Large steel structure deformation monitoring method based on TLS technology

Through the data acquisition and processing method based on TLS technology, combined with multi-link node extraction and least squares fitting algorithm, the problems of low efficiency and low accuracy in deformation monitoring of large complex steel structures are solved, and efficient, automatic, Accurate deformation monitoring meets the deformation monitoring needs of large steel structures.

CN119063643APending Publication Date: 2024-12-03HENAN POLYTECHNIC UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411150342.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing steel structure deformation monitoring technology has problems such as low efficiency, low accuracy, high cost and susceptibility to environmental factors in large and complex steel structures, especially in industrial environments, which are difficult to achieve comprehensive and accurate monitoring.

Method used

Using a large steel structure deformation monitoring method based on TLS technology, it can achieve efficient through data acquisition and preprocessing, multi-link node extraction algorithm, iterative shrinkage filtering and rotary projection filtering algorithm, least squares global search fitting algorithm and displacement calculation model. , automatic and accurate deformation monitoring.

Benefits of technology

It realizes efficient and automated node point cloud extraction and classification, accurately estimates the spherical node center, establishes a displacement calculation model, improves the monitoring accuracy and efficiency, meets the deformation monitoring needs of large steel structures, and reduces costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119063643A_ABST
    Figure CN119063643A_ABST
Patent Text Reader

Abstract

The invention discloses a large-scale steel structure deformation monitoring method based on a TLS technology. The method comprises the steps of data acquisition and preprocessing; extracting node point clouds by adopting a multi-link node extraction algorithm; separating spherical node point clouds by constructing iterative shrinkage filtering and rotating projection filtering algorithms; a global search fitting algorithm based on least squares accurately estimates the coordinates of the center of sphere; a displacement calculation model is constructed by applying a neighbor similarity theory, and safety monitoring of the large-scale steel structure is realized; and performing precision evaluation. The invention provides an efficient, automatic and accurate method for carrying out deformation monitoring on the large steel structure by using multi-period point cloud data. According to the method, a ball node center is used as a virtual monitoring point. According to the method, the ball nodes are accurately positioned, and displacement of the ball nodes in different periods is calculated based on neighbor similarity hypothesis. And the three-stage grid point cloud data is used for verifying the validity of the proposed method. The method has obvious advantages in efficiency, can meet the precision requirement, and can quickly and accurately monitor the deformation of the steel structure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of steel structure deformation monitoring, and more particularly to a large steel structure deformation monitoring method based on TLS technology. Background Art

[0002] Steel structure is a building material with excellent performance. It has the advantages of high strength, light weight, fast construction, low cost, good earthquake resistance, etc. It has been widely used in large factories, high-rise buildings and other fields. However, due to the influence of self-weight, wind, temperature, corrosion and other factors, steel structure may deform during long-term use, which will reduce the stability and safety of the structure and even cause serious safety accidents and huge economic losses. Therefore, it is very important to monitor the deformation of steel structures, especially large steel structures.

[0003] In actual deformation monitoring tasks, it is necessary to comprehensively consider multiple factors to select one or more appropriate deformation monitoring technologies. Current deformation monitoring technologies mainly include total station measurement (Zhou et al., 2003), sensor measurement (Zhu et al., 2022), terrestrial laser scanning (TLS) (Dong et al., 2020), photogrammetry (Hu et al., 2021), interferometry (Qin et al., 2021), GPS measurement (Yi et al., 2013), etc. Large-scale steel structure workshops in industry have complex spatial structures. In order to ensure the comprehensiveness of deformation monitoring, the monitoring range is wide and there are many monitoring points. In this case, the total station measurement operation is time-consuming, the number of sensors required is large and the installation is difficult. Some uncertain factors, such as temperature and vibration, will interfere with sensor measurement, and the sensors need regular maintenance and calibration, which will increase the difficulty and cost of measurement. Photogrammetry requires a large number of images to cover the entire structure, which increases the complexity and time cost of data processing, and the technology is greatly affected by factors such as light. The nature of the monitored object, the environment and other factors will affect the stability and reliability of the interferometric measurement signal, increasing the difficulty of monitoring. The accuracy of GPS measurement is easily affected by uncontrollable factors such as weather conditions and surrounding environment (Ning Ronghua et al., 2023), and it cannot be used in indoor spaces (the signal is poor). Compared with the above measurement technologies, TLS technology is very suitable for monitoring the deformation of large and complex steel structures.

[0004] TLS technology is an active remote sensing technology that can quickly generate very dense point cloud data on the surface of the object being measured. It has the advantages of non-contact, high precision, high efficiency, wide coverage, and convenient data collection (Shen et al., 2023). It plays a huge role in three-dimensional reconstruction of buildings (Ochmann et al., 2019), cultural heritage protection (Yang et al., 2022), and disaster measurement (Vetrivel et al., 2018).

[0005] The geometric shapes of large and complex steel structure components include but are not limited to plane, spherical, cylindrical, etc. Accurately and efficiently extracting and locating these key components from the point cloud of large and complex steel structures is of great significance for structural health monitoring and modeling. Summary of the invention

[0006] Therefore, the technical problem to be solved by the present invention is to provide an efficient, automatic and accurate large-scale steel structure deformation monitoring method based on TLS technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] The large-scale steel structure deformation monitoring method based on TLS technology includes the following steps:

[0009] (1) Data collection and preprocessing;

[0010] (2) Use multi-link node extraction algorithm to extract node point cloud;

[0011] (3) The ball node point cloud is separated by constructing iterative shrinkage filtering and rotation projection filtering algorithms;

[0012] (4) Accurately estimate the coordinates of the sphere center using a least squares global search fitting algorithm;

[0013] (5) Use the nearest neighbor similarity theory to construct a displacement calculation model to achieve safety monitoring of large steel structures;

[0014] (6) Conduct accuracy assessment.

[0015] In the above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, in step (1), measuring stations are arranged on-site at the large-scale steel structure, and point cloud data are collected in stages using a ground laser scanner;

[0016] The preprocessing method is as follows: first, use the software RiSCAN PRO that comes with the terrestrial laser scanner to align the point clouds between different measuring stations; then, use the software CloudCompare to separate the grid point cloud and the wall point cloud in the large structure point cloud, and use the iterative closest point algorithm ICP algorithm to align the wall point cloud after the second phase to the coordinate system of the wall point cloud of the first phase, and record the transformation matrix; finally, use the transformation matrix to align the grid point cloud after the second phase to the coordinate system of the grid point cloud of the first phase.

[0017] The above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, in step (2), uses a downsampling method to extract point clouds, including the following steps:

[0018] (2-1) Based on the original point cloud C raw Construct a KD tree based on the downsampled point cloud C sample Construct a KD tree and put p i ∈C sample ,i=1,2,…,n is marked as false; among them, C raw represents the original point cloud, C sample Represents the downsampled point cloud, n is C sample The total number of midpoints, p i C sample The i-th point in

[0019] (2-2) Select a point p marked as false in order i ∈C sample , with point p i is the query point, R outer is the search radius, and the neighbor point set Q is obtained by radius search near , calculate the neighbor point set Q near All points to points p i The Euclidean distance of inner The points are saved to the new point set Q shell ; R outer and R inner The calculation formula is as follows:

[0020]

[0021] Among them, d minN is the minimum Euclidean distance between two nodes in the spatial grid structure, R outer is the outer radius, R inner is the inner radius;

[0022] (2-3) For the new point set Q shell Perform Euclidean clustering, R cluster is the cluster radius, N clusteris the minimum number of points per class, clusters = {cluster i |i=1,2,…,m} is the clustering result; R cluster and N cluster The calculation formula is as follows:

[0023]

[0024] Among them, r minC is the minimum radius of the cylindrical component in the spatial grid structure; d minS is the minimum Euclidean distance between two points in the sampled point cloud; (the original point cloud is downsampled using the built-in spatial downsampling method of the cloudcompare software. This method requires a parameter, which indicates that the minimum Euclidean distance between points in the sampled point cloud is d minS ; and d minN Represents the minimum Euclidean distance between nodes in a spatial grid structure. )

[0025] (2-4) If m≤2, then the neighbor point set Q near Mark all points in as true and return to step (2-2);

[0026] (2-5) According to the LS principle, each point cluster is fitted as a cylinder, and the corresponding root mean square error (RMSE) is calculated; if the RMSE of a point cluster is ≤ 0.03, the point cluster is considered to be cylindrical, and its central axis L is saved in the central axis set CAS = {L i |i=1,2,…,k≤m};

[0027] (2-6) The virtual intersection of each pair of central axis lines with an angle greater than 20° in the central axis set CAS is calculated using the calculation principle of the virtual intersection of non-coplanar straight lines, and saved in the virtual intersection set (The number of clusters obtained by Euclidean clustering is m, but the number of clusters with a shape similar to a cylinder (RMSE ≤ 0.03) (hereinafter referred to as cylindrical clusters) must be less than or equal to m; select any two of these k cylindrical clusters, with a total of However, the number of combinations where the angle between the central axes of the two cylinder clusters is greater than 20° must be less than or equal to )

[0028] (2-7) If the virtual intersection point set VPS is empty, return to step (2-2) and calculate the centroid p of the virtual intersection point set VPS centroid ; If the center of mass p centroid The Euclidean distance to other centroids in the virtual intersection centroid set VPCS is ≤ R outer , then return to step (2-2);

[0029] (2-8) Save the centroid pcentroid to the virtual intersection centroid set VPCS, and C sample The distance from the center of mass is <0.9d minN The point is marked as true;

[0030] (2-9) with p centroid For the query point, follow step (2-2) to obtain the new Q shell , obtain new clusters according to steps (2-3), fit all point clusters in the new clusters into cylinders according to the LS principle, and determine the maximum radius of the cylinder, which will be used as the input parameter of the iterative shrinkage filtering algorithm

[0031] (2-10) with p centroid is the query point, 1.5r maxS is the search radius, r maxS is the maximum value of the radius of the spherical component. The neighboring point set is obtained by radius search, and this point set is the node point cloud.

[0032] In the above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, in step (3), the node point cloud obtained in step (2) is divided into two types: conventional node point cloud and ball node point cloud; it is necessary to extract a smooth spherical point cloud from the ball node point cloud;

[0033] The iterative shrinkage filtering algorithm is: the input parameter is the node point cloud C node , the maximum number of iterations N ISF , the maximum radius of the cylinder Distance threshold d ε ;

[0034] Algorithm process: First obtain the node point cloud C node A copy of point cloud C ISF , record the copy point cloud C ISF Then use the LS algorithm to fit C ISF is a sphere, with (x, y, z) as the coordinates of the sphere center and r as the radius of the sphere. ISF Center of middle distance ≤ r+d ε Save the points back to C ISF , record C ISF The number of points; keep shrinking iteratively until the two C ISF The number of points does not change or the maximum number of iterations is reached; at this time, if the radius Then the node point cloud C node It is identified as a spherical node point cloud, thereby filtering out most of the regular node point clouds; the result C of the last iteration shrinkage is output ISF , with C ISF Serves as input data for the rotation projection filtering algorithm.

[0035] The above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, rotation projection filtering algorithm: point cloud filtered by iterative shrinkage filtering algorithm ISF, first copy a copy of the input point cloud C ISF , and centroidize the copy, then rotate the copy and project it onto the XOY plane, use LS to fit the copy to a circle and calculate the RMSE until the copy is rotated to the specified angle, and finally compare the minimum RMSE with the RMSE threshold to determine the node point cloud C corresponding to the input point cloud node Type.

[0036] In the above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, in step (4), a global search fitting algorithm based on least squares is used:

[0037] Input parameters: Point cloud C ISF , the maximum number of iterations N max , the threshold value of standard deviation change Δσ ε ;

[0038] Get a C ISF A copy of C GSF , C GSF Fit it to a sphere, obtain the center (x, y, z) and radius r and calculate the standard deviation σ; if the change in the standard deviation is greater than Δσ ε , clear C GSF And C ISF Points whose distance from the center to the sphere is less than 3 times the standard deviation are saved in C GSF ; until the standard deviation change is less than or equal to Δσ ε Or until the maximum number of iterations is reached; C GSF Output point cloud for the algorithm.

[0039] In the above large-scale steel structure deformation monitoring method based on TLS technology, in step (5), the displacement calculation model directly uses the following formula to calculate the vertical displacement and horizontal displacement:

[0040]

[0041] Among them, i and j represent the ball node centers of different periods, k is the ball node center number, is the vertical displacement; is the horizontal displacement, is the center of the ball node in the horizontal direction j, is the center of the ball node in the horizontal direction i; is the center of the ball node in the vertical direction j; is the center of the ball node in the vertical direction i;

[0042] Determine the three points closest to p (p1, p2, p3). If the direct displacement value of p is greater than that of the other three points, recalculate the displacement value of p according to the following formula, while the displacement direction of p remains unchanged.

[0043]

[0044] in D1, D2 and D3 are the displacement values ​​of p1, p2 and p3 respectively.

[0045] In the above-mentioned large-scale steel structure deformation monitoring method based on TLS technology, in step (6), the node extraction rate is used to evaluate the node extraction effect;

[0046] The calculation formula of node extraction rate NER is as follows:

[0047]

[0048] Among them, n node is the number of nodes extracted, N node is the total number of nodes in the spatial grid structure.

[0049] The above-mentioned large-scale steel structure deformation monitoring method based on TLS technology uses a confusion matrix to evaluate the node classification effect; the true positive case TP is defined as the number of nodes that correctly predict the spherical nodes as spherical nodes, the false negative case FN is defined as the number of nodes that incorrectly predict the spherical nodes as regular nodes, the false positive case FP is defined as the number of nodes that incorrectly predict the regular nodes as spherical nodes, and the true negative case TN is defined as the number of nodes that correctly predict the regular nodes as regular nodes;

[0050] The calculation formulas for Producer's Accuracy, PA, User's Accuracy, UA, Overall Accuracy, Po, and Kappa Coefficient, K are as follows:

[0051]

[0052] in,

[0053] The large-scale steel structure deformation monitoring method based on TLS technology according to claim 1 is characterized in that the root mean square error Root Mean Square Error, RMSE, is used as a quantitative indicator to evaluate the fitting accuracy of the least squares global search fitting algorithm GSF;

[0054]

[0055] Among them, di is the distance from the point to the center of the fitting sphere, r fit is the radius of the fitting sphere.

[0056] The technical solution of the present invention achieves the following beneficial technical effects:

[0057] (1) An efficient and automated node point cloud extraction method is proposed. This method combines the least squares principle, Euclidean clustering, and non-planar straight lines to effectively estimate the node position and range.

[0058] (2) An automatic, fast and accurate classification method for node point clouds is proposed. The method consists of iterative shrinkage filtering and rotational projection filtering. Iterative shrinkage filtering can simplify the shape of node point clouds, making the differences between different types of node point clouds more obvious while ensuring the integrity of the ball node point clouds. Rotational projection filtering makes up for the shortcomings of iterative shrinkage filtering and further accurately separates the ball node point clouds.

[0059] (3) A method for accurately locating spherical nodes is proposed. Considering the negative impact of noise and the positive impact of non-noise, this paper proposes a global search fitting algorithm based on the least squares principle to accurately estimate the center of the spherical node.

[0060] (4) A displacement calculation model is established and a displacement analysis is performed on a large steel structure to assess its health status. There is still a slight error in the center of the spherical node, and the reliability of the directly calculated displacement is low. The proposed model takes into account the characteristics of the steel structure and the reasonable assumption that the difference in displacement values ​​of adjacent nodes is small.

[0061] This application proposes an efficient, automatic and accurate method for deformation monitoring of large steel structures using multi-period point cloud data. This method uses the center of the ball node as a virtual monitoring point. The first main step of the proposed method is the accurate positioning of the ball node, and the second main step is to calculate the displacement of the ball node in different periods based on the nearest neighbor similarity assumption. Three-phase grid point cloud data were used to verify the effectiveness of the proposed method. It was monitored that the grid sank by 1.82 cm as a whole from 2021 to 2023, and the truss unit had good stability. Compared with traditional methods, the method in this paper has obvious advantages in efficiency and is sufficient to meet the accuracy requirements, and can realize fast and accurate monitoring of steel structure deformation. Future work should try to realize the adaptive parameters of the proposed method and verify the applicability of the proposed method in other similar large steel structures. Among them, the proposed MNEM can also be applied to other similar structures, such as tree trunk bifurcation point identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 For a large-span dry coal shed in a power plant, (a) satellite image, (b) and (c) internal scenes at different locations, (d) conventional node, (e) ball node;

[0063] Figure 2 . is the station distribution map. The red dots represent the control point stations, whose PPR is 300kHz (single station takes about 12 minutes), and the black dots represent the non-control point stations, whose PPR is 75kHz (single station takes about 3 minutes). a is the collection date 2021.10.29, b is the collection date 2022.11.18, and c is the collection date 2023.10.25;

[0064] Figure 3 The grid point clouds of the three phases in the same coordinate system; red represents the first phase, green represents the second phase, and blue represents the third phase;

[0065] Figure 4 This is a flow chart of deformation monitoring of large steel structures based on TLS technology;

[0066] Figure 5 Schematic diagram of the virtual intersection calculation principle

[0067] Figure 6 Extract flow chart for mesh point cloud;

[0068] Figure 7a It is the downsampled truss point cloud; Figure 7b The point cloud is a cylindrical point cloud with point clusters; Figure 7c From the axis L to the set CAS; Figure 7d It is the virtual intersection of the two central axes; Figure 7e is the centroid set CS; Figure 7f is the node point cloud;

[0069] Figure 8a is a general node point cloud, Figure 8b It is a spherical node point cloud;

[0070] Fig. 9 It is the processing flow of node point cloud;

[0071] Fig.10 It is the iterative shrinking process of general node point cloud;

[0072] Fig.11 It is the iterative shrinkage process of the spherical node point cloud;

[0073] Fig.12 is the node distribution in the spatial grid structure;

[0074] Fig.13 is the parameter d minS The impact of

[0075] Fig.14 This is the node extraction effect. The right picture is a local enlarged picture. The blue represents the node point cloud.

[0076] Fig.15 Extract results for the nodes of phase three;

[0077] Fig.16a is the node classification effect in the first phase, Fig.16b is the node classification effect of the second phase, Fig.16c This is the node classification effect of the third phase; Fig.16d This is the separation effect of the first phase of manual separation; Fig.16e This is the separation effect of the second phase of manual separation; Fig.16f This is the separation effect of the third phase of manual separation;

[0078] Fig.17 Classify and evaluate nodes;

[0079] Fig.18 The comparative experimental results of the spherical node point cloud advance method and the RANSAC algorithm of this application;

[0080] Fig.19 is the statistical distribution of RMSE of the three phases;

[0081] Fig.20a is the vertical displacement diagram from the first period to the second period, Fig.20b It is the horizontal displacement diagram from the first period to the second period; Fig.20c is the vertical displacement diagram from the second period to the third period, Fig.20d This is the horizontal displacement diagram from the second period to the third period; Fig.20e is the vertical displacement diagram from the first to the third period, Fig.20f It is the horizontal displacement diagram from the first to the third period;

[0082] Fig.21 is the upper chord and its corresponding ball node position;

[0083] Fig.22a It is the elevation variation trend diagram of the upper chord A in three phases; Figure 22b It is the elevation variation trend diagram of the upper chord B in three phases;

[0084] Fig.23 This is the elevation change trend diagram of the horizontal rod in the third phase DETAILED DESCRIPTION

[0085] 1. Data collection and preprocessing;

[0086] 1.1 Data Collection

[0087] The research area is a large-span dry coal shed in a power plant ( Figure 1 ). The dry coal shed is mainly composed of reinforced concrete walls, steel structure grids, and color steel. The grid thickness is about 5 meters, and the grid mainly includes 16 transverse arch trusses and 7 longitudinal trusses. The trusses are composed of cylindrical components and spherical components. The cylindrical components directly intersect to form a regular node, and the cylindrical components intersect with the spherical components to form a spherical node.

[0088] The ground laser scanner Riegl VZ-1000 was used to collect three phases of point cloud data, with the collection dates being 2021.10.29, 2022.11.18, and 2023.10.25. Riegl VZ-1000 uses a thin infrared laser beam to perform high-speed, high-resolution, high-precision, and large-range scanning, and can detect multiple targets even in harsh environments. Table 1 shows the main parameters of the scanner.

[0089] Table 1

[0090]

[0091] Under the premise of meeting the point cloud density, accuracy and coverage, measurement stations are arranged in combination with on-site conditions. Figure 2 This is the distribution map of measuring stations.

[0092] 1.2 Data Preprocessing

[0093] First, the software RiSCAN PRO that comes with the scanner was used to align the point clouds between different measuring stations. The standard deviations of the three phases of alignment were 3.7 mm, 3.9 mm, and 3.1 mm, respectively. Then, the software CloudCompare was used to separate the grid point cloud and the wall point cloud in the dry coal shed point cloud. Next, the iterative closest point algorithm ICP was used to align the wall point clouds of the second and third phases to the coordinate system of the wall point cloud of the first phase, and the transformation matrix was recorded. Finally, the transformation matrix was used to align the grid point clouds of the second and third phases to the coordinate system of the grid point cloud of the first phase.

[0094] 2. Methodology

[0095] The workflow of the proposed method is as follows Figure 4 As shown. According to the characteristics of the spatial grid structure, we first determine the approximate coordinates of the node center, and then extract the node point cloud from the original point cloud. We use two filtering algorithms to filter out the spherical node point cloud from the node point cloud and simplify it into a rough spherical point cloud. Then, the proposed spherical fitting algorithm is used to estimate the spherical center coordinates of the spherical node. Finally, based on these spherical center coordinates, we analyze the stability of a single truss and use the proposed displacement calculation model to analyze the vertical and horizontal displacements of the grid structure.

[0096] (1) Introduce the calculation principle of the virtual intersection of non-coplanar straight lines.

[0097] In three-dimensional space, the vector form of the parametric equation of a line is:

[0098] l=l0+vt #(8)

[0099] Where l is the position vector of any point on the line, l0 is the vector of a known point on the line, v is the direction vector of the line, and t is a scalar parameter.

[0100] like Figure 5 As shown, L1 and L2 are skew lines, L′1 / / L1, L′2 / / L2, L3 / / L4 / / L5, and p is the position vector of the point to be determined.

[0101] Calculate the length of the common perpendicular segment d(o1o2) and the unit direction vector v3 using the following formula:

[0102]

[0103] Where v1 and v2 are the direction vectors of L1 and L2 respectively, p1 is the position vector of a point on L1, and p2 is the position vector of a point on L2.

[0104] According to the geometric meaning of the parameters of the parametric equation, there are two position vectors (p′2andp″2) on L4 that are d away from p2. Similarly, on L5, there are two position vectors (p′1and p″1) that are d away from p1. According to the spatial geometric relationship, p′2 is closer to p1 than p″2; p′1 is closer to p2 than p″1. The calculation formula is as follows:

[0105]

[0106] L′2 and L1 intersect at o1, and L′1 and L2 intersect at o2. Calculate o1 according to formula (12); calculate o2 according to formula (13).

[0107] p′2+v2t2′=p1+v1t1#(12)

[0108] Where t2′ is the scalar parameter of line L′2, and t1 is the scalar parameter of line L1.

[0109] p′1+v1t1′=p2+v2t2#(13)

[0110] Where t1′ is the scalar parameter of line L′1, and t2 is the scalar parameter of line L2.

[0111] Calculate p according to the following formula:

[0112]

[0113] Where e1 and e2 are the spatial attitude errors of the lines L1 and L2 respectively. In this paper, the error is the root mean square error (RMSE) obtained when fitting the cylinder using the least squares (LS).

[0114] (2) Use multi-link node extraction algorithm to extract node point cloud;

[0115] According to the characteristics of spatial grid, this paper proposes a multi-link node extraction method (MNEM), which can realize the fast and automatic extraction of node point cloud. Figure 6 ).

[0116] The density of the original truss point cloud is relatively high, and data processing is very time-consuming. In order to improve the data processing speed, the original truss point cloud is downsampled. The downsampling method selected in this paper is the built-in Space method of CloudCompare software (which requires a parameter d representing the minimum distance between two points). minS ), because this method does not change the coordinates of the points and does not seriously affect the local detail features of the truss.

[0117] In this section, C raw represents the original truss point cloud, C sample Represents the downsampled truss point cloud. The specific steps of MNEM are as follows:

[0118] (2-1) Based on C raw Construct KD tree based on C sample Construct a KD tree and put p i ∈C sample ,i=1,2,…,n is marked as false; among them, C raw represents the original point cloud, C sample Represents the downsampled point cloud, n is C sample The total number of midpoints, p i C sample The i-th point in .

[0119] (2-2) Select a point p marked as false in order i ∈C sample ( Figure 7a , green dot), with point p i is the query point, R outer is the search radius, and the neighbor point set Q can be obtained by radius search near , calculate the neighbor point set Q near All points to points p i The Euclidean distance of inner The points are saved to the new point set Q shell ( Figure 7a , red dots). R outer and R inner The calculation formula is as follows:

[0120]

[0121] where d minN is the minimum distance between two nodes in the spatial grid structure, R outer is the outer radius, R inner is the inner radius.

[0122] (2-3) For the new point set Q shell Perform Euclidean clustering (R cluster is the cluster radius, N cluster is the minimum number of points per class), clusters = {cluster i |i=1,2,…,m} is the clustering result. cluster and N cluster The calculation formula is as follows:

[0123]

[0124] where r minC is the minimum radius of the cylindrical component in the spatial grid structure, d minS is the minimum Euclidean distance between two points in the sampled point cloud;

[0125] The original point cloud is downsampled using the built-in spatial downsampling method of the cloudcompare software. This method requires a parameter, which indicates that the minimum Euclidean distance between points in the sampled point cloud is d. minS ; and d minN Represents the minimum Euclidean distance between nodes in a spatial grid structure.

[0126] (2-4) If m≤2, then Q near Mark all points in as true and return to step (2-2).

[0127] (2-5) According to the LS principle, each point cluster is fitted into a cylinder and the corresponding root mean square error (RMSE) is calculated. If the RMSE of a point cluster is ≤ 0.03, the point cluster is considered to be cylindrical ( Figure 7b ), and its central axis L( Figure 7c )Save to the central axis set CAS = {L i |i=1,2,…,k≤m};

[0128] (2-6) The virtual intersection of two central axes with an angle greater than 20° in the central axis set CAS is calculated using the calculation principle of the virtual intersection of non-coplanar straight lines. Figure 7d , blue point), and save it to the virtual intersection set (The number of clusters obtained by Euclidean clustering is m, but the number of clusters with a shape similar to a cylinder (RMSE ≤ 0.03) (hereinafter referred to as cylindrical clusters) must be less than or equal to m; select any two of these k cylindrical clusters, with a total of However, the number of combinations where the angle between the central axes of the two cylinder clusters is greater than 20° must be less than or equal to )

[0129] (2-7) If the virtual intersection point set VPS is empty, return to step (2-2). Calculate the centroid p of the virtual intersection point set VPS centroid ( Figure 7d , red point). If the centroid p centroid The Euclidean distance to other centroids in the virtual intersection centroid set VPCS is ≤ R outer , then return to step (2-2);

[0130] (2-8) Save the centroid p centroid to the virtual intersection centroid set VPCS, and C sample The distance from the center of mass is <0.9d minN The point is marked as true;

[0131] (2-9) Follow step (2-2) to obtain a new Q shell (The only difference between this step and step (2-2) is: centroid For query points), follow steps (2-3) to obtain new clusters. According to the LS principle, all point clusters in the clusters are fitted into cylinders, and the maximum radius of the cylinder is determined. This radius will be used as the input parameter of the iterative shrinkage filtering algorithm.

[0132] (2-10) with p centroid ( Figure 7e , red dot) is the query point, 1.5r maxS is the search radius (r maxS is the maximum value of the radius of the spherical component), and the neighboring point set can be obtained by radius search, which is the node point cloud ( Figure 7f ).

[0133] (3) The ball node point cloud is separated by constructing iterative shrinkage filtering and rotation projection filtering algorithms;

[0134] Node point clouds can be divided into two types: regular node point clouds and spherical node point clouds, such as Figure 8a and Figure 8b shown.

[0135] Our goal is to extract smooth spherical point clouds and the corresponding spherical center coordinates from the spherical node point cloud. To this end, this paper first constructs iterative shrinkage filtering (ISF) and rotational projection filtering (RPF) for node point cloud simplification and classification. Then, considering the negative impact of noise points and the positive impact of non-noise points, this paper proposes a least squares-based global search fitting method (GSF) to accurately estimate the spherical center coordinates. The node point cloud processing process is shown in Fig. 9 .

[0136] (3-1) Iterative shrinkage filtering

[0137] The spatial shapes of regular node point clouds and spherical node point clouds are relatively complex and difficult to distinguish directly. The proposed ISF can change the spatial shapes of these two types of node point clouds and filter out most of the regular node point clouds. This method believes that through continuous iterative contraction, the number of points in the node changes dynamically, but will eventually stop changing. At this point, the type of node point cloud can be determined based on the size of the spherical model radius.

[0138] Algorithm process: First obtain the node point cloud C node A copy of point cloud C ISF , record the copy point cloud C ISF The number of points; then use the LS algorithm to fit C ISF is a sphere, with (x, y, z) as the coordinates of the sphere center and r as the radius of the sphere. ISF Center of middle distance ≤ r+d ε Save the points back to C ISF , record C ISF The number of points; keep shrinking iteratively until the two C ISF The number of points does not change or the maximum number of iterations is reached; at this time, if the radius Then the node point cloud C node It is identified as a spherical node point cloud, thereby filtering out most of the regular node point clouds; the result C of the last iteration shrinkage is output ISF , with C ISF Serves as input data for the rotation projection filtering algorithm.

[0139] The specific ISF code is as follows:

[0140] Algorithm 1. Iterative shrinkage filtering.

[0141]

[0142] This paper uses two types of node point clouds to intuitively demonstrate the working process of this method. Conventional node point cloud ( Fig.10 , input) changes dynamically as the number of iterations increases, and eventually stops changing. At this point, the shape of the regular node point cloud becomes a cylindrical sheet ( Fig.10 , green dots), ball model ( Fig.10 , the radius of the orange point Therefore, the node point cloud is identified as a regular node point cloud. When the number of points no longer changes, the spherical node point cloud ( Fig.11 , input) changes to a spherical shape ( Fig.11 , green dots), ball model ( Fig.11 , orange dot) radius Therefore, the node point cloud is identified as a spherical node point cloud.

[0143] (3-2) Rotational projection filtering

[0144] Although ISF can accurately distinguish most node point clouds, some regular node point clouds on the lower chord perpendicular to the y-axis will be misclassified as spherical node point clouds. The proposed RPF can effectively solve this problem. ISF first copies a copy of the input point cloud and centroidizes the copy. Then the copy is rotated and projected onto the XOY plane, and LS is used to fit the copy to a circle and calculate the RMSE until the copy is rotated to the specified angle. Finally, the minimum RMSE is compared with the RMSE threshold to determine the node point cloud C corresponding to the input point cloud. node Type.

[0145] RPF Code:

[0146] Algorithm 2. Rotational projection filtering.

[0147]

[0148] (4) Accurately estimate the coordinates of the sphere center using the least squares global search fitting algorithm;

[0149] After two filterings, the obtained spherical point cloud is rough, and the fitting accuracy is low when the LS algorithm is used directly, which affects the reliability of subsequent deformation analysis.

[0150] Input parameters: Point cloud C ISF , the maximum number of iterations N max , the threshold value of standard deviation change Δσ ε ;

[0151] Get a C ISF A copy of C GSF , CGSF Fit it to a sphere, obtain the center (x, y, z) and radius r and calculate the standard deviation σ; if the change in the standard deviation is greater than Δσ ε , clear C GSF And C ISF Points whose distance from the center to the sphere is less than 3 times the standard deviation are saved in C GSF ; until the standard deviation change is less than or equal to Δσ ε Or until the maximum number of iterations is reached; C GSF Output point cloud for the algorithm.

[0152] Considering the need to minimize the negative impact of noise and maximize the positive impact of non-noise, this paper proposes the GSF method. In each iteration, the algorithm selects points within 3 times the standard deviation from the input point cloud. The advantage of this method is that if a noise point is identified as a non-noise point, then in subsequent iterations, the noise point may be identified as a noise point; similarly, if a non-noise point is identified as a noise point, then in subsequent iterations, the non-noise point may be identified as a non-noise point.

[0153] GSF Code:

[0154] Algorithm 3. Global search fitting based on least squares

[0155]

[0156] (5) Use the nearest neighbor similarity theory to construct a displacement calculation model to achieve safety monitoring of large steel structures;

[0157] In complex spatial grid structures, especially grid structures made by welding, the load is evenly distributed throughout the structure through the rods, thereby avoiding excessive stress concentration in local areas and ensuring the uniformity and consistency of the entire structure. In theory, this uniformity and consistency should also be reflected in the node displacement, that is, the difference in displacement of adjacent nodes is small. Based on this, this paper proposes a displacement calculation model.

[0158] Displacement calculation model, directly use the following formula to calculate vertical displacement and horizontal displacement:

[0159]

[0160] Among them, i and j represent the ball node centers of different periods, k is the number of the ball node center, is the vertical displacement; is the horizontal displacement, is the center of the ball node in the horizontal direction j, is the center of the ball node in the horizontal direction i; is the center of the ball node in the vertical direction j period; is the center of the ball node in the vertical direction i;

[0161] Determine the three points closest to p (p1, p2, p3). If the direct displacement value of p is greater than that of the other three points, recalculate the displacement value of p according to the following formula, while the displacement direction of p remains unchanged.

[0162]

[0163] in D1, D2 and D3 are the displacement values ​​of p1, p2 and p3 respectively.

[0164] (6) Conduct accuracy assessment.

[0165] The node extraction rate (NER) is defined to quantitatively evaluate the node extraction effect of the proposed method. The calculation formula of NER is as follows:

[0166]

[0167] Among them, n node is the number of nodes extracted, N node is the total number of nodes in the spatial grid structure.

[0168] A confusion matrix is ​​constructed to quantitatively evaluate the node classification effect of the proposed method (iterative shrinkage filtering, rotational projection filtering).

[0169] True Positive (TP) is defined as the number of nodes that correctly predict spherical nodes as spherical nodes, False Negative (FN) is defined as the number of nodes that incorrectly predict spherical nodes as regular nodes, False Positive (FP) is defined as the number of nodes that incorrectly predict regular nodes as spherical nodes, and True Negative (TN) is defined as the number of nodes that correctly predict regular nodes as regular nodes.

[0170] The calculation formulas for Producer's Accuracy (PA), User's Accuracy (UA), Overall Accuracy (Po) and Kappa Coefficient (K) are as follows:

[0171]

[0172] in,

[0173] The root mean square error (RMSE) is used as a quantitative indicator to evaluate the fitting accuracy of the least squares global search fitting algorithm GSF.

[0174]

[0175] Among them, d i is the distance from the point to the center of the fitting sphere, r fit is the radius of the fitting sphere.

[0176] 3. Results and Discussion

[0177] 3.1. Node point cloud extraction

[0178] The spatial grid structure has a high degree of symmetry and similarity. After careful calculation and verification, this paper finally determined that the grid contains a total of 2936 nodes ( Fig.12 ).

[0179] The proposed node point cloud extraction method mainly includes four parameters: d minS is the point spacing, reflecting the point cloud density. minC is the minimum radius of the cylindrical component. maxS The maximum radius of a spherical component. d minN is the shortest distance between two nodes in the spatial grid structure. After measurement, r minC 0.1m, r minC 0.4m, d minN is 3m. For parameter d minS We use NER and time consumption (TC) as indicators to determine its appropriate value. The experimental results are shown in Fig.13 As shown. minS gradually increases, TC and NER gradually decrease. minS >0.05m, the decreasing trend of TC gradually slows down, and the decreasing trend of NER gradually increases. In order to ensure a higher NER and lower TC, this paper will minS The value of is set to 0.05m.

[0180] Fig.14 The extraction effect of the proposed method. The NER of the three phases are 95.7%, 97.4%, and 98.1% respectively, and the average NER is 97.0% ( Fig.15 ). Overall, our method has a good extraction effect. It is worth noting that some node point clouds are missing or sparse in the three phases of point cloud data, and the NER in this paper is calculated as 2936.

[0181] 3.2. Ball Node Center Extraction

[0182] The ISF algorithm includes three parameters: N ISF , and d ε . Normally, N ISF Taking 100 is enough; It is an automatically acquired parameter; d ε is the distance from the point to the surface of the sphere model. The larger its value, the rougher the point cloud output by the algorithm is, and the smaller its value, the smoother the point cloud output by the algorithm is. After many experiments, this paper found that d ε 0.01m is more appropriate. RPF only includes one parameter: E ε , which reflects the similarity between the input point cloud and the cylinder. In this paper, the experiment achieved good results when its value was 0.03m.

[0183] The results show that our method can distinguish most nodes, which is consistent with the manual separation ( Figure 16a-16f ).like Fig.17 As shown in the figure, the overall accuracy of the three phases (P o ) were 99.0%, 98.6%, and 99.1%, respectively, with an average P o The Kappa coefficient (K) of the three phases is 0.971, 0.958, and 0.972, respectively, and the average K is 0.967. The producer accuracy (PA) of the three phases is 96.5%, 95.4%, and 96.7%, respectively, and the average PA is 96.2%. The user accuracy (UA) of the three phases is 98.9%, 98.1%, and 99.00%, respectively, and the average UA is 98.6%. This shows that the proposed method has a good classification effect.

[0184] In order to further evaluate the classification performance of the proposed method, this paper selects 585 spherical node point clouds for comparative experiments with the RANSAC algorithm. Fig.18 The experimental results are shown in Figure 2. A total of 6 experiments were conducted, and the proposed method successfully identified 583 spherical node point clouds in each experiment. In the spherical node point cloud, non-spherical points account for a large proportion, resulting in poor recognition effect of the RANSAC algorithm. In addition, the RANSAC algorithm has a certain degree of randomness, which leads to slight fluctuations in its results. Therefore, our method is better than the RANSAC algorithm.

[0185] The GSF algorithm includes two parameters: N max andΔσ ε .Usually, N max Take 100, Δσ ε 0.0001m is sufficient. Experimental results show that ( Fig.19 ) The average RMSE of the three phases were 1.22mm, 1.25mm and 1.26mm respectively. Overall, the fitting accuracy was good.

[0186] 4. Deformation analysis

[0187] 4.1. Node displacement analysis

[0188] In order to conveniently describe the vertical displacement, the vertical displacement diagram is divided into several areas ( Figure 20a-Figure 20f ).

[0189] From the first to the second period: the A area has a large amount of subsidence, with a maximum of 6.25cm, the B area has a subsidence of about 3cm, the C area has a subsidence of about 1cm, and the D area mainly rises, with a maximum of 4.83cm. The horizontal displacement direction is consistent with the vertical displacement. The average subsidence is 1.97cm, and the average horizontal displacement is 2.19cm.

[0190] From the second period to the third period: the nodes with larger sinking values ​​are distributed in area A, and the sinking value in area B is about 1 cm. The nodes in areas C and D mainly show an upward trend. The horizontal displacement direction of the nodes in this period is consistent with the vertical displacement. The average sinking amount is -0.1 cm, and the average horizontal displacement amount is 2.75 cm.

[0191] From the first to the third stage: the nodes with obvious sinking are mainly distributed in area A, followed by area B. The sinking amount in area C is about 1 cm, and the nodes in area D rise. Similarly, the horizontal displacement direction of this stage is consistent with the vertical displacement. The average sinking amount is 1.82 cm, and the average horizontal displacement is 2.58 cm.

[0192] Analysis of deformation causes: A large amount of coal is stored in the coal shed for a long time. The weight of the coal causes the ground to sink, which in turn causes the structure to sink. The structure will also sink due to factors such as heavy rain impact and snow accumulation. If these effects are uneven, it will cause uneven deformation of the structure. The foundation soil may be affected by factors such as groundwater sources, which may cause uneven soil changes, which may also cause uneven deformation of the structure. Horizontal displacement is mainly affected by the vertical displacement of the structure and is also affected by the rigidity of the structure itself. According to relevant regulations (Industry Standards of the People's Republic of China, 2010), the maximum displacement value of the grid cannot exceed 1 / 400 of the short span. Because 19000cm / 400=47.5cm>7.30cm, it meets the requirements of the specification.

[0193] 4.2 Truss stability

[0194] In this paper, the results obtained by GSF are used to analyze the stability of a truss ( Fig.21 ), the truss includes two upper chords, each of which includes 28 ball nodes (numbered 1, 2, ..., 28).

[0195] The experimental results show that the average elevations of the upper chord A in the first, second and third periods are 129.139m, 129.104m and 129.106m, respectively. From the first period to the second period, it sank by 3.5cm, and from the second period to the third period, it rose by 0.2cm. Considering the existence of errors, this rise value can be ignored. The average heights of the upper chord B in the first, second and third periods are 129.130m, 129.099m and 129.095m, respectively. From the first period to the second period, it sank by 3.1cm, and from the second period to the third period, it sank by 0.4cm. As can be seen from Figure 22, the elevation distributions of the two upper chord nodes are almost the same. In addition, for each upper chord, the elevation change trend of the node in each period is similar, indicating that the truss has good rigidity.

[0196] The distance between the upper chords A and B corresponding to the SNC (spherical node center) is calculated, and this distance can be used as the length of the transverse rod. Fig.23 The experimental results are shown in Figure 2. From a local perspective, SNC-2, SNC-17, SNC-26, and SNC-28 are relatively abnormal, with the maximum distance differences being 1.5 cm, 1.4 cm, 1.4 cm, and 1.0 cm, respectively. The maximum distance differences of other distances are all below 1.0 cm. From a general perspective, the distance between SNCs in the first period is 478.6–481.4 cm, with an average distance of 479.82 cm. The distance between SNCs in the second period is 478.4–481.5 cm, with an average distance of 479.83 cm. The distance between SNCs in the third period is 478.4–481.4 cm, with an average distance of 479.79 cm.

[0197] In general, the truss exhibits good stability in the vertical and horizontal directions, so the truss has good overall stability.

[0198] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the claims of this patent application.

Claims

1. Large steel structure deformation monitoring method based on TLS technology, It is characterized in that The steps include: (1) Data collection and preprocessing; (2) Use multi-link node extraction algorithm to extract node point cloud; (3) The ball node point cloud is separated by constructing iterative shrinkage filtering and rotation projection filtering algorithms; (4) Accurately estimate the coordinates of the sphere center using a least squares global search fitting algorithm; (5) Use the nearest neighbor similarity theory to construct a displacement calculation model to achieve safety monitoring of large steel structures; (6) Conduct accuracy assessment.

2. According to the large-scale steel structure deformation monitoring method based on TLS technology in claim 1, It is characterized in that In step (1), measuring stations are arranged on site of the large rigid structure, and point cloud data are collected in stages using a terrestrial laser scanner; The preprocessing method is as follows: first, use the software RiSCAN PRO that comes with the terrestrial laser scanner to align the point clouds between different measuring stations; then, use the software CloudCompare to separate the grid point cloud and the wall point cloud in the large structure point cloud, and use the iterative closest point algorithm ICP algorithm to align the wall point cloud after the second phase to the coordinate system of the wall point cloud of the first phase, and record the transformation matrix; finally, use the transformation matrix to align the grid point cloud after the second phase to the coordinate system of the grid point cloud of the first phase.

3. According to claim 1, the large steel structure deformation monitoring method based on TLS technology, It is characterized in that In step (2), a downsampling method is used to extract the point cloud, including the following steps: (2-1) Based on the original point cloud C raw Construct a KD tree based on the downsampled point cloud C sample Construct a KD tree and put p i ∈C sample ,i=1,2,…,n is marked as false; Among them, C raw represents the original point cloud, C sample Represents the downsampled point cloud, n is C sample The total number of midpoints, p i C sample The i-th point in ; (2-2) Select a point p marked as false in order i ∈C sample , with point p i is the query point, R outer is the search radius, and the neighbor point set Q is obtained by radius search near , calculate the neighbor point set Q near All points to points p i The Euclidean distance of inner The points are saved to the new point set Q shell ; R outer and R inner The calculation formula is as follows: Among them, d minN is the minimum Euclidean distance between two nodes in the spatial grid structure, R outer is the outer radius, R inner is the inner radius; (2-3) For the new point set Q shell Perform Euclidean clustering, R cluster is the cluster radius, N cluster is the minimum number of points per class, clusters = {cluster i |i=1,2,…,m} is the clustering result; R cluster and N cluster The calculation formula is as follows: Among them, r minC is the minimum radius of the cylindrical component in the spatial grid structure; d minS is the minimum Euclidean distance between two points in the sampled point cloud; (2-4) If m≤2, then the neighbor point set Q near Mark all points in as true and return to step (2-2); (2-5) According to the LS principle, each point cluster is fitted as a cylinder, and the corresponding root mean square error (RMSE) is calculated; if the RMSE of a point cluster is ≤ 0.03, the point cluster is considered to be cylindrical, and its central axis L is saved in the central axis set CAS = {L i |i=1,2,…,k≤m}; (2-6) The virtual intersection of each pair of central axis lines with an angle greater than 20° in the central axis set CAS is calculated using the calculation principle of the virtual intersection of non-coplanar straight lines, and saved in the virtual intersection set (2-7) If the virtual intersection point set VPS is empty, return to step (2-2) and calculate the centroid p of the virtual intersection point set VPS centroid ; If the center of mass p centroid The Euclidean distance to other centroids in the virtual intersection centroid set VPCS is ≤ R outer , then return to step (2-2); (2-8) Save the centroid p centroid to the virtual intersection centroid set VPCS, and C sample The distance from the center of mass is <0.9d minN The point is marked as true; (2-9) with p centroid For the query point, follow step (2-2) to obtain the new Q shell , obtain new clusters according to steps (2-3), fit all point clusters in the new clusters into cylinders according to the LS principle, and determine the maximum radius of the cylinder, which will be used as the input parameter of the iterative shrinkage filtering algorithm (2-10) with p centroid is the query point, 1.5r maxS is the search radius, r maxS is the maximum value of the radius of the spherical component. The neighboring point set is obtained by radius search, and this point set is the node point cloud.

4. According to claim 3, the large-scale steel structure deformation monitoring method based on TLS technology, It is characterized in that In step (3), the node point cloud obtained in step (2) is divided into two types: conventional node point cloud and ball node point cloud; it is necessary to extract a smooth spherical point cloud from the ball node point cloud; The iterative shrinkage filtering algorithm is: the input parameter is the node point cloud C node , the maximum number of iterations N ISF , the maximum radius of the cylinder Distance threshold d ε ; Algorithm process: First obtain the node point cloud C node A copy of point cloud C ISF , record the copy point cloud C ISF The number of points; then use the LS algorithm to fit C ISF is a sphere, with (x, y, z) as the coordinates of the sphere center and r as the radius of the sphere. ISF Center of middle distance ≤ r+d ε Save the points back to C ISF , record C ISF The number of points; keep shrinking iteratively until the two C ISF The number of points does not change or the maximum number of iterations is reached; at this time, if the radius Then the node point cloud C node It is identified as a spherical node point cloud, thus filtering out most of the regular node point clouds; Output the result C of the last iteration shrinkage ISF , with C ISF Serves as input data for the rotation projection filtering algorithm.

5. According to claim 4, the large steel structure deformation monitoring method based on TLS technology, It is characterized in that Rotational projection filtering algorithm: The point cloud filtered by the iterative shrinkage filtering algorithm ISF first copies a copy of the input point cloud C ISF , and centroid the copy, then rotate the copy and project it onto the XOY plane, use LS to fit the copy to a circle and calculate the RMSE until the copy is rotated to the specified angle, and finally compare the minimum RMSE with the RMSE threshold to determine the node point cloud C corresponding to the input point cloud node Type.

6. According to claim 1, the large steel structure deformation monitoring method based on TLS technology, It is characterized in that In step (4), a least squares-based global search fitting algorithm is used: Input parameters: Point cloud C ISF , the maximum number of iterations N max , the threshold value of standard deviation change Δσ ε ; Get a C ISF A copy of C GSF , C GSF Fit it to a sphere, obtain the center (x, y, z) and radius r and calculate the standard deviation σ; if the change in the standard deviation is greater than Δσ ε , clear C GSF And C ISF Points whose distance from the center to the sphere is less than 3 times the standard deviation are saved in C GSF ; until the standard deviation change is less than or equal to Δσ ε Or until the maximum number of iterations is reached; C GSF Output point cloud for the algorithm.

7. According to claim 1, the large steel structure deformation monitoring method based on TLS technology, It is characterized in that In step (5), the displacement calculation model directly uses the following formula to calculate the vertical displacement and horizontal displacement: Among them, i and j represent the ball node centers of different periods, k is the ball node center number, is the vertical displacement; is the horizontal displacement, is the center of the ball node in the horizontal direction j, is the center of the ball node in the horizontal direction i; is the center of the ball node in the vertical direction j period; is the center of the ball node in the vertical direction i; Determine the three points p closest to p 1 , p 2 , p 3 , if the direct displacement value of p is greater than the other three points, the displacement value of p is recalculated according to the following formula, and the displacement direction of p remains unchanged; in D 1 , D 2 and D 3 They are p 1 , p 2 and p 3 The displacement value.

8. According to claim 1, the large steel structure deformation monitoring method based on TLS technology, It is characterized in that In step (6), the node extraction rate is used to evaluate the node extraction effect; The calculation formula of node extraction rate NER is as follows: Among them, n node is the number of nodes extracted, N node is the total number of nodes in the spatial grid structure.

9. The large-scale steel structure deformation monitoring method based on TLS technology according to claim 1, It is characterized in that The confusion matrix is ​​used to evaluate the node classification effect; the true positive case TP is defined as the number of nodes that correctly predict the spherical nodes as spherical nodes, the false negative case FN is defined as the number of nodes that incorrectly predict the spherical nodes as regular nodes, the false positive case FP is defined as the number of nodes that incorrectly predict the regular nodes as spherical nodes, and the true negative case TN is defined as the number of nodes that correctly predict the regular nodes as regular nodes; The calculation formulas for Producer's Accuracy, PA, User's Accuracy, UA, Overall Accuracy, Po, and Kappa Coefficient, K are as follows: in, 10. The large steel structure deformation monitoring method based on TLS technology according to claim 1, It is characterized in that The root mean square error (RMSE) is used as a quantitative indicator to evaluate the fitting accuracy of the least squares global search fitting algorithm GSF. Among them, d i is the distance from the point to the center of the fitting sphere, r fit is the radius of the fitting sphere.

Citation Information

Cited By

  • Deflection non-contact detection method applied to grid structure circular steel tube rod piece

    CN120598951A

  • A deflection non-contact detection method applied to a round steel pipe rod of a net rack structure

    CN120598951B