud sets in sequence to form a one-dimensional sequence.

3. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 1, characterized in that: In step S2, the order of the Hilbert curve is adaptively determined based on the tunnel diameter, point cloud density, and detection task requirements. Specifically, the overall bounding box size of the point cloud and the desired serialization resolution are calculated, and the optimal order that prevents the space-filling curve from aliasing is selected. The order range is from 4 to 10, and dynamic adjustment is supported during the detection process.

4. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 1, characterized in that: In step S3, the Hilbert encoding and serialization process for large-scale point cloud data is performed in parallel to accelerate the process. Specifically, a GPU or distributed computing framework is used to divide the point cloud space into multiple sub-regions, calculate the Hilbert index in parallel for each sub-region, and then merge and sort the results to form a complete one-dimensional sequence to reduce serialization latency.

5. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 1, characterized in that: The method for dynamically adjusting the order of the Hilbert curve or the coordinate mapping strategy based on the detection result feedback in step S3 is as follows: the abnormal region density, confidence level and calculation time output by the vibration detection model are used as feedback indicators, and the order and mapping parameters are automatically adjusted by reinforcement learning or Bayesian optimization strategy to balance the local spatial preservation after serialization and the overall detection efficiency.

6. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 5, characterized in that: The tunnel vibration detection system includes a tunnel data acquisition module, a vibration detection model, and a data processing terminal. The tunnel data acquisition module is used to collect the original point cloud and image data, the vibration detection model is used to perform feature extraction and anomaly detection, and the data processing terminal is used to display results and issue control commands.

7. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 6, characterized in that: The tunnel data acquisition module includes a tunnel data acquisition unit, an image data preprocessing unit, and a data source information unit. The output end of the tunnel data acquisition unit is communicatively connected to the input end of the image data preprocessing unit, and the data source information unit is integrated inside the tunnel data acquisition unit. The tunnel data acquisition unit is used to acquire three-dimensional point cloud and synchronous image data of the tunnel inner wall through a laser scanner, structured light camera or depth camera; The image data preprocessing unit is used to denoise, downsample, normalize coordinates and stitch together the original point cloud to generate a standardized tunnel cross-section point cloud. The data source information unit is used to record the time, location, equipment parameters, and environmental information of each collection, forming a traceable data metadata tag.

8. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 7, characterized in that: The vibration detection model includes a model data center, a model optimization unit, and a model analysis unit. The output of the model optimization unit is communicatively connected to the input of the model data center, and the model optimization unit and the model analysis unit are bidirectionally communicatively connected. The model data center is used to store the serialized point cloud sequence, historical vibration detection results, and corresponding label data. The model optimization unit is used to train and update the parameters of the deep learning model based on contrastive learning or self-supervised learning strategies to improve the vibration recognition accuracy. The model analysis unit is used to receive real-time serialized data, run the trained model, and output vibration probability maps, abnormal region masks, and quantification indicators.

9. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 8, characterized in that: The data processing terminal includes a data processing center, a mobile control terminal, and a mobile backup database. The data processing center and the mobile backup database are connected in a two-way communication connection, and the mobile control terminal and the mobile backup database are also connected in a two-way communication connection. The data processing center is used to integrate the detection results of multiple sections, generate an overall vibration distribution map of the tunnel, and trigger graded early warnings; The mobile control terminal is used by on-site personnel to view the test results in real time, control the start and stop of the acquisition equipment, and adjust the Hilbert serialization parameters. The mobile backup database is used to locally store all point cloud sequences, intermediate model outputs, and early warning records in a network-free environment.

10. The method for detecting vibration in subway shield tunnels based on Hilbert curve point cloud serialization according to claim 9, characterized in that: The tunnel vibration detection system also includes an algorithm supplement module, which incorporates a dynamic optimization algorithm and a parallel acceleration algorithm.

Images