Post-earthquake bridge damage identification method, device and equipment and storage medium

Abstract

This invention provides a method, apparatus, device, and storage medium for post-earthquake bridge damage identification, relating to the field of damage identification technology. The method includes: acquiring bridge images after an earthquake using a drone, performing image recognition on the bridge images to obtain crack mask images of the post-earthquake bridge; extracting the skeleton from the crack mask images to obtain crack parameters of the post-earthquake bridge; inputting the structural parameters and crack parameters of the post-earthquake bridge into a deep neural network surrogate model to obtain the seismic response of each key component of the post-earthquake bridge; determining the damage evaluation result of each component based on the seismic response of each key component of the post-earthquake bridge, and determining the overall damage evaluation result of the post-earthquake bridge based on the damage evaluation results of each component. This invention combines drone visual perception, deep neural network surrogate prediction, and multi-source information fusion to achieve accurate and rapid assessment of post-earthquake bridge damage.

Description

Technical Field

[0001] This invention relates to the field of damage identification technology, and in particular to a method, apparatus, equipment and storage medium for identifying damage to bridges after an earthquake. Background Technology

[0002] Bridges are core hubs of transportation networks, and their post-earthquake structural condition directly determines the efficiency of emergency rescue and the overall progress of post-disaster reconstruction. Under earthquake action, bridge damage manifests simultaneously as visible surface cracks and internal, hidden structural stress deterioration; these two types of damage jointly determine the bridge's safety performance and load-bearing capacity. Rapid, accurate, and comprehensive damage diagnosis within a short period after an earthquake is a crucial prerequisite for making emergency access decisions and allocating repair resources.

[0003] In existing technologies, when assessing bridge damage after an earthquake, traditional manual inspections are time-consuming, labor-intensive, dangerous, and difficult to scale up, failing to meet the rapid inspection needs of a large number of bridges in a short period after an earthquake. Furthermore, the inspection results rely on the subjective experience of engineers, resulting in insufficient quantitative accuracy. Therefore, existing technologies utilize high-precision finite element simulation to model bridges and perform bridge damage analysis, or use drones for visual inspection to collect bridge images and achieve bridge damage detection.

[0004] However, finite element simulation is costly and time-consuming, making it difficult to conduct batch real-time assessments. While UAV visual inspection can quickly extract the geometric features of cracks on the bridge surface, it is difficult to infer the true damage state of the bridge's internal structure. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for identifying post-earthquake bridge damage, addressing the problems of low efficiency in post-earthquake bridge damage assessment, one-sided assessment using a single detection method, inability to effectively integrate heterogeneous data, and difficulty in handling uncertainties in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a method for identifying damage to bridges after an earthquake, including: Using drones to collect images of bridges after an earthquake, and performing image recognition on the bridge images, we obtained crack mask images of the bridges after the earthquake. Skeleton extraction was performed on the crack mask image to obtain the crack parameters of the bridge after the earthquake. The structural and crack parameters of the bridge after the earthquake are input into a deep neural network surrogate model to obtain the seismic response of each key component of the bridge after the earthquake; the deep neural network surrogate model is trained based on the seismic simulation data of the bridge after the earthquake. Based on the seismic response of each key component of the bridge after the earthquake, the damage assessment results of each component are determined, and the damage assessment results of the bridge after the earthquake are determined based on the damage assessment results of each component.

[0007] Secondly, embodiments of the present invention provide a post-earthquake bridge damage identification device, comprising: The acquisition module is used to acquire images of bridges after an earthquake using drones, and to perform image recognition on the bridge images to obtain crack mask images of the bridges after the earthquake. The extraction module is used to extract the skeleton from the crack mask image to obtain the crack parameters of the bridge after the earthquake. The prediction module is used to input the structural parameters and crack parameters of the bridge after the earthquake into the deep neural network surrogate model to obtain the seismic response of each key component of the bridge after the earthquake; the deep neural network surrogate model is trained based on the seismic simulation data of the bridge after the earthquake. The evaluation module is used to determine the damage evaluation result of each key component of the bridge after the earthquake, based on the seismic response of each component, and to determine the damage evaluation result of the bridge after the earthquake based on the damage evaluation result of each component.

[0008] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0009] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0010] In this embodiment of the invention, images of the bridge after an earthquake are collected by a UAV and image recognition is performed, which can quickly obtain fine contour information of the apparent cracks in the bridge after the earthquake. By extracting the skeleton from the crack mask image and calculating the crack parameters, the pixel-level crack contours can be transformed into quantitative crack length and width indicators that can be directly used for engineering judgment, providing standardized input for subsequent fusion analysis of apparent damage. By inputting the crack parameters and structural parameters into a pre-trained deep neural network surrogate model, the seismic response of key components such as piers, bearings, and abutments can be obtained without relying on time-consuming nonlinear finite element iterative calculations. This retains the mechanical rigor of finite element analysis and compresses the response prediction time from minutes to milliseconds. Furthermore, the damage evaluation results are determined based on the seismic response of key components, realizing a progressive judgment from local component damage to the overall damage level of the bridge. The evaluation results simultaneously cover apparent defects and internal structural responses, effectively overcoming the limitations of a single information source. This organic combination of UAV visual perception, deep neural network surrogate prediction, and multi-source information fusion enables accurate and rapid assessment of the damage to bridges after an earthquake, providing a reliable basis for the rapid repair of bridges after an earthquake. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the implementation of the post-earthquake bridge damage identification method provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the UAV data acquisition process for the post-earthquake bridge damage identification method provided in this embodiment of the invention; Figure 3 This is a structural diagram of the two-stage system architecture model of the post-earthquake bridge damage identification method provided in this embodiment of the invention; Figure 4 This is a schematic diagram of the crack skeleton extraction process in the post-earthquake bridge damage identification method provided in this embodiment of the invention; Figure 5a This is a pixel traversal diagram of the post-earthquake bridge damage identification method provided in an embodiment of the present invention; Figure 5b This is a schematic diagram of the positional relationship between pixels in the eight-neighborhood of the post-earthquake bridge damage identification method provided in this embodiment of the invention. Figure 6 This is a structural diagram of the deep neural network surrogate model of the post-earthquake bridge damage identification method provided in the embodiments of the present invention; Figure 7 This is a flowchart illustrating the implementation of step S140 of the post-earthquake bridge damage identification method provided in this embodiment of the invention. Figure 8 This is a schematic diagram of the structure of the post-earthquake bridge damage identification device provided in an embodiment of the present invention; Figure 9This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0012] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0013] See Figure 1 The document illustrates a flowchart of the post-earthquake bridge damage identification method provided in an embodiment of the present invention, which is described in detail below: Step S110: Use a drone to collect images of the bridge after the earthquake, and perform image recognition on the bridge images to obtain crack mask images of the bridge after the earthquake.

[0014] In some embodiments, a post-earthquake bridge refers to an existing bridge structure that requires damage assessment after experiencing an earthquake. A bridge image refers to a high-resolution digital image of the surface of key components such as the bridge deck, beam underside, and piers, captured on-site after an earthquake by a camera device mounted on a drone. This image contains information on the spatial distribution and morphology of apparent cracks on the bridge surface. A crack mask image is a binary image output after image recognition of the original bridge image. In this image, pixels belonging to crack areas are marked as foreground (e.g., set to white), and pixels not belonging to crack areas are marked as background (e.g., set to black), thereby separating the cracks from the complex background of the bridge surface. Crack mask images can accurately depict the contour, direction, and spatial distribution of each crack, providing fundamental data for subsequent quantitative calculations of crack parameters.

[0015] It should be noted that during the data collection process, such as... Figure 2 As shown, an unmanned aerial vehicle (UAV) equipped with a 24-megapixel three-axis gimbal camera is required to perform aerial photography, focusing on the bridge deck, beam bottom, and upper pier areas. The flight altitude is set at 5m, the flight speed at 1.5m / s, the forward overlap rate at 85%, and the lateral overlap rate at 65%. Ground control points are deployed to calibrate the camera and perform pixel-to-physical size calibration. The UAV will then collect images covering all key bridge components according to a pre-defined flight path, ensuring the images are free from overexposure, blurring, and shadow occlusion, meeting the requirements for subsequent crack detection and quantification. Furthermore, the UAV's flight parameters need to be standardized, including flight altitude, flight speed, image overlap rate, and camera and gimbal parameters. This standardization includes limiting key flight parameters to experimentally verified effective ranges (fixing flight altitude and speed, image resolution, and overlap rate parameters), standardizing equipment configuration (calibrating camera internal and external parameters, uniformly setting ground control points for pixel-to-physical size calibration), and standardizing the acquisition process (determining the starting point, executing the flight mission according to the pre-defined flight path, and archiving the data).

[0016] In one possible implementation, the specific processing method of the above step S110 is as follows: target detection is performed on the bridge image to obtain the crack region contained in the bridge image; pixel segmentation is performed on the crack region to obtain the crack mask image of each crack in the crack region.

[0017] In some embodiments, object detection refers to the process of locating image regions that may contain cracks from an entire bridge image. Since bridge images captured by drones typically contain various elements such as the bridge deck, piers, beams, and background environment, cracks only occupy localized areas in the image. Directly segmenting the entire image at the pixel level would result in high computational cost and susceptibility to background interference. Object detection can quickly filter out candidate regions containing cracks from the entire image, eliminating a large amount of irrelevant background and focusing subsequent fine-grained segmentation computational resources on the truly relevant local regions. Object detection can be implemented using a two-stage system architecture model, the structure of which is shown in the diagram below. Figure 3 As shown, the two-stage system architecture model, after being trained with pre-labeled crack image samples, can automatically output the coordinates of the localization box of the crack region in the bridge image.

[0018] In some embodiments, such as Figure 3 As shown, in the two-stage system architecture model, the object detection model first performs initial localization of the crack region, and then the segmentation model performs pixel-level fine segmentation within the localized bounding box. Together, they constitute a complete crack detection and segmentation model. When training the two-stage system architecture model, a supervised learning strategy is required, with parameter optimization based on labeled crack image samples. In the two-stage system architecture model, during object detection and pixel segmentation, the joint loss of bounding box regression and classification is used as the optimization objective. The segmentation unit in the two-stage system architecture model uses the joint loss function of the Dice coefficient and cross-entropy as the optimization objective. During training, the Adam optimizer and gradient descent algorithm are used, with a batch size of four to 16, and an early stopping strategy is introduced to prevent overfitting, thereby ensuring effective convergence of the model during training.

[0019] It's important to note that pixel segmentation refers to the process of classifying each pixel within a crack region determined by object detection as either a crack or background. Unlike object detection, which only provides coarse localization at the bounding box level, pixel segmentation achieves pixel-level classification accuracy, accurately delineating the contour boundaries of each crack. Pixel segmentation can be implemented using image segmentation models based on an encoder-decoder structure, such as U-Net and its variants. These models extract high-level semantic features of the image through downsampling paths, restore the spatial resolution of the feature map through upsampling paths, and fuse shallow details and deep semantic information through skip connections, thereby achieving accurate pixel classification while preserving the fine detail of crack edges. After pixel segmentation, the output is a binary mask image corresponding to the input image region, where pixels within the crack region are labeled as foreground and pixels in the background region are labeled as background, thus obtaining the crack mask image for each crack.

[0020] By employing a two-stage processing approach—first object detection and then pixel segmentation—high-precision pixel-level extraction of cracks in bridge images can be achieved while significantly reducing computational overhead. This avoids the computational redundancy associated with directly segmenting the entire image and overcomes the limitation of relying solely on object detection to obtain the fine contours of cracks. The crack mask image serves as the foundational input for subsequent skeleton extraction and crack parameter calculation; its accuracy directly impacts the accuracy of quantitative indicators such as crack length and width. Therefore, ensuring the quality of the mask image through two-stage recognition provides a reliable data source for the entire damage identification process.

[0021] Step S120: Extract the skeleton from the crack mask image to obtain the crack parameters of the bridge after the earthquake.

[0022] In some embodiments, while the crack mask image accurately depicts the contour region of the crack, the width of the crack region typically comprises multiple pixels. Directly calculating the crack length based on the mask image can introduce significant errors due to the interference of the width factor. Skeleton extraction refers to the process of extracting the crack centerline, which has only a single pixel width, from the crack region of the crack mask image; this is the crack skeleton processing. The crack skeleton preserves the crack's topology, extension direction, and branching relationships, while eliminating the influence of crack width on length calculation, making the quantification of crack length more accurate.

[0023] It should be noted that crack parameters refer to quantifiable indicators characterizing the geometric features of cracks, including at least crack length and crack width. Crack length reflects the extent of crack extension on the surface of bridge components and is one of the important indicators for judging the severity of cracks; crack width reflects the degree of crack opening and is closely related to the structural stress state and durability. Crack parameters provide standardized quantitative inputs for subsequent multi-source fusion with structural response information, thus providing a basis for quantifying apparent damage.

[0024] In one possible implementation, the crack parameters include crack width and crack length. Based on this, step S120 specifically involves: iteratively identifying each pixel in the crack mask image using a preset iterative algorithm to extract the crack skeleton from the crack mask image; calculating the crack length of the bridge after the earthquake based on the number of pixels in the crack skeleton, the object distance of the bridge image, the focal length of the camera on the UAV, and the sampling distance between the UAV and the bridge after the earthquake; and calculating the crack width of the bridge after the earthquake based on the minimum distance from the crack skeleton to the boundary of the crack area, the object distance of the bridge image, the focal length of the camera on the UAV, and the sampling distance between the UAV and the bridge after the earthquake.

[0025] In some embodiments, the extraction process of the fracture skeleton is as follows: Figure 4 As shown, the preset iterative algorithm is a refinement algorithm that peels away the boundary pixels of the crack region layer by layer through repeated iterations while maintaining the topological connectivity of the crack. In each iteration, the algorithm traverses every foreground pixel (i.e., the pixel belonging to the crack region) in the crack mask image, such as... Figure 5a As shown, taking this pixel as the center pixel, we analyze the value patterns of each pixel in its eight neighboring regions. Based on preset deletion conditions, we determine whether this pixel belongs to the deletable boundary redundant pixels. The positional relationships between the pixels in the eight neighboring regions are as follows: Figure 5b As shown. The design of the deletion condition needs to satisfy two constraints simultaneously: deleting the pixel must not disrupt the topological connectivity of the crack region, i.e., it cannot cause the originally connected cracks to break; deleting the pixel must not shorten the endpoints of the crack, i.e., it cannot change the position of the extended end of the crack. Through multiple iterations, redundant pixels are stripped layer by layer from the perimeter of the crack region inwards until no pixel satisfies the deletion condition after one round of traversal, at which point the iteration terminates. The remaining foreground pixels at this point constitute the crack skeleton with a width of one pixel. The iterative algorithm can adopt the Zhang-Suen thinning algorithm. By extracting the skeleton through iterative thinning, it is possible to adaptively handle crack regions of different widths and shapes without relying on manual annotation, and stably output a single-pixel skeleton that maintains the original crack topological structure, laying the foundation for the subsequent accurate calculation of crack length.

[0026] After extracting the crack skeleton, the number of pixels in the image space needs to be converted into physically meaningful actual dimensions. Since the images captured by the UAV are digital images in pixels, while engineering judgments require actual length and width in millimeters or centimeters, a mapping relationship from pixel space to physical space needs to be established. This mapping relationship is established using Ground Sampling Distance (GSD), which represents the actual physical size corresponding to a single pixel in the image. Its value is determined by the focal length of the camera and the object distance at the time of shooting. Specifically, given the focal length of the camera and the object distance, the GSD value can be calculated, reflecting the actual physical length corresponding to a unit pixel. Based on this, the formula for calculating the crack length is: The formula for calculating the crack width is: in, This represents the number of pixels in the skeleton. For object distance, Focal length The minimum distance from the skeleton to the boundary. This represents the ground sampling distance.

[0027] By employing the methods described above, extracting the crack skeleton using an iterative algorithm minimizes the interference of crack width on length calculation, making length measurement more accurate. Furthermore, by establishing a mapping relationship between pixel space and physical space, crack length and width can be transformed from dimensionless pixel quantities into engineering parameters with definite physical units, allowing them to be directly used for subsequent damage level determination. Accurate quantification of crack parameters is a prerequisite for achieving the fusion analysis of apparent damage and internal response; the quality of the quantification results directly affects the reliability of the final damage evaluation results.

[0028] Step S130: Input the structural parameters and crack parameters of the bridge after the earthquake into the deep neural network surrogate model to obtain the seismic response of each key component of the bridge after the earthquake; wherein, the deep neural network surrogate model is trained based on the seismic simulation data of the bridge after the earthquake.

[0029] In some embodiments, structural parameters refer to a set of parameters describing the mechanical properties of the bridge structure itself, including the cross-sectional shape and material properties of the superstructure, the geometric dimensions and reinforcement information of the piers, the type and mechanical parameters of the bearings, the structural form of the abutments, and the foundation type. These parameters determine the dynamic response characteristics and damage evolution of the bridge under seismic loading. Critical components refer to the components of the bridge that are critically stressed and prone to damage under seismic loading, including at least piers, bearings, and abutments. The seismic response of piers can be expressed as indicators such as curvature ductility; the seismic response of bearings can be expressed as longitudinal and lateral displacements; and the seismic response of abutments can be expressed as active, passive, and lateral displacements. These seismic response indicators of critical components are the core basis for judging the overall damage level of the bridge.

[0030] It should be noted that the structure diagram of the deep neural network surrogate model is as follows: Figure 6 As shown, the deep neural network surrogate model is a surrogate model constructed using deep neural networks to approximate and replace traditional nonlinear finite element time history analysis. Traditional finite element time history analysis establishes a refined finite element model of the bridge, inputs the seismic ground motion time history, and solves the dynamic response of the structure by step-by-step integration in the time domain. Although it has high accuracy, it is computationally extremely time-consuming and cannot meet the needs of rapid post-earthquake assessment. The deep neural network surrogate model learns from a large amount of sample data generated by finite element time history analysis, fitting a complex nonlinear mapping relationship between structural parameters and seismic ground motion parameters and the seismic response of key components. Thus, while maintaining approximate accuracy, it significantly reduces the prediction time of a single response from minutes to milliseconds. A deep neural network surrogate model can be constructed using the ReLU activation function and the Adam optimizer, with early stopping enabled to prevent overfitting. The training set to test set ratio is set to 8:2, and 10-fold cross-validation is used to complete model training. For example, after inputting structural parameters and crack parameters into the trained deep neural network surrogate model, the following prediction results can be output: pier curvature ductility 2.81, longitudinal displacement of bearing 43.6 mm, lateral displacement of bearing 38.2 mm, active displacement of abutment 9.1 mm, passive displacement of abutment 57.3 mm, and lateral displacement of abutment 10.5 mm; the model determination coefficient R0 is [missing value]. 2 >0.93, prediction time ≤0.01 seconds.

[0031] In one possible implementation, before proceeding to step S130, the deep neural network surrogate model needs to be trained. The training process is as follows: multiple sets of preset seismic parameters are input into a pre-constructed finite element model of the post-earthquake bridge. Based on the finite element model and the seismic parameters, nonlinear time history analysis is performed on the post-earthquake bridge to obtain the seismic response of the key components of the post-earthquake bridge under each set of preset seismic parameters. The multiple sets of preset seismic parameters are used as input to the deep neural network surrogate model, and the seismic response of the key components of the post-earthquake bridge under each set of preset seismic parameters is used as output to the deep neural network surrogate model. The deep neural network surrogate model is iteratively updated to obtain the deep neural network surrogate model.

[0032] In some embodiments, the finite element model refers to a numerical model established in a finite element analysis platform based on the structural parameters of the bridge. This model at least covers the superstructure, piers, bearings, abutments, and foundations, and incorporates material nonlinearities such as compressive damage and tensile cracking in the concrete constitutive model, and elastoplastic constitutive properties of the reinforcing steel, as well as geometric nonlinearities such as large deformation effects. Seismic parameters refer to quantitative indices used to describe the characteristics of seismic motion, including at least peak ground acceleration (PGA), peak ground velocity (PGV), and Arias strength. Nonlinear time history analysis refers to solving the dynamic response of the structure under seismic loading by stepwise integration in the time domain. Because it considers material and geometric nonlinear effects, this analysis can realistically reflect the damage accumulation and stiffness degradation process of the bridge during earthquakes.

[0033] In acquiring sample data, firstly, a parametric nonlinear finite element model of the bridge was established on the OpenSees platform. The model needed to cover key components such as the superstructure, piers, bearings, abutments, and foundations, and incorporate material nonlinearities such as concrete damage and steel yielding, as well as geometric nonlinearities such as large deformation. Parametric designation means defining key structural parameters such as pier height, cross-sectional dimensions, reinforcement ratio, and bearing stiffness as variables to facilitate the subsequent batch generation of models with different parameter combinations. Subsequently, 160 natural ground motion records were selected from publicly available databases such as the PEER NGA strong earthquake database, covering 11 seismic intensity indices, including peak ground acceleration, peak ground velocity, Arias intensity, spectral intensity, acceleration spectral intensity, velocity spectral intensity, spectral acceleration, root mean square acceleration, root mean square velocity, characteristic intensity, and duration. Next, each set of ground motions was input into the parametric finite element model, and nonlinear time history analysis was performed. After each analysis, preset seismic response indices for key components were extracted from the calculation results and paired with the corresponding bridge structural parameters and ground motion indices for storage, forming a complete sample record.

[0034] The seismic parameters include bridge structural parameters and ground motion parameters. The bridge structural parameters include the parameters in claim 6, specifically including 50 features such as the main beam cross-sectional shape, dimensions, material strength, pier height, cross-sectional dimensions, longitudinal reinforcement ratio, stirrup spacing, concrete strength, bearing type, stiffness, damping, abutment type, dimensions, backfill parameters, and ground motion parameters. The seismic response includes six response features: pier curvature ductility, bearing longitudinal and lateral displacement, abutment active displacement, passive displacement, and lateral displacement.

[0035] After the sample library is constructed, it is used to train the deep neural network. The deep neural network consists of an input layer, multiple hidden layers, and an output layer. The hidden layers use non-linear activation functions (such as the ReLU function) to give the network the ability to fit complex non-linear mappings. During training, the input part from the sample library is fed into the deep neural network. After forward propagation, the network obtains the predicted output. The predicted output is compared with the true output in the sample library, and the loss function (such as mean squared error loss) is calculated. The loss gradient is backpropagated layer by layer using the backpropagation algorithm, and the network's weight parameters and bias parameters are updated using an optimizer (such as the Adam optimizer). The above process of forward propagation, loss calculation, backpropagation, and parameter update is repeated until the loss function value converges to a preset threshold or a preset number of training epochs are reached. To prevent overfitting, an early stopping strategy can be introduced during training, that is, training is terminated early when the performance on the validation set no longer improves. The final deep neural network surrogate model can approximate the input-output mapping relationship of the finite element model with extremely high accuracy, and the coefficient of determination can reach above 0.93.

[0036] A training sample library was constructed using large-scale nonlinear time history analysis with a finite element model, retaining the advantages of the finite element method, such as rigorous mechanical logic and the ability to accurately reflect the nonlinear response of structures. Based on this, a deep neural network surrogate model was trained, transforming the time-consuming nonlinear time history analysis into rapid network forward propagation calculations. This reduced the time required for earthquake response prediction from minutes to milliseconds, meeting the time requirements for rapid assessment of large-scale bridge clusters after earthquakes. Furthermore, the surrogate model's input includes both bridge structural parameters and crack parameters, allowing apparent crack information to be used as a constraint in the inference of structural response, thus making the prediction results closer to the actual damage state of the bridges.

[0037] Step S140: Based on the seismic response of each key component of the bridge after the earthquake, determine the damage assessment result of each component, and determine the damage assessment result of the bridge after the earthquake based on the damage assessment result of each component.

[0038] In some embodiments, since a single numerical value is difficult to directly correlate with the damage to the project for damage level determination, it is necessary to map the response value of each component to the damage evaluation result at each damage level. The damage evaluation result refers to the degree of membership of each key component at different damage levels, reflecting the degree of conformity between the current damage state of the component and the damage standards at each level. The preset damage levels can be formulated according to the relevant highway bridge seismic performance evaluation rules and technical condition assessment standards, for example, divided into five levels: D0 (no damage), D1 (minor damage), D2 (moderate damage), D3 (severe damage), and D4 (complete failure), as shown in Table 1. After determining the damage evaluation result of each key component, it is also necessary to integrate the evaluation results of each component to obtain the overall damage evaluation result of the bridge, that is, the final damage level of the bridge.

[0039] Table 1. Standards for Classifying Five Levels of Bridge Damage After an Earthquake like Figure 7 As shown, the specific processing method of step S140 includes steps S1401-S1404, and the specific content is as follows: Step S1401: Obtain multiple initial evaluation results corresponding to the seismic response of each key component of the bridge after the earthquake.

[0040] In some embodiments, the membership thresholds for post-earthquake bridge damage indices are shown in Table 2. Initial evaluation results refer to multiple evaluation conclusions obtained after multiple experts or various judgment criteria independently determine the damage level of each key component. For the same key component, different experts may give different damage levels based on their own experience and judgment. For example, if the longitudinal displacement of a support is measured to be 100 mm, some experts may consider it minor damage (D1), some moderate damage (D2), and others severe damage (D3). By statistically analyzing multiple initial evaluation results, the probability distribution of the key component belonging to each damage level can be obtained, i.e., the membership row vector.

[0041] Table 2. Thresholds for the Level Membership Degree of Bridge Damage Indicators After Earthquake It should be noted that regarding the indicator of bridge deck crack length, if it is at the boundary point, some experts may believe that according to the specifications, it should indeed be classified as minor damage. However, some experts believe that drone measurements may have errors, and these errors should be excluded before determining no damage. In addition, there may be situations where the measured data are within a range. For example, the minimum width observed by the drone is 0.2mm, the maximum width is 1.5mm, and the average is 1.8mm, which may cover the level of no damage. Therefore, due to the subjectivity of experts, the assessment level will vary.

[0042] Step S1402: Based on the arrangement order of each key component and the multiple initial evaluation results corresponding to the seismic response of each key component, obtain the membership matrix of the bridge after the earthquake; wherein, the number of rows in the membership matrix is ​​the number of key components, and the number of columns in the membership matrix is ​​the number of preset damage levels.

[0043] In some embodiments, the membership matrix is ​​a matrix formed by stacking the membership row vectors of all key components in a preset order. Each row of the matrix corresponds to a key component, and each column corresponds to a damage level. Each element in the matrix represents the degree to which the key component corresponding to a certain row belongs to the damage level corresponding to a certain column. The membership matrix is ​​the basis for subsequent fuzzy comprehensive evaluation, as it systematically summarizes the damage information of all key components.

[0044] In one possible implementation, the specific processing method of step S1402 is as follows: For any key component, based on the initial evaluation result of the key component, determine the membership degree row vector of the key component; wherein, the number of columns of the membership degree row vector is the number of damage levels; each value in the membership degree row vector is the ratio of the number of each damage level in multiple initial evaluation results to the number of initial evaluation results; the membership degree row vectors of each key component are spliced ​​together according to the arrangement order of each key component to obtain the membership degree matrix of the post-earthquake bridge.

[0045] For example, suppose 10 experts independently evaluate the critical component of the longitudinal displacement of the bearing. Of these 10 experts, 2 classify the damage level as D1 (minor damage), 7 as D2 (moderate damage), 1 as D3 (severe damage), and none as D0 (no damage) or D4 (complete failure). The membership row vector corresponding to this critical component would then be [0, 0.2, 0.7, 0.1, 0], where 0.2 = 2 / 10, 0.7 = 7 / 10, and 0.1 = 1 / 10. Similarly, membership row vectors are constructed for each critical component, such as pier curvature ductility, bearing lateral displacement, and abutment displacement. All these row vectors are then stacked from top to bottom according to the pre-defined critical component arrangement order, resulting in the membership matrix shown in Table 3, which includes the bridge deck crack width, bearing longitudinal displacement, and bearing lateral displacement.

[0046] Table 3. Membership Matrix Diagram By constructing a membership degree row vector using expert voting statistics, the subjective judgment differences of individual experts can be quantified into a probability distribution. This not only preserves the engineering judgment value contained in the expert's experience and knowledge, but also suppresses the bias caused by extreme individual judgments through statistical averaging, so that the membership degree matrix can more objectively reflect the true damage status of each key component.

[0047] Step S1403: Based on the arrangement order of each key component, perform hierarchical analysis on each key component to obtain the weight matrix of the post-earthquake bridge; wherein, the weight matrix has 1 row and the number of columns is the number of key components.

[0048] In some embodiments, the weight matrix is ​​used to represent the importance of each key component, as determined by the analytic hierarchy process (AHP), to the overall damage assessment of the bridge. In the AHP, a pairwise comparison judgment matrix is ​​first constructed between key components. Experts score the importance of each pair of key components based on engineering experience (e.g., using a 1-9 scale). The judgment matrix is ​​then normalized and its eigenvectors are calculated. After passing a consistency check, the weight of each key component is obtained. Each element in the weight matrix represents the proportion of weight of the corresponding key component in the overall damage assessment; a larger weight indicates a greater impact of the component's damage state on the overall damage level of the bridge.

[0049] In one possible implementation, step S1403 is specifically processed as follows: Multiple sets of importance scores for any two key components are obtained; where the importance score is the evaluation result obtained after pairwise comparison of any two key components; a judgment matrix is ​​constructed based on the importance scores of any two key components; where the number of rows and columns of the judgment matrix is ​​the number of key components; each element in the judgment matrix is ​​the importance score of the two key components corresponding to that element's row and column; each judgment matrix is ​​normalized, and the geometric mean of each element in all judgment matrices is calculated, and the geometric mean of each element is determined as the weight of the key component in each row; the weights of each key component are arranged according to the order of the key components to obtain the weight matrix of the post-earthquake bridge.

[0050] In some embodiments, the importance score for any two key components in each group refers to the quantitative scoring by the same expert on the relative importance of any two key components in affecting the overall damage level of the bridge. For example, Expert A has a scoring table: 1 point: both are equally important; 3 points: the former is slightly more important than the latter; 5 points: the former is significantly more important than the latter; 7 points: the former is strongly more important than the latter; 9 points: the former is extremely more important than the latter; 2, 4, 6, 8 points: between the above (e.g., feeling slightly more important than "slightly" but not quite "significantly" important, so a score of 4 is given). For example, as shown in Table 4, if Expert A considers the pier to be "significantly more important" than the abutment, then 5 can be entered at the intersection of the row corresponding to the pier and the column corresponding to the abutment in the judgment matrix; at this time, the importance of the abutment compared to the pier is the reciprocal of the importance of the pier compared to the abutment, so 1 / 5 needs to be entered at the intersection of the row corresponding to the abutment and the column corresponding to the pier. Through such pairwise comparisons, a complete judgment matrix can be constructed. When multiple experts are simultaneously scoring importance, a judgment matrix needs to be obtained based on each expert's score. For example, if three experts are scoring importance, three judgment matrices should be obtained.

[0051] Table 4. Importance Scoring Indicator It should be noted that after the judgment matrix is ​​constructed, it needs to be normalized to calculate the weight vector. The specific normalization method is shown in Table 5. During normalization, each column of the judgment matrix is ​​summed, and then each element in that column is divided by the sum of that column to obtain the normalized matrix. The average of each row of the normalized matrix is ​​then calculated to obtain the initial weight vector: W = [0.723, 0.193, 0.084]. To eliminate potential biases from individual expert judgments, it is usually necessary to collect judgment matrices from multiple experts and take the geometric mean of the initial weight vectors calculated by each expert to obtain the comprehensive weight vector. Furthermore, to ensure the logical consistency of expert judgments, the consistency ratio of the judgment matrix needs to be calculated. When the consistency ratio is less than 0.1, the consistency of the judgment matrix is ​​considered acceptable, and the resulting weight vector is valid.

[0052] Table 5. Normalization Illustration It should be noted that after obtaining the weight vector, it is also necessary to calculate the CR value of the weight vector to determine whether there are any scoring contradictions and to prove the validity of the scoring. In calculations, it is necessary to first calculate... The largest eigenvalue corresponding to the eigenvector. The RI can be obtained by looking up the RI value table shown in Table 6.

[0053] Here, CR is the consistency ratio, which is the ratio of the consistency index CI to the random consistency index RI. This is the final index used to determine whether the matrix passes the test. CI is the consistency index, which is used to quantify the degree of inconsistency of the matrix. n is the order of the matrix, i.e. the number of indices. The larger the CI value, the worse the consistency of the matrix. RI is the random consistency index, which is the average consistency index calculated from a large number of randomly generated judgment matrices. It is related to n and can be obtained by looking up a table. , where is the largest eigenvalue, is the largest eigenvalue of the judgment matrix A, used to measure the deviation of the judgment matrix from the complete consistency matrix, and is only needed for calculating CI; A is the judgment matrix, a square matrix constructed by experts through pairwise comparisons of various indicators such as crack length and support displacement, and the element a in the matrix is ​​. ij The value of indicator i is relative to that of indicator j; W is the weight vector, which is obtained by calculating and normalizing the judgment matrix A, and each element represents the weight of the corresponding indicator in the final damage assessment.

[0054] Table 6 RI Values If CR < 0.1, the consistency check passes and the weight matrix is ​​valid.

[0055] By determining the weight matrix using the analytic hierarchy process (AHP), the qualitative experience and knowledge of experts can be transformed into quantifiable weight parameters. Furthermore, the consistency check mechanism effectively identifies contradictory scoring situations, ensuring the logical rationality of the weight allocation. At the same time, the multi-expert approach effectively suppresses individual bias, resulting in a weight matrix with high credibility and engineering applicability.

[0056] Step S1404: The damage level corresponding to the maximum value of each element in the vector product of the weight matrix and the membership matrix is ​​determined as the damage evaluation result of the bridge after the earthquake.

[0057] In some embodiments, the weight matrix is ​​multiplied by the membership matrix to obtain a comprehensive membership vector. The dimension of this vector is 1 × the number of damage levels, where each element represents the degree to which the bridge as a whole belongs to the corresponding damage level. The damage level corresponding to the maximum value in the comprehensive membership vector is then selected as the final damage evaluation result of the bridge. For example, for D1 level damage, bi = 0.1 × 0.2 + 0.2 × 0.2 + 0.15 × 1 + … = 0.25, which leads to B = [0.16, 0.25, 0.28, 0.23, 0.08]. If B(D2) is the largest, the bridge is determined to have moderate damage.

[0058] It should be noted that by controlling the weights to be perturbed within ±30%, it was found that 99.5% of the damage assessment results remained unchanged after the perturbation. When perturbing the weight matrix, other weights were kept constant, and a single weight was gradually increased or decreased in steps of 5% within ±30% of its original value. After each adjustment, the remaining weights were scaled back to their original proportions to satisfy the constraint that the sum is 1. It was observed whether the position of the maximum value of the membership vector changed when the weight changed within its allowable range. If all indicators did not change the judgment level under extreme boundary conditions, it proves that the damage assessment method provided in this application is sufficiently stable.

[0059] By using the above method, the seismic response of each key component is transformed into membership degrees that can be used for fuzzy comprehensive evaluation. The weight of each component is determined by the analytic hierarchy process, which can effectively handle the fuzziness of expert judgment and cognitive uncertainty in post-earthquake assessment. By performing a vector product operation between the weight matrix and the membership matrix, the information of all key components can be integrated under a unified framework. This allows the final output of the overall bridge damage level to reflect both the apparent crack characteristics and the internal structural response characteristics, overcoming the shortcomings of the one-sided judgment of a single information source and improving the comprehensiveness and robustness of the damage assessment results.

[0060] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0062] Figure 8 A schematic diagram of the structure of the post-earthquake bridge damage identification device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 8 As shown, the post-earthquake bridge damage identification device 8 includes: The acquisition module 81 is used to acquire images of the bridge after the earthquake using a drone, and to perform image recognition on the bridge images to obtain crack mask images of the bridge after the earthquake. Extraction module 82 is used to extract the skeleton from the crack mask image to obtain the crack parameters of the bridge after the earthquake; The prediction module 83 is used to input the structural parameters and crack parameters of the bridge after the earthquake into the deep neural network surrogate model to obtain the seismic response of each key component of the bridge after the earthquake; wherein, the deep neural network surrogate model is trained based on the seismic simulation data of the bridge after the earthquake. Evaluation module 84 is used to determine the damage evaluation result of each key component of the bridge after the earthquake based on the seismic response of each component, and to determine the damage evaluation result of the bridge after the earthquake based on the damage evaluation result of each component.

[0063] In one possible implementation, the acquisition module 81 is specifically used for: performing target detection on the bridge image to obtain the crack region contained in the bridge image; performing pixel segmentation on the crack region to obtain the crack mask image of each crack in the crack region.

[0064] In one possible implementation, the extraction module 82 is specifically used to: iteratively identify each pixel in the crack mask image based on a preset iterative algorithm, and extract the crack skeleton from the crack mask image; calculate the crack length of the bridge after the earthquake based on the number of pixels in the crack skeleton, the object distance of the bridge image, the focal length of the camera device on the UAV, and the sampling distance between the UAV and the bridge after the earthquake; and calculate the crack width of the bridge after the earthquake based on the minimum distance from the crack skeleton to the boundary of the crack area, the object distance of the bridge image, the focal length of the camera device on the UAV, and the sampling distance between the UAV and the bridge after the earthquake.

[0065] In one possible implementation, the prediction module 83 is specifically used to: input multiple sets of preset seismic parameters into a pre-constructed finite element model of the post-earthquake bridge; perform nonlinear time history analysis on the post-earthquake bridge based on the finite element model and the seismic parameters to obtain the seismic response of the key components of the post-earthquake bridge under each set of preset seismic parameters; use multiple sets of preset seismic parameters as input to a deep neural network surrogate model; use the seismic response of the key components of the post-earthquake bridge under each set of preset seismic parameters as output to the deep neural network surrogate model; iteratively update the deep neural network surrogate model to obtain the deep neural network surrogate model.

[0066] In one possible implementation, the evaluation module 84 is specifically used for: obtaining multiple initial evaluation results corresponding to the seismic response of each key component of the bridge after the earthquake; obtaining the membership matrix of the bridge after the earthquake based on the arrangement order of each key component and the multiple initial evaluation results corresponding to the seismic response of each key component; wherein the number of rows in the membership matrix is ​​the number of key components; the number of columns in the membership matrix is ​​the number of preset damage levels; performing hierarchical analysis on each key component according to the arrangement order of each key component to obtain the weight matrix of the bridge after the earthquake; wherein the number of rows in the weight matrix is ​​1; the number of columns in the weight matrix is ​​the number of key components; and determining the damage level corresponding to the maximum value of each element in the vector product of the weight matrix and the membership matrix as the damage evaluation result of the bridge after the earthquake.

[0067] In one possible implementation, the evaluation module 84 is further configured to: for any critical component, determine the membership degree row vector of any critical component based on the initial evaluation result of any critical component; wherein, the number of columns in the membership degree row vector is the number of damage levels; each value in the membership degree row vector is the ratio of the number of each damage level in multiple initial evaluation results to the number of initial evaluation results; and concatenate the membership degree row vector of each critical component according to the arrangement order of each critical component to obtain the membership degree matrix of the post-earthquake bridge.

[0068] In one possible implementation, the evaluation module 84 is further configured to: obtain multiple sets of importance scores for any two key components; wherein the importance score is the evaluation result obtained by comparing any two key components pairwise; construct a judgment matrix based on the importance scores of any two key components; wherein the number of rows and columns of the judgment matrix is ​​the number of key components; each element in the judgment matrix is ​​the importance score of the two key components corresponding to the row and column of the element; normalize each judgment matrix and calculate the geometric mean of each element in all judgment matrices, and determine the weight of the key component in each row based on the geometric mean of each element; arrange the weights of each key component according to the order of the key components to obtain the weight matrix of the post-earthquake bridge.

[0069] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 9 As shown, the electronic device 9 of this embodiment includes a processor 90 and a memory 91. The memory 91 stores a computer program 92. When the processor 90 executes the computer program 92, it implements the steps in the various method embodiments described above. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module / unit in the various device embodiments described above.

[0070] For example, computer program 92 may be divided into one or more modules / units, which are stored in memory 91 and executed by processor 90 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 92 in electronic device 9.

[0071] Electronic device 9 may include, but is not limited to, processor 90 and memory 91. Those skilled in the art will understand that... Figure 9 This is merely an example of electronic device 9 and does not constitute a limitation on electronic device 9. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 9 may also include input / output devices, network access devices, buses, etc.

[0072] The processor 90 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0073] The memory 91 can be an internal storage unit of the electronic device 9, such as a hard disk or RAM. The memory 91 can also be an external storage device of the electronic device 9, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 91 can include both internal and external storage units of the electronic device 9. The memory 91 is used to store the computer program 92 and other programs and data required by the electronic device 9. The memory 91 can also be used to temporarily store data that has been output or will be output.

[0074] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0075] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0076] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0077] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying damage to bridges after an earthquake, characterized in that, include: Using drones to collect images of bridges after an earthquake, and performing image recognition on the bridge images, crack mask images of the bridges after the earthquake are obtained. The crack parameters of the bridge after the earthquake are obtained by extracting the skeleton from the crack mask image. The structural parameters and crack parameters of the post-earthquake bridge are input into a deep neural network surrogate model to obtain the seismic response of each key component of the post-earthquake bridge; wherein, the deep neural network surrogate model is trained based on the seismic simulation data of the post-earthquake bridge; Based on the seismic response of each key component of the bridge after the earthquake, the damage assessment result of each component is determined, and based on the damage assessment result of each component, the damage assessment result of the bridge after the earthquake is determined.

2. The method for identifying post-earthquake bridge damage according to claim 1, characterized in that, The process of determining the damage assessment result of each key component of the bridge based on its seismic response after the earthquake, and determining the damage assessment result of the bridge after the earthquake based on the damage assessment results of each component, includes: Obtain multiple initial evaluation results corresponding to the seismic response of each key component of the bridge after the earthquake; Based on the arrangement order of each key component and the multiple initial evaluation results corresponding to the seismic response of each key component, the membership matrix of the post-earthquake bridge is obtained; wherein, the number of rows in the membership matrix is ​​the number of key components; and the number of columns in the membership matrix is ​​the number of preset damage levels. Based on the arrangement order of each key component, a hierarchical analysis is performed on each key component to obtain the weight matrix of the post-earthquake bridge; wherein, the weight matrix has 1 row and the number of columns is the number of key components; The damage level corresponding to the maximum value of each element in the vector product of the weight matrix and the membership matrix is ​​determined as the damage evaluation result of the bridge after the earthquake.

3. The method for identifying post-earthquake bridge damage according to claim 2, characterized in that, Based on the arrangement order of each key component and multiple initial evaluation results corresponding to the seismic response of each key component, the membership matrix of the post-earthquake bridge is obtained, including: For any key component, based on the initial evaluation result of the key component, determine the membership degree row vector of the key component; Wherein, the number of columns in the membership degree row vector is the number of damage levels; each value in the membership degree row vector is the ratio of the number of each damage level in the plurality of initial evaluation results to the number of initial evaluation results; The membership degree row vectors of each key component are concatenated according to the arrangement order of each key component to obtain the membership degree matrix of the post-earthquake bridge.

4. The method for identifying post-earthquake bridge damage according to claim 3, characterized in that, The step of performing hierarchical analysis on each key component according to its arrangement order to obtain the weight matrix of the post-earthquake bridge includes: Obtain importance scores for any two key components from multiple sets; wherein, the importance score is the evaluation result obtained by comparing any two key components pairwise. A judgment matrix is ​​constructed based on the importance scores of any two key components; wherein the number of rows and columns of the judgment matrix are the number of key components; and each element in the judgment matrix is ​​the importance score of the two key components corresponding to the row and column of that element. Normalize each judgment matrix and calculate the geometric mean of each element in all judgment matrices. Use the geometric mean of each element as the weight of the key component in each row. The weights of each key component are arranged in the order of their arrangement to obtain the weight matrix of the post-earthquake bridge.

5. The method for identifying post-earthquake bridge damage according to any one of claims 1-4, characterized in that, Before inputting the structural parameters and crack parameters of the post-earthquake bridge into the deep neural network surrogate model to obtain the seismic response of each key component of the post-earthquake bridge, the deep neural network surrogate model is trained. The training process is as follows: Multiple sets of preset seismic parameters are input into the pre-constructed finite element model of the post-earthquake bridge. Based on the finite element model and the seismic parameters, nonlinear time history analysis is performed on the post-earthquake bridge to obtain the seismic response of the key components of the post-earthquake bridge under each set of preset seismic parameters. The deep neural network proxy model is obtained by using the multiple sets of preset earthquake parameters as inputs and the earthquake response of the key components of the bridge after the earthquake under each set of preset earthquake parameters as outputs.

6. The method for identifying post-earthquake bridge damage according to claim 5, characterized in that, The step of performing image recognition on the bridge image to obtain the crack mask image of the bridge after the earthquake includes: Target detection is performed on the bridge image to obtain the crack regions contained in the bridge image; The crack region is segmented into pixels to obtain a crack mask image for each crack in the crack region.

7. The method for identifying post-earthquake bridge damage according to claim 6, characterized in that, The crack parameters include crack width and crack length; The process of extracting the skeleton from the crack mask image to obtain the crack parameters of the post-earthquake bridge includes: Based on a preset iterative algorithm, each pixel in the crack mask image is iteratively identified to extract the crack skeleton from the crack mask image. The length of the crack in the post-earthquake bridge is calculated based on the number of pixels in the crack skeleton, the object distance in the bridge image, the focal length of the camera device on the UAV, and the sampling distance between the UAV and the post-earthquake bridge. The crack width of the post-earthquake bridge is calculated based on the minimum distance from the crack skeleton to the boundary of the crack region, the object distance of the bridge image, the focal length of the camera device on the UAV, and the sampling distance between the UAV and the post-earthquake bridge.

8. A damage identification device for bridges after an earthquake, characterized in that, include: The acquisition module is used to acquire images of the bridge after the earthquake using a drone, and to perform image recognition on the bridge images to obtain crack mask images of the bridge after the earthquake. The extraction module is used to extract the skeleton from the crack mask image to obtain the crack parameters of the post-earthquake bridge. The prediction module is used to input the structural parameters and crack parameters of the post-earthquake bridge into a deep neural network surrogate model to obtain the seismic response of each key component of the post-earthquake bridge; wherein, the deep neural network surrogate model is trained based on the seismic simulation data of the post-earthquake bridge; The evaluation module is used to determine the damage evaluation result of each key component of the bridge after the earthquake based on the seismic response of each component, and to determine the damage evaluation result of the bridge after the earthquake based on the damage evaluation result of each component.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images