Self-adaptive evaluation method for mechanical response working condition of regional structure under explosion impact

By constructing a graph neural network model with high diversity of explosion impact structure mechanical response database and physical information embedded in graph neural network model, the problem of insufficient generalization ability and accuracy of regional structural mechanical response evaluation under explosion impact in the existing technology is solved, and high accuracy is achieved evaluating effect.

CN119026489AInactive Publication Date: 2024-11-26NANJING UNIV OF SCI & TECH

Patent Information

Application Number
CN202411516983.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-29
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

When evaluating the mechanical response of regional structures under explosion impact, the prior art has weak generalization ability and low accuracy, so it is impossible to achieve high-precision evaluation.

Method used

By integrating experiments with historical explosion data and numerical simulation data, a diverse explosion impact structure mechanical response database is constructed, and a graph neural network model embedded with physical information is trained and optimized, and a digital-driven intelligent model is established to evaluate regional facilities. Structural mechanical response under different explosion conditions.

Benefits of technology

High-precision evaluation of regional structural mechanical responses under explosion impact is realized, the generalization ability of evaluation is enhanced, the three-dimensional unstructured data can be processed, and the robustness and accuracy of the model are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026489A_ABST
    Figure CN119026489A_ABST
Patent Text Reader

Abstract

The invention discloses a self-adaptive evaluation method for a regional structural mechanical response working condition under explosion impact, and the method comprises the following steps: building an explosion impact structural mechanical response data integrating and processing a multi-channel data source, and obtaining multiple groups of explosion impact input data and corresponding structural mechanical response data, combining to obtain an explosion impact structural mechanical response database; building an object-driven intelligent model: dividing the database into a training data set and a verification data set, training the physical information embedded graph neural network by using the training data set, and optimizing the physical information embedded graph neural network by using the verification data set to obtain an explosion structure mechanical response object-driven intelligent model; and regional explosion structure mechanical response evaluation: constructing a field regional explosion case, and inputting the field regional explosion case into the explosion structure mechanical response number driving intelligent model to obtain the field regional explosion structure mechanical response. According to the evaluation method, the mechanical response of the facility structure can be evaluated with high precision under any explosion impact working condition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular to an adaptive evaluation method for the mechanical response of a regional structure under an explosion impact. Background Art

[0002] Regional structural mechanics response refers to the structural mechanics response of facilities within a region when the region is impacted by an explosion. These responses specifically include stress, strain, displacement, and dynamic acceleration.

[0003] Condition adaptation refers to the ability to generalize the mechanical response of facility structures under any explosion conditions, which include regional characteristics, facility characteristics, explosion characteristics, and other aspects.

[0004] The assessment of the structural mechanical response of regional facilities under explosion impact is the basis for evaluating the functional loss of regional systems. It is also an important prerequisite for intelligent decision-making on protection plans, emergency repair and construction plans, and attack plans under modern warfare conditions.

[0005] At present, the evaluation of the response of regional facilities under explosion impact is described in the paper "Beirut explosion 2020A case study for a large-scale urban blast simulation" (G. Valsamos, M.Larcher, F. Casadei, Beirut explosion 2020: A case study for a large-scale urban,blast simulation, Safety Science, Volume 137, 2021, 105190, ISSN 0925-7535.), which discloses a numerical simulation method for large-scale urban explosion simulation, including: using open source map tools to obtain urban area geospatial data; using urban area geospatial data to establish an urban finite element model; using numerical simulation methods to perform explosion simulation on the urban finite element model; using the results of numerical simulation to further evaluate the structural damage and casualties of urban buildings.

[0006] However, the above method relies on numerical simulation and makes many assumptions, which means that it can only evaluate the explosion impact response under specific working conditions and has weak generalization ability.

[0007] At present, the evaluation of the structural mechanical response of a single component under blast impact is described in the paper "Deep learning-based prediction of structural responses of RC slabs subjected to blast loading" (Xiao-Qing Zhou, Bing-Gui Huang, Xiao-You Wang, Yong Xia, Deeplearning-based prediction of structural responses of RC slabs subjected to blast loading, Engineering Structures, Volume 311, 2024, 118184, ISSN 0141-0296), which discloses a deep learning-based method for evaluating the structural response of reinforced concrete (RC) slabs, including: constructing a dataset using existing literature data and supplementary numerical simulation data; establishing a multi-layer perceptron neural network (MLP) model to predict the maximum displacement of reinforced concrete (RC) slabs under blast; establishing a one-dimensional convolutional neural network (1D-CNN) model to predict the failure mode of reinforced concrete (RC) slabs; and performing permutation feature importance analysis to determine the influence of input features on the prediction results.

[0008] However, the MLP and 1D-CNN network models used in this method are limited by the simple model structure and can only evaluate the maximum displacement and corresponding failure mode of a single component, and cannot carry out high-precision evaluation of the regional structural mechanical response under the impact of explosion. It can be seen that there is a lack of a high-precision evaluation method for the regional structural mechanical response under the impact of explosion in the field.

[0009] In summary, the problems existing in the prior art are: the generalization ability of the evaluation of the mechanical response of regional structures under explosion impact is not strong, and the evaluation accuracy is not high. Summary of the invention

[0010] The purpose of the present invention is to provide a method for adaptively evaluating the mechanical response of a regional structure under an explosion impact, which can perform a high-precision evaluation of the mechanical response of a facility structure under any explosion impact condition.

[0011] The technical solution to achieve the purpose of the present invention is: A method for adaptively evaluating the mechanical response of a regional structure under explosion impact, comprising the following steps: Establishment of explosion impact structural mechanics response database: Integrate test and historical explosion data, numerical simulation explosion data, perform data processing, obtain multiple sets of explosion impact input data and corresponding structural mechanics response data, combine the explosion impact input data and structural mechanics response data according to explosion cases, and obtain the explosion impact structural mechanics response database; Construction of a digital-physics-driven intelligent model: dividing the explosion impact structural mechanical response database into a training data set and a verification data set, using the training set to train the graph neural network embedded with physical information, and using the verification set to optimize the graph neural network embedded with physical information, to obtain a digital-physics-driven intelligent model of explosion structural mechanical response; Assessment of the mechanical response of regional explosion structures: The regional characteristics, facility characteristics and explosion characteristics of the explosion site are collected to construct a site-area explosion case, and the site-area explosion case is input into the explosion structure mechanical response numerical-physics-driven intelligent model to obtain the site-area facility explosion structure response.

[0012] Furthermore, the step of establishing the explosion impact structural mechanical response database includes: Integration of test and historical explosion data: By collecting test data on the mechanical response of structures to explosion impact and historical data on the mechanical response of structures to explosion impact, and combining the data, we can obtain test and historical explosion data; Integration of numerical simulation explosion data: by randomly designing regional characteristics, facility characteristics, and explosion characteristics, combining data, obtaining numerical simulation explosion input data, performing batch numerical simulation calculations on the above numerical simulation explosion input data, obtaining the structural mechanical response of the corresponding input, and obtaining numerical simulation explosion data; Database construction: Classify and clean the test and historical explosion data, and numerical simulation explosion data, extract the explosion data that meets the requirements, perform data conversion and data combination, obtain multiple sets of explosion impact input data and structural mechanics response data, and obtain the explosion impact structural mechanics response database after data storage.

[0013] Furthermore, the steps of constructing the digital-physical driven intelligent model include: Data set division: dividing the explosion impact structural mechanical response database into a training data set and a verification data set; Construction of neural network for mechanical response of explosion structures: Construction of a graph neural network that embeds physical information to predict the mechanical response of explosion structures; Acquisition of a digital-physics-driven intelligent model: using the constructed training set to train the explosion structure mechanical response neural network, and using the constructed verification set to optimize the explosion structure mechanical response neural network to obtain a digital-physics-driven intelligent model for evaluation.

[0014] Furthermore, the step of evaluating the mechanical response of the regional explosion structure includes: Regional explosion case construction: collect the regional characteristics, facility characteristics and explosion characteristics of the explosion site, process the data, and construct the regional explosion case on site; Assessment of the structural mechanical response of regional facilities: The on-site regional explosion case is input into the data physics-driven intelligent model of the mechanical response of the explosion impact structure, and the output regional explosion structural response is the assessment of the structural mechanical response of the regional facilities.

[0015] Compared with the prior art, the present invention has the following significant advantages: 1. Stronger generalization ability of evaluation: The present invention ensures the diversity of the database and the robustness of the neural network by establishing an explosion impact structural response database with random regional characteristics, random facility characteristics, and random explosion characteristics, thereby realizing a generalized evaluation of the structural mechanical response of regional characteristics, facility characteristics, and explosion characteristics under explosion impact.

[0016] 2. Higher evaluation accuracy: The present invention embeds the physical information of the structural mechanical response to explosion impact into a graph neural network, which can not only ensure the model's learning ability for the physical problems of the structural mechanical response under explosion impact, but also ensure the model's learning ability for three-dimensional unstructured data, thereby achieving high-precision evaluation of the regional structural mechanical response under explosion impact.

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the main flow chart of the method for adaptively evaluating the working conditions of regional structural mechanical response under explosion impact of the present invention.

[0019] Figure 2 yes Figure 1 Flow chart of the steps for establishing a database of mechanical responses of explosion impact structures.

[0020] Figure 3 This is the explosion point 1 in the example Schematic diagram of the constructed overpressure time history curve.

[0021] Figure 4 This is the explosion point 2 in the example Schematic diagram of the constructed overpressure time history curve.

[0022] Figure 5 It is a schematic diagram of the circular design of the bottom surface of the facility.

[0023] Figure 6 yes Figure 1 Flowchart of the steps for building a digital-physics-driven intelligent model.

[0024] Figure 7 yes Figure 1 Flowchart of the evaluation of mechanical response of structures subjected to explosion in the mid-region. DETAILED DESCRIPTION

[0025] like Figure 1 As shown, the method for adaptively evaluating the mechanical response of a regional structure under an explosion impact of the present invention comprises the following steps: S10. Establishment of explosion impact structural mechanics response database: integrating test and historical explosion data and numerical simulation explosion data, performing data processing, obtaining multiple sets of explosion impact input data and corresponding structural mechanics response data, combining the explosion impact input data and the structural mechanics response data according to explosion cases, and obtaining an explosion impact structural mechanics response database; like Figure 2 As shown, the step of establishing a database of explosion impact structural mechanical responses in S10 includes: S11. Integration of test and historical explosion data: By collecting test data of mechanical response of explosion impact structure and historical data of mechanical response of explosion impact structure, and combining the data, test and historical explosion data are obtained; The step of integrating the test and historical explosion data in S11 includes: S111. Data collection: Collect test data on mechanical response of structures to explosion impact by designing experiments and consulting literature; collect historical data on mechanical response of structures to explosion impact by consulting literature, databases, and accident investigation reports; S112. Data combination: Classify and combine data based on regional characteristics, facility characteristics, and explosion characteristics according to explosion cases.

[0026] S12, numerical simulation explosion data integration: by randomly designing regional characteristics, facility characteristics, explosion characteristics, combining data, obtaining numerical simulation explosion input data, performing batch numerical simulation calculations on the numerical simulation explosion input data, obtaining structural mechanical responses of corresponding inputs, and obtaining numerical simulation explosion data; The step of integrating the numerical simulation explosion data in step S12 includes: S121. Design of numerical simulation explosion input data: designing regional characteristics, facility characteristics, and explosion characteristics by random number method, and combining them according to explosion cases to obtain numerical simulation explosion input data; The step S121, designing numerical simulation explosion input data, comprises: S1211, regional characteristics acquisition: designing the regional size, facility type, facility distribution, and facility quantity by random number method to acquire regional characteristics; S1212, facility characteristics acquisition: design the geometry of each facility and the material type of each facility by random number method to acquire facility characteristics; By obtaining the facility information input data, the random facility information is obtained by arranging and transforming it to construct the facility node data.

[0027] 1. Acquisition of facility information input data: Classify facilities according to different types of facilities, such as buildings, bridges, oil tanks, etc.; obtain key information such as reasonable materials, structural composition, distance between buildings, etc. of each type of facility according to building design specifications. Randomly design reasonable building geometry and building structure, and convert them into building location information. The schematic diagram of the facility bottom circle design is as follows Figure 5 shown.

[0028] 2. Corresponding geometric facility node data: , in, represents the numerical simulation node number, Indicates No. Node facility type, Indicates No. Node location information, Indicates No. Node type information, when , indicating that the node is a common component node of the facility. , indicating that the node is a key component node of the facility, and the node will be removed when an explosion occurs. , indicating that the node belongs to the ground; Indicates Number of numerical simulation nodes material types.

[0029] For example: [1,25,25,60,1,2] means that the node belongs to the first type of facility: power facilities, indicating the location of the node: , this node is a key component of the facility, and the material type of this node is 2, concrete.

[0030] S1213, explosion characteristics acquisition: design explosion time, explosion number, explosion equivalent, and explosion center by random number method to acquire explosion characteristics; 1. Get explosive input through random construction: , in, is the explosion center node number, is the explosion center of numerical simulation, Indicates the number of explosions at the explosion center. Indicates the time when the explosion occurred. Indicated in The explosive equivalent at the moment.

[0031] For example: , Example shows that in the area and Explosions occurred at each of the two sites. There were 2 explosions: When the first explosion occurs, the explosion equivalent is , No. When the second explosion occurs, the explosion equivalent is ; Explosion point 2 1 explosion occurred: When , an explosion occurs, and the explosion equivalent is ,Here data expansion is used to align the data dimensions of the explosion events at two moments.

[0032] 2. Perform data processing to construct the explosion shock wave overpressure time history load: obtain multiple explosion shock loads through the overpressure time history curve. The key parameters for constructing the curve are: explosion time, explosion peak overpressure, and explosion positive pressure time.

[0033] The explosion center distance is obtained based on the facility node and the explosion center, as shown in the formula As shown; according to the explosion equivalent and the distance from the explosion center, the proportional distance is obtained, as shown in the formula As shown; according to different proportional distances, the corresponding peak overpressure and shock wave positive pressure time are obtained by reasonably using the Sadovskyi formula, Henrych formula, etc.; the explosion time is simplified to the shock wave arrival time.

[0034] For example: The explosion input information is as follows: , Assume the node position is .

[0035] (1), in, is the explosion center distance, and the facility node position vector is The explosion center is . Among them, the explosion point 1 Explosive distance: , explosion point 2 Explosive distance: .

[0036] (2) in, is the proportional distance, The distance from the center of explosion.

[0037] At explosion point 1 : In the When the first explosion occurs, for the node Scale distance: , in When , the second explosion occurs, for the node Scale distance: At explosion point 2 : In the When , an explosion occurs, for the node Scale distance: .

[0038] For example, Henrych's peak overpressure formula: (3).

[0039] Henrych positive pressure action time formula: (4).

[0040] Sadovskyi peak overpressure formula: (5).

[0041] Sadovskyi positive pressure action time formula: (6).

[0042] In the above empirical relationship, represents the peak overpressure, Indicates the positive pressure action time, represents the proportional distance, Indicates explosive equivalent.

[0043] At explosion point 1 : In the When the first explosion occurs, for the node Peak overpressure: , positive pressure time: , in When , the second explosion occurs, for the node Peak overpressure: , positive pressure time: At explosion point 2 : In the When , an explosion occurs, for the node Peak overpressure: , positive pressure time: .

[0044] The explosion time is set to the time when the shock wave arrives. Explosion point 1 The constructed overpressure time history curve is shown in Figure 3 As shown, explosion point 2 The constructed overpressure time history curve is shown in Figure 4 shown.

[0045] S1214, acquisition of numerical simulation explosion input data: construct three-dimensional data according to the explosion case based on the above regional characteristics and facility characteristics, and combine the above three-dimensional data with the explosion characteristics to obtain numerical simulation explosion input data: [ ].

[0046] S122. Acquisition of numerical simulation explosion data: performing batch numerical simulation calculations on the numerical simulation explosion input data according to explosion cases, acquiring structural mechanical responses of corresponding explosion cases, and combining the structural mechanical responses with the numerical simulation explosion input data to obtain numerical simulation explosion data.

[0047] 1. Structural mechanics response data example: [ ], in, Indicates the number The stress at the node, Indicates the number The strain at the node, Indicates the number The displacement of the node, For this node Dynamic response acceleration at all times. This part is also used as the training truth value.

[0048] 2. Example of numerical simulation explosion data: [ ].

[0049] Through the acquired explosion information input data and random facility construction information, the numerical calculation input of the structural mechanical response of the explosion impact facility is designed through random combination and rationality verification. Batch numerical simulation is designed using python and numerical simulation software to quickly obtain the structural mechanical response of the facility nodes of each random explosion simulation event. The facility structural mechanical response consists of stress, strain, displacement and dynamic response acceleration. The structural mechanical responses of the explosion event facility are combined to obtain the true value data for each explosion event training.

[0050] S13. Database construction: Classify and clean the test and historical explosion data, and numerical simulation explosion data, extract the explosion data that meets the requirements, perform data conversion and data combination, obtain multiple sets of explosion impact input data and structural mechanics response data, and obtain the explosion impact structural mechanics response database after data storage.

[0051] The step of constructing a database in step S13 includes: S131, explosion data extraction: classify the test and historical explosion data, and the numerical simulation explosion data according to data characteristics, clean the erroneous, incomplete or inconsistent data in the above classified data, and extract the explosion data that meets the requirements; S132, data conversion and combination: formatting the explosion data that meets the requirements according to the data type, standardizing the formatted data, and combining the standardized data according to the regional characteristics, facility characteristics, explosion characteristics, and structural mechanical responses of each explosion case, so as to achieve data conversion and combination; The step S132, data conversion and combination, includes: S1321, Data formatting: Convert the raw data in the data-driven intelligent model into a format that can be recognized by the neural network, and collect data features from discrete grid nodes; For example, 1. Reshape the original data into a continuous feature vector form, and expand or compress the data in a certain order into a storage form recognizable by the neural network through interpolation or sampling methods.

[0052] 2. Use graph-structured data to directly collect features from discrete grid nodes to avoid loss of accuracy during data reshaping, and inherit the spatial characteristics of these grids into the graph according to the graph data structure.

[0053] S1322, data standardization: according to different data features of the formatted data, a data standardization method is used to map the numerical ranges of different data features to similar intervals.

[0054] For example, the standardized method for the mechanical response data of explosion impact structures is shown in Table 1 below. is the original data, is the standardized data, parameter for The mean value, parameter for The standard deviation of .

[0055] Table 1 Standardization method .

[0056] The explosion impact input data is obtained by combining the test and historical explosion data, the explosion information and facility information under each explosion event in the random simulation model. The explosion impact input data and the corresponding input structural mechanics response data are integrated according to the explosion event to obtain the explosion impact structural mechanics response data. The data structure is as follows: S133, explosion impact input data construction: The above data obtained by combining the regional characteristics, facility characteristics, and explosion characteristics of each explosion case are used to construct the explosion impact input data according to the following data structure: [ ], in, Indicates the facility node number, Indicates the explosion center number, Indicates No. Node facility type, Indicates No. Node location information, Indicates No. Node type information, when , indicating that the node is a common component node of the facility. , indicating that the node is a key component node of the facility, and the node will be removed when an explosion occurs. , indicating that the node belongs to the ground; Indicates No. Node material type; Indicates the number of explosions at the explosion center. Indicates the time when the explosion occurred. Indicated in The explosion center The explosion equivalent at the moment, For the corresponding The moment of occurrence The distance from the center of the explosion.

[0057] S134, structural mechanics response data construction: The above data obtained by combining the structural mechanics responses of each explosion case are combined into the following data structure, structural mechanics response parameter data: [ , , , ], in, Indicates the number The stress at the node, Indicates the number The strain at the node, Indicates the number The displacement of the node, For this node Dynamic response acceleration at all times. This part is also used as the training truth value.

[0058] S135, database construction: storing the above-mentioned explosion impact input data and structural mechanics response data as an explosion impact structural mechanics response database.

[0059] Explosion impact structural mechanical response data: [ ].

[0060] The present invention ensures the diversity of the database and the robustness of the neural network by establishing an explosion impact structural response database with random regional characteristics, random facility characteristics, and random explosion characteristics, thereby realizing a generalized evaluation of the structural mechanical responses to regional characteristics, facility characteristics, and explosion characteristics under explosion impact.

[0061] S20, constructing a digital-physics-driven intelligent model: dividing the explosion impact structure mechanical response database into a training data set and a verification data set, using the training set to train the graph neural network embedded with physical information, and using the verification set to optimize the graph neural network embedded with physical information, to obtain a digital-physics-driven intelligent model of explosion structure mechanical response; like Figure 6 As shown, the step of constructing a digital-object-driven intelligent model in S20 includes: S21, data set division: dividing the explosion impact structural mechanical response database into a training data set and a verification data set; S22. Construction of neural network for mechanical response of explosion structures: Construction of a graph neural network embedded with physical information for predicting mechanical response of explosion structures; The neural network for the mechanical response of explosion structures is constructed by incorporating the key control equations in the structural mechanical response as constraints into the loss function of the graph neural network consisting of a graph neural network encoding layer, a graph neural network processing layer, and a graph neural network decoding layer.

[0062] The step S22, constructing a neural network for explosion structure mechanical response, includes: S221. Graph neural network construction: construct a graph neural network including a graph neural network encoding layer, a graph neural network processing layer, and a graph neural network decoding layer; The graph neural network encoding layer is the part that constructs the first layer of potential graph data; the graph neural network processing layer is the part that updates the graph data, and obtains the last layer of graph data through multiple iterations of the graph neural network processing layer; the graph neural network decoding layer is the part that transforms the graph data according to the model's tasks and the nature of the problem to obtain the final prediction result; The graph neural network consists of a graph neural network encoding layer, a graph neural network processing layer, and a graph neural network decoding layer. The graph neural network encoding layer is the part that constructs the first layer of potential graph data; the graph neural network processing layer is the part that updates the graph data, and obtains the last layer of graph data through multiple iterations of the graph neural network processing layer; the graph neural network decoding layer is the part that transforms the graph data according to the model's tasks and the nature of the problem to obtain the final result.

[0063] The step S221, graph neural network construction step includes: S2211, Construction of graph neural network coding layer: Use a multi-layer fully connected neural network, introduce the ReLU activation function for nonlinear processing, and construct the graph neural network coding layer; In the process of building a graph data structure suitable for graph neural network (GNN) processing, the core function of the input layer is to convert the original grid node data into Mapping to a graph representation rich in high-dimensional information ,This conversion process lays a solid foundation for the subsequent graphics processing layer.

[0064] The S2221, graph neural network input layer construction includes: 1. Automatically establish association pairs between nodes through spatial proximity (such as radius search method, K_means), and then use the graph encoding layer to construct edge features , representing the node With Node The connection relationship between them.

[0065] 2. Design a multi-layer fully connected neural network architecture, aiming to give the network powerful feature capture capabilities through embedding operations in high-dimensional space.

[0066] 3. In terms of nonlinear processing, the ReLU activation function is introduced. Its unique positive value retention property prompts the network to focus more on key features that have a significant impact on the task objectives during the learning process, thereby effectively improving the efficiency and accuracy of the training process.

[0067] Through the graph neural network encoding layer, association pairs between nodes are established and edge features are constructed to obtain the first layer of potential graph data, laying a solid foundation for subsequent graph neural network processing.

[0068] S2212. Construction of graph neural network processing layer: In the graph neural network processing layer, the potential graph data obtained by the encoding layer Iterative processing is performed through multiple 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 update; About GN blocks: 1. Update the side information of this iteration by aggregating the starting point, end point and side information of the previous iteration. This process can be expressed as: (7), Among them, The output of the th vertex and edge iteration is the input of the th, Indicates The side information input at the iteration, and Respectively represent The starting point and end point information connected to this edge entered in the iteration, Indicates The side information output by the iteration. Function The side information is learned and updated through a fully connected network.

[0069] 2. Click information The update is done by aggregating the original node information and all updated side information connected to it To achieve: (8), Among them, The output of the th vertex and edge iteration is the input of the th, Indicates The node information entered at the iteration, Indicates the node connected to The side information set after the iteration update. Indicates The node information newly generated in the iteration. Function Point information is also updated through the fully connected network.

[0070] After each GN block iteration, the updated node information and side information Will be recombined to get Graph data for iterations Through the continuous iteration of multiple GN blocks, the model can capture the complex node and edge relationships in the graph data, and finally generate highly abstract and characterized graph data. , and feed it into the graph decoding layer to generate the result.

[0071] S2213. Construction of graph neural network decoding layer: 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 neural network decoding layer is the high-dimensional graph data of the last layer of the graph neural network processing layer. The node information of the multi-layer fully connected network can be restored to obtain the node information data with stress, strain, displacement, and dynamic response acceleration: (9), in, It is the grid node information data obtained by the final iteration of the graph processing layer. are the predicted node parameters.

[0072] S222. Embedding physical information: constructing physical information constraints through explosion impact structural response control equations, adding the physical information constraints to the loss of the graph neural network, and obtaining a graph neural network embedded with physical information.

[0073] The step of embedding physical information in step S222 includes: S2221. Construction of physical information constraint loss term: Physical information constraint loss term Obtained from the residual of the elastic-plastic mechanics constraint equation under explosion impact: The dynamic equilibrium equation constraint describing the relationship between external force and stress , geometric equation constraints describing the relationship between strain and displacement : (10).

[0074] The dynamic equilibrium equation for the relationship between external forces and stresses is described by the constraint loss term: (11), in, Indicates different directions, is the predicted stress tensor , is the material density, is the external force per unit mass, is the predicted dynamic acceleration component obtained from the dynamic mechanical response.

[0075] The geometric equations describing the relationship between strain and displacement are described by the constraint loss term: (12), in, is the component of the predicted strain tensor , , represents the predicted displacement component.

[0076] S2222, Physical Information Constraint Embedding: Constructing Data Loss Term by Mean Square Error (MSE): (13), in, Indicates the use of MSE method to obtain data loss, represents the number of nodes in the graph, Indicates The truth value of each node, Indicates The predicted value of a node.

[0077] By adding physical information constraints to the loss, the embedding of physical information constraints is achieved: (14).

[0078] Embedding physical information is achieved by adding physical information constraints to the loss, which is mainly reflected in the loss term of the model. The loss term of the model is composed of the data loss term and the physical information constraint loss term Since the final prediction focuses on the mechanical response field of explosion impact, the physical information constraint loss term Loss of elastic-plastic mechanics constraint equations under explosion impact: Dynamic equilibrium equation constraints describing the relationship between external force and stress , geometric equation constraints describing the relationship between strain and displacement Since the material type is selected in the training input data, the constructed model can achieve material adaptation, so the physical equation describing the stress-strain relationship in the elastic-plastic mechanics constraint is not added to the constraint equation.

[0079] S23. Acquisition of a digital-physics-driven intelligent model: using the training data set to train the explosion structure mechanical response neural network, and using the verification data set to optimize the explosion structure mechanical response neural network, so as to obtain a digital-physics-driven intelligent model for evaluation.

[0080] S231, Model training: The Adam algorithm is used to update the model parameters. The model is trained in a supervised manner. The model parameters are randomly generated at the beginning of training. The back propagation algorithm is needed to back propagate the value of the loss function and calculate the gradient of the model parameters relative to the loss value to update the model parameters. The selected back propagation algorithm is the Adam algorithm, which combines the characteristics of gradient descent and momentum optimization and has good performance and convergence speed.

[0081] The Adam algorithm uses the following formula in each iteration: 1 Calculate the gradient : (15), In the formula, is to find the gradient of the loss function with respect to the model parameters, Represents the loss of the model. Obviously, the size of the gradient is affected by the loss function used and the model structure.

[0082] 2 Update the first-order moment estimate: (16), In the formula, is the updated first-order moment estimate, is the first-order moment estimate from the previous step, is the exponential decay rate of the first-order moment estimate, usually taken as 0.9.

[0083] 3 Update the second-order moment estimate: (17), In the formula, is the updated second-order moment estimate, is the second-order moment estimate from the previous step, is the exponential decay rate of the second-order moment estimate, usually set to 0.999.

[0084] Since the first and second moment estimates are close to zero at the beginning of the calculation, this may lead to and , and In order to correct these deviations, the Adam algorithm introduces a correction factor to correct the first-order moment estimate and the second-order moment estimate: (18), (19), In the formula, for The first-order moment estimate after the iteration correction is: for Iteratively corrected second-order moment estimate.

[0085] 4 Use the corrected first- and second-order moment estimates to compute updates to the model parameters: (20), in, are the model parameters before updating, are the updated model parameters, is the learning rate, It is a small number added to avoid the denominator being 0, usually 10e-8. This formula is the key step in updating the parameters, and each parameter will be updated in an adaptive manner. It should be noted that all parameters ( , , , ) need to be set before training and usually will not change during training.

[0086] S232. Model optimization: When performing hyperparameter tuning, you need to pay attention to a series of key hyperparameters, including the number of iterations, batch size, learning rate, and adjustment of the model structure. The key points of adjustment are as follows: 1 The number of iterations is between 50 and 300 epochs; 2 The range of batch size adjustment is between 2 and 128; 3. The learning rate is usually adjusted between 0.0001 and 0.01; 4 Adjustment of the model structure is also a key step in optimization. According to the gradient calculation formula, the design of the structure affects the direction of the gradient update. Therefore, the model structure needs to be modified according to the model prediction results, which includes adjusting the number of layers in the model, the number of units in each layer, and selecting the appropriate graph network layer configuration.

[0087] S233. Model evaluation: Model evaluation is a crucial step in the training process. It not only helps us determine whether the performance of the model meets expectations, but also guides us on how to further adjust and optimize the model.

[0088] 1. Divide the original data set into a training data set and a validation data set. The validation data set does not directly participate in model training, but is used to provide feedback on the current training effect of the model.

[0089] 2 is the prediction error of the model on the test set. The prediction results of the model will be compared point by point with the true value of the numerical calculation to calculate the relative error between them. The test set is completely independent of the original data set and is usually data that has not been seen in model training, which is conducive to further evaluating the generalization performance of the model.

[0090] In addition, the mechanical response of the facility unit: stress, strain, displacement, and acceleration of dynamic response are taken as prediction targets. In some small equivalent cases, the displacement of the facility unit is close to 0. In order to avoid the influence of the denominator, in this study, the calculation formula of the relative error is as follows: (twenty one), In the formula represents a node in the graph, is the node feature for numerical calculation, The node features predicted by the model.

[0091] (twenty two), in, represents the mean relative error of the nodes, Indicates the maximum relative error of the node.

[0092] The present invention embeds the physical information of the structural mechanical response to explosion impact into a graph neural network, which can not only ensure the model's learning ability for the physical problems of the structural mechanical response under explosion impact, but also ensure the model's learning ability for three-dimensional unstructured data, thereby achieving high-precision evaluation of the regional structural mechanical response under explosion impact.

[0093] S30. Assessment of the mechanical response of regional explosion structures: Collect the regional characteristics, facility characteristics and explosion characteristics of the explosion site to construct a regional explosion case, input the regional explosion case into the explosion structure mechanical response numerical and physical driven intelligent model, and obtain the mechanical response of the regional explosion structure.

[0094] like Figure 7 As shown, the step S30, regional explosion structure mechanical response assessment step includes: S31. Construction of regional explosion cases: Collect the regional characteristics, facility characteristics and explosion characteristics of the explosion site, process the data, and construct regional explosion cases on site; The steps of constructing the S31 regional explosion case include: S311, Data collection: Collect regional characteristics, facility characteristics and explosion characteristics through GIS information system, and classify and combine them according to explosion cases to realize data collection; The regional characteristics include: regional size, facility type, facility distribution, and facility quantity, and the regional characteristics are obtained; the facility characteristics include: the geometry of each facility and the material type of each facility; the explosion characteristics include: explosion time, number of explosions, explosion equivalent, and explosion center, and the explosion characteristics are obtained.

[0095] S312, data processing: classifying, cleaning, converting and combining the above collected data according to data features to achieve data processing; The step S312, data processing, includes: S3121. Data classification: Classify the collected regional characteristics, facility characteristics and explosion characteristics according to data features; S3122, data cleaning: deleting the erroneous, incomplete or inconsistent data in the above classified data to achieve data cleaning; S3123. Data conversion and combination: Format the cleaned data according to the data type, standardize the formatted data, and combine the standardized data according to the regional characteristics, facility characteristics, explosion characteristics, and structural mechanical responses of each explosion case to achieve data conversion and combination.

[0096] The model constructed in this paper can predict the structural mechanical response of any facility under explosion impact. Through GIS technology, the key information of the facilities in the area is obtained according to the longitude and latitude of the relevant area. The ground bottom circle of the facility is constructed according to the building distance radius corresponding to the facility type to obtain complete three-dimensional information of the facility, and the effective explosion information is sorted out. Its information composition is consistent with the facility node information in the training input, as follows: [ , in, Indicates the regional node number, Indicates No. Node facility type, Indicates No. Node location information, Indicates No. Node type information, when , indicating that the node is a common component node of the facility. , indicating that the node is a key component node of the facility, and the node will be removed when an explosion occurs. , indicating that the node belongs to the ground; Indicates Node material type.

[0097] The model constructed in this paper can predict the mechanical response of the facility structure under any explosion load under explosion impact. Through information collection technology or reasonable explosion design, the effective explosion information is sorted out to obtain the explosion information of each explosion event. Its information composition is consistent with the explosion information in the training input, as follows: , in, It is the explosion center node number of the predicted area. is the explosion center of the region, Indicates the number of explosions at the explosion center. Indicates the time when the explosion occurred. Indicated in The explosive equivalent at the moment.

[0098] S313, explosion case construction: construct an explosion case according to the following data structure using the data obtained through the above data processing: [ ], in, For the corresponding The moment of occurrence The distance from the explosion center at node No.

[0099] S32. Assessment of the mechanical response of regional facility structures: Input the on-site regional explosion case into the data physics-driven intelligent model of the mechanical response of the explosion impact structure, and output the regional explosion structural response, which is the assessment of the mechanical response of the regional facility structure.

[0100] The corresponding node data structure predicted is as follows: , , , ], in, Indicates the number The stress at the node, Indicates the number The strain of the node, Indicates the number The displacement of the node, For this node Dynamically respond to acceleration at all times.

Claims

1. A method for adaptively evaluating the mechanical response of regional structures under explosion impact, characterized in that: The steps include: Establishment of explosion impact structural mechanics response database: integrating test and historical explosion data, numerical simulation explosion data, performing data processing, obtaining multiple sets of explosion impact input data and corresponding structural mechanics response data, combining the explosion impact input data and structural mechanics response data according to explosion cases, and obtaining an explosion impact structural mechanics response database; Construction of a digital-physics-driven intelligent model: dividing the explosion impact structure mechanical response database into a training data set and a verification data set, using the training data set to train the graph neural network embedded with physical information, and using the verification data set to optimize the graph neural network embedded with physical information, to obtain a digital-physics-driven intelligent model of explosion structure mechanical response; Assessment of the mechanical response of regional explosion structures: Collect the regional characteristics, facility characteristics and explosion characteristics of the explosion site, construct a field regional explosion case, input the field regional explosion case into the explosion structure mechanical response numerical and physical driven intelligent model, and obtain the field regional facility explosion structure mechanical response.

2. The evaluation method according to claim 1, characterized in that: The step of establishing the explosion impact structural mechanical response database comprises: Integration of test and historical explosion data: By collecting test data on the mechanical response of structures to explosion impact and historical data on the mechanical response of structures to explosion impact, and combining the data, we can obtain test and historical explosion data; Integration of numerical simulation explosion data: by combining data of random design area characteristics, facility characteristics, explosion area explosion structural mechanics explosion characteristics, obtaining numerical simulation explosion input data, performing batch numerical simulation calculations on the above numerical simulation explosion input data, obtaining the structural mechanics response of the corresponding input, and obtaining numerical simulation explosion data; Database construction: Classify and clean the test and historical explosion data and numerical simulation explosion data, extract the explosion data that meets the requirements, perform data conversion and data combination, obtain multiple sets of explosion impact input data and structural mechanics response data, and obtain the explosion impact structural mechanics response database after data storage.

3. The evaluation method according to claim 2, characterized in that: The steps of integrating the test and historical explosion data include: Data collection: Collect test data on the mechanical response of structures subjected to explosion impact by designing experiments and consulting literature; collect historical data on the mechanical response of structures subjected to explosion impact by consulting literature, databases, and accident investigation reports; Data combination: Classify and combine data based on regional characteristics, facility characteristics, and explosion characteristics according to explosion cases.

4. The evaluation method according to claim 2, characterized in that: The numerical simulation explosion data integration step comprises: Design of numerical simulation explosion input data: Design regional characteristics, facility characteristics, and explosion characteristics through random number methods, and combine them according to explosion cases to obtain numerical simulation explosion input data; Acquisition of numerical simulation explosion data: performing batch numerical simulation calculations on the numerical simulation explosion input data according to explosion cases, acquiring the structural mechanics response of the corresponding explosion cases, and combining the structural mechanics response with the numerical simulation explosion input data to obtain numerical simulation explosion data.

5. The evaluation method according to claim 1, characterized in that: The steps of constructing the digital-object driven intelligent model include: Data set division: dividing the explosion impact structural mechanical response database into a training data set and a verification data set; Construction of neural network for mechanical response of explosion structures: Construction of a graph neural network that embeds physical information to predict the mechanical response of explosion structures; Acquisition of a digital-physics-driven intelligent model: using the training data set to train the explosion structure mechanical response neural network, and using the verification data set to optimize the explosion structure mechanical response neural network, so as to obtain a digital-physics-driven intelligent model for evaluation.

6. The evaluation method according to claim 5, characterized in that: The steps of constructing the neural network of the explosion structure mechanical response include: Graph neural network construction: Build a graph neural network including a graph neural network encoding layer, a graph neural network processing layer, and a graph neural network decoding layer; Embedding physical information: constructing physical information constraints through the explosion impact structural response control equation, adding the physical information constraints to the loss of the graph neural network, and obtaining a graph neural network embedded with physical information.

7. The evaluation method according to claim 6, characterized in that: The graph neural network construction steps include: Construction of graph neural network encoding layer: 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 encoding layer; Construction of graph neural network processing layer: In the graph neural network processing layer, the potential graph data obtained by the encoding layer Iterative processing is performed through multiple graph neural network blocks with the same structure to achieve in-depth analysis and feature extraction of data; each iteration of the graph neural network block includes the update of two key information: edge information and point information; Construction of graph neural network decoding layer: The input of the graph neural network decoding layer is the high-dimensional graph data of the last layer of the graph neural network processing layer. The node information of the network can be restored through multi-layer fully connected network information to obtain node information data with stress, strain, displacement and dynamic response acceleration.

8. The evaluation method according to any one of claims 1 to 7, characterized in that: The steps of evaluating the mechanical response of regional explosion structures include: Regional explosion case construction: collect the regional characteristics, facility characteristics and explosion characteristics of the explosion site, process the data, and construct the regional explosion case on site; Structural mechanics response assessment of regional facilities: The on-site regional explosion case is input into the explosion structural mechanics response numerical and physical driven intelligent model, and the output regional explosion structural response is the structural mechanics response assessment of the regional facilities.

9. The evaluation method according to claim 8, characterized in that: The steps of constructing the regional explosion case include: Data collection: regional characteristics, facility characteristics and explosion characteristics are collected through the GIS information system, and classified and combined according to explosion cases to achieve data collection; Data processing: classify, clean, convert and combine the collected data according to data characteristics; Explosion case construction: The processed data is used to construct an explosion case according to the following data structure: [ ], in, It is the explosion center node number of the predicted area. Indicates the regional node number, Indicates No. Node facility type, Indicates No. Node location information, Indicates No. Node type information, when , indicating that the node is a common component node of the facility. , indicating that the node is a key component node of the facility, and the node will be removed when an explosion occurs. , indicating that the node belongs to the ground; Indicates No. Node material type; Indicates the number of explosions at the explosion center in this area. Indicates the time when the explosion occurred. Indicated in At the node The explosion equivalent at the moment, For the corresponding The moment of occurrence The distance from the explosion center at node No.

10. The evaluation method according to claim 9, characterized in that: The data processing steps include: Data classification: Classify the collected regional characteristics, facility characteristics and explosion characteristics according to data features; Data cleaning: Delete erroneous, incomplete or inconsistent data from the classified data to achieve data cleaning; Data conversion and combination: The cleaned data is formatted and standardized according to the data type, and then combined according to the regional characteristics, facility characteristics, explosion characteristics, and structural mechanical responses of each explosion case.

Citation Information

Patent Citations

  • Power transmission tower wind vibration dynamic response prediction method based on embedded physical information learning

    CN115828698A

  • Plane stress problem solving method based on physical information neural network

    CN118410532A

  • CHON type explosive explosion shock wave parameter prediction method based on physical information neural network

    CN118821605A

Cited By

  • Explosion overpressure field physical enhancement reconstruction method and system oriented to sparse observation

    CN119397929A

  • Frame structure mechanical response and collapse evaluation method and system adaptive to component dismantling

    CN119416661A

  • Limited open space building explosive power field machine learning and escape planning method

    CN119578680A

  • Machine Learning and Escape Planning Method for the Explosion Power Field of Buildings in Limited Open Spaces

    CN119578680B

  • Brick-concrete structure explosion damage reconstruction and visualization method based on deep learning

    CN119579797A