Bridge earthquake damage rapid assessment method and system
By constructing a multi-component association graph structure and graph neural network model for bridges, the problems of long time consumption and insufficient accuracy in bridge seismic damage assessment in existing technologies are solved, realizing rapid and accurate assessment of bridge damage status, which is suitable for post-earthquake emergency assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF ENG MECHANICS CHINA EARTHQUAKE ADMINISTRATION
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-15
AI Technical Summary
Existing bridge seismic damage assessment methods are computationally time-consuming, have complex preprocessing, are highly dependent on sensor accuracy, and fail to effectively characterize the dynamic relationships among multiple components when assessing a single component, resulting in insufficient accuracy in identifying minor and moderate damage.
A graph neural network is used to construct a multi-component relationship graph structure for bridges. Component features are extracted through a residual convolutional neural network, and damage assessment is performed by combining the graph neural network model with multi-component response information.
It enables rapid and accurate assessment of bridge damage status under conditions of limited sensor quantity, improves the identification accuracy of minor and moderate damage status, and is suitable for post-earthquake emergency assessment.
Smart Images
Figure CN122046489A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bridge seismic damage assessment technology, and more specifically to a rapid bridge seismic damage assessment method and system. Background Technology
[0002] After an earthquake, the rapid identification of bridge damage, as a key infrastructure in the transportation network, is of great significance for emergency rescue and traffic restoration.
[0003] Existing bridge seismic damage assessment methods mainly include finite element analysis-based methods and acceleration integral inversion displacement methods. However, these methods generally suffer from problems such as long computation time, complex preprocessing, and high dependence on sensor accuracy, making it difficult to meet the needs of rapid post-earthquake assessment.
[0004] In recent years, deep learning methods have been introduced into the field of bridge seismic damage assessment. However, existing methods mostly use single components or simple feature splicing for assessment, which fails to effectively characterize the dynamic relationship between multiple bridge components, resulting in insufficient accuracy in identifying minor and moderate damage states.
[0005] Therefore, how to provide a rapid bridge seismic damage assessment method and system that can fully integrate the response information of multiple bridge components under the condition of a limited number of sensors, and achieve rapid and accurate assessment of the post-earthquake bridge damage status is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] In view of this, the present invention provides a method and system for rapid assessment of bridge earthquake damage, which uses graph neural networks to rapidly assess post-earthquake damage to bridges, solving the problems of insufficient utilization of multi-component information and difficulty in balancing assessment accuracy and efficiency in the prior art.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for rapid assessment of bridge seismic damage, comprising: Acquire time history data of acceleration of several key bridge components under seismic loading; The acceleration time history data is preprocessed, and the time-series response features of each key component are extracted to form a three-dimensional component feature vector. Based on the structural connection relationships and response correlations among the key components, a multi-component association diagram structure for the bridge is constructed. A graph neural network model was established and trained based on the aforementioned multi-component association graph structure of the bridge. The feature vectors of the components are input into a graph neural network model for processing, and the post-earthquake damage level of the bridge is output.
[0008] Preferably, the acceleration time history data is preprocessed, and the temporal features of each key component are extracted to form a three-dimensional component feature vector, including: The acquired acceleration time history data is preprocessed to have a uniform sampling frequency and duration. A residual convolutional neural network model is constructed, and the preprocessed acceleration time history is input into the residual convolutional neural network model to extract the temporal response features of each component.
[0009] Preferably, the bridge multi-component association diagram structure includes: Nodes are used to represent the key components of a bridge; Edges are used to represent the structural connection relationship or dynamic response correlation between components; the edge weights are determined based on the correlation coefficient between the acceleration responses of the components.
[0010] Preferably, establishing and training a graph neural network model based on the multi-component association graph structure of the bridge includes: Multiple seismic response data from bridges are acquired, and the seismic response data are sequentially subjected to data augmentation and nonlinear time history analysis to generate a training set; The graph neural network model is trained using the training set; The graph neural network model is optimized for performance during training.
[0011] Preferably, the graph neural network model includes three nodes and adopts a two-layer graph convolutional structure with full bidirectional connectivity.
[0012] Preferably, the performance of the graph neural network model is optimized during training, including: Construct a loss function and use the loss function to optimize the training process of the graph neural network model; The cross-entropy loss function is used to determine the difference between the bridge damage level output by the graph neural network model and the actual damage level, and the gradient descent algorithm is used to update the parameters of the graph neural network model.
[0013] Preferably, the key components include piers, abutments, and supports, forming three-dimensional input data.
[0014] Preferably, a rapid bridge seismic damage assessment system includes: The data acquisition module is used to acquire the acceleration time history data of multiple key bridge components under seismic loading. The feature extraction module is used to preprocess the acceleration time history data and extract the time-series response features of each key component to form a three-dimensional component feature vector. The graph structure construction module is used to construct a multi-component association graph structure for bridges based on the structural connection relationships and response correlations between key components. The model building and training module is used to build and train a graph neural network model based on the multi-component association graph structure of the bridge. The damage level output module is used to input the feature vector of the component into the graph neural network model for processing and output the post-earthquake damage level of the bridge.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for rapid assessment of bridge seismic damage, which has the following beneficial effects: 1. By constructing a multi-component relationship diagram structure for bridges, the physical connections and dynamic response relationships between components can be accurately reflected; 2. By utilizing graph neural networks to achieve joint evaluation of damage information from multiple components, the accuracy of identifying minor and moderate damage states is significantly improved; 3. No complex displacement inversion and finite element calculations are required, resulting in high evaluation efficiency and suitability for rapid emergency assessment scenarios after earthquakes. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 This is an overall flowchart of a rapid bridge seismic damage assessment method provided in an embodiment of the present invention.
[0018] Figure 2 A schematic diagram of a deep learning network structure provided for an embodiment of the present invention.
[0019] Figure 3 This is a schematic diagram of the residual convolutional layer structure provided in an embodiment of the present invention.
[0020] Figure 4 This is a schematic diagram of a graph neural network convolutional layer structure provided in an embodiment of the present invention.
[0021] Figure 5 This is a schematic diagram of the confusion matrix results of the model provided in an embodiment of the present invention on a test set of a simply supported beam bridge.
[0022] Figure 6 A comparison chart of prediction results between the integral inversion displacement method provided in this embodiment of the invention and the method (multi-component) provided in this embodiment of the invention.
[0023] Figure 7(a) is a comparison of the evaluation performance using only bridge piers provided in the embodiment of the present invention with the prediction results of the method (multi-component) provided in the embodiment of the present invention.
[0024] Figure 7(b) is a comparison of the evaluation performance using only bridge abutments provided in the embodiment of the present invention with the prediction results of the method (multi-component) provided in the embodiment of the present invention.
[0025] Figure 7(c) is a comparison of the performance of the evaluation using only supports provided in the embodiment of the present invention with the prediction results of the method (multi-component) provided in the embodiment of the present invention. Detailed Implementation
[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] This invention discloses a rapid assessment method for bridge seismic damage, such as... Figure 1 As shown, it includes: S100. Acquire acceleration time history data of multiple key components of the bridge under seismic action, including piers, abutments and bearings, to form three-dimensional input data. S200. The acceleration time history data is preprocessed, and the time-series response features of each key component are extracted to form a three-dimensional component feature vector. S300. Based on the structural connection relationships and response correlations between key components, construct a multi-component association diagram structure for the bridge. S400. Establish and train a graph neural network model based on the multi-component association graph structure of the bridge; S500. Input the feature vector of the component into the graph neural network model for processing, and output the post-earthquake damage level of the bridge.
[0028] This invention uses acceleration response data of key bridge components as input. By constructing a graph structure relating multiple bridge components, a graph neural network model is used to rapidly determine the overall seismic damage status of the bridge. The method includes: acquiring acceleration time-history data of piers, abutments, and supports under seismic loading; preprocessing the acceleration time-history data and extracting time-series features; constructing a graph structure based on the correlation of responses of each component; inputting the component features and graph structure into the graph neural network model for inference, and outputting the seismic damage level of the bridge. This invention can fully integrate damage information from multiple components under conditions of limited sensor quantity, significantly improving the accuracy and efficiency of post-earthquake damage assessment of bridges, and is suitable for post-earthquake emergency assessment and rapid bridge safety determination scenarios.
[0029] In a specific embodiment of the present invention, the specific implementation process of S100 includes: For the bridge to be evaluated, acceleration sensors are deployed at several key components to acquire dynamic response data of the bridge under seismic loading. These key components include piers, abutments, and bearings, wherein: Acceleration sensors are installed on the top of the bridge piers to collect time history data of the bridge piers' acceleration under seismic loads. Acceleration sensors were installed at the bridge abutments to collect time history data of the bridge abutments' acceleration under seismic loads. Acceleration sensors are installed at the supports to collect time history data of the acceleration of the supports under seismic loads.
[0030] When an earthquake occurs, the acceleration sensor synchronously records the acceleration time history data of each key component of the bridge under the seismic action, and combines the acceleration time history data of the piers, abutments and supports in chronological order to form three-dimensional input data containing acceleration information of multiple components.
[0031] In this embodiment of the invention, in order to ensure the data consistency and comparability between different ground motion records, the acquired acceleration time history data is preprocessed.
[0032] The preprocessing process includes: (1) Duration uniformity: The duration of all acceleration records is uniformly truncated or padded with zeros to the same length, such as 30 seconds, to ensure the consistency of the input data dimensions; (2) Sampling frequency standardization: unify the sampling frequency of all data to the same standard, such as 100Hz, in order to avoid frequency differences from interfering with the model; After preprocessing, the acceleration time history data of the piers, abutments and supports are input into the residual convolutional neural network model.
[0033] Specifically, the acceleration time history data is preprocessed, and the temporal features of each key component are extracted to form a three-dimensional component feature vector, including: The acquired acceleration time history data is preprocessed to have a uniform sampling frequency and duration. A residual convolutional neural network model is constructed, which includes an input layer, multiple residual modules, a global pooling layer, and a fully connected layer. The preprocessed acceleration time history is input into the residual convolutional neural network model, and the temporal response features of each component are extracted as node features of the graph neural network.
[0034] Specifically, such as Figure 2 , Figure 3As shown, in a specific embodiment, the residual convolutional neural network model includes an input layer, multiple residual modules, a global pooling layer, and a fully connected layer. Through convolution operations and residual connection structures in the residual modules, temporal features in the acceleration time history data are extracted layer by layer to obtain temporal feature vectors that can characterize the dynamic response features of each component.
[0035] The input layer receives acceleration time history data of key bridge components. This acceleration time history data is a one-dimensional time series with a length of T (where T is the number of data points after standardization). In this embodiment, the acceleration time histories of piers, abutments, and supports are input into the residual convolutional neural network model as independent input channels.
[0036] The residual convolutional neural network model sequentially sets up multiple residual modules, extracts features from acceleration time history data through layer-by-layer convolution and pooling operations, and finally compresses the time-series features into a fixed-dimensional one-dimensional feature vector through adaptive average pooling to characterize the dynamic response features of each bridge component, such as... Figure 4 As shown.
[0037] First, the input acceleration time history data is convolved by a first one-dimensional convolutional layer, and then combined with a batch normalization layer and a non-linear activation function to extract preliminary temporal features. Subsequently, the features are further convolved through a second one-dimensional convolutional layer to enhance the ability to express complex temporal patterns. The convolutional features are added element-wise to the original input features through residual connections to alleviate the gradient vanishing problem that may occur during deep network training and to maintain the effective transfer of low-level features to high-level features. A global pooling layer is set after some residual modules to compress the feature sequence length through downsampling operations, thereby reducing computational complexity and enhancing the model's ability to abstract temporal features.
[0038] Each convolutional layer is followed by a batch normalization layer and a non-linear activation function to improve the stability of model training and feature extraction capabilities.
[0039] After the above processing, a three-dimensional component feature vector containing the features of piers, abutments and supports is formed, which serves as the node feature input for the subsequent graph neural network model.
[0040] Specifically, in embodiment S300 of the present invention, the correlation of key components is summarized from the results of a large number of documents. The correlation coefficient between piers and abutments is 0.87, between piers and supports is 0.86, and between abutments and supports is 0.63.
[0041] Based on the above research results, the embodiments of the present invention use the above correlation coefficients as initial edge features in the process of graph structure construction, so as to more realistically reflect the response correlation between components.
[0042] Specifically, the bridge multi-component association diagram structure includes: Nodes are used to represent the key components of a bridge; Edges are used to represent the structural connection relationship or dynamic response correlation between components; the edge weights are determined based on the correlation coefficient between the acceleration responses of the components.
[0043] Specifically, nodes are used to represent the piers, abutments, and bearing components of a bridge; Edges are used to represent the dynamic response relationships between different components; The edge weights are initialized based on the response correlation coefficients between different types of components and are adaptively updated during model training to reflect changes in the dynamic coupling relationship between components.
[0044] Specifically, such as Figure 2 As shown, a graph neural network model is established and trained based on the multi-component association graph structure of the bridge, including: S410. Based on the established graph structure, construct a graph neural network model; S420. Acquire seismic response data from various sources on the bridge, and sequentially perform data augmentation and nonlinear time history analysis on the seismic response data to generate a training set; S430. Train the graph neural network model using the training set; S440. Optimize the performance of the graph neural network model during training.
[0045] Specifically, S410 includes: (1) Construction of graph neural network input The inputs to the graph neural network include: Node features are used to represent the time-series response feature vectors of bridge piers, abutments, and bearing components; Edge connectivity is used to represent the connection topology between multiple components of a bridge; Edge features are used to characterize the dynamic response correlation strength between different components.
[0046] Each node corresponds to a bridge component, and the node features are extracted by the aforementioned residual convolutional neural network; the edge features are initialized based on the response correlation coefficients between different types of components.
[0047] (2) Graph convolution operation based on attention mechanism
[0048] In this embodiment of the invention, the graph neural network convolutional layer adopts a graph convolutional structure that introduces an attention mechanism to update node features.
[0049] In each layer of graph convolution operation, for any target node, the importance of the adjacent nodes is weighted by aggregating the feature information of its neighboring nodes and combining the corresponding edge features, thereby achieving adaptive fusion of response information of different components.
[0050] Specifically, graph convolutional layers, through joint modeling of node features and edge features, enable the dynamic coupling relationship between piers, abutments, and bearing components to be fully expressed in the feature space.
[0051] (3) Introduction and updating methods of edge features
[0052] In one specific embodiment, the edge features are used to reflect the dynamic response correlation between different types of bridge components.
[0053] For different component combinations such as piers and abutments, piers and supports, and abutments and supports, corresponding edge weight parameters are set to characterize the response coupling strength between different components.
[0054] The edge weights are assigned values based on preset component correlation coefficients during the model initialization phase and are updated as trainable parameters during model training, thereby enabling the graph neural network to adaptively adjust the information transmission weights between different components according to the training samples.
[0055] (4) Feature Iterative Update of Multi-layer Graph Convolution
[0056] In this embodiment, the graph neural network model adopts a multi-layer graph convolutional structure, preferably a two-layer graph convolutional network.
[0057] The first layer of graph convolution is used to achieve preliminary information fusion between bridge components, enabling each component node to perceive the dynamic response characteristics of its adjacent components; The second-layer graph convolution, based on the feature update of the first layer, further integrates information from multiple components to obtain a high-level semantic representation that includes the dynamic features of the entire bridge structure.
[0058] Through iterative updates of multi-layer graph convolution, the features of multiple bridge components are propagated and fused layer by layer on the graph structure.
[0059] (5) Nonlinear activation and feature stabilization processing
[0060] After each layer of graph convolution operation, a nonlinear activation function is introduced to map the node features to enhance the model's ability to express complex nonlinear dynamic features.
[0061] By setting a nonlinear activation function, the graph neural network model can maintain good numerical stability and generalization ability during feature update.
[0062] In one specific embodiment of the present invention, in order to improve the prediction accuracy of the graph neural network model in the rapid assessment of bridge seismic damage, the present invention constructs a bridge seismic damage dataset for training the graph neural network model.
[0063] Specifically, the finite element method is first used to establish a dynamic analysis model of the bridge. The bridge model includes at least key components such as piers, abutments, and bearings, and the mechanical parameters and boundary conditions of each component are reasonably set according to the structural form and material properties of the bridge.
[0064] Subsequently, thousands of ground motion records selected from the KiK-net and K-NET strong earthquake databases were input into the bridge model to perform nonlinear time history analysis on the bridge, obtaining the acceleration time history response data of the piers, abutments, and bearings under seismic loading. Through the above nonlinear time history analysis process, the dynamic response characteristics of the bridge under different seismic loading intensities can be simulated.
[0065] To further expand the number of training samples and alleviate the problem of uneven distribution of samples with different damage levels, this embodiment performs data augmentation processing on the selected original ground motion records. Specifically, by scaling the peak ground acceleration of the original ground motion records, setting multiple different amplification coefficients, the same ground motion record is re-inputted into the bridge model after amplitude adjustment for nonlinear time history analysis.
[0066] Through the above amplitude scaling and repeat analysis process, bridge dynamic response data corresponding to different earthquake intensities can be generated without changing the characteristics of the ground motion spectrum. This allows for the acquisition of bridge response samples covering various post-earthquake damage states, including basically intact, slightly damaged, moderately damaged, and severely damaged.
[0067] Finally, based on the dynamic response results of the piers, abutments and bearings, and combined with the preset component damage criteria, the post-earthquake damage level of the bridge is determined, and the corresponding acceleration time history data and damage level labels are combined to form training samples to construct a bridge earthquake damage dataset for training the graph neural network model.
[0068] Specifically, such as Figure 4 As shown, the graph neural network model includes three nodes and adopts a two-layer graph convolutional structure with full bidirectional connectivity.
[0069] Specifically, the performance of the graph neural network model is optimized during training, including: Construct a loss function and use the loss function to optimize the training process of the graph neural network model; The cross-entropy loss function is used to determine the difference between the bridge damage level output by the graph neural network model and the actual damage level, and the gradient descent algorithm is used to update the parameters of the graph neural network model.
[0070] Specifically, the post-earthquake damage levels of the bridges include: basically intact, slightly damaged, moderately damaged, and severely damaged.
[0071] In one specific embodiment of the present invention, a loss function is constructed to measure the difference between the bridge damage level output by the graph neural network model and the actual damage level. In this embodiment of the present invention, the loss function adopts a multi-class cross-entropy loss function, the expression of which is: ; In the formula, It is the number of samples; It is a sample The real label is ; It is the model on the sample The predicted probability.
[0072] The calculation formula is as follows: ; In the formula, It is the first output of the model Original scores for each category.
[0073] The gradient descent algorithm is used to optimize and update the model parameters. In this embodiment, the adaptive moment estimation optimization algorithm is preferably used to iteratively optimize the model parameters to improve the convergence speed and stability of the model training process.
[0074] For iterative optimization of the parameters of the graph neural network model, the model training process is most stable and the convergence effect is best when the learning rate is set to 0.0001; a Dropout ratio of 0.1 is selected for setting.
[0075] In one specific embodiment, the output post-earthquake damage level of the bridge includes: basically intact, slightly damaged, moderately damaged, and severely damaged.
[0076] In one specific embodiment of the present invention, a rapid bridge seismic damage assessment system includes: The data acquisition module is used to acquire the acceleration time history data of multiple key components of the bridge under seismic loading. The key components include piers, abutments and bearings, forming three-dimensional input data. The feature extraction module is used to preprocess the acceleration time history data and extract the time-series response features of each key component to form a three-dimensional component feature vector. The graph structure construction module is used to construct a multi-component association graph structure for bridges based on the structural connection relationships and response correlations between key components. The model building and training module is used to build and train a graph neural network model based on the multi-component association graph structure of the bridge. The damage level output module is used to input the feature vector of the component into the graph neural network model for processing and output the post-earthquake damage level of the bridge.
[0077] To verify the performance of the method provided in this embodiment of the invention, the graph neural network model proposed in this embodiment of the invention was compared with the integral inversion displacement method and the single component evaluation method on the same experimental platform and dataset. The specific verification process is as follows: See Figure 5 As shown, the graph neural network model has good prediction accuracy for all damage levels (especially basically intact, minor damage and severe damage) on the test set.
[0078] Figure 6 Comparative experiments demonstrate that the integral inversion displacement method (see...) Figure 6 The method for assessing earthquake damage (as shown) has a much lower predictive performance than the method proposed in this embodiment of the invention, which uses multi-component acceleration input (see [example]). Figure 5 (As shown in the figure). Therefore, it is evident that introducing multi-component response information is of great significance for improving the accuracy of earthquake damage assessment.
[0079] Figure 6 Comparative experiments further demonstrate that using only a single-component method (see Figures 7(a)-7(c)) to assess earthquake damage has a significantly lower predictive performance than the method proposed in this embodiment of the invention, which uses multi-component acceleration input (see Figures 7(a)-7(c)). Figure 5 (As shown in the figure). Therefore, it is evident that introducing multi-component response information is of great significance for improving the accuracy of earthquake damage assessment.
[0080] In summary, this invention provides a method and system for rapid assessment of bridge seismic damage based on graph neural networks. By jointly modeling the acceleration time-history data of key bridge components, a multi-component correlation graph structure is constructed. Furthermore, the graph neural network is used to achieve the fusion analysis of dynamic response information between components, effectively solving the problems of insufficient utilization of multi-component information, complex assessment processes, and low accuracy in identifying minor and moderate damage states in existing technologies. This invention enables rapid and accurate assessment of bridge post-earthquake damage states even with a limited number of sensors, providing an efficient and reliable technical means for emergency assessment and safety determination of bridges in engineering practice.
[0081] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0082] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A rapid assessment method for bridge seismic damage, characterized in that, include: Acquire time history data of acceleration of several key bridge components under seismic loading; The acceleration time history data is preprocessed, and the time-series response features of each key component are extracted to form a three-dimensional component feature vector. Based on the structural connection relationships and response correlations among the key components, a multi-component association diagram structure for the bridge is constructed. A graph neural network model was established and trained based on the aforementioned multi-component association graph structure of the bridge. The feature vectors of the components are input into a graph neural network model for processing, and the post-earthquake damage level of the bridge is output.
2. The method for rapid assessment of bridge seismic damage according to claim 1, characterized in that, The acceleration time history data is preprocessed, and the temporal features of each key component are extracted to form a three-dimensional component feature vector, including: The acquired acceleration time history data is preprocessed to have a uniform sampling frequency and duration. A residual convolutional neural network model is constructed, and the preprocessed acceleration time history is input into the residual convolutional neural network model to extract the temporal response features of each component.
3. The method for rapid assessment of bridge seismic damage according to claim 1, characterized in that, The bridge multi-component association diagram structure includes: Nodes are used to represent the key components of a bridge; Edges are used to represent the structural connection relationship or dynamic response correlation between components; the edge weights are determined based on the correlation coefficient between the acceleration responses of the components.
4. The method for rapid assessment of bridge seismic damage according to claim 1, characterized in that, A graph neural network model is established and trained based on the aforementioned multi-component association graph structure of the bridge, including: Multiple seismic response data from bridges are acquired, and the seismic response data are sequentially subjected to data augmentation and nonlinear time history analysis to generate a training set; The graph neural network model is trained using the training set; The graph neural network model is optimized for performance during training.
5. The method for rapid assessment of bridge seismic damage according to claim 1, characterized in that, The graph neural network model includes three nodes and employs a two-layer graph convolutional structure with full bidirectional connectivity.
6. The method for rapid assessment of bridge seismic damage according to claim 4, characterized in that, The performance of the graph neural network model is optimized during training, including: Construct a loss function and use the loss function to optimize the training process of the graph neural network model; The cross-entropy loss function is used to determine the difference between the bridge damage level output by the graph neural network model and the actual damage level, and the gradient descent algorithm is used to update the parameters of the graph neural network model.
7. The method for rapid assessment of bridge seismic damage according to claim 1, characterized in that, The key components include piers, abutments, and supports, forming three-dimensional input data.
8. A rapid bridge seismic damage assessment system, employing the rapid bridge seismic damage assessment method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the acceleration time history data of multiple key bridge components under seismic loading. The feature extraction module is used to preprocess the acceleration time history data and extract the time-series response features of each key component to form a three-dimensional component feature vector. The graph structure construction module is used to construct a multi-component association graph structure for bridges based on the structural connection relationships and response correlations between key components. The model building and training module is used to build and train a graph neural network model based on the multi-component association graph structure of the bridge. The damage level output module is used to input the feature vector of the component into the graph neural network model for processing and output the post-earthquake damage level of the bridge.