Deep learning-based method and system for analyzing cement-based material penetration-dissolution deterioration

Abstract

The application discloses a cement-based material penetration corrosion deterioration analysis method and system based on deep learning, and relates to the technical field of data processing, and comprises the following steps: calling operation and maintenance information, collecting microstructure characterization to obtain K multi-scale characterizations of K operation and maintenance areas; aggregating the K operation and maintenance areas to obtain M groups of operation and maintenance areas; after dividing the K multi-scale characterizations, performing penetration corrosion deterioration feature networking matching to obtain M sample time sequence characterizations and M sample time sequence deterioration features; performing deterioration conduction prediction on the K operation and maintenance areas to construct a deterioration prediction graph neural network; inputting K time encoding vectors of the K operation and maintenance areas into the deterioration prediction graph neural network to perform cement-based material penetration corrosion deterioration prediction on a corrosion-prone area and output real-time operation and maintenance tasks. The application solves the technical problem of low cement-based material penetration corrosion deterioration prediction accuracy in the prior art, and achieves the technical effect of improving cement-based material deterioration prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning. Background Technology

[0002] Cement-based materials are widely used in various infrastructure constructions, especially in important projects such as hydropower dams. However, during long-term use, cement-based materials are susceptible to degradation due to factors such as penetration and corrosion. This degradation not only affects the durability of the material but may also threaten the safety of the project. Traditional methods for predicting the degradation of cement-based materials mainly rely on manual monitoring and empirical models, which cannot achieve accurate prediction of degradation trends and have low accuracy in predicting penetration and corrosion degradation of cement-based materials. Summary of the Invention

[0003] This application provides a method and system for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning, which is intended to address the technical problem of low accuracy in predicting the penetration and corrosion degradation of cement-based materials in existing technologies.

[0004] In view of the above problems, this application provides a method and system for analyzing the penetration and corrosion deterioration of cement-based materials based on deep learning.

[0005] The first aspect of this application provides a deep learning-based method for analyzing the penetration and corrosion degradation of cement-based materials, the method comprising:

[0006] After locally accessing the operation and maintenance information of the high-risk karstification area of ​​the hydropower plant dam, microstructural characterization is collected based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, wherein each K multi-scale characterization is identified by K sample metadata. The K operation and maintenance areas are aggregated based on the K sample metadata to obtain M groups of operation and maintenance areas. After dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, network matching of permeation and karstification degradation features is performed to obtain M sample time-series characterizations and M sample time-series degradation features. Based on the M sample time-series characterizations and M sample time-series degradation features, degradation propagation prediction is performed for the K operation and maintenance areas to construct a degradation prediction graph neural network. By inputting the K time-encoded vectors of the K operation and maintenance areas into the degradation prediction graph neural network, permeation and karstification degradation prediction of cement-based materials in the high-risk karstification area is performed, and real-time operation and maintenance tasks are output.

[0007] A second aspect of this application provides a deep learning-based system for analyzing the penetration and corrosion degradation of cement-based materials, the system comprising:

[0008] The microstructure characterization acquisition module is used to collect microstructure characterization data based on the operation and maintenance information of the high-incidence area of ​​hydropower plant dam after locally retrieving it, and obtain K multi-scale characterizations of K operation and maintenance areas, wherein the K multi-scale characterizations are identified by K sample metadata. The operation and maintenance area determination module is used to aggregate the K operation and maintenance areas according to the K sample metadata to obtain M groups of operation and maintenance areas. The matching module is used to perform network matching of permeation and erosion degradation features after dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, and obtain M sample time-series characterizations and M sample time-series degradation features. The network construction module is used to predict the degradation propagation of the K operation and maintenance areas based on the M sample time-series characterizations and M sample time-series degradation features, so as to construct a degradation prediction graph neural network. The degradation prediction module is used to predict the permeation and erosion degradation of cement-based materials in the high-incidence area of ​​erosion by inputting the K time-encoded vectors of the K operation and maintenance areas into the degradation prediction graph neural network, and output real-time operation and maintenance tasks.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] This application, after locally accessing the operation and maintenance information of a hydropower plant dam with a high incidence of karst erosion, collects microstructural characterization data based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, wherein each of the K multi-scale characterizations is identified by K sample metadata. The K operation and maintenance areas are aggregated based on the K sample metadata to obtain M groups of operation and maintenance areas. After dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, network matching of permeation and karst degradation features is performed to obtain M sample time-series characterizations and M sample time-series degradation features. Based on the M sample time-series characterizations and M sample time-series degradation features, degradation propagation prediction is performed on the K operation and maintenance areas to construct a degradation prediction graph neural network. By inputting the K time-encoded vectors of the K operation and maintenance areas into the degradation prediction graph neural network, permeation and karst degradation prediction of cement-based materials in the high-incidence karst erosion areas is performed, and real-time operation and maintenance tasks are output. This invention addresses the technical problem of low accuracy in predicting the penetration and corrosion deterioration of cement-based materials in existing technologies. By combining multi-scale characterization and acquisition, network matching of penetration and corrosion deterioration features, and graph neural network prediction, it achieves the technical effect of improving the accuracy of predicting the deterioration of cement-based materials. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of the process for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning, as provided in the embodiments of this application;

[0013] Figure 2 A schematic diagram of the structure of the deep learning-based cement-based material penetration corrosion and deterioration analysis system provided in this application embodiment.

[0014] Figure labeling: 11 Microstructure characterization acquisition module, 12 Operation and maintenance area determination module, 13 Matching module, 14 Network construction module, 15 Degradation prediction module. Detailed Implementation

[0015] This application provides a method and system for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning. It addresses the technical problem of low prediction accuracy of penetration and corrosion degradation of cement-based materials in existing technologies by combining multi-scale characterization and acquisition, network matching of penetration and corrosion degradation features, and graph neural network prediction, thereby improving the technical accuracy of cement-based material degradation prediction.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0018] Example 1, as Figure 1 As shown, this application provides a deep learning-based method for analyzing the penetration and corrosion degradation of cement-based materials, the method comprising:

[0019] Step S100: After calling the operation and maintenance information of the high-incidence area of ​​hydropower plant dam karst locally, microstructure characterization is collected based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, wherein the K multi-scale characterizations are identified by K sample metadata.

[0020] In this embodiment of the application, the operation and maintenance information of the hydropower plant dam karst high-incidence area is first retrieved locally, that is, by accessing historical operation and maintenance records and real-time data, relevant operation and environmental change information involving the dam area is obtained.

[0021] Then, microstructural characterization was collected based on the operation and maintenance information. Specifically, firstly, permeation and dissolution core samples were drilled from each operation and maintenance area, and the chloride ion penetration depth was measured using a colorimetric method to obtain K macroscopic characterizations. Next, X-ray tomography was used to scan the K operation and maintenance areas to obtain K interconnected porosities as millimeter-scale characterizations. Then, based on the longitudinal wave propagation characteristics of ultrasound, K relative attenuation rates were calculated as micrometer-scale features. The characterization data of each operation and maintenance area was bound and stored with sample metadata (such as repair material ratios and environmental exposure conditions), ultimately resulting in K multi-scale characterizations.

[0022] Furthermore, in the method provided in the application embodiments, before calling the operation and maintenance information of the hydropower plant dam's high-risk karstification area locally, it also includes:

[0023] By pre-embedding a distributed optical fiber sensor network in the hydropower plant dam, distributed time-series seepage velocities are extracted. Using seepage velocity thresholds and duration of exceeding the threshold as dual constraints, the distributed time-series seepage velocities are traversed to locate the spatial distribution map of high seepage areas. Local dam operation and maintenance records are used to perform defect spatial density analysis and generate a defect density heat map. After spatially overlaying the spatial distribution map of high seepage areas and the defect density heat map, multi-criteria decision fusion analysis is performed to locate the high-incidence areas of dissolution.

[0024] In this embodiment, distributed time-series seepage velocity data is extracted using a distributed fiber optic sensor network pre-embedded in the hydropower plant dam. This step involves deploying multiple sensors within the dam structure using the distributed fiber optic sensor network, employing fiber Bragg grating (FBG) sensing technology to monitor seepage velocity in real time. Each sensor collects seepage velocity data and generates time-series data; that is, each sensor records the changes in seepage velocity at different times. Through this process, the distributed time-series seepage velocity is obtained.

[0025] Next, the distributed time-series seepage velocities are traversed using both the seepage velocity threshold and the duration of exceeding the threshold as dual constraints. During this process, the distributed time-series seepage velocities are analyzed and compared with the preset seepage velocity threshold to locate the spatial distribution map of high seepage areas. First, a seepage velocity threshold is set (e.g., exceeding 1 m / s is considered abnormal seepage), and it is determined whether the seepage velocity at each monitoring point exceeds this threshold. If the seepage continuously exceeds the set threshold, the area is identified as a high seepage area. These seepage data exceeding the threshold are mapped onto the spatial coordinates of the dam to generate a spatial distribution map of high seepage areas.

[0026] Then, the historical operation and maintenance records of the dam are retrieved to perform a spatial density analysis of defects. These records contain historical defects, repair status, and detailed information on each repaired area. Statistical analysis of these historical records is used to calculate the frequency of defects in each area and generate a defect density heatmap.

[0027] Finally, the spatial distribution map of the high-permeability area and the defect density heat map are spatially overlaid, combining seepage data and defect density data to comprehensively consider the current status of seepage corrosion and the distribution of existing defects. This process, by spatially overlapping areas with seepage velocities exceeding the threshold with areas with high defect frequency, more clearly shows which areas face the dual risks of seepage corrosion and structural defects simultaneously. Through this overlay, areas where seepage and defects highly overlap are identified. Next, the Multi-Criterion Decision Analysis (MCDA) method is used to comprehensively analyze multiple factors (such as seepage velocity, defect density, and environmental conditions) to determine the overall degradation risk of each area. In this analysis, the levels of seepage velocity, defect density, and environmental conditions (such as water pressure and pH value) all influence the degree of degradation in each area. For example, in a high-permeability area, if the area also has a high defect density (such as frequent cracks or seepage problems), it indicates that the seepage corrosion problem in that area is more severe and may accelerate material degradation. Therefore, the degradation risk of this area will be assessed as high, and repair and reinforcement will be prioritized.

[0028] For example, suppose that in a certain area of ​​a dam, the seepage velocity has exceeded a threshold (e.g., 1.0 m / s), and the defect density in this area is high, showing multiple crack repair records and seepage problems. Through multi-criteria decision analysis, this area is identified as a high-incidence zone for dissolution.

[0029] Furthermore, in the method provided in the application embodiments, the microstructure characterization is collected based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, and it also includes:

[0030] Based on the maintenance information, K maintenance areas are equidistantly divided in the high-incidence area of ​​dissolution. Core samples are drilled from the K maintenance areas to obtain K permeable dissolution core samples. The K chloride ion penetration depths of the K permeable dissolution core samples are measured using a colorimetric method as K macroscopic scale characteristics. The K maintenance areas are scanned by X-ray tomography to obtain K interconnected porosities as K millimeter-scale characteristics. Based on the ultrasonic longitudinal wave propagation characteristics of the K maintenance areas, K relative attenuation rates are calculated and output as K micrometer-scale features. The K macroscopic scale characteristics, K millimeter-scale characteristics, and K micrometer-scale features are bound and stored for maintenance areas to obtain the K multi-scale characteristics.

[0031] In this embodiment of the application, based on the operation and maintenance information, K operation and maintenance areas are equally spaced in the high-incidence area of ​​dissolution, and each operation and maintenance area is a circular area with a diameter of 50 centimeters.

[0032] Next, core samples were drilled from K maintenance areas using core drilling equipment (such as handheld or mechanical drills) at predetermined locations within each maintenance area, resulting in K permeation and dissolution core samples. Subsequently, a colorimetric method was used to measure the K chloride ion penetration depths of these K core samples, serving as K macroscopic characterizations. In this process, the colorimetric method involves adding specific colorimetric reagents, allowing the chloride ion penetration depth in the material to be revealed through color changes. The colorimetric method is used here to measure the chloride ion penetration depth, and the resulting penetration depth data serves as a macroscopic characterization, revealing the surface corrosion of cement-based materials during the permeation and dissolution process. Through this step, macroscopic characterizations of the K areas are obtained.

[0033] Then, the pore structure of each maintenance area was non-destructively detected by X-ray tomography (CT scan), and the connectivity porosity of each area was obtained by analyzing the scan images, thus obtaining K connectivity porosities as K millimeter-scale representations.

[0034] Next, based on the ultrasonic longitudinal wave propagation characteristics of the K maintenance areas, the propagation speed and attenuation of the material are measured using ultrasonic longitudinal wave detection technology, thereby calculating the relative attenuation rate of each maintenance area and obtaining K relative attenuation rates as K micrometer-scale characteristics.

[0035] Finally, the K macro-scale representations, K millimeter-scale representations, and K micrometer-scale features are bound and stored for each operation and maintenance area, resulting in K multi-scale representations. Through a database management system, the macro-scale, millimeter-scale, and micrometer-scale representation data for each operation and maintenance area are integrated and stored, and then bound to the operation and maintenance records (such as repair records, environmental conditions, etc.) for each area to obtain the K multi-scale representation datasets.

[0036] Step S200: Aggregate the K operation and maintenance regions based on the K sample metadata to obtain M groups of operation and maintenance regions.

[0037] In this embodiment, when aggregating K maintenance areas based on K sample metadata, sample metadata for the K maintenance areas is extracted from the maintenance information. This metadata is used to determine the consistency of the characteristics of each maintenance area. Based on factors such as repair material ratio, environmental pH value, and hydraulic gradient, the consistency of the sample metadata is evaluated, and areas with similar characteristics are aggregated to obtain M sets of sample metadata. Finally, based on these aggregated M sets of sample metadata, the K maintenance areas are divided into M groups of maintenance areas.

[0038] Furthermore, in the method provided in the application embodiment, aggregating the K operation and maintenance regions based on the K sample metadata to obtain M groups of operation and maintenance regions further includes:

[0039] Based on the K maintenance areas, K sample metadata are retrieved from the maintenance information, wherein the sample metadata includes repair maintenance time nodes, repair material ratios, environmental pH values, and environmental hydraulic gradients; the consistency of the K sample metadata is determined based on the repair material ratios, environmental pH values, and environmental hydraulic gradients, so as to aggregate the K sample metadata to obtain M sets of sample metadata; the K maintenance areas are grouped into the M sets of maintenance areas based on the M sets of sample metadata.

[0040] In this embodiment, K sample metadata are retrieved from the operation and maintenance information based on K operation and maintenance areas. The sample metadata includes the repair and maintenance time node, repair material ratio, environmental pH value, and environmental hydraulic gradient. First, sample metadata for each operation and maintenance area is extracted from the dam's operation and maintenance management system using a data extraction method. This metadata includes the repair and maintenance time node (i.e., the time of the last repair), the repair material ratio (i.e., the ratio of cement to other components used, such as 3:1), the environmental pH value (referring to the acidity or alkalinity of the environmental water exposed to the cement-based material), and the environmental hydraulic gradient (reflecting changes in water pressure and velocity).

[0041] Next, consistency of metadata for K samples is determined based on the repair material ratio, ambient pH value, and ambient hydraulic gradient. This step uses consistency determination algorithms, such as similarity metrics (e.g., Euclidean distance or cosine similarity), to measure the similarity of repair material ratio, ambient pH value, and ambient hydraulic gradient for each maintenance area. These algorithms determine the similarity between areas based on the degree of difference in various indicators within the sample metadata. For example, if two areas have similar repair material ratios and similar pH values ​​and hydraulic gradients, they are considered relatively consistent in terms of environmental conditions and remediation strategies.

[0042] Next, the aggregated metadata of the K samples is used to obtain M groups of sample metadata. In this step, data clustering techniques (e.g., k-means clustering or hierarchical clustering) are used to group the K sample metadata with similar characteristics into M groups. Specifically, the clustering algorithm classifies the K samples based on features such as the repair material ratio, environmental pH value, and hydraulic gradient in the sample metadata, generating M groups of sample metadata with similar characteristics. For example, if certain areas use similar repair material ratios and have similar environmental pH values ​​and hydraulic gradients, these areas will be grouped together, thus forming M groups of sample metadata.

[0043] Finally, the K maintenance areas are grouped into M maintenance areas based on the metadata of the M groups of samples. This step uses a grouping method (such as label assignment based on clustering results) to divide the K maintenance areas into M groups based on the similarity of their sample metadata. Each group contains areas with similar environmental conditions and repair characteristics, thus providing a clear basis for subsequent corrosion and dissolution prediction and reinforcement decisions.

[0044] Step S300: After dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, perform network matching of penetration and corrosion degradation features to obtain M sample time series characterizations and M sample time series degradation features.

[0045] In this embodiment, after dividing the M maintenance area groups into K multi-scale representations, the multi-scale representations of the K maintenance areas are first aggregated into M groups to obtain M multi-scale representations, each group representing maintenance areas with similar characteristics. Then, based on the mapping relationship between these M multi-scale representations and sample metadata, the repair and maintenance time nodes of each group are called as time references for analysis. Using these repair time nodes, the degradation rate of each group is calculated, and M time-varying degradation rate sequences are obtained based on repair history and environmental changes, reflecting the dissolution and degradation progress of each group at different time periods. Finally, these M time-varying degradation rate sequences are used for network matching of penetration and dissolution degradation features to obtain M sample time-series representations and M sample time-series degradation features.

[0046] Furthermore, in the method provided in the application embodiment, after dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, performing network matching of penetration and corrosion degradation features to obtain M sample time-series characterizations and M sample time-series degradation features, it also includes:

[0047] The M groups of maintenance areas are divided into K multi-scale representations to obtain M groups of multi-scale representations. Based on the mapping relationship between the M groups of multi-scale representations and the M groups of sample metadata, the M groups of repair and maintenance time nodes are invoked. The degradation rate of the M groups of multi-scale representations is calculated based on the M groups of repair and maintenance time nodes to obtain M time-varying degradation rate sequences. The M time-varying degradation rate sequences are used to perform network matching of penetration and dissolution degradation features to obtain M sample time-series representations and M sample time-series degradation features.

[0048] In this embodiment of the application, K multi-scale representations are divided according to the M groups of operation and maintenance areas. First, a clustering analysis method (such as k-means clustering algorithm) is used to group the K operation and maintenance areas according to their repair material ratio, environmental pH value, environmental hydraulic gradient and other characteristics to obtain M groups of multi-scale representations.

[0049] Next, based on the mapping relationship between the M groups of multi-scale characterizations and the M groups of sample metadata, the M groups of repair and maintenance time nodes are invoked. This step uses data extraction techniques to extract the repair and maintenance time nodes for each group from the maintenance records. The repair and maintenance time nodes mark the time of the last repair for each area. By correlating these repair time nodes with the multi-scale characterization data of each group, a corresponding time reference is provided for each group of areas, ensuring that the repair history is correlated with the material degradation process, and finally obtaining the M groups of repair and maintenance time nodes.

[0050] Subsequently, based on the M groups of repair and maintenance time nodes, the degradation rate of the M groups of multi-scale characteristics is calculated. In this process, the repair and maintenance time nodes are first arranged in ascending order of time, resulting in a repair and maintenance time series. Based on these time series, each group of multi-scale characteristics (such as permeation depth, porosity, and ultrasonic attenuation) is serialized in chronological order, forming a multi-scale characteristic sequence. Next, based on the repair intervals in each group of repair and maintenance time series, the degradation rate at each time point is calculated. This calculation is performed by comparing the changes in multi-scale characteristics at adjacent time nodes. For example, the degradation rate at each time node is obtained by calculating the growth rate of permeation depth, the expansion rate of connected pores, and the change rate of ultrasonic attenuation. Finally, this step yields M time-varying degradation rate sequences.

[0051] Finally, M time-varying degradation rate sequences were used for network matching of penetration and dissolution degradation features. This step, through data augmentation and network matching techniques, matched and expanded each group's time-varying degradation rate sequence with other penetration and dissolution features (such as chloride ion penetration depth, porosity changes, etc.). This process, through network matching in the time dimension, expanded the penetration and dissolution data at each time point, ultimately obtaining M sample time-series representations and M sample time-series degradation features.

[0052] Furthermore, in the method provided in the application embodiment, the degradation rate is calculated based on the M groups of repair and maintenance time nodes to obtain M time-varying degradation rate sequences, and the method further includes:

[0053] The first group of repair and maintenance time nodes are arranged in ascending order of time to obtain the first repair and maintenance time series. Based on the first repair and maintenance time series, the first group of multi-scale representations is serialized to obtain the first multi-scale representation sequence. Based on the repair and maintenance interval of the first repair and maintenance time series, the degradation rate is calculated on the mapping of the first multi-scale representation sequence to obtain the first degradation rate time-varying sequence. Each degradation rate time-varying feature in the degradation rate time-varying sequence consists of the penetration depth growth rate, the connected pore growth rate, and the ultrasonic attenuation growth rate.

[0054] In this embodiment, the first group of repair and maintenance time nodes are arranged in ascending order of time. Then, by using a sorting algorithm, such as quicksort or mergesort, each repair and maintenance time node is arranged in chronological order to obtain the first repair and maintenance time sequence.

[0055] Subsequently, based on the first repair and maintenance time series, the first set of multi-scale representations is serialized to obtain the first multi-scale representation sequence. This process arranges the multi-scale representation data (such as permeation depth, porosity, and ultrasonic attenuation) corresponding to each time node in chronological order, forming a time series. In this process, a data serialization method is used to map and sort the multi-scale representations (which can be indicators such as permeation depth, porosity, or ultrasonic attenuation measured at different time points) of each repair time node to the time node. In this way, the first multi-scale representation sequence is obtained.

[0056] Then, based on the repair and maintenance intervals of the first repair and maintenance time series, the degradation rate of the first multi-scale characterization sequence is calculated. In this process, the time-difference method is used to calculate the rate of change of multi-scale characterization between adjacent time nodes based on the intervals between repair and maintenance time nodes. Specifically, for each pair of adjacent time nodes, the change in multi-scale characterization (such as permeability depth, porosity, ultrasonic attenuation, etc.) between them is calculated, and the degradation rate is calculated in conjunction with the time interval. For example, if the permeability depth increases from 1.2 cm to 1.5 cm within a time period of 3 months, the rate of increase in permeability depth is 0.1 cm / month. The growth rates of porosity and ultrasonic attenuation are calculated using a similar method to obtain the degradation rate for each time period. Finally, the first degradation rate time-varying sequence is obtained, where each degradation rate time-varying feature in the degradation rate time-varying sequence consists of the permeability depth growth rate, the connected pore growth rate, and the ultrasonic attenuation growth rate.

[0057] Step S400: Based on the time-series representations of the M samples and the time-series degradation features of the M samples, perform degradation propagation prediction for the K operation and maintenance areas to construct a degradation prediction graph neural network.

[0058] In this embodiment, based on the time-series representations and degradation characteristics of M samples, degradation trend fitting is first performed to obtain M material degradation curves. These degradation curves are then mapped and assigned to K maintenance areas, generating degradation curves for K regions. Next, using the hydraulic gradient direction as a constraint, degradation propagation prediction is performed on the degradation curves of these regions, and a degradation prediction graph neural network is constructed based on the prediction results. Finally, by decomposing the degradation curves of the K regions, a training dataset of K samples is obtained, and the material degradation prediction model of each node in the graph neural network is then tuned and optimized to obtain the degradation prediction graph neural network.

[0059] Furthermore, in the method provided in the application embodiments, based on the time-series representations of the M samples and the time-series degradation features of the M samples, degradation propagation prediction is performed on the K maintenance areas to construct a degradation prediction graph neural network, which further includes:

[0060] Based on the time-series representations and degradation characteristics of the M samples, degradation trend fitting is performed to output M material degradation curves. According to the M groups of operation and maintenance areas, the M material degradation curves are mapped and assigned to the K operation and maintenance areas to generate K regional degradation curves. Using the hydraulic gradient direction as a constraint, degradation propagation prediction is performed on the K regional degradation curves, and the degradation prediction graph neural network is constructed based on the prediction results. The K regional degradation curves are decomposed to obtain a K sample training dataset, and the parameters of the K material degradation prediction model for the K graph nodes in the degradation prediction graph neural network are tuned and optimized.

[0061] In this embodiment, based on the time-series characterization and time-series degradation characteristics of M samples, the degradation trend of the time-series characteristics (a three-dimensional vector of the rate of increase in penetration depth, the rate of increase in porosity, and the rate of increase in ultrasonic attenuation) of the M samples is first fitted using a time-series data fitting method (such as least squares or regression analysis), thereby obtaining M material degradation curves. Specifically, the degradation characteristics of each sample region are modeled using a time-series data fitting method, and time is correlated with these characteristics (such as the rate of increase in penetration depth, the rate of change in porosity, and the rate of increase in ultrasonic attenuation). The fitted results generate M material degradation curves.

[0062] Next, based on the M groups of maintenance areas, the M material degradation curves are mapped and assigned to K maintenance areas, generating K regional degradation curves. This step uses spatial mapping techniques, such as interpolation or weighted averaging, to map the degradation curves of the M groups of samples to the K specific maintenance areas. The material degradation curve of each area reflects the different repairs and environmental impacts experienced by that area during the corrosion process. Through this mapping, it is ensured that each of the K maintenance areas can obtain a corresponding regional degradation curve based on its specific environment and repair history.

[0063] Based on this, the hydraulic gradient (i.e., the pressure and velocity distribution of water flow) is used as a constraint to predict the transmission of degradation across the operational areas based on the degradation curves of K regions. The hydraulic gradient is a key factor influencing seepage and dissolution. Through fluid dynamics modeling or finite element analysis, the direction of the hydraulic gradient is used to predict how the dissolution effect is transmitted between the K operational areas. Specifically, the hydraulic gradient determines the flow direction and velocity of water in different regions, which directly affects the propagation path and rate of seepage and dissolution. Therefore, by combining the degradation curves of the K regions and the direction of the hydraulic gradient, a degradation transmission model is established to spatially predict the transmission of dissolution between regions. Based on these prediction results, a degradation prediction graph neural network is constructed. In this step, the K operational areas are considered as nodes in a graph, and the features of each node come from the degradation curve of that region. The graph neural network updates the state of each node by transmitting information through the connections between nodes (i.e., the influence relationship between adjacent regions). Using the graph neural network, the spatial dependencies between regions are captured, and the transmission process of the dissolution effect between different regions is accurately predicted.

[0064] Finally, by decomposing the degradation curves of K regions, K sample training datasets are obtained. These datasets are then used to tune and optimize the K material degradation prediction models for the K graph nodes in the degradation prediction graph neural network. This step decomposes the degradation curves of each of the K regions into multiple time-series data samples, each representing the dissolution characteristics of that region at different time points. Then, optimization algorithms (such as stochastic gradient descent or the Adam optimizer) are used to train and tune the parameters of each node in the graph neural network to ensure that the model can accurately predict the dissolution and degradation processes within the region. Through training and optimization, the parameters of the degradation prediction model for each node will be adjusted to improve prediction accuracy, enabling the graph neural network to better simulate the interactions between regions and the degradation propagation paths. This process completes the construction of the degradation prediction graph neural network.

[0065] Furthermore, in the method provided in the application embodiment, using the hydraulic gradient direction as a constraint, the degradation propagation prediction of the K maintenance areas is performed based on the degradation curves of the K regions, and the degradation prediction graph neural network is constructed based on the prediction results, further comprising:

[0066] K extreme values ​​of degradation extension distance are extracted from the degradation curves of the K regions; taking the regional center of the K maintenance regions as the degradation starting point and the hydraulic gradient direction as the degradation direction, degradation linkage fitting is performed on the K maintenance regions in the high-incidence dissolution area based on the K extreme values ​​of degradation extension distance to obtain K degradation impact area information; K graph nodes are constructed according to the K maintenance regions, and the K graph nodes are topologically connected according to the K degradation impact area information to complete the initialization of the degradation prediction graph neural network.

[0067] In this embodiment, K extreme values ​​of degradation propagation distance are first extracted from the degradation curves of K regions. To accomplish this, a peak detection algorithm or local extremum analysis method is used to extract the maximum degradation propagation distance from the degradation curve of each region by calculating the maximum slope change or the farthest propagation position of each degradation curve. These extreme values ​​of degradation propagation distance represent the spatial range of the dissolution effect in each region, reflecting the farthest distance the degradation extends outward from the center of the region. Each region's degradation curve contains a corresponding extreme value of propagation, which provides important spatial data for subsequent degradation propagation prediction.

[0068] Next, taking the regional centers of K maintenance areas as the starting points of degradation and the hydraulic gradient direction as the degradation direction, based on the K extreme values ​​of degradation propagation distance, a degradation linkage fitting is performed on the K maintenance areas in the high-incidence area of ​​dissolution to obtain information on the K degradation-affected regions. In this process, a fluid dynamics model or diffusion equation simulation is used, taking the regional center of each maintenance area as the starting point of dissolution, and combining the hydraulic gradient (i.e., the pressure and velocity direction of the water flow) to simulate the propagation of the dissolution effect. The hydraulic gradient affects the transmission of water flow between different areas, thus determining the propagation path of the dissolution effect. Based on the extreme value of degradation propagation distance for each area, the fluid dynamics model predicts the spread of dissolution from the degradation center to adjacent areas, ultimately generating information on the K degradation-affected regions.

[0069] Then, K graph nodes are constructed based on K maintenance areas, and these K graph nodes are topologically connected according to the information of K degradation-affected areas, completing the initialization of the degradation prediction graph neural network. In this step, a graph neural network (GNN) construction method is used, treating each maintenance area as a node in the graph. Each node represents a maintenance area, and the node's feature data comes from the degradation information of that area (e.g., permeation depth, porosity, and ultrasonic attenuation). Based on the information of the K degradation-affected areas, a topological connection method (such as a connection method based on spatial adjacency or similarity metrics) is used to represent the relationships between nodes as edges of the graph. For example, if two maintenance areas have similar dissolution effects or are spatially adjacent, they will be connected. These topological structures define the relationships between nodes in the graph neural network, forming a complete network structure. Finally, through this process, the initialization of the degradation prediction graph neural network is completed, ensuring that the dissolution effects and their impacts between areas can be effectively modeled and propagated in the graph structure.

[0070] Step S500: By inputting the K time-encoded vectors of the K maintenance areas into the degradation prediction graph neural network, the permeation and corrosion degradation of cement-based materials in the high-incidence corrosion areas is predicted, and real-time maintenance tasks are output.

[0071] In this embodiment, based on a preset maintenance interval, K time-coded vectors for K maintenance areas are first updated. These time-coded vectors are then mapped and input into K material degradation prediction models within a degradation prediction graph neural network to obtain K real-time degradation trend features. These real-time degradation trend features reflect the erosion degradation of each area over a future period. Next, the maintenance needs are determined by combining the K real-time degradation trend features and the regional boundaries of the K maintenance areas, ultimately outputting real-time maintenance tasks. These maintenance tasks include repair or maintenance action plans based on the prediction results to ensure the long-term stability of cement-based materials.

[0072] Furthermore, in the method provided in the application embodiment, by inputting the K time-encoded vectors of the K maintenance areas into the degradation prediction graph neural network to predict the penetration and degradation of cement-based materials in the high-incidence area of ​​dissolution, and outputting real-time maintenance tasks, it further includes:

[0073] The K time-encoded vectors of the K maintenance regions are updated based on a preset maintenance interval; the K time-encoded vectors are mapped and input into the K material degradation prediction models in the degradation prediction graph neural network to obtain K real-time degradation trend features; maintenance requirements are judged based on the K real-time degradation trend features and the regional boundaries of the K maintenance regions, and the real-time maintenance task is output.

[0074] In this embodiment, K time-coded vectors for K maintenance areas are updated based on preset maintenance intervals. To this end, a time-series update method is first used, combining the repair and maintenance records of each maintenance area with the preset maintenance intervals (e.g., monthly, quarterly), to update the time-coded vectors for each area, resulting in K time-coded vectors. These time-coded vectors represent the repair history and future corrosion degradation predictions of the area. This update step ensures that the time information for each of the K maintenance areas is consistent with the actual repair activities in terms of time sequence, and provides an accurate time reference for subsequent corrosion degradation predictions.

[0075] K time-encoded vectors are mapped and input into K material degradation prediction models within a degradation prediction graph neural network to obtain K real-time degradation trend features. This step uses a graph neural network (GNN) for processing, inputting the time-encoded vector of each maintenance area as a node feature into the GNN. At this stage, the GNN analyzes the interactions between nodes and their temporal characteristics to calculate the real-time degradation trend features for each area, resulting in K real-time degradation trend features. These trend features include the rate of increase in permeability depth, the rate of increase in porosity, and the rate of increase in ultrasonic attenuation, reflecting the material degradation process of each of the K areas over a future time period and providing real-time data support for subsequent maintenance decisions.

[0076] Next, maintenance requirements are determined based on K real-time degradation trend characteristics and the regional boundaries of K maintenance areas. This step uses a Spatial Analysis and Decision Support System (DSS) approach to compare the real-time degradation trend characteristics of each of the K maintenance areas with the regional boundaries to determine which areas require repair or reinforcement. When the corrosion degradation characteristics of a certain area exceed a set threshold, it indicates that the area has a high risk and needs priority maintenance. Through this analysis, the maintenance requirements of the K areas are obtained, i.e., whether repair or reinforcement tasks need to be performed.

[0077] Finally, real-time maintenance tasks are output. This step, through an automated scheduling system and task allocation algorithm, transforms the results determined based on maintenance needs into specific maintenance task instructions, including repair priority, required materials, operation time, and resource configuration.

[0078] In summary, the embodiments of this application have at least the following technical effects:

[0079] This application, after locally accessing the operation and maintenance information of a hydropower plant dam with a high incidence of karst erosion, collects microstructural characterization data based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, wherein each of the K multi-scale characterizations is identified by K sample metadata. The K operation and maintenance areas are aggregated based on the K sample metadata to obtain M groups of operation and maintenance areas. After dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, network matching of permeation and karst degradation features is performed to obtain M sample time-series characterizations and M sample time-series degradation features. Based on the M sample time-series characterizations and M sample time-series degradation features, degradation propagation prediction is performed on the K operation and maintenance areas to construct a degradation prediction graph neural network. By inputting the K time-encoded vectors of the K operation and maintenance areas into the degradation prediction graph neural network, permeation and karst degradation prediction of cement-based materials in the high-incidence karst erosion areas is performed, and real-time operation and maintenance tasks are output. This invention addresses the technical problem of low accuracy in predicting the penetration and corrosion deterioration of cement-based materials in existing technologies. By combining multi-scale characterization and acquisition, network matching of penetration and corrosion deterioration features, and graph neural network prediction, it achieves the technical effect of improving the accuracy of predicting the deterioration of cement-based materials.

[0080] Example 2, based on the same inventive concept as the deep learning-based cement-based material penetration and corrosion degradation analysis method in the previous examples, such as... Figure 2 As shown, this application provides a deep learning-based system for analyzing the penetration and corrosion degradation of cement-based materials. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0081] The microstructure characterization acquisition module 11 is used to collect microstructure characterization data based on the operation and maintenance information of the high-incidence area of ​​hydropower plant dam after calling the operation and maintenance information locally, and obtain K multi-scale characterizations of K operation and maintenance areas, wherein the K multi-scale characterizations are identified by K sample metadata; the operation and maintenance area determination module 12 is used to aggregate the K operation and maintenance areas according to the K sample metadata to obtain M groups of operation and maintenance areas; the matching module 13 is used to perform network matching of permeation and erosion degradation features after dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, and obtain M sample time-series characterizations and M sample time-series degradation features; the network construction module 14 is used to perform degradation propagation prediction of the K operation and maintenance areas based on the M sample time-series characterizations and M sample time-series degradation features, so as to construct a degradation prediction graph neural network; the degradation prediction module 15 is used to input the K time-encoded vectors of the K operation and maintenance areas into the degradation prediction graph neural network to perform permeation and erosion degradation prediction of cement-based materials in the high-incidence area of ​​erosion, and output real-time operation and maintenance tasks.

[0082] Furthermore, the system is also used to implement the following functions:

[0083] Based on the time-series representations and degradation characteristics of the M samples, degradation trend fitting is performed to output M material degradation curves. According to the M groups of operation and maintenance areas, the M material degradation curves are mapped and assigned to the K operation and maintenance areas to generate K regional degradation curves. Using the hydraulic gradient direction as a constraint, degradation propagation prediction is performed on the K regional degradation curves, and the degradation prediction graph neural network is constructed based on the prediction results. The K regional degradation curves are decomposed to obtain a K sample training dataset, and the parameters of the K material degradation prediction model for the K graph nodes in the degradation prediction graph neural network are tuned and optimized.

[0084] Furthermore, the system is also used to implement the following functions:

[0085] The K time-encoded vectors of the K maintenance regions are updated based on a preset maintenance interval; the K time-encoded vectors are mapped and input into the K material degradation prediction models in the degradation prediction graph neural network to obtain K real-time degradation trend features; maintenance requirements are judged based on the K real-time degradation trend features and the regional boundaries of the K maintenance regions, and the real-time maintenance task is output.

[0086] Furthermore, the system is also used to implement the following functions:

[0087] By pre-embedding a distributed optical fiber sensor network in the hydropower plant dam, distributed time-series seepage velocities are extracted. Using seepage velocity thresholds and duration of exceeding the threshold as dual constraints, the distributed time-series seepage velocities are traversed to locate the spatial distribution map of high seepage areas. Local dam operation and maintenance records are used to perform defect spatial density analysis and generate a defect density heat map. After spatially overlaying the spatial distribution map of high seepage areas and the defect density heat map, multi-criteria decision fusion analysis is performed to locate the high-incidence areas of dissolution.

[0088] Furthermore, the system is also used to implement the following functions:

[0089] Based on the maintenance information, K maintenance areas are equidistantly divided in the high-incidence area of ​​dissolution. Core samples are drilled from the K maintenance areas to obtain K permeable dissolution core samples. The K chloride ion penetration depths of the K permeable dissolution core samples are measured using a colorimetric method as K macroscopic scale characteristics. The K maintenance areas are scanned by X-ray tomography to obtain K interconnected porosities as K millimeter-scale characteristics. Based on the ultrasonic longitudinal wave propagation characteristics of the K maintenance areas, K relative attenuation rates are calculated and output as K micrometer-scale features. The K macroscopic scale characteristics, K millimeter-scale characteristics, and K micrometer-scale features are bound and stored for maintenance areas to obtain the K multi-scale characteristics.

[0090] Furthermore, the system is also used to implement the following functions:

[0091] Based on the K maintenance areas, K sample metadata are retrieved from the maintenance information, wherein the sample metadata includes repair maintenance time nodes, repair material ratios, environmental pH values, and environmental hydraulic gradients; the consistency of the K sample metadata is determined based on the repair material ratios, environmental pH values, and environmental hydraulic gradients, so as to aggregate the K sample metadata to obtain M sets of sample metadata; the K maintenance areas are grouped into the M sets of maintenance areas based on the M sets of sample metadata.

[0092] Furthermore, the system is also used to implement the following functions:

[0093] The M groups of maintenance areas are divided into K multi-scale representations to obtain M groups of multi-scale representations. Based on the mapping relationship between the M groups of multi-scale representations and the M groups of sample metadata, the M groups of repair and maintenance time nodes are invoked. The degradation rate of the M groups of multi-scale representations is calculated based on the M groups of repair and maintenance time nodes to obtain M time-varying degradation rate sequences. The M time-varying degradation rate sequences are used to perform network matching of penetration and dissolution degradation features to obtain M sample time-series representations and M sample time-series degradation features.

[0094] Furthermore, the system is also used to implement the following functions:

[0095] The first group of repair and maintenance time nodes are arranged in ascending order of time to obtain the first repair and maintenance time series. Based on the first repair and maintenance time series, the first group of multi-scale representations is serialized to obtain the first multi-scale representation sequence. Based on the repair and maintenance interval of the first repair and maintenance time series, the degradation rate is calculated on the mapping of the first multi-scale representation sequence to obtain the first degradation rate time-varying sequence. Each degradation rate time-varying feature in the degradation rate time-varying sequence consists of the penetration depth growth rate, the connected pore growth rate, and the ultrasonic attenuation growth rate.

[0096] Furthermore, the system is also used to implement the following functions:

[0097] K extreme values ​​of degradation extension distance are extracted from the degradation curves of the K regions; taking the regional center of the K maintenance regions as the degradation starting point and the hydraulic gradient direction as the degradation direction, degradation linkage fitting is performed on the K maintenance regions in the high-incidence dissolution area based on the K extreme values ​​of degradation extension distance to obtain K degradation impact area information; K graph nodes are constructed according to the K maintenance regions, and the K graph nodes are topologically connected according to the K degradation impact area information to complete the initialization of the degradation prediction graph neural network.

[0098] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0099] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0100] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A deep learning-based method for analyzing the penetration and corrosion degradation of cement-based materials, characterized in that, The method includes: After accessing the operation and maintenance information of the hydropower plant dam high-incidence area of ​​karstification locally, microstructure characterization is collected based on the operation and maintenance information to obtain K multi-scale characterizations of K operation and maintenance areas, wherein the K multi-scale characterizations are identified by K sample metadata. Based on the metadata of the K samples, the K operation and maintenance regions are aggregated to obtain M groups of operation and maintenance regions; After dividing the K multi-scale characterizations according to the M groups of operation and maintenance areas, the network matching of penetration and corrosion degradation features is performed to obtain M sample time series characterizations and M sample time series degradation features. Based on the time-series representations of the M samples and the time-series degradation features of the M samples, degradation propagation prediction is performed for the K operation and maintenance areas to construct a degradation prediction graph neural network. By inputting the K time-encoded vectors of the K maintenance areas into the degradation prediction graph neural network, the permeation and corrosion degradation of cement-based materials in the high-incidence corrosion areas is predicted, and real-time maintenance tasks are output. Specifically, based on the time-series representations and time-series degradation features of the M samples, degradation propagation prediction is performed for the K maintenance areas to construct a degradation prediction graph neural network. The method includes: Based on the time-series characterization of the M samples and the time-series degradation characteristics of the M samples, degradation trend fitting is performed, and M material degradation curves are output. Based on the M groups of operation and maintenance areas, the M material degradation curves are mapped and assigned to the K operation and maintenance areas to generate K area degradation curves; Using the hydraulic gradient direction as a constraint, the degradation propagation of the K maintenance areas is predicted based on the degradation curves of the K areas, and the degradation prediction graph neural network is constructed based on the prediction results. The K region degradation curves are decomposed to obtain K sample training datasets, and the parameters of the K material degradation prediction model of the K graph nodes in the degradation prediction graph neural network are tuned and optimized. The method involves collecting microstructural characterizations based on the operational information to obtain K multi-scale characterizations of K operational areas. Based on the aforementioned maintenance information, the K maintenance zones are divided at equal intervals in the high-incidence area of ​​dissolution. Core samples were drilled from the K maintenance areas to obtain K permeable and dissolved core samples; The K chloride ion penetration depths of the K permeation and dissolution core samples were measured using a colorimetric method and used as K macroscopic scale characteristics; K interconnected porosities were obtained by X-ray tomography of the K maintenance areas, which served as K millimeter-scale characteristics. Based on the ultrasonic longitudinal wave propagation characteristics of the K maintenance areas, calculate and output K relative attenuation rates as K micrometer-scale features. The K macroscopic scale features, K millimeter scale features, and K micrometer scale features are bound and stored in the operation and maintenance area to obtain the K multi-scale features; The method involves aggregating the K operation and maintenance regions based on the K sample metadata to obtain M groups of operation and maintenance regions, and includes: Based on the K maintenance areas, K sample metadata are retrieved from the maintenance information, wherein the sample metadata includes repair maintenance time nodes, repair material ratios, environmental pH values, and environmental hydraulic gradients; The consistency of the metadata of the K samples is determined based on the repair material ratio, environmental pH value and environmental hydraulic gradient, so as to aggregate the metadata of the K samples to obtain M sets of sample metadata; Based on the metadata of the M groups of samples, the K operation and maintenance regions are grouped into the M groups of operation and maintenance regions; The method involves dividing the K multi-scale representations according to the M groups of operation and maintenance areas, performing network matching of penetration and corrosion degradation features to obtain M sample time-series representations and M sample time-series degradation features. Based on the M groups of operation and maintenance areas, the K multi-scale representations are divided to obtain the M groups of multi-scale representations; Based on the mapping relationship between the M-group multi-scale representations and the M-group sample metadata, the M-group repair and maintenance time nodes are invoked; Based on the M groups of repair and maintenance time nodes, the degradation rate of the M groups of multi-scale characterizations is calculated to obtain M time-varying degradation rate sequences. Using the M time-varying sequences of degradation rates, network matching of penetration and dissolution degradation features is performed to obtain M sample time-series characterizations and M sample time-series degradation features.

2. The method for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning as described in claim 1, characterized in that, By inputting the K time-encoded vectors of the K maintenance areas into the degradation prediction graph neural network, the method predicts the penetration and corrosion degradation of cement-based materials in the high-incidence corrosion areas and outputs real-time maintenance tasks. The method includes: The K time-coded vectors of the K maintenance areas are updated based on a preset maintenance interval; The K time-encoded vectors are mapped and input into the K material degradation prediction models in the degradation prediction graph neural network to obtain K real-time degradation trend features. Based on the K real-time degradation trend characteristics and the regional boundaries of the K maintenance areas, the maintenance requirements are determined, and the real-time maintenance tasks are output.

3. The method for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning as described in claim 1, characterized in that, Prior to retrieving operation and maintenance information from a hydropower plant dam prone to karstification locally, the method included: Distributed time-series seepage velocity is extracted by pre-burying a distributed optical fiber sensor network in the dam of a hydropower plant. Using the seepage velocity threshold and the duration of exceeding the threshold as dual constraints, the distributed time-series seepage velocities are traversed to locate the spatial distribution map of high seepage areas. Locally access dam operation and maintenance records to perform defect spatial density analysis and generate a defect density heatmap; After spatially overlaying the spatial distribution map of the high-permeability area and the defect density thermal map, a multi-criteria decision fusion analysis is performed to locate the high-incidence dissolution area.

4. The method for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning as described in claim 1, characterized in that, Based on the M groups of repair and maintenance time nodes, the degradation rate of the M groups of multi-scale characterizations is calculated to obtain M time-varying degradation rate sequences. The method includes: Arrange the first group of repair and maintenance time nodes in ascending order of time to obtain the first repair and maintenance time sequence; Based on the first repair and maintenance time series, the first set of multi-scale representations is serialized to obtain the first multi-scale representation sequence; Based on the repair and maintenance interval of the first repair and maintenance time series, the degradation rate is calculated on the first multi-scale characterization sequence mapping to obtain the first degradation rate time-varying sequence, wherein each degradation rate time-varying feature in the degradation rate time-varying sequence is composed of the penetration depth growth rate, the connected pore growth rate, and the ultrasonic attenuation growth rate.

5. The method for analyzing the penetration and corrosion degradation of cement-based materials based on deep learning as described in claim 1, characterized in that, Using the hydraulic gradient direction as a constraint, the degradation propagation of the K maintenance areas is predicted based on the degradation curves of the K regions, and a degradation prediction graph neural network is constructed based on the prediction results. The method includes: Extract K extreme values ​​of degradation extension distance from the K region degradation curves; Taking the regional center of the K maintenance areas as the starting point of deterioration and the hydraulic gradient direction as the deterioration direction, based on the K extreme values ​​of deterioration expansion distance, the K maintenance areas in the high-incidence dissolution area are subjected to deterioration linkage fitting to obtain information on the K deterioration-affected areas. The K graph nodes are constructed based on the K maintenance regions, and the K graph nodes are topologically connected according to the information of the K degradation-affected regions to complete the initialization of the degradation prediction graph neural network.

6. A deep learning-based system for analyzing the penetration and corrosion degradation of cement-based materials, characterized in that, The system includes: The microstructure characterization acquisition module is used to collect microstructure characterization based on the operation and maintenance information of the high-incidence area of ​​hydropower plant dam after calling the operation and maintenance information locally, and to obtain K multi-scale characterizations of K operation and maintenance areas, wherein the K multi-scale characterizations are identified by K sample metadata. The operation and maintenance area determination module is used to aggregate the K operation and maintenance areas based on the K sample metadata to obtain M groups of operation and maintenance areas; The matching module is used to perform network matching of penetration and corrosion degradation features after dividing the K multi-scale characteristics according to the M groups of operation and maintenance areas, so as to obtain M sample time-series characteristics and M sample time-series degradation features. The network construction module is used to predict the degradation propagation of the K operation and maintenance areas based on the time-series representations of the M samples and the time-series degradation features of the M samples, so as to construct a degradation prediction graph neural network. The degradation prediction module is used to predict the penetration and corrosion degradation of cement-based materials in the high-incidence corrosion areas by inputting the K time-encoded vectors of the K maintenance areas into the degradation prediction graph neural network, and output real-time maintenance tasks. The network building module is also used to execute: Based on the time-series characterization of the M samples and the time-series degradation characteristics of the M samples, degradation trend fitting is performed, and M material degradation curves are output. Based on the M groups of operation and maintenance areas, the M material degradation curves are mapped and assigned to the K operation and maintenance areas to generate K area degradation curves; Using the hydraulic gradient direction as a constraint, the degradation propagation of the K maintenance areas is predicted based on the degradation curves of the K areas, and the degradation prediction graph neural network is constructed based on the prediction results. The K region degradation curves are decomposed to obtain K sample training datasets, and the parameters of the K material degradation prediction model of the K graph nodes in the degradation prediction graph neural network are tuned and optimized. The microstructure characterization and acquisition module is also used to perform: Based on the aforementioned maintenance information, the K maintenance zones are divided at equal intervals in the high-incidence area of ​​dissolution. Core samples were drilled from the K maintenance areas to obtain K permeable and dissolved core samples; The K chloride ion penetration depths of the K permeation and dissolution core samples were measured using a colorimetric method and used as K macroscopic scale characteristics; K interconnected porosities were obtained by X-ray tomography of the K maintenance areas, which served as K millimeter-scale characteristics. Based on the ultrasonic longitudinal wave propagation characteristics of the K maintenance areas, calculate and output K relative attenuation rates as K micrometer-scale features. The K macroscopic scale features, K millimeter scale features, and K micrometer scale features are bound and stored in the operation and maintenance area to obtain the K multi-scale features; The operation and maintenance area determination module is also used to perform: Based on the K maintenance areas, K sample metadata are retrieved from the maintenance information, wherein the sample metadata includes repair maintenance time nodes, repair material ratios, environmental pH values, and environmental hydraulic gradients; The consistency of the metadata of the K samples is determined based on the repair material ratio, environmental pH value and environmental hydraulic gradient, so as to aggregate the metadata of the K samples to obtain M sets of sample metadata; Based on the metadata of the M groups of samples, the K operation and maintenance regions are grouped into the M groups of operation and maintenance regions; The matching module is also used to perform: Based on the M groups of operation and maintenance areas, the K multi-scale representations are divided to obtain the M groups of multi-scale representations; Based on the mapping relationship between the M-group multi-scale representations and the M-group sample metadata, the M-group repair and maintenance time nodes are invoked; Based on the M groups of repair and maintenance time nodes, the degradation rate of the M groups of multi-scale characterizations is calculated to obtain M time-varying degradation rate sequences. Using the M time-varying sequences of degradation rates, network matching of penetration and dissolution degradation features is performed to obtain M sample time-series characterizations and M sample time-series degradation features.

Images