A method for identifying hidden damage of bridge structures

Abstract

The present invention relates to a method for identifying hidden damage of a bridge structure, including monitoring the corrosion, deformation and environmental data of the bridge structure; analyzing the type and degree of corrosion, identifying the type, distribution characteristics and corresponding current degree of corrosion; constructing a corrosion prediction and adhesion correlation model according to the degree of corrosion under multiple time series; evaluating the accuracy of the corrosion prediction model by comparing the monitoring data, and adjusting the corrosion prediction model; evaluating the accuracy of the correlation analysis model by comparing the monitoring data, and adjusting the corrosion prediction model; identifying potential hidden danger areas affected by corrosion, analyzing the impact of passing vehicles on these potential hidden danger areas; evaluating the degree of weakening of adhesion and classifying the hidden danger areas, and classifying the affected areas. The present invention solves the problem that traditional bridge corrosion monitoring methods are difficult to identify the type of corrosion and its development trend in real time and accurately; through precise monitoring and prediction, the safety and life of the bridge structure are effectively improved, and the maintenance cost is significantly reduced.

Description

Technical Field

[0001] The present invention relates to the technology of bridge structure damage identification, and particularly to a method for identifying hidden damage of bridge structures. Background Art

[0002] In the field of bridge maintenance and monitoring, ensuring structural safety is of utmost importance. Corrosion on the surface of steel bars and its impact on bond strength are key technical difficulties in bridge health monitoring. Since corrosion is not a homogeneous process, the morphological changes have complex and non-linear characteristics on the bond force, and this change varies with time and environmental factors. Analyzing the correlation between steel bar corrosion and bond strength has become a challenge; it is necessary to cover the corrosion morphology at the microscale starting from the steel bar surface to the measurement of bond force at the macroscale, covering the entire process of corrosion at different stages and its impact on the bond. To ensure the generalization ability of the algorithm, data collection needs to involve different types of bridges, corrosion changes in different environments, and the influence of different material properties; even with comprehensive data support, the specific conditions of different bridges may still limit the applicability of the algorithm in specific situations, such as special usage history, unique structural design, or the influence of specific environmental factors; only by deeply understanding and accurately simulating the influence mechanism of corrosion on the bond force can the progress of bridge health monitoring technology be promoted. Summary of the Invention

[0003] To solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method for identifying hidden damage of bridge structures. The method for identifying hidden damage of bridge structures solves the problem that traditional bridge corrosion monitoring methods are difficult to identify the type of corrosion and its development trend in real time and accurately; through precise monitoring and prediction, it effectively improves the safety and lifespan of bridge structures and significantly reduces maintenance costs.

[0004] A method for identifying hidden damage of bridge structures according to the present invention includes the following steps:

[0005] S1. Monitor and collect corrosion image data, structural deformation data, dynamic response data, and environmental condition data of the bridge structure;

[0006] S2. Analyze the collected corrosion images on the surface of steel bars to identify the type, distribution characteristics, and corresponding current corrosion degree of the corrosion;

[0007] S3. Construct a corrosion prediction model based on the corrosion degree under multiple time series;

[0008] S4. Conduct simulation tests on the bond force between steel bars and concrete with different preset corrosion degrees, obtain multiple groups of correlation data between the preset corrosion degrees and the bond force, and construct a correlation analysis model regarding the corrosion degree on the surface of steel bars and the bond force;

[0009] S5. Compare and analyze the corrosion prediction model with the current monitoring data of different bridge structures, evaluate the accuracy of the corrosion prediction model under the current environmental conditions, and adjust the corrosion prediction model;

[0010] S6. Compare and analyze the correlation analysis model with the current monitoring data of different bridge structures, evaluate the accuracy of the correlation analysis model under the current corrosion degree, and adjust the corrosion prediction model;

[0011] S7. According to the corrosion type, distribution characteristics and corrosion degree, as well as the comparative analysis results of the corrosion prediction model and the current monitoring data, identify the potential hazard areas affected by corrosion, and analyze the impact of passing vehicles on these potential hazard areas;

[0012] S8. According to the stress characteristics of the bridge structure, evaluate the impact of corrosion on the bonding force in the potential hazard areas. Determine the weakening degree of the bonding force caused by different corrosion degrees according to the distance from the area to the stress point, material properties and environmental conditions. Analyze the historical data and the characteristics of passing vehicles, set the threshold for the impact on the bonding force, and classify the affected areas according to this threshold.

[0013] Preferably, the step S1 specifically includes:

[0014] Use an image sensor to obtain high-definition image data on the surface of the bridge structure, segment and extract features from the images, automatically identify and locate the damaged areas, and calculate the proportion of the rust area and the crack length as the evaluation basis for rust monitoring;

[0015] Use strain sensors and accelerometers to collect the strain and vibration acceleration of the bridge structure in real time. According to the change trends and amplitude sizes of the strain and acceleration data, combined with the preset judgment thresholds, determine whether the bridge structure has deformed or vibrated abnormally, and obtain the monitoring data of the bridge structure deformation and dynamic response;

[0016] Use temperature sensors, humidity sensors and corrosion sensors to monitor the temperature, humidity and concentration of corrosive gases in the environment where the bridge is located in real time. According to the correlation between the environmental parameters and the material properties of the bridge structure, establish a quantitative evaluation model to judge the impact degree of the current environmental conditions on the durability of the bridge structure, and obtain the environmental condition monitoring results.

[0017] Preferably, the step S2 specifically includes:

[0018] According to the high-definition image data collected on the surface of the steel bars, use image preprocessing technology to denoise, enhance and standardize the images to improve the image quality and the recognizability of the rust characteristics on the surface of the steel bars;

[0019] Use a semantic segmentation algorithm based on deep learning to perform pixel-level segmentation on the rust image, accurately extract the position and boundary information of the rust area, and obtain a binary mask image of the rust area;

[0020] According to the rust area mask image, calculate the proportion of the rust area in the total surface area of the steel bar, which is an important indicator to measure the degree of rust;

[0021] By extracting the gray histogram features of the rust area, obtain the gray distribution of the rust area;

[0022] By calculating the local binary pattern features of the rust area, capture the microscopic structure and regularity of the rust texture, and combine these feature parameters into a multi-dimensional feature vector as the input for subsequent rust type discrimination and degree evaluation.

[0023] Preferably, the step S3 specifically includes:

[0024] Obtain the historical detection data of the steel bar structure, extract the rust area and maximum rust depth at different time nodes, and construct a time series dataset of rust degree indexed by time;

[0025] Conduct a stationarity test on the time series of rust degree. If it is not stationary, perform differencing until the stationarity test is passed;

[0026] Use the STL decomposition method, based on LOESS fitting, to decompose the time series into a trend term, a seasonal term, and a random term. According to the decomposition results, use linear regression or polynomial regression to model the trend term, use Fourier series or periodic functions to model the seasonal term, and use the ARMA model to model the random term;

[0027] Select a time series prediction model. According to the stationarity, autocorrelation, and trend characteristics of the rust degree time series, combined with the ACF and PACF diagrams, determine the hyperparameters of the model;

[0028] Evaluate the goodness of fit and prediction performance of different models through cross-validation and information criteria, and select the optimal model;

[0029] Introduce environmental factors as covariates, fuse them with the rust degree time series, and construct a rust prediction model for multivariate time series.

[0030] Preferably, the step S4 specifically includes:

[0031] According to the preset rust degree grade, use the method of electrochemical accelerated corrosion to perform artificial rust treatment on the steel bar specimens;

[0032] By controlling the corrosion current density and corrosion time, realize the regulation of the rust degree;

[0033] During the corrosion process, regularly monitor the mass loss rate and volume expansion rate of the steel bars to ensure that the preset corrosion degree is achieved;

[0034] Simulate the non-uniform corrosion in the actual environment, prepare steel bar specimens with different corrosion degrees, measure the corrosion area ratio and corrosion depth on their surfaces, and quantitatively characterize the corrosion degree;

[0035] Pour the steel bar specimens after corrosion treatment and concrete into standard specimens, and adopt the concrete mix ratio and curing conditions similar to those of the actual structure to ensure the representativeness of the specimens;

[0036] Set up control specimens of non-corroded steel bars and concrete to evaluate the influence degree of corrosion on the bond strength;

[0037] Conduct pull-out tests on reinforced concrete specimens with different corrosion degrees. Through displacement sensors and strain gauge test equipment, continuously record the load-displacement curve and steel bar strain distribution during the pull-out process to obtain mechanical parameters reflecting the bond strength;

[0038] According to the mechanical parameters measured in the pull-out test, calculate the bond strength distribution at the interface of reinforced concrete under different corrosion degrees;

[0039] Combined with the volume expansion effect, loose and porous characteristics of the corrosion products and the loss of the steel bar cross-section, analyze the influence of the corrosion products on the bond force transfer mechanism, and identify the key factors leading to the degradation of the bond strength due to corrosion;

[0040] Quantitatively evaluate these characteristics of the corrosion products through the cross-section analysis and microscopic morphology characterization of the pull-out specimens;

[0041] Conduct statistical analysis on the bond strength and interface stress distribution under different corrosion degrees, and extract quantitative indicators of the bond strength;

[0042] Adopt multiple linear regression or non-linear regression methods to establish a correlation analysis model between the corrosion degree on the surface of the steel bar and the bond strength.

[0043] Preferably, the step S5 specifically includes:

[0044] Extract the real-time monitoring data of different bridges, record the information of the regional type, traffic flow and load type where the bridges are located, and construct a bridge corrosion monitoring database;

[0045] Input the current corrosion state parameters of each bridge in the monitoring database into the established corrosion prediction model, obtain the predicted corrosion degree values of each bridge at the current moment, and compare them with the corrosion degree estimated by electrochemical parameters, and calculate the prediction error and relative error;

[0046] Statistically analyze the prediction errors of each bridge, calculate the mean and standard deviation of the prediction errors, and plot the distribution histogram and box plot of the prediction errors to visually evaluate the overall prediction accuracy and stability of the corrosion prediction model on different bridges.

[0047] Preferably, step S6 specifically includes:

[0048] Extract real-time bond force monitoring data, including the ultimate bond strength between the steel bar and concrete obtained by the pull-out method and the steel bar strain distribution obtained by the steel bar strain method, and invert the bond stress distribution between the steel bar and concrete according to the strain distribution;

[0049] For each monitoring section, at least 3 measuring points are arranged, and at least 3 specimens are tested in parallel for each measuring point, and the average value is obtained as the representative value of this section;

[0050] The monitoring frequency is determined according to the importance and deterioration rate of the bridge, and the monitoring data is standardized to form a bond force data set across bridges and structural types;

[0051] Input the current bond force monitoring data into the established correlation analysis model between the corrosion degree of the steel bar surface and the bond force. Inside the model, according to the current corrosion area ratio and corrosion depth parameters, the corresponding theoretical bond force value is predicted as the bond force prediction value under the current corrosion degree;

[0052] For each bridge structure, compare the bond force prediction value with the measured value, calculate the difference between the two as the prediction deviation, and quantify the size of the prediction deviation using absolute value and relative percentage indicators;

[0053] Evaluate the accuracy of the correlation analysis model under the current corrosion degree to determine whether the prediction accuracy of the model meets the actual application requirements.

[0054] Preferably, step S7 specifically includes:

[0055] According to the corrosion type identification result, divide the corrosion of bridge steel bars into three types: uniform corrosion, local corrosion and pitting corrosion;

[0056] Combined with the morphological characteristics, spatial distribution characteristics and formation mechanism of corrosion, formulate a set of multi-index comprehensive evaluation corrosion type classification criteria, use the analytic hierarchy process to determine the weights of each factor, obtain the quantitative index of the corrosion type through weighted summation, and then divide the corrosion type according to the size of the index value;

[0057] Combined with the analysis of the corrosion distribution characteristics, determine the distribution probability and spatial correlation of each type of corrosion in different parts of the bridge, and form a probability model of the bridge corrosion type and distribution;

[0058] Using the rust degree evaluation method, calculate the average rust area ratio and maximum rust depth of the steel bars at each part of the bridge. Combining with the rust type and distribution probability model, comprehensively evaluate the rust degree of the whole bridge and local parts to obtain the spatial distribution map of the bridge rust degree.

[0059] Preferably, step S8 specifically includes:

[0060] According to the force characteristics of the bridge structure, select the applicable finite element analysis software, establish the three-dimensional solid model of the bridge structure, and reasonably divide the mesh;

[0061] In the area with potential rust hazards of the bridge, according to the distribution of steel bar rust, divide the area into several sub-areas, and the rust degree in each sub-area is approximately uniform. For each sub-area, based on the cross-section loss rate of the rusted steel bars, generate the equivalent material parameters of the rusted steel bars by reducing the cross-sectional area and equivalent yield strength of the steel bars;

[0062] Adopt parametric modeling technology, preset the influence area in the geometric model or mesh model, and automatically generate different finite element models by changing the distance and size of the influence area to simulate the spatial variability of the rust area and evaluate the influence law of the position factor;

[0063] For the bond-slip behavior of rusted reinforced concrete, adopt the modified constitutive relationship of CEB-FIP Model Code 2010;

[0064] When modifying the bond-slip constitutive relationship, introduce the dimensionless parameter reflecting the rust degree, establish the quantitative correlation between each parameter and the rust parameter, and calibrate it through the pull-out test;

[0065] Embed the modified rust deterioration-bond slip constitutive relationship into the finite element model to evaluate the weakening degree of the overall bond force of the bridge under different rust degrees.

[0066] Preferably, step S8 further includes:

[0067] For the hazard area exceeding the bond force threshold, use the finite element analysis method to conduct force analysis and hazard assessment, determine its impact on the overall safety of the bridge structure, calculate the maintenance and repair scope, and formulate corresponding maintenance and repair strategies;

[0068] For the non-hazard area not exceeding the bond force threshold, according to the structural characteristics and monitoring data of the bridge, combined with the cost-benefit analysis, evaluate the long-term benefits after replacement or repair, and formulate maintenance and repair strategies.

[0069] The advantages of the method for identifying hidden damage of a bridge structure described in the present invention are as follows:

[0070] A method for identifying hidden damages in bridge structures according to the present invention solves the problem that traditional bridge corrosion monitoring methods are difficult to identify the type of corrosion and its development trend in real time and accurately. By comprehensively monitoring the bridge structure, using image processing and machine learning models to analyze the collected corrosion images, identify the type of corrosion, distribution characteristics and corrosion degree on the surface of steel bars, and construct a corrosion prediction model through time series data to predict future corrosion conditions; through simulation tests, obtain the correlation data between the corrosion degree and the bonding force between the steel bars and the concrete, construct a correlation analysis model, and compare and analyze it with the current monitoring data to evaluate and adjust the accuracy of the corrosion prediction model; according to the type of corrosion, distribution characteristics and corrosion degree, combined with the comparison results of the corrosion prediction model and the current monitoring data, identify the potential hazard areas affected by corrosion and analyze the impact of vehicle flow on these areas; use finite element analysis to evaluate the impact of corrosion on the bonding force, set a bonding force threshold, conduct stress analysis and hazard assessment on the areas exceeding the threshold, and formulate maintenance and repair strategies; through precise monitoring and prediction, effectively improve the safety and service life of the bridge structure, and significantly reduce the maintenance cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a flowchart of a method for identifying hidden damages in a bridge structure according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0072] As Figure 1 shown, a method for identifying hidden damages in a bridge structure according to the present invention includes the following steps:

[0073] S1. Monitor and collect corrosion image data, structural deformation data, dynamic response data and environmental condition data of the bridge structure; specifically, include using an image sensor to monitor corrosion according to the image data, a strain sensor and an accelerometer to monitor structural deformation and dynamic response, and a temperature, humidity and corrosion sensor to monitor environmental conditions;

[0074] S2. Analyze the collected corrosion images on the surface of the steel bars to identify the type of corrosion, distribution characteristics and the corresponding current corrosion degree;

[0075] S3. Construct a corrosion prediction model according to the corrosion degree under multiple time series;

[0076] S4. Conduct simulation tests on the bonding force between the steel bars and the concrete with different preset corrosion degrees, obtain multiple groups of correlation data between the preset corrosion degrees and the bonding force, and construct a correlation analysis model regarding the corrosion degree on the surface of the steel bars and the bonding force;

[0077] S5. Compare and analyze the corrosion prediction model with the current monitoring data of different bridge structures, evaluate the accuracy of the corrosion prediction model under the current environmental conditions, and adjust the corrosion prediction model;

[0078] S6. Compare and analyze the correlation analysis model with the current monitoring data of different bridge structures, evaluate the accuracy of the correlation analysis model under the current corrosion degree, and adjust the corrosion prediction model;

[0079] S7. According to the corrosion type, distribution characteristics and corrosion degree, as well as the comparative analysis results of the corrosion prediction model and the current monitoring data, identify the potential hazard areas affected by corrosion, and analyze the impact of passing vehicles on these potential hazard areas; specifically analyze the weight, speed and frequency of the passing vehicles.

[0080] S8. According to the stress characteristics of the bridge structure, evaluate the impact of corrosion on the bonding force in the potential hazard areas. Determine the weakening degree of different corrosion degrees on the bonding force according to the distance from the area to the stress point, material properties and environmental conditions. Analyze the historical data and the characteristics of passing vehicles, set the threshold of the bonding force impact, and classify the affected areas according to this threshold.

[0081] Further, in this embodiment, step S1 specifically includes:

[0082] Use an image sensor to obtain high-definition image data of the bridge structure surface, segment and extract features from the image, automatically identify and locate the damaged area, calculate the proportion of the rust area and the crack length, and use them as the evaluation basis for rust monitoring; use image processing algorithms such as YOLO or Faster R-CNN to segment and extract features from the image, focusing on rust and crack damage features.

[0083] Use strain sensors and accelerometers to collect the strain and vibration acceleration of the bridge structure in real time. According to the change trend and amplitude of the strain and acceleration data, combined with the preset judgment threshold, determine whether the bridge structure has deformed or vibrated abnormally, and obtain the monitoring data of the bridge structure deformation and dynamic response.

[0084] Use temperature sensors, humidity sensors and corrosion sensors to monitor the temperature, humidity and concentration of corrosive gases in the environment where the bridge is located in real time. According to the correlation between environmental parameters and the material properties of the bridge structure, establish a quantitative evaluation model to judge the impact degree of the current environmental conditions on the durability of the bridge structure, and obtain the environmental condition monitoring results.

[0085] Further, according to the results of rust monitoring, structural deformation monitoring, dynamic response monitoring and environmental condition monitoring, establish a bridge structure health assessment model, use the analytic hierarchy process to assign weights to each monitoring index, and calculate the comprehensive health index by weighted summation.

[0086] According to the comparison between the comprehensive health index and the preset threshold, divide the health state of the bridge structure into normal, mild damage, moderate damage and severe damage grades.

[0087] Evaluate the safety risk level in combination with the importance and service life of the bridge structure;

[0088] Based on the health status and risk level, formulate corresponding maintenance strategies, such as whether local repair, reinforcement, or suspension of use is required, and dynamically adjust the monitoring frequency, such as reducing the sampling interval and increasing the number of sampling points;

[0089] In the early warning mechanism, set multiple levels of early warning thresholds. For example, when the comprehensive health index is lower than 0.7, a yellow early warning is triggered; when it is lower than 0.5, an orange early warning is triggered; and when it is lower than 0.3, a red early warning is triggered;

[0090] The early warning information includes the bridge number, location, health status, risk level, and recommended measures, and is promptly notified to relevant management personnel through text messages, emails, and APP push methods;

[0091] Automatically adjust the monitoring frequency according to the early warning level. For example, when a yellow early warning is issued, increase the sampling frequency by 50%; when an orange early warning is issued, increase it by 100%; and when a red early warning is issued, continuously monitor at a high frequency;

[0092] Use big data analysis technology to mine and analyze historical monitoring data, select the Long Short-Term Memory Neural Network (LSTM) model, with the input including time series data such as strain, acceleration, temperature, and humidity, train the LSTM model, and learn the time law of the degradation of the bridge structure performance;

[0093] During the model training process, adopt early stopping and regularization methods to prevent overfitting. Use the trained LSTM model to perform rolling predictions on the health status of the bridge structure for a future period (such as 1 month, 1 year), and dynamically update the maintenance plan according to the prediction results;

[0094] Adopt three-dimensional visualization technology to intuitively display the monitoring data and analysis results. Based on the BIM model, associate the positions of the monitoring sensors and real-time data with the three-dimensional model to achieve real-time mapping of the monitoring data;

[0095] Set different color early warning levels according to the abnormality degree of the monitoring parameters. For example, green indicates normal, yellow indicates mild abnormality, orange indicates moderate abnormality, and red indicates severe abnormality;

[0096] Through the visual display of the three-dimensional model, bridge management personnel can intuitively understand the overall and local health status of the bridge. The three-dimensional model supports interactive operations such as rotation, zooming, and slicing, facilitating multi-angle and multi-level observation and analysis by management personnel, and improving the readability and usability of the monitoring data;

[0097] Examples are as follows:

[0098] When applying the YOLO algorithm for bridge surface damage recognition, the confidence threshold for damage detection can be set to 0.8, that is, when the confidence of the recognition result is greater than or equal to 0.8, it is considered that there is damage in this area;

[0099] For the identified cracks, they can be divided into minor cracks (less than 10 cm), moderate cracks (10 - 30 cm), and severe cracks (greater than 30 cm) according to their lengths;

[0100] When judging the deformation of the bridge structure, the strain threshold can be set to ±1000 με and the acceleration threshold to ±0.2 g. When the monitored data exceeds the threshold range, it is determined as abnormal deformation or vibration;

[0101] By establishing a mathematical model between environmental parameters and the degradation rate of bridge structural material properties, for example, when the humidity increases by 10%, the corrosion rate of steel bars increases by 0.05 mm / a, the impact of environmental conditions on bridge durability is quantitatively evaluated;

[0102] When calculating the comprehensive health index, the weights of corrosion monitoring, structural deformation monitoring, dynamic response monitoring, and environmental monitoring can be set to 0.3, 0.3, 0.2, and 0.2 respectively, determined by the AHP method;

[0103] When the warning level is yellow, the monitoring frequency is increased from once per hour to once every 30 minutes. When the warning level is orange, it is increased to once every 10 minutes. When the warning level is red, real-time monitoring is carried out;

[0104] When training the LSTM prediction model, the monitoring data of the past 1 year can be selected as the input, and the health index of the next 1 month as the output. The model parameters are optimized through the gradient descent algorithm, and the early stopping rounds are set to 50, and the regularization coefficient is 0.01 to prevent the model from overfitting;

[0105] In 3D visualization, different color mapping rules can be set according to the abnormal degree of monitoring parameters. For example, when the strain exceeds the normal value by less than 10%, it is displayed in yellow; when it exceeds 10% - 20%, it is displayed in orange; when it exceeds 20%, it is displayed in red;

[0106] Through interactive operations, magnification, rotation, and slice analysis of local areas of the bridge can be achieved. For example, magnify to the damaged area, rotate to the direction of the maximum crack opening, and slice analysis of the stress distribution inside the beam-column.

[0107] Furthermore, in this embodiment, step S2 specifically includes:

[0108] According to the high-definition image data collected on the surface of the steel bars, image preprocessing techniques are used to denoise, enhance, and standardize the images to improve the image quality and the recognizability of the corrosion characteristics on the surface of the steel bars; the high-definition image data on the surface of the steel bars is collected by an image sensor;

[0109] Use a semantic segmentation algorithm based on deep learning to perform pixel-level segmentation on the rust image, accurately extract the location and boundary information of the rust area, and obtain a binary mask image of the rust area; The semantic segmentation algorithm based on deep learning can select U-Net or DeepLab;

[0110] According to the rust area mask image, calculate the proportion of the rust area in the total surface area of the steel bar, which is an important indicator to measure the degree of rust;

[0111] By extracting the gray histogram features of the rust area, obtain the gray distribution of the rust area; The gray histogram features include mean, standard deviation, skewness, and kurtosis;

[0112] By calculating the local binary pattern features of the rust area, capture the microscopic structure and regularity of the rust texture, and combine these feature parameters into a multi-dimensional feature vector as the input for subsequent rust type discrimination and degree evaluation;

[0113] To train a multi-class machine learning model, a labeled rust image dataset needs to be prepared, and each image should be labeled with the class label of the rust type (such as uniform rust, pitting corrosion, and patchy corrosion);

[0114] Adopt the method of cross-validation to divide the dataset into training set, validation set, and test set, and the ratio can be set to 7:2:1;

[0115] Select a machine learning algorithm suitable for multi-class tasks, such as logistic regression, support vector machine, and random forest, to train and optimize the model;

[0116] Adjust the hyperparameters of the model through the grid search method, such as the regularization coefficient and the number of trees, to improve the classification accuracy;

[0117] Evaluate the performance of the model on the test set, calculate the classification accuracy, precision, recall rate, and F1 score, and select the model with the best performance for rust type discrimination;

[0118] If the rust type is pitting corrosion or patchy corrosion, further use image morphological operations, such as opening operation and closing operation, to perform morphological analysis on the rust area, and calculate geometric feature parameters such as the number, size, and distribution density of rust points;

[0119] Combine the discrimination results of the rust type with the morphological features of the rust area to construct a comprehensive rust evaluation index system. When establishing a mapping model between rust feature parameters and rust degree, a decision tree regression algorithm can be used;

[0120] The decision tree divides the feature space recursively into different regions, with each region corresponding to a predicted value of the rust level. Using the training dataset, the proportion of rust area, average gray value, texture features, rust type, etc. can be used as input features, and the corresponding rust levels (such as mild, moderate, severe) can be used as target values to train the decision tree model;

[0121] During the training process, by setting parameters such as the maximum depth of the tree and the minimum number of samples in the leaf nodes, the complexity and generalization ability of the tree are controlled;

[0122] The mean squared error is used as the evaluation metric, and the decision tree with the minimum mean squared error is selected as the optimal model;

[0123] When applying, the feature parameters of the new rust image are input into the decision tree, and the corresponding predicted value of the rust level can be obtained;

[0124] The example is as follows:

[0125] When using the U-Net algorithm for rust area segmentation, the input image size can be set to 512x512, the number of convolutional layers of the encoder and decoder are 4 layers and 4 layers respectively, the convolutional kernel size is 3x3, and the activation function is selected as ReLU;

[0126] The model is trained through the cross-entropy loss function and the Adam optimizer, the learning rate is set to 0.001, the batch size is 8, and the number of iterations is 100 rounds;

[0127] When extracting the LBP features of the rust area, 8 sampling points and the LBP operator with a radius of 1 can be selected, and the LBP feature map is divided into an 8x8 grid. Calculate the LBP histogram within each grid, and connect the histograms into a feature vector with a length of 8x8x256 = 16384 dimensions;

[0128] For the multi-classification model of rust type, the support vector machine algorithm can be selected, using the radial basis kernel function. The optimal penalty coefficient C and kernel function parameter gamma are selected through 5-fold cross-validation, and the model performance is evaluated on the test set, and the classification accuracy reaches more than 90%;

[0129] When constructing the decision tree model for rust level assessment, the CART algorithm is selected, and the parameters such as the maximum depth of the decision tree and the minimum number of samples for splitting are optimized through grid search. The mean squared error is used as the evaluation metric to select the optimal decision tree model;

[0130] Furthermore, identify the rust types on the steel bar surface, including uniform rust, local rust and pitting corrosion, evaluate the depth and area of the rust, analyze the distribution characteristics of the rust on the steel bar surface, determine the concentrated area and diffusion trend of the rust, and quantitatively obtain the current rust level;

[0131] Adopt an image segmentation algorithm based on deep learning, such as U-Net or Mask R-CNN, to perform semantic segmentation on the rusted images of the steel bar surface, obtain an accurate mask map of the rusted area, and thus determine the position and range information of the rust;

[0132] According to the rusted area mask map obtained by segmentation, calculate the proportion of the rusted area to the total area of the steel bar surface to obtain a quantitative evaluation result of the rusted area;

[0133] Estimate the average depth and maximum depth of the rusted area by establishing a calibration model between the gray value and the rust depth. Extract the texture, color, and shape of the rusted area;

[0134] For the texture feature, adopt the gray-level co-occurrence matrix algorithm to calculate the gray-level co-occurrence probability matrix in different directions and distances, and extract texture indicators such as energy, entropy, contrast, and homogeneity;

[0135] For the color feature, select the HSV color space and calculate the histograms of the three channels of H, S, and V. For the shape feature, calculate geometric parameters such as the area, perimeter, eccentricity, and convex hull area of the rusted area;

[0136] Construct a high-dimensional feature vector by integrating multiple features, and use machine learning algorithms such as support vector machines or random forests for training and optimization to achieve automatic classification of rust types, including uniform rust, local rust, and pitting corrosion;

[0137] Select the optimal model parameters through grid search and cross-validation to improve the accuracy and generalization ability of classification;

[0138] Divide the steel bar surface into several grid areas, count the rust area proportion and average depth in each grid, and draw a rust distribution heat map through a reasonable color mapping scheme to visually present the spatial distribution characteristics of the rust;

[0139] Adopt the K-means or DBSCAN clustering algorithm to perform clustering analysis on the heat map, find out the concentrated areas with the highest degree of rust, and provide key attention objects for subsequent analysis of rust diffusion and development;

[0140] On the boundary of the rust concentrated area, use the Canny operator for edge detection to extract the position information of the boundary pixels;

[0141] Calculate the gray-level gradient direction and magnitude of the boundary pixels through the Sobel operator, quantize the gradient direction into 8 discrete directions, and count the average gradient magnitude in each direction as the rust diffusion rate in that direction;

[0142] Comprehensively analyze the diffusion rates of the rust-concentrated areas in different directions, judge the diffusion trend of rust, simulate the changes of the rust areas over a period of time in the future, and provide a basis for predicting the development of rust. Considering multiple indicators such as the rust area ratio, average depth, maximum depth, rust type, and diffusion trend, establish a quantitative model for the degree of rust;

[0143] Use the Analytic Hierarchy Process (AHP) to rank the importance of each indicator and allocate weights, construct a judgment matrix and conduct a consistency test;

[0144] Use the Fuzzy Comprehensive Evaluation Method (FCE) to map the membership functions of each indicator to a unified evaluation scale, and obtain the comprehensive membership degree of the rust degree through fuzzy operations to determine the rust degree level;

[0145] Introduce the time dimension in the process of evaluating the rust degree, analyze the change trend of the rust degree in different time periods, and realize the dynamic evaluation of the rust process;

[0146] Track and analyze the rust images collected at different time points of the same steel bar, extract key indicators such as the rust area ratio and average depth, and construct time series data;

[0147] The example is as follows:

[0148] When using U-Net for rust area segmentation, the input image size can be set to 512x512. Both the encoder and decoder contain 4 convolutional layers, the convolutional kernel size is 3x3, and the number of feature maps is 64, 128, 256, and 512 in sequence;

[0149] Use the Dice coefficient as the loss function and the Adam optimizer for training. The learning rate is 0.001, the batch size is 8, and 100 epochs are iterated. When extracting the GLCM texture features of the rust area, the distance d = 1, directions θ = 0°, 45°, 90°, 135° can be selected, the gray level is 16, and 4 indicators of energy, entropy, contrast, and homogeneity are extracted;

[0150] Use SVM for rust type classification, use the RBF kernel function, select the penalty coefficient C = 10 through 5-fold cross-validation, the kernel function parameter γ = 0.01, and the classification accuracy on the test set reaches 95%;

[0151] When drawing the rust distribution heat map, divide the surface of the steel bar into a 50x50 grid, normalize the rust area ratio in each grid, map it to the JET pseudocolor color table, and the areas with higher area ratios are more reddish in color;

[0152] Use the K-means clustering algorithm to cluster the rust distribution heat map into 3 categories, corresponding to mild, moderate, and severe rust areas respectively.

[0153] Furthermore, in this embodiment, step S3 specifically includes:

[0154] Obtain the historical inspection data of the steel bar structure, extract the rust area and the maximum rust depth at different time nodes, and construct a rust degree time series dataset indexed by time;

[0155] Conduct a stationarity test on the rust degree time series. If it is not stationary, perform differencing until the stationarity test is passed; if it is not stationary, perform differencing until the ADF test or KPSS test is passed;

[0156] Adopt the STL decomposition method, based on LOESS fitting, decompose the time series into a trend term, a seasonal term, and a random term. According to the decomposition results, use linear regression or polynomial regression to model the trend term, use Fourier series or periodic functions to model the seasonal term, and use the ARMA model to model the random term;

[0157] Select a time series prediction model, and determine the hyperparameters of the model according to the stationarity, autocorrelation, and trend characteristics of the rust degree time series, combined with the ACF and PACF plots. The time series prediction model can be ARIMA, SARIMA, Holt-Winters; the hyperparameters include the p, d, and q values of the ARIMA model;

[0158] Evaluate the goodness of fit and prediction performance of different models through cross-validation and information criteria, and select the optimal model;

[0159] Introduce environmental factors as covariates, fuse them with the rust degree time series, and construct a rust prediction model for multivariate time series;

[0160] Furthermore, adopt a sliding window method to divide the rust degree time series into a training set and a test set, and conduct cross-validation. According to the length of the time series and the complexity of the prediction model, select an appropriate training sample length (e.g., 1 year) and prediction step length (e.g., 1 month) to improve the timeliness of prediction while ensuring the stability of the model. Evaluate the performance of the prediction model, and use indicators such as root mean square error (RMSE) and mean absolute percentage error (MAPE) to measure the deviation between the predicted value and the actual value;

[0161] Construct a prediction interval, and quantify the uncertainty of the prediction result based on the confidence level (e.g., 95%) and the error distribution assumption. Perform online update and optimization on the prediction model, adopt a sliding window method, fix the training sample length, and regularly (e.g., monthly) update the model parameters, and at the same time modify the exogenous variables in the ARIMAX or SARIMAX model to incorporate the latest environmental monitoring data;

[0162] Continuously monitor the prediction performance of the model, calculate the rolling prediction error, and trigger model retraining or model structure adjustment when the prediction error exceeds the preset threshold;

[0163] Compare the prediction results with the actual detection data, calculate the residual sequence, feedback the residuals into the model for dynamic correction, continuously improve the prediction accuracy and reliability, and adapt to the non-stationary characteristics of the steel bar corrosion process;

[0164] The example is as follows:

[0165] When obtaining the historical detection data of the steel bar structure, you can select the data of the recent 5 years, with 12 detections per year, to construct a monthly corrosion degree time series with a length of 60;

[0166] Conduct an ADF test on this time series. If the P-value is greater than 0.05, perform a first-order difference operation and re-conduct the ADF test until the P-value is less than 0.05 to ensure the stationarity of the time series;

[0167] Adopt the STL decomposition method, set the seasonal period to 12, the window span of the trend term to 5, and the cut-off frequency of the low-pass filter to 0.5 to extract the trend term, seasonal term, and random term of the time series;

[0168] Perform quadratic polynomial regression on the trend term, expand the seasonal term using a 6th-order Fourier series, and fit the random term using an ARMA(1,1) model;

[0169] When selecting a time series prediction model, train and evaluate the ARIMA(1,1,1), SARIMA(1,1,1)(1,1,1)_12, and Holt-Winters models respectively, and select the model through the AIC criterion. The results show that the SARIMA model has the smallest AIC value, so SARIMA(1,1,1)(1,1,1)_12 is selected as the optimal model;

[0170] When introducing environmental factors, select the monthly average temperature, monthly average relative humidity, and monthly total rainfall related to the corrosion process as covariates, conduct a partial correlation analysis with the corrosion degree time series. If the P-value of the partial correlation coefficient is less than 0.05, include this environmental factor in the SARIMAX model to construct a SARIMAX(1,1,1)(1,1,1)_12 model;

[0171] When dividing the training set and the test set, set the training sample length to 48 and the prediction step to 6, that is, use the data of the first 4 years to train the model, predict the corrosion degree of the next half year, and use the expanding window method for cross-validation to obtain a series of prediction error values;

[0172] For model performance evaluation, the RMSE between the predicted value and the true value is calculated to be 0.02, and the MAPE is 3%, indicating a relatively high precision of the prediction model;

[0173] Based on the bootstrap method, a 95% prediction interval is constructed, and the average interval width is 0.05, reflecting the degree of uncertainty of the prediction results;

[0174] When the model is updated online, whenever new detection data arrives, it is added to the training set. At the same time, the earliest observed value is removed, the SARIMAX model is retrained, and the model parameters are updated;

[0175] When the RMSE of the prediction error exceeds 0.05 for three consecutive months, the model structure adjustment is triggered. The differencing order, autoregressive order, and moving average order are reselected, and new environmental factors are introduced to continuously optimize the prediction performance of the model. Through online update and feedback correction, the prediction model can adapt to the dynamic changes of the steel bar corrosion process, improving the prediction accuracy and reliability.

[0176] Further, in this embodiment, step S4 specifically includes:

[0177] According to the preset rust level, the artificial rust treatment of the steel bar specimens is carried out by means of electrochemical accelerated corrosion;

[0178] By controlling the corrosion current density and corrosion time, the regulation of the rust level is realized; the corrosion current density is 0.1 - 1 mA / cm 2 and the corrosion time is 1 - 30 days;

[0179] During the corrosion process, the mass loss rate and volume expansion rate of the steel bars are regularly monitored to ensure that the preset rust level is reached;

[0180] Simulate the non-uniform corrosion in the actual environment, prepare steel bar specimens with different rust levels, measure the rust area ratio and rust depth on their surfaces, and quantitatively characterize the rust level; the simulation of non-uniform corrosion in the actual environment can prefabricate some defects on the steel bar surface, such as scratches and pits, to induce local corrosion;

[0181] The rust-treated steel bar specimens are cast into standard specimens with concrete, and the concrete mix ratio and curing conditions similar to those of the actual structure are adopted to ensure the representativeness of the specimens;

[0182] Set the control group specimens of non-rusted steel bars and concrete to evaluate the influence degree of rust on the bond strength;

[0183] Pull-out tests were conducted on reinforced concrete specimens with different degrees of corrosion. Through displacement sensors and strain gauge test equipment, the load-displacement curves and steel bar strain distributions during the pull-out process were continuously recorded to obtain mechanical parameters reflecting the bond strength. The mechanical parameters reflecting the bond strength include the ultimate bond strength and the yield bond strength.

[0184] Based on the mechanical parameters measured from the pull-out tests, the bond strength distributions at the reinforced concrete interfaces with different degrees of corrosion were calculated.

[0185] Combined with the volume expansion effect, loose and porous characteristics of the corrosion products, and the cross-sectional loss of the steel bars, the influence of the corrosion products on the bond force transfer mechanism was analyzed to identify the key factors leading to the degradation of the bond strength due to corrosion.

[0186] Through the cross-section analysis and microscopic morphology characterization of the pull-out specimens, these characteristics of the corrosion products were quantitatively evaluated. Quantitatively evaluating these characteristics of the corrosion products can provide a basis for revealing the corrosion damage mechanism.

[0187] Statistical analysis was carried out on the bond strengths and interface stress distributions at different degrees of corrosion to extract quantitative indicators of the bond force. The quantitative indicators of the bond force include the peak bond strength, the residual bond strength, and the stress distribution uniformity.

[0188] Using multiple linear regression or non-linear regression methods, a correlation analysis model between the corrosion degree on the steel bar surface and the bond force was established.

[0189] Taking multiple linear regression as an example, the proportion of the corrosion area, the average corrosion depth, and the rust layer thickness were selected as independent variables, and the ultimate bond strength and the residual bond strength were selected as dependent variables, and the regression equation was fitted by the least squares method.

[0190] During the fitting process, outlier detection and elimination were performed on the sample data to improve the robustness of the regression model.

[0191] Model diagnosis was carried out, such as residual analysis and multicollinearity test, to ensure the statistical significance and predictive ability of the model.

[0192] For complex non-linear relationships, polynomial regression, exponential regression, logarithmic regression model forms can be used, or machine learning algorithms, such as support vector machines and neural networks, can be used for non-linear modeling.

[0193] Based on the above test data and correlation analysis model, a numerical simulation method, such as finite element analysis, was used to numerically simulate the bond behavior at the reinforced concrete interface. Considering the degradation of the interface properties caused by corrosion, such as stiffness reduction and strength decline, parametric characterization of corrosion damage was realized in the numerical simulation.

[0194] Verify the rationality of the correlation model through the comparison between the numerical simulation results and the pull-out test results, and provide a theoretical explanation for the test phenomena;

[0195] When applying the correlation analysis model to practical engineering, it is necessary to obtain the corrosion monitoring data of reinforced concrete structures. On the one hand, use the structural health monitoring system to deploy corrosion sensors at key positions, such as linear polarization resistance sensors and electrochemical impedance sensors, to monitor the corrosion state of the steel bars in real time and obtain data such as the corrosion area ratio and corrosion current density; on the other hand, adopt non-destructive testing techniques, such as ultrasonic testing and radar detection, to regularly inspect the structure and obtain the spatial distribution information of the steel bar corrosion;

[0196] Input the monitoring data and detection data into the correlation analysis model to predict the bond force index of the structure, and consider the influence of environmental factors and load factors to correct and dynamically update the correlation analysis model. The prediction results can be used as an important basis for structural performance evaluation and repair and reinforcement decision-making, providing technical support for the safe operation of existing reinforced concrete structures;

[0197] The examples are as follows:

[0198] During the electrochemical accelerated corrosion process, HRB400 grade deformed bars with a diameter of 16 mm and a length of 500 mm are selected, and 3.5% NaCl solution is used as the corrosion medium. The corrosion current density is set to 0.5 mA / cm 2 The corrosion time is 5 days, and slightly corroded specimens with a corrosion area ratio of 5% are prepared;

[0199] Pour the corroded steel bars and concrete of strength grade C30 into cylindrical pull-out specimens with a diameter of 150 mm and a height of 300 mm, and compare them with non-corroded steel bar specimens. Apply a monotonic pull-out load through a universal testing machine, control the pull-out rate at 0.1 kN / s, synchronously collect the load and displacement data, and draw the load-displacement curve;

[0200] Identify the ultimate bond strength through the mutation points on the curve. The average value of the ultimate bond strength of the slightly corroded specimens is calculated to be 5.2 MPa, which is 15% lower than that of the non-corroded specimens. Observe the cross-section of the pull-out specimens by scanning electron microscopy, and it is found that a layer of loose corrosion products adheres to the surface of the steel bars, with a thickness of about 120 μm. Based on the ultimate bond strength data of specimens with different corrosion degrees, use the least square method to fit the linear regression equation of the corrosion area ratio x and the ultimate bond strength y: y = -0.35x + 6.8, and the correlation coefficient R 2 = 0.92, indicating a significant negative correlation between the two;

[0201] In finite element numerical simulation, the cohesive-zone model is adopted to describe the bond-slip behavior between steel bars and concrete. The bond constitutive parameters of the interface are set according to the proportion of the rusted area. For example, in the 5% rusted state, the bond strength of the interface is reduced by 20%, and the fracture energy is reduced by 15%.

[0202] The pull-out load-displacement curve obtained by numerical simulation is in good agreement with the test results, verifying the rationality of the bond-slip model.

[0203] This method is applied to the reinforced concrete sheet piles in a port project. The instantaneous corrosion rate of the steel bars is monitored by embedded corrosion sensors, and the average rusted area proportion of the steel bars is obtained as 3% through radar non-destructive testing. Substituting it into the correlation analysis model, it is predicted that the ultimate bond strength of the sheet piles has decreased by 10%. Based on this, the bearing capacity of the sheet piles is checked and the safety assessment is carried out.

[0204] Furthermore, in this embodiment, step S5 specifically includes:

[0205] Extract the real-time monitoring data of different bridges, record the information of the regional type, traffic flow and load type where the bridges are located, and construct a bridge rust monitoring database. Specifically, extracting the real-time monitoring data of different bridges includes the key factors affecting rust, such as environmental temperature, humidity, and chloride ion concentration, as well as the electrochemical parameters on the surface of the steel bars, such as the natural corrosion potential and polarization resistance. For environmental temperature and humidity, high-precision sensors are selected, with a measurement range covering -20°C to 60°C and 0 to 100% RH, and a resolution of 0.1°C and 0.1% RH. Data is collected every 1 hour. For chloride ion concentration, an electrochemical sensor, such as an Ag / AgCl reference electrode, is selected, with a measurement range of 0.001 to 5 mol / L, and data is collected every 24 hours. For the electrochemical parameters of the steel bars, the linear polarization method or the electrochemical impedance spectroscopy method is used to measure the natural corrosion potential and polarization resistance, and the corrosion current density is estimated according to the Stern-Geary formula. Data is collected every 7 days.

[0206] Input the current rust state parameters of each bridge in the monitoring database into the constructed rust prediction model to obtain the predicted rust degree values of each bridge at the current moment, and compare them with the rust degree estimated by electrochemical parameters. Calculate the prediction error and relative error. Conduct a statistical analysis of the prediction errors of each bridge, calculate the mean and standard deviation of the prediction errors, and draw the distribution histogram and box plot of the prediction errors to visually evaluate the overall prediction accuracy and stability of the rust prediction model on different bridges.

[0207] Furthermore, the Pearson correlation coefficient, Spearman rank correlation coefficient, partial correlation analysis, and scatter plot matrix method are used to analyze the correlation between the prediction deviation and factors such as bridge environmental conditions, structural characteristics, and load conditions, quantitatively and qualitatively evaluate the influence degree of each factor on the prediction deviation, and determine the applicable conditions and limitations of the corrosion prediction model in practical applications;

[0208] For bridges with large prediction deviations, extract their historical monitoring data and existing inspection data, and make targeted modifications and optimizations to the internal structure and parameters of the corrosion prediction model;

[0209] Combined with the results of the correlation analysis, targetedly adjust the input features of the model, modify the weight coefficients of each feature, and add non-linear terms to improve the expression ability and fitting accuracy of the model; Apply the optimized corrosion prediction model to the current monitoring data of each bridge again, and use the k-fold cross-validation method to randomly divide the monitoring data of different bridges into k mutually exclusive subsets. Each time, select k-1 of these subsets as the training set, and the remaining 1 subset as the test set. Repeat the experiment k times, and take the average value of the k experimental results as the performance index of the optimized model;

[0210] Evaluate the improvement effect of the prediction accuracy before and after model optimization through the root mean square error, mean absolute percentage error, and coefficient of determination, and objectively evaluate the effectiveness of model optimization;

[0211] Integrate the monitoring data, model prediction results, and model optimization conditions of each bridge to establish a knowledge base and case base for bridge corrosion prediction; Summarize the key influencing factors of corrosion prediction for different types of bridges, the applicable conditions of the prediction model, and the selection principles of optimization methods, and form a set of standardized corrosion prediction model application and update processes;

[0212] Embed the process into the bridge health monitoring system to realize the automated and intelligent application of the corrosion prediction model, and continuously improve the accuracy and reliability of corrosion prediction through continuous data accumulation and model optimization, providing a scientific basis for bridge maintenance decision-making and safety assessment;

[0213] Examples are as follows:

[0214] In the health monitoring system of a certain cross-sea bridge, 20 temperature and humidity sensors and 10 chloride ion concentration sensors are arranged. The data collection frequencies are 1 hour and 1 day respectively. At the same time, electrochemical sensors are installed at 200 key parts of the bridge, and the natural corrosion potential and polarization resistance data of the steel bars are collected every 7 days. The monitoring data shows that the temperature range of the environment where the bridge is located is -5°C to 40°C, the humidity range is 50% to 90%, and the chloride ion concentration range is 0.01 to 0.5 mol / L;

[0215] According to the Stern-Geary formula, the average corrosion current density of the steel bars is estimated to be 1.5 μA / cm 2 ;

[0216] Input the above monitoring parameters into the established corrosion prediction model, and the corrosion area percentage of the bridge in the current state is obtained as 2%, while the actually detected corrosion area percentage is 2.5%, and the relative error is 20%;

[0217] Through the statistical analysis of the prediction errors at 100 historical time points, the mean value of the prediction errors is obtained as 0.5%, and the standard deviation is 0.3%; draw the distribution histogram of the prediction errors and find that the errors are mainly concentrated between 0.2% and 0.8%, showing a normal distribution;

[0218] Further calculate that the Pearson correlation coefficient between the prediction error and the environmental temperature is 0.8, and the Spearman rank correlation coefficient with the chloride ion concentration is 0.6, indicating that there is a significant positive correlation between the prediction error and the environmental factors;

[0219] Under the condition of keeping other factors unchanged, for every 10 °C increase in the environmental temperature, the prediction error increases by 0.2%; for every 0.1 mol / L increase in the chloride ion concentration, the prediction error increases by 0.15%;

[0220] For the case where the prediction error of the bridge exceeds 0.5%, take the environmental temperature and chloride ion concentration as the new model input features, and use the support vector machine algorithm to retrain the model. Under 5-fold cross-validation, the root mean square error is reduced by 20%, the mean absolute percentage error is reduced by 15%, and the coefficient of determination is increased by 0.1, indicating that the prediction accuracy of the model has been significantly improved;

[0221] Based on the above analysis results, a bridge corrosion prediction knowledge base containing 1000 actual engineering cases is established, covering the value ranges of key influencing factors of different types of bridges, the applicable conditions of the prediction model, and the selection principles of optimization algorithms, forming a set of standardized application processes for bridge corrosion prediction, and embedding them into the bridge health monitoring system to achieve the automated and intelligent application of the corrosion prediction model;

[0222] Furthermore, through comparative analysis, identify the deviation between the predicted corrosion degree and the current corrosion degree of the bond force change model under the current environmental conditions, and analyze the influence of temperature and humidity on the corrosion degree;

[0223] Adopt the established bond force change model, input the parameters such as the current measured steel bar corrosion area ratio and average corrosion depth, predict the bond force between the steel bar and the concrete under the current environmental conditions, and compare it with the bond force estimated by theoretical calculation or empirical formula, and calculate the absolute value and relative percentage of the prediction deviation;

[0224] Obtain the predicted values and estimated values of the bonding force over a historical period (e.g., 1 year), plot the time series curve of the prediction deviation, analyze the change trend, fluctuation period, and mutation time point of the prediction deviation, and compare it with the ambient temperature and humidity data in the same period to identify the correlation between the prediction deviation and the changes in environmental factors;

[0225] Calculate the statistical indicators of the prediction deviation, such as mean, standard deviation, skewness, and kurtosis. Based on the numerical size and change trend of the indicators, judge the distribution type and dispersion degree of the prediction deviation, and calculate the Pearson correlation coefficient or Spearman rank correlation coefficient between the prediction deviation and the ambient temperature and humidity to quantitatively measure the correlation strength and direction between the prediction deviation and environmental factors, providing a basis for subsequent model correction;

[0226] Use statistical methods such as partial correlation analysis and multiple regression analysis to quantitatively evaluate the influence degree of temperature and humidity on the prediction deviation under the condition of controlling other factors unchanged, obtain the partial correlation coefficient, regression coefficient, etc. of temperature and humidity, establish a quantitative association model between the prediction deviation and environmental factors, and on this basis, correct and optimize the original bonding force change model;

[0227] Determine the influence magnitude and direction of temperature and humidity on the prediction deviation according to the quantitative association model, and compare the influence weights of the two factors; select a suitable functional form (e.g., linear function, quadratic function, exponential function), introduce temperature and humidity into the bonding force change model, estimate the function coefficients and other model parameters using methods such as regression analysis, and conduct statistical tests and diagnoses to ensure that the corrected model is reasonable statistically;

[0228] Use the corrected model to re-predict the historical data, calculate the prediction deviation, compare it with the prediction deviation before correction, and evaluate the effect of model correction. By repeatedly iterating the above steps, continuously optimize the internal structure and parameters of the bonding force change model to improve the model's expression ability and applicable range;

[0229] Adopt the sensitivity analysis method to evaluate the sensitivity of the corrected bonding force change model to temperature and humidity changes, that is, change the values of temperature and humidity respectively, observe the change range of the model prediction deviation, and determine the adaptability and stability of the model to environmental factors; in engineering applications, real-time collect the temperature and humidity data of the environment where the reinforced concrete structure is located, input it into the corrected bonding force change model, and obtain the predicted value of the bonding force considering environmental influence;

[0230] According to the quantitative association model between environmental factors and prediction deviation, calculate the prediction deviation compensation value under the current environmental conditions, calibrate the predicted value of the bonding force, and obtain a relatively reliable estimated value of the bonding force. However, when applying the estimated value, the influence of other factors (e.g., load level, member size) also needs to be considered;

[0231] For the prediction of bond strength under different environmental conditions (such as high temperature and high humidity, low temperature and low humidity), a stratified sampling method is adopted to select representative reinforced concrete structures (such as bridges, wharves, buildings) for key test observations. A detailed test plan and operating procedures are formulated to clarify the purpose, content, method, frequency, and personnel division of the test, ensuring the standardization of the test process and the reliability of the data.

[0232] During the test process, the changes in environmental conditions are strictly controlled, and the interference of other factors is minimized as much as possible. The observed data is sorted, analyzed, and fed back in a timely manner, which is used to verify and optimize the bond strength change model and also provides experience and guidance for other similar projects.

[0233] Examples are as follows:

[0234] Taking a caisson wharf in a seaport project as an example, using the established bond strength change model, inputting the currently measured average rust area ratio of steel bars of 3% and the average rust depth of 0.2 mm, it is predicted that under the current environmental conditions (temperature 25°C, relative humidity 75%), the bond strength between the steel bars and concrete is 8.5 MPa. According to the empirical formula in the port engineering industry standard JTS167-4-2019, the actual bond strength is 7.8 MPa, the absolute value of the prediction deviation is 0.7 MPa, and the relative error is 9.0%. Through the analysis of the time series of prediction deviations in the past 1 year, it is found that the mean of the prediction deviations is 0.6 MPa, the standard deviation is 0.3 MPa, the skewness is 0.5, and the kurtosis is 2.8, which basically follows a normal distribution.

[0235] The Pearson correlation coefficient between the prediction deviation and the environmental temperature is 0.7 (P<0.01), and the Spearman rank correlation coefficient with the relative humidity is 0.6 (P<0.01), showing a significant positive correlation.

[0236] Further using partial correlation analysis, keeping other factors unchanged, for every 1°C increase in temperature, the prediction deviation increases by 0.05 MPa; for every 1% increase in relative humidity, the prediction deviation increases by 0.03 MPa.

[0237] On this basis, selecting the quadratic function form, introducing the temperature t and relative humidity h into the bond strength change model, the modified model is obtained: y = a0 + a1x1 + a2x2 + b1t + b2t 2 + c1h + c2h 2， where y is the predicted bond strength value, x1 is the rust area ratio, x2 is the average rust depth, and a0, a1, a2, b1, b2, c1, c2 are parameters to be estimated.

[0238] Parameter estimation is carried out using the data of the most recent year, and the root mean square error of the corrected model on the training set is 0.4 MPa, and the coefficient of determination is 0.85, and the prediction accuracy is significantly better than that before correction;

[0239] Using sensitivity analysis, it is found that within the common range of temperature 20 - 30 °C and relative humidity 60% - 80%, the prediction deviation of the corrected model is less than 0.5 MPa, indicating that the model has good adaptability to environmental factors;

[0240] In subsequent wharf maintenance, the environmental temperature and humidity are monitored in real time, input into the corrected bond strength change model, and combined with the prediction deviation compensation value of 0.2 MPa under the current environmental conditions, the dynamically calibrated bond strength prediction value is obtained, providing an important reference for safety assessment and service life prediction.

[0241] Furthermore, in this embodiment, step S6 specifically includes:

[0242] Extract real-time bond strength monitoring data, including the ultimate bond strength between the steel bar and the concrete obtained by the pull-out method, and the steel bar strain distribution obtained by the steel bar strain method, and invert the bond stress distribution between the steel bar and the concrete according to the strain distribution;

[0243] For each monitoring section, at least 3 measuring points are arranged, and each measuring point tests at least 3 specimens in parallel, and the average value is obtained as the representative value of this section;

[0244] The monitoring frequency is determined according to the importance and deterioration rate of the bridge, and the monitoring data is standardized to form a bond strength data set across bridges and structural types;

[0245] Input the current bond strength monitoring data into the established correlation analysis model between the corrosion degree of the steel bar surface and the bond strength. Inside the model, according to the current corrosion area ratio and corrosion depth parameters, the corresponding theoretical bond strength value is predicted as the bond strength prediction value under the current corrosion degree;

[0246] For each bridge structure, compare the bond strength prediction value with the measured value, calculate the difference between the two as the prediction deviation, and quantify the size of the prediction deviation using absolute value and relative percentage indicators;

[0247] Evaluate the accuracy of the correlation analysis model under the current corrosion degree, and determine whether the prediction accuracy of the model meets the actual application requirements. Evaluating the accuracy of the correlation analysis model under the current corrosion degree requires special attention to the deviation between the prediction value and the measured value;

[0248] According to the evaluation results, if it is found that the prediction accuracy of the model is not ideal, the rust prediction model needs to be adjusted in a timely manner, including optimizing the model parameters and introducing more influencing factors to improve the prediction performance of the model under different rust levels. At the same time, summarize and analyze the prediction deviations of different bridge structures to obtain the average prediction deviation and dispersion degree of the correlation analysis model under different structural types and different rust levels;

[0249] Adopt statistical methods such as hypothesis testing. At a given confidence level (e.g., 95%), judge whether the prediction deviation of the correlation analysis model is significantly different from zero under the current rust level condition, that is, whether there is a systematic deviation between the predicted value and the measured value of the model;

[0250] According to the distribution type and sample size of the deviation data, select an appropriate hypothesis testing method, such as one-sample t-test, Wilcoxon signed-rank test. Before testing, perform a normality test (e.g., Shapiro-Wilk test) and a heteroscedasticity test (e.g., Levene test) on the deviation data to determine the applicable test method;

[0251] If there is a significant deviation, it means that there is a prediction deviation in the model under the current working conditions, and the model needs to be adjusted and corrected; comprehensively analyze the key factors affecting the relationship between the degree of steel bar rust and bond strength based on existing theoretical research results and a large amount of experimental data. In addition to the rust area ratio and rust depth, for example, rust layer thickness, type of rust products, concrete strength, and use these factors as new model inputs;

[0252] Adopt data mining methods such as association rule mining, decision tree, and random forest to mine the association rules between various influencing factors and the bond strength prediction deviation. For example, after discretizing each factor and the prediction deviation according to a certain threshold, use the Apriori and FP-growth algorithms to mine the frequent item sets and association rules between each factor and the prediction deviation;

[0253] Use the mined association rules to guide the structural optimization of the correlation analysis model. For example, adjust the input features of the model and embed physical mechanisms according to the rules to improve the prediction accuracy and applicability of the model;

[0254] For different types of bridge structures, construct a rust-bond strength correlation analysis sub-model based on the structural characteristics. For example, for prestressed concrete bridges and steel-concrete composite bridges, train and optimize the correlation analysis models applicable to specific structures respectively;

[0255] In the process of constructing the sub-model, make full use of the optimized overall correlation analysis model in steps 3-5, and combine the monitoring data of various bridge structures to make targeted adjustments and improvements to the input, output, and algorithm of the sub-model;

[0256] By means of model fusion, ensemble learning and other methods, combine each sub-model with the optimized overall model to form a comprehensive and cross-structure applicable correlation analysis model, so as to improve the applicability and generalization ability of the model;

[0257] Establish a model continuous optimization and update mechanism, regularly collect the bond force monitoring data and corrosion state evaluation data of each bridge structure, supplement them to the existing training data set, and combine the latest theoretical research results to iteratively update the structure and parameters of the correlation analysis model; formulate the standards and trigger conditions for model update. For example, when the amount of newly collected monitoring data reaches a certain scale (such as an increase of 20%), or the distribution characteristics of the new data change significantly (such as the mean and variance exceed the original range), trigger the model update;

[0258] The examples are as follows:

[0259] In the health monitoring of a certain cross-sea bridge, using the distributed optical fiber sensing technology, fiber Bragg grating sensors are arranged at 800 monitoring sections of the whole bridge to collect the steel bar strain data in real time; every six months, use the wedge-shaped pressing block method to conduct on-site detection of the bond performance between the steel bars and the concrete at 100 representative sections, and obtain indicators such as the ultimate bond strength;

[0260] Through the correlation analysis model constructed by the BP neural network, predict the theoretical value of the bond strength of the whole bridge under the current corrosion state (average corrosion area 3.2%, average corrosion depth 0.25mm), and compare it with the measured value. It is found that the mean value of the prediction deviation is 0.42MPa, and the relative error is 8.1%. After a one-sample t-test (confidence level 95%), it is judged that the mean value of the deviation is significantly greater than 0, that is, there is a systematic overestimation phenomenon in the model;

[0261] Further adopt the Apriori association rule mining algorithm, and find that when the corrosion depth is greater than 0.3mm and the concrete strength is lower than 40MPa, the probability that the prediction deviation exceeds 0.5MPa reaches 65%; accordingly, introduce a quantitative index of concrete strength into the model, and through multiple nonlinear regression, construct a quantitative relationship between the corrosion state, concrete strength and bond strength;

[0262] For the prestressed concrete T-beam of the upper structure of this bridge, train a sub-model, and describe the influence law of the corrosion difference between the web and the bottom plate of the T-beam on the bond performance through piecewise linear fitting; finally, adopt the Stacking fusion strategy to weight and combine the sub-model and the overall model, which significantly improves the prediction accuracy of this type of bridge (the relative error is reduced to 5.3%);

[0263] Within the next six months, 1000 new sets of bond performance monitoring data were added, with 120 sets coming from a newly built steel-concrete composite girder cable-stayed bridge. When the new data volume exceeds the threshold (15%), the incremental learning process is automatically triggered. Through the Mini-batch gradient descent algorithm, the model is updated online to keep the performance indicators stable within the warning threshold (relative error of 6% and absolute error of 0.5 MPa). The model knowledge base was continuously enriched, forming a set of knowledge collections containing 20 rules and 5 cases, which are used to guide the bond performance prediction and evaluation of similar bridges.

[0264] Furthermore, in this embodiment, step S7 specifically includes:

[0265] According to the rust type identification results, the corrosion of bridge steel bars is divided into three types: uniform corrosion, local corrosion, and pitting corrosion.

[0266] Combined with the morphological characteristics, spatial distribution characteristics, and formation mechanism of rust, a set of multi-index comprehensive evaluation criteria for rust type classification is formulated. The analytic hierarchy process is used to determine the weights of various factors, and the quantitative index of rust type is obtained through weighted summation. Then, the rust type is divided according to the size of the index value.

[0267] Combined with the analysis of the rust distribution characteristics, the distribution probability and spatial correlation of each type of rust in different parts of the bridge are determined, forming a probability model of the bridge rust type and distribution. Each type of rust in different parts of the bridge includes the bridge deck, bridge pier, and bearing.

[0268] Using the rust degree evaluation method, calculate the average rust area ratio and maximum rust depth of the steel bars in each part of the bridge. Combining with the rust type and distribution probability model, comprehensively evaluate the rust degree of the whole bridge and local parts to obtain the spatial distribution map of the bridge rust degree. Input the spatial distribution map of the rust degree into the established rust prediction model. The model predicts the development trend of the bridge rust degree within a certain time range (such as 1 year, 5 years) based on environmental parameters (temperature, humidity) and time series data, and obtains a dynamically updated rust degree prediction distribution map.

[0269] Compare and analyze the spatial distribution of the rust degree reflected in the rust degree prediction distribution map and the real-time monitoring data. Using spatial statistical methods such as Kriging interpolation, convert the discrete monitoring point data into grid data with the same resolution as the prediction distribution map, and use the Pearson correlation coefficient to measure the linear correlation strength between the two grid data.

[0270] Using the moving window method, select a window of a certain size (such as 3x3) on the prediction map and the monitoring map, calculate the mean and variance of the difference between the grid values within the two windows, and then divide the difference level according to the size of the mean and variance. Traverse the entire area through the moving window to obtain the spatial distribution map of the local difference index.

[0271] Identify areas with significant differences, severe corrosion types (such as pitting corrosion), and large distribution ranges (such as contiguous distribution) as potential hidden danger areas affected by corrosion; obtain the passing vehicle data of the bridge parts where the hidden danger areas are located, including vehicle weight (axle weight), vehicle speed, and passing frequency, and establish a load action model for the hidden danger areas;

[0272] Adopt a refined finite element analysis method to calculate the stress distribution and deformation characteristics of the hidden danger areas under different corrosion degrees, and establish a structural response model for the hidden danger areas; use the load action model and the structural response model to analyze the correlation mechanism between vehicle loads and corrosion development, and quantitatively evaluate the impact of vehicle loads on corrosion aging;

[0273] By establishing a finite element model that couples corrosion expansion and structural stress, set the corrosion parameters of the hidden danger areas in the model, and apply vehicle loads under different working conditions; by the method of controlling variables, keep other parameters unchanged and only change the values of vehicle weight (or vehicle speed, passing frequency), observe the change rules of response indicators such as the stress level and the range of plastic zones in the corrosion areas, and quantitatively evaluate the relative contribution degree and sensitivity of vehicle weight (or vehicle speed, passing frequency) to corrosion effects using methods such as the response surface method and sensitivity coefficient, and identify the controlling factors that have the most significant impact on corrosion aging, providing a basis for subsequent use, maintenance, and management;

[0274] Based on the comprehensive results of corrosion development prediction, load effect analysis, and structural performance evaluation of the hidden danger areas, dynamically evaluate the corrosion risk level of the hidden danger areas; formulate targeted safety warning and maintenance and reinforcement measures, and dynamically optimize the implementation timing and frequency of the measures according to vehicle load characteristics and traffic flow data to slow down corrosion development to the greatest extent and extend the service life of the bridge; during the implementation process, continuously monitor the corrosion aging process and vehicle load effects, dynamically update the corrosion prediction model and risk assessment model, and continuously improve the maintenance decision-making plan to form a long-term mechanism for bridge corrosion risk assessment and maintenance decision-making based on the vehicle-bridge coupling action mechanism;

[0275] Examples are as follows:

[0276] Taking a prestressed concrete continuous rigid frame bridge on a certain expressway as an example, through on-site investigation and testing, it is found that there are varying degrees of steel bar corrosion problems in parts such as the bridge deck, bridge piers, and pier caps. According to the corrosion appearance characteristics, using the fuzzy comprehensive evaluation method, select 4 indicators of corrosion color, corrosion area, rust layer thickness, and corrosion pit density, and assign weight coefficients of 0.2, 0.3, 0.3, and 0.2 respectively, and divide the corrosion types into mild corrosion (membership degree 0.8), moderate corrosion (membership degree 0.6), and severe corrosion (membership degree 0.9);

[0277] Using the GIS spatial interpolation method, the distribution probability maps of various types of corrosion are generated. It is found that the bridge deck and pier cap are mainly moderately corroded, while the bridge pier is mainly slightly corroded;

[0278] The average cross-sectional loss rate of the corroded steel bars is measured to be 8% by the Archimedes method, and the local maximum corrosion depth is 2 mm. These data are input into the established grey Markov chain prediction model. It is predicted that within the next 5 years, under the current climate conditions, the proportion of the corroded area will increase to 12%, and the maximum corrosion depth will reach 3.5 mm;

[0279] Using the kriging method, the data of 20 corrosion monitoring points of the bridge are interpolated onto a 100 m × 100 m grid and compared with the prediction results. The calculated Marangoni difference index is 0.28, indicating that the spatial distribution difference between the two is small;

[0280] Near the mid-span and supports, the local difference index exceeds 0.5, and the predicted value is significantly higher than the monitored value, which is judged as a corrosion hazard area. An automatic vehicle detection system is installed in the hazard area, and the vehicle weight distribution is obtained as a normal distribution N(30t, 5t), the average vehicle speed is 80 km / h, and the daily traffic volume is 10,000 vehicles;

[0281] In the finite element model, the corrosion parameters of the hazard area are set, and different vehicle weights (20t, 30t, 40t) and vehicle speeds (60 km / h, 80 km / h, 100 km / h) working conditions are applied. The deflection and stress responses of the bridge under normal service conditions and severe corrosion conditions are calculated, and the response surface equation is fitted; through variance analysis, the sensitivity coefficients of vehicle weight and vehicle speed are 0.6 and 0.3 respectively, indicating that vehicle weight is the main control factor affecting corrosion aging;

[0282] In the dynamically updated corrosion risk prediction model, the vehicle weight distribution is used as an important input parameter. When the proportion of the corroded area exceeds 15%, an early warning is issued in a timely manner, and the weight limit standard is adjusted from 30t to 25t. At the same time, combined with the corrosion development trend and traffic characteristics, the best maintenance time window is set within 2 years, and the maintenance plan is optimized. By replacing the corroded steel bars, adding cathodic protection and other measures, the remaining service life of the bridge is extended by more than 10 years.

[0283] Furthermore, in this embodiment, step S8 specifically includes:

[0284] According to the mechanical characteristics of the bridge structure, a suitable finite element analysis software is selected to establish a three-dimensional solid model of the bridge structure and reasonably divide the mesh; the mechanical characteristics of the bridge structure include beam type, arch type, and rigid frame type; the mesh size should not be too large or too small, generally taking 1 / 10 - 1 / 8 of the beam height;

[0285] In the area with potential bridge corrosion hazards, according to the distribution of steel bar corrosion, the area is divided into several sub-areas, and the corrosion degree within each sub-area is approximately uniform. For each sub-area, based on the cross-sectional loss rate of the corroded steel bars, equivalent corroded steel bar material parameters are generated by reducing the cross-sectional area and equivalent yield strength of the steel bars; when selecting the key stress points of the bridge, in addition to considering the load magnitude, the corrosion sensitivity should also be considered; the areas with high corrosion sensitivity mainly include the areas where positive and negative bending moments act alternately, the areas where shear force and torque act intensively, the anchorage areas and the lap joints; in the finite element model, attention should be focused on the above areas, and the calculation accuracy should be improved by means of local mesh refinement, element type refinement, etc.

[0286] Using parametric modeling technology, influence areas are preset in the geometric model or mesh model, and different finite element models are automatically generated by changing the distance and size of the influence areas to simulate the spatial variability of the corroded area and evaluate the influence law of position factors.

[0287] For the bond-slip behavior of corroded reinforced concrete, the modified CEB-FIP Model Code 2010 constitutive relationship is adopted.

[0288] When modifying the bond-slip constitutive relationship, a dimensionless parameter reflecting the corrosion degree is introduced, a quantitative correlation between each parameter and the corrosion parameter is established, and it is calibrated by means of pull-out test; the dimensionless parameters reflecting the corrosion degree include the volume expansion rate and the residual cross-sectional rate.

[0289] Embed the modified corrosion degradation-bond slip constitutive relationship into the finite element model to evaluate the weakening degree of the overall bond force of the bridge under different corrosion degrees; input the meteorological parameters of the area where the bridge is located, such as temperature and humidity, in the finite element model, and adopt the multi-field coupling analysis of environment-material-corrosion degradation to explore the interaction mechanism between environmental conditions and material properties on the corrosion development and bond force degradation, where the influence of temperature on the corrosion reaction kinetics and the physical and mechanical properties of concrete, and the influence of humidity on the ion migration of the corrosion medium and the pore structure of concrete need to be considered.

[0290] When analyzing the influence mechanism of vehicle load on corrosion and bond force, the relationship between the stress amplitude and cycle number of steel bars induced by vehicle dynamic load and the corrosion electrochemistry reaction kinetics needs to be considered, study the evolution law of traffic flow characteristics and corrosion current density, product composition and distribution, and quantitatively evaluate the load-corrosion-bond chain effect.

[0291] Collect the vehicle passing data since the bridge was built, including vehicle type, vehicle weight, vehicle speed, traffic volume, use the rain flow counting method to count the load spectrum, and combine with the finite element analysis results to quantitatively evaluate the influence of vehicle repeated load on the fatigue performance of corroded steel bars and obtain the predicted values of fatigue life and reliability.

[0292] Compare the finite element simulation results with the actual bridge test results, and adopt parameter identification methods such as partial least squares regression and artificial neural networks to correct the material parameters, load parameters, and boundary conditions in the finite element model, etc., so that the predicted results after model calibration are in the best fit with the measured values and improve the prediction accuracy;

[0293] Based on the structural reliability theory of probability theory, through stochastic finite element analysis, calculate the variation law of the structural failure probability under different corrosion deterioration levels; taking the failure probability in the 100-year design reference period as a reference, when the increase in the structural failure probability caused by the degradation of the bond force in the corroded area reaches one order of magnitude, the corresponding bond force degradation rate can be used as the critical threshold;

[0294] Adopt the sensitivity analysis method to further quantify the influence weight of the bond force degradation rate on the failure probability; combined with the risk matrix method, classify and grade the risk scenarios formed by different combinations of bond degradation rates and failure consequence levels, so as to determine the risk levels of different corroded areas and formulate differentiated maintenance strategies;

[0295] The example is as follows:

[0296] Taking a prestressed concrete continuous rigid frame bridge as an example, use the ABAQUS finite element software to establish a full-bridge geometric solid model, and use the C3D8R eight-node linear reduced integration element for mesh division, generating about 500,000 elements and 600,000 nodes in total;

[0297] According to on-site inspections, define 10% of the bridge deck area and 5% of the pier area as corroded hidden danger areas. Based on the inspection results of a 20% reduction in the cross-sectional area of corroded steel bars, use the geometric scaling method to generate an equivalent elastic modulus of 200 GPa and a yield strength of 300 MPa in the corroded area; identify the mid-span, supports, pier bottoms, etc. as corrosion-sensitive areas, encrypt the mesh size to 0.1 m, and introduce the Gaussian point positioning corrosion element method to automatically generate finite element models of 10 different corrosion distribution scenarios; adopt the modified CEB-FIP bond-slip constitutive model, introduce the volume expansion rate (ranging from 0.1 to 0.5) as the corrosion degradation characteristic parameter, and correlate parameters such as slip strength, residual bond stress, and characteristic slip amount with it, and calibrate through 18 groups of pull-out test data;

[0298] Use COMSOL software for multi-field coupling analysis of temperature-humidity-corrosion, consider the influence of temperature gradient on the pore water chemistry of concrete, use the non-Fick diffusion law to describe the migration of chloride ions in the corroded area, and couple it with the adsorption / desorption mechanism driven by the potential difference. At the same time, introduce the vehicle-bridge coupling vibration equation, adopt transient dynamic analysis, simulate the stress-strain response of corroded steel bars under vehicle loads, extract the stress amplitude and the number of cycles, and substitute them into the fatigue life prediction model based on the Paris formula;

[0299] Collect the traffic flow data of the bridge over the past 5 years. After statistics, the proportion of fatigue vehicles is about 30%, the average axle load is 80 kN, and the daily traffic volume is 10,000 vehicle trips;

[0300] Perform partial least squares regression on the finite element analysis results and on-site detection data. Through Leave-one-out cross-validation, determine the correction values of material parameters (such as the elastic modulus of concrete) and load parameters (such as the dynamic load amplification factor). The relative error of the predicted overall stiffness of the calibrated bridge is reduced to within 5%;

[0301] Combined with the principle of reliability analysis, conduct stochastic finite element simulation on the key rust areas. Considering random variables such as the rust current density and concrete strength, use the first-order second-moment method (FORM) to calculate the structural failure probability. When the bond strength in the rust area degrades by 20%, the failure probability increases from 3% to 38%, approaching the threshold (40%) of the 100-year design reference period;

[0302] Using the Sobol sensitivity analysis method, it shows that the importance weight of the bond strength degradation rate is as high as 0.65. Therefore, 20% is determined as the key threshold for bond degradation. Then, divide the risk levels according to the bond degradation rate ≤ 10%, 10% - 20%, ≥ 20%, and formulate corresponding maintenance strategies.

[0303] Furthermore, in this embodiment, step S8 further includes:

[0304] For the hazard areas exceeding the bond force threshold, use the finite element analysis method to conduct stress analysis and hazard assessment, determine its impact on the overall safety of the bridge structure, calculate the maintenance and repair scope, and formulate corresponding maintenance and repair strategies;

[0305] Specifically, according to the bridge design drawings and on-site detection data, obtain the specific location, area, and shape of the hazard areas exceeding the bond force threshold;

[0306] Based on these data, determine the relative position of the hazard areas in the entire bridge structure and their potential impact on the overall structure; adopt testing methods in materials science, such as tensile testing and compression testing, to obtain the material properties of the hazard areas, including but not limited to the elastic modulus and Poisson's ratio;

[0307] Based on these material properties, judge the mechanical behavior and possible failure modes of the materials. Through the structural stress analysis tool and load data, determine the stress analysis state where the hazard areas are located, including the magnitude and distribution of the loads;

[0308] After obtaining this data, the structural safety of the hazardous area under actual operating conditions can be evaluated. The connection mode and boundary conditions between the hazardous area and the surrounding structure are obtained, and the stress distribution and potential weaknesses of the entire structure are determined by analyzing the stability of the connection and the constraints of the boundary conditions; and the material properties and geometric dimensions of the overall structure of the bridge are obtained by reviewing the overall design parameters of the bridge. This information is crucial for evaluating the mechanical properties and durability of the entire structure;

[0309] Use appropriate finite element software tools to select appropriate finite element meshing methods and unit types according to the complexity of the hazardous area, the accuracy requirements of the analysis, and the force analysis; after obtaining the mesh model, more accurate numerical simulation and structural analysis can be performed; by setting the convergence accuracy and the number of iterations, the numerical solution of the finite element analysis can be performed; after obtaining the simulation results, the specific impact of the hazardous area on the overall safety of the bridge structure can be determined, and the scope and methods of maintenance and repair that may be required can be evaluated;

[0310] The following is an example:

[0311] Through the analysis of the bridge design drawings and on-site inspection data, it was determined that the hazardous area is located at the bottom of the third T-beam of the second span of the main span, with an area of ​​about 5 square meters and an irregular ellipse; this area is located at 1 / 4 of the entire T-beam length, which has a certain impact on the stress of the overall structure. Tensile and compression tests were conducted on the materials in the hazardous area, and the elastic modulus of the material was 30GPa and the Poisson's ratio was 2, which shows that the material belongs to the category of normal concrete and is not prone to brittle fracture under normal loads; and through structural stress analysis, it was determined that the maximum compressive stress in the hazardous area under the combined action of dead load and live load is 10MPa, and the maximum tensile stress is 2MPa, both of which do not exceed the material strength limit. However, long-term load effects may cause crack expansion;

[0312] The hazard area is connected to the adjacent T-beam through continuous longitudinal reinforcement, and the end of the T-beam is consolidated with the pier cap, so the overall restraint is good. However, if the cracks expand further, the effect of these connections may be weakened. The overall design parameters of the bridge are reviewed. The T-beam of the bridge adopts C50 concrete, with a design elastic modulus of 35GPa, a mid-span T-beam height of 8 meters, a web thickness of 2 meters, a flange width of 2 meters, and a thickness of 25 meters.

[0313] The hexahedral unit was used to mesh the hazardous area and adjacent structures, and the mesh was locally encrypted, with a total number of about 500,000 units. Through the solver settings, numerical simulations of stress, displacement and cracks were performed. The finite element analysis results show that the stress level in the hazardous area is controllable under the current state, but the stress concentration factor at the crack tip is high, and it is necessary to adopt surface sealing and crack embedding methods for repair, and continuously monitor the crack expansion to ensure safe use;

[0314] ​For non-hazardous areas where the bonding force does not exceed the threshold, based on the structural characteristics of the bridge and the monitoring data, combined with cost-benefit analysis, evaluate the long-term benefits after replacement or repair, and formulate maintenance and repair strategies;

[0315] Specifically, through Geographic Information System (GIS) and on-site investigation, obtain the specific location and boundary data of the non-hazardous areas where the bonding force does not exceed the threshold, and obtain accurate regional geographical location information;

[0316] According to the obtained regional geographical location information, use area measurement tools such as GIS or CAD software to calculate the total area of the non-hazardous area, obtain the accurate area size of the area, use a bonding force test instrument, such as a tensile testing machine, to obtain the specific value of the bonding force within the area, determine the specific gap between it and the safety threshold, and obtain the bonding force value and its comparison result with the threshold; obtain the environmental monitoring data of the area, including the readings of humidity and temperature sensors, and by analyzing these data, judge the influence of surrounding environmental factors (such as humidity, temperature) on the bonding force of the area, and obtain the influence evaluation result of environmental factors;

[0317] According to the design documents and structural analysis reports of the bridge, evaluate the role and influence degree of the area in the overall structure of the bridge, obtain the impact evaluation of the area on the bridge safety, estimate the material cost and labor cost, obtain the cost data for repairing or replacing the area, calculate the comprehensive cost, and obtain the total cost estimate required for repairing the area;

[0318] According to the bonding force test results before and after repair, evaluate the effect of the repair measures, such as whether the bonding force reaches the expected enhanced value, obtain the repair effect and its expected improvement in service life, analyze the repair effect and the expected remaining service life of the bridge, calculate the long-term economic benefits after repair, and obtain the long-term benefit analysis result;

[0319] Adopt maintenance history records and expert consultations to formulate a daily maintenance plan suitable for the area, including the frequency and measures of maintenance, and obtain the maintenance plan and the effect of expected extended service life; comprehensively consider the maintenance plan, cost, repair effect and long-term benefits, and through optimization models and decision analysis tools, determine the best repair timing, and formulate a detailed repair plan and implementation plan to obtain the final repair decision and implementation plan;

[0320] Examples are as follows:

[0321] Through GIS and on-site investigation, a non-hazardous area with a length of 100 meters and a width of 50 meters was determined. Its specific boundary coordinates are (12234E, 3123N), (12236E, 3123N), (12236E, 3121N), and (12234E, 3121N), forming a rectangular area. Using the GIS area measurement tool, the area of this rectangular area was calculated as 100 meters × 50 meters = 5000 square meters. The average bond strength within this area was measured to be 2 MPa using a tensile testing machine, and the safety threshold was 5 MPa. The bond strength value was 3 MPa lower than the threshold. By analyzing the environmental monitoring data, it was found that the average humidity around this area was 60% and the average temperature was 20°C. It was evaluated that the humidity and temperature were within the normal range and had little impact on the bond strength. According to the design document and structural analysis, it was evaluated that this area is located at the edge of the bridge, has a relatively small impact on the overall structure, and the impact level on the bridge safety is low risk.

[0322] It was estimated that the material cost for repairing this area would be 200,000 yuan, the labor cost would be 100,000 yuan, and the comprehensive cost would be 300,000 yuan. The bond strength before repair was 2 MPa, and after repair, the measured bond strength increased to 6 MPa, reaching the expected enhanced value of 5 MPa. It was expected to extend the service life of this area by 10 years. Through analysis, it was obtained that the expected remaining service life of this bridge was 30 years, and after repair, it could be extended by 10 years. Estimated based on an annual toll income of 1 million yuan, the long-term economic benefit would be 10 million yuan. A daily maintenance plan was formulated to conduct inspections once every quarter, repair problems in a timely manner, and conduct a comprehensive inspection once a year. It was expected to extend the service life by 5 years. Through optimized model calculation, considering the maintenance cost, repair effect, and long-term benefit comprehensively, the best repair time was determined to be 2 years later, and a detailed phased implementation plan was formulated. The total construction period was expected to be 3 months.

[0323] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by orientation words such as "front, back, up, down, left, right", "horizontal, vertical, level" and "top, bottom", etc. is usually based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description. Without contrary explanation, these orientation words do not indicate and imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the protection scope of the present invention.

[0324] For those skilled in the art, various corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all such changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A method for identifying hidden damage of a bridge structure, characterized in that: The following steps are involved: S1. Monitor and collect corrosion image data, structural deformation data, dynamic response data and environmental condition data of bridge structures; S2, analyzing the collected steel bar surface corrosion images to identify the type, distribution characteristics and corresponding current corrosion degree of the corrosion; S3. Construct a corrosion prediction model based on the corrosion degree under multiple time series; S4. Conducting simulation tests on the bonding strength between steel bars and concrete with different preset corrosion degrees, obtaining multiple sets of correlation data between the preset corrosion degrees and the bonding strength, and constructing a correlation analysis model between the steel bar surface corrosion degree and the bonding strength; S5, comparing and analyzing the corrosion prediction model with current monitoring data of different bridge structures, evaluating the accuracy of the corrosion prediction model under current environmental conditions, and adjusting the corrosion prediction model; S6. Compare and analyze the correlation analysis model with current monitoring data of different bridge structures, evaluate the accuracy of the correlation analysis model under the current corrosion degree, and adjust the corrosion prediction model; S7. Based on the corrosion type, distribution characteristics and degree of corrosion, as well as the comparative analysis results of the corrosion prediction model and the current monitoring data, identify the potential risk areas affected by corrosion, and analyze the impact of passing vehicles on these potential risk areas; S8. Based on the stress characteristics of the bridge structure, evaluate the impact of corrosion on adhesion in the potential risk area, determine the degree of weakening of adhesion due to different degrees of corrosion based on the distance from the area to the stress point, material properties and environmental conditions, analyze historical data and the characteristics of passing vehicles, set a threshold for the impact of adhesion, and classify the affected areas based on this threshold; The step S3 specifically includes: Obtain historical inspection data of steel structure, extract the corrosion area and maximum corrosion depth at different time nodes, and construct a time series data set of corrosion degree indexed by time; Perform a stationarity test on the time series of the degree of corrosion. If it is not stationary, perform differential processing until it passes the stationarity test. Adopting the STL decomposition method and based on LOESS fitting, the time series is decomposed into trend term, seasonal term and random term. According to the decomposition results, linear regression or polynomial regression modeling is adopted for trend term, Fourier series or periodic function modeling is adopted for seasonal term, and ARMA modeling is adopted for random term. Select the time series prediction model, and determine the hyperparameters of the model based on the stationarity, autocorrelation and trend characteristics of the corrosion degree time series, combined with the ACF and PACF graphs; The goodness of fit and predictive performance of different models were evaluated by cross-validation and information criteria, and the optimal model was selected; Environmental factors are introduced as covariates, which are integrated with the corrosion degree time series to build a corrosion prediction model for multivariate time series. The step S6 specifically includes: Extract real-time bond monitoring data, including the ultimate bond strength between steel bars and concrete obtained by the pull-out method and the steel bar strain distribution obtained by the steel bar strain method, and invert the bond stress distribution between steel bars and concrete based on the strain distribution; For each monitoring section, at least three measuring points are arranged, and at least three specimens are tested in parallel at each measuring point, and the average value is obtained as the representative value of the section; The monitoring frequency is determined based on the importance and deterioration rate of the bridge, and the monitoring data are normalized to form a bond data set across bridges and structure types; The current bonding force monitoring data is input into the established correlation analysis model between the steel bar surface corrosion degree and bonding force. The model predicts the corresponding theoretical bonding force value according to the current corrosion area ratio and corrosion depth parameters as the bonding force prediction value under the current corrosion degree. For each bridge structure, the predicted bond strength value is compared with the measured value, and the difference between the two is calculated as the prediction deviation. The magnitude of the prediction deviation is quantified using absolute value and relative percentage indicators. Evaluate the accuracy of the correlation analysis model under the current corrosion level and determine whether the prediction accuracy of the model meets the actual application requirements.

2. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S1 specifically includes: Use image sensors to obtain high-definition image data of the bridge structure surface, segment and extract features from the image, automatically identify and locate the damaged area, and calculate the corrosion area ratio and crack length as the evaluation basis for corrosion monitoring; Use strain sensors and accelerometers to collect the strain and vibration acceleration of the bridge structure in real time. According to the change trend and amplitude of the strain and acceleration data, combined with the preset judgment threshold, determine whether the bridge structure has deformed or vibrated abnormally, and obtain monitoring data on the deformation and dynamic response of the bridge structure; Temperature sensors, humidity sensors and corrosion sensors are used to monitor the temperature, humidity and corrosive gas concentration of the bridge environment in real time. According to the correlation between environmental parameters and the characteristics of bridge structure materials, a quantitative evaluation model is established to determine the impact of current environmental conditions on the durability of the bridge structure and obtain environmental condition monitoring results.

3. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S2 specifically includes: Based on the high-definition image data collected from the steel bar surface, image preprocessing technology is used to denoise, enhance and standardize the image to improve the image quality and the recognizability of the steel bar surface corrosion features; A semantic segmentation algorithm based on deep learning is used to perform pixel-level segmentation on the rust image, accurately extract the location and boundary information of the rust area, and obtain a binary mask map of the rust area; According to the mask map of the corrosion area, the ratio of the corrosion area to the total surface area of ​​the steel bar is calculated as an important indicator to measure the degree of corrosion; By extracting the grayscale histogram features of the rusted area, the grayscale distribution of the rusted area is obtained; By calculating the local binary pattern features of the rust area, the microstructure and regularity of the rust texture are captured, and these feature parameters are combined into a multidimensional feature vector as the input for subsequent rust type discrimination and degree assessment.

4. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S4 specifically includes: According to the preset corrosion degree level, the steel bar specimens are artificially corroded by using the electrochemical accelerated corrosion method; By controlling the corrosion current density and corrosion time, the degree of corrosion can be regulated; During the corrosion process, the mass loss rate and volume expansion rate of the steel bars are regularly monitored to ensure that the preset corrosion degree is reached; Simulate the uneven corrosion in the actual environment, prepare steel bar specimens with different corrosion degrees, measure the corrosion area ratio and corrosion depth on the surface, and quantitatively characterize the corrosion degree; The steel bar specimens after corrosion treatment were cast with concrete into standard specimens, and the concrete mix ratio and curing conditions similar to the actual structure were adopted to ensure the representativeness of the specimens; A control group of uncorroded steel bars and concrete specimens was set up to evaluate the effect of corrosion on the bond strength; The pull-out test was carried out on reinforced concrete specimens with different corrosion degrees. The load-displacement curve and steel bar strain distribution during the pull-out process were continuously recorded by displacement sensors and strain gauge testing equipment to obtain the mechanical parameters reflecting the bonding force. According to the mechanical parameters measured by the pull-out test, the bond force distribution of the reinforced concrete interface under different corrosion degrees is calculated; Combined with the volume expansion effect, loose porous characteristics and cross-sectional loss of the corrosion products, the influence of the corrosion products on the bond transfer mechanism is analyzed to identify the key factors that lead to bond degradation due to corrosion. These characteristics of the corrosion products were quantitatively evaluated through cross-sectional analysis and microscopic morphology characterization of the drawn specimens; Statistical analysis was performed on the bond strength and interfacial stress distribution under different corrosion levels to extract quantitative indicators of bond strength; The correlation analysis model between the degree of steel bar surface corrosion and bonding strength is established using multiple linear regression or nonlinear regression methods.

5. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S5 specifically includes: Extract real-time monitoring data of different bridges, record the area type, traffic flow and load type information of the bridge, and build a bridge corrosion monitoring database; The current corrosion state parameters of each bridge in the monitoring database are input into the constructed corrosion prediction model to obtain the predicted value of the corrosion degree of each bridge at the current moment, and compared with the corrosion degree estimated by electrochemical parameters to calculate the prediction error and relative error; The prediction errors of each bridge were statistically analyzed, the mean and standard deviation of the prediction errors were calculated, and the distribution histogram and box plot of the prediction errors were drawn to intuitively evaluate the overall prediction accuracy and stability of the corrosion prediction model on different bridges.

6. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S7 specifically includes: According to the corrosion type identification results, the corrosion of bridge reinforcement is divided into three types: uniform corrosion, local corrosion and pitting corrosion. Combining the morphological characteristics, spatial distribution characteristics and causal mechanism of rust, a set of rust type classification standards with multi-index comprehensive evaluation is formulated. The weight of each factor is determined by the hierarchical analysis method. The quantitative index of the rust type is obtained by weighted summation, and the rust type is then divided according to the size of the index value. Combined with the analysis of corrosion distribution characteristics, the distribution probability and spatial correlation of various types of corrosion in different parts of the bridge are determined to form a probability model of bridge corrosion type and distribution; The corrosion degree assessment method was used to calculate the average corrosion area ratio and maximum corrosion depth of the steel bars in various parts of the bridge. Combined with the corrosion type and distribution probability model, a comprehensive assessment of the overall and local corrosion degree of the bridge was conducted to obtain the spatial distribution map of the bridge corrosion degree.

7. The bridge structure hidden damage identification method according to claim 1 is characterized in that: The step S8 specifically includes: According to the stress characteristics of the bridge structure, select the appropriate finite element analysis software, establish a three-dimensional solid model of the bridge structure, and divide the grid reasonably; In the bridge corrosion risk area, the area is divided into several sub-areas according to the distribution of steel bar corrosion. The degree of corrosion in each sub-area is approximately uniform. For each sub-area, the equivalent corroded steel bar material parameters are generated by reducing the cross-sectional area and equivalent yield strength of the steel bar according to the cross-sectional loss rate of the corroded steel bar. Using parametric modeling technology, the influence area is preset in the geometric model or mesh model. By changing the distance and size of the influence area, different finite element models are automatically generated to simulate the spatial variability of the corrosion area and evaluate the influence of location factors. The modified CEB-FIPModelCode2010 constitutive relationship is used to study the bond-slip behavior of corroded steel bars and concrete; When modifying the bond-slip constitutive model, dimensionless parameters reflecting the degree of corrosion are introduced, and quantitative correlations between each parameter and the corrosion parameter are established, and calibration is performed through pull-out test. The modified corrosion degradation-bond-slip constitutive model was embedded in the finite element model to evaluate the degree of weakening of the overall bond strength of the bridge due to different corrosion degrees.

8. The bridge structure hidden damage identification method according to claim 7 is characterized in that: The step S8 further comprises: For the hazardous areas that exceed the bond strength threshold, the finite element analysis method is used to conduct stress analysis and hazard assessment, determine its impact on the overall safety of the bridge structure, calculate the maintenance and repair scope, and formulate corresponding maintenance and repair strategies; For non-hazardous areas that do not exceed the bond strength threshold, the long-term benefits after replacement or repair are evaluated and maintenance and repair strategies are formulated based on the structural characteristics of the bridge and monitoring data, combined with cost-benefit analysis.

Images