Reconstruction and visualization method of explosion damage in brick-concrete structures based on deep learning

Abstract

The present invention discloses a method for reconstruction and visualization of explosion damage of brick-concrete structures based on deep learning. Through numerical simulation, a damage data set of brick-concrete structures under explosions in multiple scenarios is obtained. Through data processing and conversion, a training database is constructed, and a two-stage brick-concrete structure damage prediction model based on a graph neural network is established. Error evaluation and parameter tuning are performed on the model to improve the prediction ability of the damage prediction model. According to the node displacement and node stress predicted by the two-stage brick-concrete structure damage prediction model, the damage effect of the brick-concrete structure under the impact of various types of explosions is visualized. The present invention can quickly obtain the damage results of shock waves under different equivalents and different explosion positions, and visualize the explosion damage effect, which greatly improves the calculation efficiency while ensuring the calculation accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of explosion damage assessment, and in particular to a deep learning-based method for reconstructing and visualizing explosion damage in brick-concrete structures. Background Art

[0002] The visualization technology of explosion damage effects is of great significance in improving the efficiency of emergency response to accidents, optimizing disaster relief training, and realizing real-time monitoring and early warning.

[0003] However, existing technologies face many challenges in visualizing explosion damage effects. The characteristics of explosion loads are complex and changeable. For example, different explosion sources and explosion distances will significantly affect the damage effect. Existing explosion damage effect visualization methods show insufficient real-time performance under different explosion conditions, and it is difficult to vividly and vividly display the three-dimensional damage performance of the structure. Taking brick-concrete structures as an example, they show complex nonlinear destructive behavior under explosion impact, which further increases the technical difficulty of visualizing related explosion damage effects.

[0004] At present, for the visualization method of explosion damage effects, as described in the Chinese invention patent "Explosion Disaster Visualization Display and Evaluation System Platform" (publication number: CN112380685A, publication date: 2023-07-21), the physical model of the explosion disaster scene is created based on multiple parameter information of the actual explosion disaster scene, and the target numerical method is used to generate the explosion disaster process simulation data, and the parameter change curve is generated based on the explosion disaster process simulation data, and the three-dimensional parameter change cloud map that changes with time is simulated to display the explosion disaster process. However, the three-dimensional damage performance of the structure has not yet been vividly and vividly displayed.

[0005] Similarly, in the Chinese invention patent "Visual simulation analysis method for explosion of dangerous goods in complex building structures in large spaces" (publication number: CN112464335A, publication date: 2022-10-11), a three-dimensional model of complex building structures in large spaces and the surrounding environment is established to analyze the damage and overpressure of the hazardous source room and surrounding rooms, so that the equivalent pressure of the shock wave is represented by a ball of different colors at different time nodes, and the damaged area of ​​the complex building structure in large spaces is represented by different color blocks. The above methods mainly rely on traditional numerical simulation or empirical formulas. When faced with complex stress conditions or multi-dimensional mechanical parameter changes, it often takes a lot of simulation time, and it is difficult to achieve the real-time requirements of visualization of explosion damage effects.

[0006] In summary, the problems existing in the prior art are: low efficiency and accuracy in visualizing explosion damage effects and limited adaptability. Summary of the invention

[0007] The purpose of the present invention is to provide a deep learning-based brick-concrete structure explosion damage reconstruction and visualization method, which has strong real-time performance and high computational efficiency, and can take into account the influence of shock waves on the damage effect under different equivalents and different explosion positions.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] In a first aspect, a method for reconstructing explosion damage of a brick-concrete structure based on deep learning is provided, comprising the following steps:

[0010] Based on graph neural network, a damage prediction model for brick-concrete structures is built;

[0011] The damage prediction model of brick-concrete structure is used to predict the damage effect of the brick-concrete structure under the explosion scenario, and the damage effect prediction results are obtained; wherein the damage prediction results include the node displacement and node stress of the brick-concrete structure;

[0012] According to the set node displacement threshold, the nodes of the brick-concrete structure are screened, and the remaining nodes after screening are used to reconstruct the brick-concrete structure.

[0013] As a further optimization scheme of the present invention, the steps for constructing a damage prediction model for brick-concrete structures are as follows:

[0014] Establish a brick-concrete structure damage database including training data set and validation data set;

[0015] Based on the graph neural network, a damage classification sub-model and a damage prediction sub-model are built. The input of the damage classification sub-model is the damage data of the brick-concrete structure, and the input of the damage prediction sub-model is the output of the damage classification sub-model and the damage data of the brick-concrete structure.

[0016] The damage classification sub-model and the damage prediction sub-model are trained and optimized using the training data set and the validation data set to obtain the brick-concrete structure damage prediction model.

[0017] As a further optimization solution of the present invention, the steps of constructing a brick-concrete structure damage database include:

[0018] Based on the explosion conditions, geometry and material information of the brick-concrete structure, the damage effect of the brick-concrete structure under the explosion impact is solved through Abaqus numerical simulation, and the damage data of the brick-concrete structure under different explosion conditions is obtained;

[0019] According to the explosion conditions and brick-concrete structures involved in different scenarios, multi-source data fusion is performed on the damage data of brick-concrete structures under different explosion conditions to obtain the multi-source explosion scene fusion vector under the corresponding scenario, and all scenarios and corresponding brick-concrete structure response results are stacked. The explosion scene feature vector of multi-scenario multi-source information fusion is used as the input feature, and the brick-concrete structure response set is used as the output result to construct a set;

[0020] The set is divided to obtain a brick-concrete structure explosion damage database including a training data set and a validation data set.

[0021] As a further optimization scheme of the present invention, the steps of acquiring damage data of brick-concrete structures under different explosion conditions include:

[0022] Based on the explosion conditions, the geometric shape, size and material properties of the brick-concrete structure, the damage data of the brick-concrete structure is obtained through Abaqus numerical simulation;

[0023] Associating the damage data with corresponding critical explosion parameters, including explosion equivalent, explosion location, impact direction, and impact distance;

[0024] The geometry, dimensions, and material properties of the brick-concrete structure are coupled with damage data. The material properties include elastic modulus, yield strength, and fracture toughness.

[0025] The key explosion parameters associated with the damage data are coupled with the geometric shape, size, and material properties of the brick-concrete structure, and a one-dimensional semantic information vector is obtained through data pruning and normalization.

[0026] Converting the geometric data of the brick-concrete structure into a three-dimensional semantic information vector, wherein the geometric data includes geometric shape and size;

[0027] The one-dimensional semantic information vector and the three-dimensional semantic information vector are spliced ​​and integrated to obtain the explosion scene feature vector of multi-source information fusion.

[0028] As a further optimization scheme of the present invention, the geometric data of the brick-concrete structure is converted into a three-dimensional semantic information vector based on a geometric feature fusion network and a Transformer network; wherein the geometric feature fusion network includes an edge information extraction layer and a skip connection fusion layer;

[0029] In the edge information extraction layer, the geometric data of the brick-concrete structure is input, and the edge features of the brick-concrete structure are obtained based on the three-dimensional pooling convolution layer;

[0030] In the skip connection fusion layer, the geometric data of the brick-concrete structure is input, and based on the three-dimensional convolutional layer, the multi-level feature mapping of the brick-concrete structure is established in sequence by adopting the hierarchical skip connection method, and the skip connection fusion layer of the three-dimensional feature is obtained to obtain the local features of the brick-concrete structure;

[0031] In the Transformer network, the local features of the brick-concrete structure are used as the input of the Transformer network to obtain the global features of the brick-concrete structure, and are combined with the edge features and local features obtained by the geometric feature fusion network to obtain a three-dimensional semantic information vector.

[0032] As a further optimization scheme of the present invention,

[0033] The damage classification sub-model is a graph neural network composed of a graph encoder, a graph processor, and a graph decoder in series, and the loss function is set to Binary_crossentropy;

[0034] The damage prediction sub-model is a graph neural network composed of a graph encoder, a graph processor and a graph decoder connected in series, and the loss function is set to MeanSquaredError.

[0035] As a further optimization scheme of the present invention, the brick-concrete structure reconstruction includes the concrete structure reconstruction and the quick structure reconstruction:

[0036] According to the node coordinates of the remaining nodes, space is divided, the remaining nodes are allocated to voxels of fixed size, and a hexahedral topological structure is constructed based on the voxel data, and these topological structures constitute a new space grid;

[0037] The vertices of each hexahedron are numbered according to a fixed arrangement rule: the bottom vertices are numbered 0, 1, 2, 3 clockwise, and the top vertices are numbered 4, 5, 6, 7 clockwise;

[0038] Generate 12 edges of the hexahedron according to the set rules. Each edge is represented by a vertex pair consisting of two vertices. The vertices in the vertex pair are sorted in ascending order.

[0039] Use edge sets to store the edges of all hexahedrons and remove duplicate edges;

[0040] The faces of a single hexahedron are generated according to the geometric definition of the faces, and the duplicate faces are removed. Finally, all the faces of the brick-concrete structure are drawn to complete the reconstruction of the brick-concrete structure.

[0041] As a further optimization solution of the present invention, the new spatial grid generation step includes:

[0042] Discretize the node coordinates of the remaining nodes into a fixed-size voxel grid, where each voxel represents a small cubic unit in space;

[0043] Based on the sparse distribution of nodes, invalid voxels are eliminated and only valid voxels containing nodes are retained;

[0044] For each valid voxel, generate the 3D coordinates of its eight vertices:

[0045] ;

[0046] in, represents the node coordinates, Indicates the preset voxel size;

[0047] The three-dimensional coordinates of the vertices are used to define the hexahedral topological structure of each voxel to form a new spatial grid for the reconstruction of brick-concrete structures.

[0048] As a further optimization scheme of the present invention, the node coordinates of the remaining nodes are discretized into a voxel grid of a fixed size, including:

[0049] Traverse all remaining nodes and find their minimum and maximum coordinates and ;

[0050] Based on preset voxel size , calculate the 3D size of the voxel grid:

[0051]

[0052] in , , is the voxel grid in The number of voxels in the direction;

[0053] Create an empty three-dimensional array to store valid voxel information and initialize the voxel grid; wherein the valid voxel information includes the coordinates of the eight vertices of the corresponding hexahedron;

[0054] Based on node coordinates , calculate the voxel index corresponding to the node:

[0055]

[0056] Assign the node to the corresponding voxel index and record its voxel index number ;

[0057] Input the valid voxel information into the initialized voxel grid according to the index number and save the voxel result.

[0058] In a second aspect, a method for visualizing explosion damage of a brick-concrete structure is also provided, comprising:

[0059] Based on the reconstruction method described above, the brick-concrete structure under the explosion scene is reconstructed;

[0060] The node stress is embedded into the reconstructed brick-concrete structure, and the numerical difference of the node stress is used to render different stress characteristics, thereby obtaining the rendered visualization result.

[0061] Compared with the prior art, the present invention has the following significant advantages:

[0062] 1. Strong real-time performance: This invention combines the advantages of Python library functions in rendering visualization results and the ability of graph neural networks in gradually refining damage effects, thereby enhancing the real-time performance of visualization of damage effects under explosive loads;

[0063] 2. Good adaptability: The present invention takes into account the complex nonlinear destructive behavior under explosion impact and can adapt to the visualization of the destructive effect of shock waves on target structures under different equivalents and different explosion positions;

[0064] 3. High calculation efficiency: The present invention can quickly obtain the damage results of shock waves under different equivalents and different explosion positions, and visualize the explosion damage effects, which greatly improves the calculation efficiency while ensuring the calculation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is the main flow chart of the deep learning-based brick-concrete structure explosion damage reconstruction and visualization method of the present invention.

[0066] Figure 2 It is a flowchart of three-dimensional data processing.

[0067] Figure 3 It is a structural diagram of the double-stage brick-concrete structure damage prediction model. DETAILED DESCRIPTION

[0068] like Figure 1 As shown, the method for reconstructing and visualizing explosion damage of brick-concrete structures based on deep learning of the present invention comprises the following steps:

[0069] S10. Multi-source data fusion: Based on multi-scenario and multi-dimensional actual working conditions, the damage effect of brick-concrete structure under explosion impact is solved through Abaqus numerical simulation, and the damage data of brick-concrete structure under different explosion conditions is obtained. For one-dimensional semantic information, data processing is performed through normalization method; for three-dimensional spatial information, feature fusion is performed through the method of combining geometric feature fusion network with Transformer network.

[0070] The multi-source data fusion step comprises:

[0071] S11. Damage data collection: Based on the explosion conditions, geometry and material information, the damage effect of the brick-concrete structure under the explosion impact is solved through Abaqus numerical simulation to obtain damage data.

[0072] S12. Explosion condition data association: Associating the brick-concrete structure damage data with corresponding explosion key condition parameters, wherein the explosion key condition parameters include explosion equivalent, explosion position, impact direction, and impact distance.

[0073] S13. Integration of geometry and material information: The geometry, dimensions, and material properties (such as elastic modulus, yield strength, and fracture toughness) of the brick-concrete structure are coupled with damage data. The material properties include elastic modulus, yield strength, and fracture toughness.

[0074] Steps S12 and S13 complete the preliminary extraction of data, forming a corresponding relationship between input data such as brick-concrete geometry, material information, and explosion conditions and output data on brick-concrete structure damage, ensuring that the subsequent damage prediction model can perform damage prediction based on the characteristics of the brick-concrete structure and the explosion conditions.

[0075] S14, one-dimensional data processing: coupling the critical explosion operating condition parameters associated with the brick-concrete structure damage data with the material properties of the brick-concrete structure, and obtaining a one-dimensional semantic information vector through data pruning and normalization.

[0076] The step S14, one-dimensional data processing, comprises:

[0077] S141. Data pruning: pruning the data of each vector of material properties, explosion equivalent, explosion location, explosion impact load and damage degree, and integrating vectors of different dimensions into the same dimension;

[0078] S142, data normalization: After data pruning is completed, the data needs to be normalized to the minimum and maximum values. This method scales all features to the range of 0 and 1. The calculation formula can be expressed as:

[0079]

[0080] In the formula Represents the original data, Represents normalized data.

[0081] S15, 3D data processing: Construct a geometric feature fusion network, introduce an edge information extraction layer and a jump connection fusion layer to fuse the geometric features of the brick-concrete structure, and combine it with the Transformer network to process the 3D geometric data of the brick-concrete structure and realize the fusion of the geometric features of the brick-concrete structure, such as Figure 2 shown.

[0082] The three-dimensional data processing step comprises:

[0083] S151, edge information extraction layer application: input three-dimensional data of the brick-concrete structure, and obtain edge features of the brick-concrete structure based on the three-dimensional pooling convolution layer;

[0084] S152, application of skip connection fusion layer: input the three-dimensional data of the brick-concrete structure, based on the three-dimensional convolution layer, adopt the hierarchical skip connection method, sequentially establish the multi-level feature mapping of the brick-concrete structure, obtain the skip connection fusion layer of the three-dimensional feature, fuse the mapped shallow features with the deep features, and obtain the local features of the brick-concrete structure;

[0085] S153, Transformer network application: The local features of the brick-concrete structure are used as the input of the Transformer, and the global features are captured through the self-attention mechanism. They are combined with the edge features and local features extracted by the geometric feature fusion network to obtain a three-dimensional semantic information vector.

[0086] S16, data integration: splice and integrate the normalized one-dimensional semantic information vector and the processed three-dimensional semantic information vector to obtain the explosion scene feature vector of multi-source information fusion.

[0087] The explosion scene feature vector is a standardized damage data that integrates multiple information sources.

[0088] S20. Establishment of a brick-concrete structure damage database: The damage data of brick-concrete structures under different explosion conditions are fused according to a multi-source data fusion method to obtain a multi-source information fused explosion scene feature vector, store the data, and obtain a brick-concrete structure explosion damage database including a training data set and a verification data set.

[0089] The steps of establishing the brick-concrete structure damage database include:

[0090] S21. Dataset generation: For the explosion conditions and brick-concrete structures involved in different scenarios, each scenario is fused in turn through the multi-source data fusion method to obtain the multi-source explosion scenario fusion vector under the corresponding scenario, and all scenarios and the corresponding brick-concrete structure response results are stacked. The explosion scenario feature vector fused by multi-scenario and multi-source information is used as the input feature, and the brick-concrete structure response set is used as the output result.

[0091] S22, database generation: divide the above-mentioned feature vector set, divide all data sets into a ratio of 7:3 and enter them into the database, so as to obtain an explosion damage database including a training data set and a verification data set.

[0092] S30. Establishment of a two-stage brick-concrete structure damage prediction model: Based on a graph neural network composed of a graph encoder, a graph processor and a graph decoder connected in series, a damage classification sub-model and a damage prediction sub-model are built in sequence, and the sub-models are trained and optimized using the training data set and the validation data set to obtain a two-stage brick-concrete structure damage prediction model based on a graph neural network.

[0093] This step is based on the idea of ​​preliminary damage classification to refined damage prediction. With the graph encoder, graph processor and graph decoder as the basic framework, a damage classification sub-model and a damage prediction sub-model are built in sequence. The brick-concrete structure explosion damage training set and its verification set are used to train and optimize the sub-models, and a two-stage brick-concrete structure damage prediction model is established.

[0094] The steps of establishing the brick-concrete structure damage prediction model include:

[0095] S31. Construction of damage classification sub-model: Based on the graph neural network composed of a graph encoder, a graph processor and a graph decoder connected in series, a damage classification sub-model is constructed.

[0096] The damage classification sub-model construction step comprises:

[0097] S311, graph coding layer construction: a multi-layer fully connected neural network is used, and the ReLU activation function is introduced for nonlinear processing to construct the graph neural network coding layer to obtain the coding result of the input data .

[0098] S312, graph processing layer construction: In the graph processing layer, the Iterative processing is performed through m graph neural network (GN) blocks with the same structure to achieve in-depth analysis and feature extraction of data; each GN block iteration includes the update of two key information, namely edge information and point information, and finally the abstract features are obtained. .

[0099] S313, graph decoding layer construction: through multi-layer upsampling fully connected neural network, and finally introduce Sigmoid activation function to perform classification probability transformation to construct the graph neural network decoding layer.

[0100] S314. Loss function setting: The loss function of the damage classification model is set to Binary_crossentropy to adapt to the classification of multiple damage types.

[0101] S315. Establishment of damage classification prediction sub-model: By combining the graph coding layer, graph processing layer and graph decoding layer, as well as the Binary_crossentropy loss function, a damage classification sub-model based on graph neural network is established.

[0102] S32. Construction of damage prediction sub-model: Based on a graph neural network composed of a graph encoder, a graph processor and a graph decoder connected in series, a damage prediction sub-model is constructed.

[0103] The damage prediction sub-model construction step comprises:

[0104] S321, graph coding layer construction: a multi-layer fully connected neural network is used, and the ReLU activation function is introduced for nonlinear processing to construct the graph coding layer to obtain the encoding result of the input data ;

[0105] S322, graph processing layer construction: In the graph processing layer, the Iterative processing is performed through n graph neural network (GN) blocks with the same structure to achieve in-depth analysis and feature extraction of data; each GN block iteration includes the update of two key information, namely edge information and point information, and finally the abstract features are obtained. .

[0106] S323, graph decoding layer construction: Through a multi-layer upsampling fully connected neural network, the ReLU activation function is introduced for nonlinear transformation to construct a graph neural network decoding layer.

[0107] The main task of the graph decoding layer is to convert the highly abstract graph data output by the graph processing layer into Mapped back to the original space with actual physical meaning. The input of the graph decoding layer is the high-dimensional graph data of the last layer of the graph processing layer. The node information of the component unit can be restored through multi-layer fully connected network information to obtain the node information data with component unit displacement.

[0108] S324. Loss function setting: The loss function of the damage prediction model is set to MeanSquaredError to adapt to damage prediction.

[0109] S325. Establishment of damage prediction sub-model: By combining the graph encoding layer, the graph processing layer and the graph decoding layer, as well as the MeanSquaredError loss function, a damage prediction sub-model based on the graph neural network is established.

[0110] S33. Construction of damage prediction model: Combine the damage classification sub-model and the damage prediction sub-model in order, combine the output results of the damage classification sub-model and the damage data of the brick-concrete structure, and use them as input features of the damage prediction model, connect the two sub-models in series, and then construct a two-stage brick-concrete structure damage prediction model.

[0111] S34, model training and structural optimization: using the training data set, the two-stage brick-concrete structure damage prediction model is trained in a distributed subgraph manner, and the trained prediction model is evaluated and optimized using the validation data set to obtain a two-stage brick-concrete structure damage prediction model based on a graph neural network, such as Figure 3 shown.

[0112] The model training and structure tuning steps include:

[0113] S341, model pre-training: randomly generate model training parameters and pre-train the model in a data-supervised manner.

[0114] S342. Select training parameters: monitor the model pre-training status, such as computing memory usage, graphics processor memory usage, and model convergence performance information. When model training cannot be implemented normally, reselect model training parameters.

[0115] S343. Adjust the graph processing layer structure: Input the validation set into the prediction model after iterative training, check the mean square error loss (MSE) and mean absolute error (MAE) of the model output results, and determine whether both have converged and decreased.

[0116] S344. Repeat steps S341, S342 and S343 until the MSE and MAE converge and decrease.

[0117] In the present invention, the training parameters of the frame-type building explosion damage effect prediction model finally optimized are: the training batch size is 32, the loss function is the mean absolute error (MAE), and the training iteration step is 10,000 times; the processing layer uses a total of 8 GN blocks, and each GN block contains 128 nodes.

[0118] S40, damage effect visualization: The damage of the brick-concrete structure under the explosion scenario is predicted according to the brick-concrete structure damage prediction model to obtain the damage prediction result; wherein, the damage prediction result includes the node displacement and node stress of the brick-concrete structure. A displacement threshold is set, and when the node displacement is greater than the threshold, the node is removed. The three-dimensional brick-concrete structure is reconstructed using the remaining nodes. The reconstructed brick-concrete structure is rendered with the stress information of each remaining node to obtain the damage effect visualization result.

[0119] The damage effect visualization step comprises:

[0120] S41. Construction of prediction input data set: Obtain the brick-concrete structure, material type, and key conditions of on-site explosion under the required prediction scenario to obtain the input data set.

[0121] The steps for constructing the prediction input data set are specifically as follows:

[0122] According to the parameters such as brick-concrete structure, material type, explosion equivalent, explosion distance, etc. in the required prediction scenario, a multi-source information fusion data set is obtained through data preprocessing and feature extraction, and a prediction input data set is constructed through data standardization and data conversion.

[0123] S42. Brick-concrete structure damage prediction: Predict brick-concrete structure damage according to the above-mentioned brick-concrete structure damage prediction model based on graph neural network.

[0124] S43, node threshold setting: according to the collapse criterion of brick-concrete structure, the displacement threshold of the brick-concrete node is set, and when the predicted node displacement is greater than the displacement threshold, the node is removed;

[0125] S44, spatial grid generation: divide the space according to the spatial position information of the remaining nodes, assign the nodes to voxels of fixed size, and build a hexahedral topological structure based on the voxel data. These hexahedral topological structures will constitute a new spatial grid for the subsequent three-dimensional reconstruction of the brick-concrete structure.

[0126] The step of generating the spatial grid comprises:

[0127] S441, Node voxelization: Discretize the remaining node spatial position information into a fixed-size voxel grid, where each voxel represents a small cubic unit in space. Based on the sparse distribution of nodes, invalid voxels are eliminated, and only valid voxels containing nodes are retained.

[0128] The node voxelization step comprises:

[0129] S4411. Get node boundaries: traverse all remaining nodes and find their minimum and maximum coordinates and .

[0130] S4412, Voxel grid size calculation: manually set the voxel size according to needs , calculate the three-dimensional size of the voxel grid:

[0131]

[0132] in , , is the voxel grid in The number of voxels in the direction.

[0133] S4413, Voxel grid initialization: Create an empty three-dimensional array to store valid voxel information and initialize the voxel grid.

[0134] S4414, Voxel allocation: For spatial position information in the node , calculate the corresponding voxel index:

[0135]

[0136] Assign nodes to corresponding voxels and record their voxel index numbers .

[0137] S4415, voxel result saving: input the valid voxel carrying information into the initialized voxel grid according to the index number, and save the voxel result.

[0138] S442, Vertex Construction: For each valid voxel, generate the three-dimensional coordinates of its eight vertices and avoid duplicate records through the vertex mapping mechanism. and the node position index Calculate the actual space coordinates of each vertex. The 8 vertex coordinates of the voxel can be expressed by the following formula:

[0139] .

[0140] S443, Hexahedron construction: Use a set of vertex coordinates to define the hexahedral structure of each voxel to ensure the integrity of the topological structure, thereby constructing a hexahedral mesh corresponding to each remaining node.

[0141] S45, 3D reconstruction of brick-concrete structure: Based on the output hexahedral mesh vertex coordinate array and topological relationship matrix, the algorithm is compiled to construct the outer surface of the brick-concrete structure through complex geometric and topological operations;

[0142] The complex geometric and topological operation steps include:

[0143] S451. Hexahedron vertex drawing: The vertices in each hexahedron mesh vertex coordinate array have a fixed arrangement rule, numbered from 0 to 7, bottom surface vertices: 0, 1, 2, 3 (arranged clockwise); top surface vertices: 4, 5, 6, 7 (arranged clockwise).

[0144] S452, Hexahedron edge drawing: Each cube has 12 edges, each edge is connected by two vertices, bottom edge: (0,1), (1,3), (3,2), (2,0); top edge: (4,5), (5,7), (7,6), (6,4); inter-face connecting edge: (0,4), (1,5), (2,6), (3,7). Generate the 12 edges of the cube according to the above rules, sort the vertex pairs of each edge in ascending order, use a set to store all edges, if two cubes are adjacent, they will share some edges, and automatically remove duplicate edges.

[0145] S453, brick-concrete structure surface drawing: Generate the surface of a single cube according to the geometric definition of the surface. When there are multiple cubes, process the shared surfaces between adjacent cubes to remove duplicate surfaces, and finally draw all the surfaces of the brick-concrete structure in combination with the topological relationship matrix.

[0146] The brick-concrete structure surface drawing step comprises:

[0147] S4531, Single cube face generation: bottom face is composed of vertices The top surface is composed of vertices Composition; four sides: the front side is , rear side , the left side is , the right side is .

[0148] S4532. Face processing of multiple cubes: If two cubes have the same 4 vertices, they share the face and only keep a set of shared faces. Sort the vertex index of each face in ascending order, use a set to store all faces, and automatically remove duplicate faces.

[0149] S4533, Draw all cube faces: Extract the 8 vertex indexes of each cube, generate the 4 vertex indexes of each face according to the topological relationship matrix, sort the face vertices to ensure uniqueness. Store the faces in a set and remove duplicate faces. Finally, extract the faces from the set and organize them into the final result.

[0150] S46, rendering and visualization: embed the stress information of each remaining node into the three-dimensional reconstructed brick-concrete structure, and use the numerical difference of the information to render and represent different stress characteristics, so as to obtain a rendered visualization result.

[0151] Through the above steps, a method for visualizing the damage effect of explosion on brick-concrete structures is obtained. Through numerical simulation, a damage data set of brick-concrete structures under explosions in multiple scenarios is obtained. Through data processing and conversion, a training database is constructed, and a two-stage brick-concrete structure damage prediction model based on graph neural network is established. Error evaluation and parameter tuning of the model are performed to improve the prediction ability of the damage prediction model. According to the node displacement and node stress predicted by the two-stage brick-concrete structure damage prediction model, the damage effect of the brick-concrete structure under various types of explosion impacts can be visualized.

[0152] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes the above-mentioned deep learning-based brick-concrete structure explosion damage reconstruction and visualization method.

[0153] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned deep learning-based brick-concrete structure explosion damage reconstruction and visualization method.

[0154] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0155] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0156] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0158] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A deep learning-based reconstruction method for explosion damage in brick-concrete structures, characterized in that: The steps include: Based on graph neural network, a damage prediction model for brick-concrete structures is built; The damage prediction model of brick-concrete structure is used to predict the damage effect of the brick-concrete structure under the explosion scenario, and the damage effect prediction results are obtained; wherein the damage prediction results include the node displacement and node stress of the brick-concrete structure; According to the set node displacement threshold, the nodes of the brick-concrete structure are screened, and the remaining nodes after screening are used to reconstruct the brick-concrete structure; The steps for constructing the damage prediction model of brick-concrete structure are as follows: Establish a brick-concrete structure damage database including training data set and validation data set; Based on the graph neural network, a damage classification sub-model and a damage prediction sub-model are built. The input of the damage classification sub-model is the damage data of the brick-concrete structure, and the input of the damage prediction sub-model is the output of the damage classification sub-model and the damage data of the brick-concrete structure. The damage classification sub-model and the damage prediction sub-model are trained and optimized using the training data set and the validation data set to obtain the brick-concrete structure damage prediction model. The damage classification sub-model is a graph neural network composed of a graph encoder, a graph processor, and a graph decoder in series, and the loss function is set to Binary_crossentropy; The damage prediction sub-model is a graph neural network composed of a graph encoder, a graph processor and a graph decoder connected in series, and the loss function is set to MeanSquaredError.

2. The reconstruction method according to claim 1, characterized in that: The steps to construct the brick-concrete structure damage database include: Based on the explosion conditions, geometry and material information of the brick-concrete structure, the damage effect of the brick-concrete structure under the explosion impact is solved through Abaqus numerical simulation, and the damage data of the brick-concrete structure under different explosion conditions is obtained; According to the explosion conditions and brick-concrete structures involved in different scenarios, multi-source data fusion is performed on the damage data of brick-concrete structures under different explosion conditions to obtain the multi-source explosion scene fusion vector under the corresponding scenario, and all scenarios and corresponding brick-concrete structure response results are stacked. The explosion scene feature vector of multi-scenario multi-source information fusion is used as the input feature, and the brick-concrete structure response set is used as the output result to construct a set; The set is divided to obtain a brick-concrete structure explosion damage database including a training data set and a validation data set.

3. The reconstruction method according to claim 2, characterized in that: The steps for obtaining damage data of brick-concrete structures under different explosion conditions include: Based on the explosion conditions, the geometric shape, size and material properties of the brick-concrete structure, the damage data of the brick-concrete structure is obtained through Abaqus numerical simulation; Associating the damage data with corresponding critical explosion parameters, including explosion equivalent, explosion location, impact direction, and impact distance; The geometry, dimensions, and material properties of the brick-concrete structure are coupled with damage data. The material properties include elastic modulus, yield strength, and fracture toughness. The key explosion parameters associated with the damage data are coupled with the geometric shape, size, and material properties of the brick-concrete structure, and a one-dimensional semantic information vector is obtained through data pruning and normalization. Converting the geometric data of the brick-concrete structure into a three-dimensional semantic information vector, wherein the geometric data includes geometric shape and size; The one-dimensional semantic information vector and the three-dimensional semantic information vector are spliced ​​and integrated to obtain the explosion scene feature vector of multi-source information fusion.

4. The reconstruction method according to claim 3, characterized in that: Based on the geometric feature fusion network and the Transformer network, the geometric data of the brick-concrete structure is converted into a three-dimensional semantic information vector; wherein the geometric feature fusion network includes an edge information extraction layer and a skip connection fusion layer; In the edge information extraction layer, the geometric data of the brick-concrete structure is input, and the edge features of the brick-concrete structure are obtained based on the three-dimensional pooling convolution layer; In the skip connection fusion layer, the geometric data of the brick-concrete structure is input, and based on the three-dimensional convolutional layer, a hierarchical skip connection method is used to sequentially establish a multi-level feature map of the brick-concrete structure, and a skip connection fusion layer of three-dimensional features is obtained to obtain the local features of the brick-concrete structure; In the Transformer network, the local features of the brick-concrete structure are used as the input of the Transformer network to obtain the global features of the brick-concrete structure, and are combined with the edge features and local features obtained by the geometric feature fusion network to obtain a three-dimensional semantic information vector.

5. The reconstruction method according to claim 1, characterized in that: Brick-concrete structure reconstruction includes concrete structure reconstruction and quick structure reconstruction: According to the node coordinates of the remaining nodes, space is divided, the remaining nodes are allocated to voxels of fixed size, and a hexahedral topological structure is constructed based on the voxel data, and these topological structures constitute a new space grid; The vertices of each hexahedron are numbered according to a fixed arrangement rule: the bottom vertices are numbered 0, 1, 2, 3 clockwise, and the top vertices are numbered 4, 5, 6, 7 clockwise; Generate 12 edges of the hexahedron according to the set rules. Each edge is represented by a vertex pair consisting of two vertices. The vertices in the vertex pair are sorted in ascending order. Use edge sets to store the edges of all hexahedrons and remove duplicate edges; The faces of a single hexahedron are generated according to the geometric definition of the faces, and the duplicate faces are removed. Finally, all the faces of the brick-concrete structure are drawn to complete the reconstruction of the brick-concrete structure.

6. The reconstruction method according to claim 5, characterized in that: The new spatial mesh generation steps include: Discretize the node coordinates of the remaining nodes into a fixed-size voxel grid, where each voxel represents a small cubic unit in space; Based on the sparse distribution of nodes, invalid voxels are eliminated and only valid voxels containing nodes are retained; For each valid voxel, generate the 3D coordinates of its eight vertices: ； in, represents the node coordinates, Indicates the preset voxel size; The hexahedral topological structure of each voxel is defined using the three-dimensional coordinates of the vertices to form a new spatial grid for the reconstruction of brick-concrete structures.

7. The reconstruction method according to claim 6, characterized in that: Discretize the node coordinates of the remaining nodes into a fixed-size voxel grid consisting of: Traverse all remaining nodes and find their minimum and maximum coordinates and ; Calculate the 3D size of the voxel grid based on the preset voxel size: ， in , , is the voxel grid in The number of voxels in the direction; Create an empty three-dimensional array to store valid voxel information and initialize the voxel grid; wherein the valid voxel information includes the coordinates of the eight vertices of the corresponding hexahedron; Based on node coordinates , calculate the voxel index corresponding to the node: ， Assign the node to the corresponding voxel index and record its voxel index number ; Input the valid voxel information into the initialized voxel grid according to the index number and save the voxel result.

8. A method for visualizing explosion damage of brick-concrete structures, characterized in that: include: Reconstructing the brick-concrete structure under the explosion scene based on the reconstruction method as described in any one of claims 1 to 7; The node stress is embedded into the reconstructed brick-concrete structure, and the numerical difference of the node stress is used to render different stress characteristics, thereby obtaining the rendered visualization result.

Images