Machine learning-based rapid assessment method for earthquake damage performance of high-speed railway bridge

Abstract

The invention relates to the technical field of bridge earthquake damage performance evaluation, and particularly discloses a high-speed railway bridge earthquake damage performance rapid evaluation method based on machine learning, and the method comprises the steps: collecting multi-source data of a bridge structure of a high-speed railway, a historical earthquake database and a BIM model of the bridge structure in real time through a heterogeneous data collection system; constructing a multi-modal feature space containing partial features of time-frequency domain features, spatial features and probability features; constructing a hierarchical machine learning framework based on a dynamic Bayesian network damage propagation model, a mixed model of a multi-scale convolutional neural network and a long-short term memory network and an uncertainty quantization and dynamic updating mechanism, and establishing an earthquake damage performance evaluation model based on the hierarchical machine learning framework; carrying out seismic damage performance evaluation on the multi-modal feature space by adopting a seismic damage performance evaluation model, and generating damage probability distribution; potential earthquake damage related information behind the data is effectively mined, and the earthquake damage condition of the bridge in an earthquake can be more accurately simulated and predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic damage performance evaluation of bridges, and particularly to a rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning. Background Art

[0002] At present, in the modern transportation system, as an important infrastructure, the safety and stability of high-speed railways play a crucial role in ensuring people's travel and economic development. However, the potential threat of earthquake disasters poses a huge challenge to the safety of high-speed railway bridges. Once an earthquake occurs, accurately and quickly evaluating the seismic damage performance of bridges is crucial for taking effective emergency measures in a timely manner, ensuring the railway to resume operation as soon as possible, and reducing the losses of secondary disasters.

[0003] Traditional methods for evaluating the seismic damage performance of high-speed railway bridges are mainly based on structural mechanics theory and numerical simulation. These methods establish complex mechanical models and calculate the response of bridges under earthquake action according to theoretical formulas. Although they can analyze the seismic damage situation of bridges to a certain extent, the calculation process is cumbersome, requiring a large amount of professional knowledge and computing resources, and taking a long time, making it difficult to quickly give an evaluation result in the emergency situation after an earthquake. With the rapid development of sensor technology, data acquisition and storage technology, and machine learning algorithms, a rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning shows great potential. However, relying solely on structural mechanics models cannot fully integrate multi-source heterogeneous data, and the utilization of a single data type limits the comprehensive and accurate evaluation of the seismic damage performance of bridges. In addition, earthquakes themselves are uncertain, and there are many uncertain factors in the response of bridge structures under earthquake action. Traditional evaluation methods have limited means in dealing with these uncertainties and cannot accurately quantify the impact of uncertainties on the evaluation results of seismic damage performance, resulting in a reduction in the reliability of the evaluation results.

[0004] Therefore, the present invention proposes a rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning. Summary of the Invention

[0005] The present invention provides a method for rapid assessment of earthquake damage performance of high-speed railway bridges based on machine learning, including: real-time acquisition of multi-source data of bridge structures through a plurality of distributed sensor arrays in a heterogeneous data acquisition system, and combining the acquired historical earthquake database and the BIM model of the bridge structure, providing a rich and comprehensive data basis for subsequent analysis. The multi-source data covers the real-time operation status information of the bridge, the historical earthquake database provides a reference for past earthquakes and bridge responses, and the BIM model presents the details of the bridge from the perspective of three-dimensional spatial structure. The combination of the three makes the understanding of the bridge more in-depth and three-dimensional, laying a solid foundation for accurately assessing earthquake damage performance. A multimodal feature space including partial features such as time-frequency domain features, spatial features, and probability features is constructed, and feature extraction and integration of bridge data are performed from multiple dimensions. The time-frequency domain features can reflect the changing laws of data in time and frequency, the spatial features reflect the spatial characteristics of the bridge structure, and the probability features take into account the uncertainty of the occurrence of events. The establishment of a multimodal feature space helps to more comprehensively and accurately characterize the characteristics of the bridge under the action of earthquakes, and to mine the potential information related to earthquake damage behind the data. A hierarchical machine learning architecture is constructed based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM (convolutional neural network and long short-term memory network) hybrid model, and the uncertainty quantification and dynamic update mechanism, and a seismic damage performance evaluation model is established. The dynamic Bayesian network can effectively process uncertainty and dynamic change information and analyze the damage propagation process; the multi-scale CNN-LSTM hybrid model is good at extracting spatial and temporal features of different scales; the uncertainty quantification and dynamic update mechanism ensure that the model can adapt to the dynamic changes and uncertainties of the data. This complex and sophisticated architecture gives the evaluation model powerful learning and analysis capabilities, and can more accurately simulate and predict the seismic damage of bridges in earthquakes. The seismic damage performance evaluation model is used to evaluate the seismic damage performance of multimodal feature space and generate a damage probability distribution, which provides intuitive and quantitative evaluation results for the seismic damage performance of high-speed railway bridges. The damage probability distribution can clearly show the possibility of different degrees of damage, help decision makers quickly understand the risk of damage to bridges after an earthquake, and thus formulate targeted maintenance, repair or emergency plans, effectively ensure the safe operation of high-speed railway bridges, and reduce the losses that may be caused by earthquakes.

[0006] The present invention provides a method for rapid assessment of earthquake damage performance of high-speed railway bridges based on machine learning, comprising:

[0007] S1: Use a large number of distributed sensor arrays deployed in a heterogeneous data acquisition system to collect multi-source data of high-speed railway bridge structures in real time, and obtain historical earthquake databases and BIM models of bridge structures;

[0008] S2: Combine the multi-source data with the historical earthquake database and the BIM model of the bridge structure to construct a multi-modal feature space containing partial features such as time-frequency domain features, spatial features, and probability features;

[0009] S3: Based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism, construct a hierarchical machine learning architecture, and establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture;

[0010] S4: Use the seismic damage performance evaluation model to evaluate the seismic damage performance of the multi-modal feature space and generate the damage probability distribution of high-speed railway bridges.

[0011] Preferably, for the rapid evaluation method of the seismic damage performance of high-speed railway bridges based on machine learning, S1: Real-time collect the multi-source data of the bridge structure of high-speed railway through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system, and obtain the historical earthquake database and the BIM model of the bridge structure, including:

[0012] Real-time collect the track slab deformation data, seismic wave response data, overall bridge displacement, and speckle images of each local structure under different intensity earthquakes of the bridge structure of high-speed railway through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system as the multi-source data of the bridge structure of high-speed railway;

[0013] Obtain the historical earthquake database and the BIM model of the bridge structure, where the BIM model of the bridge structure includes bridge set parameters, material properties, and connection methods.

[0014] Preferably, for the rapid evaluation method of the seismic damage performance of high-speed railway bridges based on machine learning, S2: Combine the multi-source data with the historical earthquake database and the BIM model of the bridge structure to construct a multi-modal feature space containing partial features such as time-frequency domain features, spatial features, and probability features, including:

[0015] Perform spatio-temporal alignment on the multi-source data and the BIM model of the bridge structure based on the spatio-temporal alignment algorithm to obtain spatio-temporally aligned source data;

[0016] Extract time-frequency domain features and spatial features from the spatio-temporally aligned source data;

[0017] Based on the historical earthquake database and introduce Monte Carlo simulation to generate parameter perturbation samples, and then obtain probability features by fusing sensor measurement errors and model parameter uncertainties based on the evidence theory;

[0018] Construct a multi-modal feature space based on partial features such as time-frequency domain features, spatial features, and probability features.

[0019] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, time-frequency domain features and spatial features are extracted from the spatio-temporally aligned source data, including:

[0020] Extract the preset frequency band energy distribution from the multi-source data aligned in the spatio-temporally aligned source data through short-time Fourier transform, and then decompose it through Hilbert-Huang transform to obtain multiple intrinsic mode functions as time-frequency domain features;

[0021] Construct the bridge structure stiffness matrix based on the BIM model of the bridge structure aligned in the spatio-temporally aligned source data;

[0022] Extract damage-sensitive features from the spatio-temporally aligned source data;

[0023] Among them, the spatial features include the bridge structure stiffness matrix and damage-sensitive features.

[0024] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, damage-sensitive features are extracted from the spatio-temporally aligned source data, including:

[0025] Based on the phase correlation method, perform preset pixel-level registration on the speckle images of the local structures at different intensities of seismic action in the multi-source data aligned in the spatio-temporally aligned source data, and construct a full-field displacement vector field based on all the speckle images after pixel-level registration;

[0026] Construct a two-dimensional Delaunay triangular mesh based on the full-field displacement vector field, and statistically calculate the displacement gradient of the mesh nodes to generate a strain nephogram of the corresponding level by calculating the Green strain tension;

[0027] Calculate the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers based on various strain components in the strain nephogram;

[0028] Perform modal analysis on the spatio-temporally aligned source data to extract the first ten-order frequency change rates of the bridge piers;

[0029] Regard the warping deformation of the track slab and the first ten-order frequency change rates of the bridge piers as damage-sensitive features.

[0030] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, calculate the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers based on various strain components in the strain nephogram, including:

[0031] Establish a warping deformation calculation model based on strain components and a damage calculation model for each component;

[0032] Denoise and handle outliers of the strain nephogram to obtain a preprocessed strain nephogram;

[0033] Input the preprocessed strain nephogram into the warping deformation calculation model to obtain the warping deformation of the track slab, and input the preprocessed strain nephogram into the damage calculation models of each component to obtain the damage of the track layer connectors, the damage of the bearings, and the damage of the piers.

[0034] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, perform modal analysis on the spatio-temporally aligned source data to extract the first ten-order frequency change rates of the piers, including:

[0035] Extract the vibration response signals of the piers from the spatio-temporally aligned source data, and separate the partial pier vibration response signals within the frequency band of interest in the pry response signals through band-pass filtering;

[0036] Use the frequency domain decomposition method to perform spectral analysis on the partial pier vibration response signals within the frequency band of interest to identify the first ten-order natural frequencies;

[0037] Based on the first ten-order natural frequencies and the first ten-order natural frequencies of the piers in the reference state, calculate the first ten-order frequency change rates of the piers.

[0038] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, construct a multi-modal feature space based on some features of time-frequency domain features, spatial features, and probability features, including:

[0039] Visualize the feature contribution status of time-frequency domain features, spatial features, and probability features based on the SHAP value analysis method;

[0040] Based on the recursive feature elimination method and the feature contribution status of time-frequency domain features, spatial features, and probability features, retain the partial features with a contribution degree not less than the contribution degree threshold in time-frequency domain features, spatial features, and probability features and perform feature dimensionality reduction processing to obtain a multi-modal feature space.

[0041] Preferably, for the rapid assessment method of the seismic damage performance of high-speed railway bridges based on machine learning, S3: Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism, and establish a seismic damage performance assessment model based on the hierarchical machine learning architecture, including:

[0042] Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism;

[0043] Construct a multi-modal feature space of a large number of samples, and generate a system state vector for each sample based on the multi-modal feature space of each sample;

[0044] A seismic damage performance assessment model is established based on a hierarchical machine learning architecture and the system state vectors of all samples.

[0045] Preferably, the rapid assessment method for earthquake damage performance of high-speed railway bridges based on machine learning establishes an earthquake damage performance assessment model based on a hierarchical machine learning architecture and system state vectors of all samples, including:

[0046] Based on the dynamic Bayesian network damage propagation model in the hierarchical machine learning architecture, the system state vectors of all samples are decomposed into multiple damage levels, and the conditional damage probability matrix of all samples is generated based on the system state vectors of all samples and the corresponding damage levels;

[0047] Based on the conditional damage probability matrix of each sample, the transition probability of the local structural damage state of the bridge in all samples is defined;

[0048] The bridge damage propagation path is modeled based on a discrete-time Markov chain, and the joint probability of the bridge damage propagation path of each sample is calculated based on the transition probability of the local structural damage state of each bridge in each sample;

[0049] Based on the multi-scale CNN-LSTM hybrid model in the hierarchical machine learning architecture, the joint probability of the bridge damage propagation path of each sample is multi-scale fused to obtain the damage probability distribution of each sample bridge;

[0050] At the same time, uncertainty quantification results are obtained based on the uncertainty quantification and dynamic update mechanism, and the model update frequency is driven based on the uncertainty quantification results until a seismic damage performance evaluation model is established.

[0051] The beneficial effects of the present invention compared to the prior art are as follows: with the help of numerous distributed sensor arrays in a heterogeneous data acquisition system, multi-source data of the bridge structure can be obtained in real time, including vibration, strain, displacement and other information. At the same time, combined with historical earthquake data and the BIM model of the bridge structure, a rich multimodal feature space can be constructed. With its powerful data mining and pattern recognition capabilities, the machine learning algorithm can automatically learn the complex relationship between features and earthquake damage performance from massive data, and establish an efficient and accurate earthquake damage performance assessment model. This method can not only quickly obtain evaluation results, but also improve the accuracy and reliability of the evaluation. It is of great significance to enhance the emergency response capabilities of high-speed railways under earthquake disasters, and is expected to be widely used in the field of railway engineering safety assurance in the future.

[0052] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0054] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention. In the accompanying drawings:

[0055] Figure 1 is the architecture diagram of the rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning in the embodiment of the present invention;

[0056] Figure 2 is the architecture diagram of multi-modal feature fusion in the embodiment of the present invention;

[0057] Figure 3 is the network architecture diagram of the multi-scale CNN-LSTM hybrid model in the embodiment of the present invention. Detailed Embodiments

[0058] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0059] Embodiment 1:

[0060] The present invention provides a rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning, referring to Figure 1 , including:

[0061] S1: Real-time collect multi-source data of the bridge structure of high-speed railway through numerous distributed sensor arrays deployed in the heterogeneous data collection system, and obtain the historical earthquake database and the BIM model of the bridge structure;

[0062] This step aims to collect a comprehensive data basis for subsequent seismic damage performance evaluation. Through numerous distributed sensor arrays in the heterogeneous data collection system, multi-source data of the bridge structure of high-speed railway are collected in real time. These sensors may include acceleration sensors, strain gauges, displacement sensors, etc. The collected data, such as track slab deformation data, seismic wave response data, overall displacement of the bridge, and strain of pier materials, reflect the real-time state of the bridge during an earthquake from multiple dimensions. At the same time, obtain the historical earthquake database, which contains relevant information of past earthquake events, such as earthquake intensity, duration, and bridge response, providing a historical reference for analyzing the impact of earthquakes on bridges. In addition, obtain the BIM (Building Information Model) of the bridge structure, which covers detailed information such as geometric parameters, material properties, and connection methods of the bridge, accurately presenting the bridge structure from a three-dimensional space perspective and assisting in understanding the physical characteristics of the bridge and potential seismic damage correlations.

[0063] S2: Combine multi-source data with historical earthquake database and BIM model of bridge structure to construct a multimodal feature space including some features of time-frequency domain, spatial features, and probability features;

[0064] This step first uses the time-space alignment algorithm to align the multi-source data with the BIM model of the bridge structure in time and space, ensuring that the data from different sources are consistent in time and space, thereby obtaining time-space aligned source data. For example, the data collected by the sensor at different time and space positions are matched with the corresponding positions and time points in the BIM model. Then, the time-frequency domain features and spatial features are extracted from the time-space aligned source data. The time-frequency domain features can reflect the change law of the data in time and frequency. For example, the energy distribution of the preset frequency band is extracted by short-time Fourier transform, and then multiple intrinsic mode functions are obtained as time-frequency domain features by Hilbert-Huang transform decomposition. Spatial features reflect the spatial characteristics of the bridge structure, including the construction of the bridge structure stiffness matrix based on the BIM model, and the extraction of damage-sensitive features, such as the warping deformation of the track plate and the frequency change rate of the bridge pier obtained by analyzing the speckle image. Based on the historical earthquake database, Monte Carlo simulation is introduced to generate parameter perturbation samples, and the sensor measurement error and model parameter uncertainty are fused by evidence theory to obtain probabilistic features, which take into account the uncertainty factors in earthquakes and measurements. Finally, a multimodal feature space is constructed based on the time-frequency domain characteristics, spatial characteristics, and probability characteristics to comprehensively characterize the characteristics of the bridge under the action of earthquakes and to explore potential information related to earthquake damage.

[0065] S3: A hierarchical machine learning architecture is constructed based on the dynamic Bayesian network damage propagation model and the multi-scale CNN-LSTM hybrid model as well as the uncertainty quantification and dynamic update mechanism, and a seismic damage performance evaluation model is established based on the hierarchical machine learning architecture;

[0066] A hierarchical machine learning architecture is constructed based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism. The dynamic Bayesian network is good at processing uncertainty and dynamic change information, and can analyze the propagation process of damage in the bridge structure; the multi-scale CNN-LSTM hybrid model can extract spatial and temporal features of different scales and fully explore the patterns in the data; the uncertainty quantification and dynamic update mechanism ensure that the model can adapt to the dynamic changes and uncertainties of the data.

[0067] A multimodal feature space of a large number of samples is constructed, and a system state vector is generated based on the multimodal feature space of each sample, which comprehensively describes the characteristics of the sample.

[0068] An earthquake damage performance evaluation model is established using a hierarchical machine learning architecture and the system state vectors of all samples. For example, the system state vectors are decomposed into multiple damage levels through a dynamic Bayesian network damage propagation model to generate a conditional damage probability matrix, and the transition probabilities of the local structural damage states at various locations of the bridge are defined. A discrete-time Markov chain is used to model the bridge damage propagation path and calculate the joint probability. Then, a multi-scale CNN-LSTM hybrid model performs multi-scale fusion on the joint probability to obtain the damage probability distribution. At the same time, the model is driven to update based on the uncertainty quantification and dynamic update mechanism until a complete earthquake damage performance evaluation model is established.

[0069] S4: Use the earthquake damage performance evaluation model to evaluate the earthquake damage performance of the multi-modal feature space, and generate the damage probability distribution of the high-speed railway bridge;

[0070] Use the established earthquake damage performance evaluation model to analyze the multi-modal feature space, so as to evaluate the earthquake damage performance of the high-speed railway bridge, and finally generate the damage probability distribution. This damage probability distribution intuitively shows the possibility of different degrees of damage occurring. For example, it can show the probabilities of different levels of damage such as slight damage, moderate damage, severe damage, and complete damage of the bridge, providing a quantitative evaluation result for decision-makers to quickly understand the damage risk of the bridge after an earthquake, and then formulate targeted maintenance, repair, or emergency plans to ensure the safe operation of the high-speed railway bridge and reduce the losses caused by the earthquake.

[0071] Embodiment 2:

[0072] Based on the earthquake damage performance rapid evaluation method for high-speed railway bridges based on machine learning in Embodiment 1, S1: Real-time collect multi-source data of the bridge structure of the high-speed railway through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system, and obtain the historical earthquake database and the BIM model of the bridge structure, including:

[0073] Real-time collect the track slab deformation data, seismic wave response data, overall displacement of the bridge, and speckle images of each local structure under different intensity earthquakes of the bridge structure of the high-speed railway as multi-source data of the bridge structure of the high-speed railway through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system;

[0074] Obtain the historical earthquake database and the BIM model of the bridge structure, where the BIM model of the bridge structure includes bridge set parameters, material properties, and connection methods.

[0075] In this embodiment:

[0076] Multi-source data collection: Use numerous distributed sensor arrays in the heterogeneous data acquisition system to collect specific types of data of the high-speed railway bridge structure in real time.

[0077] Track slab deformation data: As the part directly bearing the train load on high-speed railway bridges, the deformation of the track slab is crucial for the overall performance of the bridge and the safety of train operation. By monitoring the track slab deformation data in real time through sensors, the impact of earthquakes on the track slab can be detected in a timely manner. For example, whether the track slab has warping, twisting and other deformations, which may further affect the smoothness and safety of train operation.

[0078] Seismic wave response data: Seismic waves are the waves that transmit energy during an earthquake, and various responses will occur to the bridge under the action of an earthquake. Collecting seismic wave response data can understand the reaction of the bridge to seismic waves with different characteristics, such as the amplitude and frequency of vibration, etc. These information helps to analyze the action intensity and mode of the earthquake on the bridge structure and is one of the key data for evaluating the seismic damage performance of the bridge.

[0079] Overall displacement of the bridge: Monitoring the overall displacement data of the bridge can intuitively reflect the change of the spatial position of the bridge during an earthquake. Excessive overall displacement may indicate that the stability of the bridge structure is threatened. By obtaining this data in real time, the overall state of the bridge during an earthquake can be quickly judged, providing an important basis for subsequent evaluation.

[0080] The strain of the pier refers to the relative deformation inside the pier when it is subjected to external forces (such as earthquakes, vehicle loads, etc.). In actual situations, when external forces are applied to the pier, such as the impact of seismic waves on the pier during an earthquake, or the pressure generated on the pier when a train is running, the shape or size of the pier material will change, and this degree of change is measured by strain. Strain is usually expressed by a dimensionless ratio, reflecting the elongation or shortening degree of the pier material in the direction of the force. Information related to the strain of the pier is very crucial for evaluating the seismic damage of the bridge. By monitoring the strain of the pier, the stress state borne by the pier can be understood, and it can be judged whether it exceeds the bearing range of the material, and then it can be inferred whether the pier has damage and the degree of damage, providing an important basis for taking maintenance or repair measures in a timely manner. For example, if the strain in a certain area of the pier suddenly increases and exceeds the normal range, it may mean that there are damages such as cracks in this area, which requires further detection and treatment.

[0081] Speckle images of each local structure under seismic actions of different intensities: Speckle image technology is often used to measure the micro-deformation of the object surface. By collecting the speckle images of each local structure of the bridge under seismic actions of different intensities and analyzing these images, the subtle changes of the local structure, such as strain, displacement and other information, can be obtained. Seismic actions of different intensities simulate various situations that the bridge may encounter during actual operation. Combining speckle image analysis can more comprehensively understand the damage situation of the local structure under the combined action of earthquakes and different loads.

[0082] Other data acquisition:

[0083] Historical earthquake database: It contains records of past earthquake events and the corresponding bridge responses. These historical data can be used as a reference to help analyze the similarity between the current earthquake and historical earthquakes, as well as the damage patterns and degrees of past bridges under similar earthquake conditions, providing empirical support for the assessment of the seismic damage performance this time.

[0084] BIM model of the bridge structure: This model includes the geometric parameters of the bridge (such as the length and height of the bridge structure and track structure), material properties (such as the mechanical property parameters of concrete and steel), and connection methods (such as the connection form between the pier and the bridge deck, the connection node characteristics between the layers of the track structure). These detailed information accurately describes the bridge structure from a physical level, which is very crucial for understanding the mechanical response and potential damage mechanism of the bridge under earthquake action, and provides the basic structural information for subsequent construction of the multi-modal feature space and establishment of the seismic damage performance assessment model.

[0085] Example 3:

[0086] Based on the method for rapid assessment of seismic damage performance of high-speed railway bridges based on machine learning in Example 1, S2: Combine multi-source data with the historical earthquake database and the BIM model of the bridge structure to construct a multi-modal feature space containing partial features of time-frequency domain features, spatial features, and probability features, including:

[0087] Perform spatio-temporal alignment of the multi-source data and the BIM model of the bridge structure based on the spatio-temporal alignment algorithm to obtain spatio-temporally aligned source data;

[0088] Extract time-frequency domain features and spatial features from the spatio-temporally aligned source data;

[0089] Generate parameter perturbation samples based on the historical earthquake database and introduce Monte Carlo simulation, and then obtain probability features by fusing sensor measurement errors and model parameter uncertainties based on the evidence theory;

[0090] Construct a multi-modal feature space based on the partial features of time-frequency domain features, spatial features, and probability features.

[0091] In this example:

[0092] Spatio-temporal alignment: Using the spatio-temporal alignment algorithm, multi-source data collected through a distributed sensor array is spatio-temporally aligned with the BIM model of the bridge structure. Since the multi-source data comes from different types of sensors, there may be differences in the time and space dimensions of the data they collect, while the BIM model is a digital representation of the bridge structure from the perspectives of three-dimensional space and time. Through spatio-temporal alignment, the multi-source data and the BIM model can correspond to each other in time and space, forming unified spatio-temporally aligned source data. For example, the vibration data of a certain part of the bridge collected at a certain moment is accurately matched to the corresponding spatial position and time point in the BIM model, which can ensure the consistency and relevance of the subsequent extracted features and lay a foundation for accurately analyzing the bridge state.

[0093] Extracting time-frequency domain and spatial features includes:

[0094] Time-frequency domain feature extraction: Time-frequency domain features can reflect the variation characteristics of data in the time and frequency dimensions. Extracting time-frequency domain features from the spatio-temporally aligned source data helps to capture the dynamic variation laws of the bridge structure response over time and frequency. For example, during an earthquake, the vibration response of the bridge will exhibit specific patterns at different times and frequencies, and by analyzing these patterns, the damage condition of the bridge structure can be inferred.

[0095] Spatial feature extraction: Spatial features mainly reflect the spatial characteristics of the bridge structure, including information such as the geometric shape of the bridge, the distribution of structural stiffness, and local structural changes. These features are of great significance for understanding the mechanical response and damage propagation of the bridge in space. For example, by constructing the bridge structure stiffness matrix, the ability of each part of the bridge to resist deformation can be quantified, and then the force conditions and potential damage areas of different parts under earthquake action can be analyzed.

[0096] Probability feature extraction: Monte Carlo simulation to generate parameter perturbation samples: Based on the historical earthquake database, the Monte Carlo simulation method is introduced. Monte Carlo simulation is a technique that simulates uncertain factors through random sampling. In this case, various parameters in the historical earthquake data (such as earthquake intensity, duration, spectral characteristics, etc.) are used to generate parameter perturbation samples through multiple random samplings to simulate the uncertainty during the earthquake process.

[0097] Fusing Uncertainties Based on the Evidence Theory: Combining the measurement errors of sensors and the uncertainties of model parameters, and using the evidence theory for fusion to obtain probability characteristics. When sensors collect data, measurement errors inevitably exist. At the same time, when establishing a bridge model, there are also certain uncertainties in the model parameters. The evidence theory can effectively process this uncertain information, fuse them together, and obtain a probability characteristic that reflects the seismic damage probability characteristics of the bridge. This probability characteristic can quantify the possibility of the bridge appearing in different damage states under seismic action, providing probabilistic information for the assessment of seismic damage performance.

[0098] Constructing a Multimodal Feature Space: Integrate the extracted time-frequency domain features, spatial features, and probability features to construct a multimodal feature space. The multimodal feature space can comprehensively describe the characteristics of the bridge under seismic action from multiple dimensions, fully mining the potential information related to seismic damage behind the data. In this way, it provides rich, comprehensive, and representative feature data for the subsequent seismic damage performance assessment model based on machine learning, helping to improve the accuracy and reliability of model assessment.

[0099] Example 4:

[0100] Based on the method for rapid assessment of seismic damage performance of high-speed railway bridges based on machine learning in Example 3, extract time-frequency domain features and spatial features from the spatio-temporal aligned source data, including:

[0101] Extract the energy distribution of a preset frequency band from the multi-source data aligned in the spatio-temporal aligned source data through short-time Fourier transform, and then decompose it through Hilbert-Huang transform to obtain multiple intrinsic mode functions as time-frequency domain features;

[0102] Construct the bridge structure stiffness matrix based on the BIM model of the bridge structure in the spatio-temporal aligned source data;

[0103] Extract damage-sensitive features from the spatio-temporal aligned source data;

[0104] Among them, the spatial features include the bridge structure stiffness matrix and damage-sensitive features.

[0105] In this embodiment:

[0106] Short-time Fourier transform to extract the energy distribution in the preset frequency band: The short-time Fourier transform (STFT) can transform a time-domain signal into the time-frequency domain, thereby revealing the frequency composition of the signal at different time points. In this embodiment, STFT is applied to the aligned multi-source data in the spatio-temporally aligned source data, with a focus on the energy distribution within the preset frequency band. Because the energy distribution in different frequency bands is often related to specific responses or damage states of the bridge structure. For example, changes in the energy of certain frequency bands (such as 0 - 50 Hz) may reflect increased vibration in a certain part of the bridge structure, which may be a signal of potential damage. By analyzing the energy distribution in these preset frequency bands, characteristic information about the bridge structure response in the time and frequency dimensions can be initially obtained.

[0107] Hilbert-Huang transform to obtain intrinsic mode functions: After obtaining the energy distribution in the preset frequency band, the Hilbert-Huang transform (HHT) is further used. HHT is a method suitable for analyzing non-linear and non-stationary signals, which can decompose a complex signal into multiple intrinsic mode functions (IMFs). These IMFs represent the oscillation modes at different time scales in the signal, and each IMF has a certain physical meaning, reflecting different characteristic components of the original signal. The multiple IMFs obtained through HHT decomposition, as time-frequency domain features, can more finely characterize the time-frequency characteristics of the bridge structure response, helping to discover the complex vibration modes and potential damage characteristics of the bridge structure under seismic action.

[0108] Construct the bridge structure stiffness matrix based on the BIM model: Using the BIM model of the bridge structure in the aligned spatio-temporally aligned source data, construct the bridge structure stiffness matrix. The bridge structure stiffness matrix is an important parameter matrix that describes the ability of the bridge structure to resist deformation, which reflects the relationship between the displacements of each node of the bridge and the applied loads. Through the detailed information such as the bridge geometric parameters, material properties, and connection methods provided by the BIM model, the bridge structure stiffness matrix can be accurately calculated. This matrix is crucial for analyzing the mechanical response of the bridge under seismic action. For example, the deformation of each part of the bridge under a given load (such as seismic force) can be calculated through the stiffness matrix, thereby evaluating the stability of the bridge structure and potential damage areas. The formula is:

[0109]

[0110] In the formula, K global is the bridge structure stiffness matrix, K j is the stiffness matrix of the jth beam element, T j is the coordinate transformation matrix, is the transpose matrix of the coordinate transformation matrix, and m is the total number of beam elements of the bridge structure.

[0111] Extracting damage-sensitive features: Damage-sensitive features refer to those characteristic parameters that are relatively sensitive to the damage of the bridge structure. Through the changes in these features, it is possible to effectively detect whether the bridge structure has suffered damage and the degree of damage. Damage-sensitive features are extracted from the spatio-temporally aligned source data, for example, by analyzing data such as speckle images and vibration responses. These damage-sensitive features and the bridge structure stiffness matrix together constitute the spatial features, which can comprehensively reflect the state of the bridge structure from a spatial perspective and provide key information for the evaluation of the seismic damage performance of the bridge. For example, the changes in certain damage-sensitive features may directly indicate the damage of the local structure of the bridge. Combining the analysis of the bridge structure stiffness matrix, the damage location can be more accurately located and the impact of the damage on the overall structure can be evaluated.

[0112] Example 5:

[0113] Based on the method for rapid evaluation of the seismic damage performance of high-speed railway bridges using machine learning in Example 4, damage-sensitive features are extracted from the spatio-temporally aligned source data, including:

[0114] Based on the phase correlation method, perform preset pixel-level registration on the speckle images of the local structure at different intensities of seismic action in the aligned multi-source data in the spatio-temporally aligned source data, and construct a full-field displacement vector field based on all the speckle images after pixel-level registration;

[0115] Construct a two-dimensional Delaunay triangular mesh based on the full-field displacement vector field, and calculate the Green's strain tension by statistically analyzing the displacement gradients of the grid nodes to generate the corresponding level of strain nephogram;

[0116] Calculate the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers based on various strain components in the strain nephogram;

[0117] Perform modal analysis on the spatio-temporally aligned source data to extract the change rate of the first ten natural frequencies of the bridge piers;

[0118] Take the warping deformation of the track slab and the change rate of the first ten natural frequencies of the bridge piers as damage-sensitive features.

[0119] In this embodiment:

[0120] Speckle Image Registration and Construction of Full-Field Displacement Vector Field Based on Phase Correlation Method: First, use the phase correlation method to perform preset pixel-level registration on speckle images under different intensity seismic actions in the spatio-temporally aligned source data. The speckle images record the random speckle distributions on the surface of the local structure of the bridge. Under different loadings, these speckles will displace due to structural deformation. The phase correlation method can accurately calculate the relative displacement between adjacent speckle images by analyzing the phase information of the images, thereby accurately registering them at the pixel level. For example, assume that in the speckle image of a certain local structure, a speckle has undergone a small movement before and after the load change. The phase correlation method can accurately measure the distance and direction of this movement;

[0121] Based on all the speckle images after pixel-level registration, construct a full-field displacement vector field. The full-field displacement vector field reflects the displacement conditions of the entire local structure surface under different intensity seismic actions. Each point corresponds to a displacement vector, containing information about the magnitude and direction of the displacement. Through this full-field displacement vector field, the deformation trend of the local structure under loading can be intuitively understood, providing basic data for subsequent analysis.

[0122] Two-Dimensional Delaunay Triangulation Grid Construction and Generation of Strain Cloud Map: Based on the full-field displacement vector field, construct a two-dimensional Delaunay triangulation grid. The Delaunay triangulation grid is a method for triangulating discrete points on a plane. It can connect the discrete points in the full-field displacement vector field into a triangular grid, and these triangles have the property of an empty circumcircle, which can better reflect the spatial relationship between data points. For example, use the key displacement points in the full-field displacement vector field as grid nodes to construct a Delaunay triangulation grid;

[0123] Statistically calculate the Green strain tension by calculating the displacement gradient of grid nodes, and then generate a strain cloud map corresponding to the level. The displacement gradient reflects the rate of change of displacement in space. By calculating the displacement gradient of grid nodes, the strain information in each triangular region can be obtained. The Green strain tension is a physical quantity describing material deformation. Calculating the Green strain tension based on the displacement gradient can more accurately reflect the strain state of the local structure. According to these strain values, generate a strain cloud map. The strain cloud map visually shows the strain magnitudes in different regions of the local structure in different colors. The darker or lighter the color, the greater or smaller the strain, which is convenient for intuitively analyzing the strain distribution of the local structure.

[0124] Calculation of the Warping Deformation Quantity of the Track Slab: Calculate the warping deformation quantity of the track slab, the damage quantity of the track layer connectors, the damage quantity of the bearings, and the damage quantity of the bridge piers based on various strain components in the strain cloud map. The track slab is an important part of high-speed railway bridges, and its warping deformation quantity has an important impact on the safety of train operation. First, establish a calculation model for the warping deformation quantity based on strain components. This model considers the comprehensive influence of strain components in different directions in the strain cloud map on the warping of the track slab;

[0125] Denoise and process the strain nephogram to obtain the preprocessed strain nephogram. Since the actually collected data may be affected by noise interference or there may be outliers, these noises and outliers may affect the accuracy of the calculation of the warping deformation amount of the track slab. By denoising and processing outliers, the data quality can be improved and irrelevant or incorrect data points can be removed. For example, a filtering algorithm is used to remove high-frequency noise, and outliers are removed by setting a reasonable data range;

[0126] Input the preprocessed strain nephogram into the warping deformation amount calculation model to obtain the warping deformation amount of the track slab, and input the preprocessed strain nephogram into each component damage amount calculation model to obtain the damage amount of the track layer connectors, the damage amount of the bearings, and the damage amount of the bridge piers. The processed strain nephogram data is more accurate and reliable. When input into the calculation model, more accurate results of the warping deformation amount of the track slab, the damage amount of the track layer connectors, the damage amount of the bearings, and the damage amount of the bridge piers can be obtained. These results can be used as one of the important indicators for evaluating the local structural damage of the bridge.

[0127] Extraction of the first ten-order frequency change rate of the bridge pier: Extract the vibration response signal of the bridge pier from the spatio-temporal aligned source data, and separate the partial vibration response signal of the bridge pier within the frequency band of interest in the vibration response signal through band-pass filtering. As the supporting structure of the bridge, the vibration characteristics of the bridge pier are closely related to the overall stability of the bridge. Band-pass filtering can select signals within a specific frequency range, remove interference signals of other frequencies, and only retain the signals within the frequency band of interest because the signals in this frequency band may be related to specific vibration modes or damage states of the bridge pier. For example, the vibration in certain frequency bands may be related to the first, second, or third vibration modes of the bridge pier;

[0128] Use the frequency domain decomposition method to perform spectral analysis on the partial vibration response signal of the bridge pier within the frequency band of interest to identify the first ten natural frequencies. The frequency domain decomposition method is a method that converts a time-domain signal to the frequency domain for analysis. Through this method, the energy distribution of the signal at different frequencies can be obtained, so as to identify the first ten natural frequencies of the bridge pier. The natural frequency is an inherent characteristic of the structure. When the bridge pier is damaged, its natural frequency often changes;

[0129] Based on the first ten natural frequencies and the first ten natural frequencies of the bridge pier in the reference state, calculate the first ten-order frequency change rate of the bridge pier. By comparing the currently measured first ten natural frequencies with the natural frequencies in the reference state (such as when the bridge is not affected by an earthquake or in normal operation), the frequency change rate can be obtained. The frequency change rate can intuitively reflect the change in the structural state of the bridge pier. If the frequency change rate is large, it may mean that the structure of the bridge pier has been damaged. Therefore, it is also an important damage-sensitive feature.

[0130] Determination of damage-sensitive features: The warping deformation of the track slab and the change rate of the first ten natural frequencies of the bridge pier are regarded as damage-sensitive features. These two features reflect the damage condition of the bridge structure from different aspects. The warping deformation of the track slab reflects the local damage of the superstructure of the bridge, while the change rate of the natural frequency of the bridge pier reflects the damage of the bridge support structure. Together, they provide key feature data for the rapid assessment of the seismic damage performance of high-speed railway bridges based on machine learning, which helps to more accurately evaluate the damage degree and location of the bridge under earthquake action.

[0131] Example 6:

[0132] Based on the method for rapid assessment of the seismic damage performance of high-speed railway bridges based on machine learning in Example 5, the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers are calculated based on various strain components in the strain contour map, including:

[0133] Establish a calculation model for warping deformation based on strain components and calculation models for the damage of each component;

[0134] Denoise and handle outliers of the strain contour map to obtain a preprocessed strain contour map;

[0135] Input the preprocessed strain contour map into the warping deformation calculation model to obtain the warping deformation of the track slab, and input the preprocessed strain contour map into the calculation models for the damage of each component to obtain the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers.

[0136] In this example:

[0137] Establish a calculation model for warping deformation based on strain components: Under the action of external forces such as earthquakes, the internal strain state of the track slab is complex and diverse, and strain components in different directions and types jointly affect the warping deformation of the track slab. Therefore, a special calculation model needs to be established to comprehensively consider the relationship between these strain components and the warping deformation. This model may be based on theories such as material mechanics and structural mechanics, and through mathematical formulas and algorithms, various strain components (such as normal strain, shear strain, etc.) in the strain contour map are used as input parameters to calculate the warping deformation of the track slab. For example, according to the principles of elasticity, combined with the geometric shape, material properties of the track slab, and the interaction relationship between strain components, a corresponding mathematical model is constructed to accurately describe the conversion process from strain to warping deformation.

[0138] Establish calculation models for the damage amounts of various components: Similarly, based on the strain components in the strain cloud diagram, models for calculating the damage amounts of the interlayer connectors of the track, bearings, and bridge piers are established respectively. These models consider the damage mechanisms of each component under different strain states. For example, the interlayer connectors of the track may experience damage such as loosening and fracture under specific strain combinations. The model calculates its damage amount by analyzing the strain information in the area related to the connectors in the strain cloud diagram and combining the mechanical properties and structural characteristics of the connectors. For bearings and bridge piers, based on similar principles, the damage amounts are calculated using strain components according to their respective structural functions and force characteristics.

[0139] Denoising and outlier processing of the strain cloud diagram include:

[0140] Denoising: During the actual acquisition and generation of the strain cloud diagram, various noises will inevitably be introduced. These noises may come from measurement errors of sensors, environmental interference, or errors in the image processing process, etc. The noises will interfere with the accurate acquisition of the real strain information and affect the accuracy of calculating the warping deformation amount of the track slab. Therefore, appropriate denoising methods, such as filtering techniques (such as Gaussian filtering, median filtering, etc.), are needed to remove the noise signals in the strain cloud diagram, make the strain cloud diagram smoother, and highlight the real strain change trend.

[0141] Outlier processing: In addition to noises, there may also be outliers in the strain cloud diagram. These outliers may be caused by sensor failures, data transmission errors, or local sudden interferences, etc. The outliers will seriously affect the data analysis results. If not processed, it may cause a large deviation in the calculated warping deformation amount of the track slab. By setting a reasonable threshold range or using statistical methods (such as methods based on the mean and standard deviation), these outliers are identified and corrected or removed, so as to obtain the preprocessed strain cloud diagram, providing reliable data for accurately calculating the warping deformation amount of the track slab subsequently.

[0142] Input the preprocessed strain cloud diagram into the warping deformation amount calculation model: After denoising and outlier processing, the data quality of the strain cloud diagram is significantly improved, and it can more accurately reflect the actual strain state of the track slab. Input such a strain cloud diagram into the previously established warping deformation amount calculation model. The model analyzes and calculates each strain component according to the set algorithms and formulas, and finally outputs the result of the warping deformation amount of the track slab. This result can more accurately reflect the warping degree of the track slab under the current strain state, providing an important basis for evaluating the damage condition of the bridge track slab under earthquake action. For example, if the calculated warping deformation amount of the track slab exceeds a certain safety threshold, it indicates that the track slab may have suffered damage that affects the safe operation of the train.

[0143] The preprocessed strain nephogram is simultaneously input into the damage quantity calculation models of each component. According to their respective set algorithms, these models combine the strain information in the strain nephogram and the structural and mechanical characteristics of each component to calculate the damage quantity of the track interlayer connectors, the damage quantity of the bearings, and the damage quantity of the bridge piers respectively. These damage quantity data can help the evaluators comprehensively understand the damage degree of each key component of the bridge and provide a quantitative basis for formulating targeted maintenance and repair strategies. For example, by comparing the damage quantity of the bridge piers calculated at different times, the development trend of the bridge pier damage can be judged, and corresponding measures can be taken in time to ensure the safe operation of the bridge.

[0144] Example 7:

[0145] Based on the method for rapid evaluation of seismic damage performance of high-speed railway bridges based on machine learning in Example 5, modal analysis is performed on the spatio-temporally aligned source data to extract the first ten-order frequency change rates of the bridge piers, including:

[0146] Extract the vibration response signal of the bridge pier from the spatio-temporally aligned source data, and separate a part of the vibration response signal of the bridge pier within the frequency band of interest in the pry response signal through band-pass filtering;

[0147] Perform spectral analysis on a part of the vibration response signal of the bridge pier within the frequency band of interest using the frequency domain decomposition method to identify the first ten-order natural frequencies;

[0148] Based on the first ten-order natural frequencies and the first ten-order natural frequencies of the bridge pier in the reference state, calculate the first ten-order frequency change rates of the bridge pier.

[0149] In this embodiment:

[0150] Extract and filter the vibration response signal of the bridge pier: First, extract the vibration response signal of the bridge pier from the spatio-temporally aligned source data, and these signals contain the vibration information generated by the bridge pier under various external forces (such as earthquakes, train running, etc.). However, the original vibration response signals often contain components in multiple frequency bands, and some of these frequency bands may have nothing to do with the key vibration characteristics or damage conditions of the bridge pier, and may even interfere with the extraction of effective information. Therefore, through band-pass filtering technology, a part of the vibration response signal of the bridge pier within the frequency band of interest is separated from the original vibration response signal of the bridge pier. The band-pass filter can set a specific frequency range, allowing only the signals within this range to pass through and blocking the signals of other frequencies. For example, it is known that certain natural vibration frequencies of the bridge pier are concentrated in a specific frequency band. By setting the band-pass filter, only the vibration response signals within this frequency band are retained, thus highlighting the information related to the key vibration characteristics of the bridge pier and laying a foundation for accurately analyzing the state of the bridge pier in the follow-up.

[0151] Spectrum analysis to identify the first ten natural frequencies: The frequency domain decomposition method is used to perform spectrum analysis on the vibration response signals of some bridge piers within the frequency band of interest after band-pass filtering. The frequency domain decomposition method is an effective means of converting time-domain signals to the frequency domain for analysis, which can reveal the energy distribution of signals at different frequencies. Through this analysis method, the first ten natural frequencies of the bridge piers can be identified from complex vibration response signals. The natural frequency is an inherent property of the bridge pier structure, which depends on factors such as the geometric shape, material properties, and boundary conditions of the bridge pier. Under normal circumstances, the bridge pier has specific natural frequency values. When the bridge pier structure is damaged, its natural frequency often changes. Therefore, accurately identifying the natural frequency is crucial for evaluating the damage state of the bridge pier.

[0152] Calculate the change rate of the first ten frequencies: After identifying the first ten natural frequencies of the bridge pier in the current state (i.e., f 当前 ), compare them with the first ten natural frequencies of the bridge pier in the reference state (i.e., f 基准 ):

[0153]

[0154] The reference state usually refers to the situation when the bridge pier is not damaged or in a normal working state, and the natural frequency at this time is used as a reference standard. By calculating the difference between the current first ten natural frequencies and the corresponding frequencies in the reference state, and normalizing the difference with the reference frequency, the change rate of the first ten frequencies of the bridge pier is obtained. This frequency change rate can intuitively reflect the degree of change in the bridge pier structure state. For example, if the change rate of a certain order frequency is large, it may mean that there is damage or change in the structural part related to the vibration mode of this order of the bridge pier, providing a quantitative key index for evaluating the seismic damage of the bridge pier, helping to detect potential structural problems in a timely manner, and providing an important basis for the safety assessment of high-speed railway bridges.

[0155] Example 8:

[0156] Based on the method for rapid assessment of seismic damage performance of high-speed railway bridges based on machine learning in Example 3, a multi-modal feature space is constructed based on some features of time-frequency domain features, spatial features, and probability features, referring to Figure 2 , including:

[0157] Visualize the feature contribution status of time-frequency domain features, spatial features, and probability features based on the SHAP value analysis method;

[0158] Based on the recursive feature elimination method and the feature contribution status of time-frequency domain features, spatial features, and probability features, retain the part of the features with a contribution degree not less than the contribution degree threshold in the time-frequency domain features, spatial features, and probability features and perform feature dimensionality reduction processing to obtain a multi-modal feature space.

[0159] In this embodiment:

[0160] Visualize the feature contribution status based on the SHAP value analysis method (such as the influence coefficient of bearing stiffness on the damage probability): The SHAP (SHapley Additive exPlanations) value analysis method is a technique used to explain the prediction results of machine learning models. It can evaluate the contribution degree of each feature to the model output. During the construction of the multi-modal feature space, this method is used to analyze time-frequency domain features, spatial features, and probability features, so as to clearly understand the importance of each feature in describing the seismic damage performance of bridges. For example, through SHAP value analysis, it may be found that some time-frequency domain features (such as the energy distribution in a specific frequency band) have a high contribution to predicting the probability of a certain type of damage to the bridge, while some spatial features (such as the stiffness change in a specific part) play a significant role in reflecting the overall structural stability of the bridge. By visually displaying these feature contribution states, presenting the magnitude and sign of the SHAP value of each feature in the form of charts or graphs, the relative importance of each feature in the model can be intuitively shown, providing a basis for subsequent feature screening. This helps to understand the influence mechanism of different types of features on the seismic damage performance assessment of bridges, which features are key and which features have relatively little influence.

[0161] Feature screening and dimensionality reduction based on the recursive feature elimination method include:

[0162] Recursive feature elimination method: This is a method of gradually eliminating features with less contribution to the model performance. Combining the feature contribution status obtained from the SHAP value analysis, a contribution threshold (thirty percent) is set. For time-frequency domain features, spatial features, and probability features, only part of the features with a contribution degree not less than this threshold are retained. For example, if the SHAP value of a certain time-frequency domain feature is low, indicating that its contribution to the seismic damage performance assessment model of the bridge is small, it may be eliminated in this step.

[0163] Feature dimensionality reduction processing: While retaining important features, this screening process actually also achieves feature dimensionality reduction. Feature dimensionality reduction is of great significance. On the one hand, too many redundant or unimportant features may increase the computational amount and complexity of model training, and even lead to overfitting, affecting the generalization ability of the model; on the other hand, retaining key features can more prominently highlight the factors that actually affect the seismic damage performance of the bridge, making the model more concise and efficient. In this way, a multi-modal feature space is finally obtained. This space contains both the most important feature information for the seismic damage performance assessment of the bridge and avoids the interference of too many irrelevant or redundant features, which helps to improve the accuracy and efficiency of the seismic damage performance assessment model established based on this multi-modal feature space.

[0164] The above methods can improve the inter-class distance of features, the cross-dataset test accuracy rate, and reduce the 95% confidence interval width.

[0165] Example 9:

[0166] Based on the method for rapid assessment of seismic damage performance of high-speed railway bridges using machine learning in Example 1, S3: Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism, and establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture. Refer to Figure 3 , including:

[0167] Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism;

[0168] Construct a multi-modal feature space for a large number of samples (such as 100,000+ samples), and generate a system state vector for each sample based on the multi-modal feature space of each sample;

[0169] Establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture and the system state vectors of all samples.

[0170] In this example:

[0171] Constructing the hierarchical machine learning architecture includes:

[0172] Fuse multiple models and mechanisms: This step aims to build a comprehensive and multi-level machine learning architecture that integrates the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism.

[0173] Dynamic Bayesian network damage propagation model: The dynamic Bayesian network is good at dealing with information with time series characteristics and uncertainty, and can effectively analyze the propagation process of bridge damage in the structure. It can infer how damage propagates from one part to other parts based on the changes in the bridge structure state at different time points, as well as the transition probabilities between different damage states. For example, during an earthquake, the dynamic Bayesian network can simulate the process of damage propagating from the bridge pier to the bridge deck, and the possibility of damage state changes at each stage.

[0174] Multi-scale CNN-LSTM Hybrid Model: Convolutional Neural Network (CNN) is good at capturing spatial features in data, while Long Short-Term Memory Network (LSTM) performs excellently in dealing with long-term dependencies in time series data. Combining the two to form a multi-scale CNN-LSTM hybrid model can extract spatio-temporal features of bridge-related data from different scales. For example, CNN can extract local spatial features from image data of bridge structures (such as speckle images), and LSTM can process time series data such as bridge vibration responses to learn their long-term change trends. The combination of the two helps to comprehensively mine complex patterns in bridge data.

[0175] Uncertainty Quantification and Dynamic Update Mechanism: Considering the uncertainty of earthquakes and the variability of bridge structure responses, an uncertainty quantification and dynamic update mechanism is introduced. This mechanism can quantify various uncertainty factors in the model, such as the uncertainty of earthquake parameters, sensor measurement errors, etc. At the same time, the model is dynamically updated according to new data and actual situations, enabling the model to better adapt to changing conditions and maintain the accuracy of evaluation.

[0176] The generated sample system state vector includes:

[0177] Constructing Multi-modal Feature Space Samples: First, construct the multi-modal feature space of a large number of samples, which cover various states of bridges under different earthquake scenarios. The multi-modal feature space includes time-frequency domain features, spatial features, probability features, etc., comprehensively describing the characteristics of bridges under earthquake actions from multiple dimensions. For example, different samples may correspond to the responses of bridges under earthquakes of different intensities, and the multi-modal feature space of each sample includes corresponding time-frequency domain features (such as energy distribution in different frequency bands), spatial features (such as changes in bridge structure stiffness, damage-sensitive features), and probability features (probability distribution considering earthquake uncertainty and measurement errors).

[0178] Generating System State Vectors: Generate system state vectors based on the multi-modal feature space of each sample. The system state vector integrates and encodes various features in the multi-modal feature space to form a vector that can represent the state of the bridge in this sample. This vector contains key information about the bridge under a specific earthquake scenario, providing a unified data format for subsequent model learning and evaluation. For example, arrange the time-frequency domain features, spatial features, and probability features in a certain order or fuse them into a vector through a certain algorithm as the system state vector of the bridge in this sample.

[0179] Establishing the Earthquake Damage Performance Evaluation Model includes:

[0180] Basis of Model Training: Use the constructed hierarchical machine learning architecture and the system state vectors of all samples to establish a seismic damage performance evaluation model. The hierarchical machine learning architecture provides a powerful learning framework, while the system state vectors of a large number of samples provide rich data for model training.

[0181] Learning and Modeling: The model learns from these sample system state vectors, discovers the laws and patterns contained therein, and establishes a mapping relationship between the input features (system state vectors) and the output results (seismic damage performance of bridges). For example, by continuously adjusting the parameters of the model, the model can accurately predict the damage probability distribution, damage level and other seismic damage performance indicators of the bridge under the corresponding seismic scenarios based on the input system state vectors, so as to effectively evaluate the seismic damage performance of high-speed railway bridges.

[0182] Example 10:

[0183] Based on the method for rapid evaluation of seismic damage performance of high-speed railway bridges based on machine learning in Example 9, an earthquake damage performance evaluation model is established based on a hierarchical machine learning architecture and the system state vectors of all samples, including:

[0184] Based on the dynamic Bayesian network damage propagation model in the hierarchical machine learning architecture, the system state vectors of all samples are decomposed into multiple damage levels, and a conditional damage probability matrix of all samples is generated based on the system state vectors of all samples and the corresponding damage levels;

[0185] Define the transition probabilities of the local structural damage states at various locations of the bridges in all samples based on the conditional damage probability matrix of each sample;

[0186] Model the bridge damage propagation path based on a discrete-time Markov chain, and calculate the joint probability of the bridge damage propagation path of each sample based on the transition probabilities of the local structural damage states at various locations of the bridge in each sample;

[0187] Based on the multi-scale CNN-LSTM hybrid model in the hierarchical machine learning architecture, perform multi-scale fusion on the joint probability of the bridge damage propagation path of each sample to obtain the damage probability distribution of the bridge of each sample;

[0188] At the same time, obtain the uncertainty quantification result based on the uncertainty quantification and dynamic update mechanism, and drive the model update frequency based on the uncertainty quantification result until the seismic damage performance evaluation model is established.

[0189] In this embodiment:

[0190] Damage level decomposition: Using the dynamic Bayesian network damage propagation model in the hierarchical machine learning architecture, analyze the system state vectors of all samples and decompose them into multiple damage levels. The system state vector contains comprehensive information extracted from the multi-modal feature space of the bridge under specific seismic scenarios. The dynamic Bayesian network judges the damage degree of the bridge based on this information and divides it into different levels, such as slight damage, moderate damage, severe damage, and complete damage, etc. This step lays the foundation for subsequent quantification of damage probability. For example, by analyzing the characteristics such as strain, displacement, and frequency change of the bridge structure in the system state vector, determine which damage level the bridge belongs to.

[0191] Conditional damage probability matrix generation: Based on the system state vectors of all samples and the corresponding damage levels, generate a conditional damage probability matrix. This matrix reflects the probability that the bridge is in different damage levels under the condition of a given system state vector. For example, for a specific system state vector, the conditional damage probability matrix will give the probability that the bridge is in the slight damage level, the probability of the moderate damage level, etc. This matrix quantifies the probability relationship between the system state and the damage level, which helps to further analyze the possibility of bridge damage. The formula is:

[0192]

[0193] In the formula, P(S t+1 |S t ,D t ) is the conditional damage probability matrix, S t is the system state vector at time t (including the damage states of n components), S t+1 is the system state vector at time t+1 (including the damage states of n components), D t is the set of parent node damage states at time t, D i,t+1 is the damage level of component i at time t, P(D i,t+1 |D pa(i),t+1 ) is the conditional probability table based on the parent node damage;

[0194] Define the local structure damage state transition probability: Based on the conditional damage probability matrix of each sample, define the transition probability of the local structure damage states at various locations of the bridge in all samples. The bridge consists of multiple local structures. Under the action of an earthquake, the damage states of different local structures will change and affect each other. Through the conditional damage probability matrix, the probability that a local structure changes from one damage state to another can be determined. For example, the probability of changing from the undamaged state of the pier to the slight damage state, or the probability of changing from the slight damage of the bridge deck to the moderate damage, etc. These transition probabilities describe the possibility of local structure damage propagation of the bridge and provide key parameters for subsequent modeling of the damage propagation path.

[0195] Discrete-time Markov chain modeling: A discrete-time Markov chain is used to model the bridge damage propagation path. A Markov chain is a stochastic process. In this model, it is assumed that the damage state of the bridge at a certain moment depends only on the damage state at the previous moment and is independent of earlier states. Based on this assumption, combined with the locally-structural damage state transition probabilities defined earlier, a propagation path model of bridge damage between different local structures and different time points can be constructed. For example, simulate how damage propagates from one part of the bridge to other parts successively during the duration of an earthquake.

[0196] Calculating the joint probability: Based on the transition probabilities of the locally-structural damage states at various locations of the bridge in each sample, calculate the joint probability of the bridge damage propagation path for each sample. The joint probability represents the probability that the damage states of all local structures of the bridge occur simultaneously under a specific damage propagation path. By calculating the joint probability, the likelihood of different damage propagation paths can be evaluated, thus providing a more comprehensive understanding of the damage evolution of the bridge during an earthquake. For example, calculate the joint probability of the propagation path starting from a slightly damaged pier and successively propagating to different damage states at different parts of the bridge deck to determine the likelihood of this path occurring during an actual earthquake.

[0197] Obtaining the damage probability distribution through multi-scale fusion: Use a multi-scale CNN-LSTM hybrid model in a hierarchical machine learning architecture to perform multi-scale fusion on the joint probability of the bridge damage propagation path for each sample. The multi-scale CNN-LSTM hybrid model can extract and analyze features from different scales. In this step, it processes the joint probability at different scales, for example, from the macroscopic scale of the overall bridge damage propagation to the microscopic scale of the details of local structural damage, fully mining the information therein. Through multi-scale fusion, the damage probability distribution of the bridge for each sample is obtained. This distribution shows the probability of the bridge at different damage levels, providing intuitive and quantitative results for evaluating the seismic damage performance of the bridge. For example, obtain that the probability of the bridge having slight damage is X%, moderate damage is Y%, severe damage is Z%, etc.

[0198] Principle of multi-scale fusion:

[0199] F = α·F low + β·F high

[0200] In the formula, F is the feature obtained after multi-scale fusion, F low is the low-resolution feature (with a large receptive field), F high is the high-resolution feature (rich in details), and α and β are adaptive fusion weights.

[0201] The model update based on uncertainty quantification includes:

[0202] Uncertainty quantification: By using uncertainty quantification and dynamic update mechanisms, uncertainty quantification results are obtained. During the earthquake occurrence process and the response of bridge structures, there are many uncertain factors, such as the randomness of the earthquake itself, sensor measurement errors, and the uncertainty of model parameters. The uncertainty quantification mechanism analyzes and quantifies these factors to obtain results that can reflect the degree of model uncertainty. For example, through the uncertainty analysis of earthquake parameters, the error range of the model prediction results is given.

[0203] Driving model update: Based on the uncertainty quantification results, the model update frequency is driven. If the uncertainty quantification results show a high level of uncertainty in the model, for example, a large error range in the prediction results, it indicates that the model may need to be updated more frequently to adapt to changes in the actual situation. By continuously adjusting the model parameters according to new data and uncertainty quantification results and optimizing the model structure, until a stable and accurate earthquake damage performance evaluation model is established. This process ensures that the model can always maintain a high evaluation accuracy when facing complex and changing actual situations, providing a reliable tool for the earthquake damage performance evaluation of high-speed railway bridges. The formula is:

[0204] τ = γ·max(U1, U2, U3)

[0205] In the formula, τ is the model update frequency, γ is the threshold coefficient, and U1, U2, U3 are the uncertainties of each module respectively.

[0206] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A rapid evaluation method for the seismic damage performance of high-speed railway bridges based on machine learning, characterized in that Including: S1: Real-time collect multi-source data of the bridge structure of high-speed railways through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system, and obtain the historical earthquake database and the BIM model of the bridge structure; S2: Combine the multi-source data with the historical earthquake database and the BIM model of the bridge structure to construct a multi-modal feature space containing partial features of time-frequency domain features, spatial features, and probability features; S3: Build a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism, and establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture; S4: Use the seismic damage performance evaluation model to evaluate the seismic damage performance of the multi-modal feature space, and generate the damage probability distribution of high-speed railway bridges.

2. The rapid assessment method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 1, characterized in that S1: Real-time collect multi-source data of the bridge structure of high-speed railways through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system, and obtain the historical earthquake database and the BIM model of the bridge structure, including: Real-time collect the track slab deformation data, seismic wave response data, overall bridge displacement, and speckle images of each local structure under different intensity earthquakes of the bridge structure of high-speed railways as multi-source data of the bridge structure of high-speed railways through numerous distributed sensor arrays deployed in the heterogeneous data acquisition system; Obtain the historical earthquake database and the BIM model of the bridge structure, where the BIM model of the bridge structure includes bridge set parameters, material properties, and connection methods.

3. The rapid evaluation method for the seismic damage performance of high-speed railway bridges based on machine learning according to claim 1, characterized in that S2: Combine the multi-source data with the historical earthquake database and the BIM model of the bridge structure to construct a multi-modal feature space containing partial features of time-frequency domain features, spatial features, and probability features, including: Perform spatio-temporal alignment of the multi-source data and the BIM model of the bridge structure based on the spatio-temporal alignment algorithm to obtain spatio-temporally aligned source data; Extract time-frequency domain features and spatial features from the spatio-temporally aligned source data; Generate parameter perturbation samples based on the historical earthquake database and introduce Monte Carlo simulation, and then obtain probability features by fusing sensor measurement errors and model parameter uncertainties based on the evidence theory; Construct a multi-modal feature space based on partial features of time-frequency domain features, spatial features, and probability features.

4. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 3, wherein Extract time-frequency domain features and spatial features from the spatio-temporally aligned source data, including: Extract the energy distribution of a preset frequency band from the multi-source data aligned in the spatio-temporally aligned source data through short-time Fourier transform, and then decompose it through Hilbert-Huang transform to obtain multiple intrinsic mode functions as time-frequency domain features; Construct the bridge structure stiffness matrix based on the BIM model of the bridge structure aligned in the spatio-temporally aligned source data; Extract damage-sensitive features from the spatio-temporally aligned source data; Among them, the spatial features include the bridge structure stiffness matrix and damage-sensitive features.

5. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 4, characterized in that, Extract damage-sensitive features from the spatio-temporally aligned source data, including: Perform preset pixel-level registration on the speckle images of each local structure under different intensity earthquakes in the multi-source data aligned in the spatio-temporally aligned source data based on the phase correlation method, and construct a full-field displacement vector field based on all the speckle images after pixel-level registration; Construct a two-dimensional Delaunay triangular mesh based on the full-field displacement vector field, and calculate the Green's strain tension by statistically analyzing the displacement gradients of the mesh nodes to generate strain nephograms at corresponding levels. Calculate the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers based on various strain components in the strain nephogram. Perform modal analysis on the spatio-temporally aligned source data to extract the first ten order frequency change rates of the bridge piers. Regard the warping deformation of the track slab and the first ten order frequency change rates of the bridge piers as damage-sensitive features.

6. The rapid assessment method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 5, characterized in that Calculate the warping deformation of the track slab, the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers based on various strain components in the strain nephogram, including: Establish a warping deformation calculation model based on strain components and damage calculation models for each component. Denoise and process outliers of the strain nephogram to obtain a preprocessed strain nephogram. Input the preprocessed strain nephogram into the warping deformation calculation model to obtain the warping deformation of the track slab, and input the preprocessed strain nephogram into the damage calculation models for each component to obtain the damage of the track layer connectors, the damage of the bearings, and the damage of the bridge piers.

7. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 5, characterized in that Perform modal analysis on the spatio-temporally aligned source data to extract the first ten order frequency change rates of the bridge piers, including: Extract the vibration response signal of the bridge pier from the spatio-temporally aligned source data, and separate the partial bridge pier vibration response signal within the frequency band of interest in the pry response signal through band-pass filtering. Use the frequency domain decomposition method to perform spectral analysis on the partial bridge pier vibration response signal within the frequency band of interest to identify the first ten order natural frequencies. Calculate the first ten order frequency change rates of the bridge pier based on the first ten order natural frequencies and the first ten order natural frequencies of the bridge pier in the reference state.

8. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 3, characterized in that Construct a multi-modal feature space based on partial features of time-frequency domain features, spatial features, and probability features, including: Visualize the feature contribution status of time-frequency domain features, spatial features, and probability features based on the SHAP value analysis method. Based on the recursive feature elimination method and the feature contribution status of time-frequency domain features, spatial features, and probability features, retain the partial features with a contribution degree not less than the contribution degree threshold in time-frequency domain features, spatial features, and probability features and perform feature dimensionality reduction processing to obtain a multi-modal feature space.

9. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 1, characterized in that, S3: Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism, and establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture, including: Construct a hierarchical machine learning architecture based on the dynamic Bayesian network damage propagation model, the multi-scale CNN-LSTM hybrid model, and the uncertainty quantification and dynamic update mechanism. Construct a multi-modal feature space for a large number of samples, and generate a system state vector for each sample based on the multi-modal feature space of each sample. Establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture and the system state vectors of all samples.

10. The rapid evaluation method for seismic damage performance of high-speed railway bridges based on machine learning according to claim 9, characterized in that Establish a seismic damage performance evaluation model based on the hierarchical machine learning architecture and the system state vectors of all samples, including: Based on the dynamic Bayesian network damage propagation model in the hierarchical machine learning architecture, the system state vectors of all samples are decomposed into multiple damage levels, and a conditional damage probability matrix of all samples is generated based on the system state vectors of all samples and the corresponding damage levels; Based on the conditional damage probability matrix of each sample, the transition probabilities of the local structural damage states at various locations of the bridge in all samples are defined; Based on the discrete-time Markov chain, the bridge damage propagation path is modeled, and the joint probability of the bridge damage propagation path of each sample is calculated based on the transition probabilities of the local structural damage states at various locations of the bridge in each sample; Based on the multi-scale CNN-LSTM hybrid model in the hierarchical machine learning architecture, multi-scale fusion is performed on the joint probability of the bridge damage propagation path of each sample to obtain the damage probability distribution of the bridge of each sample; Meanwhile, based on the uncertainty quantification and dynamic update mechanism, the uncertainty quantification result is obtained, and the model update frequency is driven based on the uncertainty quantification result until the seismic damage performance evaluation model is established.

Images