Curve tunnel deformation monitoring data fusion processing method based on multi-station robot

Abstract

The application discloses a curve tunnel deformation monitoring data fusion processing method based on multi-station robots, and belongs to the technical field of tunnel measurement. The method comprises the following steps: acquiring tunnel original point cloud data collected by multiple mobile robots and corresponding initial pose data to construct a multi-station dynamic unified coordinate system; performing preliminary registration based on tunnel structure feature point sets in the extracted tunnel original point cloud data to generate preliminary fusion point cloud; performing weighted fusion on the preliminary fusion point cloud according to fusion error distribution parameters to generate final fusion point cloud; performing geometric modeling on the final fusion point cloud to generate a deformation monitoring reference model; comparing and analyzing the deformation monitoring reference model and a tunnel design model to obtain a deformation distribution graph; and generating a deformation trend prediction report. The method can realize high-precision three-dimensional modeling and dynamic deformation trend early warning of the curve tunnel by constructing a unified coordinate system, weightedly fusing multi-source point cloud and performing time sequence prediction.

Description

Technical Field

[0001] This invention relates to the field of tunnel measurement technology, and in particular to a method for data fusion and processing of deformation monitoring data of curved tunnels based on multi-station robots. Background Technology

[0002] As a critical infrastructure in transportation networks, the structural safety of curved tunnels is of paramount importance. Deformation monitoring is a core technological means to assess tunnel health and ensure operational safety, involving the precise measurement and analysis of changes in tunnel geometry over time and under external loads. Traditional monitoring mainly relies on the manual placement of monitoring points, but in recent years, the use of mobile robots equipped with laser scanners for automation and high-density 3D point cloud data acquisition has become a development trend.

[0003] In existing technologies, when using mobile robots for tunnel monitoring, a single robot typically moves back and forth within the tunnel, using Simultaneous Localization and Mapping (SLAM) technology to stitch together point cloud data collected at different locations to construct a 3D model of the tunnel. For long-distance or large curved tunnels, to improve efficiency, multiple robots are sometimes used to work in segments, and then the data from each segment is stitched together later. This approach mainly relies on geometric feature matching between point clouds or the robot's own odometry data to achieve data alignment.

[0004] However, the aforementioned existing technical solutions have significant drawbacks. Single-robot operations in long-distance curved tunnels inevitably generate significant cumulative drift errors, leading to a severe decrease in the global accuracy of the model and the appearance of bending or distortion. Furthermore, when multiple robots operate in segments, the lack of a unified, high-precision global coordinate reference makes it difficult to achieve accurate and seamless stitching between the independently generated local maps of each robot, especially in sections with sparse or repetitive features, where the stitching error is even greater. This directly affects the reliability of subsequent deformation analysis. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a data fusion processing method for curved tunnel deformation monitoring based on a multi-station robot. By constructing a unified coordinate system, weighted fusion of multi-source point clouds, and time series prediction, it is possible to achieve high-precision 3D modeling and dynamic deformation trend early warning for curved tunnels.

[0006] The above objectives can be achieved through the following approach:

[0007] A data fusion processing method for curve tunnel deformation monitoring based on multi-station robots includes: acquiring original tunnel point cloud data and corresponding initial pose data collected by multiple mobile robots; identifying environmental structural feature points in the original tunnel point cloud data, and constructing a multi-station dynamic unified coordinate system based on the initial pose data and the environmental structural feature points; extracting the tunnel structural feature point set from the original tunnel point cloud data, and performing preliminary registration of the original tunnel point cloud data based on the multi-station dynamic unified coordinate system to generate a preliminary fused point cloud; calculating the fusion error distribution parameters of the preliminary fused point cloud in the overlapping area, and performing weighted fusion of the preliminary fused point cloud according to the fusion error distribution parameters to generate a final fused point cloud; performing geometric modeling on the final fused point cloud to generate a deformation monitoring benchmark model; acquiring the tunnel design model, and comparing and analyzing the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map; extracting the current deformation parameters from the deformation distribution map for time series prediction, and generating a deformation trend prediction report.

[0008] Optionally, the construction of the multi-station dynamic unified coordinate system includes: acquiring a feature point library containing the designed locations of natural feature points of the tunnel; identifying environmental structural feature points in the original point cloud data of the tunnel, matching the environmental structural feature points with the feature point library, and calculating the relative pose correction amount; using the relative pose correction amount to correct the initial pose data and generate corrected pose data; and establishing a multi-station dynamic unified coordinate system based on the corrected pose data of all robots.

[0009] Optionally, the calculation of relative pose correction includes: acquiring image data of the tunnel inner wall and extracting environmental structural feature points from it, the environmental structural feature points including environmental corner features and edge features; matching the corner features and the edge features with the feature point library and calculating the feature matching confidence; determining whether the feature matching confidence meets the preset positioning requirements; if so, calculating the relative pose correction amount based on the corresponding corner features and the corresponding edge features.

[0010] Optionally, generating the preliminary fused point cloud includes: extracting the tunnel structure feature point set from the original tunnel point cloud data; distinguishing circumferential joint feature points and longitudinal joint feature points from the tunnel structure feature point set; calculating the similarity matrix between the circumferential joint feature points and the longitudinal joint feature points between different station cloud data; performing initial alignment based on the similarity matrix to generate an initial registration result; calculating the point cloud density distribution of the initial registration result in the overlapping area based on the multi-station dynamic unified coordinate system; adjusting the preset registration weights according to the point cloud density distribution and performing weighted optimization to generate the preliminary fused point cloud.

[0011] Optionally, the step of adjusting the preset registration weights based on the point cloud density distribution and performing weighted optimization to generate a preliminary fused point cloud includes: calculating the point cloud density difference value based on the point cloud density distribution; adjusting the preset registration weights based on the point cloud density difference value; establishing an objective function to characterize the registration error based on the point cloud density distribution and the adjusted registration weights, using the rotation matrix and translation as variables; solving the objective function with the preset registration error as the objective to obtain the final rotation matrix and the final translation; and generating a preliminary fused point cloud using the final rotation matrix and the final translation.

[0012] Optionally, generating the final fused point cloud includes: calculating the fusion error distribution parameters of the preliminary fused point cloud in the overlapping region, and extracting angle deviation values ​​and distance deviation values ​​from the fusion error distribution parameters; calculating an angle weight coefficient based on the angle deviation values, and calculating a distance weight coefficient based on the distance deviation values; fusing the angle weight coefficient and the distance weight coefficient to generate an adaptive weight matrix; and using the adaptive weight matrix to perform weighted interpolation processing on the overlapping region of the preliminary fused point cloud to generate the final fused point cloud.

[0013] Optionally, the generation of the deformation monitoring benchmark model includes: extracting cross-sectional contour points from the final fused point cloud; performing ellipse fitting on the cross-sectional contour points to obtain multiple cross-sectional fitted ellipses; extracting the center coordinates of the multiple cross-sectional fitted ellipses, and generating the tunnel centerline by fitting the center coordinates; and constructing the deformation monitoring benchmark model based on the tunnel centerline and the final fused point cloud.

[0014] Optionally, obtaining the deformation distribution map includes: acquiring a tunnel design model; spatially aligning the deformation monitoring benchmark model with the tunnel design model; calculating deformation parameters including radial displacement and convergent deformation; identifying abnormal deformation regions based on the deformation parameters; extracting the point cloud density and curvature features of the abnormal deformation regions; and generating a deformation distribution map.

[0015] Optionally, generating the deformation trend prediction report includes: obtaining historical deformation parameters and extracting current deformation parameters from the deformation distribution map; performing time series analysis based on the historical deformation parameters and the current deformation parameters to calculate the deformation rate trend change; comparing the deformation rate trend change with a preset safety threshold to generate an early warning signal; and marking key areas according to the early warning signal and generating a deformation trend prediction report.

[0016] Based on the same inventive concept, this invention also provides a data fusion processing system for curve tunnel deformation monitoring based on multi-station robots. The system includes: a data acquisition module for acquiring raw tunnel point cloud data and corresponding initial pose data collected by multiple mobile robots; a coordinate system construction module for identifying environmental structural feature points in the raw tunnel point cloud data and constructing a multi-station dynamic unified coordinate system based on the initial pose data and the environmental structural feature points; and a point cloud registration module for extracting a set of tunnel structural feature points from the raw tunnel point cloud data and performing preliminary registration of the raw tunnel point cloud data based on the multi-station dynamic unified coordinate system. The system generates a preliminary fused point cloud; a data fusion module calculates the fusion error distribution parameters of the preliminary fused point cloud in the overlapping area, and performs weighted fusion on the preliminary fused point cloud according to the fusion error distribution parameters to generate a final fused point cloud; a model generation module performs geometric modeling on the final fused point cloud to generate a deformation monitoring benchmark model; a deformation analysis module obtains the tunnel design model, compares and analyzes the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map; and a trend prediction module extracts the current deformation parameters from the deformation distribution map for time series prediction and generates a deformation trend prediction report.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. This invention fundamentally solves the problem of inconsistent spatial references caused by accumulated errors between data sources when multiple robots are working together by constructing a dynamic unified coordinate system for multiple stations and using the inherent environmental structural feature points of the tunnel for pose correction, thus ensuring the accuracy and consistency of global data fusion.

[0019] 2. This invention not only utilizes stable tunnel structure features to ensure the robustness of initial registration, but also performs differential processing on the data by analyzing point cloud density and fusion error distribution, effectively suppressing the influence of noise and uneven data quality, and significantly improving the geometric fidelity of the final fusion model.

[0020] 3. This invention, by establishing a time-series model of deformation and analyzing its rate of change, can provide early warning of potential risk areas where deformation is accelerating. This transforms tunnel safety management from a passive, delayed response to a proactive, forward-looking prevention approach, greatly enhancing the practical value and timeliness of deformation monitoring. This invention forms an automated, integrated processing workflow from raw data acquisition to final decision support. This workflow covers key aspects such as data registration and fusion, 3D modeling, deformation quantification analysis, and trend prediction, reducing manual intervention, improving processing efficiency and the objectivity of results, and providing a complete and efficient technical solution for the long-term health monitoring of curved tunnels.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the structure of the curve tunnel deformation monitoring data fusion processing method based on a multi-station robot according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the structure of the curve tunnel deformation monitoring data fusion processing system based on a multi-station robot according to an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram comparing the pose correction effects of an embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram of the deformation analysis data of a curved tunnel according to an embodiment of the present invention.

[0027] Figure 5 This is a schematic diagram of the deformation analysis of the tunnel K1+295 section according to an embodiment of the present invention.

[0028] Figure 6 This is a schematic diagram of deformation time series analysis and trend prediction according to an embodiment of the present invention.

[0029] Figure 7 This is a schematic diagram of the deformation trend prediction and early warning data of tunnel section K1+295 in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Reference Figure 1One embodiment of the present invention proposes a data fusion processing method for deformation monitoring of curved tunnels based on multi-station robots. By constructing a unified coordinate system, weighted fusion of multi-source point clouds, and time series prediction, it is possible to achieve high-precision three-dimensional modeling and dynamic deformation trend early warning of curved tunnels.

[0032] The method described in this embodiment specifically includes:

[0033] Acquire raw point cloud data of the tunnel and corresponding initial pose data collected by multiple mobile robots;

[0034] Identify environmental structural feature points in the original point cloud data of the tunnel, and construct a dynamic unified coordinate system for multiple stations based on the initial pose data and the environmental structural feature points;

[0035] Extract the tunnel structure feature point set from the original tunnel point cloud data, and perform preliminary registration of the original tunnel point cloud data based on the multi-station dynamic unified coordinate system to generate a preliminary fused point cloud;

[0036] Calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping region, and perform weighted fusion on the preliminary fused point cloud according to the fusion error distribution parameters to generate the final fused point cloud;

[0037] Geometric modeling is performed on the final fused point cloud to generate a deformation monitoring benchmark model;

[0038] Obtain the tunnel design model, and compare and analyze the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map;

[0039] The current deformation parameters are extracted from the deformation distribution map for time series prediction, and a deformation trend prediction report is generated.

[0040] Specifically, this invention overcomes the inherent data registration and fusion challenges in multi-robot collaborative operations by constructing a unified coordinate system and implementing weighted fusion, ensuring the integrity and high accuracy of the final generated 3D model. By comparing and analyzing the measured model with the design model, precise quantification and intuitive visualization of tunnel deformation are achieved, enabling engineers to clearly understand the location, shape, and extent of deformation. More importantly, by introducing time series prediction, this method elevates traditional static deformation monitoring to a new level of dynamic trend early warning, enabling the early identification of potential risk areas where deformation is accelerating. This provides a scientific and reliable basis for decision-making in the preventative maintenance and safety management of tunnels, transforming passive, reactive repair into proactive, predictive maintenance.

[0041] Optionally, the construction of a multi-station dynamic unified coordinate system includes:

[0042] Obtain a feature point library containing the designed locations of natural feature points in the tunnel;

[0043] Identify environmental structural feature points in the original point cloud data of the tunnel, match the environmental structural feature points with the feature point library, and calculate the relative pose correction amount;

[0044] The initial pose data is corrected using the relative pose correction amount to generate corrected pose data;

[0045] Based on the corrected pose data of all robots, a multi-station dynamic unified coordinate system is established.

[0046] Specifically, firstly, multiple mobile robots deployed within the curved tunnel perform data acquisition tasks. Each robot is equipped with sensors such as laser scanners to acquire raw point cloud data of the tunnel at its location. Simultaneously, the robot's internal navigation system, including its inertial measurement unit and wheeled odometer, records initial pose data, describing the robot's position and orientation in its own motion coordinate system. However, this data accumulates errors as the travel distance increases. To eliminate this accumulated error, this method introduces an external, stable reference benchmark. This benchmark is a pre-established feature point library, based on tunnel design drawings, containing fixed natural feature points within the tunnel, such as the three-dimensional design coordinates of specific bolts, pipe joints, or structural edges. These coordinates are precise and located in a globally unified coordinate system. Next, a relative pose correction is calculated. This correction essentially describes the deviation between the robot's initial pose data and its actual pose, including both rotation and translation. Subsequently, the calculated relative pose correction is applied to the corresponding initial pose data to correct it, generating corrected pose data. This correction process can be represented as:

[0047] ,

[0048] in, Representing the corrected pose data, it is a 4x4 homogeneous transformation matrix that can accurately transform the point cloud data in the robot's local coordinate system to the global coordinate system. It is the calculated relative pose correction amount, which is also a 4x4 homogeneous transformation matrix used to compensate for the error in the initial pose. This is the initial pose data provided by the robot navigation system. By performing the above correction process on the mobile robots at all stations, the point cloud data of each station has a high-precision pose under a unified reference. Finally, based on the corrected pose data obtained by all robots, a multi-station dynamic unified coordinate system covering the entire monitoring area is successfully established, ensuring that all data sources are expressed under the same high-precision spatial reference.

[0049] Optionally, the calculation of relative pose correction includes:

[0050] Collect image data of the tunnel inner wall and extract environmental structural feature points from it. The environmental structural feature points include environmental corner features and edge features.

[0051] The corner features and edge features are matched with the feature point database, and the feature matching confidence is calculated.

[0052] Determine whether the feature matching confidence level meets the preset positioning requirements;

[0053] If so, the relative pose correction amount is calculated based on the corresponding corner features and the corresponding edge features.

[0054] Specifically, this method details the implementation path for calculating relative pose correction. First, the mobile robot uses its onboard industrial camera and other vision sensors to acquire high-resolution images of the tunnel walls as it moves through the tunnel. Then, the system uses image processing algorithms to analyze the acquired image data to extract stable environmental structural feature points. These feature points are mainly divided into two categories: environmental corner features, which are points in the image with drastic gradient changes and diverse directions, and can be identified using corner detection algorithms such as Harris or Shi-Tomasi; and edge features, which are linear or curvilinear features formed by changes in object contours or textures in the image, and can be extracted using edge detection algorithms such as Canny. After extracting these environmental structural feature points located in the two-dimensional image plane, the next step is to match them with a pre-built three-dimensional feature point library. This feature point library stores the precise three-dimensional coordinates of these feature points in the tunnel design model. The matching process is achieved by calculating feature descriptors; that is, a numerical vector describing the information of its neighborhood image is generated for each extracted feature point, and this vector is compared with the descriptors of feature points in the feature point library to find the best matching pair. To ensure the reliability of the matching, the system calculates the feature matching confidence score for each matching pair. This confidence score is a quantitative indicator used to evaluate the quality of the matching, typically determined based on the distance between descriptors or the uniqueness of the match. The system then determines whether the calculated feature matching confidence score meets a preset localization requirement. This requirement is a threshold; only when the confidence score is higher than this threshold is the matching pair considered valid and reliable. This filtering step eliminates fuzzy and incorrect matches caused by changes in lighting, viewpoint differences, or scene repetition, thus ensuring the quality of the data used for pose calculation. Finally, for all valid matching pairs that pass the confidence score test, the system obtains a set of precise 2D-to-3D point correspondences, i.e., the 2D pixel coordinates in the image and their 3D spatial coordinates in the global coordinate system. Based on these correspondences, the Perspective-n-Point (PnP) algorithm is used to solve for the camera (i.e., the robot) pose. This process aims to solve for a rigid body transformation that minimizes the reprojection error of 3D points on the image plane, and its mathematical model can be expressed as:

[0055] ,

[0056] in, It represents the two-dimensional pixel coordinates of the i-th environmental structural feature point on the image. It is its corresponding three-dimensional global coordinate in the feature point library. This represents the camera's internal parameter matrix, obtained through pre-calibration. It is a scale factor. The core objective of this algorithm is to obtain the rotation matrix. Translation vector These two parameters together constitute the robot's precise pose relative to the global coordinate system. Combining them into a homogeneous transformation matrix yields the final output relative pose correction, used to correct the robot's initial pose data.

[0057] Optionally, generating the preliminary fused point cloud includes:

[0058] Extract the set of tunnel structural feature points from the original tunnel point cloud data;

[0059] Distinguish circumferential joint feature points and longitudinal joint feature points from the set of tunnel structural feature points;

[0060] Calculate the similarity matrix between the circumferential joint feature points and the longitudinal joint feature points between different measuring stations;

[0061] Initial alignment is performed based on the similarity matrix to generate initial registration results;

[0062] Based on the aforementioned multi-station dynamic unified coordinate system, the point cloud density distribution of the initial registration result in the overlapping area is calculated;

[0063] The preset registration weights are adjusted according to the point cloud density distribution, and weighted optimization is performed to generate a preliminary fused point cloud.

[0064] Specifically, the first step is to perform feature analysis on the raw tunnel point cloud data collected by each monitoring station to extract key information that characterizes the tunnel's geometry, namely, the tunnel structural feature point set. In shield tunnels, these feature points mainly manifest as joints formed by segment assembly. Therefore, the system further distinguishes two types of orthogonal features from this feature point set: circumferential joint feature points distributed radially along the tunnel and longitudinal joint feature points distributed axially along the tunnel. These two types of features form a stable grid structure within the tunnel, providing robust geometric constraints for matching between point clouds. After obtaining the joint features of different monitoring station cloud data, the system calculates a similarity matrix between circumferential and longitudinal joint feature points between adjacent or overlapping monitoring stations for initial alignment. This matrix evaluates the probability of correspondence between joint features in different point clouds by quantifying and comparing the geometric attributes of the features, such as length, curvature, and spatial orientation. Based on the highest-scoring matching pair in the similarity matrix, the system can solve for an initial rigid body transformation matrix, transforming the point cloud of one station to the coordinate system of another station, thus completing the coarse alignment between the point clouds and generating the initial registration result. Subsequently, it enters the weighted optimization fine registration stage, ultimately generating a high-precision preliminary fused point cloud.

[0065] Optionally, the step of adjusting the preset registration weights based on the point cloud density distribution and performing weighted optimization to generate a preliminary fused point cloud includes:

[0066] Calculate the point cloud density difference value based on the point cloud density distribution;

[0067] Based on the point cloud density difference value, adjust the preset registration weight;

[0068] Based on the point cloud density distribution and the adjusted registration weights, an objective function is established to characterize the registration error, using the rotation matrix and translation as variables.

[0069] With the preset registration error as the target, the objective function is solved to obtain the final rotation matrix and the final translation amount;

[0070] The initial fused point cloud is generated using the final rotation matrix and the final translation.

[0071] Specifically, firstly, within the overlapping region of the point clouds, the local neighborhood of each point is analyzed, and the local point cloud density is calculated by counting the number of points within its neighborhood. Based on this, a point cloud density difference value is further calculated. This difference value quantifies the deviation of the local point cloud density from an ideal or average density, thus reflecting the quality and reliability of data acquisition in that area. Subsequently, the system uses this point cloud density difference value to dynamically adjust a preset registration weight. The core idea of ​​this process is to assign higher weights to areas with high point cloud density and more reliable data, while assigning lower weights to areas with low density that may contain noise or insufficient information. This adjustment mechanism ensures that high-quality data points have a greater influence on the registration results during subsequent optimization. Next, based on the registration weights adjusted for density, the system establishes an objective function to characterize the registration error. This objective function uses the rotation matrix and translation to be solved as variables, and its purpose is to mathematically describe the degree of alignment between two point clouds. This objective function is usually defined as the sum of the weighted squared distances between all corresponding point pairs, and can be expressed as:

[0072] ,

[0073] in, The registration error objective function is defined in terms of the rotation matrix. Translation vector The function. It is a point in the source point cloud. It is the point that corresponds to it in the target point cloud. It is calculated based on the point cloud density difference value and assigned to point pairs. The adjusted registration weights. and These represent the rotation matrix and translation to be optimized. The physical meaning of this formula is to find a rigid body transformation such that the weighted sum of distances between points in the source point cloud and corresponding points in the target point cloud is minimized after the transformation. The next step of the system is to solve the above objective function using a numerical optimization algorithm, with the goal of minimizing the registration error. This process iteratively adjusts... and until the objective function The value converges or falls below a preset registration error threshold. After solving, the optimal final rotation matrix and final translation amount are obtained. Finally, this final transformation matrix is ​​applied to the entire source point cloud to precisely align it with the target point cloud. By merging all the multi-measurement point clouds that have undergone this optimization and alignment, a preliminary fused point cloud with higher overall geometric consistency is generated.

[0074] Optionally, generating the final fused point cloud includes:

[0075] Calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping region, and extract the angle deviation value and distance deviation value from the fusion error distribution parameters;

[0076] Calculate the angle weighting coefficient based on the angle deviation value, and calculate the distance weighting coefficient based on the distance deviation value;

[0077] The angle weight coefficient and the distance weight coefficient are combined to generate an adaptive weight matrix;

[0078] The adaptive weight matrix is ​​used to perform weighted interpolation on the overlapping regions of the preliminary fused point cloud to generate the final fused point cloud.

[0079] Specifically, this method aims to refine the preliminary fused point cloud to eliminate minor deviations in overlapping areas of multi-source data, thereby generating a seamless and high-precision final fused point cloud. This process first focuses on the overlapping areas in the preliminary fused point cloud, which are composed of data from different stations. Within this area, the system quantifies the geometric consistency of each local location, i.e., calculates the fusion error distribution parameters. For this purpose, the system extracts two key indicators: distance deviation, which represents the Euclidean distance between points from different original point clouds at the same physical location, reflecting the residual error of location registration; and angle deviation, obtained by calculating the local surface normal vectors of the corresponding point's neighborhood and solving for the angle between these normal vectors, reflecting the consistency of surface geometry. After obtaining the distance and angle deviation distributions for the entire overlapping area, the system determines the weight of each data point based on these error indicators. Specifically, it calculates the angle weight coefficient based on the angle deviation value and the distance weight coefficient based on the distance deviation value. The calculation of these two coefficients follows a basic principle: the smaller the deviation, the larger the weight coefficient, indicating high consistency and strong reliability of the data at that point. Conversely, the greater the deviation, the smaller the weight coefficient. Subsequently, the system fuses the angle and distance weight coefficients to generate a comprehensive weight value for each point in the overlapping region. These weight values ​​together constitute an adaptive weight matrix. The core function of this matrix is ​​to provide a quantitative confidence reference that varies with spatial location for subsequent data processing. The final step is to use this adaptive weight matrix to perform weighted interpolation on the overlapping region of the initially fused point cloud. This process can be viewed as a weighted average; it no longer simply retains or removes overlapping points, but calculates a new, optimal spatial location based on the weight of each point. The final fused point location within a local region. It can be calculated in the following ways:

[0080] ,

[0081] in, These are the original points from different stations within this local area. The result is obtained through the adaptive weight matrix. The corresponding weight values. This calculation process ensures that points with high consistency, small deviations, and large weights contribute more to the final fused position. By performing this process on all overlapping regions, the system can effectively smooth and correct stitching marks, ultimately generating a geometrically continuous, smooth-surfaced final fused point cloud.

[0082] Optionally, the generation of the deformation monitoring benchmark model includes:

[0083] Extract cross-sectional contour points from the final fused point cloud;

[0084] Ellipse fitting is performed on the cross-sectional contour points to obtain multiple cross-sectional fitting ellipses;

[0085] Extract the center coordinates of the fitted ellipse of the multiple cross sections, and generate the tunnel centerline by fitting the center coordinates;

[0086] Based on the tunnel centerline and the final fused point cloud, a deformation monitoring benchmark model is constructed.

[0087] Specifically, this method details how to transform a high-precision final fused point cloud into a structured deformation monitoring benchmark model that can be used for quantitative analysis. This process begins by slicing the final fused point cloud to extract cross-sectional profile points that represent the tunnel's cross-sectional morphology. Specifically, a series of parallel virtual cutting planes are defined along the general direction of the tunnel; these planes are substantially perpendicular to the tunnel axis. The portions of the final fused point cloud that intersect with these cutting planes constitute a series of discrete two-dimensional point sets, i.e., the cross-sectional profile points.

[0088] After acquiring cross-sectional profile points at multiple locations, the system performs ellipse fitting on each set of two-dimensional points. An ellipse, rather than a perfect circle, is chosen because the cross-section of a tunnel, after actual stress and deformation, typically exhibits a non-circular elliptical shape. Ellipse fitting uses optimization algorithms such as the least squares method to find an ellipse parameter that best approximates all the profile points of the cross-section, including the ellipse's center coordinates, major and minor axis lengths, and rotation angle. By performing this operation on all cross-sectional profile point sets, the system generates multiple cross-sectional fitting ellipses, which mathematically and precisely describe the actual cross-sectional shape of the tunnel at different locations.

[0089] Next, the system extracts the center coordinates of all the generated cross-sectional fitting ellipses. These center coordinate points form a discrete sequence of points in three-dimensional space, which depicts the trajectory of the tunnel's actual geometric centerline. To obtain a continuous and smooth tunnel centerline, the system uses methods such as B-spline curves or polynomial curves to fit these center coordinate points. The curve generated by this fitting is the actual centerline of the tunnel, which accurately reflects the tunnel's macroscopic orientation and shape in three-dimensional space.

[0090] Finally, the system organically combines this precise tunnel centerline with the original, final fused point cloud containing rich surface details to jointly construct a deformation monitoring benchmark model. This model not only includes high-precision three-dimensional geometric information of the tunnel surface but also possesses precisely extracted and parameterized structural axes and cross-sectional features. This provides the tunnel's three-dimensional solid model with a clear geometric reference benchmark, laying the foundation for subsequent precise comparative analysis with the design model.

[0091] Optionally, obtaining the deformation distribution map includes:

[0092] Obtain the tunnel design model, spatially align the deformation monitoring benchmark model with the tunnel design model, and calculate the deformation parameters including radial displacement and convergent deformation.

[0093] The abnormal deformation area is identified based on the deformation parameters;

[0094] Extract the point cloud density and curvature features of the deformed anomaly region to generate a deformation distribution map.

[0095] Specifically, the first step is to obtain an authoritative tunnel design model, typically a 3D CAD model that precisely defines the ideal geometry of the tunnel upon completion, including its centerline position, cross-sectional profile, and dimensions. Next, the deformation monitoring benchmark model generated in the previous steps is precisely aligned with this tunnel design model in the same 3D space—a process known as spatial alignment. This alignment process usually uses the centerlines of the two models as a reference to ensure that subsequent comparisons are conducted within a unified and meaningful coordinate framework. After alignment, the system begins point-by-point or cross-sectional comparative calculations to obtain a series of quantified deformation parameters. Radial displacement is obtained by calculating the shortest normal distance from any point on the surface of the deformation monitoring benchmark model to the corresponding location on the surface of the tunnel design model; it directly reflects the degree of inward convexity or outward stretching of the tunnel wall. Convergence deformation is calculated on the cross-section by comparing the major and minor axes of the fitted ellipse in the deformation monitoring benchmark model or the chord length in a specific direction with the dimensions of the corresponding cross-section in the design model to quantify the degree of flattening or stretching of the cross-section. After obtaining deformation parameters covering the entire tunnel surface, the system evaluates these parameters according to preset engineering safety specifications or warning thresholds, automatically identifying areas where deformation exceeds the allowable range and marking them as abnormal deformation areas. To further diagnose these critical areas, the system extracts the local geometric features of the point cloud in those areas. Specifically, the system calculates the point cloud density of the abnormal deformation areas to help determine if surface defects such as spalling or flaking exist. Simultaneously, the system calculates the surface curvature characteristics of these areas, as abrupt changes in curvature are often closely related to stress concentration, joint misalignment, or structural cracks. Finally, the system comprehensively visualizes all the calculation and analysis results, generating an intuitive deformation distribution map. This map is typically presented as a 3D model and uses pseudo-color rendering technology to map different deformation parameter values ​​to different colors. For example, red may indicate severe radial displacement or convergence deformation, while blue represents stable areas. The map can also highlight identified abnormal deformation areas and overlay their point cloud density or curvature characteristics, providing engineers with a comprehensive, intuitive, and multi-dimensional tunnel health report.

[0096] Optionally, the generation of the deformation trend prediction report includes:

[0097] Obtain historical deformation parameters and extract current deformation parameters from the deformation distribution map;

[0098] Based on the historical deformation parameters and the current deformation parameters, a time series analysis is performed to calculate the trend change in deformation rate.

[0099] The deformation rate trend change is compared with a preset safety threshold to generate an early warning signal;

[0100] Based on the early warning information, key areas are marked and a deformation trend prediction report is generated.

[0101] Specifically, the first step is to establish a time-dimensional deformation database. The system acquires historical deformation parameters from each monitoring cycle and adds the current deformation parameters extracted from the deformation distribution map as a new data point to the corresponding time series. Each key monitoring point or area will have a dataset consisting of a series of deformation values ​​sorted by time. Based on this complete time series containing historical and current data, the system will perform time series analysis, with the core objective of calculating the trend change in deformation rate, i.e., the acceleration of deformation. First, the system calculates the deformation rate between adjacent time points, i.e., the increment of deformation per unit time. Subsequently, the system further analyzes the variation of these deformation rates over time, calculating the rate of change, which is the trend change in deformation rate. This parameter is crucial because it reveals whether the deformation process is stabilizing, maintaining a uniform rate, or accelerating. A positive and continuously increasing trend change in deformation rate is a strong signal of increased structural instability risk. Next, the system will rigorously compare the calculated trend change in deformation rate with a safety threshold preset according to engineering specifications and geological conditions. This safety threshold defines the acceptable upper limit of deformation acceleration. Once the deformation rate trend in a certain area exceeds a certain threshold, the system will automatically trigger and generate an early warning signal. This signal precisely points to the potential source of risk, indicating that the deformation development in that area has entered a stage requiring close attention. Finally, the system will use this early warning signal to highlight the corresponding key areas on the 3D deformation distribution map or engineering drawings, and automatically generate a deformation trend prediction report. This report will not only clearly list all the key areas that triggered the warning, but also include historical deformation curves, deformation rate change graphs, and quantitative predictions of deformation development in the near future for these areas. This report provides tunnel managers with direct, clear, and forward-looking decision support information.

[0102] Based on the same inventive concept, such as Figure 2As shown, the present invention also provides a data fusion and processing system for curve tunnel deformation monitoring based on a multi-station robot, the system comprising:

[0103] The data acquisition module is used to acquire raw point cloud data of the tunnel and corresponding initial pose data collected by multiple mobile robots;

[0104] The coordinate system construction module is used to identify environmental structural feature points in the original point cloud data of the tunnel, and to construct a dynamic unified coordinate system for multiple stations based on the initial pose data and the environmental structural feature points.

[0105] The point cloud registration module is used to extract the tunnel structure feature point set from the original tunnel point cloud data, and perform preliminary registration of the original tunnel point cloud data based on the multi-station dynamic unified coordinate system to generate a preliminary fused point cloud.

[0106] The data fusion module is used to calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping area, and to perform weighted fusion of the preliminary fused point cloud according to the fusion error distribution parameters to generate the final fused point cloud;

[0107] The model generation module is used to perform geometric modeling on the final fused point cloud and generate a deformation monitoring benchmark model.

[0108] The deformation analysis module is used to acquire the tunnel design model and compare the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map.

[0109] The trend prediction module is used to extract the current deformation parameters from the deformation distribution map, perform time series prediction, and generate a deformation trend prediction report.

[0110] To verify the feasibility of this invention in practice, it was applied to a structural health monitoring project for a curved tunnel section of an extended subway line in a certain city. This tunnel section has complex geological conditions and is constantly affected by surface traffic loads and surrounding construction. Traditional manual contact-based monitoring methods are inefficient, have limited accuracy, and interfere with normal subway operations. The operating unit hopes to use this invention to deploy multiple monitoring robots to achieve high-precision, automated deformation monitoring and trend early warning for this curved tunnel section.

[0111] In this embodiment, the project team deployed three mobile robots equipped with laser scanners and industrial cameras to work collaboratively along the tunnel track. The system first acquires the raw point cloud data of the tunnel collected by each robot and the initial pose data recorded by its internal navigation system through the data acquisition module. Subsequently, the coordinate system construction module uses a pre-established feature point library containing designed locations such as tunnel segment bolts and cable tray fixing points to analyze the images acquired by the robots, identify environmental corner points and edge features, calculate relative pose corrections, and thus construct a globally unified multi-station dynamic coordinate system. Figure 3 The comparison of pose correction effects is shown. Based on this, the point cloud registration module extracts the circumferential and longitudinal joint features of the tunnel wall for initial alignment, and then performs weighted optimization registration based on the point cloud density distribution in the overlapping area. The data fusion module then performs adaptive weighted interpolation based on the fusion error (angle and distance deviation) of the registered point cloud in the overlapping area to generate the final fused point cloud. Finally, the model generation, deformation analysis, and trend prediction modules process the data sequentially, generating the deformation monitoring benchmark model and deformation distribution map for this tunnel section, and predicting the deformation trend in key areas.

[0112] To verify the beneficial effects of this invention, the project team conducted a full-coverage data collection once a month during a certain period and performed a detailed analysis of the collected data. The following are the data analysis and effect verification results during the experiment.

[0113] During the coordinate system construction and point cloud fusion stage, for example, near mileage K1+350, the initial pose data of a robot experienced a cumulative error of approximately 120mm due to slippage of the wheeled odometer. Using a pose correction method, the system matched the robot's environmental structural feature points with a feature point database to calculate a precise relative pose correction. After correction, the absolute error of the pose was controlled within 5mm. During data fusion, for overlapping areas of data from different stations, the system calculated an average distance deviation of approximately 3.5mm and an average normal vector angle deviation of approximately 2.1°. Through adaptive weighted fusion processing, the stitching marks in the overlapping areas of the final fused point cloud were effectively eliminated, significantly improving geometric continuity.

[0114] During the deformation analysis phase, the system compares the generated deformation monitoring benchmark model with the tunnel design model. For example... Figure 4 As shown, significant convergence deformation occurred in the region from mileage K1+280 to K1+310. Specifically, as... Figure 5 As shown, the maximum radial displacement at section K1+295 is -15.8 mm, pointing towards the tunnel interior, while the vertical convergence deformation reaches 11.2 mm, exceeding the design allowable value. The system automatically marks this area as an abnormal deformation zone and displays it on the deformation distribution map, providing an intuitive basis for maintenance decisions.

[0115] like Figure 6 As shown, the system retrieved historical deformation parameters of section K1+295 over the past 6 months and performed time series analysis. Figure 7 As shown, the vertical convergence deformation rate of this section gradually increased from 0.5 mm / month in February to 2.2 mm / month in June. The system calculated that the change in its deformation rate trend had exceeded the preset safety threshold of 1.0 mm / month², indicating that the deformation in this area was accelerating. The system then generated an early warning signal and pointed out in the deformation trend prediction report that there was a high structural safety risk in this area, recommending immediate manual review and reinforcement intervention.

[0116] In summary, this invention can efficiently and accurately complete the deformation monitoring task of curved tunnels. By constructing a dynamic unified coordinate system and a multi-level fusion strategy, the global consistency and high accuracy of multi-source data are ensured. Through comparative analysis with the design model and time series prediction, it can not only accurately quantify the current deformation state of the tunnel, but also provide early warning of potential safety risks, realizing a leap from static measurement to dynamic prediction, and providing strong technical support for ensuring the safe operation of urban rail transit.

[0117] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0118] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for data fusion and processing of curve tunnel deformation monitoring based on a multi-station robot, characterized in that, The method includes: Acquire raw point cloud data of the tunnel and corresponding initial pose data collected by multiple mobile robots; The process involves identifying environmental structural feature points in the original tunnel point cloud data, and constructing a multi-station dynamic unified coordinate system based on the initial pose data and the environmental structural feature points. This includes: acquiring a feature point library containing the designed locations of natural feature points in the tunnel; identifying environmental structural feature points in the original tunnel point cloud data, matching the environmental structural feature points with the feature point library, and calculating relative pose correction amounts; using the relative pose correction amounts to correct the initial pose data, generating corrected pose data; and establishing a multi-station dynamic unified coordinate system based on the corrected pose data of all robots. Extracting the tunnel structural feature point set from the original tunnel point cloud data and performing preliminary registration of the original tunnel point cloud data based on the multi-station dynamic unified coordinate system to generate a preliminary fused point cloud; including: extracting the tunnel structural feature point set from the original tunnel point cloud data; distinguishing circumferential joint feature points and longitudinal joint feature points from the tunnel structural feature point set; calculating the similarity matrix between the circumferential joint feature points and the longitudinal joint feature points between different station point clouds; performing initial alignment based on the similarity matrix to generate an initial registration result; calculating the point cloud density distribution of the initial registration result in the overlapping area based on the multi-station dynamic unified coordinate system; adjusting the preset registration weights according to the point cloud density distribution and performing weighted optimization to generate a preliminary fused point cloud; Calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping region, and perform weighted fusion on the preliminary fused point cloud according to the fusion error distribution parameters to generate the final fused point cloud; Geometric modeling is performed on the final fused point cloud to generate a deformation monitoring benchmark model; Obtain the tunnel design model, and compare and analyze the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map; The current deformation parameters are extracted from the deformation distribution map for time series prediction, and a deformation trend prediction report is generated.

2. The data fusion and processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 1, characterized in that, The calculation of relative pose correction includes: Collect image data of the tunnel inner wall and extract environmental structural feature points from it. The environmental structural feature points include environmental corner features and edge features. The corner features and edge features are matched with the feature point database, and the feature matching confidence is calculated. Determine whether the feature matching confidence level meets the preset positioning requirements; If so, the relative pose correction amount is calculated based on the corresponding corner features and the corresponding edge features.

3. The data fusion processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 1, characterized in that, The step of adjusting the preset registration weights based on the point cloud density distribution and performing weighted optimization to generate a preliminary fused point cloud includes: Calculate the point cloud density difference value based on the point cloud density distribution; Based on the point cloud density difference value, adjust the preset registration weight; Based on the point cloud density distribution and the adjusted registration weights, an objective function is established to characterize the registration error, using the rotation matrix and translation as variables. With the preset registration error as the target, the objective function is solved to obtain the final rotation matrix and the final translation amount; The initial fused point cloud is generated using the final rotation matrix and the final translation.

4. The data fusion processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 1, characterized in that, The generation of the final fused point cloud includes: Calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping region, and extract the angle deviation value and distance deviation value from the fusion error distribution parameters; Calculate the angle weighting coefficient based on the angle deviation value, and calculate the distance weighting coefficient based on the distance deviation value; The angle weight coefficient and the distance weight coefficient are combined to generate an adaptive weight matrix; The adaptive weight matrix is ​​used to perform weighted interpolation on the overlapping regions of the preliminary fused point cloud to generate the final fused point cloud.

5. The data fusion and processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 4, characterized in that, The generated deformation monitoring benchmark model includes: Extract cross-sectional contour points from the final fused point cloud; Ellipse fitting is performed on the cross-sectional contour points to obtain multiple cross-sectional fitting ellipses; Extract the center coordinates of the fitted ellipse of the multiple cross sections, and generate the tunnel centerline by fitting the center coordinates; Based on the tunnel centerline and the final fused point cloud, a deformation monitoring benchmark model is constructed.

6. The data fusion processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 5, characterized in that, The obtained deformation distribution map includes: Obtain the tunnel design model, spatially align the deformation monitoring benchmark model with the tunnel design model, and calculate the deformation parameters including radial displacement and convergent deformation. The abnormal deformation area is identified based on the deformation parameters; Extract the point cloud density and curvature features of the deformed anomaly region to generate a deformation distribution map.

7. The data fusion processing method for curve tunnel deformation monitoring based on a multi-station robot according to claim 6, characterized in that, The generated deformation trend prediction report includes: Obtain historical deformation parameters and extract current deformation parameters from the deformation distribution map; Based on the historical deformation parameters and the current deformation parameters, a time series analysis is performed to calculate the trend change in deformation rate. The deformation rate trend change is compared with a preset safety threshold to generate an early warning signal; Based on the warning signal, key areas are marked and a deformation trend prediction report is generated.

8. A data fusion and processing system for curve tunnel deformation monitoring based on a multi-station robot, characterized in that, The system includes: The data acquisition module is used to acquire raw point cloud data of the tunnel and corresponding initial pose data collected by multiple mobile robots; A coordinate system construction module is used to identify environmental structural feature points in the original tunnel point cloud data, and to construct a multi-station dynamic unified coordinate system based on the initial pose data and the environmental structural feature points. This includes: acquiring a feature point library containing the designed locations of natural feature points in the tunnel; identifying environmental structural feature points in the original tunnel point cloud data, matching the environmental structural feature points with the feature point library, and calculating relative pose correction amounts; using the relative pose correction amounts to correct the initial pose data, generating corrected pose data; and establishing a multi-station dynamic unified coordinate system based on the corrected pose data of all robots. The point cloud registration module is used to extract the set of tunnel structural feature points from the original tunnel point cloud data and perform preliminary registration of the original tunnel point cloud data based on the multi-station dynamic unified coordinate system to generate a preliminary fused point cloud. This includes: extracting the set of tunnel structural feature points from the original tunnel point cloud data; distinguishing circumferential joint feature points and longitudinal joint feature points from the set of tunnel structural feature points; calculating the similarity matrix between the circumferential joint feature points and the longitudinal joint feature points in the point clouds of different stations; performing initial alignment based on the similarity matrix to generate an initial registration result; calculating the point cloud density distribution of the initial registration result in the overlapping area based on the multi-station dynamic unified coordinate system; adjusting the preset registration weights according to the point cloud density distribution and performing weighted optimization to generate a preliminary fused point cloud. The data fusion module is used to calculate the fusion error distribution parameters of the preliminary fused point cloud in the overlapping area, and to perform weighted fusion of the preliminary fused point cloud according to the fusion error distribution parameters to generate the final fused point cloud; The model generation module is used to perform geometric modeling on the final fused point cloud and generate a deformation monitoring benchmark model. The deformation analysis module is used to acquire the tunnel design model and compare the deformation monitoring benchmark model with the tunnel design model to obtain a deformation distribution map. The trend prediction module is used to extract the current deformation parameters from the deformation distribution map, perform time series prediction, and generate a deformation trend prediction report.

Images