Method and system for analyzing mechanical performance of fabricated bridge and culvert joint based on neural network

Abstract

The application provides a kind of based on neural network's fabricated bridge and aqueduct node mechanical performance analysis method and system, it is related to data processing technical field, including obtaining the real-time monitoring data of fabricated bridge and aqueduct node, the real-time monitoring data includes strain data, vibration data and load data;Real-time monitoring data of the fabricated bridge and aqueduct node is carried out feature extraction and optimization, and nonlinear feature set is obtained;The nonlinear feature set is carried out mechanical performance evaluation, and real-time mechanical performance evaluation result is obtained;Based on the real-time mechanical performance evaluation result and preset historical monitoring data, model construction and optimization are carried out, and the optimized mechanical performance change trend prediction model is obtained;The real-time monitoring data is sent to the optimized mechanical performance change trend prediction model for prediction, and the mechanical performance of fabricated bridge and aqueduct node in the preset time period is obtained.The application improves the precision, timeliness and reliability of the fabricated bridge and aqueduct node mechanical performance monitoring.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks. Background Technology

[0002] With the rapid development of urban infrastructure, prefabricated bridges and culverts, as an important form of underground structure, are widely used in bridge engineering. As a key component, the joints of prefabricated bridges and culverts bear significant mechanical loads and are easily affected by factors such as geological changes, traffic loads, and construction quality. Therefore, accurately assessing the mechanical properties of prefabricated bridge and culvert joints and carrying out timely reinforcement and maintenance have become important issues for ensuring project safety and extending the service life of the structure.

[0003] Currently, many engineering projects employ traditional monitoring methods, such as manual inspections and conventional instrument measurements. These methods typically rely on human experience and lack the ability for real-time monitoring and dynamic assessment. Although some sensor-based monitoring systems have been proposed, existing technologies largely depend on simple data acquisition and analysis methods, which cannot effectively handle complex mechanical performance data and nonlinear characteristics, nor can they timely and accurately assess the long-term changing trends and damage risks of structures. This results in many potential structural problems failing to be detected in their early stages, thereby increasing maintenance costs and safety hazards.

[0004] Therefore, there is an urgent need for a neural network-based method and system for analyzing the mechanical performance of prefabricated bridge and culvert nodes to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for analyzing the mechanical performance of prefabricated bridge and culvert joints based on neural networks, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0006] Firstly, this application provides a method for analyzing the mechanical performance of prefabricated bridge and culvert joints based on neural networks, including:

[0007] Acquire real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data, and load data;

[0008] Feature extraction and optimization are performed on the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set;

[0009] The nonlinear feature set is subjected to mechanical performance evaluation to obtain real-time mechanical performance evaluation results;

[0010] Based on the real-time mechanical performance evaluation results and the preset historical monitoring data, the model is constructed and optimized to obtain the optimized mechanical performance change trend prediction model.

[0011] The real-time monitoring data is sent to the optimized mechanical performance change trend prediction model for prediction, and the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period is obtained.

[0012] Secondly, this application also provides a neural network-based system for analyzing the mechanical performance of prefabricated bridge and culvert joints, including:

[0013] The acquisition unit is used to acquire real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data and load data.

[0014] The processing unit is used to extract and optimize the features of the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set;

[0015] An evaluation unit is used to evaluate the mechanical properties of the nonlinear feature set and obtain real-time mechanical property evaluation results.

[0016] The construction unit is used to build and optimize the model based on the real-time mechanical performance evaluation results and preset historical monitoring data to obtain an optimized mechanical performance change trend prediction model.

[0017] The prediction unit is used to send the real-time monitoring data to the optimized mechanical performance change trend prediction model for prediction, so as to obtain the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention achieves efficient extraction and prediction of the mechanical properties of prefabricated bridge and culvert joints by collecting real-time strain, vibration, and load data and combining algorithms such as deep belief networks, gradient boosting machines, and long short-term memory networks. Specifically, this invention optimizes the monitoring data using a feature extraction model, evaluates mechanical properties using nonlinear feature sets, and constructs a predictive model for mechanical property change trends based on historical data and real-time monitoring results, ultimately achieving an accurate assessment of the long-term health status of prefabricated bridge and culvert joints. Furthermore, this invention combines adversarial networks, isolated forest algorithms, and variational autoencoders for anomaly detection and damage analysis, and optimizes reinforcement strategies based on fuzzy rule reasoning and multi-objective optimization functions to ensure maximum utilization of reinforcement resources. This invention effectively overcomes the shortcomings of existing technologies and improves the accuracy, timeliness, and reliability of mechanical property monitoring for prefabricated bridge and culvert joints.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the process for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks, as described in this embodiment of the invention.

[0023] Figure 2 This is a schematic diagram of the structure of the neural network-based prefabricated bridge and culvert node mechanical performance analysis system described in this embodiment of the invention.

[0024] In the diagram: 701, acquisition unit; 702, processing unit; 703, evaluation unit; 704, construction unit; 705, prediction unit. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] Example 1:

[0028] This embodiment provides a method for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks.

[0029] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0030] Step S1: Obtain real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data and load data;

[0031] Understandably, this step first involves collecting real-time monitoring information across multiple dimensions, including strain, vibration, and load data, from sensors deployed on the bridge and culvert nodes. These prefabricated bridge and culvert nodes typically refer to the connecting joints between different parts of the prefabricated bridge and culvert structure. In the construction of prefabricated bridges and culverts, the bridge and culvert are assembled on-site from multiple prefabricated components (such as box slabs, cover plates, and sidewalls) through connection nodes. These connections (i.e., nodes) are crucial parts of the structure, responsible for transmitting forces, stresses, and loads between different components. This data represents the physical response of the prefabricated bridge and culvert nodes under different working conditions, reflecting the dynamic changes of the structure under external loads and internal forces. Strain data helps monitor material deformation and thus infer the stress state of the structure; vibration data reveals abnormal vibration modes that may exist during operation, such as crack propagation or loosening at node connections; and load data primarily reflects the magnitude and direction of external forces acting on the bridge and culvert nodes.

[0032] Step S2: Extract and optimize the features from the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set;

[0033] Understandably, this step extracts representative nonlinear features from real-time monitoring data of prefabricated bridge and culvert nodes. These features accurately reflect changes in the structure's mechanical properties. To address the complex mechanical behavior and response of bridge and culvert nodes, we need to employ appropriate algorithms for feature extraction and optimize these features to ensure the accuracy and predictability of subsequent analyses. In this step, step S2 includes steps S21, S22, S23, and S24.

[0034] Step S21: Perform factor analysis based on the real-time monitoring data of prefabricated bridge and culvert nodes to obtain a set of mechanical characteristic factors for prefabricated bridge and culvert nodes.

[0035] Understandably, this step extracts a small but representative set of mechanical characteristic factors from the real-time monitoring data of prefabricated bridge and culvert nodes. These factors can reveal the main mechanical behaviors of prefabricated bridge and culvert nodes, such as the potential relationships between multi-dimensional data like strain, vibration modes, and loads. Since the real-time monitoring data of prefabricated bridge and culvert nodes often includes multiple complex interactions, factor analysis can effectively reduce the dimensionality of the data and reveal the main influencing factors within it.

[0036] First, based on the standardized monitoring data, the covariance matrix between each dimension of the data is calculated. The covariance matrix describes the linear relationship between the various data dimensions. For monitoring data of prefabricated bridge and culvert nodes, the covariance matrix helps us identify the relationships between variables such as strain, vibration, and load.

[0037] Next, eigenvalue decomposition is performed, decomposing the covariance matrix into eigenvalues ​​and eigenvectors. Eigenvalues ​​reflect the variance of the data along the corresponding eigenvector direction, while eigenvectors represent the direction of each factor. Larger eigenvalues ​​correspond to dominant factors, while smaller eigenvalues ​​correspond to noise or less important factors. For example, eigenvalue decomposition can extract factors related to the mechanical performance of prefabricated bridge and culvert joints, which may include stress, deformation, and vibration modes. Each factor corresponds to an eigenvector, representing a combination of different dimensions of the monitoring data.

[0038] Then, the variance maximization rotation method is used to rotate the factors to obtain the mechanical characteristic factors. Through factor analysis, the original data is transformed from multidimensional to a few key factors, which simplifies the complexity of subsequent analysis.

[0039] Step S22: Unsupervised learning of the set of mechanical feature factors is performed on the autoencoder of the deep belief network to obtain the nonlinear features of the set of mechanical feature factors.

[0040] It is understandable that in this step, the input data is the set of mechanical feature factors obtained from factor analysis. These factors include various mechanical features such as stress, deformation, and vibration modes. The data is usually high-dimensional and may have complex nonlinear relationships. The autoencoder will progressively compress and learn low-dimensional representations of these features through multiple hidden layers;

[0041] The primary task of the encoder is to map the input data into a low-dimensional space. In this process, the encoder performs a non-linear mapping of the data through multiple neural network layers, outputting compressed hidden features, known as "encoders." The encoder's goal is to learn a low-dimensional feature that effectively represents the original data by minimizing the reconstruction error between the input and output. For example, assuming the input data consists of features of strain and vibration modes, the encoder network compresses this data into a low-dimensional representation containing crucial information about the node characteristics.

[0042] The decoder's task is to recover the input data from the encoded low-dimensional representation. Through the decoder, information loss during the encoding process can be checked and optimized to make the reconstructed data as close as possible to the original input. The decoder's output is the non-linear feature learned through inverse propagation.

[0043] The training objective of an autoencoder is to minimize the error between the input data and the reconstructed data. A commonly used error metric is mean squared error (MSE), which calculates the difference between the reconstructed and input data. The network parameters are continuously adjusted using backpropagation to optimize the data representation. In deep belief networks, the ReLU activation function in the hidden layers effectively captures nonlinear relationships in the input data. For example, when processing strain and vibration data, these activation functions help the network learn the stress distribution patterns of nodes under high pressure or the deformation trends that may occur during long-term use.

[0044] Furthermore, assuming that under certain working conditions, the strain data of prefabricated bridge and culvert joints exhibits a nonlinear growth trend, and traditional linear analysis methods cannot capture this change. Through an autoencoder, the model can learn the nonlinear characteristics in the strain data, thus providing more accurate input data for subsequent mechanical performance evaluation. For example, joints may experience localized plastic deformation under certain ultimate loads, changes that might be ignored in traditional analysis methods. However, through the nonlinear feature extraction of the autoencoder, these changes can be effectively captured and used for performance evaluation.

[0045] Step S23: Based on the nonlinear characteristic processing of the mechanical characteristic factor set by the gradient boosting machine, the optimized mechanical characteristic set is obtained;

[0046] Understandably, the gradient booster first builds a simple initial model (usually a shallow decision tree) to predict mechanical characteristic factors. This model's predictions produce residuals (i.e., the difference between the predicted and actual values). Next, the gradient booster trains a new decision tree by calculating the residuals of the current predictions. The new decision tree specifically learns from the residuals, aiming to correct errors in the previous prediction through a new round of learning. This process is repeated multiple times, training a new decision tree in each round to reduce the residuals from the previous step. The weights of each tree are updated using gradient descent, and the final model is a weighted sum of the predictions from all decision trees. Through the gradient booster, we can further process and optimize these nonlinear characteristics. We can identify which characteristics (such as specific strain modes or vibration frequencies) are most critical to the structural performance of nodes and progressively adjust the model based on the residuals, making the predictions of mechanical characteristics more accurate.

[0047] Step S24: Aggregate all decision trees in the gradient booster machine based on the ensemble learning method to obtain the weighted average of multiple decision trees and obtain the optimized nonlinear feature set.

[0048] Understandably, in a gradient booster, each decision tree is trained based on the residuals of the previous prediction, thus learning different error patterns. All decision trees together form an ensemble model, with each contributing to the final prediction. Ensemble learning combines the outputs of multiple decision trees through weighted averaging to generate the final prediction. The weight of each tree is typically determined based on its performance on the training set, with better-performing trees receiving higher weights. By integrating the predictions of multiple decision trees, ensemble learning effectively reduces the risk of overfitting from a single decision tree and enhances the model's generalization ability on new data. The biases learned by each decision tree are corrected by the learning results of other trees, ultimately resulting in a more stable and reliable prediction. In a specific example, strain, vibration, and load data of prefabricated bridge and culvert nodes are used as input features, and a gradient booster generates multiple decision trees. At some nodes, due to excessive load, there may be a complex relationship between vibration patterns and strain characteristics. Through ensemble learning, we weight and integrate the predictions of all decision trees to obtain the final optimized feature set. The advantage of this process is that ensemble learning can extract the most valuable features by weighting and averaging multiple decision trees, thereby helping to provide more accurate input data for subsequent mechanical performance evaluation and prediction models.

[0049] Step S3: Perform mechanical performance evaluation on the nonlinear feature set to obtain real-time mechanical performance evaluation results;

[0050] It is understandable that this step utilizes a nonlinear feature set for mechanical performance evaluation, providing accurate and real-time predictions of mechanical performance. Compared to traditional evaluation methods, this machine learning and deep learning-based approach can better capture the mechanical response of prefabricated bridge and culvert nodes under complex working conditions, improving the accuracy and real-time performance of the evaluation. Furthermore, the dynamically updated model can adapt to environmental changes in real time, ensuring that the evaluation results remain up-to-date and providing a reliable basis for structural health monitoring and fault prediction. In this step, step S3 includes steps S31, S32, and S33.

[0051] Step S31: Based on the kernel principal component analysis method, the nonlinear feature set is mapped to a preset high-dimensional space to obtain the nonlinear feature set in the high-dimensional space;

[0052] Understandably, this step first inputs the nonlinear feature set of the prefabricated bridge and culvert nodes into the kernel principal component analysis model. This feature set typically contains nonlinear features extracted from node monitoring data (strain, vibration, and load, etc.). These features reflect the mechanical behavior of the nodes under complex working conditions. Through the radial basis function kernel, the nonlinear features are mapped to a high-dimensional space. At this point, the nonlinear structure of the data in the original space will be transformed into a linear structure in the high-dimensional space. Then, by calculating the covariance matrix of the mapped data, eigenvalues ​​and eigenvectors are obtained. Eigenvalues ​​represent the variance of the data in different directions, while eigenvectors represent the principal components of the data. Finally, by selecting eigenvectors with larger eigenvalues, most of the variance and key information are retained, thus obtaining the mapped nonlinear feature set. The data directions corresponding to the selected eigenvectors are the principal feature directions in the high-dimensional space.

[0053] By employing kernel principal component analysis to map the nonlinear feature set to a high-dimensional space, we can effectively process the complex nonlinear data of prefabricated bridge and culvert nodes, thereby extracting more meaningful high-dimensional feature sets. This process helps reveal the potential mechanical characteristics of the nodes, providing more accurate evaluation results, and thus offering strong data support for fault prediction and health monitoring.

[0054] Step S32: Establish a regression model based on the support vector machine regression algorithm and the nonlinear feature set in the high-dimensional space, and calculate the error between the predicted value of the support vector regression model and the actual observed value to obtain the feature prediction value after minimizing the error.

[0055] It is understandable that this step establishes a regression model based on the support vector machine (SVM) regression algorithm and a nonlinear feature set in high-dimensional space, and calculates the error between the predicted value and the actual observed value of the SVM regression model to obtain the feature prediction value after minimizing the error. The purpose of this step is to use the SVM regression algorithm to perform regression analysis on nonlinear features to obtain more accurate mechanical performance prediction results. Using these nonlinear features mapped to high-dimensional space as input to the SVM model, the goal is to establish a regression model to predict the mechanical performance of prefabricated bridge and culvert nodes. A radial basis function (RBF) kernel is used as the kernel function, allowing the feature set in high-dimensional space to be better mapped into the regression model, thereby capturing the complex nonlinear relationships in node behavior. The model then calculates the optimal regression hyperplane (i.e., regression function) based on the training data, where the prediction error is the error between the predicted value and the actual observed value, and the regression error is minimized by adjusting the kernel function parameters. In the practical application of this invention, strain, vibration, and load data of the nodes over a past period are collected by sensors. These data, after feature extraction and optimization in the aforementioned steps, yield a nonlinear feature set. Next, the SVM model is used to perform regression analysis on these features, with the goal of predicting the future mechanical performance of the nodes. During training, by comparing historical data with actual observations, the support vector machine model can learn the response patterns of nodes under different load conditions. Suppose we want to predict the stress response of a node at a future point in time, the support vector machine model will use the training data to build a model and output a predicted value, thus obtaining a predicted value that minimizes the error.

[0056] Step S33: Generate a knowledge graph from the feature prediction values ​​after minimizing the error, and evaluate the knowledge graph and the preset feature evaluation threshold to obtain real-time mechanical performance evaluation results.

[0057] Understandably, in this step, firstly, a knowledge graph is constructed based on the predicted mechanical characteristics with minimized errors obtained in the previous steps. The key to this process is to structure all nonlinear features, prediction results, and actual performance indicators of nodes (such as stress and vibration), and construct the graph based on their interrelationships. Nodes represent different parts or mechanical states of the prefabricated bridge or culvert (such as stress, deformation, and vibration frequency), while edges represent causal relationships or similarities between these parts or states (such as the possible functional relationship between stress and deformation, and the correlation between vibration and load). Then, real-time evaluation is performed using the constructed knowledge graph and preset feature evaluation thresholds. These evaluation thresholds are critical values ​​obtained based on engineering standards, historical experience, or simulation, used to measure whether a node is within its normal operating range or whether potential faults or structural problems may occur. In a specific example of this invention, nodes in the knowledge graph may represent different structural parts of the bridge or culvert (such as bases, pipe joints, etc.), while edges represent the relationships between them (such as stress transmission, vibration mode transmission, etc.). The predicted value of each node is mapped to the relevant nodes in the graph, such as the predicted value of stress and the predicted value of vibration frequency. Assuming that the safety threshold of stress is 50 MPa, if it exceeds this threshold, it is considered that there is a structural problem. Then, the evaluation result of "normal" or "abnormal" is obtained through the method in this step.

[0058] Step S4: Based on the real-time mechanical performance evaluation results and the preset historical monitoring data, the model is constructed and optimized to obtain the optimized mechanical performance change trend prediction model;

[0059] Understandably, this step establishes a model capable of accurately predicting future mechanical performance trends of prefabricated bridge and culvert joints by combining real-time evaluation results with historical data. This process not only considers current real-time monitoring data but also effectively integrates historical data, using data-driven methods to scientifically predict future mechanical performance changes. In this step, step S4 includes steps S41, S42, and S43.

[0060] Step S41: Standardize the real-time mechanical performance evaluation results and the preset historical monitoring data to obtain dimensionally consistent fused data;

[0061] Understandably, this step employs standardization methods to unify the units of measurement for both real-time and historical monitoring data. For example, real-time stress data is measured in MPa, while historical temperature data is measured in °C. By applying Z-score standardization, stress and temperature data can be standardized to unit standard deviation, allowing for comparison on the same scale. After standardization, both real-time and historical monitoring data are converted to the same units. Next, these data are aligned by timestamp and fused. The fused dataset includes both real-time assessment results and historical data, providing consistent input for subsequent modeling and prediction.

[0062] Step S42: Divide the fused data with consistent dimensions into windows according to a preset time series, and input the fused data of multiple preset time periods into a preset long short-term memory network for learning, to obtain a trained long short-term memory network model.

[0063] Understandably, this step divides the standardized and fused time-series data into predefined window sizes. Here, "window" refers to the monitoring data collected within each time period. The window size is determined based on the data sampling frequency and actual application requirements. These divided time-window data are then input into a Long Short-Term Memory (LSTM) network and transformed into the LTM network's input format (number of samples, time step, number of features). The divided time-series data is then fed into the predefined LTM network for training. The LTM network includes input gates, forget gates, and output gates to determine which information needs to be retained and which needs to be discarded, thereby effectively capturing long-term dependencies in the time series. The LTM network includes LTM network unit layers and can be combined with other layers such as fully connected layers and Dropout layers to capture more complex time-series features. During training, the network uses a backpropagation algorithm to adjust the network parameters, enabling the model to minimize prediction errors. This invention uses mean squared error as the loss function.

[0064] Step S43: Optimize the hyperparameters based on the simulated annealing algorithm, and substitute the optimized hyperparameters into the trained long short-term memory network model to obtain the optimized long short-term memory network model.

[0065] Understandably, this step involves selecting the learning rate, number of hidden layer units, batch size, and time step as hyperparameters, and then randomly selecting a set of hyperparameters as the initial solution, for example, a learning rate of 0.01, 50 hidden layer units, a batch size of 32, and a time step of 30. A search is then performed within the neighborhood of the current solution to generate new combinations of hyperparameters. For example, neighborhood solutions can be generated by increasing or decreasing the learning rate or adjusting the number of hidden layer units. Next, for each new hyperparameter combination, an LSTM model is trained, and its prediction error on the validation set is calculated. If the error of the new solution is less than a preset threshold, the solution is accepted; if the error is greater than the preset threshold, the solution is accepted with a preset probability. As the simulated annealing process progresses, the temperature is gradually reduced. Lowering the temperature means that the algorithm's probability of accepting inferior solutions decreases, eventually tending towards the global optimum, until the temperature reaches a set minimum value, or no better solution is found consecutively, at which point the algorithm terminates. By leveraging the global optimization characteristics of the simulated annealing algorithm, it is possible to find the optimal or near-optimal combination of hyperparameters within a large search space, thus avoiding the computational burden of traditional grid search or random search methods.

[0066] Step S5: Send the real-time monitoring data to the optimized mechanical performance change trend prediction model for prediction to obtain the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period.

[0067] It is understandable that in this step, by utilizing the optimized mechanical performance change trend prediction model and combining it with real-time monitoring data, the future mechanical performance of the node can be accurately predicted, especially the trend of stress change at the node. Step S5 is followed by steps S51 and S52.

[0068] Step S51: Construct an adversarial network based on the optimized mechanical performance change trend prediction model and real-time monitoring data, generate final simulation data with an error of less than a threshold between the adversarial network and the real-time monitoring data, and perform anomaly identification on the final simulation data based on the isolated forest algorithm to obtain the real-time monitoring data corresponding to the abnormal nodes.

[0069] Understandably, this step generates simulated data similar to actual data based on the optimized mechanical performance change trend prediction model and real-time monitoring data. The adversarial network includes a generator and a discriminator. The generator aims to produce simulated data as similar to real data as possible, ensuring that the error between the simulated data and the actual monitoring data is less than a preset threshold. The generator can reduce the error by continuously optimizing its parameters. Then, a discriminator is built based on the generator's data. The discriminator aims to improve the generator's generation ability by distinguishing between real data and generated data. Through repeated training, the generator and discriminator compete against each other. Ultimately, the generator can generate data with the smallest error to the real data. In each training round, the generator attempts to generate simulated data close to the actual data, while the discriminator judges the authenticity of the generated data and feeds back error information to the generator to optimize its parameters.

[0070] It's understandable that the Isolation Forest algorithm is an efficient anomaly detection algorithm. It constructs multiple trees and uses the degree of "isolation" between data points and the trees to determine whether they are anomalies. First, this step inputs the generated final simulation data into the Isolation Forest model for processing;

[0071] Isolation Forest isolates data by randomly selecting features and constructing multiple trees. For each data point, it calculates its degree of isolation; data points with higher isolation are more likely to be outliers. For each data point, the Isolation Forest algorithm generates an anomaly score. If the score exceeds a certain threshold, the data point is considered anomalous. In a specific example of this invention, vibration data showed a sudden change during a certain period. The generator produced simulated data based on historical data, but the Isolation Forest algorithm identified that the vibration data for this period differed significantly from the pattern of most data, resulting in a high anomaly score. Therefore, Isolation Forest marked this data as an anomaly, thus providing engineers with potential warnings and preventing undetected potential faults.

[0072] Step S52: Based on the variational autoencoder, perform damage analysis on the real-time monitoring data corresponding to the abnormal node to obtain the damage data in the real-time monitoring data corresponding to the abnormal node.

[0073] Understandably, in this step, damage typically manifests as abnormal changes in the structure's strain, vibration, or load data. For example, a sharp increase in strain values ​​may indicate cracks in a node or material fatigue; abnormal fluctuations in vibration amplitude may be related to loosening or deformation of a node. Therefore, damage data includes sudden changes in strain data, abnormal fluctuations in vibration data, and abrupt changes in load.

[0074] In this step, the variational autoencoder maps the input real-time monitoring data (including strain, vibration, load, etc.) to a latent space. At this point, the information in the monitoring data is compressed into low-dimensional latent variables. The encoder of the variational autoencoder finds these latent variables, enabling them to effectively represent the input data. Each point in the latent space corresponds to a specific representation of the input data. Then, the decoder extracts the latent variables from the latent space and maps them back to the original data space, thereby reconstructing an output that is as similar as possible to the original data. Furthermore, the variational autoencoder is trained by minimizing the reconstruction error and the KL divergence of the latent space to obtain damage data. In a specific example of this invention, it is assumed that the strain data at a certain moment fluctuates little under normal conditions, but after being input into the variational autoencoder, the reconstruction error increases significantly. This may indicate that damage has occurred in the strain data at that time. For example, microcracks or material fatigue may have occurred near the strain sensor, causing a large change in the strain value at that node. The variational autoencoder can accurately identify these anomalies and mark them as damage data by the increase in reconstruction error. The steps S52 are followed by S53, S54, S55 and S56.

[0075] Step S53: The damage data of the bridge and culvert nodes is fuzzed to obtain damage data of multiple damage levels.

[0076] The first step, as is understandable, is to define the membership function for each damage level. For example, the membership function for minor damage might resemble a triangle, indicating that minor damage data has a high membership degree within a specific range, while the membership degree decreases rapidly outside that range. Then, based on the actual damage values ​​(such as strain, vibration, and load) in the monitoring data, these data are mapped to multiple damage levels using the membership function. Finally, by inputting the damage data (such as strain, vibration, and load) into the fuzzification model, the membership degree corresponding to each data point is calculated one by one according to the preset membership function. For each monitoring data point, its membership degree under each damage level is calculated, and fuzzy damage levels are generated based on these membership degree values.

[0077] In this step, three membership functions are constructed: a membership function for mild damage, a membership function for moderate damage, and a membership function for severe damage. The membership function for mild damage is shown below:

[0078] ;

[0079] in, This represents the membership function value for mild damage. The damage value of the node. and All are the boundaries of areas with mild damage.

[0080] The membership function for moderate injury is shown below:

[0081] ;

[0082] in, This represents the membership function value for moderate damage. The damage value of the node. The boundary of the area with mild damage. and All are the boundaries of the intervals with moderate damage.

[0083] ;

[0084] in, This represents the membership function value for severe injury. The damage value of the node. and All are the boundaries of the intervals with moderate damage.

[0085] Step S54: Convert the historical reinforcement scheme into reinforcement schemes with multiple damage levels to obtain reinforcement schemes with multiple damage levels.

[0086] It is understandable that in practical applications, the relationship between the damage level of a structure and the reinforcement scheme is non-linear. Generally, the more severe the damage, the more intensive the reinforcement measures required. Damage levels are typically classified based on monitoring data such as strain, vibration, and load of the structure. After fuzzification, different damage levels are obtained. Based on experience with historical reinforcement schemes, a mapping relationship can be established between these damage levels and corresponding reinforcement schemes. For example: Minor damage: Only local reinforcement is needed, using simple repair materials such as carbon fiber patches and adhesives. Moderate damage: Reinforcement of nodes or connections is required, using more complex reinforcement schemes such as steel plates, reinforcing ribs, or concrete pouring. Severe damage: Comprehensive large-scale reinforcement is required, using measures such as prestressed steel cables and external concrete encasing.

[0087] Step S55: Select the reinforcement scheme corresponding to the damage data of each bridge and culvert node based on the multi-objective optimization function to obtain the preliminary reinforcement strategy for all damaged bridge and culvert nodes.

[0088] Understandably, this step inputs the damage data of each node obtained in the previous steps into the multi-objective optimization algorithm. The damage data for each node may include monitoring data such as strain, vibration, and load, as well as damage level and damage characteristic data generated based on damage analysis. Based on this data, the algorithm will evaluate the degree of damage to each node and select the corresponding reinforcement scheme according to the preset damage level.

[0089] The multi-objective optimization function is shown below:

[0090] ;

[0091] in, These represent different objective functions used to measure optimization goals in various aspects, such as cost, efficiency, and resource utilization. This represents a vector of optimization decision variables, which represents different values ​​of reinforcement schemes or design variables. The constraint function represents certain constraints, such as resource limitations and structural constraints, that affect decision variables. Perform the calculations and ensure they are within the given corresponding upper limits. Within the range, This indicates the upper limit of a constraint, representing the maximum allowable value for a specific constraint, such as the maximum limit of resources and the maximum load capacity of a structure. Indicates the start time of the task or hardening process. Indicates based on decision variables The reinforcement scheme is calculated to determine the completion time of the reinforcement process. This indicates the maximum allowed completion time, representing the upper limit of the time required to complete the reinforcement task. Based on decision variables The total damage level of the reinforcement scheme was calculated, with minor damage level 1, moderate damage level 2, and severe damage level 3. This indicates the maximum permissible level of damage, used to limit the damage control limitations of reinforcement schemes. Indicates the decision variables The amount of additional resources required in the resulting reinforcement solution. Indicates the maximum amount of additional resources that can be accepted. This indicates the cost of the reinforcement solution. Indicates the maximum allowed cost.

[0092] Step S56: Perform resource maximization analysis on the preliminary reinforcement strategies for all damaged bridge and culvert nodes, and reselect all the preliminary reinforcement strategies based on the analysis results to obtain the optimal reinforcement strategy.

[0093] Understandably, this step defines the objective function as the ratio between the cost and expected benefits (e.g., the increase in node load-bearing capacity) of the reinforcement scheme. Resource constraints are the core constraints of this model, mainly including budget constraints, construction resource constraints (such as labor and equipment), and time constraints (such as construction period and project delivery date). The decision variable can be the selection of a reinforcement scheme for each node, represented as a 0-1 decision variable, i.e., whether a certain node adopts a certain reinforcement method. Then, based on the results of the resource maximization analysis, under the condition of satisfying resource constraints, the system will re-evaluate and select all preliminary reinforcement strategies. The reinforcement schemes for each node will be re-ranked according to their required resource consumption and contribution to the overall reinforcement effect, and the optimal reinforcement strategy will be selected. The final generated reinforcement strategy will be the reinforcement scheme with optimal resource utilization, ensuring that the project achieves the best reinforcement effect under resource constraints.

[0094] Example 2:

[0095] like Figure 2 As shown, this embodiment provides a neural network-based system for analyzing the mechanical performance of prefabricated bridge and culvert joints. (See also...) Figure 2 The system includes an acquisition unit 701, a processing unit 702, an evaluation unit 703, a construction unit 704, and a prediction unit 705.

[0096] The acquisition unit 701 is used to acquire real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data and load data.

[0097] The processing unit 702 is used to extract and optimize the features of the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set;

[0098] The evaluation unit 703 is used to evaluate the mechanical properties of the nonlinear feature set and obtain real-time mechanical property evaluation results.

[0099] The construction unit 704 is used to construct and optimize the model based on the real-time mechanical performance evaluation results and preset historical monitoring data to obtain an optimized mechanical performance change trend prediction model.

[0100] The prediction unit 705 is used to send the real-time monitoring data to the optimized mechanical performance change trend prediction model for prediction, so as to obtain the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period.

[0101] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0103] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for analyzing the mechanical performance of prefabricated bridge and culvert joints based on neural networks, characterized in that, include: Acquire real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data, and load data; Feature extraction and optimization are performed on the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set; The nonlinear feature set is subjected to mechanical performance evaluation to obtain real-time mechanical performance evaluation results; Based on the real-time mechanical performance evaluation results and the preset historical monitoring data, the model is constructed and optimized to obtain the optimized mechanical performance change trend prediction model. The real-time monitoring data is sent to the optimized mechanical performance change trend prediction model for prediction, and the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period is obtained. Among them, feature extraction and optimization are performed on the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set, including: Factor analysis was performed on the real-time monitoring data of prefabricated bridge and culvert nodes to obtain a set of mechanical characteristic factors for prefabricated bridge and culvert nodes. Unsupervised learning of the set of mechanical feature factors is performed using an autoencoder of a deep belief network to obtain the nonlinear features of the set of mechanical feature factors. Based on the nonlinear characteristic processing of the mechanical characteristic factor set by the gradient booster, the optimized mechanical characteristic set is obtained; By aggregating all decision trees within the gradient booster machine using ensemble learning, a weighted average of the multiple decision trees is obtained, resulting in an optimized nonlinear feature set.

2. The method for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks according to claim 1, characterized in that... The mechanical performance evaluation of the nonlinear feature set includes: Based on the kernel principal component analysis method, the nonlinear feature set is mapped to a preset high-dimensional space to obtain the nonlinear feature set in the high-dimensional space. A regression model is established based on the support vector machine regression algorithm and a nonlinear feature set in a high-dimensional space. The error between the predicted value of the support vector regression model and the actual observed value is calculated, and the feature prediction value after minimizing the error is obtained. The predicted feature values ​​after minimizing the error are used to generate a knowledge graph, and the evaluation is performed based on the knowledge graph and a preset feature evaluation threshold to obtain real-time mechanical performance evaluation results.

3. The method for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks according to claim 1, characterized in that... Based on the real-time mechanical performance evaluation results and preset historical monitoring data, model construction and optimization are performed, including: The real-time mechanical performance evaluation results and the preset historical monitoring data are standardized to obtain dimensionally consistent fused data. The fused data with consistent dimensions is divided into windows according to a preset time series, and the fused data of multiple preset time periods are input into a preset long short-term memory network for learning, so as to obtain a trained long short-term memory network model. By selecting the learning rate, the number of hidden layer units, the batch size, and the number of time steps as hyperparameters, the hyperparameters are optimized based on the simulated annealing algorithm. The optimized hyperparameters are then substituted into the trained long short-term memory network model to obtain the optimized long short-term memory network model.

4. The method for analyzing the mechanical performance of prefabricated bridge and culvert nodes based on neural networks according to claim 1, characterized in that... After obtaining the mechanical properties of prefabricated bridge and culvert joints over a preset time period, the process also includes: An adversarial network is constructed based on the optimized mechanical performance change trend prediction model and real-time monitoring data. The final simulation data with an error of less than a threshold between the adversarial network and the real-time monitoring data is generated. Anomalies are identified in the final simulation data based on the isolated forest algorithm to obtain the real-time monitoring data corresponding to the abnormal nodes. Based on the variational autoencoder, the damage data in the real-time monitoring data corresponding to the abnormal node is obtained through damage analysis.

5. A neural network-based system for analyzing the mechanical performance of prefabricated bridge and culvert joints, characterized in that, include: The acquisition unit is used to acquire real-time monitoring data of prefabricated bridge and culvert nodes, including strain data, vibration data and load data. The processing unit is used to extract and optimize the features of the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a nonlinear feature set; An evaluation unit is used to evaluate the mechanical properties of the nonlinear feature set and obtain real-time mechanical property evaluation results. The construction unit is used to build and optimize the model based on the real-time mechanical performance evaluation results and preset historical monitoring data to obtain an optimized mechanical performance change trend prediction model. The prediction unit is used to send the real-time monitoring data to the optimized mechanical performance change trend prediction model for prediction, and obtain the mechanical performance of the prefabricated bridge and culvert nodes for a preset time period. The processing unit includes: The first processing subunit is used to perform factor analysis based on the real-time monitoring data of the prefabricated bridge and culvert nodes to obtain a set of mechanical characteristic factors for the prefabricated bridge and culvert nodes. The second processing subunit is used to perform unsupervised learning on the set of mechanical feature factors through the autoencoder of the deep belief network to obtain the nonlinear features of the set of mechanical feature factors. The third processing subunit is used to obtain the optimized set of mechanical features based on the nonlinear feature processing of the set of mechanical feature factors by the gradient boosting machine. The fourth processing subunit is used to aggregate all decision trees in the gradient booster machine based on the ensemble learning method, obtain the weighted average of multiple decision trees, and obtain the optimized nonlinear feature set.

6. The neural network-based mechanical performance analysis system for prefabricated bridge and culvert joints according to claim 5, characterized in that, The evaluation unit includes: The first evaluation subunit is used to map the nonlinear feature set to a preset high-dimensional space based on the kernel principal component analysis method, so as to obtain the nonlinear feature set in the high-dimensional space. The second evaluation subunit is used to establish a regression model based on the support vector machine regression algorithm and the nonlinear feature set in the high-dimensional space, and to calculate the error between the predicted value of the support vector regression model and the actual observed value, so as to obtain the feature prediction value after minimizing the error. The third evaluation subunit is used to generate a knowledge graph from the feature prediction values ​​after minimizing the error, and to evaluate the knowledge graph and the preset feature evaluation threshold to obtain real-time mechanical performance evaluation results.

7. The neural network-based mechanical performance analysis system for prefabricated bridge and culvert joints according to claim 5, characterized in that, The building unit includes: The first construction subunit is used to standardize the real-time mechanical performance evaluation results and the preset historical monitoring data to obtain dimensionally consistent fused data. The second construction subunit is used to divide the fused data with consistent dimensions into windows according to a preset time series, and input the fused data of multiple preset time periods into a preset long short-term memory network for learning, so as to obtain a trained long short-term memory network model. The third sub-unit is used to optimize the hyperparameters by selecting the learning rate, the number of hidden layer units, the batch size, and the number of time steps, based on the simulated annealing algorithm. The optimized hyperparameters are then substituted into the trained long short-term memory network model to obtain the optimized long short-term memory network model.

8. The neural network-based mechanical performance analysis system for prefabricated bridge and culvert joints according to claim 5, characterized in that, Following the prediction unit, the system further includes: The first judgment subunit is used to construct an adversarial network based on the optimized mechanical performance change trend prediction model and real-time monitoring data, generate final simulation data with an error of less than a threshold based on the adversarial network, and perform anomaly identification on the final simulation data based on the isolated forest algorithm to obtain the real-time monitoring data corresponding to the abnormal nodes. The second judgment subunit is used to perform damage analysis on the real-time monitoring data corresponding to the abnormal node based on the variational autoencoder, and to obtain the damage data in the real-time monitoring data corresponding to the abnormal node.

Images