Variable structure size building blast cloud prediction method based on generative adversarial network
A method for predicting shock wave cloud maps of buildings with varying structural dimensions is constructed by using generative adversarial networks. This method solves the problem of the difficulty in quickly and accurately predicting the explosive load inside buildings with varying structural dimensions in existing technologies. It enables the rapid generation and accurate prediction of shock wave cloud maps and is applicable to a variety of building structures.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAN MODERN CHEM RES INST
- Filing Date
- 2026-01-30
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies struggle to quickly and accurately predict internal explosion loads in buildings with varying structural dimensions, especially shock wave cloud maps of complex building structures, and lack effective prediction models.
A method for predicting shock wave cloud images of buildings with varying structural dimensions is constructed using generative adversarial networks. By constructing, normalizing, and segmenting the dataset, and training the model using generative adversarial networks, a rapid prediction of shock wave cloud images can be achieved.
It enables rapid and accurate prediction of explosive loads within buildings, reducing cloud map generation time from days to seconds. It is applicable to buildings with different room sizes and corridor widths, reducing testing costs and limitations.
Smart Images
Figure CN122154283A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of damage assessment technology, and relates to shock wave cloud maps, specifically a method for rapid prediction of variable structure size shock wave cloud maps based on generative adversarial networks. Background Technology
[0002] In recent years, the demand for rapid damage assessment technology for non-cooperative building targets on the battlefield has become increasingly prominent. The structural forms and dimensions of building targets vary greatly, and achieving rapid calculation of internal blast loads in buildings of different sizes is a crucial step in realizing rapid overall building damage calculation. However, the propagation law of blast shock waves inside buildings is complex and still under research. While empirical formulas for internal blasts exist for some simple geometric structures, their applicability is limited, and there is a lack of rapid calculation models for internal blast loads in complex building structures and buildings of varying sizes. In recent years, deep learning technology has continuously developed, and its self-learning capabilities have improved, enabling it to efficiently handle complex tasks and make effective predictions in many fields. Therefore, it is necessary to conduct research on rapid prediction methods for internal blast loads in buildings based on deep learning. Summary of the Invention
[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks, thereby solving the technical problem that existing prediction methods are unable to achieve fast and accurate prediction.
[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution.
[0005] A method for predicting shock wave cloud images of buildings with varying structural dimensions based on generative adversarial networks, comprising the following steps.
[0006] Step 1: Dataset Construction.
[0007] Step 101: Construction of the shock wave cloud map dataset.
[0008] When generating simulated building conditions in batches, room dimensions, corridor dimensions, explosive equivalent, and blast point location are set as dynamic values to generate multiple simulated building conditions in batches.
[0009] Finite element simulations were performed on each simulated building condition, and shock wave cloud maps of the cross-section of the floor where the epicenter was located at different times were saved as images to construct a shock wave cloud map dataset.
[0010] Step 102: Construction of the explosion load dataset inside the building.
[0011] The building explosion load dataset consists of multiple data pairs. Each data pair includes an output shock wave cloud image and the corresponding input parameter tensor and input space tensor.
[0012] Step 103, normalization.
[0013] A linear normalization method is used to map all parameters in the building explosion load dataset constructed in step 102 to the interval [0,1].
[0014] Step 104, dataset splitting.
[0015] The building explosion load dataset after normalization in step 103 is segmented to form a training set, a validation set, and a test set.
[0016] Step 2: Construction of a shock wave cloud map prediction model for buildings with variable structural dimensions.
[0017] Based on generative adversarial networks, a shock wave cloud image prediction model for buildings with variable structural dimensions is constructed, using input parameter tensors and input space tensors as inputs and shock wave cloud images as outputs. The model is then trained, validated, and tested using the training, validation, and test sets constructed in step one, resulting in the completed shock wave cloud image prediction model for buildings with variable structural dimensions.
[0018] The input parameter tensor includes the location of the detonation point, the mass of the explosive, the room size, and the time.
[0019] The input spatial tensor includes a room layout diagram.
[0020] Step 3: Predicting shock wave cloud maps for buildings with variable structural dimensions.
[0021] The shock wave cloud map prediction model for buildings with variable structural dimensions, which was constructed in step two, was used to predict the shock wave cloud map of buildings with variable structural dimensions.
[0022] The present invention also has the following technical features.
[0023] In step 101, multiple simulated building conditions are generated in batches using a rapid batch generation method for building conditions in finite element simulation; the rapid batch generation method for building conditions in finite element simulation includes the following steps.
[0024] Step 10101: Analyze the building description parameters and explosive description parameters, and divide them into two categories: fixed parameters and dynamic parameters. For fixed parameters, clarify the values of the fixed parameters; for dynamic parameters, clarify the range of values and the discrete value step size of the dynamic parameters.
[0025] The architectural description parameters include four categories: spatial structural parameters, component size parameters, component reinforcement parameters, and grid division size parameters.
[0026] The explosive description parameters include the detonation point coordinates x, y, and z, and the explosive mass m.
[0027] Step 10102: Based on the building description parameters, construct the mapping relationship between the building description parameters and the modeling instructions of the finite element simulation software, and use the modeling instructions in conjunction with the modeling software to quickly generate a building model file under one working condition; set the value range and step size of the momentum parameter of the building description parameters, and quickly generate building model files under multiple working conditions for finite element simulation in batches.
[0028] Step 10103: Based on the building description parameters, establish the mapping relationship between the building description parameters and the coordinates of each shock wave measuring point on the building, and compile the location information of each measuring point into keywords. The coordinate position corresponding to DATABASE_TRACER enables control of the shock wave measurement point position; the value range and step size of momentum are set to quickly generate shock wave pressure measurement point files for building models in batches.
[0029] Step 10104: Select a cylindrical explosive charge with a length-to-diameter ratio of 3:1, based on the charge mass. ,in, This indicates the density of the cylindrical charge. Represents the cross-sectional radius of the cylindrical charge; calculate the diameter of the cylindrical charge. Establish the coordinates of the detonation point and the charge mass. The mapping relationship between the center coordinates of the upper and lower surfaces of the cylindrical explosive and the cylinder diameter is compiled into keywords. The shape of the explosive can be controlled within `INITIAL_VOLUME_FRACTION_GEOMETRY`; the coordinates of the detonation point are compiled into keywords. Within INITIAL_DETONATION, the position of the explosive is controlled; the range and step size of the momentum are set to quickly generate files containing the explosive detonation points in batches.
[0030] Step 10105: Establish the mapping relationship between building description parameters, explosion point locations, air mesh sizes, and the air mesh boundaries and number of generated meshes in the x, y, and z directions of the building; compile keywords. ALE_STRUCTURED_MESH_CONTROL_POINTS enables control over the size and editing of air meshes; it allows setting the momentum value range and step size to quickly generate variable-size air mesh files.
[0031] In step 10102, the method of establishing the mapping relationship between building description parameters and modeling instructions of finite element simulation software, and applying modeling instructions in combination with modeling software to quickly generate a building model file under a working condition includes the following steps.
[0032] Step 1010201, single room concrete structure.
[0033] Step 1010202: Cut the solid structure of a single room.
[0034] Step 1010203: Mirror copy the entire structure.
[0035] Step 1010204, Concrete mesh generation.
[0036] Step 1010205: Generate the initial steel reinforcement structure.
[0037] Step 1010206, all room steel reinforcement.
[0038] Step 1010207: Complete the reinforcement of the corridor wall.
[0039] Step 1010208, Reinforcement mesh division.
[0040] Step 1010209: Compile all command streams to generate .k files.
[0041] Step 1010210: Extract node and element information to generate the final simulation model.
[0042] Compared with the prior art, the present invention has the following technical effects.
[0043] (I) The method of the present invention can quickly determine the building description parameters and the shock wave cloud map at any explosion time on the floor where the explosion center is located. The shock wave cloud map covers all rooms on the floor where the explosion center is located, and the cloud map generation time is reduced from days in finite element simulation to seconds.
[0044] (II) The method of the present invention is applicable to the rapid prediction of internal explosion shock waves under different blast point locations and explosive charge masses in buildings with different room sizes and corridor widths. The applicable scope of building description parameters and explosive parameters can be dynamically supplemented and expanded according to user needs.
[0045] (III) The rapid batch generation method for building conditions in finite element simulation in this invention can be dynamically supplemented and expanded according to user needs. It can generate building conditions for finite element simulation in batches. The generation time for a single condition is shortened to the minute level, and the algorithm includes dozens of variable structural parameters, which can change room size, floors, and number of spans, greatly expanding the variable dimensions of the building.
[0046] (IV) To address the challenges of complex explosion propagation patterns and difficulty in rapid and accurate prediction within buildings with varying structural dimensions, the present invention employs a simulation + deep learning model approach. This approach offers low experimental costs and fewer limitations, effectively overcoming the constraints of limited ground-based experiments and high costs. Attached Figure Description
[0047] Figure 1 This is a schematic diagram illustrating the principle of a method for rapidly generating building conditions in batches for finite element simulation.
[0048] Figure 2 It is a structural schematic diagram of the architectural model.
[0049] Figure 3 This is a flowchart illustrating the method for establishing the mapping relationship between building description parameters and modeling instructions in finite element simulation software.
[0050] Figure 4 This is a schematic diagram of the structure of the shock wave cloud image dataset.
[0051] Figure 5 This is a schematic diagram illustrating the principle of constructing an explosion load dataset within a building.
[0052] Figure 6 This is a schematic diagram of the structure of the explosion load dataset inside the building.
[0053] Figure 7 This is the result of shock wave cloud map test (prediction).
[0054] Figure 8 This is the prediction chart for operating condition 12 (epoch=4000).
[0055] Figure 9 This is the prediction chart for operating condition 7 (epoch=800).
[0056] The specific content of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Detailed Implementation
[0057] It should be noted that, unless otherwise specified, all software, networks, and evaluation metrics in this invention are based on software, networks, and evaluation metrics known in the prior art.
[0058] Following the above technical solutions, specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent modifications made based on the technical solutions of this application fall within the protection scope of the present invention.
[0059] Example 1:
[0060] This embodiment presents a method for rapidly generating building conditions in batches for finite element simulation, such as... Figure 1 As shown, the method includes the following steps.
[0061] Step 10101: Analyze the building description parameters and explosive description parameters, and divide them into two categories: fixed parameters and dynamic parameters. For fixed parameters, clarify the values of the fixed parameters; for dynamic parameters, clarify the range of values and the discrete value step size of the dynamic parameters.
[0062] Building description parameters include four categories: spatial structural parameters, component size parameters, component reinforcement parameters, and grid division size parameters.
[0063] The explosive description parameters include the x-coordinate of the detonation point, the y-coordinate of the detonation point, the z-coordinate of the detonation point, and the explosive mass m.
[0064] In this embodiment, the spatial structure parameters include the number of floors (N_floor), the number of rooms per floor (N_room), the room length (R_Leng), the room width (R_Width), the room height (R_Height), and the corridor width (C_Width), such as... Figure 2 As shown.
[0065] In this embodiment, the component size parameters include floor slab thickness (TH_floor), wall thickness (Wa_thick), square window size (Wi_width), window height from the ground (Wi_height), distance between two windows (Wi_dist), door height (Do_height), door width (Do_width), and column width (Col_width).
[0066] In this embodiment, the component reinforcement parameters include the diameter of the inner reinforcement bars in the floor slab (Reb_dia_roof), the spacing of the floor slab reinforcement mesh (Reb_dis_roof), the distance between the outer layer reinforcement bars and the outer floor slab (TR_dis_roof), the distance between the inner layer reinforcement bars and the outer floor slab (LR_dis_roof), the distance between the outer layer horizontal reinforcement bars and the center line of the corridor (TR_dis_bis), the distance between the inner layer longitudinal reinforcement bars and the center line of the left wall (LR_dis_bis), the diameter of the wall reinforcement bars (Reb_dia_wall), the spacing of the wall reinforcement mesh (Reb_dis_wall), and the distance between the outer layer longitudinal reinforcement bars and the outer wall panel (LR_dis_). The following parameters are considered: wall, distance from outer longitudinal reinforcement to column (LR_dis_dis), distance from inner transverse reinforcement to outer wall panel (TR_dis_wall), height of inner transverse reinforcement from ground (TR_dis_dis), number of reinforcement bars in column (Reb_num_col), diameter of reinforcement bars in column (Reb_dia_col), distance from inner reinforcement bars to outer wall (LR_dis_col), diameter of stirrups in column (ST_dia_col), stirrup spacing in column (ST_dis), distance from stirrups to outer wall (ST_dis_col), and distance from stirrups to ground (ST_dis_zz).
[0067] In this embodiment, the grid division size parameters include the steel mesh size (gridsize_r) and the concrete mesh size (gridsize_c).
[0068] Step 10102: Based on the building description parameters, construct the mapping relationship between the building description parameters and the modeling instructions of the finite element simulation software, and use the modeling instructions in conjunction with the modeling software to quickly generate a building model file under one working condition; set the value range and step size of the momentum parameter of the building description parameters, and quickly generate building model files under multiple working conditions for finite element simulation in batches.
[0069] In this embodiment, the finite element simulation software used is lsdyna, a commonly known finite element simulation software in the art. The modeling software used is MATLAB, a commonly known modeling software in the art. MATLAB is used to compile and generate a command stream .txt file, which is then imported into lsdyna to generate the simulation software.
[0070] Step 10103: Based on the building description parameters, establish the mapping relationship between the building description parameters and the coordinates of each shock wave measuring point on the building, and compile the location information of each measuring point into keywords. The coordinate position corresponding to DATABASE_TRACER enables control of the shock wave measurement point position; the value range and step size of momentum are set to quickly generate shock wave pressure measurement point files for building models in batches.
[0071] Step 10104: Select a cylindrical explosive charge with a length-to-diameter ratio of 3:1, based on the charge mass. ,in, This indicates the density of the cylindrical charge. Represents the cross-sectional radius of the cylindrical charge; calculate the diameter of the cylindrical charge. Establish the coordinates of the detonation point and the charge mass. The mapping relationship between the center coordinates of the upper and lower surfaces of the cylindrical explosive and the cylinder diameter is compiled into keywords. The shape of the explosive can be controlled within `INITIAL_VOLUME_FRACTION_GEOMETRY`; the coordinates of the detonation point are compiled into keywords. Within INITIAL_DETONATION, the position of the explosive is controlled; the range and step size of the momentum are set to quickly generate files containing the explosive detonation points in batches.
[0072] Step 10105: Establish the mapping relationship between building description parameters, explosion point locations, air mesh sizes, and the air mesh boundaries and number of generated meshes in the x, y, and z directions of the building; compile keywords. ALE_STRUCTURED_MESH_CONTROL_POINTS enables control over the size and editing of air meshes; it allows setting the momentum value range and step size to quickly generate variable-size air mesh files.
[0073] As a preferred embodiment, in step 10102, such as Figure 3 As shown, the method for constructing the mapping relationship between building description parameters and modeling instructions of finite element simulation software includes the following steps.
[0074] Step 1010201, single room concrete structure.
[0075] Based on the values of the building description parameters, the concrete elements of the left wall of the corridor are quickly generated using finite element simulation software. This element is numbered 1. The concrete elements of room 1-1 are then constructed, and the concrete elements of room 1-1 are numbered starting from 2. The modeling instructions of the concrete elements of room 1-1 are extracted, and the mapping relationship between the building description parameters and the modeling instructions of the relevant component dimensions in room.txt is established using programming software to generate a new single room concrete structure.
[0076] Specifically, in this embodiment, the modeling instructions for the concrete unit of room 1-1 are extracted to the room.txt file, the variable names of the building description parameters are defined, and the mapping relationship between the building description parameters and the relevant component size modeling instructions in room.txt is established using programming software. This allows the building description parameters to be edited in the programming software, the program to be run, and the parameters in the corresponding modeling instructions in the room.txt file to be automatically changed. The newly compiled modeling instructions are imported into the modeling software to generate a new single room concrete structure.
[0077] Step 1010202: Cut the solid structure of a single room.
[0078] The solid structure of a single room is cut using cutting commands. The modeling commands for a single room are extracted, and the mapping relationship between architectural description parameters and relevant cutting positions is established using programming software. The concrete structure of a single room is then cut.
[0079] In this specific embodiment, to enable subsequent mesh generation using mapped meshes, the Divide command is used for solid cutting. Considering the complexity of solid cutting for the entire building, it is chosen to first cut the solids of individual rooms. The Divide command is used to cut the solids, extracting the modeling instructions for each room into a divide.txt file. Programming software is then used to establish a mapping relationship between the building description parameters and the relevant cutting positions in divide.txt. This allows the building description parameters to be edited in the programming software, and upon running the program, the parameters in the corresponding modeling instructions in the divide.txt file are automatically updated. The newly compiled modeling instructions are then imported into the modeling software to automatically cut the concrete structure of each room.
[0080] Step 1010203: Mirror copy the entire structure.
[0081] Select the unit numbers of all units in a single room after solid cutting, and apply the command flow to mirror and copy all numbered units to generate the concrete structure of all rooms in the entire building; then select unit number 1 of the concrete wall on the left side of the corridor, and complete the concrete structure of the two walls at the left and right ends of the corridor on each floor of the building by mirroring and copying, and finally generate the concrete structure model of the entire building; extract the modeling instructions of the concrete structure model of the entire building, and use programming software to program and establish the mapping relationship between the building description parameters and the relevant copy position coordinates, and copy and mirror the entire concrete structure of the building.
[0082] Specifically, in this embodiment, the modeling instructions for the entire building's concrete structure model are extracted into a building.txt file. Programming software is used to establish a mapping relationship between the building description parameters and the relevant copy location coordinates in building.txt. This allows the building description parameters to be edited in the programming software, the program to be run, and the parameters in the corresponding location modeling instructions in the building.txt file to be automatically changed. The newly compiled modeling instructions are imported into the modeling software and automatically copied and mirrored into the overall concrete structure of the building.
[0083] Step 1010204, Concrete mesh generation.
[0084] The overall concrete structure of the building is meshed, the modeling instructions for meshing are extracted, and the mapping relationship between the concrete mesh size parameters and the meshing modeling instructions is established using programming software.
[0085] In this specific embodiment, the modeling instructions for mesh generation are extracted into a mesh.txt file. Programming software is used to establish a mapping relationship between the concrete mesh size parameters and the mesh generation modeling instructions in mesh.txt. This allows the concrete mesh parameters to be edited in the programming software. When the program is run, the parameters in the corresponding modeling instructions in the mesh.txt file are automatically changed. The newly compiled modeling instructions are imported into the modeling software to automatically perform mesh generation on the overall concrete structure of the building.
[0086] Step 1010205: Generate the initial steel reinforcement structure.
[0087] In this embodiment, considering that the application of copy and mirror commands is to copy and mirror the selected numbered units, and when generating doors and windows using Boolean operations, the original steel bar unit numbering rules will be disrupted.
[0088] First, construct the reinforcement of the left corridor wall on the first floor of the building. The reinforcement of the left corridor wall on the first floor starts from 1. Calculate the number N of the last reinforcement of this wall based on parameters such as the corridor size and reinforcement spacing. Then, construct the reinforcement of the left wall and two columns of room 1-1 based on the values of the building description parameters. Apply mirroring to generate the reinforcement of the left wall and columns of room 1-4. The reinforcement of the left walls and columns of the two rooms is numbered from N+1 to M.
[0089] Then, generate the reinforcing bars for the wall containing the door and window of room 1-1, the right wall, the other two columns, and the roof slab. The reinforcing bars for the wall containing the door and window of room 1-1, the right wall, the other two columns, and the roof slab are numbered starting from M+1. Calculate the number P of the last reinforcing bar based on the room dimensions and the spacing between the reinforcing bars. Extract the modeling instructions for the reinforcing bars of the wall containing the door and window of room 1-1, the right wall, the other two columns, and the roof slab. Use programming software to establish the mapping relationship between the building description parameters and the relevant reinforcing bar unit dimensions, and generate the new reinforcing bar structure for the left wall and the reinforcing bar structure for room 1-1.
[0090] Specifically, in this embodiment, the application programming software is used to program and establish a mapping relationship between the building description parameters and the relevant steel reinforcement unit dimensions in part1.txt. This allows the building description parameters to be edited in the programming software, the program to be run, and the parameters in the corresponding modeling instructions in the part1.txt file to be automatically changed. The newly compiled modeling instructions are imported into the modeling software to generate new steel reinforcement structures for the left wall and room 1-1.
[0091] Step 1010206, all room steel reinforcement.
[0092] Select the steel reinforcement units numbered M+1 to P, apply the mirror and copy commands to generate the room structure of the entire building; extract the modeling instructions for the room structure of the entire building, apply programming software to program and establish the mapping relationship between the building description parameters and the copy position coordinates of the relevant steel reinforcement units, and generate the room structure of the entire building.
[0093] Specifically, in this embodiment, the modeling instructions for the entire building's room structure are extracted into a part2.txt file. Programming software is used to establish a mapping relationship between the building description parameters and the coordinates of the copy positions of the relevant steel reinforcement units in part2.txt. This allows the building description parameters to be edited in the programming software, the program to be run, and the parameters in the corresponding modeling instructions in the part2.txt file to be automatically changed. The newly compiled modeling instructions are then imported into the modeling software to generate the entire building's room structure.
[0094] Step 1010207: Complete the reinforcement of the corridor wall.
[0095] Select the steel reinforcement units numbered 1 to N, apply the copy command to complete the wall reinforcement at the end of the corridor on each floor; extract the modeling instructions for the wall reinforcement at the end of the corridor on each floor, apply programming software to program and establish the mapping relationship between the building description parameters and the copy position coordinates of the relevant steel reinforcement units, and generate the overall steel reinforcement structure of the building.
[0096] Specifically, in this embodiment, the modeling instructions for the wall reinforcement at the end of each floor's corridor are extracted into a part3.txt file. Programming software is used to establish a mapping relationship between the building description parameters and the coordinates of the copy positions of the relevant reinforcement units in part3.txt. This allows the building description parameters to be edited in the programming software, the program to be run, and the parameters in the corresponding modeling instructions in the part3.txt file to be automatically changed. The newly compiled modeling instructions are imported into the modeling software to complete the corridor wall structure of the entire building. At this point, the overall reinforcement structure of the building is generated.
[0097] Step 1010208, Reinforcement mesh division.
[0098] The overall steel reinforcement structure of the building is meshed, the modeling instructions for meshing are extracted, and the mapping relationship between the steel reinforcement mesh size and the meshing modeling instructions is established using programming software; the overall steel reinforcement structure of the building is then meshed.
[0099] In this specific embodiment, the modeling instructions for mesh generation are extracted to the part4.txt file. Programming software is used to establish a mapping relationship between the steel mesh size and the mesh generation modeling instructions in part4.txt. This allows the steel mesh parameters to be edited in the programming software. When the program is run, the parameters in the corresponding modeling instructions in the part4.txt file are automatically changed. The newly compiled modeling instructions are imported into the modeling software to automatically perform mesh generation on the overall steel reinforcement structure of the building.
[0100] Step 1010209: Compile all command streams to generate .k files.
[0101] The application software compilation process compiles all modeling instructions from steps 1010201 to 1010208 into a single .k file. Importing the .k file into the modeling software will automatically generate a meshed reinforced concrete building structure model.
[0102] Step 1010210: Extract node and element information to generate the final simulation model.
[0103] Save the reinforced concrete building structure model generated by the modeling software according to the command flow as a .k file, and compile and retain it. NODE keyword and The file contains two keywords, ELEMENT. After deleting other keywords from the file, the .k file becomes a building model file generated under a specific working condition for finite element simulation.
[0104] Example 2: This embodiment presents a method for predicting shock wave cloud maps of buildings with varying structural dimensions based on generative adversarial networks. The method is characterized by the following steps.
[0105] Step 1: Dataset Construction.
[0106] Step 101: Construction of the shock wave cloud map dataset.
[0107] When generating simulated building conditions in batches, room dimensions, corridor dimensions, explosive equivalent, and blast point location are set as dynamic values to generate multiple simulated building conditions in batches.
[0108] Finite element simulations were performed on each simulated building condition, and shock wave cloud maps of the cross-section of the floor where the epicenter was located at different times were saved as images to construct a shock wave cloud map dataset.
[0109] In step 101, multiple simulated building conditions are generated in batches using the rapid batch generation method for building conditions in finite element simulation given in Example 1.
[0110] In step 101, each simulated building case includes a building model file model.k, an air mesh file sale.k, a shock wave measuring point file tracer.k, and a main file main.k containing the location and dimensions of the explosives.
[0111] In this embodiment, the constructed shock wave cloud image dataset is as follows: Figure 4 As shown.
[0112] Step 102: Construction of the explosion load dataset inside the building.
[0113] like Figure 5 As shown, the explosion load dataset inside the building consists of multiple data pairs. Each data pair includes an output shock wave cloud image and the corresponding input parameter tensor and input spatial tensor.
[0114] like Figure 5 As shown, the location of the explosion point, the mass of the explosive, the room size, and the time are extracted from each simulated building condition obtained in step 101, and the location of the explosion point, the mass of the explosive, the room size, and the time are combined to form an input parameter tensor.
[0115] like Figure 5 As shown, room layout diagrams are extracted from each simulated building condition obtained in step 101, and the room layout diagrams are used as input spatial tensors.
[0116] like Figure 5 As shown, the shock wave cloud image in the shock wave cloud image dataset obtained in step 101 is used as the output shock wave cloud image.
[0117] In this embodiment, the data set of explosive loads inside the building is associated with data through a unified 6-digit numbering system.
[0118]
[0119] In this embodiment, the input and output parameters are structured. The input parameter tensors are stored in .json files, using both .json (human-readable) and .npy (machine-readable) storage formats. An example of the JSON format is as follows: .
[0120] Step 103, normalization.
[0121] like Figure 5 As shown, the linear normalization method is used to map all parameters in the building explosion load dataset constructed in step 102 to the interval [0,1].
[0122] Step 104, dataset splitting.
[0123] like Figure 5 As shown, the building explosion load dataset after normalization in step 103 is segmented to form a training set, a validation set, and a test set.
[0124] In this embodiment, the data partitioning rules for the training set, validation set, and test set are as follows: each group consists of 10 samples, and in each group: the first sample → test set, the second sample → validation set, and the rest → training set.
[0125] Step 2: Construction of a shock wave cloud map prediction model for buildings with variable structural dimensions.
[0126] Based on generative adversarial networks, a shock wave cloud image prediction model for buildings with variable structural dimensions is constructed, using input parameter tensors and input space tensors as inputs and shock wave cloud images as outputs. The model is then trained, validated, and tested using the training, validation, and test sets constructed in step one, resulting in the completed shock wave cloud image prediction model for buildings with variable structural dimensions.
[0127] The input parameter tensor includes the location of the explosion point, the mass of the explosive, the room size, and the time.
[0128] The input spatial tensor includes a room layout diagram.
[0129] In this embodiment, the overall architecture of the building explosion load dataset is as follows: Figure 6 As shown.
[0130] In this embodiment, the performance of the shock wave cloud map prediction model for buildings with variable structural dimensions is evaluated using the evaluation metrics Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM).
[0131] The smaller the mean squared error (MSE) value, the better (0 indicates a perfect match). Larger errors are penalized more severely (squared terms). The range is from 0 to ∞.
[0132] The higher the peak signal-to-noise ratio (PSNR) value, the better (∞ indicates a perfect match). It is calculated based on MSE, but using a logarithmic scale is more in line with human perception. The range is 0 to ∞ (typically 20-40dB).
[0133] The closer the Structural Similarity Index (SSIM) value is to 1, the better. The range is -1 to 1 (usually 0 to 1), which is more in line with human visual perception.
[0134] Step 3: Predicting shock wave cloud maps for buildings with variable structural dimensions.
[0135] The shock wave cloud map prediction model for buildings with variable structural dimensions, which was constructed in step two, was used to predict the shock wave cloud map of buildings with variable structural dimensions.
[0136] Application example: This application example presents a method for predicting shock wave cloud maps of buildings with variable structural dimensions based on generative adversarial networks, according to Embodiment 2 above.
[0137] First, taking a three-story, three-span building with a corridor as an example, multiple simulated building conditions are generated by changing the room size, corridor size, explosive equivalent, and blast point location.
[0138] Second, the shock wave cloud map simulation dataset was constructed. The fixed values of the working condition parameters are shown in Table 1, and the dynamic values are shown in Table 2. By changing the room size, corridor size, explosive equivalent, and detonation point location, multiple simulated building working conditions were generated using a rapid batch generation method based on finite element simulation. Each working condition contains a building model file (model.k), an air mesh file (sale.k), a shock wave measuring point file (tracer.k), and a main file (main.k) containing the explosive location dimensions. Finite element simulations were performed on each working condition, and the shock wave cloud maps of the cross-section of the floor where the detonation center was located at different times were saved as images to construct the shock wave cloud map dataset.
[0139] Table 1. Operating Parameter Settings (Unit: mm)
[0140] Table 2 Dynamic values of building parameters (unit: mm)
[0141] Third, construct an explosion load dataset inside the building. The explosion point location, explosive mass, room size, and time are combined into an input parameter tensor. The room layout map is used as the input spatial tensor, and the shock wave cloud map at the corresponding time is used as the output shock wave cloud map image to construct the explosion load dataset inside the building.
[0142] Next, the dataset is normalized and split to generate training, validation, and test (prediction) datasets.
[0143] Fourth, a shock wave cloud image prediction model for buildings with variable structural dimensions was constructed based on generative adversarial networks. The model was iterated and optimized 4000 times. The evaluation results of the MSE, PSNR, and SSIM of the test set for the model are as follows: Figure 7 As shown, from Figure 7 As can be seen, after 4000 iterations, the average PSNR and SSIM values of the test set reached the evaluation criteria.
[0144] Fifth, the model for predicting shockwave cloud patterns in buildings with varying structural dimensions is presented, showing the prediction results for buildings of different spatial sizes and at different blast locations. Figure 8 The comparison results of experimental and predicted contour maps at certain times during condition 12 and 4000 iterations are shown. Figure 9 The comparison results of experimental and predicted cloud maps at certain moments under condition 7 and 800 iterations show that the cloud map prediction effect of the shock wave cloud map prediction model for buildings with variable structural dimensions continuously improves as the number of iterations increases.
Claims
1. A method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks, characterized in that, The method includes the following steps: Step 1, Dataset Construction: Step 101, Construction of the shock wave cloud image dataset: When generating simulated building conditions in batches, room size, corridor size, explosive equivalent, and blast point location are set as dynamic values to generate multiple simulated building conditions in batches. Finite element simulations were performed on each simulated building condition, and shock wave cloud images of the cross-section of the floor where the explosion center was located at different times were saved as images to construct a shock wave cloud image dataset. Step 102, Construction of the Explosive Load Dataset Inside the Building: The building explosion load dataset consists of multiple data pairs. Each data pair includes an output shock wave cloud image and the corresponding input parameter tensor and input space tensor. Step 103, Normalization: A linear normalization method is used to map all parameters in the building explosion load dataset constructed in step 102 to the interval [0,1]. Step 104, Dataset Splitting: The building explosion load dataset after normalization in step 103 is segmented to form a training set, a validation set, and a test set. Step 2: Construction of a shock wave cloud image prediction model for buildings with variable structural dimensions: Based on generative adversarial networks, a shock wave cloud image prediction model for buildings with variable structural dimensions is constructed, using input parameter tensors and input space tensors as inputs and shock wave cloud images as outputs. The model is then trained, validated, and tested using the training, validation, and test sets constructed in step one, resulting in the completed shock wave cloud image prediction model for buildings with variable structural dimensions. The input parameter tensor includes the detonation point location, explosive mass, room dimensions, and time. The input spatial tensor includes a room layout diagram; Step 3: Prediction of shock wave cloud map for buildings with variable structural dimensions: The shock wave cloud map prediction model for buildings with variable structural dimensions, which was constructed in step two, was used to predict the shock wave cloud map of buildings with variable structural dimensions.
2. The method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks as described in claim 1, characterized in that, In step 101, a rapid batch generation method for building conditions in finite element simulation is used to generate multiple simulated building conditions in batches. The method for rapid batch generation of building conditions for finite element simulation includes the following steps: Step 10101: Analyze the building description parameters and explosive description parameters, and divide them into two categories: constant parameters and dynamic parameters. For constant parameters, clarify the values of the constant parameters; for dynamic parameters, clarify the range of values and the discrete value step size of the dynamic parameters. The architectural description parameters include four categories: spatial structural parameters, component size parameters, component reinforcement parameters, and mesh division size parameters. The explosive description parameters include the detonation point coordinates x, y, and z, and the explosive mass m. Step 10102: Based on the building description parameters, construct the mapping relationship between the building description parameters and the modeling instructions of the finite element simulation software, and use the modeling instructions in combination with the modeling software to quickly generate a building model file under a working condition; set the value range and step size of the momentum parameter of the building description parameters, and quickly generate building model files under multiple working conditions for finite element simulation in batches. Step 10103: Based on the building description parameters, establish the mapping relationship between the building description parameters and the coordinates of each shock wave measuring point on the building, and compile the location information of each measuring point into keywords. The coordinate position corresponding to DATABASE_TRACER enables control of the shock wave measurement point position; the value range and step size of momentum are set to quickly generate shock wave pressure measurement point files for building models in batches. Step 10104: Select a cylindrical explosive charge with a length-to-diameter ratio of 3:1, based on the charge mass. ,in, This indicates the density of the cylindrical charge. Represents the cross-sectional radius of the cylindrical charge; calculate the diameter of the cylindrical charge. Establish the coordinates of the detonation point and the charge mass. The mapping relationship between the center coordinates of the upper and lower surfaces of the cylindrical explosive and the cylinder diameter is compiled into keywords. The shape of the explosive can be controlled within `INITIAL_VOLUME_FRACTION_GEOMETRY`; the coordinates of the detonation point are compiled into keywords. Within INITIAL_DETONATION, the position of the explosive is controlled; the range and step size of the momentum are set to quickly generate files containing the explosive detonation points in batches. Step 10105: Establish the mapping relationship between building description parameters, explosion point locations, air mesh sizes, and the air mesh boundaries and number of generated meshes in the x, y, and z directions of the building; compile keywords. ALE_STRUCTURED_MESH_CONTROL_POINTS enables control over the size and editing of air meshes; it allows setting the momentum value range and step size to quickly generate variable-size air mesh files.
3. The method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks as described in claim 2, characterized in that, In step 10102, the method of establishing the mapping relationship between building description parameters and modeling commands of finite element simulation software, and applying modeling commands in combination with modeling software to quickly generate a building model file under a working condition, includes the following steps: Step 1010201, Single room concrete structure: Based on the values of the building description parameters, the concrete elements of the left wall of the corridor are quickly generated using finite element simulation software. The element is numbered 1. The concrete elements of room 1-1 are then constructed, and the concrete elements of room 1-1 are numbered starting from 2. The modeling instructions of the concrete elements of room 1-1 are extracted, and the mapping relationship between the building description parameters and the modeling instructions of the relevant component dimensions is established using programming software to generate a new single room concrete structure. Step 1010202, Single room solid cutting: The solid structure of a single room is cut using cutting commands. The modeling commands for a single room are extracted, and the mapping relationship between architectural description parameters and relevant cutting positions is established using programming software to cut the concrete structure of a single room. Step 1010203: Mirror copy to create the entire structure: Select the unit numbers of all units in a single room after solid cutting, apply command flow to mirror and copy all numbered units to generate the concrete structure of all rooms in the entire building; then select unit number 1 of the concrete wall on the left side of the corridor, and complete the concrete structure of the two walls at the left and right ends of the corridor on each floor of the building by mirroring and copying, finally generating the concrete structure model of the entire building; extract the modeling instructions of the concrete structure model of the entire building, apply programming software to program and establish the mapping relationship between the building description parameters and the relevant copy position coordinates, and copy and mirror to form the overall concrete structure of the building. Step 1010204, Concrete mesh generation: The overall concrete structure of the building is meshed, the modeling instructions for meshing are extracted, and the mapping relationship between the concrete mesh size parameters and the meshing modeling instructions is established by programming software. The overall concrete structure of the building is then meshed. Step 1010205, Generate the initial steel reinforcement structure: First, construct the reinforcement of the left corridor wall on the first floor of the building. The reinforcement of the left corridor wall on the first floor of the building starts from 1. Calculate the number N of the last reinforcement of this wall based on parameters such as the corridor size and reinforcement spacing. Then, construct the reinforcement of the left wall and two columns of room 1-1 based on the values of the building description parameters. Apply mirroring to generate the reinforcement of the left wall and columns of room 1-4. The reinforcement of the left walls and columns of the two rooms is numbered from N+1 to M. Then, generate the reinforcing bars for the door and window wall, right side wall, two other columns, and roof slab of room 1-1. The reinforcing bars for the door and window wall, right side wall, two other columns, and roof slab of room 1-1 are numbered starting from M+1. Calculate the number P of the last reinforcing bar based on the room size and the spacing between the reinforcing bars. Extract the modeling instructions for the reinforcing bars for the door and window wall, right side wall, two other columns, and roof slab of room 1-1. Use programming software to establish the mapping relationship between architectural description parameters and relevant reinforcing bar unit dimensions, and generate the new left wall reinforcing bar structure and the reinforcing bar structure of room 1-1. Step 1010206, Reinforcing steel bars in all rooms: Select the steel reinforcement units numbered M+1 to P, apply the mirror and copy commands to generate the room structure of the entire building; extract the modeling instructions for the room structure of the entire building, apply programming software to program and establish the mapping relationship between the building description parameters and the copy position coordinates of the relevant steel reinforcement units, and generate the room structure of the entire building. Step 1010207, complete the reinforcement of the corridor wall: Select the steel reinforcement units numbered 1 to N, apply the copy command to complete the wall reinforcement at the end of the corridor on each floor; extract the modeling instructions for the wall reinforcement at the end of the corridor on each floor, apply programming software to program and establish the mapping relationship between the building description parameters and the copy position coordinates of the relevant steel reinforcement units, and generate the overall steel reinforcement structure of the building. Step 1010208, Reinforcement mesh division: The overall steel reinforcement structure of the building is meshed, the modeling instructions for meshing are extracted, and the mapping relationship between the steel reinforcement mesh size and the meshing modeling instructions is established using programming software; the overall steel reinforcement structure of the building is meshed. Step 1010209: Compile all command streams to generate .k files: The application software compilation process compiles all modeling instructions from steps 1010201 to 1010208 into a single .k file. Importing the .k file into the modeling software will automatically generate a meshed reinforced concrete building structure model. Step 1010210: Extract node and element information to generate the final simulation model. Save the reinforced concrete building structure model generated by the modeling software according to the command flow as a .k file, and compile and retain it. NODE keyword and The file contains two keywords, ELEMENT. After deleting other keywords from the file, the .k file becomes a building model file generated under a specific working condition for finite element simulation.
4. The method for predicting shock wave cloud images of buildings with varying structural dimensions based on generative adversarial networks as described in claim 1, characterized in that, In step 101, each of the simulated building conditions includes a building model file model.k, an air grid file sale.k, a shock wave measuring point file tracer.k, and a main file main.k containing the location and dimensions of the explosives.
5. The method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks as described in claim 1, characterized in that, In step 102, the location of the explosion point, the mass of the explosive, the room size, and the time are extracted from each simulated building condition obtained in step 101, and the location of the explosion point, the mass of the explosive, the room size, and the time are combined into an input parameter tensor.
6. The method for predicting shock wave cloud images of buildings with variable structural dimensions based on generative adversarial networks as described in claim 1, characterized in that, In step 102, room layout diagrams are extracted from each simulated building condition obtained in step 101, and the room layout diagrams are used as input spatial tensors.
7. The method for predicting shock wave cloud images of buildings with varying structural dimensions based on generative adversarial networks as described in claim 1, characterized in that, In step 102, the shock wave cloud image from the shock wave cloud image dataset obtained in step 101 is used as the output shock wave cloud image.