A bridge crack three-dimensional size measurement method based on multi-source data fusion

Abstract

The application discloses a bridge crack three-dimensional size measurement method based on multi-source data fusion, and comprises the following steps: S1, three-dimensional reconstruction of a crack apparent contour based on binocular vision and semantic segmentation; S2, finite element mechanics positive analysis with a measured contour as a boundary condition; S3, calculation of errors between measured values and simulation values of a crack center line and a crack width; and S4, establishment of global optimal constraints of multi-source data fusion and acquisition of crack three-dimensional size information. The application can effectively solve problems such as coordinate system registration, heterogeneous data alignment and center line structured expression, and can realize rapid alignment of measured values and simulation values.

Description

Technical Field

[0001] This invention belongs to the field of bridge construction monitoring and structural health monitoring technology, specifically involving a method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion. Background Technology

[0002] During bridge construction, processes such as concrete pouring, prestressing tensioning, formwork removal, and application of construction loads can easily induce structural cracking. The length, width (surface opening), and depth of cracks are core indicators for assessing the safety, durability, and construction quality of a bridge structure. However, existing detection technologies have the following significant shortcomings: 1) Traditional manual inspection methods (such as crack observation instruments, feeler gauges, and ultrasonic detectors) are inefficient, rely on the experience of the inspectors, and can only perform single-point measurements, failing to obtain the continuous width distribution and spatial three-dimensional morphology of cracks along their direction.

[0003] 2) While binocular vision inspection technology can reconstruct the 3D contour of a crack surface through stereo matching, existing methods typically only reconstruct the crack edge lines, lacking accurate extraction of the crack centerline. More importantly, due to the non-uniform variation in crack width along its direction, establishing accurate correlations between each point on the centerline and its corresponding edge points to obtain a continuous width distribution function along the crack direction remains a technical challenge that has not yet been fully resolved in the field of visual inspection. Furthermore, existing methods mostly use discrete point-to-point distances to calculate the width, lacking a geometrically precise solution strategy based on tangent and normal vectors.

[0004] 3) While pure finite element method (FE) simulations (such as the extended finite element method XFEM) can simulate crack initiation and propagation and output crack profile nodal coordinates, their simulation accuracy is highly dependent on the accuracy of model parameters (elastic modulus, fracture energy, etc.) and construction-period load boundary conditions. Due to the complex and variable loads during construction and the difficulty in accurately quantifying temperature field distribution, the crack size (especially depth information) predicted by finite element methods often deviates significantly from the actual situation. More importantly, the crack information output by finite element methods is discrete nodal coordinates, lacking a structured representation along the crack centerline (i.e., the centerline trajectory and the corresponding width function), making direct comparison and analysis with visually measured data difficult.

[0005] 4) Existing multi-source data fusion methods face two major technical bottlenecks when comparing and analyzing visually measured data with finite element simulation results: First, the measured coordinate system (camera coordinate system) and the simulation coordinate system (finite element global coordinate system) are inconsistent, and there is an unknown rigid body transformation relationship between the two; Second, the number of measured point clouds (usually thousands to tens of thousands of pixels) and the number of finite element nodes (usually hundreds to thousands of mesh nodes) differ by orders of magnitude and there is no natural one-to-one correspondence. Existing registration algorithms (such as the ICP iterative nearest point method) have high computational cost, slow convergence speed, and are prone to getting trapped in local optima when processing such heterogeneous point clouds, making it difficult to meet the engineering requirements of rapid detection during the construction period.

[0006] 5) Most existing depth inversion methods rely solely on the geometric features of cracks for inverse correction, failing to fully utilize the structural response data provided by pre-embedded sensors (strain gauges, displacement gauges) during construction. This can easily lead to the inversion results falling into local optima, meaning that while the geometric error on the crack surface is small, the depth information is severely distorted. Summary of the Invention

[0007] The purpose of this invention is to provide a method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion, which can effectively solve problems such as coordinate system registration, heterogeneous data alignment, and structured expression of centerline, and can achieve rapid alignment between measured values ​​and simulated values.

[0008] To achieve the above objectives, this invention provides a method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion, comprising the following steps: S1, 3D Reconstruction of Crack Appearance Contour Based on Binocular Vision and Semantic Segmentation: Binocular images of cracks during bridge construction are acquired. Pixel masks of the crack regions are extracted from these images. The crack centerline and edgeline are extracted from the pixel masks. The corresponding 3D point clouds of the crack edgeline and centerline are reconstructed, forming a 3D point cloud set of the measured contour in the camera coordinate system. and the three-dimensional point cloud set of the measured center line ,pass Get the crack width Variation along the length direction; S2, Finite Element Mechanical Forward Analysis with Measured Profile as Boundary Condition: Loads during bridge construction are collected, a finite element cracking mesh model is established, the loads are added to the finite element cracking mesh model to obtain the cracking results under the current load, and the three-dimensional spatial coordinate set of the crack centerline in the finite element coordinate system is extracted. and the crack width in the finite element coordinate system ; S3, calculate the error between the measured and simulated values ​​of the crack centerline and crack width: The measured values ​​obtained in S1... , The simulated values ​​obtained from S2 , After alignment, calculate and , and Error_match between them; S4. Establish global optimal constraints for multi-source data fusion to obtain three-dimensional crack size information: collect monitoring data from the bridge construction monitoring system: a collection of measured strain values ​​from different locations. A collection of measured displacement values ​​from different locations And extract the set of strain simulation values ​​at corresponding locations in the finite element cracking network model. and the set of displacement simulation values The normalized error Error_S between the measured strain value and the simulated strain value, and the normalized error Error_Dis between the measured displacement value and the simulated displacement value are calculated. A multi-objective global error function is constructed to determine the accuracy of the finite element cracking mesh model, and crack depth information values ​​are extracted from it to form the three-dimensional size information of the crack.

[0009] A further aspect of this invention: In S4, a multi-objective global error function is constructed. ; in, Indicates the multiplication sign. This indicates the weight of the crack width error Error_match. This represents the weight of the strain error Error_S. This indicates the weight of the displacement error Error_Dis; ; if If the value is less than the specified threshold, the finite element simulation accuracy is considered good, and the crack depth information can be directly extracted from the finite element cracked mesh model, denoted as . , and , Together, they constitute the three-dimensional dimensional information of the crack; if If the value does not meet the requirements, the finite element cracked mesh model is modified until... The value is less than the specified threshold.

[0010] As a further aspect of the present invention: Step S1 specifically includes: S1.1, The binocular camera was calibrated using the Zhang Zhengyou calibration method, and then the binocular camera was used to acquire binocular images of cracks during the bridge construction process; S1.2, use a semantic segmentation network to extract the pixel mask of the crack region in the binocular image of the crack, then use a binary image morphology algorithm to extract the crack center line in the pixel mask, and use the Canny edge detection algorithm to extract the crack edge line in the pixel mask; S1.3, using the principles of stereo matching and triangulation, reconstructs the three-dimensional point cloud corresponding to the crack edge line and crack center line, forming a three-dimensional point cloud set of the measured contour: 3D point set of the measured center line: ; S1.4, through Obtain the crack width The variation along the length direction is denoted as Therefore, there is a one-to-one correspondence between the points on the center line of the crack and its width, that is... = .

[0011] As a further aspect of the present invention: the node coordinates are three-dimensional coordinates based on the finite element coordinate system. , The method for extracting the corresponding centerline coordinates and the width value along the centerline includes the following steps: S2.1, find nodes PO, PX, and PY within the finite element plane where the crack is located, such that the vector... Vertical vector , and All are unit vectors arranged in a row and a column; S2.2, Construct the rotation and translation matrix R: ; in, It is a vector and The cross product, where T represents the transpose; S2.3, The three-dimensional coordinates are calculated using the rotation and translation matrix R. Projecting this onto a two-dimensional plane yields the set of crack profile coordinates within that plane. : [ ]=R [ ] T ; in, The true value is always equal to a certain fixed constant C, and its absolute value is equal to the vertical distance from the origin of the finite element coordinate system to the plane where the crack is located. S2.4, Identification The two endpoints along the major axis, based on these two endpoints Divide into two, represented as , Then to , Pair the points in the graph to determine the set of all two-dimensional points on the crack centerline in the two-dimensional plane. Then, in a two-dimensional plane, using and The two-dimensional width value corresponding to the center point of each crack is calculated to obtain the crack width on the two-dimensional plane. The variation along the length of the centerline is denoted as ; S2.5, for and Performing an inverse transformation yields the set of three-dimensional spatial coordinates of the crack centerline in the finite element coordinate system. and the crack width in the finite element coordinate system : ; ; ; = .

[0012] As a further aspect of the present invention: in S2.4 , The point pairing process in the process includes: pairing Each point PD on -left ,exist Find the point PD with the closest Euclidean distance in the middle. -right As a pairing, after pairing is completed, PD is calculated. -left PD -right The midpoint between the two points is used as a point on the center line of the crack.

[0013] As a further aspect of the present invention, the specific steps for aligning the measured values ​​with the simulated values ​​include: S3.1, in the finite element cracked network model, select three reference points PGO, PG1, and PG2, satisfying: vector Vertical vector ,in, , It is a unit vector with three rows and one column; S3.2, using a binocular vision 3D reconstruction algorithm, calculates the 3D spatial coordinates W of three reference points PGO, PG1, and PG2 in the camera coordinate system. PCO W PG1 W PG2 Among them, W PCO =[W PCOX W PCOYW PCOZ ] T W PC1 =[W PC1X W PC1Y W PC1Z ] T W PC2 =[W PC2X W PC2Y W PC2Z ] T ; S3.3, let W PC1O =W PC1 -W PCO W PC2O =W PC2 -W PCO W PC1O W PC2O Normalized to a unit vector W PC1O_unit W PC2O_unit Then construct the rotation and translation matrix R. I2C : ; in, Represents the cross product of vectors; S3.4, using the rotation and translation matrix R I2C Will Transform to the camera coordinate system to obtain the simulation results of the crack centerline in the camera coordinate system. ; width The rotation and translation matrix R is not affected. I2C The influence, therefore the width in the camera coordinate system = ; S3.5, the measured crack centerline value The crack centerline value compared with the finite element simulation Perform matching and filtering and All successfully paired points in the array are denoted as follows: and ; S3.6, based on the pairing results of the center lines and ,turn up and Pairing results and Alignment complete; S3.7, calculate the error_match between the measured and simulated values ​​of the crack centerline and crack width. Error_match is characterized by the matching success rate, i.e.: Error_match=1-max(PERCENT1, PERCENT2); PERCENT1=NUM( ) / NUM( ); PERCENT2=NUM( ) / NUM( ); Where NUM() represents the number of values ​​a variable can take, and max() represents taking the maximum value; the lower the value of Error_match, the better the finite element simulation result; the larger the value of Error_match, the worse the error simulation result.

[0014] As a further aspect of the present invention: in S3.5 and Pairing method: Find the relationship between each 3D point in the data and... Find the nearest neighbor in the array and calculate the Eulerian distance between these two points. If the distance is less than a threshold, the match is considered successful; if no nearest neighbor that meets the threshold is found, the pairing fails and the point is discarded.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The centerline extraction method based on coordinate orthogonal projection establishes a local coordinate system for the contour and performs orthogonal projection transformation on the contour node coordinates. This method can automatically and accurately extract the three-dimensional coordinates of the crack centerline from crack meshes of any irregular shape, completely avoiding the subjective errors and computational instabilities introduced by manually selecting the centerline or relying on complex curve fitting. At the same time, based on the extracted centerline, by calculating the normal of each point and intersecting it with the contours on both sides, the simulated crack width value at each point along the centerline can be obtained simultaneously. This ensures that the simulated data is structurally consistent with the measured data (each point on the centerline uniquely corresponds to a width value), providing a unified and reliable data foundation for subsequent point-by-point error calculation between the measured and simulated data. The method for rapid alignment of measured and simulated values ​​using coordinate normalization and adjacent threshold discrimination uses the camera coordinate system of the measured centerline point cloud as a reference. It performs coordinate system transformation on the simulated centerline point cloud in the finite element coordinate system, normalizing it to a unified camera coordinate system. This fundamentally eliminates the translation and scale differences between the camera coordinate system and the finite element coordinate system. Then, it constructs a discrimination threshold using the Euclidean distance between adjacent points. The optimal matching between measured and simulated points can be quickly completed in a single traversal, solving the matching difficulties caused by inconsistent point cloud numbers without iterative optimization. Compared with traditional iterative registration algorithms such as ICP, it significantly improves computational efficiency. More importantly, this method can establish a one-to-one correspondence between measured and simulated points along the crack direction, allowing key geometric parameters such as crack length and width to be compared point by point along the entire crack. This accurately calculates the matching error (Error_match) of the crack geometry, providing a clear and robust metric for error quantification in multi-source data fusion.

[0016] The alignment of centerline extraction with measured and simulated values ​​breaks down the data barrier between binocular vision measured data and finite element simulation data. This allows apparent measured information such as the width and length of the crack surface to be directly used as global constraints in the error assessment and iterative correction of the finite element cracking network model. Based on this, a global error function is constructed by combining multi-source data such as strain and displacement obtained from the construction monitoring system. When the error meets the threshold requirement, the crack depth can be directly extracted from the finite element cracking model, thereby obtaining complete three-dimensional information on the crack length, width distribution along the length, and depth in one go. This significantly improves the automation, accuracy, and engineering practicality of crack size inversion during bridge construction. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion, as described in this invention.

[0018] Figure 2 This is a binocular image of the slit.

[0019] Figure 3 A segmentation mask for the left eye image of a certain crack branch.

[0020] Figure 4 The center line and edge line of the crack.

[0021] Figure 5 A 3D point cloud guided by a crack branch mask map.

[0022] Figure 6 This is a schematic diagram of the crack width.

[0023] Figure 7 A schematic diagram for establishing local normals and boundary points.

[0024] Figure 8 The results of finite element crack location and fine simulation are presented.

[0025] Figure 9 This is a schematic diagram showing the crack width values ​​at different centerline locations.

[0026] Figure 10 This is a schematic diagram showing the three-dimensional dimensions of the crack. Detailed Implementation

[0027] The present invention will be further illustrated by the following examples.

[0028] like Figure 1 As shown, a method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion includes the following steps: S1, 3D Reconstruction of Crack Appearance Contour Based on Binocular Vision and Semantic Segmentation: Binocular images of cracks during bridge construction are acquired. Pixel masks of the crack regions are extracted from these images. The crack centerline and edgeline are extracted from the pixel masks. The corresponding 3D point clouds of the crack edgeline and centerline are reconstructed, forming a 3D point cloud set of the measured contour in the camera coordinate system. and the three-dimensional point cloud set of the measured center line ,pass Obtain the crack width The variation along the length direction.

[0029] The specific steps include: S1.1 The stereo camera was calibrated using Zhang Zhengyou's calibration method. The calibration results of the stereo camera parameters are shown in Table 1.

[0030] Table 1: Results of Binocular Camera Parameter Calibration

[0031] Then, binocular images of cracks during bridge construction were acquired using a binocular camera. The binocular images of the cracks are shown below. Figure 2 As shown; S1.2, using semantic segmentation networks such as DeepLabv3+, the pixel mask of the crack region in the binocular image of the crack is extracted. The segmentation mask of the left eye image of a certain crack branch is as follows: Figure 3 As shown, the center line of the crack in the pixel mask is then extracted using a binary image morphology algorithm, and the edge line of the crack in the pixel mask is extracted using the Canny edge detection algorithm, as shown. Figure 4 The image shows the extracted crack centerline and crack edgeline. S1.3, using the principles of SGBM stereo matching and triangulation, reconstructs the three-dimensional point cloud corresponding to the crack edge line and crack center line, forming a three-dimensional point cloud set of the measured profile: 3D point set of the measured center line: A 3D point cloud guided by a crack branch mask image, such as... Figure 5 As shown; S1.4, through Obtain the crack width The variation along the length direction is denoted as ; Specifically, for The measured length is obtained by summing the distances between adjacent points in the data. In addition, for For each point P1 in the crack centerline, multiple adjacent points are selected for fitting calculation to obtain the tangent vector of point P1 on the crack centerline. Then, the normal vector of point P1 on the crack centerline is calculated. The two intersection points P2 and P3 of this normal vector with the 3D point cloud of the contour lines on both sides of the centerline are calculated. The 3D spatial distance between these two intersection points P2 and P3 is the width of the crack at point P1. From the crack centerline... Starting from the first point on the surface, follow the crack towards the opposite direction. By traversing all points on the surface, the width of the crack can be obtained. The variation along the length direction, i.e. A schematic diagram of the crack width is shown below. Figure 6 As shown, Figure 6 In the diagram: C1 represents the left boundary line of the crack, C2 represents the right boundary line of the crack, and L... t L represents the normal to a point on the center line. P P represents the tangent line at a point on the center line. 11 P represents the point on the left boundary line that is closest to the normal line. 12 Let P be the point on the left boundary line that is closest to the other side of the normal. 21 P represents the point on the right boundary line that is closest to the normal line. 22 This represents the point on the right boundary line that is closest to the other side of the normal. A schematic diagram illustrating the determination of normals and boundary points is shown below. Figure 7 As shown, Figure 7 In the middle: the blue line represents the center line of the crack, the two red dots represent the start and end points of the width, i.e. the boundary points, the green dot represents a point on the selected center line (the corresponding width should be calculated based on this point), and the short yellow line represents the normal of the green point on the center line; Therefore, there is a one-to-one correspondence between the points on the center line of the crack and its width, that is... = .

[0032] S2, Finite element mechanical forward analysis with measured profile as boundary condition: Using the existing bridge construction monitoring system, the loads during bridge construction are collected: structural self-weight G, construction load. Temperature field Shrinkage and creep parameters In finite element software such as Abaqus, the extended finite element method is used to establish a high-fidelity finite element cracking mesh model that allows crack propagation. The loads during construction are then added to the finite element cracking mesh model to begin forward crack propagation analysis, obtaining the cracking results under the current construction load. Figure 8 As shown, the coordinates of the nodes located on the crack profile after the crack has opened are extracted. ,pass The set of three-dimensional spatial coordinates of the crack centerline in the finite element coordinate system was extracted. and the crack width in the finite element coordinate system ; Calculate the crack width values ​​at different centerline locations, such as Figure 9 As shown.

[0033] Node coordinates are three-dimensional coordinates based on the finite element coordinate system. In order to obtain quickly The corresponding centerline coordinates and the width along the centerline, The method for extracting the corresponding centerline coordinates and the width value along the centerline includes the following steps: S2.1, find nodes PO, PX, and PY within the finite element plane where the crack is located, such that the vector... Vertical vector , It is a unit vector with three rows and one column, specifically obtained by subtracting the three-dimensional spatial coordinates of PO from the three-dimensional spatial coordinates of PX and then dividing by its vector length. It is a unit vector with three rows and one column, specifically obtained by subtracting the three-dimensional spatial coordinates of PO from the three-dimensional spatial coordinates of PY and then dividing by its vector length; S2.2, Construct the rotation and translation matrix R: ; in, It is a vector and The cross product, where T denotes transpose; S2.3, The three-dimensional coordinates are calculated using the rotation and translation matrix R. Projecting this onto a two-dimensional plane yields the set of crack profile coordinates within that plane. : [ ]=R [ ]; in, The true value is always equal to a certain fixed constant C, and its absolute value is equal to the perpendicular distance from the origin of the finite element coordinate system to the plane where the crack is located; because The true value is always equal to C. It is actually two-dimensional; S2.4, using the convex hull and rotational caliper algorithm for identification. The two endpoints along the major axis, and the coordinates of the closed crack profile based on these two endpoints. Divide into two, represented as , Then to , Pair the points in the graph to determine the set of all two-dimensional points on the crack centerline in the two-dimensional plane. Then, in a two-dimensional plane, using and Following the same approach as calculating the width using normals in S1.4, the two-dimensional width value corresponding to the center point of each crack is calculated, thus obtaining the crack width on the two-dimensional plane. The variation along the length of the centerline is denoted as ; Furthermore, , The point pairing process in the process includes: pairing Each point PD -left ,exist Find the point PD with the closest Euclidean distance in the middle. -right As a pairing, after pairing is completed, PD is calculated. -left PD -right The midpoint between the two points is used as a point on the center line of the crack.

[0034] S2.5, for and Performing an inverse transformation yields the set of three-dimensional spatial coordinates of the crack centerline in the finite element coordinate system. and the crack width in the finite element coordinate system : ; ; ; = .

[0035] Since the width value is an absolute value, it is valid in both the finite element coordinate system and the two-dimensional plane coordinate system. = .

[0036] S3, calculate the error between the measured and simulated values ​​of the crack centerline and crack width: The measured values ​​obtained in S1... , The simulated values ​​obtained from S2 , After alignment, calculate and , and Error_match between them.

[0037] Crack length measurement results: Finite element measurement result: 1465.20 mm. Binocular vision measurement result: 1402.3 mm. Error_match=2.6%, indicating that the finite element measurement result and the binocular vision measurement result are very close.

[0038] Table 2 shows a comparison between the crack width measured by finite element method and the width measured by binocular vision.

[0039] Table 2: Comparison of crack width measured by finite element method and binocular vision method

[0040] Two problems exist when calculating errors: First, the coordinate systems of measured and simulated values ​​are inconsistent. Measured values ​​are 3D coordinates obtained based on the camera coordinate system, while simulated values ​​are 3D coordinates obtained based on the finite element coordinate system. Second, the number of 3D points in the measured and simulated values ​​is inconsistent. Therefore, it is necessary to align the measured and simulated values. To this end, a fast alignment method for measured and simulated values ​​based on coordinate normalization and adjacent threshold discrimination is proposed, which adopts the following steps: S3.1, select three reference points PGO, PG1, and PG2 in the finite element crack network model, ensuring that these three points are easily identifiable both in the finite element crack network model and in the stereoscopic image of the crack. For example, select corner points in the finite element crack network model, and satisfy the following conditions: vector Vertical vector ,in, It is a unit vector with three rows and one column, specifically obtained by subtracting the three-dimensional spatial coordinates of PO from the three-dimensional spatial coordinates of PG1 and then dividing by its vector length (with the finite element coordinate system as the reference frame). It is a unit vector with a row of three rows and one column, specifically composed of... Subtract the three-dimensional spatial coordinates The three-dimensional spatial coordinates are then divided by its vector length to obtain the result (with the finite element coordinate system as the reference system). S3.2, using a binocular vision 3D reconstruction algorithm, calculates the 3D spatial coordinates W of three reference points PGO, PG1, and PG2 in the camera coordinate system. PCO W PG1 W PG2 Among them, W PCO =[W PCOX W PCOY W PCOZ ] TW PC1 =[W PC1X W PC1Y W PC1Z ] T W PC2 =[W PC2X W PC2Y W PC2Z ] T ; S3.3, let W PC1O =W PC1 -W PCO W PC2O =W PC2 -W PCO W PC1O W PC2O Normalized to a unit vector W PC1O_unit W PC2O_unit Then construct the rotation and translation matrix R. I2C : ; in, Represents the cross product of vectors; S3.4, using the rotation and translation matrix R I2C Will Transform to the camera coordinate system to obtain the simulation results of the crack centerline in the camera coordinate system. ; width The rotation and translation matrix R is not affected. I2C The influence, therefore the width in the camera coordinate system = ; S3.5, the measured crack centerline value The crack centerline value compared with the finite element simulation Perform matching and filtering and All successfully paired points in the array are denoted as follows: and ; Furthermore, and Pairing method: Find the relationship between each 3D point in the data and... Find the nearest neighbor in the array and calculate the Eulerian distance between these two points. If the distance is less than a threshold, the match is considered successful; if no nearest neighbor that meets the threshold is found, the pairing fails and the point is discarded.

[0041] S3.6, Since the crack width value and the crack centerline value are in one-to-one correspondence, based on the pairing results of the centerlines... and ,turn up and Pairing results and Alignment complete; S3.7, calculate the error_match between the measured and simulated values ​​of the crack centerline and crack width. Error_match is characterized by the matching success rate, i.e.: Error_match=1-max(PERCENT1, PERCENT2); PERCENT1=NUM( ) / NUM( ); PERCENT2=NUM( ) / NUM( ); Where NUM() represents the number of values ​​a variable can take, and max() represents taking the maximum value; the lower the value of Error_match, the better the finite element simulation result; the larger the value of Error_match, the worse the error simulation result.

[0042] S4, establish global optimal constraints for multi-source data fusion, such as Figure 10 As shown, the three-dimensional size information of the crack was obtained by collecting monitoring data from the bridge construction monitoring system: a collection of measured strain values ​​from different locations. A collection of measured displacement values ​​from different locations And extract the set of strain simulation values ​​at corresponding locations in the finite element cracking network model. and the set of displacement simulation values The normalization error between the measured strain value and the simulated strain value, Error_S, and the normalization error between the measured displacement value and the simulated displacement value, Error_Dis, are calculated using methods such as the NRMSE maximum value normalization model. Construct a multi-objective global error function: ; in, Indicates the multiplication sign. This indicates the weight of the crack width error Error_match. This represents the weight of the strain error Error_S. This indicates the weight of the displacement error Error_Dis; , usually take ; if If the value is less than a specified threshold (the threshold is determined based on accuracy requirements, such as 0.05), the finite element simulation accuracy is considered good, and no further optimization of the finite element cracked mesh model is needed. The crack depth information can be directly extracted from the finite element cracked mesh model using methods such as the interior point method, denoted as . , and , Together, they constitute the three-dimensional dimensional information of the crack; if If the value does not meet the requirements, then commonly used classical methods such as adjoint gradient optimization and Bayesian inversion are used to correct the finite element cracked mesh model until... The value is less than the specified threshold.

[0043] Unlike traditional single-vision measurement (which only provides length and width geometric information), this invention employs multi-source data fusion (providing length, width distribution, and depth geometric information). While conventional methods for obtaining crack depth information require ultrasonic assistance, this invention uses a threshold to determine the accuracy of the finite element cracking network model, enabling direct extraction of crack depth information from the model with relatively small errors. This invention can adapt to the three-dimensional measurement of critical cracks throughout the entire construction period, significantly improving the automation, accuracy, and engineering practicality of crack size inversion during bridge construction.

Claims

1. A method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion, characterized in that, Includes the following steps: S1. 3D Reconstruction of Crack Appearance Contour Based on Binocular Vision and Semantic Segmentation: Binocular images of cracks during bridge construction are acquired. Pixel masks of the crack regions are extracted from these images. The crack centerline and edge lines are extracted from the pixel masks. The corresponding 3D point clouds of the crack edge and centerline are reconstructed, forming a 3D point cloud set of the measured contour in the camera coordinate system. and the three-dimensional point cloud set of the measured center line ,pass Get the crack width Variation along the length direction; S2, Finite Element Mechanical Forward Analysis with Measured Profile as Boundary Condition: Loads during bridge construction are collected, a finite element cracking mesh model is established, the loads are added to the finite element cracking mesh model to obtain the cracking results under the current load, and the three-dimensional spatial coordinate set of the crack centerline in the finite element coordinate system is extracted. and the crack width in the finite element coordinate system ; Node coordinates are three-dimensional coordinates based on the finite element coordinate system. , The method for extracting the corresponding centerline coordinates and the width value along the centerline includes the following steps: S2.1, find nodes PO, PX, and PY within the finite element plane where the crack is located, such that the vector... Vertical vector , and All are unit vectors arranged in a row and a column; S2.2, Construct the rotation and translation matrix R: ； in, It is a vector and The cross product, where T represents the transpose; S2.3, The three-dimensional coordinates are calculated using the rotation and translation matrix R. Projecting this onto a two-dimensional plane yields the set of crack profile coordinates within that plane. : [ ]=R*[ ] T ； in, The true value is always equal to a certain fixed constant C, and its absolute value is equal to the vertical distance from the origin of the finite element coordinate system to the plane where the crack is located. S2.4, Identification The two endpoints along the major axis, based on these two endpoints Divide into two, represented as , Then to , Pair the points in the graph to determine the set of all two-dimensional points on the crack centerline in the two-dimensional plane. Then, in a two-dimensional plane, using and The two-dimensional width value corresponding to the center point of each crack is calculated to obtain the crack width on the two-dimensional plane. The variation along the length of the centerline is denoted as ; S2.5, for and Performing an inverse transformation yields the set of three-dimensional spatial coordinates of the crack centerline in the finite element coordinate system. and the crack width in the finite element coordinate system : ； ； ； = ； S3, calculate the error between the measured and simulated values ​​of the crack centerline and crack width: The measured values ​​obtained in S1... , The simulated values ​​obtained from S2 , After alignment, calculate and , and Error_match between them; S4. Establish global optimal constraints for multi-source data fusion to obtain three-dimensional crack size information: collect monitoring data from the bridge construction monitoring system: a collection of measured strain values ​​from different locations. A collection of measured displacement values ​​from different locations And extract the set of strain simulation values ​​at corresponding locations in the finite element cracking network model. and the set of displacement simulation values The normalized error Error_S between the measured strain value and the simulated strain value, and the normalized error Error_Dis between the measured displacement value and the simulated displacement value are calculated. A multi-objective global error function is constructed to determine the accuracy of the finite element cracking mesh model, and crack depth information values ​​are extracted from it to form the three-dimensional size information of the crack.

2. The method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion according to claim 1, characterized in that, Constructing a multi-objective global error function in S4: ； In this context, · represents a multiplication sign. This indicates the weight of the crack width error Error_match. This represents the weight of the strain error Error_S. This indicates the weight of the displacement error Error_Dis; ; if If the value is less than the specified threshold, the finite element simulation accuracy is considered good, and the crack depth information can be directly extracted from the finite element cracked mesh model, denoted as . , and , Together, they constitute the three-dimensional dimensional information of the crack; if If the value does not meet the requirements, the finite element cracked mesh model is modified until... The value is less than the specified threshold.

3. The method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion according to claim 1, characterized in that, The specific steps in S1 include: S1.1, The binocular camera was calibrated using the Zhang Zhengyou calibration method, and then the binocular camera was used to acquire binocular images of cracks during the bridge construction process; S1.2, use a semantic segmentation network to extract the pixel mask of the crack region in the binocular image of the crack, then use a binary image morphology algorithm to extract the crack center line in the pixel mask, and use the Canny edge detection algorithm to extract the crack edge line in the pixel mask; S1.3, using the principles of stereo matching and triangulation, reconstructs the three-dimensional point cloud corresponding to the crack edge line and crack center line, forming a three-dimensional point cloud set of the measured contour: 3D point set of the measured center line: ; S1.4, through Get the crack width The variation along the length direction is denoted as Therefore, there is a one-to-one correspondence between the points on the center line of the crack and its width, that is... = .

4. The method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion according to claim 1, characterized in that, S2.4 , The point pairing process in the process includes: pairing Each point PD on -left ,exist Find the point PD with the closest Euclidean distance in the middle. -right As a pairing, after pairing is completed, PD is calculated. -left PD -right The midpoint between the two points is used as a point on the center line of the crack.

5. The method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion according to claim 1, characterized in that, The specific steps for aligning measured values ​​with simulated values ​​include: S3.1, in the finite element cracked network model, select three reference points PGO, PG1, and PG2, satisfying: vector Vertical vector ,in, , It is a unit vector with three rows and one column; S3.2, using a binocular vision 3D reconstruction algorithm, calculates the 3D spatial coordinates W of three reference points PGO, PG1, and PG2 in the camera coordinate system. PCO W PG1 W PG2 Among them, W PCO =[W PCOX W PCOY W PCOZ ] T W PC1 =[W PC1X W PC1Y W PC1Z ] T W PC2 =[W PC2X W PC2Y W PC2Z ] T ; S3.3, let W PC1O =W PC1 -W PCO W PC2O =W PC2 -W PCO W PC1O W PC2O Normalized to a unit vector W PC1O_unit W PC2O_unit Then construct the rotation and translation matrix R. I2C : ； in, Represents the cross product of vectors; S3.4, using the rotation and translation matrix R I2C Will Transform to the camera coordinate system to obtain the simulation results of the crack centerline in the camera coordinate system. ; width The rotation and translation matrix R is not affected. I2C The influence, therefore the width in the camera coordinate system = ; S3.5, the measured crack centerline value The crack centerline value compared with the finite element simulation Perform matching and filtering and All successfully paired points in the array are denoted as follows: and ; S3.6, based on the pairing results of the center lines and ,turn up and Pairing results and Alignment complete; S3.7, calculate the error_match between the measured and simulated values ​​of the crack centerline and crack width. Error_match is characterized by the matching success rate, i.e.: Error_match=1-max(PERCENT1, PERCENT2); PERCENT1=NUM( ) / NUM( ); PERCENT2=NUM( ) / NUM( ); Where NUM() represents the number of values ​​a variable can take, and max() represents taking the maximum value; the lower the value of Error_match, the better the finite element simulation result; the larger the value of Error_match, the worse the error simulation result.

6. The method for measuring the three-dimensional dimensions of bridge cracks based on multi-source data fusion according to claim 5, characterized in that, S3.5 and Pairing method: Find the relationship between each 3D point in the data and... Find the nearest neighbor in the array and calculate the Eulerian distance between these two points. If the distance is less than a threshold, the match is considered successful; if no nearest neighbor that meets the threshold is found, the pairing fails and the point is discarded.

Images