Mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame

Abstract

A method for reconstructing a mechanical model based on intelligent prediction of the internal information of a reinforced concrete frame belongs to the field of earthquake engineering. It solves the problem that the existing method for constructing a three-dimensional mechanical model of a reinforced concrete frame must rely entirely on known data information for construction, and the data information is difficult to obtain and the construction process is complicated. The present invention infers the internal structural information of a reinforced concrete frame based on a graph neural network. It can reconstruct the three-dimensional structural model and material usage of beams, slabs, and columns in a reinforced concrete frame when the contour, seismic fortification intensity, site category, floor height, and number of floors are known, thereby constructing a three-dimensional mechanical model. The present invention is mainly used for predicting the internal information of a reinforced concrete frame and reconstructing a three-dimensional mechanical model of the structure of its reinforced concrete frame.

Description

Technical Field

[0001] The present invention belongs to the field of earthquake engineering, and in particular relates to the ability to infer structural information inside a reinforced concrete frame when some readily available information is known. Background Art

[0002] With the construction of cities, people are paying more and more attention to the seismic performance of buildings. Reinforced concrete frames are an important type of building widely used in urban buildings. If a fast method for urban buildings can be proposed to obtain the internal information of the structure before an earthquake, analyze the seismic performance of the building, and reinforce buildings with poor seismic performance, it will be able to effectively protect people's lives and property.

[0003] The three-dimensional structural model suitable for seismic performance analysis of reinforced concrete frame structures consists of two basic parts, namely the three-dimensional geometric model and the material usage. At present, there is no method to reconstruct the internal structural geometric information and material information of the reinforced concrete frame when the easily available information such as the outline, floor height, and seismic fortification intensity is known. In terms of three-dimensional geometric model reconstruction, when collecting building information for urban agglomerations, remote sensing and aerial photography are usually used. Based on the data obtained by these means, the current technical means can usually only construct the external three-dimensional surface model of the building, but cannot realize the inference of the geometric dimensions of the internal components. If you want to realize the inference of the geometric dimensions of the internal components, you need to scan the reinforced concrete frame building layer by layer to realize the inference of the geometric dimensions of its internal components, so as to obtain the three-dimensional mechanical model of the reinforced concrete frame. The three-dimensional mechanical model is used for mechanical calculations, seismic performance evaluation, etc. However, the process of obtaining the three-dimensional mechanical model of the reinforced concrete frame in the above method is complicated, the required data type is comprehensive, and the data information is difficult to obtain. It must rely on known data information to construct the three-dimensional mechanical model of the reinforced concrete frame, and the construction structure is complex. Therefore, it is necessary to propose a new method to construct a three-dimensional mechanical model of reinforced concrete frame structures, so as to reconstruct the internal component size information and component material usage under the condition of known and readily available information, predict the internal structure information of the reinforced concrete frame, and thus obtain a three-dimensional mechanical model of the reinforced concrete frame. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that the existing methods for constructing a three-dimensional mechanical model of a reinforced concrete frame must rely entirely on known data information for construction, and the data information is difficult to obtain and the construction process is complicated. The present invention provides a three-dimensional mechanical model reconstruction method for predicting the internal structural information of a reinforced concrete frame based on known (easy to obtain) information, and in specific applications, a physical mechanism is added to the graph neural network, so that the dynamic characteristics of the reconstructed three-dimensional mechanical model can achieve better accuracy than the real three-dimensional mechanical model.

[0005] A mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame, the method comprises the following steps:

[0006] Training phase:

[0007] According to the actual drawings of the reinforced concrete frame of each scene, the geometric plane outline information diagram, seismic fortification intensity, seismic fortification category, site category and column size at each position of any layer of the reinforced concrete frame of each scene in the top view are obtained in proportion; each layer structure in the reinforced concrete frame is composed of beams, floor slabs and columns, and the layout positions of beams, floor slabs and columns of each layer are exactly the same;

[0008] According to the beam-column positioning method, the geometric plane contour information map is preprocessed to obtain a beam-column plane topological map with node features and edge features; wherein the node feature is the position information of the column, the edge feature is the length information of the beam, and the beam-column plane topological map is a topological map formed by taking the column as the node and the connecting line between the nodes as the beam;

[0009] The plane topology diagrams of each beam and column, and their corresponding seismic fortification intensity, seismic fortification category, and site category are used as input data, and the dimensions of columns at each position in the scene corresponding to the plane topology diagram of the beam and column are used as output data to train the graph neural network to obtain a trained graph neural network;

[0010] Reasoning phase:

[0011] The geometric plane contour information map of the reinforced concrete frame of the scene to be tested in the top-down direction is obtained by aerial photography in equal proportion, and the geometric plane contour information map is preprocessed according to the beam-column positioning method to obtain the beam-column plane topology map;

[0012] The obtained beam-column plane topology map and the seismic fortification intensity, seismic fortification category and site category of the area where the scene to be tested is located are input into the trained graph neural network to predict the column size at each location;

[0013] The predicted column size at each position, known floor thickness, known longitudinal section size of beam, beam-column plane topology, known storey height and number of storeys are integrated by parametric modeling to obtain a three-dimensional geometric model of reinforced concrete frame.

[0014] Then, the three-dimensional geometric model of the reinforced concrete frame, the seismic fortification intensity of the area where the scene to be tested is located, the load corresponding to the scene to be tested, and the steel and concrete strength grades are calculated by parametric modeling to obtain the reinforcement of any layer of the reinforced concrete frame.

[0015] Finite element analysis is performed on the three-dimensional geometric model of the reinforced concrete frame, the reinforcement condition of any layer of the structure in the reinforced concrete frame, the seismic fortification intensity of the area where the test scenario is located, the load corresponding to the test scenario, and the strength grade of steel bars and concrete to obtain the three-dimensional mechanical model of the reinforced concrete frame and complete the reconstruction of the three-dimensional mechanical model of the reinforced concrete frame.

[0016] Preferably, the implementation method of preprocessing the geometric plane profile information graph according to the beam-column positioning method includes:

[0017] S1. Extend each boundary line of the original geometric contour figure in the geometric plane contour information graph, and the extended line does not exceed the range of the original geometric contour figure, to obtain a geometric contour figure after the extended boundary line;

[0018] S2. Generate nodes by arranging axes: create horizontal axes and vertical axes at equal intervals for the geometric outline figure after extending the boundary line, and obtain node position information formed by the intersection of lines; wherein the node position information is used as the column position information;

[0019] S3, node elimination: eliminate the nodes outside the area where the original geometric contour graphic is located;

[0020] S4. Generate beam and beam-column plane topology diagram: connect the nodes in the geometric outline figure after removing the nodes according to the beam generation principle to obtain the beam-column plane topology diagram; the connection line between the nodes is used as the beam, and the length of the connection line is used as the length information of the beam.

[0021] Preferably, the beam generation principle includes a transverse beam generation principle and a longitudinal beam generation principle within a plane;

[0022] The principle of generating a transverse beam is: when the longitudinal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a transverse beam;

[0023] The principle of generating a longitudinal beam is: when the horizontal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a longitudinal beam.

[0024] Preferably, the graph neural network comprises three parts, namely, a graph representation module, a feature fusion module and a regression module, which are arranged in sequence. The graph representation module is composed of three cascaded graphSAGE layers, the feature fusion module is composed of one feature fusion layer, and the regression module includes three layers of residual blocks with an attention mechanism added.

[0025] The beam-column plane topology diagram, seismic fortification intensity, seismic fortification category and site category are used as the input data of the graph neural network;

[0026] The first graphSAGE layer in the graph representation module is used to extract features from nodes and edges in the received beam-column plane topology graph, and the obtained first feature graph is sent to the second graphSAGE layer;

[0027] The second graphSAGE layer is used to extract features from nodes and edges in the received first feature graph, and the obtained second feature graph is sent to the third graphSAGE layer;

[0028] The third graphSAGE layer is used to extract features from nodes and edges in the received second feature graph, and the obtained third feature graph is sent to the feature fusion layer;

[0029] The feature fusion layer is used to add the features of the adjacent edges of each node in the third feature graph to obtain the adjacent edge features of the node; then the node features, the adjacent edge features of the node, the seismic fortification intensity, the seismic fortification category and the site category are concatenated in the same direction to obtain the fused feature vector of the node;

[0030] The first layer adds a residual block of the attention mechanism, which is used to receive the fused feature vectors of each node in turn, and obtain the first new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the received fused feature vectors of each node;

[0031] The second layer adds a residual block of the attention mechanism to obtain the second new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the first new fused feature vector of each node received;

[0032] The third layer adds a residual block of the attention mechanism to obtain the third new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the second new fused feature vector of each node received;

[0033] The third new fused feature vector of each node is a vector of size 2×1, and the elements in the first row and first column represent the length of the rectangular structure formed by the transverse section of the column corresponding to the node, and the elements in the first row and first column represent the width of the rectangular structure formed by the transverse section of the column corresponding to the node;

[0034] The third new fused feature vectors of all nodes are used as the output data of the graph neural network.

[0035] Preferably, the scenes include residences, hospitals, schools, office buildings and libraries.

[0036] Preferably, the site category is an indicator type that reflects the comprehensive amplification effect of different site conditions on bedrock earthquake vibrations.

[0037] Preferably, the column size is the length and width of a rectangular structure formed by the transverse section of the column, and the longitudinal section of the beam is rectangular.

[0038] Preferably, the parametric modeling approach is implemented using architectural structure design software.

[0039] Preferably, the reinforcement calculation is implemented by building structure design software.

[0040] Preferably, the geometric contour figures in the geometric plane contour information graph are L-shaped, concave, T-shaped, square and U-shaped.

[0041] Advantages of the present invention:

[0042] This method can be used to achieve component-level reconstruction of reinforced concrete frame structures. Physical mechanism-enhanced graph neural network-based inference of the internal structural information of reinforced concrete frames can reconstruct the three-dimensional structural model and material usage of the beams, slabs, and columns of reinforced concrete frames when the outline, seismic fortification intensity, site category, floor height, and number of floors are known, and a three-dimensional mechanical model can be constructed.

[0043] The reconstruction results can provide a reasonable basis for the static and dynamic analysis of reinforced concrete frame buildings under limited information background. It has been verified that the three-dimensional mechanical model reconstructed by this method has a similar period and dynamic response to the real three-dimensional mechanical model of the original reinforced concrete frame structure, and can reflect the mechanical and dynamic characteristics of the original structure to a certain extent, providing a relatively reliable basis for the mechanical analysis of reinforced concrete frame structures under limited information background.

[0044] The present invention is conducive to improving the seismic performance of urban building complexes. Based on the fast reconstruction speed of the reinforced concrete frame structure information under limited information conditions, the reconstruction of a single reinforced concrete frame structure can be completed within 20 seconds. Therefore, the present invention method is suitable for the reconstruction of structural information of urban building complexes with a large number of buildings. The three-dimensional mechanical model reconstructed by the present invention method can serve the seismic performance evaluation of buildings and provide a basis for reinforcement for decision-making departments, which will help improve the seismic performance of urban building complexes. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1It is a schematic diagram of the principle of the mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame in the reasoning stage of the present invention;

[0046] Figure 2 This is a schematic diagram of the structure of the graph neural network;

[0047] Figure 3 It is the shape diagram of geometric contour figures in the geometric plane contour information diagram and the key control range of various geometric contour figures;

[0048] Figure 4 It is a schematic diagram of the three-dimensional geometric model of the reinforced concrete frame;

[0049] Figure 5 It is a schematic diagram of the three-dimensional mechanical model of the reinforced concrete frame;

[0050] Figure 6 This is the EI Centro ground motion acceleration time history curve;

[0051] Figures 7a to 7d Comparison of displacement time histories between the real 3D mechanical model and the reconstructed 3D mechanical model of the reinforced concrete frames of the four selected buildings under the amplitude-modulated EI Centro earthquake motion;

[0052] Figure 7a This is a displacement time history comparison diagram of the real 3D mechanical model and the reconstructed 3D mechanical model corresponding to the reinforced concrete frame of the first building;

[0053] Figure 7b This is a displacement time history comparison diagram of the real 3D mechanical model and the reconstructed 3D mechanical model corresponding to the reinforced concrete frame of the second building;

[0054] Figure 7c This is a displacement time history comparison diagram of the real 3D mechanical model and the reconstructed 3D mechanical model corresponding to the reinforced concrete frame of the third building;

[0055] Figure 7d This is a comparison chart of the displacement time history between the real 3D mechanical model and the reconstructed 3D mechanical model corresponding to the reinforced concrete frame of the fourth building. DETAILED DESCRIPTION

[0056] 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 only 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 are within the scope of protection of the present invention.

[0057] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0058] The reinforced concrete frame structure is mainly composed of plates, beams, and columns. The columns are important lateral force resisting components. The prediction accuracy of the column size in the frame structure is directly related to the mechanical properties of the structure. Therefore, firstly based on the known information obtained, the topological relationship of the building plane beams and columns can be further constructed, and the organized topological relationship can be better processed by the graph neural network. In the present invention, a graph neural network for reconstructing reinforced concrete frame structures is constructed for reconstructing the column size. The network input data is a beam-column plane topological map, where the edge feature of the topological map is the beam length, and the node feature is the position coordinate of the corresponding column. The organized topological information with known features is input into the graph neural network model, and the graph neural network can predict the size information of each column.

[0059] Embodiment 1:

[0060] See also Figure 1 The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame described in this embodiment is described, and the method comprises the following steps:

[0061] (I) Training phase:

[0062] A1. Data collection and acquisition of geometric plane profile information diagram:

[0063] According to the actual drawings of the reinforced concrete frame of each scene, the geometric plane outline information diagram, seismic fortification intensity, seismic fortification category, site category and column size at each position of any layer of the reinforced concrete frame of each scene in the top view are obtained in proportion; each layer structure in the reinforced concrete frame is composed of beams, floor slabs and columns, and the layout positions of beams, floor slabs and columns of each layer are exactly the same;

[0064] Among them, the scenes include residences, hospitals, schools, office buildings and libraries. In specific applications, the proportion of residences can be 47.2%, office buildings account for 16.4%, schools account for 25.5%, hospitals account for 9.1%, and libraries account for 1.8%. Collect drawings for subsequent research.

[0065] A2. Preprocess the geometric plane profile information map to obtain the beam-column plane topology map:

[0066] According to the beam-column positioning method, the geometric plane contour information map is preprocessed to obtain a beam-column plane topological map with node features and edge features; wherein the node feature is the position information of the column, the edge feature is the length information of the beam, and the beam-column plane topological map is a topological map formed by taking the column as the node and the connecting line between the nodes as the beam;

[0067] A3. Training the graph neural network:

[0068] The plane topology diagrams of each beam and column, and their corresponding seismic fortification intensity, seismic fortification category, and site category are used as input data, and the dimensions of columns at each position in the scene corresponding to the plane topology diagram of the beam and column are used as output data to train the graph neural network to obtain a trained graph neural network;

[0069] (II) Reasoning stage:

[0070] B1. Obtain the geometric plane profile information map and pre-process it to obtain the beam-column plane topology map:

[0071] The geometric plane contour information map of the reinforced concrete frame of the scene to be tested in the top-down direction is obtained by aerial photography in equal proportion, and the geometric plane contour information map is preprocessed according to the beam-column positioning method to obtain the beam-column plane topology map;

[0072] B2. Predict the column size at each position:

[0073] The obtained beam-column plane topology map and the seismic fortification intensity, seismic fortification category and site category of the area where the scene to be tested is located are input into the trained graph neural network to predict the column size at each location;

[0074] Among them, the site category is an indicator type that reflects the comprehensive amplification effect of different site conditions on bedrock earthquake vibrations. B3. Obtain the three-dimensional geometric model of the reinforced concrete frame:

[0075] The predicted column size at each position, known floor thickness, known longitudinal section size of beam, beam-column plane topology, known storey height and number of storeys are integrated by parametric modeling to obtain a three-dimensional geometric model of reinforced concrete frame.

[0076] In specific applications, parametric modeling is implemented using architectural structure design software. The column size is the length and width of the rectangular structure formed by the column transverse section. The floor thickness is generally 120mm. The longitudinal section of the beam is rectangular, and the width × height of the longitudinal section of the beam is generally 300mm × 600mm.

[0077] B4. Reinforcement calculation:

[0078] Then, the three-dimensional geometric model of the reinforced concrete frame, the seismic fortification intensity of the area where the scene to be tested is located, the load corresponding to the scene to be tested, and the steel and concrete strength grades are calculated by parametric modeling to obtain the reinforcement of any layer of the reinforced concrete frame.

[0079] In specific applications, the reinforcement conditions are determined through building structure design software.

[0080] B4. Reconstruction of three-dimensional mechanical model:

[0081] Finite element analysis is performed on the three-dimensional geometric model of the reinforced concrete frame, the reinforcement of any layer of the structure in the reinforced concrete frame, the seismic fortification intensity of the area where the scene to be tested is located, the load corresponding to the scene to be tested, and the strength grade of steel bars and concrete to obtain the three-dimensional mechanical model of the reinforced concrete frame and complete the reconstruction of the three-dimensional mechanical model of the reinforced concrete frame. In specific applications, finite element analysis can be achieved through finite element analysis software.

[0082] In this embodiment, the reinforced concrete frame is regarded as a one-story or multi-story reinforced concrete structure building with a single standard layer, so that the layout positions of the beams, floor slabs and columns of each layer are exactly the same; and when the reinforced concrete frame building is a one-story or multi-story reinforced concrete structure with multiple standard layers, the layout positions of the beams, floor slabs and columns of each layer in the one or more multi-story reinforced concrete structure corresponding to each standard layer are exactly the same; therefore, the corresponding three-dimensional mechanical model of each standard layer can be reconstructed according to this embodiment, and finally the corresponding multiple three-dimensional mechanical models corresponding to all multiple standard layers are superimposed.

[0083] For specific applications, see Figure 3 , the geometric contour figures in the geometric plane contour information diagram are L-shaped, concave, T-shaped, square and U-shaped. The frame structure has various body types. The beam and column determination method under a single rule will lead to unreasonable beam and column classification in certain ranges. In order to make the beam and column position arrangement result more reasonable and make the present invention applicable to buildings of different shapes, key control ranges are set for different body types, and additional rules are set within the key control ranges for inspection and adjustment. Figure 1 , the implementation methods of preprocessing the geometric plane profile information map according to the beam-column positioning method include:

[0084] S1. Extend each boundary line of the original geometric contour figure in the geometric plane contour information graph, and the extended line does not exceed the range of the original geometric contour figure, to obtain a geometric contour figure after the extended boundary line;

[0085] S2. Generate nodes by arranging axes: create horizontal axes and vertical axes at equal intervals for the geometric outline figure after extending the boundary line, and obtain node position information formed by the intersection of lines; wherein the node position information is used as the column position information;

[0086] S3, node elimination: eliminate the nodes outside the area where the original geometric contour graphic is located;

[0087] S4. Generate beam and beam-column plane topology diagram: connect the nodes in the geometric outline figure after removing the nodes according to the beam generation principle to obtain the beam-column plane topology diagram; the connection line between the nodes is used as the beam, and the length of the connection line is used as the length information of the beam.

[0088] Specifically, the beam generation principle includes the transverse beam generation principle and the longitudinal beam generation principle in the plane;

[0089] The principle of generating a transverse beam is: when the longitudinal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a transverse beam;

[0090] The principle of generating a longitudinal beam is: when the horizontal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a longitudinal beam.

[0091] See also Figure 2 ,The graph neural network consists of three parts: graph representation module, feature fusion module and regression module, which are arranged in sequence. The graph representation module is composed of three cascaded graphSAGE layers, the feature fusion module is composed of one feature fusion layer, and the regression module includes three layers of residual blocks with attention mechanism added;

[0092] The beam-column plane topology diagram, seismic fortification intensity, seismic fortification category and site category are used as the input data of the graph neural network;

[0093] The first graphSAGE layer in the graph representation module is used to extract features from nodes and edges in the received beam-column plane topology graph, and the obtained first feature graph is sent to the second graphSAGE layer;

[0094] The second graphSAGE layer is used to extract features from nodes and edges in the received first feature graph, and the obtained second feature graph is sent to the third graphSAGE layer;

[0095] The third graphSAGE layer is used to extract features from nodes and edges in the received second feature graph, and the obtained third feature graph is sent to the feature fusion layer;

[0096] The feature fusion layer is used to add the features of the adjacent edges of each node in the third feature graph to obtain the adjacent edge features of the node; then the node features, the adjacent edge features of the node, the seismic fortification intensity, the seismic fortification category and the site category are concatenated in the same direction to obtain the fused feature vector of the node;

[0097] The first layer adds a residual block of the attention mechanism, which is used to receive the fused feature vectors of each node in turn, and obtain the first new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the received fused feature vectors of each node;

[0098] The second layer adds a residual block of the attention mechanism to obtain the second new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the first new fused feature vector of each node received;

[0099] The third layer adds a residual block of the attention mechanism to obtain the third new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the second new fused feature vector of each node received;

[0100] The third new fused feature vector of each node is a vector of size 2×1, and the elements in the first row and first column represent the length of the rectangular structure formed by the transverse section of the column corresponding to the node, and the elements in the first row and first column represent the width of the rectangular structure formed by the transverse section of the column corresponding to the node;

[0101] The third new fused feature vectors of all nodes are used as the output data of the graph neural network.

[0102] The above figure shows that the module uses graphSAGE for message passing and aggregation. The neural network selected in the regression module is the single-channel residual network ResNet1D-Attention with an attention mechanism. The ResNet1D-Attention regression module consists of three layers of residual blocks with an attention mechanism.

[0103] The graph neural network of the present invention is referred to as RC-FSR GNN. After the physical mechanism begins to intervene in the training process of adding the physical mechanism, the node features generated by RC-FSR GNN will be added to the information of the topological graph. The topological graph with node features combined with other information will continue to be input into the sub-model of the physical mechanism. The trained sub-model will output the predicted value of the cycle, which will be compared with the actual label, and the generated physical mechanism loss L mse-PHY And the mean square error loss L of node prediction during training mse The sum is added in a certain ratio, and the loss L is back-propagated to update the model parameters. The loss function with the physical mechanism added is calculated as follows:

[0104] L=L mse +λL mse-PHY

[0105] The loss function consists of two parts. The first part is the conventional mean square error (MSE) loss L. mse , the second part is the physical mechanism loss L mse-PHY , different weights are assigned to the two losses and added together. The MSE loss weight of this part is determined to be 1, and the physical mechanism loss weight is set to λ. Through optimization, the value of λ is set to 8.

[0106] Through the analysis of the collected drawings, it can be found that when the seismic fortification intensity is within the range of 6 degrees (0.05g) to 7 degrees (0.15g), the concrete strength used in most buildings is C30. When the seismic fortification intensity is 8 degrees (0.20g), the concrete strength is concentrated in C35 and C40, and the buildings with concrete strength above C40 are all concentrated above 6 floors, and the strength of C35 concrete is below 6 floors. In the collected drawings, the representative steel bar strengths used are HPB300 and HRB400, among which HPB300 is used for distribution bars and stirrups, and HRB400 is used for force bars. Based on the above conclusions, the material usage of concrete and steel bars for RC frame structure buildings under different conditions is formulated. Among them, RC frame structure is reinforced concrete structure, and its English name is: reinforced concrete structure, referred to as RC.

[0107] Table 1 Selection rules for steel bars and concrete

[0108]

[0109] In specific applications, with the help of building structure design software, the reinforcement conditions of the structure can be generated according to the seismic fortification intensity of the area where the scene to be tested is located, the load corresponding to the scene to be tested, the strength grade of steel bars and concrete, and the geometric model of the building.

[0110] Verification test:

[0111] In order to verify the accuracy of the 3D mechanical model reconstructed by this method, 4 independent buildings were selected from the test set for verification experiments. Table 2 shows the comparison of the first-order period of the 3D mechanical model reconstructed by these 4 buildings and the real 3D mechanical model. The mean relative error MRE of the first-order period of the 3D mechanical model reconstructed by these 4 buildings and the real 3D mechanical model is 5.23%, which is obtained from the mean of the relative error in Table 2. The smaller the mean relative error MRE, the closer the structure of the reconstructed 3D mechanical model is to the dynamic characteristics of the real 3D mechanical model.

[0112] The relative error mean MRE value between the first-order period of the approximate model constructed by the traditional multi-degree-of-freedom simplified approximation method and the first-order period of the real structure is more than 20%. The three-dimensional mechanical model reconstructed by the reconstruction method proposed in the present invention can better describe the dynamic characteristics of the real model compared to the approximate model constructed by the traditional multi-degree-of-freedom simplified approximation method. The relative error mean MRE of the present invention is 5.23%, which is much smaller than the relative error mean 20% corresponding to the traditional multi-degree-of-freedom simplified approximation method. The three-dimensional mechanical model reconstructed by this method can better reflect the dynamic characteristics of the real mechanical model.

[0113] Table 2 Comparison of the periods between the reconstructed 3D mechanical model and the real 3D mechanical model

[0114]

[0115] In addition, the present invention also compares the seismic response of the real 3D mechanical model and the reconstructed 3D mechanical model under the same seismic motion, see Table 3. The selected seismic motion is the EI Centro seismic motion, such as Figure 6 As shown, the earthquake acceleration amplitude is adjusted to 300 mm / s 2 , ground motion is a given load.

[0116] Table 3 Comparison of seismic responses of the real 3D mechanical model and the reconstructed 3D mechanical model of the building

[0117]

[0118]

[0119] Table 3 and Figures 7a to 7d The comparison of the seismic responses of the four reinforced concrete frame buildings calculated using this method shows that the three-dimensional mechanical model reconstructed by the method of the present invention and the real three-dimensional mechanical model created using drawings respond similarly under earthquake action. The reconstructed three-dimensional mechanical model can better describe the reinforced concrete frame building with real structure, providing a reference for further seismic response analysis. Through the above comparative study, under the background of limited information, the three-dimensional mechanical model reconstructed by the method of the present invention has better accuracy in mechanical and dynamic characteristics than the three-dimensional mechanical model created using drawings with known information.

[0120] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame, characterized in that: The method comprises the following steps: Training phase: According to the actual drawings of the reinforced concrete frame of each scene, the geometric plane outline information diagram, seismic fortification intensity, seismic fortification category, site category and column size at each position of any layer of the reinforced concrete frame of each scene in the top view are obtained in proportion; each layer structure in the reinforced concrete frame is composed of beams, floor slabs and columns, and the layout positions of beams, floor slabs and columns of each layer are exactly the same; According to the beam-column positioning method, the geometric plane contour information map is preprocessed to obtain a beam-column plane topological map with node features and edge features; wherein the node feature is the position information of the column, the edge feature is the length information of the beam, and the beam-column plane topological map is a topological map formed by taking the column as the node and the connecting line between the nodes as the beam; The plane topology diagrams of each beam and column, and their corresponding seismic fortification intensity, seismic fortification category, and site category are used as input data, and the dimensions of columns at each position in the scene corresponding to the plane topology diagram of the beam and column are used as output data to train the graph neural network to obtain a trained graph neural network; Reasoning stage: The geometric plane contour information map of the reinforced concrete frame of the scene to be tested in the top-down direction is obtained by aerial photography in equal proportion, and the geometric plane contour information map is preprocessed according to the beam-column positioning method to obtain the beam-column plane topology map; The obtained beam-column plane topology map and the seismic fortification intensity, seismic fortification category and site category of the area where the scene to be tested is located are input into the trained graph neural network to predict the column size at each location; The predicted column size at each position, known floor thickness, known longitudinal section size of beam, beam-column plane topology, known storey height and number of storeys are integrated by parametric modeling to obtain a three-dimensional geometric model of reinforced concrete frame. Then, the three-dimensional geometric model of the reinforced concrete frame, the seismic fortification intensity of the area where the scene to be tested is located, the load corresponding to the scene to be tested, and the steel and concrete strength grades are calculated by parametric modeling to obtain the reinforcement of any layer of the reinforced concrete frame. Finite element analysis is performed on the three-dimensional geometric model of the reinforced concrete frame, the reinforcement of any layer of the structure in the reinforced concrete frame, the seismic fortification intensity of the area where the test scene is located, the load corresponding to the test scene, and the strength grade of steel bars and concrete to obtain the three-dimensional mechanical model of the reinforced concrete frame and complete the reconstruction of the three-dimensional mechanical model of the reinforced concrete frame; The graph neural network consists of three parts: a graph representation module, a feature fusion module, and a regression module. The graph representation module consists of three cascaded graphSAGE layers, the feature fusion module consists of one feature fusion layer, and the regression module includes three residual blocks with an attention mechanism added. The beam-column plane topology diagram, seismic fortification intensity, seismic fortification category and site category are used as the input data of the graph neural network; The first graphSAGE layer in the graph representation module is used to extract features from nodes and edges in the received beam-column plane topology graph, and the obtained first feature graph is sent to the second graphSAGE layer; The second graphSAGE layer is used to extract features from nodes and edges in the received first feature graph, and the obtained second feature graph is sent to the third graphSAGE layer; The third graphSAGE layer is used to extract features from the nodes and edges in the received second feature graph, and the obtained third feature graph is sent to the feature fusion layer; The feature fusion layer is used to add the features of the adjacent edges of each node in the third feature graph to obtain the adjacent edge features of the node; then the node features, the adjacent edge features of the node, the seismic fortification intensity, the seismic fortification category and the site category are concatenated in the same direction to obtain the fused feature vector of the node; The first layer adds a residual block of the attention mechanism, which is used to receive the fused feature vectors of each node in turn, and obtain the first new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the received fused feature vectors of each node; The second layer adds a residual block of the attention mechanism to obtain the second new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the first new fused feature vector of each node received; The third layer adds a residual block of the attention mechanism to obtain the third new fused feature vector of each node after performing convolution, pooling and cross-layer connection operations on the second new fused feature vector of each node received; The third new fused feature vector of each node is a vector of size 2×1, and the elements in the first row and first column represent the length of the rectangular structure formed by the transverse section of the column corresponding to the node, and the elements in the first row and first column represent the width of the rectangular structure formed by the transverse section of the column corresponding to the node; The third new fused feature vectors of all nodes are used as the output data of the graph neural network.

2. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: The implementation methods of preprocessing the geometric plane profile information map according to the beam-column positioning method include: S1. Extend each boundary line of the original geometric contour figure in the geometric plane contour information graph, and the extended line does not exceed the range of the original geometric contour figure, to obtain a geometric contour figure after the extended boundary line; S2. Generate nodes by arranging axes: create horizontal axes and vertical axes at equal intervals for the geometric outline figure after extending the boundary line, and obtain node position information formed by the intersection of lines; wherein the node position information is used as the column position information; S3, node elimination: eliminate the nodes outside the area where the original geometric contour figure is located; S4. Generate beam and beam-column plane topology diagram: connect the nodes in the geometric outline figure after removing the nodes according to the beam generation principle to obtain the beam-column plane topology diagram; the connection line between the nodes is used as the beam, and the length of the connection line is used as the length information of the beam.

3. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 2 is characterized in that: The beam generation principle includes the horizontal beam generation principle and the longitudinal beam generation principle in the plane; The principle of generating a transverse beam is: when the longitudinal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a transverse beam; The principle of generating a longitudinal beam is: when the horizontal coordinates of any two nodes are the same and there is no other node between the two nodes, the two nodes are connected to form a longitudinal beam.

4. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: Scenes include residences, hospitals, schools, office buildings, and libraries.

5. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: Site category is an indicator type that reflects the comprehensive amplification effect of different site conditions on bedrock earthquake vibrations.

6. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: The column size is the length and width of the rectangular structure formed by the transverse section of the column, and the longitudinal section of the beam is a rectangle.

7. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: Parametric modeling is achieved using architectural structure design software.

8. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: Reinforcement calculation is implemented through building structure design software.

9. The mechanical model reconstruction method based on intelligent prediction of internal information of reinforced concrete frame according to claim 1 is characterized in that: The geometric outline shapes in the geometric plane outline infographic are L-shape, concave shape, T-shape, square shape and U-shape.

Images