Three-dimensional modeling analysis processing method based on precast beam field and steel bar net

Abstract

The present application belongs to the technical field of three-dimensional modeling, and relates to a three-dimensional modeling analysis processing method based on a precast beam field and a steel bar network. The present application extracts multi-scale geometric and material characteristics by fusing laser radar point cloud and depth camera image, and introduces a component physical property vector to generate stress field characteristics. At the same time, based on historical frame data, the topological connection characteristics of the steel bar network are constructed and enhanced. This method solves the problems of multi-scale feature extraction distortion caused by fixed analysis scale, mechanical analysis deviation from reality caused by lack of physical property fusion, and inability to perceive structure time evolution caused by dependence on single-frame static modeling in the prior art. Finally, by combining calibrated stress field and enhanced topological characteristics and performing abnormality determination according to the mechanical equilibrium equation, high-precision and dynamic analysis of the precast beam structure from geometric shape to stress state is realized, and the automation level of quality detection and the reliability of early risk warning are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional modeling technology and relates to a three-dimensional modeling analysis and processing method based on precast beam yards and steel mesh. Background Technology

[0002] As key load-bearing components in bridges, buildings, and other engineering projects, the manufacturing quality of precast beams directly affects the safety and durability of the overall structure. Traditional quality control in precast beam yards mainly relies on manual inspections and sampling, which suffers from low efficiency, strong subjectivity, and difficulty in achieving full coverage. With the development of 3D sensing and modeling technologies, using equipment such as LiDAR and depth cameras for 3D digital acquisition and analysis of beam yards and reinforcing mesh has become an important direction for improving the level of intelligent quality control.

[0003] However, existing technologies for 3D modeling and analysis based on precast beam yards and steel mesh still have the following key drawbacks: First, traditional point cloud processing methods typically employ a fixed, single neighborhood scale. Precast beam scenes simultaneously contain steel reinforcement nodes with millimeter-level precision and beam surfaces with meter-level scale; a single scale cannot adequately capture both: large scales smooth node details, leading to distortion in node recognition and geometric modeling; small scales fail to capture the overall beam shape and impose a large computational load, affecting the integrity and efficiency of 3D modeling.

[0004] Secondly, existing methods are mostly limited to geometric shape recognition and fail to effectively integrate material appearance information with the physical and mechanical properties of components. This results in models that lack physical meaning and cannot truly reflect the behavior of components under actual stress, making mechanical analysis and anomaly judgment based on the model lack practical basis, and warning results often deviate from the actual structural condition.

[0005] Finally, most existing 3D modeling methods are based on single-frame data, which falls under static modeling and analysis. Reinforcing bar nodes may undergo gradual changes during construction. Single-point-of-time modeling cannot capture the temporal evolution of node connection states, resulting in the model failing to reflect changes in structural stability and thus making it difficult to achieve early risk identification and warning based on time-series data. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a three-dimensional modeling and analysis method based on precast beam yard and steel mesh is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: a three-dimensional modeling and analysis method based on precast beam yard and steel mesh, including: acquiring three-dimensional point cloud data of precast beam yard and steel mesh through lidar, and synchronously acquiring corresponding image data through depth camera.

[0008] Based on 3D point cloud data, local geometric features of steel bar nodes and global morphological features of beam surface are extracted through multi-scale neighborhood analysis, and material difference features are extracted based on image data.

[0009] The weights of local geometric features and global morphological features are dynamically assigned according to the component type, and the stress field features are generated by integrating material difference features with the preset component physical property vector. The stress field features are then calibrated based on the material properties to obtain the calibrated stress field features.

[0010] Based on historical frame data, the adjacency relationship between rebar nodes is constructed and topological connection features are generated. By verifying and enhancing the stability of the adjacency relationship, enhanced topological features are obtained.

[0011] By integrating the calibration stress field characteristics and enhanced topological characteristics, constraint calibration is performed based on the mechanical equilibrium equation to determine the structural anomaly parameters; if the structural anomaly parameters exceed the preset threshold, an anomaly warning signal is output.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention effectively solves the modeling contradiction that exists when the traditional fixed scale method processes millimeter-level steel bar nodes and meter-level beam surfaces by dynamically allocating the analysis scale according to the component type and adopting a multi-scale neighborhood analysis strategy. This method can accurately extract the local details of the nodes and the global shape of the beam, thereby improving the modeling efficiency while ensuring the geometric integrity and detail authenticity of the three-dimensional model.

[0013] (2) This invention integrates material difference features with component physical property vectors, dynamically allocates the weights of local geometric features and global morphological features, and calibrates the features based on material properties. This method solves the problem of the disconnect between geometric shape recognition and physical mechanical properties in the prior art, enabling the model to have physical meaning and truly reflect the behavior of the component under actual stress, thereby improving the accuracy of mechanical analysis and the reliability of anomaly judgment, and ensuring that the early warning results are highly consistent with the actual structural condition.

[0014] (3) This invention constructs the adjacency relationship between rebar nodes using historical frame data, verifies and enhances the stability of the adjacency relationship, and dynamically updates the topological connection features by combining time-series data. This method solves the problem that traditional single-frame data modeling cannot capture the temporal evolution of the structure, realizes continuous monitoring of the changes in the connection state of rebar nodes, can promptly detect gradual structural changes, significantly improves the early risk identification capability and early warning timeliness, and provides reliable protection for construction safety. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a step diagram of the three-dimensional modeling and analysis method based on precast beam yard and steel mesh in this invention.

[0017] Figure 2 This is a flowchart of the method for obtaining the calibration stress field characteristics in this invention.

[0018] Figure 3 This is a flowchart of the method for obtaining enhanced topological features in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, the present invention provides a three-dimensional modeling and analysis processing method based on precast beam yard and steel mesh, including: S1, collecting three-dimensional point cloud data of precast beam yard and steel mesh through lidar, and synchronously collecting corresponding image data through depth camera.

[0021] Considering that digital modeling and analysis of precast beam yards requires high-precision, high-completeness 3D geometric information, as well as visual information that can help distinguish materials and surface conditions, a single sensor is insufficient. Therefore, a LiDAR and an RGB-D depth camera are used for synchronous collaborative acquisition to obtain complementary point cloud data and image data.

[0022] In this embodiment, a sensor system configured with specific parameters performs synchronous data acquisition. The angular resolution of the lidar must ensure that the spatial resolution of the point cloud is better than 2 cm, and the effective detection range covers the working area within 10 meters. The depth camera uses a camera with integrated RGB-D function to acquire image data containing color and depth information to help distinguish differences in surface materials.

[0023] After data acquisition, for the 3D point cloud data: First, a statistical outlier removal algorithm is used to identify and remove outlier noise points by calculating the mean and standard deviation of the distances within the neighborhood of each point; then, motion estimation and coarse registration are performed on multiple temporally continuous point clouds based on the LeGO-LOAM algorithm to eliminate distortion introduced by the movement of the scanning platform; finally, the iterative nearest point algorithm is used to align and stitch point cloud segments from different scanning perspectives to form 3D point cloud data.

[0024] Meanwhile, for RGB-D image data: denoising processing, including bilateral filtering, is performed to smooth the image and preserve edges, and color and depth values ​​are normalized to enhance the stability of subsequent feature extraction.

[0025] S2. Based on 3D point cloud data, local geometric features of steel bar nodes and global morphological features of beam surface are extracted through multi-scale neighborhood analysis, and material difference features are extracted based on image data.

[0026] Given that the reinforcing mesh and the beam have different structural dimensions and focuses: the reinforcing mesh nodes require detailed local geometric descriptions to detect connection anomalies, while the beam surface requires macroscopic morphological descriptions to assess overall deformation. Furthermore, differences in material properties such as rust and stains on the surface are crucial for judging the condition of the component, and geometric information alone is insufficient for effective identification.

[0027] Therefore, by implementing multi-scale neighborhood analysis, local and global geometric features are extracted respectively, and material difference features are extracted from image data at the same time.

[0028] In one specific embodiment, the neighborhood scale is first dynamically set according to the component type to which each point in the point cloud belongs, such as a rebar node or a beam surface.

[0029] For points classified as rebar nodes, a small-scale neighborhood (e.g., radius r=5cm) is used. By constructing a local graph within its first-order neighborhood and performing graph convolution operations, local geometric features such as node curvature and normal vector distribution are extracted.

[0030] For points classified as beam surfaces, a large-scale neighborhood (e.g., radius r=20cm) is used. Graph convolution and feature aggregation are performed within its extended third-order neighborhood to extract global morphological features such as surface continuity and curvature variation trends.

[0031] Meanwhile, a multi-channel tensor containing information on surface color distribution, texture features, and crack depth is extracted from RGB-D image data as material difference features.

[0032] The weight parameters of the graph convolutional network used to extract geometric features can be obtained through supervised training on a dataset containing a large number of labeled rebar nodes and concrete surface point cloud samples.

[0033] S3. Dynamically allocate the weights of local geometric features and global morphological features according to the component type, and integrate material difference features with the preset component physical attribute vector to generate stress field features; calibrate the stress field features based on material properties to obtain calibrated stress field features.

[0034] Considering that different component types have different stress characteristics, and the influence of local geometric features and global morphological features on their stress state varies, using a fixed weight allocation method will lead to inaccurate stress field characterization.

[0035] Meanwhile, material differences, such as steel corrosion, can change the actual material properties of components, thereby affecting stress distribution. If this factor is ignored and the original stress field characteristics are used directly, the reliability of subsequent structural anomaly detection will be reduced. Therefore, it is necessary to dynamically allocate feature weights and calibrate the stress field characteristics in combination with material properties.

[0036] Therefore, the dynamic allocation of feature weights is achieved through gating signals, stress field features are generated through multimodal feature fusion, and calibrated stress field features are obtained based on the identification of rusted areas and the mapping of calibration coefficients.

[0037] In one specific embodiment, the method for obtaining the stress field features is as follows: First, a gating signal is generated based on the component type. Then, the gating signal is used as a weight to perform a weighted summation of local geometric features and global morphological features. Specifically, the gating signal is obtained by processing the component type encoding vector through a single fully connected layer and then calculating it using a sigmoid activation function. Its output value is within the range [0, 1].

[0038] Then, the weighted fused features and material difference features are spliced ​​together using the channel dimension, and then processed by batch normalization to achieve multimodal feature fusion. Finally, the features are spliced ​​together with the preset component physical attribute vector, which provides physical constraints and prior guidance for the mapping process from geometric and visual features to mechanical quantities, forming stress field features that characterize the spatial stress state.

[0039] The component physical property vector includes the yield strength of the steel reinforcement and the elastic modulus of the concrete, both of which have been normalized.

[0040] Further, please refer to Figure 2 As shown, the method for obtaining the calibration stress field features is as follows: based on the color distribution and crack depth information in the material difference features, the corrosion area is identified. The specific method for identifying the corrosion area is as follows: convert RGB colors to HSV space, extract pixels with Hue values ​​between 30 and 60 degrees as corrosion candidate points; screen out candidate points with crack depth greater than 1.2 times the average value of the neighborhood, and then remove areas with an area of ​​less than 50 pixels through connected component analysis to complete the corrosion area identification.

[0041] Then, from the material difference features corresponding to the identified rusted areas, the color histogram features are extracted, and the proportion of the total number of pixels in the rusted area to the total number of pixels on the surface of the entire component is calculated to obtain the rusted area percentage.

[0042] The material property calibration coefficient is determined based on the preset mapping relationship table between the rust area ratio and the calibration coefficient.

[0043] For example, the percentage of the rusted area With calibration coefficient The mapping relationship can be represented as ,in, and The attenuation parameters were obtained by conducting accelerated corrosion tests on standard steel reinforcement specimens in the laboratory, along with simultaneous tensile tests, collecting data on corrosion area and yield strength attenuation, and then fitting the data using the least squares method. The term represents the amount of material performance degradation caused by damage quantified by the percentage of corroded area. This indicates the remaining proportion of material properties relative to the uncorroded and intact state. This represents the lower limit of the calibration coefficient corresponding to the residual strength of the material properties, which is an empirical value greater than zero, such as 0.3.

[0044] This mapping relationship indicates that the material performance calibration coefficient decreases non-linearly with the expansion of the rust area, and there is a lower limit to the decrease. This mathematically quantifies the physical law that the more severe the rust, the lower the effective performance of the material.

[0045] The stress field characteristics are multiplied element by element with the material property calibration coefficients, and the result of the multiplication is used as the calibration stress field characteristics. This characteristic represents the spatial internal force distribution that conforms to the current true load-bearing state of the component after the material properties have been reduced and corrected according to the actual corrosion state of the component surface. It is used for subsequent structural safety analysis and anomaly diagnosis based on mechanical equilibrium equations.

[0046] S4. Based on historical frame data, construct the adjacency relationship between rebar nodes and generate topological connection features. Then, by verifying and enhancing the stability of the adjacency relationship, obtain enhanced topological features.

[0047] Considering that the connection relationship between steel bar nodes is the key to reflecting the stability of the precast beam yard and steel mesh structure, it is difficult to fully capture the dynamic changes and stability status of the connection relationship based solely on the current frame data. Furthermore, in real-world scenarios, there may be issues such as connection breaks and occlusions, leading to inaccurate topology connection features and affecting the completeness of structural anomaly detection. Therefore, it is necessary to construct topology connection features by combining historical frame data and perform stability enhancement processing.

[0048] In one specific embodiment, the method for obtaining the topological connection features is as follows: rebar node identification is performed on the three-dimensional point cloud data of historical frames, and a clustering algorithm such as DBSCAN algorithm is used, with a neighborhood radius of 2cm and a minimum number of points of 3, and the center of the point cloud cluster that meets the conditions is extracted as the coordinates of the rebar node.

[0049] Adjacency relationships are constructed based on the spatial Euclidean distance between nodes (e.g., distance less than 5cm) and the connection angle relationship (e.g., the angle between nodes is close to 90 degrees), forming a historical topology connection graph that represents the historical stable connection state.

[0050] The source point cloud is obtained by detecting the rebar node coordinates in the current frame. The node coordinates recorded in the historical topology connection map are used as input. The optimal spatial transformation parameters between the two are calculated by the iterative nearest point algorithm. Based on these parameters, the node coordinates of the current frame are rigidly transformed, and the node coordinates of the current frame are output to be aligned with the historical reference coordinate system. This eliminates inter-frame motion errors and generates topology connection data.

[0051] From the multi-channel tensor of material difference features, the mean color, texture roughness, and depth variance of the local region corresponding to each rebar node in the topological connection data are extracted to form the appearance state descriptor of the node region. Then, this descriptor is directly vector-concatenated with the topological connection data such as node coordinates and adjacency relationships to generate topological connection features containing node connection strength information.

[0052] This feature represents the spatial connection relationship between rebar nodes and their visual appearance quality, which is used for subsequent topological stability verification, connection anomaly detection, and provides spatial constraint relationship input for mechanical equilibrium analysis.

[0053] Further, please refer to Figure 3 As shown, the method for obtaining enhanced topological features is as follows: the topological connection data is feature-encoded through a graph neural network, and the original connection strength of the steel bar intersection is calculated. Specifically, the topological connection data, including node coordinates and adjacency relationships, is constructed into a graph structure and input into a graph convolutional network for message passing and aggregation.

[0054] The network outputs the embedding vector of each node. For each pair of adjacent nodes, the embedding vectors of their two endpoints are concatenated and regressed through a multilayer perceptron to output a scalar value. This value represents the theoretical connection strength of the connection in an ideal and intact state, and is called the original connection strength.

[0055] Calculate the difference matrix of the adjacency relationship between the rebar nodes in the current frame and the historical frames. The element of the difference matrix is ​​1, which indicates that the adjacency relationship is broken, and 0 indicates that the adjacency relationship is preserved.

[0056] The stability index is obtained by dividing the number of elements with a value of 1 in the statistical difference matrix by the total number of rebar nodes. This index represents the retention rate or the degree of drastic change in the connectivity of the topology between adjacent scanning cycles. A higher index indicates greater changes in the connectivity between nodes, a more unstable structure, or a large number of occlusion misjudgments; a lower index indicates more stable connectivity and more reliable structure and scanning.

[0057] When the stability index is below the set threshold, it indicates that the overall topology remains stable. In this case, the breakage of a few adjacent relationships is more likely to be due to real local damage such as cracks rather than global errors. If the stability index is above the threshold, it indicates that the overall structure has changed significantly, which may be due to construction stage or scanning errors. In this case, no compensation is made, and an overall instability warning signal is directly output.

[0058] The threshold value can be an empirical value between 0.05 and 0.15, and the implementer can set it according to the specific situation.

[0059] The compensation strength of the fracture connection is determined based on the crack depth information in the material difference features. Specifically, pixels whose crack depth values ​​in the material difference features are greater than a preset multiple, such as 1.5 times, of the average crack depth of the adjacent area are marked as crack areas.

[0060] Calculate the arithmetic mean of the depth values ​​of all pixels within the crack area to obtain the average depth value.

[0061] The compensation strength of the fractured connection is determined based on the average depth value and a segmented threshold rule. For example, the segmented threshold rule is as follows: when the average depth value is ≤2mm, the compensation strength = 1.0, indicating a minor crack that does not affect the connection; when 2mm < average depth value ≤5mm, the compensation strength = 0.7, indicating a moderate crack that partially weakens the connection; when the average depth value >5mm, the compensation strength = 0.3, indicating a severe crack that significantly weakens the connection.

[0062] The specific threshold and compensation coefficient of this rule can be determined with reference to relevant industry norms and standards. This invention does not further limit the exemplary values.

[0063] The compensation strength of the broken connection is multiplied by the original connection strength, and the result is limited to not less than 0.1, to obtain the reconstructed adjacency relationship.

[0064] By concatenating the reconstructed adjacency relationships with the topological connectivity features along the channel dimension, an enhanced topological feature is obtained. This feature represents the enhanced connectivity of the steel reinforcement network, which integrates historical stability and local damage compensation. It is used to provide spatial constraints that better reflect the actual damage state in subsequent mechanical analysis.

[0065] S5. Integrate the calibration stress field characteristics and enhanced topology characteristics, perform constraint calibration based on the mechanical equilibrium equation, and determine the structural anomaly parameters; if the structural anomaly parameters exceed the preset threshold, an anomaly prompt signal is output.

[0066] Considering that the accuracy of structural anomaly parameters directly depends on the rationality of feature fusion and the effectiveness of constraints, calibrating stress field features reflects the stress state of components, and enhancing topological features reflects the connection stability of the structure. Together, they determine the structural health of the precast beam yard and steel mesh.

[0067] Furthermore, the stress and deformation of the structure must follow the mechanical equilibrium equation. If this constraint is ignored and abnormal parameters are calculated directly, the detection results will deviate from the actual situation. Therefore, it is necessary to integrate the two types of features and perform constraint calibration based on the mechanical equilibrium equation.

[0068] In one specific embodiment, the method for determining the structural anomaly parameters is as follows: the divergence of the calibration stress field characteristics is calculated to obtain the stress field divergence distribution. Specifically, the calibration stress field characteristics that characterize the spatial vector field, i.e., the three-dimensional tensor, are discretized. Then, at each internal node of the grid, the sum of its partial derivatives along the three coordinate axes is calculated using the central difference method, thereby obtaining the stress field divergence distribution describing the distribution of internal stress sources.

[0069] Read the pre-stored design parameters of the precast beam yard, such as design load values ​​and self-weight coefficients, and perform dimensionless processing to obtain a standardized design parameter vector. Perform Hadamard product operation on the enhanced topological features and the standardized design parameter vector. This operation means that the standardized theoretical load or force is distributed, transmitted and transformed into equivalent force acting on the internal nodes of the structure according to the spatial connection relationship of the enhanced steel reinforcement network, generating an external force density distribution.

[0070] The difference between the stress field divergence distribution and the external force density distribution is calculated based on the mechanical equilibrium equation. This calculation is used to verify whether the inferred internal stress distribution is in equilibrium with the theoretical external force, and the equilibrium residual distribution is obtained.

[0071] Calculating the L2 norm of the equilibrium residual distribution is equivalent to calculating the square root of the sum of squares of all values ​​in the distribution. This norm represents a comprehensive scalar index of the degree of mechanical imbalance within the overall analysis area, and the L2 norm is used as a structural anomaly parameter.

[0072] If the abnormal structural parameters exceed a preset threshold, an abnormality warning signal will be output. The preset threshold can be a value between 0.1 and 0.3, and the implementer can set it according to the specific circumstances.

[0073] The abnormality warning signal includes the determination result of whether the structural abnormality parameter exceeds the threshold and its specific value, and is output through a graphical interface or an audible and visual alarm device.

[0074] In summary, this invention first acquires three-dimensional point cloud and image data of the precast beam yard and steel mesh simultaneously using lidar and depth camera, and performs registration and filtering preprocessing; then, based on multi-scale neighborhood analysis, it extracts the local geometric features of the steel bar nodes and the global morphological features of the beam surface from the point cloud, while extracting material difference features from the image.

[0075] Then, the geometric features are dynamically integrated through a gating fusion mechanism and spliced ​​with a preset physical property vector to form an initial stress field feature. The material properties of this feature are then calibrated based on the rusted areas and their area ratios identified from the image. Simultaneously, the topological connection features of the steel reinforcement network are constructed and enhanced based on historical and current frame data.

[0076] Finally, the calibrated stress field characteristics are combined with the enhanced topological characteristics, and the L2 norm of the overall structural equilibrium residual is calculated based on the mechanical equilibrium equation, which is used as a structural anomaly parameter for safety determination.

[0077] This method realizes a complete closed loop from multi-source data acquisition and physical information fusion to mechanical principle verification, which improves the automation level, damage identification accuracy and robustness to complex working conditions of precast beam structure state analysis.

[0078] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0079] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0080] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0082] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for three-dimensional modeling analysis and processing based on a precast beam yard and a reinforcement mesh, characterized in that, The method comprises the following steps: Collecting three-dimensional point cloud data of the prefabricated beam field and the reinforcement mesh by a laser radar, and synchronously collecting corresponding image data by a depth camera; Based on the three-dimensional point cloud data, the local geometric features of the reinforcement nodes and the global morphological features of the beam surface are extracted by multi-scale neighborhood analysis, and the material difference features are extracted based on the image data; According to the type of the component, the weights of the local geometric features and the global morphological features are dynamically allocated, and the material difference features and the preset component physical property vector are fused to generate stress field features; The stress field features are calibrated based on the material properties to obtain calibrated stress field features; Based on the historical frame data, the adjacency relationship between the reinforcement nodes is constructed and the topological connection features are generated, and the enhanced topological features are obtained by stability verification and enhancement of the adjacency relationship; The calibrated stress field features and the enhanced topological features are fused, and the structure abnormal parameters are determined by constraint calibration according to the mechanical equilibrium equation; If the structure abnormal parameter exceeds the preset threshold, an abnormal prompt signal is outputted; The determination method of the structure abnormal parameter is as follows: the divergence of the calibrated stress field features is calculated to obtain the stress field divergence distribution; the prefabricated beam field design parameters stored in advance are read and dimensionless processing is performed to obtain a standardized design parameter vector; the enhanced topological features and the standardized design parameter vector are multiplied to generate an external force density distribution; the stress field divergence distribution and the external force density distribution are difference calculated according to the mechanical equilibrium equation to obtain a balance residual distribution; the L2 norm of the balance residual distribution is calculated, and the L2 norm is taken as the structure abnormal parameter.

2. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The method for extracting the local geometric features of the reinforcement nodes and the global morphological features of the beam surface by multi-scale neighborhood analysis comprises the following steps: According to the type of the component, the neighborhood scale is dynamically set, wherein a small-scale neighborhood is set for the reinforcement node area, and a large-scale neighborhood is set for the beam surface area; In the small-scale neighborhood, the local geometric features of the reinforcement nodes are extracted by a first-order neighborhood graph convolution operation, and the local geometric features include node curvature and normal vector distribution; In the large-scale neighborhood, the global morphological features of the beam surface are extracted by a third-order neighborhood graph convolution operation, and the global morphological features include surface continuity and curvature variation trend.

3. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The material difference features are multi-channel tensors containing surface color distribution, texture features and crack depth information.

4. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The method for dynamically allocating the weights of the local geometric features and the global morphological features according to the type of the component comprises the following steps: Based on the type of the component, a gating signal is generated, and the gating signal is used as a weight to perform weighted summation on the local geometric features and the global morphological features.

5. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The method for obtaining the stress field features comprises the following steps: Multi-modal feature fusion is performed on the weighted and fused local geometric features, global morphological features and material difference features to obtain a multi-modal feature vector; The multi-modal feature vector and the preset component physical property vector are spliced to form a stress field feature representing the stress state of the component; The component physical property vector includes the yield strength of the reinforcement and the elastic modulus of the concrete.

6. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The method for obtaining the calibrated stress field features comprises the following steps: The rust area is identified based on the material difference feature, and a color histogram feature is extracted from the material difference feature of the rust area, and a rust area proportion is calculated; A material attribute calibration coefficient is determined according to a preset mapping relationship table of the rust area proportion and the calibration coefficient; The stress field feature is multiplied by the material attribute calibration coefficient element by element, and the multiplication result is taken as a calibrated stress field feature.

7. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 1, wherein, The method for obtaining the topological connection feature is: Based on historical frame data, a steel bar node is identified, and a node adjacency relationship is constructed to form a historical topological connection graph; The steel bar node coordinate point cloud of the current frame is spatiotemporally aligned with the historical topological connection graph through an iterative closest point algorithm to generate topological connection data; The topological connection feature is generated by fusing the material difference feature and the topological connection data.

8. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 7, wherein, The method for obtaining the enhanced topological feature is: The topological connection data is feature-encoded through a graph neural network to calculate an original connection strength of a steel bar intersection point; A difference matrix of the adjacency relationship of the steel bar node in the current frame and the historical frame is calculated, wherein a difference matrix element is 1, indicating that the adjacency relationship is broken, and 0, indicating that the adjacency relationship is retained; The number of elements with a value of 1 in the difference matrix is counted, and the number is divided by the total number of steel bar nodes to obtain a stability index; When the stability index is lower than a set threshold, a compensation strength of a broken connection is determined based on crack depth information in the material difference feature; The compensation strength of the broken connection is multiplied by the original connection strength to obtain a reconstructed adjacency relationship after compensation; The reconstructed adjacency relationship after compensation is concatenated with the topological connection feature in the channel dimension to obtain an enhanced topological feature.

9. The precast beam yard and mesh based three-dimensional modeling analysis processing method according to claim 8, wherein, The compensation strength of the broken connection is determined based on the crack depth information in the material difference feature, specifically: Pixel points with a crack depth value greater than a preset multiple of the average crack depth of adjacent regions in the material difference feature are marked as crack regions; An arithmetic mean of the depth values of all pixel points in the crack region is calculated to obtain an average depth value; The compensation strength of the broken connection is determined based on a segmented threshold rule according to the average depth value.

Images