A Parametric Modeling Method for Geometric Digital Twins of Linear Engineering Structures Based on Point Clouds

By employing a parametric modeling method based on polar angle distribution and bidirectional normal intersection, the problems of model updating and stability of linear engineering structures are solved, resulting in a high-fidelity, editable digital twin model applicable to various cross-sections and complex structures.

CN123069488APending Publication Date: 2026-11-10CHONGQING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611001670.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-07
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional methods are difficult to reflect the measured geometric state of linear engineering structures, have high update costs, lack parametric representation in mesh reconstruction, and are difficult to cover diversity and transition sections. Point cloud noise and density inhomogeneity lead to unstable fitting, and cross-section fitting results in poor model continuity and smoothness.

Method used

Boundary points are extracted based on polar angle distribution features, the central axis is obtained using the bidirectional normal intersection method, a parametric cross-section template model library is constructed, the abrupt change interval of cross-sectional geometry is identified by polar radius difference and median absolute deviation, and the global optimization model is solved by combining robust fitness function and parameter association mechanism, and the teaching and learning optimization algorithm is used to reconstruct the parametric geometric model.

Benefits of technology

It realizes an editable and updatable high-fidelity digital twin model of linear engineering structures, applicable to multiple types of cross sections, improving the stability and continuity of the model, reducing the dimensionality of optimization variables, and improving optimization efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069488A_ABST
    Figure CN123069488A_ABST
Patent Text Reader

Abstract

This invention discloses a point cloud-based parametric modeling method for linear engineering structures using digital twin geometry. First, boundary points are extracted from the point cloud of the linear engineering structure to obtain planar point cloud boundary points, and the three-dimensional central axis is extracted. The three-dimensional central axis is sampled, and a local coordinate system is established, reducing the dimensionality of the three-dimensional point cloud to a sequence of two-dimensional point sets. A parametric cross-section template model is then constructed. Second, abrupt changes in the cross-sectional geometry within the two-dimensional point set sequence are identified. Based on the parametric cross-section template model, typical cross-sectional types for each two-dimensional point set in the sequence are identified, and a robust fitness function incorporating an additional error compensation mechanism and a weighted strategy is constructed. Next, parameter association, parameter sharing, and parameter regularization mechanisms are constructed. Finally, a global optimization model incorporating boundary penalty terms is constructed, and the model is solved to obtain spatially continuous and smooth parametric geometric model parameters, reconstructing the three-dimensional solid model of the linear engineering structure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of 3D laser scanning point cloud processing, reverse engineering, and infrastructure digital twin technology, specifically a parametric modeling method for linear engineering structure geometric digital twins based on point clouds. Background Technology

[0002] Linear engineering structures such as tunnels and integrated utility tunnels are characterized by long-distance extension along the route, diverse cross-sectional types, and potential axial variations. Traditional modeling methods relying on two-dimensional drawings or BIM models struggle to reflect the actual measured geometric state after completion and are costly to update. Three-dimensional laser scanning can acquire high-density point clouds, providing a data foundation for constructing high-fidelity geometric digital twins. However, existing point cloud reverse modeling methods still have the following shortcomings:

[0003] (1) Although the mesh reconstruction method can generate surface models, it lacks parametric expression and constraint semantics, making it difficult to support editable and updatable digital twins;

[0004] (2) Parametric methods are mostly applicable to finite cross-section types, and are difficult to cover the diversity of linear engineering cross-sections and transition sections;

[0005] (3) Engineering point clouds often have noise, local missing and uneven density, which leads to unstable fitting and sensitivity to data quality;

[0006] (4) Independent fitting of each section can easily lead to irregular fluctuations of parameters along the axial direction, resulting in poor model continuity and smoothness. Summary of the Invention

[0007] The purpose of this invention is to provide a parametric modeling method for linear engineering structures based on point cloud geometric digital twins, comprising the following steps:

[0008] Step 1) Extract boundary points from the point cloud of the linear engineering structure based on the polar angle distribution characteristics to obtain the boundary points of the planar point cloud;

[0009] The three-dimensional centerline of the boundary points of the planar point cloud is extracted using the bidirectional normal intersection method;

[0010] Step 2)

[0011] The three-dimensional central axis is sampled, and a local coordinate system is established at each sampling point to reduce the dimensionality of the preprocessed three-dimensional point cloud into a two-dimensional point set sequence.

[0012] Based on preset typical cross-section types, a parametric cross-section template model library is constructed, which is characterized by finite geometric parameters and geometric constraints to represent the cross-section profile.

[0013] Step 3) Use the extreme radius difference and median absolute deviation to identify the abrupt change intervals of cross-sectional geometry in the two-dimensional point set sequence, and divide the two-dimensional point set sequence into abrupt change intervals and reserved intervals based on the abrupt change intervals of cross-sectional geometry;

[0014] Step 4) Based on the parametric cross-section template model library, a classification method based on normalized radial distance features is used to identify the typical cross-section type of each two-dimensional point set in the two-dimensional point set sequence;

[0015] Step 5) Based on the typical cross-sectional types of two-dimensional point sets, construct a robust fitness function that incorporates an additional error compensation mechanism and a weighted strategy;

[0016] Step 6) Based on the typical cross-section types and robust fitness functions of two-dimensional point sets, construct a parameter association mechanism to form a unified basis parameter space;

[0017] A parameter sharing mechanism is constructed based on a unified base parameter space;

[0018] Based on parameter association and parameter sharing mechanisms, a parameter regularization mechanism is constructed.

[0019] Step 7) Based on the parameter association mechanism, parameter sharing mechanism, and parameter regularization mechanism, construct a global optimization model that introduces a boundary penalty term;

[0020] The teaching and learning optimization algorithm is used to solve the global optimization model, and the parameters of the spatially continuous and smooth parameterized geometric model are obtained.

[0021] Based on the parameters of the parametric geometric model, a three-dimensional solid model of a linear engineering structure is reconstructed.

[0022] Furthermore, in step 1), the steps for obtaining the boundary points of the planar point cloud include:

[0023] Using a K-dimensional tree data structure to perform neighborhood search and obtain points of The nearest neighbor points; with points Establish a polar coordinate system for the pole and calculate the polar angles of all nearest neighbor points. And sorted in ascending order;

[0024] Calculate the polar angle differences between adjacent points for all neighboring points, as well as the polar angle difference between the last polar angle and the first polar angle, and determine the maximum value of the polar angle differences. ;

[0025] If satisfied Then point These are the boundary points of a planar point cloud. The threshold angle;

[0026] The difference between adjacent polar angles is shown below:

[0027] (1)

[0028] The polar angle difference between the last polar angle and the first polar angle is shown below:

[0029] (2)

[0030] The maximum values ​​of all adjacent polar angle differences are shown below:

[0031] (3)

[0032] In the formula, The difference between adjacent polar angles, This is the difference between the last polar angle and the first polar angle.

[0033] Furthermore, the steps for determining the three-dimensional centerline include:

[0034] Points were obtained by uniformly sampling on boundary curve 1. ,Pass Draw the normal to boundary curve 1 to intersect boundary curve 2 at... ,Pick and The midpoint is ;Pass Draw the normal to boundary curve 2 so that it intersects boundary curve 1 at... ,Pick and The midpoint is ;by and midpoint As the central axis point;

[0035] A bidirectional normal intersection method is performed on all sampling points to obtain the central axis point set, and a random sampling consensus algorithm is used to fit the three-dimensional central axis.

[0036] Boundary curve 1 is shown below:

[0037] (4)

[0038] Boundary curve 2 is shown below:

[0039] (5)

[0040] In the formula, Let $T$ be the degree of polynomial fitting. Let x be the j-th power of the x-axis variable.

[0041] Furthermore, in step 2), the preprocessing includes denoising and segmentation operations;

[0042] The steps to reduce the dimensionality of the preprocessed 3D point cloud to a 2D point set sequence include:

[0043] set up To control the sampling density, the central axis parameter range is set. Divided into equal parts A portion, received 1 sampling point; with each odd number of sampling points The plane is defined with the origin at that point and the tangent vector of the central axis at that point as the plane's normal vector; it lies on the front boundary plane. and back boundary plane The point cloud between them is the sampling point. The corresponding point cloud region;

[0044] Points in the point cloud region Projecting onto a plane yields the projection point, i.e.:

[0045] (6)

[0046] In the formula, The three-dimensional centerline at the sampling point The unit tangent vector at that point;

[0047] At the sampling point Establish a two-dimensional local coordinate system at the location, with the basis vectors as shown below:

[0048] (7)

[0049] (8)

[0050] In the formula, Sampling points The first and second unit basis vectors of the two-dimensional local coordinate system; , , They are respectively Components in the global x, y, and z directions;

[0051] Transform the projected points to a two-dimensional local coordinate system to obtain a two-dimensional point set, i.e.:

[0052] (9)

[0053] In the formula, The transformation matrix;

[0054] The parametric section template model library is shown below:

[0055] (10)

[0056] In the formula, For parameter set; For constraint sets; The number of geometric elements;

[0057] The parameter set includes position parameters and shape parameters;

[0058] The constraint set includes horizontal constraints. Vertical constraints Symmetry constraints Equality constraints .

[0059] Furthermore, in step 3), the step of identifying the abrupt change intervals in the cross-sectional geometry includes:

[0060] Step 3.1) A sequence of two-dimensional point sets Viewed as a spatial sequence, each two-dimensional point set is shown below:

[0061] (11)

[0062] In the formula, and They are the i-th two-dimensional point sets The x and y coordinates of the j-th point in the array;

[0063] Step 3.2) The scope is divided into Equally spaced angular regions, namely:

[0064] (12)

[0065] In the formula, ;

[0066] Step 3.3) For each angular region Extract the maximum polar diameter and minimum polarity The polar radius difference between adjacent point sets is calculated as follows:

[0067] (13)

[0068] Step 3.4) Calculate the median of the absolute value sequence of polar radius differences, as shown below:

[0069] (14)

[0070] The absolute deviation of the median is shown below:

[0071] (15)

[0072] Step 3.5) Define the mutation detection threshold as follows:

[0073] (16)

[0074] In the formula, The median absolute deviation factor;

[0075] when At that time, it was determined that a mutation was detected;

[0076] Step 3.6) Count the number of mutations detected in each angular region for each two-dimensional point set, and sum them up to obtain the total mutation score. ;

[0077] like Then determine the two-dimensional point set The set of mutation points is identified as the intervals of abrupt changes in cross-sectional geometry.

[0078] Furthermore, in step 4), the step of identifying the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence includes:

[0079] Step 4.1) For a two-dimensional point set, take the geometric center of the bounding box of the rectangle as the pole. Divide the whole week into Each angle area ; for the angular region Extracting the polar radius set Calculate the mean polar radius within this angular region. ;

[0080] Step 4.2) Repeated calculations were performed on each angular region to obtain... The 3D radial distance vector is shown below:

[0081] (17)

[0082] Maximum value normalization A dimensional radial distance vector, i.e.:

[0083] (18)

[0084] The normalized radial distance vector is shown below:

[0085] (19)

[0086] Step 4.3) Extract the multidimensional peak and valley information vector, concatenate the normalized radial distance vector with the multidimensional peak and valley information vector to form a multidimensional feature vector, use a random forest classifier for classification, and output the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence.

[0087] Furthermore, in step 5), the robust fitness function is as follows:

[0088] (20)

[0089] (twenty one)

[0090] In the formula, The number of geometric elements in the parametric section template model; The parameter vector of the parameterized section template model; This refers to the r-th parameter in the parameter vector; This represents the total number of parameters in the parameter vector; For the first in the parameter vector One parameter; , The r-th parameter The lower limit and upper limit of the value; Let be the weight of the j-th geometric element; It is the sum of the distances from each point assigned to the j-th geometric element to that element; The additional error for the j-th geometric element;

[0091] The weights of the j-th geometric element are shown below:

[0092] (twenty two)

[0093] In the formula, The number of points assigned to the j-th geometric element. The total number of points in the point set. As a safety factor;

[0094] The additional error of the j-th geometric element is shown below:

[0095] (twenty three)

[0096] In the formula, For the purpose of borrowing points, For the j-th geometric element, To borrow the number of points, For the first Minimum number of constraint points required for a geometric element The actual number of points earned. This is the distance from the borrowed point to the j-th geometric element.

[0097] Furthermore, in step 6), the parameter association mechanism is as follows:

[0098] (twenty four)

[0099] (25)

[0100] (26)

[0101] (27)

[0102] In the formula, For the first Types of parameter vectors For dimension The basis parameter vector, Let be the mapping function, and m be the number of independent constraint equations between different typical section types. ; The number of typical cross-section types contained in a two-dimensional point set;

[0103] The parameter sharing mechanism refers to the fact that when multiple segments are processed along the same central axis, two-dimensional point sets are generated at the same sampling point position on the central axis for different segments; if the two-dimensional point sets correspond to the cross-sectional profile at the same spatial position and have the same typical cross-sectional type, then the instantiated basis parameters of the two-dimensional point sets share the same set of instantiated basis parameter values ​​in a unified basis parameter space.

[0104] Shared groups are identified by matching the centerline parameters: if the centerline parameters of two two-dimensional point sets are consistent and the typical cross-section types are consistent, they are classified into the same shared group.

[0105] Parametric regularization mechanisms include polynomial regularization, cosine function regularization, and sawtooth wave regularization.

[0106] Furthermore, the polynomial regularization is as follows:

[0107] (28)

[0108] In the formula, For the first One basic parameter, For the first The variation pattern of each basic parameter The order of the polynomial; For polynomial coefficients, These are the parameter values ​​for the sampling points;

[0109] The cosine function regularization is shown below:

[0110] (29)

[0111] In the formula, For amplitude, Angular frequency, This is the initial phase. This is the offset;

[0112] Sawtooth wave regularization is shown below:

[0113] (30)

[0114] In the formula, For period Local location within, The slope This is the starting value for the cycle.

[0115] Furthermore, in step 7), the global fitness function is as follows:

[0116] (31)

[0117] (32)

[0118] In the formula, For point set The weighted fitness function, As weight, The penalty coefficient is... For boundary penalty terms, The number of points in the two-dimensional point set. This is the mapping function from the regularization coefficient vector to the parameter vector; The dimension of the regularization coefficient vector; Let l be the l-th regularization coefficient; , These are the l-th regularization coefficients. The lower limit and upper limit of the value;

[0119] The mapping function from the regularization coefficient vector to the parameter vector is shown below:

[0120] (33)

[0121] In the formula, For point set The parameter vector corresponding to the parameterized section template model, This is the regularization coefficient vector;

[0122] The boundary penalty terms are as follows:

[0123] (34)

[0124] In the formula, For the i-th two-dimensional point set The number of parameters corresponding to the parametric section template model; For the i-th two-dimensional point set The r-th parameter in the parameter vector ; , Parameters The lower limit and upper limit of the value;

[0125] The steps to solve the global optimization model include:

[0126] exist Internal random generation vector of regularization coefficients The fitness is calculated as follows:

[0127] (35)

[0128] During the teacher phase, the process for each individual is as follows:

[0129] (36)

[0130] In the formula, This refers to the individual with the best fitness in the current population. is the mean vector of all individuals in the population; r is a random vector; As a teaching factor; Element-wise multiplication of vectors; This is the new regularization coefficient vector generated for the i-th individual after the teacher stage update;

[0131] During the student phase, another individual is randomly selected. The learning direction is determined based on fitness level, that is:

[0132] (37)

[0133] In the formula, For the i-th individual, Another individual randomly selected during the student phase; , Individuals and The corresponding fitness function value;

[0134] The individual is updated using a greedy selection strategy;

[0135] A convergence detection mechanism based on a sliding window is introduced, with a window size of [value missing]. The relative rate of change threshold is When continuous The relative rate of change of the optimal value in the next iteration is less than If so, terminate early;

[0136] The steps for outputting the reconstructed 3D solid model include:

[0137] The regularization coefficient vector obtained from global optimization Through mapping function Convert into parameter vectors of the parameterized cross-section template model instantiation corresponding to each two-dimensional point set. ;

[0138] Based on the geometric definition of the parametric section template model, determine the coordinates of the two-dimensional control points of each geometric element;

[0139] By inverse coordinate transformation, the two-dimensional control points are mapped back to three-dimensional space to obtain the three-dimensional control points, i.e.:

[0140] (38)

[0141] In the formula, The transformation matrix is... This is the sampling point for the central axis.

[0142] The technical effects of this invention are undeniable, and compared with the prior art, this invention has the following beneficial effects:

[0143] (1) By constructing a parameterized cross-section template model, the present invention represents the cross-sectional profile of a linear engineering structure as a parameterized model determined by finite geometric parameters and geometric constraints. Compared with the traditional mesh reconstruction method, it can retain the parameterized expression and constraint relationship of the cross-sectional geometry, making the reconstructed model have the characteristics of being editable and updatable, and is more suitable for geometric digital twin modeling and subsequent operation and maintenance updates.

[0144] (2) By pre-setting multiple typical cross-section types and combining a classification method based on normalized radial distance features, the present invention achieves automatic matching between two-dimensional point sets and corresponding parameterized cross-section template models. It can be applied to linear engineering structures with different cross-section shapes, such as tunnels and integrated pipe corridors, thus improving the applicability of the method to multiple types of cross-section structures.

[0145] (3) The present invention uses the extreme diameter difference and the absolute deviation of the median to identify the abrupt change interval of the cross-sectional geometry, which can distinguish between the stable interval and the abrupt change interval along the central axis, and avoid directly incorporating the mixed cross-sectional shape point set into a single template fitting, thereby improving the stability of cross-section classification, parameter fitting and global optimization.

[0146] (4) The present invention introduces an additional error compensation mechanism and a weighting strategy into the robust fitness function. The additional error compensation mechanism can provide indirect constraints for geometric elements that are constrained due to local missing points in the point cloud, and the weighting strategy can balance the differences in error contribution caused by uneven point density of different geometric elements, thereby improving the method's adaptability to local missing points and uneven density in the point cloud.

[0147] (5) This invention constructs a unified global parameter expression method through parameter association mechanism, parameter sharing mechanism and parameter regularization mechanism, which can reduce the dimension of optimization variables and constrain the parameters to change continuously along the central axis, thereby avoiding the irregular fluctuation of parameters caused by independent fitting of each section, so that the reconstructed linear engineering structure three-dimensional solid model has good spatial continuity and smoothness.

[0148] (6) The present invention uses the teaching and learning optimization algorithm to solve the global optimization model, which can achieve parameter optimization solution when the objective function is not differentiable and there are multiple local optima. It also reduces redundant iterations and improves optimization efficiency through the sliding window convergence detection mechanism.

[0149] (7) The three-dimensional solid model reconstructed by the present invention is determined by the parameters of the parametric geometric model. It can be parametrically edited, locally updated and reconstructed as needed, providing a high-fidelity, computable, editable and updatable geometric model foundation for the geometric digital twin of linear engineering structures, and has good engineering applicability. Attached Figure Description

[0150] Figure 1 This is a flowchart of a point cloud-based geometric digital twin parametric modeling method for linear engineering structures according to the present invention.

[0151] Figure 2 This is a schematic diagram illustrating the polar angle distribution characteristics of planar point cloud boundary points according to the present invention.

[0152] Figure 3 This is a schematic diagram of the planar point cloud boundary point extraction effect of the present invention; wherein, (a) is the original planar point cloud, and (b) is the boundary point extraction result;

[0153] Figure 4 This is a schematic diagram illustrating the generation of the axis point in the bidirectional normal intersection method of the present invention;

[0154] Figure 5 This is a schematic diagram illustrating the conversion from a linear engineering structure point cloud to a two-dimensional point set according to the present invention.

[0155] Figure 6 This is a schematic diagram of a typical linear engineering structure section PSTM type of the present invention; wherein, (a) is an arch, (b) is a circle, (c) is a rectangle, (d) is a symmetrical octagon, (e) is an eight-segment polygon, (f) is a nine-segment polygon, (g) is a square arch, and (h) is a horseshoe shape.

[0156] Figure 7 This is a schematic diagram illustrating the identification of abrupt changes in the geometric shape of a linear engineering section according to the present invention.

[0157] Figure 8 This is a schematic diagram of the two-dimensional point set classification framework of the present invention;

[0158] Figure 9 This is a schematic diagram illustrating the effect of local point cloud defects on PSTM fitting according to the present invention; where (a) is a two-dimensional point set with local defects, and (b) is the fitting result with MAE as the fitness function;

[0159] Figure 10 This is a schematic diagram of the parameter association mechanism of the present invention;

[0160] Figure 11 This is a schematic diagram of the three-dimensional solid model generated by the cross-sectional contour lofting of the present invention; wherein, (a) is the sequence of inner and outer cross-sectional contours, and (b) is the three-dimensional solid model generated by the lofting. Detailed Implementation

[0161] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0162] Example 1:

[0163] See Figures 1 to 11 A parametric modeling method for linear engineering structures based on point cloud geometric digital twins includes the following steps:

[0164] Step 1) Extract boundary points from the point cloud of linear engineering structures such as tunnels and integrated utility tunnels based on polar angle distribution characteristics to obtain the boundary points of the planar point cloud;

[0165] The three-dimensional centerline of the boundary points of the planar point cloud is extracted using the bidirectional normal intersection method;

[0166] Step 2) Sample the three-dimensional central axis, establish a local coordinate system at each sampling point, and reduce the dimensionality of the preprocessed three-dimensional point cloud to a two-dimensional point set sequence;

[0167] Based on preset typical cross-section types, a parametric cross-section template model library is constructed, which is characterized by finite geometric parameters and geometric constraints to represent the cross-section profile.

[0168] Step 3) Use the extreme radius difference and median absolute deviation to identify the abrupt change intervals of cross-sectional geometry in the two-dimensional point set sequence, and divide the two-dimensional point set sequence into abrupt change intervals and reserved intervals based on the abrupt change intervals of cross-sectional geometry;

[0169] Among them, the two-dimensional point set within the retention interval is used for subsequent typical cross-section type identification, parameter fitting and global optimization, while the abrupt change interval is used to determine the segmentation position and parameter correlation between different typical cross-section types;

[0170] Step 4) Based on the parametric cross-section template model library, a classification method based on normalized radial distance features is used to identify the typical cross-section type of each two-dimensional point set in the two-dimensional point set sequence;

[0171] Step 5) Based on the typical cross-sectional types of two-dimensional point sets, construct a robust fitness function that incorporates an additional error compensation mechanism and a weighted strategy;

[0172] Step 6) Based on the typical cross-section types and robust fitness functions of two-dimensional point sets, construct a parameter association mechanism to form a unified basis parameter space;

[0173] A parameter sharing mechanism is constructed based on a unified base parameter space;

[0174] Based on parameter association and parameter sharing mechanisms, a parameter regularization mechanism is constructed.

[0175] Step 7) Based on the parameter association mechanism, parameter sharing mechanism, and parameter regularization mechanism, construct a global optimization model that introduces a boundary penalty term;

[0176] The teaching and learning optimization algorithm is used to solve the global optimization model, and the parameters of the spatially continuous and smooth parameterized geometric model are obtained.

[0177] Based on the parameters of the parametric geometric model, a three-dimensional solid model of a linear engineering structure is reconstructed.

[0178] Example 2:

[0179] The main structure of this embodiment is the same as that of Embodiment 1. Further, in step 1), the step of obtaining the boundary points of the planar point cloud includes:

[0180] Using a K-dimensional tree data structure to perform neighborhood search and obtain points of The nearest neighbor points; with points Establish a polar coordinate system for the pole and calculate the polar angles of all nearest neighbor points. And sorted in ascending order;

[0181] Calculate the polar angle differences between adjacent points for all neighboring points, as well as the polar angle difference between the last polar angle and the first polar angle, and determine the maximum value of the polar angle differences. ;

[0182] If satisfied Then point These are the boundary points of a planar point cloud. The threshold angle is used to control the maximum polar angle interval. near The tolerance for judgment at that time is usually taken as It can be adjusted according to point cloud density, noise level and number of neighboring points;

[0183] The difference between adjacent polar angles is shown below:

[0184] (1)

[0185] The polar angle difference between the last polar angle and the first polar angle is shown below:

[0186] (2)

[0187] The maximum values ​​of all adjacent polar angle differences are shown below:

[0188] (3)

[0189] In the formula, The difference between adjacent polar angles, This is the difference between the last polar angle and the first polar angle.

[0190] Example 3:

[0191] The main structure of this embodiment is the same as any one of embodiments 1-2. Furthermore, the steps for determining the three-dimensional centerline include:

[0192] Points were obtained by uniformly sampling on boundary curve 1. ,Pass Draw the normal to boundary curve 1 to intersect boundary curve 2 at... ,Pick and The midpoint is ;Pass Draw the normal to boundary curve 2 so that it intersects boundary curve 1 at... ,Pick and The midpoint is ;by and midpoint As the central axis point;

[0193] A bidirectional normal intersection method is performed on all sampling points to obtain the central axis point set, and a random sampling consensus algorithm is used to fit the three-dimensional central axis.

[0194] Boundary curve 1 is shown below:

[0195] (4)

[0196] Boundary curve 2 is shown below:

[0197] (5)

[0198] In the formula, Let $T$ be the degree of polynomial fitting. Let be the j-th power of the x-axis variable x, and be the j-th term in the polynomial fitting, where j is the polynomial term index. ,when hour, , corresponding to the constant term.

[0199] Example 4:

[0200] The main structure of this embodiment is the same as any one of embodiments 1-3. Further, in step 2), the preprocessing includes noise reduction and segmentation operations.

[0201] The steps to reduce the dimensionality of the preprocessed 3D point cloud to a 2D point set sequence include:

[0202] set up To control the sampling density, the central axis parameter range is set. Divided into equal parts A portion, received 1 sampling point; with each odd number of sampling points The plane is defined with the origin at that point and the tangent vector of the central axis at that point as the plane's normal vector; it lies on the front boundary plane. and back boundary plane The point cloud between them is the sampling point. The corresponding point cloud region;

[0203] Points in the point cloud region Projecting onto a plane yields the projection point, i.e.:

[0204] (6)

[0205] In the formula, The three-dimensional centerline at the sampling point The unit tangent vector at that point is also the sampling point. The unit normal vector of the plane corresponding to the cross section;

[0206] At the sampling point Establish a two-dimensional local coordinate system at the location, with the basis vectors as shown below:

[0207] (7)

[0208] (8)

[0209] In the formula, Sampling points The first and second unit basis vectors of the two-dimensional local coordinate system; , , They are respectively Components in the global x, y, and z directions;

[0210] Transform the projected points to a two-dimensional local coordinate system to obtain a two-dimensional point set, i.e.:

[0211] (9)

[0212] In the formula, The transformation matrix;

[0213] The parametric section template model library is shown below:

[0214] (10)

[0215] In the formula, For parameter set; For constraint sets; The number of geometric elements;

[0216] The parameter set includes position parameters and shape parameters;

[0217] The constraint set includes horizontal constraints. Vertical constraints Symmetry constraints Equality constraints ;

[0218] Horizontal constraints are used to limit the corresponding geometric elements to remain horizontal, vertical constraints are used to limit the corresponding geometric elements to remain vertical, symmetry constraints are used to limit the cross-sectional profile or related geometric elements to remain symmetrical about a specified axis, and equality constraints are used to limit the related geometric parameters or geometric quantities to remain equal.

[0219] The constraint objects and constraint relationships differ in different typical cross-section templates, and the specific constraint relationships are determined by the geometric construction of each parametric cross-section template model.

[0220] Example 5:

[0221] The main structure of this embodiment is the same as any one of embodiments 1-4. Further, in step 3), the step of identifying the abrupt change interval of the cross-sectional geometry includes:

[0222] Step 3.1) A sequence of two-dimensional point sets Viewed as a spatial sequence, each two-dimensional point set is shown below:

[0223] (11)

[0224] In the formula, and They are the i-th two-dimensional point sets The x and y coordinates of the j-th point in the array;

[0225] Step 3.2) The scope is divided into Equally spaced angular regions, namely:

[0226] (12)

[0227] In the formula, ;

[0228] Step 3.3) For each angular region Extract the maximum polar diameter and minimum polarity The polar radius difference between adjacent point sets is calculated as follows:

[0229] (13)

[0230] Step 3.4) Calculate the median of the absolute value sequence of polar radius differences, as shown below:

[0231] (14)

[0232] The absolute deviation of the median is shown below:

[0233] (15)

[0234] Step 3.5) Define the mutation detection threshold as follows:

[0235] (16)

[0236] In the formula, The median absolute deviation factor;

[0237] when At that time, it was determined that a mutation was detected;

[0238] Step 3.6) Count the number of mutations detected in each angular region for each two-dimensional point set, and sum them up to obtain the total mutation score. ;

[0239] like Then determine the two-dimensional point set The set of mutation points is identified as abruptly changed intervals in the cross-sectional geometry.

[0240] The threshold for determining the total mutation score is preferably set to 2 in this embodiment;

[0241] Used to determine the total mutation score of a two-dimensional point set Determine whether the two-dimensional point set is a set of mutation points. It is not a fixed constant, but is adaptively calculated based on the total mutation score sequence using the median absolute deviation method, as shown below:

[0242]

[0243] In the formula, The median of the total mutation score sequence. This is the absolute deviation of the median. It is a multiple factor.

[0244] The abrupt change interval of cross-sectional geometry refers to the set of intervals formed by the parameter intervals corresponding to the set of abrupt change points along the central axis; when the parameter intervals corresponding to multiple sets of abrupt change points are adjacent, they can be merged into a continuous abrupt change interval.

[0245] Example 6:

[0246] The main structure of this embodiment is the same as any one of embodiments 1-5. Further, in step 4), the step of identifying the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence includes:

[0247] Step 4.1) For a two-dimensional point set, take the geometric center of the bounding box of the rectangle as the pole. Divide the whole week into Each angle area ; for the angular region Extracting the polar radius set Calculate the mean polar radius within this angular region. ;

[0248] Step 4.2) Repeated calculations were performed on each angular region to obtain... The 3D radial distance vector is shown below:

[0249] (17)

[0250] Maximum value normalization A dimensional radial distance vector, i.e.:

[0251] (18)

[0252] The normalized radial distance vector is shown below:

[0253] (19)

[0254] Step 4.3) Extract the multidimensional peak and valley information vector, concatenate the normalized radial distance vector with the multidimensional peak and valley information vector to form a multidimensional feature vector, use a random forest classifier for classification, and output the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence.

[0255] The random forest classifier is used to identify the typical cross-section type to which a two-dimensional point set belongs. Its input is the multi-dimensional feature vector extracted from the two-dimensional point set, and its output is the label of the typical cross-section type corresponding to the two-dimensional point set.

[0256] The input feature vector is composed of a K-dimensional normalized radial distance vector and a multi-dimensional peak-valley information vector; in this embodiment, K is 360, and the peak-valley information vector is 22-dimensional, so the input feature vector is 382-dimensional. The output categories include eight typical cross-sectional types: arch, circle, rectangle, symmetrical octagon, eight-segment polygonal polygon, nine-segment polygonal polygon, square arch, and horseshoe.

[0257] The random forest classifier consists of multiple decision trees. In this embodiment, the random forest contains 100 decision trees. Each decision tree is trained using bootstrapping, and when splitting a node, a subset of features is randomly selected from all features as candidate features. The optimal splitting feature and splitting threshold are determined based on minimizing Gini impurity. Each decision tree outputs its classification result, and the typical cross-sectional type of the two-dimensional point set is finally determined by majority voting. The classification results are shown below:

[0258]

[0259] In the formula, x is the input feature vector. Let T be the output class of the t-th decision tree, and T be the number of decision trees. This represents a typical cross-sectional type output by a random forest classifier.

[0260] The training data comes from a simulated synthetic 2D point set generated based on a parametric cross-section template model. A standard synthetic 2D point set is generated by randomly sampling within the parameter range of the parametric cross-section template model corresponding to each typical cross-section type. Perturbations such as measurement errors, outliers, local occlusion, and uneven point density are further superimposed to form a simulated synthetic 2D point set, which is used to train the random forest classifier. The 2D point set obtained by converting real engineering point clouds is used to test and verify the actual recognition performance of the classifier.

[0261] Example 7:

[0262] The main structure of this embodiment is the same as any one of embodiments 1-6. Further, in step 5), the robust fitness function is as follows:

[0263] (20)

[0264] (twenty one)

[0265] In the formula, The number of geometric elements in the parametric section template model; The parameter vector of the parameterized section template model; Let r be the r-th parameter in the parameter vector; This represents the total number of parameters in the parameter vector; For the first in the parameter vector One parameter; , The r-th parameter The lower limit and upper limit of the value; Let be the weight of the j-th geometric element; It is the sum of the distances from each point assigned to the j-th geometric element to that element; The additional error for the j-th geometric element;

[0266] The weights of the j-th geometric element are shown below:

[0267] (twenty two)

[0268] In the formula, The number of points assigned to the j-th geometric element. The total number of points in the point set. As a safety factor;

[0269] The additional error of the j-th geometric element is as follows:

[0270] (twenty three)

[0271] In the formula, For the purpose of borrowing points, For the j-th geometric element, To borrow the number of points, For the first Minimum number of constraint points required for a geometric element The actual number of points earned. This is the distance from the borrowed point to the j-th geometric element.

[0272] When the linear engineering structure is a shield tunnel, a grouping allocation strategy is used to allocate the point cloud of each segment;

[0273] The grouping and allocation strategy refers to the process where, after the shield tunnel point cloud is preprocessed to obtain point clouds of each segment with color labels, points with the same color label are treated as a whole, and the average distance to each arc is calculated in groups. The entire group is then assigned to the arc with the smallest average distance. After the grouping and allocation are completed, the robust fitness function value is calculated according to the weighted robust fitness function.

[0274] The robust fitness function expression used in shield tunnel scenarios is the same as in other scenarios, the difference lies in the point-to-geometric element allocation strategy. For general linear engineering sections, points are allocated point-by-point according to the shortest distance to each geometric element; for shield tunnels, since their sections are composed of arcs corresponding to multiple segments, to maintain physical topology consistency, the point clouds of segments carrying the same color label are treated as a whole, and the average distance to each arc is calculated in groups. The entire group is then allocated to the arc with the smallest average distance. After allocation, the function value is still calculated using the same weighted robust fitness function.

[0275] Example 8:

[0276] The main structure of this embodiment is the same as any one of embodiments 1-7. Further, in step 6), the parameter association mechanism is as follows:

[0277] (twenty four)

[0278] (25)

[0279] (26)

[0280] (27)

[0281] In the formula, For the first Types of parameter vectors For dimension The basis parameter vector, Let be the mapping function, and m be the number of independent constraint equations between different typical section types. ; The number of typical cross-section types contained in a two-dimensional point set;

[0282] The parameter sharing mechanism refers to the fact that when multiple segments are processed along the same central axis, two-dimensional point sets are generated at the same sampling point position on the central axis for different segments; if the two-dimensional point sets correspond to the cross-sectional profile at the same spatial position and have the same typical cross-sectional type, then the instantiated basis parameters of the two-dimensional point sets share the same set of instantiated basis parameter values ​​in a unified basis parameter space.

[0283] Shared groups are identified by matching the centerline parameters: if the centerline parameters of two two-dimensional point sets are consistent and the typical cross-section types are consistent, they are classified into the same shared group.

[0284] Parametric regularization mechanisms include polynomial regularization, cosine function regularization, and sawtooth wave regularization.

[0285] Example 9:

[0286] The main structure of this embodiment is the same as any one of embodiments 1-8. Furthermore, the polynomial regularization is as follows:

[0287] (28)

[0288] In the formula, For the first One basic parameter, For the first The variation pattern of each basic parameter The order of the polynomial; For polynomial coefficients, These are the parameter values ​​for the sampling points;

[0289] The cosine function regularization is shown below:

[0290] (29)

[0291] In the formula, For amplitude, Angular frequency, This is the initial phase. This is the offset;

[0292] Sawtooth wave regularization is shown below:

[0293] (30)

[0294] In the formula, For period Local location within, The slope This is the starting value for the cycle.

[0295] Example 10:

[0296] The main structure of this embodiment is the same as any one of embodiments 1-9. Further, in step 7), the global fitness function is as follows:

[0297] (31)

[0298] (32)

[0299] In the formula, For point set The weighted fitness function, As weight, The penalty coefficient is... For boundary penalty terms, The number of points in the two-dimensional point set. This is the mapping function from the regularization coefficient vector to the parameter vector; This represents the dimension of the regularization coefficient vector, i.e., the total number of regularization coefficients. Let l be the l-th regularization coefficient; , These are the l-th regularization coefficients. The lower limit and upper limit of the value;

[0300] The mapping function from the regularization coefficient vector to the parameter vector is shown below:

[0301] (33)

[0302] In the formula, For point set The parameter vector corresponding to the parameterized section template model, This is the regularization coefficient vector;

[0303] The boundary penalty terms are as follows:

[0304] (34)

[0305] In the formula, in the formula, For the i-th two-dimensional point set The number of parameters corresponding to the parametric section template model; For the i-th two-dimensional point set The r-th parameter in the parameter vector ; , Parameters The lower limit and upper limit of the value;

[0306] The steps to solve the global optimization model include:

[0307] exist Internal random generation vector of regularization coefficients The fitness is calculated as follows:

[0308] (35)

[0309] During the teacher phase, the process for each individual is as follows:

[0310] (36)

[0311] In the formula, This refers to the individual with the best fitness in the current population. is the mean vector of all individuals in the population; r is a random vector; As a teaching factor; Element-wise multiplication of vectors; This is the new regularization coefficient vector generated for the i-th individual after the teacher stage update;

[0312] During the student phase, another individual is randomly selected. The learning direction is determined based on fitness level, that is:

[0313] (37)

[0314] In the formula, For the i-th individual, Another individual randomly selected during the student phase; , Individuals and The corresponding fitness function value;

[0315] The individual is updated using a greedy selection strategy;

[0316] A convergence detection mechanism based on a sliding window is introduced, with a window size of [value missing]. The relative rate of change threshold is When continuous The relative rate of change of the optimal value in the next iteration is less than If so, terminate early;

[0317] The steps for outputting the reconstructed 3D solid model include:

[0318] The regularization coefficient vector obtained from global optimization Through mapping function Convert into parameter vectors of the parameterized cross-section template model instantiation corresponding to each two-dimensional point set. ;

[0319] Based on the geometric definition of the parametric section template model, determine the coordinates of the two-dimensional control points of each geometric element;

[0320] By inverse coordinate transformation, the two-dimensional control points are mapped back to three-dimensional space to obtain the three-dimensional control points, i.e.:

[0321] (38)

[0322] In the formula, The transformation matrix is... This is the sampling point for the central axis.

[0323] Example 11:

[0324] like Figure 1 As shown, this invention discloses a flowchart of a method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins.

[0325] S101, Point Cloud Data Preprocessing - Boundary Point Extraction: Extracting Planar Point Cloud Boundary Points Based on Polar Angle Distribution Features.

[0326] In specific implementation, such as Figure 2 As shown, this is for determining a point in a planar point cloud. To determine if a point is a boundary point, a neighborhood search is first performed using a KD-Tree data structure to obtain the point's location. of The nearest neighbor points; with points Establish a polar coordinate system for the pole and calculate the polar angles of all nearest neighbor points. Arranged in ascending order; the difference between adjacent polar angles is defined as:

[0327]

[0328] The difference between the last polar angle and the first polar angle is:

[0329]

[0330] Define the maximum polar angle interval The maximum value of all adjacent polar angle differences:

[0331]

[0332] Set threshold angle If satisfied Then the decision point These are boundary points. The boundary point extraction result is as follows: Figure 3 As shown.

[0333] S102, Point Cloud Data Preprocessing - Centerline Extraction: The three-dimensional centerline is extracted using the bidirectional normal intersection method.

[0334] In specific implementation, such as Figure 4 As shown, the 3D point cloud is projected onto... plane and Plane, extract boundary points and use Curve fitting is performed using a polynomial of degree n, and the equation of the boundary curve is:

[0335]

[0336]

[0337] Robust fitting was performed using the RANSAC algorithm; points were obtained by uniform sampling on boundary curve 1. ,Pass Draw the normal to boundary curve 1 to intersect boundary curve 2 at... ,Pick and The midpoint is ;Pass Draw the normal to boundary curve 2 so that it intersects boundary curve 1 at... ,Pick and The midpoint is ;by and midpoint As the central axis point, the above operation is performed on all sampled points to obtain the central axis point set, and the RANSAC algorithm is used to fit it to obtain a smooth and continuous three-dimensional central axis.

[0338] S103, Point Cloud Data Preprocessing – Point Cloud Dimensionality Reduction and Transformation: Through central axis sampling and local coordinate system transformation, the three-dimensional point cloud is converted into a two-dimensional point set sequence.

[0339] In specific implementation, such as Figure 5 As shown, let To control the sampling density, the central axis parameter range is set. Divided into equal parts A portion, received 1 sampling point; with each odd number of sampling points The cross-sectional plane is defined with the origin of the plane and the tangent vector of the central axis at that point as the plane's normal vector; it lies on the front boundary plane. With the back boundary plane The point cloud between them is the sampling point. The corresponding point cloud region. The points in the region's point cloud... Projected onto a plane The projection point is obtained:

[0340]

[0341] At the sampling point Establish a two-dimensional local coordinate system at point , with the basis vectors as follows:

[0342]

[0343]

[0344] Transformation matrix Transform the projection points to a two-dimensional local coordinate system:

[0345]

[0346] This yields a two-dimensional point set sequence distributed along the central axis, providing a data foundation for subsequent classification and fitting.

[0347] S104, PSTM Construction: PSTM is constructed for eight typical section types.

[0348] In specific implementation, such as Figure 6 As shown, PSTM is defined as:

[0349]

[0350] in, This is a set of parameters, including positional and shape parameters; The constraint set includes horizontal constraints. Vertical constraints Symmetry constraints Equality constraints ; The number of geometric elements is specified. Corresponding PSTM templates are created for eight typical cross-sectional types: arch, circle, rectangle, symmetrical octagon, eight-segment polygonal, nine-segment polygonal, square arch, and horseshoe.

[0351] S105. Identification of abrupt changes in cross-sectional geometry: The method based on polar radius difference and MAD is used to identify the abrupt change intervals in cross-sectional geometry.

[0352] In specific implementation, such as Figure 7 As shown, A set of two-dimensional points Consider it as a spatial sequence; for each two-dimensional point set The geometric center of the rectangular bounding box is used as the pole. :

[0353]

[0354] Will The scope is divided into Equally spaced angular regions ,in, ;

[0355] Extract the maximum and minimum polar radii for each angular region and calculate the polar radii difference between adjacent point sets:

[0356]

[0357] Calculate the median and MAD for a series of difference absolute values:

[0358]

[0359]

[0360] Constructing a threshold to achieve mutation detection:

[0361]

[0362] In the formula, The median absolute deviation factor; when At that time, it was determined that a mutation was detected;

[0363] The total mutation score was obtained statistically. ,when Exceeding the threshold The set of points is determined to be a set of mutation points, thereby locating the mutation interval.

[0364] S106, Two-dimensional point set classification: Identify the PSTM type of two-dimensional point sets using a classification method based on normalized radial distance features.

[0365] In specific implementation, such as Figure 8 As shown, for a two-dimensional point set Divide the angular region with the geometric center of the rectangular bounding box as the pole. Extract the mean polar radius of each region to construct the radial distance vector:

[0366]

[0367] Perform maximum value normalization:

[0368]

[0369] The normalized radial distance vector is obtained as follows:

[0370]

[0371] Further extract peak and valley information vectors, including the number of peaks, the number of valleys, the position and amplitude of peaks, and the position and amplitude of valleys; concatenate the normalized radial distance vector with the peak and valley information vector to form a 382-dimensional feature vector; use a random forest classifier to complete the PSTM type recognition class. The random forest contains 100 decision trees, each with a maximum depth of 3 layers and a minimum number of samples for node splitting of 2.

[0372] S107. Additional error compensation mechanism: An additional error compensation mechanism is introduced to address the issue of missing point clouds.

[0373] In specific implementation, such as Figure 9 As shown, let PSTM be composed of Composed of geometric elements, the first Geometric elements The minimum number of constraint points required is The actual number of points obtained is ;when At that time, select a distance from the distribution points of other geometric elements. Recent Each point is used as a set of borrowing points. Additional error Defined as:

[0374]

[0375] in , Indicates the borrowing point arrive The distance.

[0376] S108. Weighted Strategy: A weighted strategy is adopted to address the problem of uneven point density.

[0377] In practical implementation, geometric elements are defined. weight for:

[0378]

[0379] in The total number of points in the point set. As a safety factor, a robust fitness function for a single two-dimensional point set is constructed by combining additional error compensation and weighting strategies:

[0380]

[0381]

[0382] And satisfy parameter boundary constraints .

[0383] S109. Grouping and allocation strategy for shield tunnels: A grouping and allocation strategy is adopted for shield tunnels.

[0384] In practice, after the shield tunnel point cloud is preprocessed, each segment point cloud is obtained with a color label. Points with the same color label are regarded as a whole, and the average distance to each arc is calculated in groups. The whole group is assigned to the arc with the smallest average distance to maintain physical topology consistency. After the group assignment is completed, the objective function value is calculated according to the weighted fitness function in step S108.

[0385] S110, Parameter Association Mechanism: Construct a unified base parameter space to handle geometric continuity constraints between different PSTM types.

[0386] In specific implementation, such as Figure 10 As shown, when a two-dimensional point set sequence contains When the PSTM type is 1, the first The parameter vector of type is Concatenate them into a joint parameter vector:

[0387]

[0388] in Geometric analysis was used to identify the differences between different PSTM types. There are 3 independent constraint equations, constructing a dimension of . basis parameter vectors This ensures that the parameter vectors instantiated by each PSTM type can be uniquely derived from the mapping function:

[0389]

[0390] S111, Parameter sharing mechanism: Identify repeated point sets at the same spatial location and share basic parameters.

[0391] In practice, when multiple segments are processed along the same central axis, different segments may generate two-dimensional point sets at the same sampling point position on the central axis. If these two-dimensional point sets correspond to the cross-sectional profile at the same spatial position and have the same PSTM type, they share the same set of instantiated basis parameter values ​​in a unified basis parameter space. The shared group is identified by matching the central axis parameters: if the central axis parameters corresponding to two two-dimensional point sets are consistent and the PSTM type is consistent, they are classified into the same shared group.

[0392] S112. Parameter regularization mechanism: The variation law of the basis parameters along the axis is described by regularization functions such as polynomials, cosine functions or sawtooth waves.

[0393] In practice, polynomial regularization: the first... One basis parameter The pattern of change is Polynomial description of order:

[0394]

[0395] in, These are the parameter values ​​for the sampling points. These are the polynomial coefficients;

[0396] Cosine function regularization is expressed as follows: using the cosine function to describe the periodic change of the parameter:

[0397]

[0398] in, For amplitude, Angular frequency, This is the initial phase. This is the offset;

[0399] Sawtooth wave regularization is represented as follows: the parameters change linearly within each period, and reset abruptly at the period boundaries.

[0400]

[0401] in, For period Local location within, The slope This is the starting value for the cycle.

[0402] By transforming the discrete basis parameter distribution into an optimization problem with a small number of regularization coefficients, we can achieve continuous and smooth changes in parameters along the central axis and reduce the optimization dimensionality.

[0403] S113. Construction of global fitness function: Construct a global fitness function with regularization coefficient vector as optimization variable and introduce a boundary penalty term.

[0404] In specific implementation, set After parameter association, sharing, and regularization processing, let the point set be... Corresponding PSTM parameter vector It can be obtained from the regularization coefficient vector The mapping function is uniquely determined as follows:

[0405]

[0406] The global fitness function is:

[0407]

[0408]

[0409] and satisfy Among them, the boundary penalty item Used to constrain parameters from going out of bounds.

[0410] in, For point set The weighted fitness function, As weight, The penalty coefficient is... For boundary penalty terms:

[0411]

[0412] S114, TLBO Global Optimization Solution: TLBO is used to solve the global fitness function, and a sliding window convergence detection mechanism is introduced.

[0413] In specific implementation, Initialize the population randomly; generate randomly vector of regularization coefficients Calculate fitness In the teacher training phase, the best individual at present is selected as the teacher. Calculate the population mean and update the individual:

[0414]

[0415] in, As teaching factors, from the set Uniform random sampling in the middle, It is a random vector;

[0416] During the student phase, another individual is randomly selected, and their learning direction is determined based on their fitness:

[0417]

[0418] A greedy strategy is used to update individuals; a sliding window convergence detection mechanism is introduced, with a window size of [value missing]. The relative rate of change threshold is When continuous The relative rate of change of the optimal value in the next iteration is less than the threshold. The iteration is terminated early.

[0419] S115. Geometric Model Reconstruction: Based on the optimization results, control points are analyzed and mapped back to three-dimensional space to generate a three-dimensional solid model.

[0420] In specific implementation, such as Figure 11 As shown, based on the optimized regularization coefficient vector Through mapping function Convert to PSTM instantiation parameter vectors corresponding to each 2D point set The coordinates of the two-dimensional control points are analytically determined based on the geometric definition of PSTM; the two-dimensional control points are then mapped back to three-dimensional space through inverse coordinate transformation.

[0421]

[0422] in, The transformation matrix is... The sampling point is the centerline; the cross-sectional profile is constructed using 3D control points in the Dynamo platform: Line.ByStartPointEndPoint nodes are used for straight line segments, and Arc.ByCenterPointStartPointEndPoint nodes are used for arcs, connecting them to form closed polylines; combined with the outer contour parameters determined by the design drawings, the outer contour sequence and inner contour sequence are lofted to generate outer and inner solids respectively; the Boolean difference operation is performed using the Solid.Difference node to subtract the inner solid from the outer solid to obtain a 3D solid model with wall thickness.

Claims

1. A parametric modeling method for linear engineering structures based on point cloud geometric digital twins, characterized in that, Includes the following steps: Step 1) Extract boundary points from the point cloud of the linear engineering structure based on the polar angle distribution characteristics to obtain the boundary points of the planar point cloud; The three-dimensional centerline of the boundary points of the planar point cloud is extracted using the bidirectional normal intersection method; Step 2) Sample the three-dimensional central axis, establish a local coordinate system at each sampling point, and reduce the dimensionality of the preprocessed three-dimensional point cloud to a two-dimensional point set sequence; Based on preset typical cross-section types, a parametric cross-section template model library is constructed, which is characterized by finite geometric parameters and geometric constraints to represent the cross-section profile. Step 3) Use the extreme radius difference and median absolute deviation to identify the abrupt change intervals of cross-sectional geometry in the two-dimensional point set sequence, and divide the two-dimensional point set sequence into abrupt change intervals and reserved intervals based on the abrupt change intervals of cross-sectional geometry; Step 4) Based on the parametric cross-section template model library, a classification method based on normalized radial distance features is used to identify the typical cross-section type of each two-dimensional point set in the two-dimensional point set sequence; Step 5) Based on the typical cross-sectional types of two-dimensional point sets, construct a robust fitness function that incorporates an additional error compensation mechanism and a weighted strategy; Step 6) Based on the typical cross-section types and robust fitness functions of two-dimensional point sets, construct a parameter association mechanism to form a unified basis parameter space; A parameter sharing mechanism is constructed based on a unified base parameter space; Based on parameter association and parameter sharing mechanisms, a parameter regularization mechanism is constructed. Step 7) Based on the parameter association mechanism, parameter sharing mechanism, and parameter regularization mechanism, construct a global optimization model that introduces a boundary penalty term; The teaching and learning optimization algorithm is used to solve the global optimization model, and the parameters of the spatially continuous and smooth parameterized geometric model are obtained. Based on the parameters of the parametric geometric model, a three-dimensional solid model of a linear engineering structure is reconstructed.

2. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, Step 1) involves obtaining the boundary points of the planar point cloud, including: Using a K-dimensional tree data structure to perform neighborhood search and obtain points of The nearest neighbor points; with points Establish a polar coordinate system for the pole and calculate the polar angles of all nearest neighbor points. And sorted in ascending order; Calculate the polar angle differences between adjacent points for all neighboring points, as well as the polar angle difference between the last polar angle and the first polar angle, and determine the maximum value of the polar angle differences. ; If satisfied Then point These are the boundary points of a planar point cloud. The threshold angle; The difference between adjacent polar angles is shown below: (1) The polar angle difference between the last polar angle and the first polar angle is shown below: (2) The maximum values ​​of all adjacent polar angle differences are shown below: (3) In the formula, The difference between adjacent polar angles, This is the difference between the last polar angle and the first polar angle.

3. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 2, characterized in that, The steps to determine the three-dimensional centerline include: Points were obtained by uniformly sampling on boundary curve 1. ,Pass Draw the normal to boundary curve 1 to intersect boundary curve 2 at... ,Pick and The midpoint is ;Pass Draw the normal to boundary curve 2 so that it intersects boundary curve 1 at... ,Pick and The midpoint is ;by and midpoint As the central axis point; A bidirectional normal intersection method is performed on all sampling points to obtain the central axis point set, and a random sampling consensus algorithm is used to fit the three-dimensional central axis. Boundary curve 1 is shown below: (4) Boundary curve 2 is shown below: (5) In the formula, Let $T$ be the degree of polynomial fitting. Let x be the j-th power of the x-axis variable.

4. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, In step 2), preprocessing includes denoising and segmentation operations; The steps to reduce the dimensionality of the preprocessed 3D point cloud to a 2D point set sequence include: set up To control the sampling density, the central axis parameter range is set. Divided into equal parts A portion, received 1 sampling point; with each odd number of sampling points The plane is defined with the origin at that point and the tangent vector of the central axis at that point as the plane's normal vector; it lies on the front boundary plane. and back boundary plane The point cloud between them is the sampling point. The corresponding point cloud region; Points in the point cloud region Projecting onto a plane yields the projection point, i.e.: (6) In the formula, The three-dimensional centerline at the sampling point The unit tangent vector at that point; At the sampling point Establish a two-dimensional local coordinate system at the location, with the basis vectors as shown below: (7) (8) In the formula, Sampling points The first and second unit basis vectors of the two-dimensional local coordinate system; , , They are respectively Components in the global x, y, and z directions; Transform the projected points to a two-dimensional local coordinate system to obtain a two-dimensional point set, i.e.: (9) In the formula, The transformation matrix; The parametric section template model library is shown below: (10) In the formula, For parameter set; For constraint sets; The number of geometric elements; The parameter set includes position parameters and shape parameters; The constraint set includes horizontal constraints. Vertical constraints Symmetry constraints Equality constraints .

5. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, Step 3), the step of identifying the abrupt change intervals in the cross-sectional geometry includes: Step 3.1) A sequence of two-dimensional point sets Viewed as a spatial sequence, each two-dimensional point set is shown below: (11) In the formula, and They are the i-th two-dimensional point sets The x and y coordinates of the j-th point in the array; Step 3.2) The scope is divided into Equally spaced angular regions, namely: (12) In the formula, ; Step 3.3) For each angular region Extract the maximum polar diameter and minimum polarity The polar radius difference between adjacent point sets is calculated as follows: (13) Step 3.4) Calculate the median of the absolute value sequence of polar radius differences, as shown below: (14) The absolute deviation of the median is shown below: (15) Step 3.5) Define the mutation detection threshold as follows: (16) In the formula, The median absolute deviation factor; when At that time, it was determined that a mutation was detected; Step 3.6) Count the number of mutations detected in each angular region for each two-dimensional point set, and sum them up to obtain the total mutation score. ; like Then determine the two-dimensional point set The set of mutation points is identified as the intervals of abrupt changes in cross-sectional geometry.

6. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 5, characterized in that, Step 4), identifying the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence, includes: Step 4.1) For a two-dimensional point set, take the geometric center of the bounding box of the rectangle as the pole. Divide the whole week into Each angle area ; for the angular region Extracting the polar radius set Calculate the mean polar radius within this angular region. ; Step 4.2) Repeated calculations were performed on each angular region to obtain... The 3D radial distance vector is shown below: (17) Maximum value normalization A dimensional radial distance vector, i.e.: (18) The normalized radial distance vector is shown below: (19) Step 4.3) Extract the multidimensional peak and valley information vector, concatenate the normalized radial distance vector with the multidimensional peak and valley information vector to form a multidimensional feature vector, use a random forest classifier for classification, and output the typical cross-sectional type of each two-dimensional point set in the two-dimensional point set sequence.

7. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, In step 5), the robust fitness function is as follows: (20) (21) In the formula, The number of geometric elements in the parametric section template model; The parameter vector of the parameterized section template model; Let r be the r-th parameter in the parameter vector; This represents the total number of parameters in the parameter vector; For the first in the parameter vector One parameter; , The r-th parameter The lower limit and upper limit of the value; Let be the weight of the j-th geometric element; It is the sum of the distances from each point assigned to the j-th geometric element to that element; The additional error for the j-th geometric element; The weights of the j-th geometric element are shown below: (22) In the formula, The number of points assigned to the j-th geometric element. The total number of points in the point set. As a safety factor; The additional error of the j-th geometric element is as follows: (23) In the formula, For the purpose of borrowing points, For the j-th geometric element, To borrow the number of points, For the first Minimum number of constraint points required for a geometric element The actual number of points earned. This is the distance from the borrowed point to the j-th geometric element.

8. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, In step 6), the parameter association mechanism is as follows: (24) (25) (26) (27) In the formula, For the first Types of parameter vectors For dimension The basis parameter vector, Let be the mapping function, and m be the number of independent constraint equations between different typical section types. ; The number of typical cross-section types contained in a two-dimensional point set; The parameter sharing mechanism refers to the fact that when multiple segments are processed along the same central axis, two-dimensional point sets are generated at the same sampling point position on the central axis for different segments; if the two-dimensional point sets correspond to the cross-sectional profile at the same spatial position and have the same typical cross-sectional type, then the instantiated basis parameters of the two-dimensional point sets share the same set of instantiated basis parameter values ​​in a unified basis parameter space. Shared groups are identified by matching the centerline parameters: if the centerline parameters of two two-dimensional point sets are consistent and the typical cross-section types are consistent, they are classified into the same shared group. Parametric regularization mechanisms include polynomial regularization, cosine function regularization, and sawtooth wave regularization.

9. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 8, characterized in that, Polynomial regularization is shown below: (28) In the formula, For the first One basic parameter, For the first The variation pattern of each basic parameter The order of the polynomial; For polynomial coefficients, These are the parameter values ​​for the sampling points; The cosine function regularization is shown below: (29) In the formula, For amplitude, Angular frequency, This is the initial phase. This is the offset; Sawtooth wave regularization is shown below: (30) In the formula, For period Local location within, The slope This is the starting value for the cycle.

10. The method for parametric modeling of linear engineering structures using point cloud-based geometric digital twins according to claim 1, characterized in that, In step 7), the global fitness function is as follows: (31) (32) In the formula, For point set The weighted fitness function, As weight, The penalty coefficient is... For boundary penalty terms, The number of points in the two-dimensional point set. This is the mapping function from the regularization coefficient vector to the parameter vector; The dimension of the regularization coefficient vector; Let l be the l-th regularization coefficient; , These are the l-th regularization coefficients. The lower limit and upper limit of the value; The mapping function from the regularization coefficient vector to the parameter vector is shown below: (33) In the formula, For point set The parameter vector corresponding to the parameterized section template model, This is the regularization coefficient vector; The boundary penalty terms are as follows: (34) In the formula, For the i-th two-dimensional point set (Q_i) represents the number of parameters in the parameterized section template model; For the i-th two-dimensional point set (Q_i) corresponds to the r-th parameter in the parameter vector; , Parameters The lower limit and upper limit of the value; The steps to solve the global optimization model include: exist Internal random generation vector of regularization coefficients The fitness is calculated as follows: (35) During the teacher phase, the process for each individual is as follows: (36) In the formula, This refers to the individual with the best fitness in the current population. is the mean vector of all individuals in the population; r is a random vector; As a teaching factor; Element-wise multiplication of vectors; This is the new regularization coefficient vector generated for the i-th individual after the teacher stage update; During the student phase, another individual is randomly selected. The learning direction is determined based on fitness level, that is: (37) In the formula, For the i-th individual, Another individual randomly selected during the student phase; , Individuals and The corresponding fitness function value; The individual is updated using a greedy selection strategy; A convergence detection mechanism based on a sliding window is introduced, with a window size of [value missing]. The relative rate of change threshold is When continuous The relative rate of change of the optimal value in the next iteration is less than If so, terminate early; The steps for outputting the reconstructed 3D solid model include: The regularization coefficient vector obtained from global optimization Through mapping function Convert into parameter vectors of the parameterized cross-section template model instantiation corresponding to each two-dimensional point set. ; Based on the geometric definition of the parametric section template model, determine the coordinates of the two-dimensional control points of each geometric element; By inverse coordinate transformation, the two-dimensional control points are mapped back to three-dimensional space to obtain the three-dimensional control points, i.e.: (38) In the formula, The transformation matrix is... This is the sampling point for the central axis.