Automatic mechanical model establishing method based on point cloud geometric model

Abstract

The invention relates to the field of structural health detection in civil engineering, in particular to an automatic mechanical model establishing method based on a point cloud geometric model. The method comprises the steps that semantic segmentation of a deep learning network is adopted, and a to-be-analyzed structure is segmented into independent components through different materials; embedding the segmented individual components into the divided grids based on a finite cell method to obtain a grid model of an overall structure; setting a displacement boundary condition and a force boundary condition according to the grid model of the overall structure; and performing automatic mechanical property analysis based on the grid model of the overall structure, the displacement boundary condition and the force displacement boundary condition, forming a structural stiffness matrix by applying a three-dimensional topological relation judgment rule, and calculating a node load vector of a discrete surface by applying a conversion process from area integral to domain integral, thereby completing mechanical analysis of the to-be-analyzed structure. According to the point cloud geometric model-based mechanical model automatic establishment method provided by the invention, a finite cell method is introduced as a theoretical support, and point cloud model-based autodynamics modeling analysis which does not need surface reconstruction and parameterization measurement and has high noise robustness is completed.

Description

Technical Field

[0001] The present invention relates to the field of structural health detection in civil engineering, and in particular to a method for automatically establishing a mechanical model based on a point cloud geometric model. Background Art

[0002] Bridge structures play a key role in the transportation system, so it is crucial to ensure their long-term safe operation. Such structures are subject to the combined effects of a variety of complex factors during their daily service, including direct effects such as structural deadweight and vehicle loads, as well as indirect effects such as changes in temperature and humidity, shrinkage and creep of concrete, etc. Bridge structure systems are usually large and complex, and some completed bridges are very old and lack effective management and maintenance systems. In addition, with the rapid economic development of my country, the traffic and passenger flow have increased rapidly, the bridge operating environment has continued to deteriorate, and bridge collapses have occurred from time to time. Therefore, mechanical evaluation of bridges is an important part of the long-term operation and maintenance of bridges. Based on the results of mechanical evaluation, engineers can formulate appropriate maintenance measures in a timely manner to maximize the life of the bridge and ensure the safety and smoothness of transportation.

[0003] Traditional mechanical evaluation methods mostly rely on commercial finite element software such as ABAQUS, ANSYS, and Midas, and rely heavily on manual work and design drawings. Not only does it require a lot of manpower and material resources, but also due to structural damage and long-term deformation, the modeling results will deviate from the actual structure. To solve these problems, researchers began to use standing laser scanners and depth cameras to obtain three-dimensional point clouds of structures, and use them to extract detailed geometric information to establish geometric models as the basis for evaluating mechanical properties, in order to improve modeling accuracy and reduce manpower investment. With the development of computer vision technology, the method of extracting geometric information using point cloud models has become more mature, and various optimization algorithms have emerged in an endless stream, but these methods still face a series of significant challenges, including sensitivity to noise and high parameter dependence. These limitations significantly affect the practicality and robustness of these methods in engineering applications. In recent years, deep learning technology has spawned cutting-edge technological innovations in a wide range of disciplines. The application of vision-based deep learning technology does not require parameter adjustment and is also noise-robust. It can use point cloud models as input and directly output the corresponding dimensional parameters, but this method requires a large amount of data sets as support, which is contrary to the current application status of bridge point cloud models. Therefore, if the process of extracting geometric information can be ignored and the point cloud model that implicitly contains geometric information is directly used as the basis for mechanical modeling, mechanical analysis can be performed more reasonably, thus providing a suitable solution for bridge performance evaluation.

[0004] Theoretical research on finite element automatic modeling based on point cloud geometric models has always been a cutting-edge topic. Existing methods mainly use surface reconstruction algorithms to convert geometric models into three-dimensional solid models such as triangulation or Poisson reconstruction, and then further convert them into mesh models to complete mechanical analysis. However, the effectiveness of surface reconstruction of these methods still has the problem of sensitivity to noise, which may be seriously affected by data loss or noise in the geometric model, thus adversely affecting the accuracy of finite element modeling. In recent years, with the further expansion of finite element theory, this problem has been gradually solved. The rise of immersed boundary methods such as the finite cell method provides a modeling idea that only needs to judge the relationship between the position of the integration point and the boundary, eliminating the necessity of surface reconstruction of the geometric model. However, since it is only applied to the analysis of a single individual, it lacks adaptability to the automatic modeling and analysis of complex large-scale civil engineering structures composed of multiple components. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method for automatically establishing a mechanical model based on a point cloud geometric model, aiming to make full use of the geometric information of the point cloud model and combine the existing mechanical analysis algorithm to realize the automatic dynamics analysis of the civil engineering structure, so as to improve the efficiency of bridge structure performance evaluation.

[0006] The present invention adopts the following technical solution:

[0007] The method for automatically establishing a mechanical model based on a point cloud geometric model of the present invention comprises the following steps:

[0008] Step S1, using semantic segmentation of a deep learning network to segment the structure to be analyzed into separate components based on different materials;

[0009] Step S2, embedding the individual components segmented in step S1 into the segmented grid based on the finite cell method to obtain a grid model of the overall structure;

[0010] Step S3, setting displacement boundary conditions and force boundary conditions according to the grid model of the overall structure;

[0011] Step S4, automatically analyze the mechanical properties based on the mesh model, displacement boundary conditions, and force displacement boundary conditions of the overall structure, use the three-dimensional topological relationship judgment rule to form the structural stiffness matrix, use the surface integral to domain integral transformation process to calculate the node load vector of the discrete surface, and complete the mechanical analysis of the structure to be analyzed.

[0012] The automatic establishment method of a mechanical model based on a point cloud geometric model described in the present invention forms a semantically hierarchical component segmentation in step S1 through a point cloud semantic segmentation network or a traditional point cloud semantic segmentation algorithm, so as to fully separate components with different material properties in the structure to be tested.

[0013] In the method for automatically establishing a mechanical model based on a point cloud geometric model of the present invention, the segmentation and cutting steps of step S2 are as follows:

[0014] Step S21, setting the initial position and the end position of the embedded grid, the area enclosed by the initial position and the end position fully includes the structure to be measured;

[0015] Step S22, setting the number of grids and the degree of grid polynomials;

[0016] Step S23, setting the level of fine meshing, controlling the integration points by parameters to approach the boundary of the structure to be measured, and performing octree fine meshing on an initial cubic mesh for each layer, until the final number of subdivision levels is reached.

[0017] In the method for automatically establishing a mechanical model based on a point cloud geometric model of the present invention, the displacement boundary conditions and the force boundary conditions in step S3 are as follows:

[0018] Set the displacement boundary condition as follows: the displacement boundary fits the mesh boundary, and set the displacement constraint type;

[0019] The boundary conditions for setting the force are: representing the force area in the form of a point cloud and setting the magnitude of the force.

[0020] The method for automatically establishing a mechanical model based on a point cloud geometric model of the present invention, step S4 is specifically as follows:

[0021] Step S41, the positional relationship between the integral points in each divided grid and the boundaries of the point clouds of different components segmented in S1 is determined by the three-dimensional topological relationship judgment rule, so as to assign different material properties and calculation weights to different integral points to complete the calculation of the stiffness matrix. The specific steps are as follows:

[0022] Perform preliminary filtering on the point clouds corresponding to different components of the structure that have been segmented in step S1. For any integral point gi = (xgi, ygi, zgi),

[0023] The effective interval is intercepted along the PCA principal axis direction of the above component point cloud (assuming it is the X-axis direction):

[0024] X ini ={(x i ,y i ,z i ),x i ∈[x i -Δ 1 ,x i +Δ 1 ]}

[0025] And project the point cloud in the valid interval along the X-axis to the YOZ plane where the Gaussian integral point is located;

[0026] Extract the effective interval along the Y and Z axes in the two-dimensional plane:

[0027] Y ini ={(x gi ,y i ,z i ),z i ∈[z i -Δ 2 ,z i +Δ 2 ]}

[0028] Z ini ={(x gi ,y i ,z i ),y i ∈[y i -Δ 2 ,y i +Δ 2 ]}

[0029] Filter out the interval with too large Z-axis projection density in the effective area of ​​Y direction,

[0030] In the effective area of ​​Z direction, filter out the interval with excessive Y-axis projection density.

[0031] Based on the extracted final effective intervals of the Y direction and the Z direction, the integration point affiliation is determined in the final effective interval according to the basic determination principle;

[0032] If all three directions are inside the point cloud, it is an internal integration point, otherwise it is an external integration point;

[0033] The steps to calculate different weights α according to the internal and external positions of the integration points are as follows:

[0034]

[0035] Where Ω phys Represents the physical domain, and all internal integration points belong to the physical domain; Ω fict represents the virtual domain, and all external integral points belong to the virtual domain;

[0036] According to the material properties of the component area where the integration point is located, the material matrix D of the integration point is assigned:

[0037]

[0038] Where E is the elastic modulus and ν is Poisson's ratio;

[0039] Apply the calculation weight α to the material matrix D to form the stiffness matrix K, which is expressed as follows:

[0040] K = ∫Ω (LN) T αD(LN)dV

[0041] Step S42, by converting the surface integral into a domain integral, the surface integral calculation of the point cloud surface is realized, and the integration of the node load vector is completed; the specific steps are as follows:

[0042] Conversion of surface integrals to volume integrals in the calculation of nodal load vectors:

[0043]

[0044] In the formula, δ(x) is the distance weight function, and the distance weight function expression is as follows:

[0045]

[0046] During use, it is regularized to obtain the regularized distance weight function δ ε (x):

[0047]

[0048] By using the regularized distance weight function δ ε (x), the force load vector F is calculated by the following formula:

[0049]

[0050] where d PCA is the calculation function of the distance between the integral point and the boundary point, which is the distance between any integral point and the fitted straight line of the n points closest to the boundary point set;

[0051] The following linear equations are solved by using the stiffness matrix K calculated in step S41 and the load vector F obtained in step S42 to obtain the node displacement vector u to be determined:

[0052] K×u=F

[0053] According to the displacement interpolation function N, the full-field displacement field of the structure to be analyzed is calculated:

[0054]

[0055] Step S43, calculate the stress and strain fields of the structure to be analyzed according to the geometric equation and the constitutive equation to complete the mechanical analysis:

[0056] ε=L×U;σ=D×ε。

[0057] Beneficial Effects

[0058] 1) Currently, the finite element modeling method based on point cloud model mostly relies on parametric measurement geometry information or surface reconstruction technology, which is highly sensitive to noise and highly dependent on the set parameters or data sets. This method consumes more computing resources and the calculation results are not robust enough.

[0059] The present invention provides a method for automatically establishing a mechanical model based on a point cloud geometric model, which introduces the finite cell method as a theoretical support and completes the self-dynamic modeling analysis based on the point cloud model with high noise robustness without the need for surface reconstruction and parametric measurement.

[0060] 2) Traditional automatic analysis based on the finite cell method is only used for the analysis of single non-engineering structural components because it cannot determine the material affiliation of the point cloud.

[0061] In order to achieve an overall analysis of complex engineering structures, the present invention provides an automatic mechanical model establishment method based on a point cloud geometric model, introduces a point cloud semantic segmentation algorithm as a support, realizes the semantic segmentation of complex structures at a semantic level, completes the assignment of material properties to the integral points of the complex structure, and realizes the actual engineering structure autodynamics analysis based on the point cloud geometric model.

[0062] 3) Existing algorithms for judging integration points and point cloud boundaries usually rely on the normal vectors or display surfaces of the point cloud model to improve their detection accuracy, but this limits their feasibility in practical engineering applications with limited computing resources. Therefore, unlike the more complex algorithm designs in the past, the present invention focuses on engineering practicability and aims to get rid of the dependence on high-performance hardware to achieve fast and accurate acquisition of the spatial subordination of integration points. By cleverly conceiving the three-dimensional topological relationship judgment rule, the spatial position subordination relationship between the integration point and the point cloud boundary of components of different materials is successfully judged, and the automatic integration of the stiffness matrix is ​​completed.

[0063] 4) Currently, the surface integral of discrete surfaces (point cloud surfaces without surface reconstruction) is inconvenient to calculate, resulting in the discrete mechanical analysis of point clouds being able to only calculate the response under the action of volume forces, but unable to calculate the results under the action of surface forces. The present invention focuses on the application method of discrete surface boundary conditions, and proposes a calculation method for converting surface integrals into volume integrals, which smoothly applies force boundary conditions in the point cloud model, aligns the structural displacement boundary with the grid boundary, applies displacement boundary conditions by forcibly modifying the stiffness matrix, smoothly completes the modification of the stiffness matrix and the automatic calculation of the load vector, and establishes a group of equilibrium equations for the entire structure.

[0064] 5) The goal of the present invention is to evaluate the real response of the structure. Therefore, the present invention not only obtains the displacement field, strain field and stress field of the structure under load, but also analyzes the dynamic response of the structure, aiming to more comprehensively analyze the actual operation and maintenance of the structure under long-term load, and more reasonably evaluate its safety, applicability and durability. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 Schematic diagram of the overall framework of the present invention;

[0066] Figure 2 A schematic diagram of the semantic segmentation proposed by the present invention;

[0067] Figure 3 A schematic diagram of the grid division proposed by the present invention;

[0068] Figure 4 Schematic diagram of the grid subdivision principle proposed in the present invention;

[0069] Figure 5 Schematic diagram of the basic principle of judging the integral points and point cloud boundaries proposed by the present invention;

[0070] Figure 6 Schematic diagram of the filtering effect of the effective area of ​​the integral point proposed by the present invention;

[0071] Wherein (a) is a schematic diagram of the subordination determination of the integral points before the final filtering, and (b) is a schematic diagram of the subordination determination of the integral points after the final filtering;

[0072] Figure 7 Schematic diagram of the process of assigning integral point materials according to the present invention;

[0073] Among them, (a) is the integration point in material 1, (b) is the principle diagram, (c) is the integration in material 2, and (d) is the integration point in the virtual domain;

[0074] Figure 8 Schematic diagram of the canonical function for converting surface integral into volume integral according to the present invention;

[0075] Fig. 9 Detailed diagram of distance function calculation in the process of converting surface integral to volume integral proposed by the present invention;

[0076] Fig.10 The bridge test diagram mentioned in the example;

[0077] Fig.11 The point cloud model diagram of the bridge mentioned in the example;

[0078] Fig.12 The semantic segmentation effect diagram mentioned in the example;

[0079] Fig.13The meshing effect diagram mentioned in the example;

[0080] Fig.14 Comparison chart of calculation results mentioned in the example;

[0081] In the figure, (a) is a comparison diagram of displacement and stress solution under 20KN load, (b) is a comparison diagram of displacement and stress solution under 40KN load, and (c) is a comparison diagram of displacement and stress solution under 60KN load. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0083] The present invention is a method for automatically establishing a mechanical model based on a point cloud geometric model. First, in order to overcome the problem that the traditional use of point cloud models requires the generation of surfaces and is inconvenient to automatically divide different materials, a point cloud semantic segmentation strategy is introduced to achieve semantic segmentation of components at the real semantic level, and the entire complex civil engineering structure is split into a combination of individual components with different material properties. Subsequently, based on the finite cell method as the basic principle, a three-dimensional topological relationship judgment method for the grid integral point and the point cloud boundary is designed to achieve the integration of the stiffness matrix. Secondly, the surface integral is converted into a domain integral to achieve the calculation of the node load vector. The linear equations formed by the final calculation obtain the displacement, stress and strain response of the structure to be tested, so as to realize the evaluation of the self-dynamic performance of the structure.

[0084] like Figure 1 , Figure 2 As shown: The present invention is a method for automatically establishing a mechanical model based on a point cloud geometric model. The overall process of the solution includes the following components:

[0085] Step 1: Complex structure component splitting based on semantic segmentation. This method uses deep learning networks or other classic semantic segmentation algorithms to split complex structures into semantically hierarchical components, providing support for the subsequent assignment of different material properties, solving the problem that traditional methods are inconvenient to automatically assign materials.

[0086] The specific details of the complex structure component splitting based on semantic segmentation in the first step are as follows:

[0087] First, the point cloud model is filtered and denoised to remove the influence of significant noise; then, a semantic hierarchical component segmentation is formed through a point cloud semantic segmentation network such as PointNet, PointNet++, RandLa-Net, SCF-Net or a traditional point cloud semantic segmentation algorithm such as an algorithm based on projection density slicing, an algorithm based on clustering, etc., which requires that the components with different material properties in the structure to be tested be fully separated, and the point cloud segmentation accuracy of the components should be greater than 95%.

[0088] like Figure 3 , Figure 4 As shown:

[0089] Step 2: Mesh division based on the basic theory of finite cell method. In order to facilitate subsequent calculation and analysis, the point cloud model of the structure to be tested should be fully embedded in the divided grid, and the structure subdivision grid should be kept close to the structure boundary through continuous refinement. The specific steps are as follows:

[0090] (1) Set the initial position and end position of the embedded grid. The area enclosed by the initial position and the end position is required to fully contain the structure to be measured, and the displacement constraint of the structure to be measured should be as close to the boundary of the area as possible.

[0091] (2) Set the number of grids and the degree of grid polynomial. These two parameters control the number of degrees of freedom in the final calculation. Setting the number of grids includes setting the number of grids on the X, Y, and Z axes. It should be ensured that all the divided grids are regular hexahedral grids.

[0092] (3) Set the mesh refinement level. This parameter controls the accuracy of the integration point approaching the boundary of the structure to be measured. Each time a layer is deepened, an initial cube mesh is refined by an octree until the final number of subdivision levels is reached. During the division process, the criterion for whether to refine is whether the mesh contains part of the point cloud model. If the mesh contains a point cloud, it is regarded as a boundary mesh and the next level of refinement is performed. Otherwise, it is regarded as an internal mesh or an external mesh and no refinement operation is performed. The execution process is shown in Figure 4 .

[0093] Step 3: Setting boundary conditions. The boundary conditions set include force boundary conditions and displacement boundary conditions to ensure that the mechanical model based on the point cloud model component can be solved.

[0094] The setting of boundary conditions includes the following steps:

[0095] The displacement boundary conditions are set as follows: the displacement boundary is required to fit the mesh boundary as closely as possible, and the displacement constraint type is set accurately.

[0096] The boundary conditions for setting the force are: the force area must be represented in the form of a point cloud set, and the magnitude of the force must be accurately set.

[0097] like Figure 5-Figure 9 As shown:

[0098] Step 4: Automatic analysis of mechanical properties based on the point cloud model and its mesh. The stiffness matrix is ​​integrated by judging the three-dimensional topological relationship between the integration point and the boundary point cloud, and the node load vector is calculated by converting the surface integral into the volume integral. The automatic analysis of mechanical properties based on the point cloud model and its mesh includes the following steps:

[0099] Step 1: Determine the weight of the integration points based on the three-dimensional spatial relationship between the integration points in the grid and the boundary point cloud, and integrate the stiffness matrix. The basic criteria for determining the positional relationship between the integration points and the boundary point cloud are as follows: Figure 5 As shown in the figure, if there are point clouds on both sides of the integration point cross-section plane in both directions, the integration point is considered an inner integration point, otherwise it is an outer integration point. The judgment algorithm example is as follows:

[0100] 1) Perform preliminary filtering on the point cloud. For any integral point g i =(x gi ,y gi ,z gi ), the effective interval along the PCA main axis direction of the point cloud model (assuming it is the X-axis direction) is:

[0101] X ini ={(x i ,y i ,z i ),x i ∈[x i -Δ 1 ,x i +Δ 1 ]}

[0102] The point cloud in the valid interval is projected along the X axis to the YOZ plane where the Gaussian integral point is located. During the calculation process, Δ 1 Take 3 times the point spacing in the point cloud model;

[0103] 2) Extract the effective interval along the Y and Z axes in the two-dimensional plane:

[0104] Y ini ={(x gi ,y i ,z i ),z i ∈[z i -Δ 2 ,z i +Δ 2 ]}

[0105] Z ini ={(x gi ,y i ,zi ),y i ∈[y i -Δ 2 ,y i +Δ 2 ]}

[0106] During the calculation process, Δ 2 Take 5 times the point spacing in the point cloud model;

[0107] Then, in the Y-direction effective area, the interval with too large Z-axis projection density is filtered out, and in the Z-direction effective area, the interval with too large Y-axis projection density is filtered out to form the final filtered effective interval. The purpose of this filtering step is to eliminate the judgment error caused by the integration point being too close to the boundary. For example, a two-dimensional example Figure 6 shown.

[0108] Finally, in the final effective interval of Y and Z directions, the integration point is determined to be subordinate according to the basic judgment principle. It is worth noting that if the component has no obvious principal axis, it is directly determined whether its X, Y, and Z directions are all located inside the point cloud. If all three directions are located inside the point cloud, it is an internal integration point, otherwise it is an external integration point.

[0109] Calculate different calculation weights α according to the position of the integration point:

[0110]

[0111] Among them, Ωphys represents the physical domain, and all inner integration points belong to the physical domain; Ωfict represents the virtual domain, and all outer integration points belong to the virtual domain.

[0112] 3) According to the material properties of the component area where the integration point is located, the material matrix D of the integration point is assigned:

[0113]

[0114] Where E is the elastic modulus and ν is Poisson's ratio. These parameters are only related to the material of the component itself. The process of classifying different materials is shown in Figure 7 .

[0115] 4) Apply the calculation weight α to the material matrix D to form the stiffness matrix K:

[0116] K = ∫ Ω (LN) T αD(LN)dV

[0117] Step 2: Calculate the node load vector by converting the surface integral into the volume integral. The specific method is as follows:

[0118] The key link is to realize the conversion from surface integral to volume integral in the process of calculating the node load vector:

[0119]

[0120] In the formula, δ(x) is the distance weight function, which has the following basic properties:

[0121]

[0122] During use, it is regularized to obtain the regularized distance weight function δ ε (x):

[0123]

[0124] The function value can be found in Figure 8 , the calculation of the force load vector is done by using this function:

[0125]

[0126] where d PCA is the calculation function of the distance between the integral point and the boundary point. This function calculates the distance between any integral point g and the straight line fitted by the n points closest to g in the boundary point set. PCA The calculation principle of Fig. 9 .

[0127] Step 3: Solve the linear equations and complete the mechanical analysis of the structure. The specific method is as follows:

[0128] By calculating the stiffness matrix K and the load vector F, the following linear equations are solved to obtain the node displacement vector u:

[0129] K×u=F

[0130] According to the displacement interpolation function N, the full-field displacement field of the structure to be analyzed is calculated:

[0131]

[0132] According to the geometric equation and constitutive equation, the stress and strain fields of the structure to be analyzed are calculated to complete the mechanical analysis of the structure:

[0133] ε=L×U;σ=D×ε

[0134] Embodiment 1:

[0135] The proposed scheme was tested on a steel arch bridge. The experimental arrangement and node numbering are shown in Fig.10. The steel used for the bridge is Q235, and its elastic modulus, density and Poisson's ratio are 200Gpa, 7.85g / cm3 and 0.3 respectively. The elastic modulus, density and Poisson's ratio of the suspension cable are 150Gpa, 2.68g / cm3 and 0.33 respectively. The displacement constraint of the structure is fixed at both ends, and the surface load is applied to the beam between nodes 4-5 (P4) and 7-8 (P7), with loading sizes of 20KN, 40KN and 60KN respectively. In order to monitor the structural response, strain gauges are installed on the nodes, and 18 Esco PT8510 rope displacement sensors are installed on the beams, and the displacement data are collected using the NIPXIe-1082 data acquisition system.

[0136] First, point cloud acquisition and component segmentation are performed. The bridge point cloud is acquired by multi-view geometric reconstruction. The arch bridge is photographed using a Canon EOS800D series camera. During the shooting process, the image overlap rate between multiple perspectives is guaranteed to be greater than 50%. The pixel size of the captured image is 5328×4000. After the image is captured, the DJ I series software is used for 3D reconstruction. After reconstruction, the background points are manually filtered to obtain a point cloud model (the manual filtering here is the only part that requires manual operation in the calculation process), and the point cloud model is statistically filtered and denoised to obtain the final arch bridge point cloud, as shown in Figure 1. Fig.11 The original component is divided into three structural parts: arch, hanger and I-beam. The segmentation results are shown in Fig.12 As shown in Figure 2, the local PCA direction of the corresponding component has been calculated during the segmentation process to facilitate the determination of the subordinate relationship between the integral point and the point cloud boundary.

[0137] Then, the segmented arch bridge is meshed, using an initial mesh of 31×8×5, a subdivision level of 3, and a maximum mesh polynomial degree of 3. Fig.13 shown.

[0138] Secondly, the boundary conditions are set: apply fixed displacement constraints in the z and y directions at the ends, and apply loads of 20KN, 40KN, and 60KN respectively to the point cloud set in the force action area.

[0139] Finally, the mechanical analysis based on the mesh and point cloud model is completed to calculate the displacement field and stress-strain field. The experimental results of nodes 2 to 10 (the measurement points where sensors are arranged), the traditional finite element calculation results (FEM results) and the calculation results of the proposed method (point cloud FCM results) are shown in Fig.14 The traditional finite element model was established using ANSYS APDL based on the drawings, using Beam188 units to simulate I-beams and arches, SOLID85 units to simulate the nodes at the displacement constraints at both ends of the arch, and Link8 units to simulate the hangers.

[0140] Error 1 is the error between the proposed method and the experiment. Error 2 is the error between the proposed method and the traditional finite element method. For displacement analysis, the average errors (average of errors of all nodes) between the proposed point cloud FCM method and the experimental results are 10.1%, 9%, and 10.9% at three loading levels, respectively, and the average errors with the finite element results are 1.7%, 1.5%, and 1.2% at three loading levels, respectively. In terms of stress, the average errors between the point cloud FCM method and the experiment are 10.9%, 9.0%, and 9.1%, respectively, and the average errors with the finite element results are 3.7%, 3.2%, and 3.4%, respectively. In practical applications, neither the finite element model nor the point cloud FCM model has been model-corrected, resulting in a certain deviation from the experimental data, but the average error of the proposed method is consistent with the average deviation of the finite element method, which shows that the accuracy of this method is close to that of the traditional FEM method.

[0141] In summary, the specific embodiments verify the effectiveness of the solution proposed in the present invention and its applicability to complex projects.

[0142] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for automatically establishing a mechanical model based on a point cloud geometric model, characterized in that: Here are the steps: Step S1, using semantic segmentation of a deep learning network to segment the structure to be analyzed into separate components based on different materials; Step S2, embedding the individual components segmented in step S1 into the segmented grid based on the finite cell method to obtain a grid model of the overall structure; Step S3, setting displacement boundary conditions and force boundary conditions according to the grid model of the overall structure; Step S4, based on the mesh model, displacement boundary conditions, and force displacement boundary conditions of the overall structure, automatic mechanical performance analysis is performed, the structural stiffness matrix is ​​formed using the three-dimensional topological relationship judgment rule, and the node load vector of the discrete surface is calculated using the surface integral to domain integral transformation process to complete the mechanical analysis of the structure to be analyzed.

2. The method for automatically establishing a mechanical model based on a point cloud geometric model according to claim 1, characterized in that: In step S1, a semantic-level component segmentation is formed through a point cloud semantic segmentation network or a traditional point cloud semantic segmentation algorithm, so as to fully separate components with different material properties in the structure to be tested.

3. The method for automatically establishing a mechanical model based on a point cloud geometric model according to claim 1, characterized in that: The segmentation and cutting steps of step S2 are as follows: Step S21, setting the initial position and the end position of the embedded grid, the area enclosed by the initial position and the end position fully includes the structure to be measured; Step S22, setting the number of grids and the degree of grid polynomials; Step S23, setting the level of fine meshing, controlling the integration points by parameters to approach the boundary of the structure to be measured, and performing octree fine meshing on an initial cubic mesh for each layer, until the final number of subdivision levels is reached.

4. The method for automatically establishing a mechanical model based on a point cloud geometric model according to claim 1, characterized in that: The displacement boundary conditions and force boundary conditions in step S3 are as follows: Set the displacement boundary condition as follows: the displacement boundary fits the mesh boundary, and set the displacement constraint type; The boundary conditions for setting the force are: representing the force area in the form of a point cloud and setting the magnitude of the force.

5. The method for automatically establishing a mechanical model based on a point cloud geometric model according to claim 1, characterized in that: Step S4 is as follows: Step S41, using the three-dimensional topological relationship judgment rule to judge the positional relationship between the integral points in each divided grid and the boundaries of the segmented point clouds of different components, assigning different material properties and calculation weights to different integral points to complete the calculation of the stiffness matrix, the specific steps are as follows: The point clouds corresponding to the different components of the segmented structure are initially filtered. For any integral point g i =(x gi ,y gi ,z gi ), The effective interval is intercepted along the PCA main axis direction of the component point cloud: X ini ={(x i ,y i ,z i ),x i ∈[x i -Δ1,x i +Δ1]} And project the point cloud in the valid interval along the X-axis to the YOZ plane where the Gaussian integral point is located; Extract the effective interval along the Y and Z axes in the two-dimensional plane: Y ini ={(x gi ,y i ,z i ),z i ∈[z i -Δ2,z i +Δ2]} Z ini ={(x gi ,y i ,z i ),y i ∈[y i -Δ2,y i +Δ2]} Filter out the interval with too large Z-axis projection density in the effective area of ​​Y direction, In the effective area in the Z direction, filter out the area with excessive Y-axis projection density. Based on the extracted final effective intervals of the Y direction and the Z direction, the integration point affiliation is determined in the final effective interval according to the basic determination principle; If all three directions are inside the point cloud, it is an internal integration point, otherwise it is an external integration point; The steps to calculate different weights α according to the internal and external positions of the integration points are as follows: Where Ω phys Represents the physical domain, and all internal integration points belong to the physical domain; Ω fict represents the virtual domain, and all external integral points belong to the virtual domain; According to the material properties of the component area where the integration point is located, the material matrix D of the integration point is assigned: Where E is the elastic modulus and ν is Poisson's ratio; Apply the calculation weight α to the material matrix D to form the stiffness matrix K, which is expressed as follows: K=∫ Ω (LN) T αD(LN)dV Step S42, by converting the surface integral into a domain integral, the surface integral calculation of the point cloud surface is realized, and the integration of the node load vector is completed; the specific steps are as follows: Conversion of surface integrals to volume integrals in the calculation of nodal load vectors: In the formula, δ(x) is the distance weight function, and the distance weight function expression is as follows: During use, it is regularized to obtain the regularized distance weight function δ ε (x): By using the regularized distance weight function δ ε (x), the force load vector F is calculated by the following formula: where d PCA is the calculation function of the distance between the integral point and the boundary point, which is the distance between any integral point and the fitted straight line of the n points closest to the boundary point set; The following linear equations are solved by using the stiffness matrix K calculated in step S41 and the load vector F obtained in step S42 to obtain the node displacement vector u: K×u=F According to the displacement interpolation function N, the full-field displacement field of the structure to be analyzed is calculated: Step S43, calculate the stress and strain fields of the structure to be analyzed according to the geometric equation and the constitutive equation to complete the mechanical analysis: ε=L×U;σ=D×ε。

Images