Method for monitoring and predicting carbonation depth of concrete under dry-wet cycle conditions in brackish water area

Abstract

The present application relates to the technical field of concrete monitoring, in particular to a method for monitoring and predicting carbonation depth of concrete under dry-wet cycle conditions in salt-fresh water areas. The method comprises the following steps: by laying multiple source carbonation monitoring nodes, collecting carbonation related characteristic data, and using the maximum information coefficient algorithm to screen the dominant factor; using fuzzy C-means dynamic clustering algorithm to identify carbonation stage and spatial partition; constructing partition discriminant model through multiple discriminant analysis, and combining extreme learning machine to realize carbonation depth prediction and risk assessment, and finally generating carbonation risk early warning report. The present application realizes the whole process integration of multi-source data driving, feature screening, dynamic partition clustering, partition discrimination and intelligent prediction; improves the monitoring accuracy, risk identification rate and management response efficiency of concrete members, and can provide scientific and effective technical support for concrete durability evaluation and operation management in special environments such as ocean and estuary.

Description

Technical Field

[0001] This invention relates to the field of concrete monitoring technology, and more particularly to a method for monitoring and predicting the carbonation depth of concrete under dry-wet cycle conditions in brackish water areas. Background Technology

[0002] Concrete carbonation refers to the process by which alkaline components such as calcium hydroxide inside concrete react chemically with atmospheric carbon dioxide to form calcium carbonate, leading to a decrease in the alkalinity of the pore solution in the concrete. Carbonation not only affects the durability of concrete structures but also accelerates the corrosion of reinforcing steel, thus threatening the overall safety of the concrete structure. Especially in brackish water areas, where concrete components are constantly subjected to the combined effects of alternating infiltration of seawater and freshwater and wet-dry cycles, the carbonation process is significantly affected by environmental changes, exhibiting a more complex and nonlinear evolutionary pattern.

[0003] Existing technologies have conducted extensive research on methods for monitoring and predicting concrete carbonation under general environments, mainly including physical sampling methods, chemical titration methods, and surface spraying with phenolphthalein indicator methods. However, these methods are mostly based on static laboratory conditions or single environmental parameters, and cannot truly reflect the dynamic evolution characteristics of concrete carbonation under wet-dry cycles in brackish water areas. Currently, the following problems remain: Existing monitoring methods often lack detailed analysis of carbonation differences in different spatial zones and service stages, and cannot dynamically reflect the evolution characteristics of carbonation in different regions and stages, resulting in insufficient targeting for risk identification and management; existing carbonation depth prediction methods mostly employ traditional statistical or shallow machine learning methods, which are difficult to fully utilize complex multi-source features for high-precision modeling, and also lack automated and intelligent risk warning capabilities, failing to meet the actual needs of durability management and proactive maintenance decision-making. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method for monitoring and predicting the carbonation depth of concrete under dry-wet cycle conditions in brackish water areas. This method effectively acquires multi-source carbonation characteristic data of concrete in complex dry-wet cycle environments at the brackish water interface, accurately identifies differences in carbonation evolution across different spatial zones and service stages, and overcomes the limitations of existing monitoring methods in capturing dynamic processes, identifying spatially coarsely, and predicting with limited accuracy. This improves the monitoring accuracy, risk identification rate, and management response efficiency of concrete components.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas includes the following steps: S1: Carbonation monitoring nodes are set from the surface layer to the internal structure of concrete components to periodically collect multi-source carbonation characteristic data; S2: Based on the multi-source carbonization feature data, the correlation between the multi-source carbonization feature data and the carbonization depth is analyzed using the maximum information coefficient algorithm to screen out the dominant carbonization feature factors. S3: Based on the carbonization dominant characteristic factors, a fuzzy C-means dynamic clustering algorithm is used to perform cluster analysis on carbonization characteristic data of different periods and different monitoring points, dynamically identify the carbonization development stage and spatial partition, and generate dynamic carbonization clustering labels. S4: Based on the carbonization dynamic clustering labels, a multi-discriminant analysis algorithm is used to model the dominant factors of each partition, establish a discriminant function for the change of carbonization depth within the partition, and generate partition carbonization discriminant values; S5: Based on the carbonation discrimination value and dominant carbonation characteristic factor of the partition, an extreme value learning machine regression algorithm is used to construct a carbonation depth prediction model, which intelligently predicts and assesses the carbonation depth of each concrete partition under dry-wet cycle conditions in brackish water areas, and generates a carbonation risk warning report.

[0006] Furthermore, the multi-source carbonation characteristic data includes the carbonation depth of concrete components, resistivity change rate, carbonation interface migration speed, surface color grayscale gradient, carbonation product accumulation index, porosity change, and alkalinity decay trajectory.

[0007] Furthermore, step S2 includes the following steps: The multi-source carbonization characteristic data are used to construct a feature sequence matrix according to the monitoring nodes and time order. The feature sequence matrix uses the data group of each monitoring period as the row vector and the feature type as the column vector. Based on the feature sequence matrix, the maximum information coefficient algorithm is used to measure the degree of nonlinear correlation between each type of carbonization feature data and carbonization depth, forming a feature-depth coupling correlation map. By combining the temporal evolution trend and spatial variability of the feature dimensions, multi-level principal factor identification is performed on the coupling spectrum, and carbonization-dominant feature factors that exhibit a strong spatiotemporal coupling mode with carbonization depth are screened out, including the resistivity change rate-grayscale gradient joint trajectory, porosity dynamic change rate and alkalinity decay coupling term.

[0008] Furthermore, the formula for the maximum information coefficient algorithm is as follows:

[0009] in, This represents the maximum information coefficient value; The f-th dominant carbonation characteristic factor is represented; D represents the carbonation depth of concrete. Representation of features The joint probability of the bin with depth D in the i-th row and j-th column; Representation of features Marginal probabilities within the i-th bin; The marginal probability of carbonization depth D within the j-th bin is represented by k; k represents the feature value. The number of bins; l represents the number of bins for carbonization depth D; n represents the total number of samples; This represents the binning penalty factor.

[0010] Furthermore, step S3 includes the following steps: Based on the dominant carbonization characteristic factors, a multi-temporal-multi-monitoring-point three-dimensional data structure is constructed, wherein each dimension represents the sampling time, monitoring spatial location and characteristic factor value, forming a three-dimensional carbonization characteristic matrix that reflects the carbonization evolution process. The local coefficient of variation method was used to perform pre-clustering sensitivity analysis on the three-dimensional carbonization feature matrix to extract the sensitive regions of dry-wet cycle response, and the carbonization susceptibility field was constructed using the water-salt flux index as the weight correction input for the clustering algorithm. Based on the carbonization feature matrix after weight correction, the fuzzy C-means dynamic clustering algorithm is used to iteratively cluster the carbonization development state under the influence of dry and wet cycles, and construct a set of carbonization state labels including time stationarity, spatial diffusivity and factor coupling degree. After each round of dynamic clustering, based on the set of carbonization state labels, according to the rate of change of the spatial distribution centroid of the cluster centers and the rate of change of the factor gradient, a carbonization development stage transition index is defined, and the clustering results are labeled as carbonization dynamic clustering labels including stage identifiers and partition identifiers.

[0011] Furthermore, the formula for the fuzzy C-means dynamic clustering algorithm is as follows:

[0012] in, denoted by the objective function value of temporal carbonization clustering; p represents the number of effective samples in the multi-temporal-multi-monitoring-point three-dimensional carbonization feature matrix; Q represents different carbonization development stages or spatial partitions. This indicates the probability of a current carbonization monitoring point being assigned to a certain stage / zone; m represents the fuzzy index. Represents the carbonization-dominant feature vector of the p-th sample; Let represent the cluster center feature vector of the q-th cluster; This represents the q-th cluster center formed in the previous round of dynamic clustering; This indicates that the weighting coefficient is dynamically adjusted.

[0013] Furthermore, step S4 includes the following steps: Based on the carbonization dynamic clustering label, carbonization feature data from different periods and spatial locations are divided according to stage identifiers and partition identifiers to construct partition feature subsets, and the corresponding dominant carbonization feature factors in each subset are extracted as input variables. For each feature subset of the partition, a set of linear discriminant functions reflecting the level of carbonization depth variation is constructed using a multi-discriminant analysis algorithm; Based on the set of linear discriminant functions, the carbonization level of the characteristic data of each carbonization monitoring node is discriminated, and the carbonization evolution level label is output. Combined with the time series evolution trend of the monitoring node's partition, it is used as the partition carbonization discriminant value.

[0014] Furthermore, step S5 includes the following steps: Based on the partition carbonization discrimination value and its corresponding dominant carbonization feature factor, combined with the spatial location and stage label of the monitoring node, a structured training sample set is constructed. Each sample contains spatial partition, stage identifier, dominant feature combination value and carbonization level label. Based on the constructed sample set, an extreme value learning machine regression algorithm is used to establish a carbonization depth prediction model. The carbonization depth prediction model takes the combination of dominant features as input and outputs the carbonization level or depth value. The carbonization characteristic data of the monitoring nodes collected in real time are input into the carbonization depth prediction model, and the predicted carbonization depth value of each monitoring node is output. Based on the historical evolution trend and structural design, carbonization tolerance is set and multi-level carbonization risk thresholds are set. By comparing the prediction results with the carbonization threshold, and combining the stage and spatial partition of the node, the carbonization risk level is determined. Furthermore, by integrating spatial distribution characteristics, stage evolution trends, and discrimination label weights, a carbonization risk early warning matrix based on time-space joint mapping is formed. Based on the aforementioned risk warning matrix, a carbonization risk warning report is automatically generated.

[0015] Furthermore, the formula for the carbonization depth prediction model is as follows:

[0016] in, This represents the predicted carbonization depth of the s-th partition; This represents the standardized value of the h-th dominant carbonization characteristic factor within the s-th partition; This represents the spatial heterogeneity correction term for the s-th partition; This represents the input weight from the h-th carbonization dominant feature factor to the r-th hidden layer node; represents the output weight coefficient of the r-th hidden layer node; R represents the number of hidden layer nodes; This represents the bias term of the r-th hidden layer node; Represents the set of dominant carbonization characteristic factors; This represents the activation function.

[0017] Furthermore, the carbonization risk warning report includes high-risk area identification, phased carbonization development trends, regional risk level assessment, and carbonization intervention recommendations.

[0018] The beneficial effects of this invention are as follows: This invention achieves multi-dimensional, full-cycle dynamic monitoring of the concrete carbonation process by periodically collecting multi-source carbonation characteristic data through the deployment of carbonation monitoring nodes from the surface to the internal structure of concrete components. This improves the comprehensiveness and timeliness of carbonation monitoring. Utilizing the maximum information coefficient algorithm, dominant characteristic factors highly correlated with carbonation depth are scientifically selected from the multi-source carbonation characteristic data, effectively avoiding redundant data interference and ensuring the accuracy and efficiency of subsequent modeling. Through the fuzzy C-means dynamic clustering algorithm, intelligent clustering of carbonation characteristic data from different monitoring points and different periods can be performed, dynamically identifying the carbonation development stage and spatial partitions, achieving fine-grained partitioning and labeling management of the carbonation process. Employing a multi-discriminant analysis algorithm to model the dominant factors of each partition enables the establishment of specific discriminant functions for carbonation depth changes in different spatial partitions, improving the accuracy and targeting of partitioned carbonation process identification. Based on the extreme value learning machine regression algorithm, combined with the regional carbonation discrimination value and dominant feature factor, a regional carbonation depth prediction model is constructed. This model can achieve high-precision intelligent prediction and risk assessment of the carbonation depth of concrete in different regions, and output timely carbonation risk warning reports, providing a scientific basis for durability management and maintenance decisions of concrete structures in brackish water areas. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the method for monitoring and predicting the carbonation depth of concrete under dry-wet cycle conditions in brackish water areas according to the present invention.

[0020] Figure 2 This is a flowchart illustrating the specific steps of step S3 provided in an embodiment of the present invention.

[0021] Figure 3 This is a flowchart illustrating the specific steps of step S5 provided in an embodiment of the present invention. Detailed Implementation

[0022] Please see Figure 1-3 As shown, this invention relates to a method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas.

[0023] Example

[0024] A method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas includes the following steps: S1: Carbonation monitoring nodes are set from the surface layer to the internal structure of the concrete component to periodically collect multi-source carbonation characteristic data; the multi-source carbonation characteristic data includes the carbonation depth of the concrete component, resistivity change rate, carbonation interface migration speed, surface color grayscale gradient, carbonation product accumulation index, porosity change and alkalinity decay trajectory.

[0025] Specifically, several concrete beam and slab components were selected as monitoring objects. Based on the component's geometric dimensions and structural layers, multiple carbonation monitoring nodes were pre-embedded at different depths, including the surface layer, middle layer, and near the reinforcement protective layer.

[0026] The data collection process is as follows: Carbonization depth: Miniature pH sensors are pre-embedded at each node, and pH values ​​are measured periodically using an automated multi-channel data acquisition instrument, combined with phenolphthalein spraying to assist in verifying the carbonization front.

[0027] Resistivity change rate: Four-electrode resistivity sensors are pre-embedded at each node. Using the AC impedance method, the resistivity is automatically collected monthly and the change rate is compared with the previous period.

[0028] Carbonization interface migration speed: Bidirectional miniature temperature / humidity sensors and digital cameras or fiber optic sensing elements are embedded in the surface, shallow and middle layers. Combined with a high-resolution imaging system, images are acquired periodically, and the migration distance and speed of the carbonization interface are automatically calculated through AI boundary recognition algorithms.

[0029] Surface color grayscale gradient: The surface image is periodically captured using a high-resolution imaging device, and the color grayscale distribution is extracted using image analysis algorithms to analyze its gradient changes.

[0030] Carbonation product accumulation index: Surface material was periodically sampled, and the amount of carbonate accumulation was quantitatively analyzed using thermogravimetric analysis (TG) and X-ray diffraction (XRD) techniques.

[0031] Porosity variation: Nuclear magnetic resonance (NMR) and X-ray CT were used to perform non-destructive testing on the micro-porosity of concrete at different period nodes.

[0032] Alkalinity decay trajectory: By collecting pore liquid and periodically analyzing the pH changes at each monitoring node, the alkalinity decay curve over time is obtained.

[0033] All data is collected synchronously using a unified encoding and timestamp, and uploaded to the data center to form a multi-source carbonization characteristic database.

[0034] S2: Based on the multi-source carbonization feature data, the correlation between the multi-source carbonization feature data and the carbonization depth is analyzed using the maximum information coefficient algorithm to screen out the dominant carbonization feature factors. Step S2 includes the following steps: The multi-source carbonization characteristic data are used to construct a feature sequence matrix according to the monitoring nodes and time order. The feature sequence matrix uses the data group of each monitoring period as the row vector and the feature type as the column vector. Based on the feature sequence matrix, the maximum information coefficient algorithm is used to measure the degree of nonlinear correlation between each type of carbonization feature data and carbonization depth, forming a feature-depth coupling correlation map. Specifically, in a data analysis platform (such as a Python data mining environment integrating tools like minepy, pandas, and numpy), the standardized feature sequence matrix is ​​input one by one. For each type of feature and carbonization depth, the maximum information coefficient (MIC) value is calculated. The process is as follows: select a feature (e.g., resistivity change rate), and input the feature and carbonization depth as a two-dimensional vector into the MIC analysis function. After obtaining the MIC value, it is automatically stored in the feature correlation index table. This process iterates through each feature until the MIC values ​​for all features and carbonization depths have been calculated.

[0035] Data visualization tools (such as Matplotlib, Seaborn, Plotly, etc.) are used to automatically generate a "feature-carbonization depth correlation heatmap." The rows and columns of the heatmap represent different features and carbonization depths, respectively, and the color intensity indicates the MIC value. A "correlation network diagram" is also generated, with nodes representing different features and edge thickness or color representing the strength of the correlation. The correlation graph is visualized on the web or mobile version of the engineering monitoring platform. Engineers can click on specific features through an interactive interface to view the details of their correlation changes at different nodes and periods.

[0036] Furthermore, the correlation is overlaid and displayed according to its temporal and spatial distribution. The horizontal axis represents the monitoring period, and the vertical axis represents the node position. Color or height represents the correlation value. Spatiotemporal thermal animations can be generated to dynamically show the evolution trajectory of the correlation between a certain feature and carbonization depth in different periods and spatial locations.

[0037] By combining the temporal evolution trend and spatial variability of the feature dimensions, multi-level principal factor identification is performed on the coupling spectrum, and carbonization-dominant feature factors that exhibit a strong spatiotemporal coupling mode with carbonization depth are screened out, including the resistivity change rate-grayscale gradient joint trajectory, porosity dynamic change rate and alkalinity decay coupling term.

[0038] Specifically, features exhibiting high MIC values ​​in heatmaps and network diagrams are statistically analyzed, and the consistency and prominence of these features across different time periods (periodic sequences) and spatial distributions (monitoring node distributions) are compared. Principal component analysis (PCA) or hierarchical clustering methods are introduced to further explore the coupling between features and the dominant spatiotemporal trends. For example, if a feature is found to show high correlation in most periods / most nodes, it is initially identified as a candidate dominant factor.

[0039] The joint evolution between features is explored in depth, and feature combinations exhibiting synergistic changes are selected. For example, the joint trajectory of "resistivity change rate" and "surface gray-scale gradient" is interactively analyzed to observe the degree to which they change synchronously with carbonization depth. In some regions (such as the water-facing side of bridges), synergistic fluctuations in "porosity dynamic change rate" and "alkalinity decay" may also be found, which are presented through dual-feature coupling trend plots.

[0040] It not only considers the high correlation of a single feature, but also comprehensively evaluates: its stability in the time dimension (whether multiple periods are consistently highly correlated); its breadth in the spatial dimension (whether multiple monitoring points all perform well); and its degree of synergy with other dominant factors. Multiple rounds of expert review meetings are organized to conduct engineering interpretation of the algorithm's initial screening results. Combined with experience in concrete carbonation mechanisms, noisy features are screened out, and only dominant factors with strong mechanism explanatory power and outstanding predictive effects are retained.

[0041] Ultimately, a list of dominant characteristic factors is formed. For example, the "resistivity change rate-grayscale gradient joint trajectory" is determined to be the dominant carbonization factor in region A; the "porosity dynamic change rate and alkalinity decay coupling term" becomes the dominant factor in region B. Each dominant factor is archived and recorded according to region, monitoring point and period, and factor files are automatically generated as the core input for subsequent cluster analysis, discriminant modeling and prediction evaluation.

[0042] Furthermore, the formula for the maximum information coefficient algorithm is as follows:

[0043] in, This represents the maximum information coefficient value, used to quantify the degree of nonlinear coupling, and is the basis for selecting the dominant factor; The f-th dominant carbonation characteristic factor is represented; D represents the carbonation depth of concrete. Representation of features The joint probability of the bin with depth D in the i-th row and j-th column; Representation of features Marginal probabilities within the i-th bin; The marginal probability of carbonization depth D within the j-th bin is represented by k; k represents the feature value. The number of bins; l represents the number of bins for carbonization depth D; n represents the total number of samples; This represents the binning penalty factor.

[0044] S3: Based on the carbonization dominant characteristic factors, a fuzzy C-means dynamic clustering algorithm is used to perform cluster analysis on carbonization characteristic data of different periods and different monitoring points, dynamically identify the carbonization development stage and spatial partition, and generate dynamic carbonization clustering labels. Step S3 includes the following steps: Based on the dominant carbonization characteristic factors, a multi-temporal-multi-monitoring-point three-dimensional data structure is constructed, wherein each dimension represents the sampling time, monitoring spatial location and characteristic factor value, forming a three-dimensional carbonization characteristic matrix that reflects the carbonization evolution process. It should be noted that the dimension is defined as follows: The first dimension is the sampling time, which is sampled on a weekly, monthly, or quarterly basis.

[0045] The second dimension is the monitoring spatial location, including various monitoring nodes of structures such as bridges and docks (numbered according to component hierarchy and geographical distribution).

[0046] The third dimension is the specific value of the dominant characteristic factor.

[0047] Matrix generation: Monitoring data from different times, monitoring points, and characteristic factors are aggregated into a three-dimensional data matrix according to a unified standard. For example, monthly data are aggregated into a three-dimensional "data cube," with each data block corresponding to a value of a specific time, location, and characteristic. All three-dimensional matrices are stored uniformly in a cloud database for easy batch processing and visualization.

[0048] The local coefficient of variation method was used to perform pre-clustering sensitivity analysis on the three-dimensional carbonization feature matrix to extract the sensitive regions of dry-wet cycle response, and the carbonization susceptibility field was constructed using the water-salt flux index as the weight correction input for the clustering algorithm. Specifically, based on a three-dimensional feature matrix, the data is sliced ​​spatially (for each node) and temporally (for each sampling period) to obtain the time series of each spatial point and the spatial distribution of each time point. For the feature value sequence of each node throughout all sampling periods, its mean and standard deviation are calculated to obtain the coefficient of variation (standard deviation / mean). The higher the coefficient of variation, the more sensitive the monitoring point is to the wet-dry cycle in terms of that feature, and the more easily it is affected by the external environment. Using heat maps or three-dimensional scatter plots, the spatial distribution of the coefficient of variation of all monitoring points is visualized, and highly sensitive areas (such as nodes near bridge surfaces or exposed to humid air) are clearly identified. Further analysis of the changes in the coefficient of variation over time reveals which periods (such as typhoon season or periods of extreme temperature difference) show a surge in overall sensitivity, capturing the most dangerous windows for carbonization transitions.

[0049] Water content and ion-selective microelectrodes were deployed at sensitive nodes to periodically measure pore water content and chloride / sulfate concentrations. By continuously measuring water and salinity data over different periods, the rates of increase / decrease were analyzed to evaluate water-salt mobility. The coefficient of variation was fused with water-salt flux data to generate a "carbonization susceptibility field" at the node level. Nodes with high susceptibility were assigned higher weights, while low-susceptibility regions were assigned lower weights. These weights were input as additional parameters into the clustering algorithm to dynamically adjust the influence of each node / period of data during clustering, allowing the algorithm to prioritize regions and time periods most likely to exhibit carbonization transitions. All sensitivity and susceptibility field analysis results were periodically reviewed by a structural durability expert team to identify abnormal fluctuations and data noise, ensuring that subsequent clustering weights were scientifically sound and reasonable.

[0050] Based on the carbonization feature matrix after weight correction, the fuzzy C-means dynamic clustering algorithm is used to iteratively cluster the carbonization development state under the influence of dry and wet cycles, and construct a set of carbonization state labels including time stationarity, spatial diffusivity and factor coupling degree. Specifically, according to the weights given by the aforementioned carbonization susceptibility field, a weight label is assigned to each data unit (i.e., each spatiotemporal feature point) of the three-dimensional feature matrix. Before clustering, the weight distribution is checked to prevent individual abnormal weights from biasing the overall clustering results. If necessary, extreme weights can be normalized.

[0051] Based on the actual development process of carbonization, the number of clusters is generally set to 3-5, such as "initial stage," "acceleration stage," "stable stage," and "local high-risk stage." A fuzziness coefficient is set to allow a monitoring point / period to have different membership degrees in multiple cluster categories (i.e., reflecting the transitional and fuzzy nature of the actual working conditions). The clustering process involves multiple iterations, each round combining the latest weight matrix with historical cluster labels to automatically adjust partitions and stage divisions. When new data flows in (such as the latest monitoring period), incremental clustering can be automatically triggered to maintain the timeliness of stage and partition labels. Each monitoring point / period will receive a set of labels, including membership degrees (reflecting the probability of belonging to each stage / partition), and these labels will be stored in the label database.

[0052] The tags not only include "development stage" (e.g., accelerated / initial / stable), but also "spatial partition" (e.g., surface / middle / protective layer) and factor coupling state (e.g., high coupling / low coupling). The engineering management platform supports visualized tag evolution animations, intuitively displaying the carbonization development stages and the spread and transition trajectories of partitions throughout the entire structure over time. The system can recall historical data at any time to review the stage changes and partition assignment changes of a specific segment in the clustering history, which helps to detect abnormal spread trends in advance and take structural reinforcement measures.

[0053] After each round of dynamic clustering, based on the set of carbonization state labels, according to the rate of change of the spatial distribution centroid of the cluster centers and the rate of change of the factor gradient, a carbonization development stage transition index is defined, and the clustering results are labeled as carbonization dynamic clustering labels including stage identifiers and partition identifiers.

[0054] Specifically, after each round of clustering, the changes in the centroid and characteristic gradients of the cluster centers (i.e., representative states of each development stage) in spatial distribution are analyzed. For example, if a cluster center moves deeper into the structure or to a new region within a continuous time period, it is considered that there has been a significant leap in carbonization development. Based on parameters such as the movement rate of cluster centers and changes in spatial coverage, transition thresholds for carbonization development stages are defined to determine the timing of structural carbonization transitions from "slow-accelerated-stable" stages. The clustering results are combined into a final label using "stage identifier + partition identifier." For example, "accelerated stage - mid-layer partition," "stable stage - protective layer partition," etc. The clustering labels for all monitoring points and all time periods are automatically archived in the database for subsequent identification, modeling, and early warning.

[0055] Furthermore, the formula for the fuzzy C-means dynamic clustering algorithm is as follows:

[0056] in, denoted by , represents the objective function value of temporal carbonation clustering, which is the overall error metric for dynamically clustering the concrete carbonation process under wet-dry cycle conditions, used to simultaneously minimize sample clustering error and time iteration disturbance; p represents the number of effective samples in the multi-temporal-multi-monitoring-point three-dimensional carbonation feature matrix; Q represents different carbonation development stages or spatial partitions. This indicates the probability of the current carbonization monitoring point belonging to a certain stage / zone; m represents the fuzziness index, used to adjust the degree of cluster fuzziness, which is usually set to 2; Represents the carbonization-dominant feature vector of the p-th sample; Let represent the cluster center feature vector of the q-th cluster; This represents the q-th cluster center formed in the previous round of dynamic clustering; This represents a dynamically adjusted weighting coefficient, used to strengthen the stability constraint of cluster centers during the time evolution process. The specific value can be set according to the changes in the dry and wet cycle.

[0057] S4: Based on the carbonization dynamic clustering labels, a multi-discriminant analysis algorithm is used to model the dominant factors of each partition, establish a discriminant function for the change of carbonization depth within the partition, and generate partition carbonization discriminant values; Step S4 includes the following steps: Based on the carbonization dynamic clustering label, carbonization feature data from different periods and spatial locations are divided according to stage identifiers and partition identifiers to construct partition feature subsets, and the corresponding dominant carbonization feature factors in each subset are extracted as input variables. It should be noted that the carbonization dynamic tags of "stage identifier + partition identifier" output by the clustering in step S3 are used to automatically classify all historical and current period monitoring data according to the tags.

[0058] For example, monitoring data labeled "Acceleration Phase - Mid-Layer Partition" is categorized into the "Mid-Layer - Acceleration Phase" feature subset; while "Stable Phase - Surface Partition" is categorized into the "Surface - Stable Phase" feature subset.

[0059] Within the data platform, data from all monitoring nodes and periods are grouped and stored according to tags. Each group contains information on the corresponding partition, stage, time point, and spatial point. For each partition-stage subset, the dominant feature factors (such as resistivity change rate, porosity dynamic change rate, alkalinity decay, etc.) selected in step S2 are invoked to extract corresponding features in batches from the subset data table as modeling input. The extraction process is automatically completed by the data processing script, ensuring that the feature dimensions of different partitions and stages are completely corresponding.

[0060] For each feature subset of the partition, a set of linear discriminant functions reflecting the level of carbonization depth variation is constructed using a multi-discriminant analysis algorithm; Specifically, the dominant feature factor data in each partition feature subset are first uniformly processed for missing values ​​(e.g., using mean imputation, outlier removal, etc.). Z-score standardization is then used to transform each feature data into a standard distribution with a mean of 0 and a variance of 1, eliminating the influence of dimensions and ensuring direct comparability between different features. For each partition feature subset, each data point is ensured to have a true carbonization depth level label, and periodic marking is performed for easy traceability and grouping. Based on historical experiments and structural specifications, carbonization depth is divided into several levels (e.g., slight, moderate, severe), and each monitoring data point is manually or automatically labeled with its corresponding carbonization level, forming a supervised dataset.

[0061] In data analysis software (such as SPSS, R, or the sklearn package for Python), select the multi-class linear discriminant analysis (LDA) or discriminant function analysis (DFA) module. Use the dominant feature factors in the partition subset as input variables and the labeled carbonization level as the target variable. Model each partition separately, training a set of linear discriminant functions. Each function should be able to delineate the decision boundaries for different carbonization levels in the feature space based on the input features. Set up cross-validation (such as K-fold cross-validation) to test the model's generalization ability under different sample combinations and prevent overfitting. Record evaluation metrics such as the discrimination accuracy, confusion matrix, and feature contribution for each model, and retain the optimal model parameters. During model training, analyze the contribution of each dominant factor to the discrimination results. If some features have no significant effect on the discrimination, they can be automatically or manually removed to improve model simplicity and stability. Archive the linear discriminant functions corresponding to each partition in the system model library for easy subsequent use. Equip each set of discriminant functions with metadata (such as applicable partition, sample range, feature list, training time, etc.) to ensure full traceability.

[0062] Based on the set of linear discriminant functions, the carbonization level of the characteristic data of each carbonization monitoring node is discriminated, and the carbonization evolution level label is output. Combined with the time series evolution trend of the monitoring node's partition, it is used as the partition carbonization discriminant value.

[0063] Specifically, the system automatically extracts dominant features (such as resistivity change rate and porosity) from the monitoring data in real time and performs the same standardization operations as during training to ensure model input consistency. The system calls the linear discriminant function for that partition, inputs the standardized feature data into the model, and the model outputs the corresponding carbonization level (e.g., "Level II - Medium"), and archives the results with labels. Each discrimination result also outputs a confidence score. If the confidence score of the result is too low or falls near the boundary of multiple categories, expert review or supplementary data collection is automatically triggered.

[0064] Carbonization level labels (by zone and by time) for all nodes are automatically recorded in the database. The platform can retrieve the carbonization level change curves for a specific section or the entire structure at any time to assist engineers in judging dynamic changes in carbonization risk. For each zone, the platform statistically analyzes the carbonization level distribution of all monitored nodes in the current period, such as calculating indicators like "proportion of severely carbonized nodes," "average carbonization level," and "maximum carbonization level transition rate," to form one or more carbonization discrimination values ​​for each zone. If a zone experiences a carbonization level transition or a significant increase in the proportion of severely carbonized nodes for several consecutive periods, the system will automatically mark the zone as in a warning state, facilitating early maintenance and structural reinforcement decisions.

[0065] The system dynamically displays all carbonization level labels and zoning criteria values ​​using various methods such as line charts, bar charts, and heat maps, allowing engineers to intuitively see the carbonization evolution process and risk distribution in each zone. The platform regularly generates carbonization monitoring and criterion results reports, pushing them to structural maintenance and management teams to support data-driven maintenance decisions and risk management. If abnormal carbonization development is subsequently detected in a particular zone, managers can trace historical monitoring data, criterion labels, and model usage with a single click, comprehensively reviewing each step of the operation and decision-making process to support scientific verification and subsequent optimization.

[0066] Furthermore, the formula for the multi-discriminant analysis algorithm is as follows:

[0067] in, The composite discrimination score represents the r-th carbonization evolution level, which is used to determine the carbonization level of a certain monitoring node. Represents the bias term of class r; Represents the j-th dominant carbonization characteristic factor; This represents the b-th dominant carbonization characteristic factor; Indicates dominant characteristic factor The weight coefficients of the first-order term in the r-th class discriminant function; Characteristic factors and The coefficient of the quadratic term in the r-th class discriminant function of the interaction term (product term) reflects the influence of the nonlinear interaction between factors on the discrimination of carbonization level.

[0068] S5: Based on the carbonation discrimination value and dominant carbonation characteristic factor of the partition, an extreme value learning machine regression algorithm is used to construct a carbonation depth prediction model, which intelligently predicts and assesses the carbonation depth of each concrete partition under dry-wet cycle conditions in brackish water areas, and generates a carbonation risk warning report.

[0069] Step S5 includes the following steps: Based on the partition carbonization discrimination value and its corresponding dominant carbonization feature factor, combined with the spatial location and stage label of the monitoring node, a structured training sample set is constructed. Each sample contains spatial partition, stage identifier, dominant feature combination value and carbonization level label. Specifically, the carbonization discrimination values ​​and dominant characteristic factors (such as resistivity change rate, porosity change, alkalinity decay, etc.) of each zone during historical monitoring periods are collected. Each sample includes the spatial zone of the monitoring node, stage identifier (such as "acceleration stage - surface zone"), combined values ​​of dominant characteristic factors, and the corresponding carbonization level label or the actual measured carbonization depth. The spatial location (such as component number, zone label) and stage (such as initial, acceleration, and stable stages) of each data point are clearly marked to ensure that the sample set can be flexibly retrieved and expanded according to zone and stage.

[0070] Based on the constructed sample set, an extreme value learning machine regression algorithm is used to establish a carbonization depth prediction model. The carbonization depth prediction model takes the combination of dominant features as input and outputs the carbonization level or depth value. Specifically, before modeling, the collected dominant feature data are analyzed column by column. Outliers are detected using box plots or quantile methods, and outliers are automatically marked and removed. For missing values, imputation is preferentially performed using the mean of adjacent periods. If the missing values ​​are continuous or account for a large proportion, the data is removed to ensure the quality of the training samples. All feature values ​​(such as resistivity change rate, porosity change, etc.) are processed in batches using normalization tools (such as MinMaxScaler or StandardScaler) to ensure consistent input range, which is beneficial for stable model convergence. The same normalization model is strictly used for the training set and subsequent inference (prediction) stages to avoid input distribution drift.

[0071] Set the number of hidden layer nodes (e.g., 50-200) and activation function type (e.g., sigmoid, tanh), and automatically iterate through multiple parameter combinations to select the optimal model structure. Train / test the model on both the training and independent test sets, outputting various evaluation metrics (e.g., MSE, R²), and select the ELM model with the best prediction performance and strongest generalization ability. Archive the final model structure, parameters, performance report, and normalized parameters together. The platform automatically generates model metadata for future traceability and version comparison.

[0072] The carbonization characteristic data of the monitoring nodes collected in real time are input into the carbonization depth prediction model, and the predicted carbonization depth value of each monitoring node is output. Based on the historical evolution trend and structural design, carbonization tolerance is set and multi-level carbonization risk thresholds are set. Specifically, the platform automatically calls the ELM prediction model in batches, inputs the normalized features of all nodes, outputs the predicted carbonization depth of each node, and archives it in batches in the structural database. The system automatically generates a three-dimensional prediction matrix of "zone-node-prediction cycle," which can display the carbonization depth prediction map, risk distribution of zones, and other multi-dimensional views in real time on the management interface. Combining structural durability specifications and historical carbonization monitoring data of this project, the expert group discusses and determines multiple threshold levels for carbonization depth, including Level I (safe), Level II (early warning), and Level III (high risk). The thresholds can be flexibly adjusted according to multiple dimensions such as zone, structural type, and environmental exposure level. For example, a lower early warning threshold is set for areas with thinner bridge deck protective layers, while the threshold is appropriately relaxed in the reinforcement protection zone. The system supports adaptive threshold updates. If the predicted values ​​for multiple consecutive cycles approach or exceed a certain threshold level, a threshold review suggestion is automatically triggered to promptly adapt to risk changes due to new working conditions or sudden environmental changes. The platform automatically compares the predicted depth with historical evolution trends. If it finds an accelerated increase in the predicted depth of a single node or a sudden change in the average predicted value of a zone in the short term, it is automatically marked as "risk aggravated" and classified into a higher risk level.

[0073] By comparing the prediction results with the carbonization threshold, and combining the stage and spatial partition of the node, the carbonization risk level is determined. Furthermore, by integrating spatial distribution characteristics, stage evolution trends, and discrimination label weights, a carbonization risk early warning matrix based on time-space joint mapping is formed. Specifically, the predicted carbonization depth for each node and each cycle is compared with the risk threshold corresponding to the partition, and the system automatically assigns a risk label (such as "safe", "warning", "high risk") to each data point. The system automatically integrates the node's historical clustering labels and the weight of the discrimination label (such as the node being in the acceleration phase continuously, or having a high discrimination value), and assigns it a higher risk priority.

[0074] Time-space risk matrix generation: Using nodes (or partitions) as the horizontal axis and periods as the vertical axis, fill the risk level of each spatiotemporal point in a two-dimensional matrix.

[0075] Each matrix unit carries rich information, including prediction depth, risk level, discrimination weight, stage identifier, etc., to achieve a multi-dimensional mapping of "structure-time-feature-risk".

[0076] The system continuously tracks the aggregation, spread, and transition of "high-risk" units in the matrix. If a high-risk area is detected to be spreading towards the core structural area, the warning level is automatically upgraded. Spatial clustering of high-risk units in the matrix identifies "risk hotspots," supporting subsequent on-site intensive monitoring or maintenance actions. Each high-risk determination records relevant data, models, parameters, and historical trends, allowing engineers to review them with a single click, conduct manual verification, or automatically receive risk reports and recommendations.

[0077] Based on the aforementioned risk warning matrix, a carbonization risk warning report is automatically generated; the carbonization risk warning report includes high-risk area identification, phased carbonization development trends, regional risk level assessment, and carbonization intervention recommendations.

[0078] It should be noted that the carbonization risk warning report includes the following: High-risk area identification: The platform automatically filters high-risk units in the early warning matrix and locates the partitions and nodes in the structure that are most likely to undergo excessive carbonization in the current / future period.

[0079] Development Trends and Risk Level Assessment: Through trend analysis, carbonization development curves and risk level evolution diagrams for each zone / node are generated, visually demonstrating the trajectory of carbonization risk over time.

[0080] Zonal Risk Report and Intervention Recommendations: The report details the risk level, predicted carbonization depth, and trend for each zone and stage, and provides targeted intervention recommendations for high-risk areas (such as increasing monitoring frequency, accelerating surface protection, and early structural reinforcement).

[0081] Automated push and historical archiving: All risk reports can be automatically pushed to the engineering management team and structural maintenance department, and archived in the database for easy review and long-term trend tracking.

[0082] Furthermore, the formula for the carbonization depth prediction model is as follows:

[0083] in, This represents the predicted carbonization depth of the s-th partition; This represents the standardized value of the h-th dominant carbonization characteristic factor within the s-th partition; This represents the spatial heterogeneity correction term for the s-th partition; This represents the input weight from the h-th carbonization dominant feature factor to the r-th hidden layer node; represents the output weight coefficient of the r-th hidden layer node; R represents the number of hidden layer nodes; This represents the bias term of the r-th hidden layer node; Represents the set of dominant carbonization characteristic factors; This represents the activation function.

[0084] In summary, this invention comprehensively acquires multi-source characteristic data of concrete under complex environments by setting up various types of carbonation monitoring nodes, including micro-pH, resistivity, temperature and humidity, image, NMR, and XRD. Compared to traditional single monitoring methods that rely solely on carbonation depth or pH, this method achieves a multi-dimensional characterization of the carbonation process, providing a more accurate data foundation for subsequent modeling and risk assessment.

[0085] This invention quantifies the nonlinear coupling relationship between various features and carbonization depth using the Maximum Information Coefficient (MIC) algorithm. Combined with heatmaps, network diagrams, and spatiotemporal evolution analysis, it scientifically selects dominant feature factors that possess stability, spatial breadth, and synergy, significantly improving the input quality of subsequent clustering and prediction models and reducing redundant noise interference. A fuzzy C-means dynamic clustering algorithm is employed, combined with the local coefficient of variation method and a carbonization susceptibility field constructed using water-salt flux, to dynamically identify carbonization evolution stages and spatial partitions. By setting clustering weights, it focuses on highly sensitive regions and key transition cycles, enhancing the relevance and practicality of the clustering results and solving the problem that traditional static clustering struggles to handle time-varying processes.

[0086] This invention addresses different spatial partitions and carbonization stages by constructing linear discriminant functions using multi-discriminant analysis to automatically classify and attribute carbonization levels within each partition. By introducing interaction term weights, the model can identify the impact of nonlinear interactions between features on carbonization levels, enhancing the explanatory power and accuracy of level discrimination and effectively supporting engineering partition management and maintenance plan formulation. A carbonization depth prediction model based on Extreme Value Learning Machine (ELM) is constructed, leveraging its high-speed training capability and strong nonlinear fitting ability to achieve rapid prediction of partition carbonization depth in complex environments. The model has a simple structure, is easy to deploy, and can flexibly incorporate newly added monitoring data for rapid retraining, supporting online learning and real-time early warning.

[0087] This invention integrates predicted depth values, historical evolution trends, spatial zoning, and carbonization stage information to construct a multi-dimensional "carbonization risk early warning matrix," dynamically tracking the evolution path of carbonization risks. The system supports hot zone clustering, transition identification, and risk level classification, and automatically generates early warning reports to assist engineers in developing targeted intervention measures, shifting from "passive response" to "proactive prevention." From data collection, feature extraction, model training, risk assessment to result early warning, a closed-loop process is formed. The system features data labeling, model version control, and a traceable decision-making process mechanism, providing standardized and regulated technical support for structural health monitoring and significantly improving the intelligence and scientific level of engineering monitoring.

[0088] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas, characterized in that, Includes the following steps: S1: Carbonation monitoring nodes are set from the surface layer to the internal structure of concrete components to periodically collect multi-source carbonation characteristic data; S2: Based on the multi-source carbonization feature data, the correlation between the multi-source carbonization feature data and the carbonization depth is analyzed using the maximum information coefficient algorithm to screen out the dominant carbonization feature factors. S3: Based on the carbonization dominant characteristic factors, a fuzzy C-means dynamic clustering algorithm is used to perform cluster analysis on carbonization characteristic data of different periods and different monitoring points, dynamically identify the carbonization development stage and spatial partition, and generate dynamic carbonization clustering labels. S4: Based on the carbonization dynamic clustering labels, a multi-discriminant analysis algorithm is used to model the dominant factors of each partition, establish a discriminant function for the change of carbonization depth within the partition, and generate partition carbonization discriminant values; S5: Based on the carbonation discrimination value and dominant carbonation characteristic factor of the partition, an extreme value learning machine regression algorithm is used to construct a carbonation depth prediction model, which intelligently predicts and assesses the carbonation depth of each concrete partition under dry-wet cycle conditions in brackish water areas, and generates a carbonation risk warning report.

2. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 1, characterized in that, The multi-source carbonization characteristic data includes the carbonization depth of concrete components, resistivity change rate, carbonization interface migration speed, surface color grayscale gradient, cumulative carbonization product index, porosity change, and alkalinity decay trajectory.

3. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 1, characterized in that, Step S2 includes the following steps: The multi-source carbonization characteristic data are used to construct a feature sequence matrix according to the monitoring nodes and time order. The feature sequence matrix uses the data group of each monitoring period as the row vector and the feature type as the column vector. Based on the feature sequence matrix, the maximum information coefficient algorithm is used to measure the degree of nonlinear correlation between each type of carbonization feature data and carbonization depth, forming a feature-depth coupling correlation map. By combining the temporal evolution trend and spatial variability of the feature dimensions, multi-level principal factor identification is performed on the coupling spectrum, and carbonization-dominant feature factors that exhibit a strong spatiotemporal coupling mode with carbonization depth are screened out, including the resistivity change rate-grayscale gradient joint trajectory, porosity dynamic change rate and alkalinity decay coupling term.

4. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 3, characterized in that, The formula for the maximum information coefficient algorithm is as follows: in, This represents the maximum information coefficient value; The f-th dominant carbonation characteristic factor is represented; D represents the carbonation depth of concrete. Representation of features The joint probability of the bin with depth D in the i-th row and j-th column; Representation of features Marginal probabilities within the i-th bin; The marginal probability of carbonization depth D within the j-th bin is represented by k; k represents the feature value. The number of bins; l represents the number of bins for carbonization depth D; n represents the total number of samples; This represents the binning penalty factor.

5. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 1, characterized in that, Step S3 includes the following steps: Based on the dominant carbonization characteristic factors, a multi-temporal-multi-monitoring-point three-dimensional data structure is constructed, wherein each dimension represents the sampling time, monitoring spatial location and characteristic factor value, forming a three-dimensional carbonization characteristic matrix that reflects the carbonization evolution process. The local coefficient of variation method was used to perform pre-clustering sensitivity analysis on the three-dimensional carbonization feature matrix to extract the sensitive regions of dry-wet cycle response, and the carbonization susceptibility field was constructed using the water-salt flux index as the weight correction input for the clustering algorithm. Based on the carbonization feature matrix after weight correction, the fuzzy C-means dynamic clustering algorithm is used to iteratively cluster the carbonization development state under the influence of dry and wet cycles, and construct a set of carbonization state labels including time stationarity, spatial diffusivity and factor coupling degree. After each round of dynamic clustering, based on the set of carbonization state labels, according to the rate of change of the spatial distribution centroid of the cluster centers and the rate of change of the factor gradient, a carbonization development stage transition index is defined, and the clustering results are labeled as carbonization dynamic clustering labels including stage identifiers and partition identifiers.

6. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 5, characterized in that, The formula for the fuzzy C-means dynamic clustering algorithm is as follows: in, denoted by the objective function value of temporal carbonization clustering; p represents the number of effective samples in the multi-temporal-multi-monitoring-point three-dimensional carbonization feature matrix; Q represents different carbonization development stages or spatial partitions. This indicates the probability of a current carbonization monitoring point being assigned to a certain stage / zone; m represents the fuzzy index. Represents the carbonization-dominant feature vector of the p-th sample; Let represent the cluster center feature vector of the q-th cluster; This represents the q-th cluster center formed in the previous round of dynamic clustering; This indicates that the weighting coefficient is dynamically adjusted.

7. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 1, characterized in that, Step S4 includes the following steps: Based on the carbonization dynamic clustering label, carbonization feature data from different periods and spatial locations are divided according to stage identifiers and partition identifiers to construct partition feature subsets, and the corresponding dominant carbonization feature factors in each subset are extracted as input variables. For each feature subset of the partition, a set of linear discriminant functions reflecting the level of carbonization depth variation is constructed using a multi-discriminant analysis algorithm; Based on the set of linear discriminant functions, the carbonization level of the characteristic data of each carbonization monitoring node is discriminated, and the carbonization evolution level label is output. Combined with the time series evolution trend of the monitoring node's partition, it is used as the partition carbonization discriminant value.

8. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 1, characterized in that, Step S5 includes the following steps: Based on the partition carbonization discrimination value and its corresponding dominant carbonization feature factor, combined with the spatial location and stage label of the monitoring node, a structured training sample set is constructed. Each sample contains spatial partition, stage identifier, dominant feature combination value and carbonization level label. Based on the constructed sample set, an extreme value learning machine regression algorithm is used to establish a carbonization depth prediction model. The carbonization depth prediction model takes the combination of dominant features as input and outputs the carbonization level or depth value. The carbonization characteristic data of the monitoring nodes collected in real time are input into the carbonization depth prediction model, and the predicted carbonization depth value of each monitoring node is output. Based on the historical evolution trend and structural design, carbonization tolerance is set and multi-level carbonization risk thresholds are set. By comparing the prediction results with the carbonization threshold, and combining the stage and spatial partition of the node, the carbonization risk level is determined. Furthermore, by integrating spatial distribution characteristics, stage evolution trends, and discrimination label weights, a carbonization risk early warning matrix based on time-space joint mapping is formed. Based on the aforementioned risk warning matrix, a carbonization risk warning report is automatically generated.

9. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 8, characterized in that, The formula for the carbonization depth prediction model is as follows: in, This represents the predicted carbonization depth of the s-th partition; This represents the standardized value of the h-th dominant carbonization characteristic factor within the s-th partition; This represents the spatial heterogeneity correction term for the s-th partition; This represents the input weight from the h-th carbonization dominant feature factor to the r-th hidden layer node; represents the output weight coefficient of the r-th hidden layer node; R represents the number of hidden layer nodes; This represents the bias term of the r-th hidden layer node; Represents the set of dominant carbonization characteristic factors; This represents the activation function.

10. The method for monitoring and predicting the carbonation depth of concrete under wet-dry cycle conditions in brackish water areas according to claim 8, characterized in that, The carbonization risk warning report includes identification of high-risk areas, phased carbonization development trends, regional risk level assessment, and carbonization intervention recommendations.

Images