Bridge structure intelligent damage identification method based on physical information neural network

Abstract

The invention belongs to the technical field of bridge damage identification, and relates to a bridge structure intelligent damage identification method based on a physical information neural network, which comprises the following steps: 1, acquiring spatial relation data of a bridge structure, 2, acquiring real-time data, 3, preprocessing the data after the data is acquired, 4, performing hyper-parameter adjustment, 5, constructing a model, 6, carrying out model training and evaluation; 7, carrying out early warning and repairing scheme matching; the health state of the bridge can be monitored in real time, data support can be provided for future maintenance and repair, the safety of the bridge is improved, and the service life of the bridge is prolonged; therefore, a novel efficient tool is provided for bridge health monitoring and damage assessment through a unique data and physical driving method, an efficient and intelligent bridge monitoring and assessment system is constructed, and the safety and maintenance efficiency of the bridge are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of bridge damage identification and relates to an intelligent damage identification method for bridge structure based on physical information neural network. Background Art

[0002] With the acceleration of urbanization, bridges play an increasingly important role in urban transportation networks. However, as the service life of bridges increases, fatigue and aging of structural materials will lead to a decrease in the safety and durability of bridges, and eventually damage their structures. Therefore, it is necessary to monitor the health of bridge structures. Health monitoring mainly includes three parts: sensor optimization layout, structural moving load identification, and structural damage identification. Structural damage identification is the core of health monitoring. In the field of bridge damage identification, traditional research methods mainly rely on static and dynamic tests and intelligent algorithms.

[0003] However, in the prior art, although these methods have achieved certain results, they still have some obvious defects and deficiencies. For example, methods based on static testing can usually only provide limited measurement information such as displacement and strain, which may make it difficult to obtain ideal recognition results; and methods based on dynamic testing, although they have more information, are greatly affected by factors such as the test environment and the location of the measuring point, and are also greatly affected by noise. Although intelligent algorithms have high noise resistance, the recognition accuracy of neural networks is greatly affected by training samples and requires a large number of training samples.

[0004] At the same time, although the static response-based structural damage identification method is relatively mature and the static data obtained is intuitive, this method can only measure a small number of damage identification parameters, and it is difficult to process complex data for large bridge structures; the use of natural frequency for damage identification has a strong anti-interference ability against environmental noise, but the frequency change will be affected by many factors, and it is not sensitive enough to the slightest damage of the structure; although the use of modal vibration morphology for structural damage identification solves the problem of small damage identification, it usually requires the deployment of a large number of sensors, which will be limited by equipment cost and structure in practical applications; the use of model correction methods for structural damage detection can achieve highly accurate damage identification, but it depends on the integrity of the model, and large-scale projects will require a lot of time and effort.

[0005] The shortcomings of existing bridge damage identification technology are also reflected in the following aspects:

[0006] 1. Data dependence: Traditional methods such as finite element model modification require a large amount of accurate data, which is difficult to achieve in actual bridge maintenance. The lack of real scene data limits model training and affects the accuracy of damage identification.

[0007] 2. Accuracy and efficiency: In complex environments, existing technologies have difficulty handling highly nonlinear damage characteristics, resulting in uncertainty in identification results. In addition, real-time or near-real-time feedback requires improvements in computing efficiency, which is particularly important in monitoring critical infrastructure.

[0008] 3. Generalization ability: The model’s performance outside the training data distribution is often unsatisfactory. This limitation makes it difficult to maintain consistent recognition results when the model is applied under different bridge or environmental conditions.

[0009] 4. Physical interpretability: Many data-driven methods lack physical background, making it difficult for engineers to understand the prediction mechanism when applying the model. This lack of interpretability reduces the credibility of the model and hinders its promotion in practical engineering.

[0010] 5. Computing resources: Traditional methods often require a lot of computing resources and time to solve complex problems, especially in large-scale simulations. This not only increases the economic burden, but also limits the possibility of real-time monitoring.

[0011] Therefore, a bridge structure damage identification method is needed to solve the above technical problems by effectively identifying damage even when resources are limited and improving the accuracy and efficiency of damage identification. Summary of the invention

[0012] The intelligent bridge damage identification method based on physical information neural network (PINN) proposed in the present invention provides an effective solution to the deficiencies of the prior art. By adding physical laws to the loss function, PINN can ensure that the model output not only conforms to the data characteristics but also conforms to the physical laws, thereby improving the accuracy of damage identification. The method of the present invention is particularly suitable for solving partial differential equations (PDE) and other physical problems, especially when data is scarce or the problem is complex; and PINN reduces the computing cost through more effective algorithms, making the solution process more efficient, especially when dealing with complex bridge structure problems, by intelligently utilizing physical knowledge, reducing the demand for computing resources, so that effective damage identification can be performed even when resources are limited.

[0013] The present invention aims to achieve smarter and more efficient damage identification and promote the development of bridge monitoring technology. Through these systematic improvements, the shortcomings of the existing technology can be effectively overcome, providing a more accurate, applicable and scarce data solution for intelligent damage identification of bridge structures.

[0014] The technical solution adopted by the present invention to solve the technical problem is: a bridge structure intelligent damage identification method based on physical information neural network, the identification method comprises the following steps:

[0015] Step 1, obtaining the spatial relationship data of the bridge structure;

[0016] Step 2: Real-time data collection;

[0017] Step 3: After data acquisition, data preprocessing;

[0018] Step 4: Adjust hyperparameters;

[0019] Step 5: Model construction: During the model construction process, the physical equations are determined;

[0020] Step 6: Model training and evaluation;

[0021] Step 7: Match the warning with the repair plan.

[0022] Preferably, in step 1, the geometric features of the bridge are captured by three-dimensional laser scanning or photogrammetry, and the geometric features include: size, shape and relative position.

[0023] Preferably, in step 2, the equipment used for real-time data collection includes: displacement meters, strain gauges and temperature sensors; the data collected in real-time data include: deflection, strain and temperature status data of the bridge to be monitored.

[0024] Preferably, in step 2, first, when selecting displacement meters, strain gauges and temperature sensors, their accuracy, stability and durability should be considered to adapt to the specific environment and load conditions of the bridge; the arrangement of sensors should be based on structural analysis to ensure coverage of high stress areas and potential failure points to obtain comprehensive status information; secondly, the data collection frequency should be flexibly adjusted according to the use of the bridge, traffic flow and environmental changes to capture instantaneous changes in a timely manner.

[0025] Preferably, in step 3, firstly, outliers need to be eliminated and missing values ​​need to be supplemented; after data preprocessing, the data are normalized.

[0026] Preferably, in step 3, when outliers are removed and missing values ​​are supplemented, the missing values ​​in the monitoring data are intelligently filled using the data processing library in Python and the built-in functions, wherein the intelligent filling includes: mean filling, median filling or interpolation; at the same time, statistical methods are used to detect and remove outliers;

[0027] When normalizing the data, Z-score standardization was used.

[0028] Preferably, the physical equation in step 5 includes:

[0029]

[0030] Where w(x) is the deflection, M(x′) is the bending moment, E is the elastic modulus of the material, I is the moment of inertia of the section, L is the length of the beam, and C1 and C2 are constants.

[0031] Preferably, in step 5, the physical equations are embedded into the model, the equations are combined with the structure of the neural network, and physical knowledge is introduced during the training process.

[0032] Preferably, in step 6, model training and evaluation specifically include: inputting the processed data set into the trained bridge structure condition monitoring and evaluation model, optimizing the model parameters through multiple iterative learning; after completing the training, using the test data set to evaluate the prediction ability of the model, and generating evaluation and analysis results of the bridge structure condition.

[0033] Preferably, in step 7, the matching of the warning and the repair plan specifically includes: setting a corresponding warning mechanism based on the status assessment analysis results output by the model; when a potential damage risk is detected, the system automatically triggers a warning to prompt relevant personnel; at the same time, matching a preset repair plan according to the assessment results.

[0034] The beneficial effects of the present invention are:

[0035] 1. The present invention can improve the accuracy of bridge structure damage identification; traditional damage identification methods often rely on experience or simple statistical analysis, which may lead to errors in identification results. The present invention combines the advantages of physical models and data-driven, and can more accurately capture tiny damage signals by deeply learning the structural response characteristics. The method of the present invention shows stronger robustness when dealing with complex structures and multiple damage modes.

[0036] 2. The present invention can monitor bridge structural damage in real time; the present invention can realize real-time data collection and analysis, and continuously monitor the bridge status using a sensor network. Real-time monitoring can help engineers identify potential structural problems in a timely manner, take maintenance measures in advance, significantly reduce accident risks, and improve public safety.

[0037] 3. The present invention can reduce the maintenance cost for bridge inspection; traditional bridge inspection methods usually require regular manual inspections, which are time-consuming and costly. Through intelligent damage identification technology, the frequency of manual inspections can be reduced, maintenance costs can be reduced, and inspection efficiency can be improved. This intelligent management method not only saves human resources, but also optimizes the maintenance decision-making process.

[0038] 4. The present invention can improve the resource utilization rate of bridge maintenance and management; the introduction of the method of the present invention makes the maintenance and management of bridges more intelligent, and the allocation and use of resources more efficient. According to the actual monitoring data, targeted maintenance can be carried out to avoid unnecessary maintenance and waste of resources.

[0039] 5. The present invention promotes intelligent management of bridge structure health monitoring; with the continuous advancement of technology, the bridge structure health monitoring system is gradually developing towards intelligence and automation. The method of the present invention provides important support for this transformation, and can realize real-time analysis and feedback of data, thereby promoting the innovation of bridge management mode.

[0040] 6. The present invention can establish a long-term monitoring and early warning mechanism; the method of the present invention can establish a long-term monitoring system, form an accurate prediction model through the combination of historical data and models, warn of potential damage, and enhance the risk resistance of the bridge.

[0041] 7. The present invention can provide data-driven decision support for the maintenance and management strategies of bridge structures; the present invention utilizes the large amount of collected data, combined with machine learning and artificial intelligence technology, to provide a more scientific basis for decision-making and help engineers develop more effective maintenance and management strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the model construction process of a bridge structure intelligent damage identification method based on physical information neural network of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the relevant technologies in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] The intelligent bridge damage identification technology based on physical information neural network (PINN) in this implementation is an innovative method that combines the advantages of deep learning and physical modeling to detect and locate damage in bridge structures. The emergence of this technology is based on a deep understanding of the damage problems caused by environmental erosion, traffic loads, material aging and other factors during the long-term service of bridges, as well as a clear understanding of the limitations of traditional damage identification methods.

[0045] The core of PINN technology is to embed the physical model of the bridge directly into the training process of the neural network. This not only enables the network to learn from limited observation data, but also ensures that the learning process strictly abides by the laws of physics, shows better robustness in the face of noise and uncertainty, and can be better generalized to unknown data, improving the accuracy and generalization ability of the model. By adding physical information terms to the loss function, PINN ensures that the model output not only conforms to the data characteristics, but also follows the laws of physics, thereby improving the interpretability and credibility of the model. The data error term and the physical information error term are used to jointly optimize the network parameters, which enables effective training and prediction even when data is scarce.

[0046] In practical applications, the physical information neural network (PINN) model shows unique advantages in bridge health monitoring. By analyzing the response data of the bridge under different working conditions, such as vibration frequency, vibration mode and strain, PINN can effectively identify structural damage. This invention mainly focuses on monitoring the deflection change of the bridge, which is an important indicator for evaluating the health of the bridge. The deflection change is affected by many factors, including temperature changes, dynamic acceleration and local strain. Therefore, the PINN model takes these factors into account during the training process to improve the sensitivity and accuracy of damage. In terms of data processing, PINN can extract abnormal patterns from a large amount of sensor data through deep learning technology. These patterns are often closely related to potential damage inside the bridge. For example, a specific vibration mode may change when the structure is damaged, and PINN can capture these subtle changes, thereby providing early warning. In addition, another major advantage of PINN is its predictive ability. By learning from historical data, PINN can not only identify the current damage state, but also predict the development trend of future damage. This provides a scientific basis for the maintenance and repair of bridges, making maintenance decisions more timely and effective.

[0047] Example

[0048] 1. Collect data

[0049] A large cross-sea bridge was selected as the monitoring object. The collected data include: displacement data: high-precision displacement meters are installed at key locations of the bridge to collect displacement change data in different time periods. For example, at the top of the main piers and the mid-span of the main beam, the collected displacement data show certain fluctuations in different seasons and traffic flows. Strain data: strain meters are installed at key stress-bearing locations of the bridge structure, such as the bottom and sides of the beam. The collected strain data reflects the stress conditions of the bridge under vehicle loads and environmental factors. Temperature data: temperature sensors are installed at different locations on the bridge, and the collected temperature data shows obvious day and night temperature differences and seasonal changes.

[0050] 2. Data processing

[0051] Outlier processing: Statistical methods are used to detect outliers in the collected data. For example, for displacement data, if the displacement value at a certain moment deviates too much from that at adjacent moments, it is determined to be an outlier and removed.

[0052] Missing value supplementation: For missing values ​​caused by sensor failure and other reasons, interpolation is used to supplement them. For example, for the missing part of the strain data, linear interpolation is performed based on the strain data at adjacent positions and times.

[0053] Normalization processing: All processed displacement, strain and temperature data are normalized using Z-score to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1 for subsequent model processing.

[0054] 3. Model building and training process

[0055] Model establishment: According to the structural characteristics of the bridge and the data type collected, the physical equation is determined to be the Euler-Bernoulli beam theory equation, and it is embedded in the physical information neural network (PINN) model. At the same time, the structural parameters of the neural network are set, such as the number of hidden layers is 3, the number of neurons in each layer is 50, 30, and 20 respectively, and the activation function is selected as ReLU.

[0056] Hyperparameter adjustment: After multiple experiments, the learning rate was adjusted to 0.001 and the batch size was 32. The cross-validation method was used to calculate the root mean square error (RMSE) and the coefficient of determination (R 2 ) and other indicators to evaluate the performance of the model under different hyperparameter combinations and ultimately determine the optimal hyperparameter settings.

[0057] Model training: The preprocessed data set is divided into a training set and a test set in a ratio of 7:3. The training set is input into the PINN model for multiple iterations of learning. Each iteration adjusts the model parameters according to the loss function (including data error terms and physical information error terms) until the model converges, that is, the loss function value reaches the minimum.

[0058] 4. Early warning data feedback

[0059] Model evaluation: The trained model was evaluated using the test set, and the error indicators between the predicted value and the true value were calculated, such as the root mean square error (RMSE) of 0.05 and the coefficient of determination (R2) of 0.92, indicating that the model has good predictive ability.

[0060] Early warning mechanism: According to the bridge structure status evaluation and analysis results output by the model, the early warning threshold is set. For example, when the predicted deflection change exceeds the set threshold, the system automatically triggers an early warning. For this cross-sea bridge, the deflection change threshold is set to 5mm. When the model predicts that the deflection may exceed 5mm at a certain moment, the system immediately sends an early warning message to the bridge management department, prompting relevant personnel to pay attention to the bridge status, and matching the preset repair plan according to the evaluation results, such as detailed inspection and maintenance of the parts that may be damaged.

[0061] It can be seen from the above embodiments that the method of the present invention has a good application effect in bridge monitoring, can effectively identify bridge structure damage, and provide strong support for bridge maintenance and management.

[0062] refer to Figure 1 , the specific steps of this embodiment are:

[0063] 1. Obtaining spatial relationship data of bridge structures is a critical first step in analyzing bridge performance and safety. Through 3D laser scanning or photogrammetry, the geometric characteristics of the bridge, including size, shape and relative position, can be accurately captured. These technologies can not only obtain static data, but also record the deformation of the bridge under different environmental conditions. Subsequently, this data can be used to build a detailed 3D model to help engineers deeply analyze the bridge's bearing capacity, fatigue characteristics and potential damage, thereby providing a scientific basis for maintenance and reinforcement.

[0064] 2. Real-time data acquisition is crucial in bridge monitoring and involves multiple optimization aspects. First, when selecting displacement meters, strain gauges, and temperature sensors, consider their accuracy, stability, and durability to adapt to the specific environment and load conditions of the bridge. The arrangement of sensors should be based on structural analysis to ensure coverage of high-stress areas and potential failure points, such as piers and supports, to obtain comprehensive status information. In addition, the frequency of data collection should be flexibly adjusted according to bridge usage, traffic flow, and environmental changes, such as increasing the collection frequency during peak hours or in bad weather to capture instantaneous changes in a timely manner. Combined with machine learning technology, the intelligent collection mechanism can automatically identify and optimize the frequency and improve the system responsiveness. In terms of data processing, advanced analysis techniques such as time series analysis and anomaly detection are used to extract valuable information from real-time data to identify potential risks in a timely manner.

[0065] 3. After data acquisition, data preprocessing is a crucial step. First, outliers need to be eliminated and missing values ​​need to be supplemented. This embodiment uses various data processing libraries in Python, such as Pandas and NumPy, and uses built-in functions to intelligently fill in missing values ​​in monitoring data. Common methods include mean filling, median filling, or interpolation. At the same time, in order to improve data quality, this embodiment also uses statistical methods, such as the Z-score method, to detect and remove outliers. After these processes, this embodiment normalizes all data and uses Z-score standardization to ensure that the data is under the same standard, which is convenient for subsequent comparison and analysis. This series of operations can not only improve the accuracy of the data, but also ensure the stability and effectiveness of subsequent model training, thereby laying a solid foundation for the performance of the model.

[0066] 4. Performing hyperparameter adjustment is one of the key steps in building a neural network model. Before building the model, various hyperparameters must be carefully adjusted, such as learning rate, number of hidden layers, activation function selection, and batch size. Typically, this process involves manual debugging and cross-validation to ensure that the performance of the model is optimal under different parameter combinations. By systematically evaluating preset evaluation indicators such as determination coefficient, root mean square error, and mean error, this embodiment selects the optimal hyperparameter settings. This not only helps to improve the performance of the model, but also lays a solid foundation for building a stable and efficient neural network framework. Ultimately, this process can significantly improve the generalization ability of the model, making it perform better in practical applications.

[0067] 5. In the process of model building, determining the physical equations is a key step. By utilizing the characteristics of physical information neural networks, this embodiment embeds relevant physical equations into the model. These equations not only provide the basic laws of the physical system, but also set constraints for the model output, thereby ensuring that the prediction results are physically reasonable. When determining the physical equations, the classical equations of bridge deflection can be considered, such as the Euler-Bernoulli beam theory. This theory describes the deflection of the beam through the following equation:

[0068]

[0069] Where w(x) is the deflection, M(x′) is the bending moment, E is the elastic modulus of the material, I is the moment of inertia of the section, L is the length of the beam, and C1 and C2 are constants.

[0070] By combining the equation with the structure of the neural network, this embodiment introduces physical knowledge into the training process, so that even when data is scarce, the model can still generate output that conforms to physical laws. This method not only improves the reliability of the model, but also enhances its generalization ability.

[0071] 6. Model training and evaluation: The processed data set is input into the trained bridge structure condition monitoring and evaluation model, and the model parameters are optimized through multiple iterative learning. After the training is completed, the prediction ability of the model is evaluated using the test data set, and the evaluation and analysis results of the bridge structure condition are generated.

[0072] 7. Matching of early warning and repair scheme: Based on the status assessment and analysis results output by the model, set up the corresponding early warning mechanism. When potential damage risk is detected, the system automatically triggers an early warning to prompt relevant personnel. At the same time, the preset repair scheme is matched according to the assessment results to provide scientific basis and guidance for subsequent bridge maintenance to ensure the safety and stability of the bridge.

[0073] Through these steps, this embodiment constructs an efficient and intelligent bridge monitoring and evaluation system to improve the safety and maintenance efficiency of the bridge.

[0074] The key point of this embodiment is

[0075] 1. Innovation of monitoring method: Using the PINN (Physical Information Neural Network) model to identify bridge damage in combination with deflection changes, temperature, acceleration and strain data provides a new monitoring method. This method not only improves the flexibility and accuracy of monitoring, but also processes multiple data in real time, enhancing the comprehensive assessment of the health status of the bridge. In addition, the adaptive learning ability of the PINN model enables it to be optimized under different environments and conditions, reducing dependence on traditional technologies and promoting the intelligent development of bridge monitoring technology. By in-depth exploration of the application of this innovative method, more reliable support can be provided for bridge safety management.

[0076] 2. Reduce data dependence: The effective construction of the model can be achieved by integrating physical knowledge with data-driven methods. In practical applications, many fields face the challenge of insufficient data, so combining existing physical theories can provide a more solid foundation for damage identification. This method not only improves the robustness of the model, but also helps identify potential damage situations through simulation or theoretical deduction when data is scarce. In addition, the introduction of physical knowledge can guide the data collection process, optimize sensor layout and measurement strategies, and further improve recognition accuracy and efficiency.

[0077] Therefore, the advantage of this embodiment is that the damage identification method integrates the physical model and deep learning technology to provide an efficient damage identification mechanism. First, the method establishes a physical model of the bridge, considers the material properties and structural characteristics, and then combines the dynamic response data collected by the sensor in real time, and uses a neural network to perform deep learning on the damage characteristics. This not only improves the accuracy of damage identification, but also effectively identifies different types of damage patterns.

[0078] In addition, the present invention exhibits strong anti-interference capabilities, can handle complex environmental noise, and ensure the reliability of identification results. At the same time, real-time data processing and feedback mechanisms make damage identification automated and efficient, and can provide timely warnings when damage occurs, reducing safety hazards. The present invention also has good integration capabilities and can be seamlessly connected with existing structural health monitoring systems, intelligent traffic management platforms, etc. to achieve overall intelligent management. In addition, accurate damage identification can reduce unnecessary comprehensive inspections, thereby significantly reducing maintenance costs, optimizing resource utilization, and ultimately improving the safety and service life of bridges.

[0079] This implementation solves the defects of traditional methods by embedding physical laws into the training process of neural networks, and improves the accuracy and efficiency of damage identification. In addition, PINN has a wide range of applicability, not only for bridge structures, but also for damage identification of other civil engineering structures, providing an effective tool for structural health monitoring and maintenance. Through experimental verification, PINN's accuracy and noise resistance in bridge damage identification are better than those of traditional BP neural network methods, and its advantages in damage location identification are more obvious.

[0080] In summary, the application of the present invention in bridge health monitoring can not only monitor the health status of the bridge in real time, but also provide data support for future maintenance and repair, thereby improving the safety and service life of the bridge. This innovative method has opened up a new direction for the intelligent management of infrastructure. Therefore, the PINN-based bridge intelligent damage identification technology, through its unique data and physics-driven approach, provides a new and efficient tool for bridge health monitoring and damage assessment, which has broad application prospects.

[0081] It should be emphasized that the above are only preferred embodiments of the present invention and do not limit the present invention in any form. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A bridge structure intelligent damage identification method based on physical information neural network, characterized in that: The identification method comprises the following steps: Step 1, obtaining the spatial relationship data of the bridge structure; Step 2: Real-time data collection; Step 3: After data acquisition, data preprocessing; Step 4: Adjust hyperparameters; Step 5: Model construction: During the model construction process, the physical equations are determined; Step 6: Model training and evaluation; Step 7: Match the warning with the repair plan.

2. According to claim 1, a bridge structure intelligent damage identification method based on physical information neural network is characterized in that: In step 1, the geometric features of the bridge are captured by three-dimensional laser scanning or photogrammetry, and the geometric features include: size, shape and relative position.

3. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 1 is characterized in that: In step 2, the equipment used for real-time data collection includes: displacement meter, strain gauge and temperature sensor; the data collected in real-time data includes: deflection, strain and temperature status data of the bridge to be monitored.

4. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 3 is characterized in that: In step 2, first, when selecting displacement meters, strain gauges and temperature sensors, their accuracy, stability and durability should be considered to adapt to the specific environment and load conditions of the bridge; the arrangement of sensors should be based on structural analysis to ensure that high stress areas and potential failure points are covered to obtain comprehensive status information; Secondly, the frequency of data collection should be flexibly adjusted according to bridge usage, traffic flow and environmental changes to capture instantaneous changes in a timely manner.

5. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 1 is characterized in that: In step 3, firstly, outliers need to be eliminated and missing values ​​need to be supplemented; after data preprocessing, the data is normalized.

6. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 5 is characterized in that: In step 3, when outliers are removed and missing values ​​are supplemented, the data processing library in Python is used to intelligently fill the missing values ​​in the monitoring data using built-in functions, and the intelligent filling includes: mean filling, median filling or interpolation; at the same time, statistical methods are used to detect and remove outliers; When normalizing the data, Z-score standardization was used.

7. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 1 is characterized in that: The physical equations in step 5 include: Where w(x) is the deflection, M(x′) is the bending moment, E is the elastic modulus of the material, I is the moment of inertia of the cross section, L is the length of the beam, and C1 and C2 are constants.

8. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 7 is characterized in that: In step 5, the physical equations are embedded into the model, the equations are combined with the structure of the neural network, and physical knowledge is introduced into the training process.

9. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 1 is characterized in that: In step 6, model training and evaluation specifically include: The processed data set is input into the trained bridge structure condition monitoring and evaluation model, and the model parameters are optimized through multiple iterative learning. After the training is completed, the test data set is used to evaluate the prediction ability of the model and generate evaluation and analysis results of the bridge structure condition.

10. The intelligent damage identification method for bridge structure based on physical information neural network according to claim 1 is characterized in that: In step 7, the matching of the warning and the repair plan specifically includes: setting a corresponding warning mechanism based on the status assessment analysis results output by the model; when a potential damage risk is detected, the system automatically triggers a warning to prompt relevant personnel; at the same time, matching a preset repair plan according to the assessment results.

Images