Early warning method for erosion damage of concrete member influenced by multiple factors in salt-fresh water area

Abstract

The invention relates to the field of monitoring and early warning, in particular to a concrete member erosion damage early warning method influenced by multiple factors in a salt and fresh water area. Comprising the following steps: collecting environmental parameters of a concrete member in a salt-fresh water intersection area through a sensor network, constructing a corrosion data matrix, extracting a disturbance factor characteristic spectrum by adopting a tensor column block decomposition algorithm, and constructing a corrosion evolution trajectory model based on a heterogeneous graph neural network. A dynamic Bayesian network and Monte Carlo simulation are combined to predict a multi-stage corrosion risk, short-term, medium-term and long-term early warning indexes are generated, finally, graded alarm and protection strategy generation are realized through a fuzzy logic mechanism, and intelligent sensing, modeling and prediction of a corrosion process are realized. According to the method, intelligent sensing, modeling and prediction of the erosion process of the concrete member are realized, and the timeliness, accuracy and interpretability of erosion early warning of the concrete member are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of monitoring and early warning, and particularly relates to a method for early warning of erosion and damage of concrete members in a salt-fresh water region under the influence of multiple factors. BACKGROUND

[0002] Concrete structures are widely used as basic materials in coastal and hydraulic structures, such as port terminals, cross-sea bridges, dam gates and other facilities. However, in the salt-fresh water intersection region, due to the complex and changeable environmental medium, the concrete members are exposed to the natural environment with strong corrosion, volatility and interaction for a long time, which is prone to problems such as structural performance degradation, crack propagation and durability decline, seriously threatening the engineering safety and service life. The current problems are as follows: the existing methods mostly rely on static or single-point monitoring equipment, which is difficult to comprehensively and dynamically collect the complex and changeable environmental factors in the salt-fresh water intersection region; the existing system usually lacks the ability of deep mining of environmental data and high-dimensional feature modeling, cannot effectively extract potential damage driving factors, lacks the ability of modeling the heterogeneous relationship between environmental disturbance factors, and is difficult to depict the spatial coupling and temporal evolution of the corrosion process; the existing methods mostly only provide threshold trigger type alarm based on a single index, which is difficult to meet the actual needs of fine maintenance management. SUMMARY

[0003] To solve the above problems, the present application provides a method for early warning of erosion and damage of concrete members in a salt-fresh water region under the influence of multiple factors, which solves the problem of how to break through the limitations of traditional monitoring methods in the complex and changeable corrosion environment of the salt-fresh water intersection region, and accurately realize the dynamic perception of multiple source environmental factors of concrete members, the coupling modeling of high-dimensional disturbance factors and the spatio-temporal evolution prediction of the corrosion process, thereby improving the timeliness, accuracy and interpretability of the erosion early warning of concrete members.

[0004] To achieve the above purpose, the technical scheme adopted by the present application is as follows: The method for early warning of erosion and damage of concrete members in a salt-fresh water region under the influence of multiple factors comprises the following steps: S1: collecting concrete member environmental parameter data in real time through a sensor network in the salt-fresh water intersection region, and constructing an environmental corrosion original data matrix; S2: based on the environmental corrosion original data matrix, using a multi-modal extraction algorithm based on tensor column block decomposition for dimension compression and potential coupling factor extraction, and constructing a disturbance factor feature spectrum; the disturbance factor feature spectrum includes chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity and cross-correlation degree of sea wind corrosion index; S3: Based on the disturbance factor feature graph, a heterogeneous graph neural network algorithm is used to construct a concrete member corrosion evolution trajectory model, different types of factors in the disturbance factor feature graph are taken as heterogeneous nodes, the coupling relationship between factors is constructed, and a risk situation graph is output; S4: Based on the risk situation graph, a dynamic Bayesian network is used for multi-stage corrosion evolution prediction, and a Monte Carlo simulation is used to calculate the corrosion-induced cracking time of the concrete cover and the degradation path of the bearing capacity under different environmental scenarios, and a multi-level warning index set including short-term warning level, medium-term risk index and long-term service life prediction is generated; S5: Based on the multi-level warning index set, a fuzzy logic warning decision mechanism is established, and when the prediction index exceeds the safety threshold, a hierarchical alarm is triggered and a protection decision scheme is generated.

[0005] Further, the concrete member environmental parameter data includes the chloride ion concentration, sulfate concentration, water temperature, pH value, flow rate, dry-wet cycle frequency, atmospheric humidity and sea wind corrosion index of the region.

[0006] Further, the step S2 includes the following steps: The environmental corrosion original data matrix is modeled as a three-order tensor, and a three-dimensional tensor representation structure is constructed with time series, spatial monitoring points and corrosion related parameters as input expression forms of multi-modal corrosion environment data; The three-dimensional tensor is divided into blocks according to column dimension, and the tensor is decomposed into several column sub-tensor blocks along the spatial dimension according to the spatial clustering results of the monitoring area, and the low-rank reconstruction and principal component extraction are performed on each sub-tensor block by the tensor column block decomposition method, and the disturbance principal component factors of the region under different time scales are extracted; Based on the extracted disturbance principal component factors, the high-order mutual information and frequency domain cross spectrum density between each disturbance factor are calculated, and a disturbance factor feature graph is constructed; The nodes in the disturbance factor feature graph include chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity and sea wind corrosion index cross correlation degree, and the edge weight of the graph represents the degree of cooperative change between different factors; The joint identification criterion of disturbance factor change rate and mutation intensity is used to analyze the entropy increase of the node disturbance value in the feature graph, identify the potential high-risk factor combination area, and take the result as the input basis for constructing the corrosion evolution trajectory model.

[0007] Further, the joint identification criterion is specifically determined by comprehensively considering the disturbance factor change rate and the mutation intensity, the disturbance factor change rate is calculated based on the normalized first-order difference of the disturbance principal component in the adjacent time period, and the mutation intensity is extracted based on the joint of the sample entropy change trend and the frequency domain mutation point identification algorithm of the disturbance factor sequence.

[0008] Further, the step S3 comprises the following steps: Based on each type of disturbance factor in the disturbance factor feature map, a corresponding type of heterogeneous node set is constructed, and the high-order mutual information and frequency domain cross spectral density calculated between the disturbance factors are taken as the edge weight parameters in the heterogeneous graph to form an initial heterogeneous factor graph; Based on the heterogeneous factor graph, a heterogeneous graph neural network is used to construct a concrete member corrosion evolution trajectory model, the node evolution path of each disturbance factor is nonlinearly modeled, and key corrosion state features in the multi-stage evolution process are extracted; Through the feature dynamic embedding and edge coupling weight change of the heterogeneous factor in different time windows, a member-level corrosion state evolution sequence is output; The corrosion state evolution sequence is visualized as a corrosion evolution trajectory map, and the nodes of the map represent each evolution stage, and the path of the map represents the corrosion trend change trajectory driven by the factor.

[0009] Further, the formula of the concrete member corrosion evolution trajectory model is as follows: ; Wherein, represents the evolution state vector of the disturbance factor node v at the t+1 layer of the heterogeneous graph neural network; represents the state vector of the disturbance factor node u which has a coupling relationship with node v at the t layer; represents all adjacent node sets that have a relationship with node v under edge type r; R represents the total edge type set contained in the heterogeneous graph, including the high-order mutual information coupling relationship between the disturbance factor pairs; and respectively represent the degree of nodes v and u under edge type r; represents feature transformation and representation enhancement of the relationship between disturbance factor nodes; represents the bias vector corresponding to the edge type r; represents the activation function.

[0010] Further, the step S4 comprises the following steps: Based on the corrosion evolution trajectory map, the state evolution sequence of each disturbance factor in the time dimension is extracted, and the corrosion state stage set is constructed based on the node state transition path, and is divided into the evolution stages of mild corrosion, crack initiation, protection layer damage and structural performance degradation; A dynamic Bayesian network model is constructed with the corrosion stage evolution relationship driven by the disturbance factor as the node, and the node embedding vector evolution trend of the disturbance factor is used as the input feature to conditionally model the transition probability between stages; The Monte Carlo algorithm is used to simulate and calculate diversified evolution paths of disturbance factors under different environmental condition combinations, and the dynamic Bayesian network model is sampled for multiple rounds to obtain the probability density distribution of rust expansion cracking time and the degradation curve of the bearing capacity of the component. According to the multiple Monte Carlo simulation results, the quantile intervals are extracted, and a multi-level early warning index set is constructed, wherein the short-term early warning level is based on the mutation probability of the disturbance factor and the change degree of the Bayesian node edge weight, the medium-term risk index is weighted calculated according to the cumulative transition probability in the corrosion state transition chain, and the long-term service life prediction value is obtained by fitting the time decay function of the structural bearing capacity.

[0011] Further, the formula of the Monte Carlo algorithm is as follows:

[0012] Among them, T k represents the predicted component rust expansion or cracking time in the kth Monte Carlo simulation; q represents the current time step; n represents the number of disturbance factors; and S represents the number of corrosion evolution stages. wi,s represents the time-varying disturbance weight of the disturbance factor i at stage s. β i,s represents the risk driving strength of the disturbance factor i to stage s in the kth sampling; δ i,s represents the cumulative damage amount of the disturbance factor i at stage s. η q represents the dynamic environmental tolerance factor under the current environmental combination. λ s represents the risk penalty coefficient of stage s. μ represents the total risk tolerance threshold of the structural durability.

[0013] Further, the determination condition that the prediction index exceeds the safety threshold is determined based on a multi-dimensional index fuzzy fusion mechanism, including the following steps: Set the fuzzy interval set corresponding to the short-term early warning level, the medium-term risk index and the long-term life prediction value, and respectively map to the low risk, the alert, the danger and the limit four types of safety levels. Based on the fuzzy membership function, the fuzzy deviation degree of each index relative to the corresponding safety threshold is calculated. A fuzzy rule base is constructed, the membership strength and weight coefficient of each type of index are fused, fuzzy reasoning is performed, and the comprehensive risk level is output. When the fuzzy reasoning result level exceeds the specified risk level threshold, the alarm mechanism is triggered.

[0014] Further, the protection decision scheme includes concrete component corrosion protection suggestions, maintenance priority sorting and sensor layout optimization.

[0015] The beneficial effects of the present application are: The present application can dynamically collect various corrosion environment parameters including chloride ion concentration, pH value, dry-wet cycle, etc. by deploying a sensing network in the salt-fresh water intersection area, construct a high-dimensional original data matrix, provide a solid data foundation for subsequent analysis, and improve the response capability of the early warning system to the changes of complex environmental factors. The multi-modal extraction algorithm based on tensor column block decomposition is introduced, which effectively compresses the multi-dimensional data dimension, extracts the potential coupled disturbance factors, and the constructed disturbance factor feature spectrum can accurately represent the key environmental variables affecting the corrosion evolution of concrete and their cross coupling effect. Through the heterogeneous graph neural network algorithm, the heterogeneous association between disturbance factors is simulated by using the graph structure, the corrosion evolution trajectory model is constructed and the risk situation spectrum is output, realizing the visualization and traceability of the corrosion development process in the space and time dimensions. Fusion of dynamic Bayesian network and Monte Carlo simulation method, a multi-stage corrosion evolution prediction model is established, which can not only evaluate the time sequence of rust expansion damage of concrete members under different working conditions, but also depict the degradation path of bearing capacity, support accurate life cycle management. Combined with the multi-level early warning index set and the fuzzy logic reasoning mechanism, not only can the corresponding risk alarm be triggered according to different evolution stages, but also the customized protection scheme can be dynamically generated according to the prediction results, effectively guiding the subsequent maintenance and reinforcement strategy, and improving the intelligent response and decision support capability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 FIG. 1 is a flowchart of the concrete member corrosion damage early warning method in the salt-fresh water area multi-factor influence provided by the present application.

[0017] Fig. 2 FIG. 3 is a flowchart of the specific steps of step S3 provided by an embodiment of the present application.

[0018] Fig. 3 FIG. 4 is a flowchart of the specific steps of step S4 provided by an embodiment of the present application. DETAILED DESCRIPTION

[0019] Referring to FIG. 1, the present application relates to a concrete member corrosion damage early warning method in a salt-fresh water area multi-factor influence. Figs. 1-3

[0020] EMBODIMENT

[0021] The concrete member corrosion damage early warning method in the salt-fresh water area multi-factor influence comprises the following steps: S1: Real-time collection of concrete member environment parameter data through the sensing network in the salt-fresh water intersection area, construction of environment corrosion original data matrix; the concrete member environment parameter data includes chloride ion concentration, sulfate concentration, water temperature, pH value, flow rate, dry-wet cycle frequency, atmospheric humidity and sea wind corrosion index in the region.

[0022] ​In one embodiment, in the salt and fresh water intersection area, concrete members including seawalls, bridge piers and port pile foundations are selected as monitoring objects. To achieve full-cycle, multi-dimensional corrosion environment perception, the specific implementation process is as follows: according to the geographical location of the concrete member, it is divided into three types of monitoring points of high tide zone, middle tide zone and low tide zone, to ensure that the environmental differences of different seawater erosion heights are fully covered. In each type of area, 3-5 monitoring points are arranged, and a plurality of multi-parameter sensors are arranged at each monitoring point; an embedded sensor module is arranged at a distance of 5 cm from the surface of the member at each monitoring point, to ensure that the data source is close to the real erosion surface.

[0023] The collected parameters are as follows: Chloride ion concentration and sulfate concentration: a potential ion selective electrode is used, and the detection hole is embedded in the member concrete protective layer through a closed epoxy cavity.

[0024] Water temperature and pH value: a composite electrode type sensor with automatic temperature compensation function is used, and is fixed on the surface layer of the member or in the surrounding water body.

[0025] Flow rate: an ultrasonic flow rate module is used, which is installed at the key section of the tidal flow channel to reflect the water flow intensity in real time.

[0026] Dry-wet cycle frequency: the dry-wet conversion frequency per unit time is recorded by an infrared humidity sensing film + timing control unit.

[0027] Atmospheric humidity: a micro-capacitive humidity sensor is selected, which is installed 1 meter above the member to exclude the direct influence of water.

[0028] Sea wind corrosion index: calculated according to wind speed (measured by an ultrasonic anemometer), wind direction and salt mist content in the air, the sensor is integrated on a special bracket at the windward opening of the seawall.

[0029] All collected parameters are constructed into a three-dimensional data matrix with member number as primary key and time as index: Dimension one: time (in units of hours or days); Dimension two: monitoring point (member + direction); Dimension three: environmental parameter type (a total of eight types: chloride ion, sulfate, water temperature, pH, flow rate, dry-wet frequency, humidity, and wind erosion index).

[0030] The constructed data matrix is regularly uploaded to the cloud platform and called by the subsequent tensor decomposition and graph neural modeling modules.

[0031] S2: based on the environmental corrosion original data matrix, using a multi-modal extraction algorithm based on tensor column block decomposition, dimension compression and latent coupling factor extraction, and constructing a disturbance factor characteristic map; the disturbance factor characteristic map includes the cross correlation of chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity and sea wind corrosion index; wherein the step S2 comprises the following steps: The environmental corrosion original data matrix is modeled as a three-order tensor, and a three-dimensional tensor representation structure is constructed with time series, spatial monitoring points and corrosion related parameters as the input expression form of multi-modal corrosion environment data. Specifically, after obtaining the environmental corrosion parameter data collected by multiple monitoring points in the salt and fresh water intersection area, the data is organized into a data tensor with three dimensions: Time dimension: selecting a rolling time window of nearly 90 days with an hour sampling period; Spatial dimension: corresponding to multiple monitoring points arranged along the bridge pier, wharf and other structures; Parameter dimension: including chloride ion concentration, pH value, water temperature, flow rate, dry-wet frequency, sea wind index and other corrosion related parameters.

[0032] The tensor not only integrates time evolution information, but also contains spatial distribution and parameter coupling characteristics, which is the basic input form for multi-modal modeling.

[0033] The three-dimensional tensor is divided into blocks according to the column dimension, and the tensor is decomposed into several column sub-tensor blocks along the spatial dimension according to the spatial clustering results of the monitoring area. Each sub-tensor block is reconstructed by low rank and principal component extraction through the tensor column block decomposition method, and the disturbance principal component factors of the area at different time scales are extracted; Specifically, after the three-dimensional tensor is constructed, the column block and principal component disturbance factor extraction stage is entered, and the operation process is as follows: The monitoring points are clustered using spatial coordinates and structural properties, such as grouping based on component location (upper structure / pier / column / foundation), distance from tide line, and wind erosion exposure level; each class is regarded as a structural "functional area", such as: pier column area above tide; near-sea foundation area; windward side structure; then the original tensor is divided into column blocks along the spatial dimension, forming several sub-tensor blocks (such as each block dimension: time × parameter × number of monitoring points in the area).

[0034] For each sub-tensor block, the tensor column block decomposition technology (such as CP-ALS algorithm or Tucker model combined with column direction compression) is used to compress the original dimension and eliminate noise; the processing result retains the main components and reduces environmental redundant interference, for example, parameters with insignificant fluctuations or noise dominance will be weakened.

[0035] For each sub-block, the system extracts the main disturbance factors in the time dimension to identify which parameters have regular abnormal fluctuations within a certain time period. The extracted factors include: frequency-enhanced disturbance (such as rapid oscillation of pH value); trend deviation disturbance (such as long-term increase in chloride ion concentration); sudden peak disturbance (such as sharp change in sea wind erosion index before and after a typhoon).

[0036] Based on the extracted disturbance principal component factors, the high-order mutual information and frequency domain cross spectral density between each disturbance factor are calculated to construct a disturbance factor feature graph. The nodes in the graph include chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity, and sea wind erosion index cross correlation degree. The edge weight of the graph represents the degree of cooperative change between different factors. It should be noted that after obtaining the multiple regional disturbance principal components, the system converts them into a structured graph for depicting the mutual influence and joint pattern of multiple factors. Each disturbance principal component corresponds to a node in the graph. The nodes are classified into the following four categories: Chloride ion flux disturbance node: represents the migration instability in the salt intrusion process; pH value fluctuation frequency node: measures the dynamic change of the local alkaline environment; Dry-wet cycle linkage intensity node: represents the coupling degree of water exchange and microcrack dynamics; Sea wind erosion index cross correlation degree node: comprehensively reflects the combined effect of wind speed, wind direction, and salt-containing aerosol intensity on other factors.

[0037] An edge is established between each two disturbance factors according to their linkage degree in the time series. The weight of the edge is derived from: High-order mutual information: reveals strong nonlinear correlation; Frequency domain cross spectral density: identifies the consistency or coupling of frequency components; For example, if the chloride ion disturbance and dry-wet cycle have strong spectral resonance within 1-2 day cycles, the edge weight is set to high. If the pH value fluctuation and sea wind erosion change frequency are completely mismatched, the edge weight tends to zero. The system finally forms a weighted undirected graph, with each region corresponding to a disturbance factor network.

[0038] Using the joint identification criterion of disturbance factor change rate and mutation intensity, the entropy increase analysis is performed on the node disturbance values in the feature graph to identify potential high-risk factor combination regions. The results are used as input basis for constructing the corrosion evolution trajectory model.

[0039] The joint identification criterion specifically determines by comprehensively combining the disturbance factor change rate and the mutation intensity, the disturbance factor change rate is calculated based on the normalized first-order difference of the disturbance principal component in adjacent time periods, and the mutation intensity is extracted based on the sample entropy change trend of the disturbance factor sequence and the frequency domain mutation point identification algorithm.

[0040] Specifically, after the atlas is constructed, to identify potential high-risk points, the system further implements a joint identification criterion.

[0041] (1) Disturbance factor change rate calculation For each disturbance principal component time series, perform first-order difference normalization processing on an hourly and daily scale; calculate the disturbance value change amplitude in each time period to identify factors with high change rates; for example, during a humid and hot period, the short-term change of chloride ion flux may increase sharply, i.e., marked as a "high change rate factor", and the calculation formula is as follows:

[0042] wherein, represents the normalized change rate of the i-th disturbance factor at time step k; represents the principal component value of the i-th disturbance factor (such as chloride ion flux disturbance, pH fluctuation amplitude, etc.); represents the principal component value of the disturbance factor at the previous time step; represents a very small positive number to prevent the denominator from being 0.

[0043] (2) Mutation intensity detection Sample entropy trend analysis is used: if the complexity of the disturbance sequence increases significantly in a short period of time, it indicates that the system state is mutating; joint use of frequency domain mutation identification techniques (such as wavelet packet reconstruction difference identification or short-time Fourier window difference method) to identify "jump points" in the frequency domain; if the dry-wet cycle frequency suddenly appears a new high-frequency component in a certain period of time, it is determined to be a mutation, and the sample entropy growth rate calculation formula is as follows:

[0044] wherein, represents the sample entropy growth rate of the i-th disturbance factor; represents the sample entropy of the disturbance factor i at the current time step k; represents the sample entropy at the previous time step; represents a very small positive number to prevent the denominator from being 0.

[0045] The frequency domain mutation intensity estimation formula is as follows:

[0046] wherein, represents the frequency domain mutation intensity of the i-th disturbance factor at time step k; represents the frequency domain energy of the perturbation factor i at time step k and frequency f; represents a set of frequencies.

[0047] (3) Joint criterion determination Set a double-threshold determination mechanism: if the change rate of a certain perturbation node ≥ threshold 1 and the mutation intensity ≥ threshold 2, the node is identified as a “high-risk perturbation factor”; the system further identifies the local subgraph formed by the high-risk factors and marks it as a “high-risk perturbation combination area”; such an area is often a sensitive area where corrosion acceleration occurs, and is the key input target of the S3 corrosion trajectory model.

[0048] S3: Based on the perturbation factor characteristic map, a heterogeneous graph neural network algorithm is used to construct a concrete member corrosion evolution trajectory model, different types of factors in the perturbation factor characteristic map are used as heterogeneous nodes, the coupling relationship between the nodes is constructed, and a risk situation map is output; The step S3 includes the following steps: Based on each type of perturbation factor in the perturbation factor characteristic map, a set of heterogeneous nodes of the corresponding type is constructed, and the high-order mutual information and frequency domain cross-spectral density calculated between the perturbation factors are used as the edge weight parameters in the heterogeneous graph to form an initial heterogeneous factor graph; It should be noted that the perturbation factor characteristic map has been generated in step S2, and the map contains the following four types of perturbation factor nodes: C-type node: Chloride ion flux perturbation (representing corrosion medium permeability) P-type node: pH value fluctuation frequency (representing the degree of alkaline protection failure) D-type node: Dry-wet cycle linkage intensity (representing environmental driving period) W-type node: Cross-correlation degree of sea wind corrosion index (representing external wind erosion load) Each type corresponds to a node type in the heterogeneous graph, and the system constructs the initial heterogeneous graph structure according to the following mechanism: the original nodes are retained in the graph, but they are classified according to the above types to form four heterogeneous node sets; the edges between the nodes are generated according to the factor relationship extracted in the early stage: if the chloride ion and the dry-wet cycle node frequently fluctuate together, an edge is created; if the pH value and the sea wind corrosion index exist cross-frequency mutual driving, the edge weight is higher; each edge is attached with a coupling strength label, which can be weighted based on the factor linkage matrix obtained by internal preprocessing of the system, and is used for information transmission constraint of graph neural network.

[0049] Based on the heterogeneous factor graph, a heterogeneous graph neural network is used to construct a concrete member corrosion evolution trajectory model, to nonlinearly model the node evolution path of each perturbation factor, and to extract key corrosion state features in the multi-stage evolution process; Specifically, the constructed heterogeneous graph is sent to a modeling engine deploying a heterogeneous graph neural network (HGNN) model for modeling the nonlinear path of the evolution of the corrosion of the component over time, mainly including the following steps: The system segments the historical disturbance values of each node by time window, such as setting 24 hours as a window; time slice indexes are added to the graph, so that the state of the same node in different time windows becomes a "time slice node"; in each time slice, the original heterogeneous structure between nodes is still retained, forming a time series of heterogeneous graphs.

[0050] The dynamic change vector of each disturbance factor is input into the network and encoded as the feature representation of the corresponding node; different types of nodes use different embedding mapping networks (such as multi-channel encoders) to achieve type-aware feature extraction; the network automatically learns the influence weight of each type of factor on the corrosion evolution, such as the C-type node being more sensitive to short-term corrosion and the W-type node affecting the medium and long-term trend.

[0051] The heterogeneous GNN distinguishes by edge type to realize edge weight guided message passing; in each time window, the network aggregates the disturbance features of adjacent factors to generate the "corrosion state vector" of the current stage; the network further tracks the evolution trajectory of these states between different time slices to form a complete corrosion evolution path model.

[0052] Through the dynamic embedding of heterogeneous factors in different time windows and the change of edge coupling weights, the corrosion state evolution sequence of the component level is output; Specifically, after the network model is trained, it can output the corrosion state sequence of each monitored component within a period of time: Each time slice outputs a state identifier, such as "mild corrosion", "moderate cracking", "crack development acceleration", and "bearing capacity criticality"; The system can identify key turning points, such as the state of a node jumping from "stable" to "rapid deterioration"; The key driving factors and their action windows are recorded in the state sequence, for example: the "corrosion jump" is mainly affected by the combined influence of the sudden increase in sea wind frequency and the sudden drop in pH.

[0053] The corrosion state evolution sequence is visualized as a corrosion evolution trajectory atlas, with graph nodes representing each evolution stage and graph paths representing the corrosion trend change trajectory driven by factors.

[0054] It should be noted that the above corrosion state sequence is finally visualized as a structured "corrosion evolution trajectory atlas", realizing the following information presentation: Each node represents an evolution stage (such as "light corrosion", "moderate corrosion", "cracking", "rust expansion and spalling", etc.); the color of the node represents the corrosion level (green -> yellow -> red); the edge represents the transition path of the corrosion state, and the dominant disturbance factor and its key time window are marked on the edge.

[0055] Each atlas corresponds to a component; the trajectory atlas of multiple components can be superimposed in the same pier / breakwater structure three-dimensional model; the atlas can also be exported as an animation for construction site personnel to make corrosion trend prediction and protection timing decision.

[0056] Further, the formula of the concrete component corrosion evolution trajectory model is as follows: ; Among them, represents the evolution state vector of the disturbance factor node v at the t+1 layer of the heterogeneous graph neural network; represents the state vector of the disturbance factor node u which has a coupling relationship with node v at the t layer; represents the set of all adjacent nodes that have a relationship with node v under edge type r; R represents the set of all edge types contained in the heterogeneous graph, including the high-order mutual information coupling relationship between disturbance factor pairs; and respectively represent the degree of nodes v and u under edge type r; represents the feature transformation and representation enhancement of the relationship between disturbance factor nodes; represents the bias vector corresponding to the edge type r; represents the activation function.

[0057] S4: Based on the risk situation atlas, multi-stage corrosion evolution prediction is carried out by using dynamic Bayesian network, and the rust expansion and cracking time and bearing capacity degradation path of the concrete protective layer under different environmental scenarios are calculated by combining Monte Carlo simulation, to generate a multi-level early warning index set including short-term warning level, medium-term risk index and long-term service life prediction; Among them, the step S4 includes the following steps: Based on the corrosion evolution trajectory atlas, the state evolution sequence of each disturbance factor in the time dimension is extracted, and the corrosion state stage set is constructed based on the node state transition path, which is divided into evolution stages of light corrosion, crack initiation, protective layer damage and structural performance degradation; Specifically, the time slice sequence (such as 1 state node every 24 hours) is extracted from the evolution trajectory of each disturbance node; based on the state characteristic value and the transition mode, the time sequence is mapped to four standard corrosion stages: Light corrosion stage: the disturbance factor fluctuates stably, and no obvious structural change is seen; Crack initiation stage: the frequency of chloride ion disturbance and pH fluctuation increases significantly, and fine cracks appear on the surface of the component; Protection layer destruction stage: dry-wet cycle linkage is enhanced, concrete protection layer fails, and steel bars begin to be exposed; Structural performance degradation stage: the corrosion area expands, concrete spalls, and the load-carrying capacity of the component decreases.

[0058] A dynamic Bayesian network model is constructed based on the evolution relationship of the corrosion stages driven by disturbance factors, and the evolution trend of the node embedding vector of the disturbance factors is used as the input feature to conditionally model the transition probability between stages. Specifically, the aforementioned four corrosion stages are regarded as state nodes in the Bayesian network; the evolution embedding vector of the disturbance factor (C-type, P-type, D-type, W-type) node is used as the model input; the embedding vector contains information such as recent fluctuation characteristics, cross-stage change trend, and frequency domain disturbance density of the disturbance factor; the network learns the conditional probability between the time variation pattern of the disturbance factor and the corrosion stage transition to form the transition probability matrix between stages; during the training process, the system can sample and expand the trajectories under different time periods and environmental scenarios to enhance the generalization ability of the model.

[0059] Monte Carlo algorithm is used to simulate the diversified evolution paths of disturbance factors under different environmental condition combinations, and the dynamic Bayesian network model is sampled multiple times to obtain the probability density distribution of the rust expansion and cracking time and the load-carrying capacity degradation curve of the component; Specifically, for different environmental scenarios (such as "high humidity + strong wind", "weak alkali + slow drying" combination, etc.), the random fluctuation path of the disturbance factor is simulated; each simulation drives the Bayesian network to run once for corrosion state transition, generating a complete "stage path"; the simulation number can be set according to the demand (such as 1000 or 10000), which is used to build a sufficient corrosion evolution sample set; each simulation path contains: corrosion state transition time point; rust expansion and cracking time; time when the performance of the component decreases to the critical level; finally, a large number of path samples are obtained to support statistical analysis and early warning evaluation.

[0060] According to the results of multiple Monte Carlo simulations, the quantile intervals are extracted to construct a multi-level early warning index set, where the short-term early warning level is based on the mutation probability of the disturbance factor and the change degree of the Bayesian node edge weight, the medium-term risk index is calculated by weighting the cumulative transition probability in the corrosion state transition chain, and the long-term service life prediction value is obtained by fitting the time decay function of the structural load-carrying capacity.

[0061] In one embodiment, a large number of corrosion path results obtained by simulation are analyzed in layers to extract three-dimensional early warning indicators corresponding to short-term, medium-term, and long-term corrosion risks.

[0062] (1) Short-term early warning level: The mutation probability based on the disturbance factor (such as the probability of chloride disturbance rising within 24 hours > 80%); At the same time, the sensitivity of the edge weight change in the Bayesian model is considered, that is, the probability of transition in a certain stage rapidly rises within a continuous time period; Output such as: "high-level warning (orange)", "medium-level warning (yellow)", "no warning (green)" three types of labels; It is suitable for predicting whether a significant corrosion transition will occur in the next 1-3 days.

[0063] (2) Medium-term risk index: Calculate the cumulative transition probability from the current stage to the "protection layer damage" or "structural performance degradation" stage within the next 7-30 days; Weighted integration according to the category of disturbance factors to obtain the overall risk value (0-1 interval); used to determine the priority level of medium-term maintenance or on-site detection.

[0064] (3) Long-term service life prediction: Use the time point distribution of bearing capacity decline in the Monte Carlo sample to estimate the life probability interval (such as 25%, 50%, and 75% quantile points); Combine the component structure model to fit the "bearing capacity-time" decay curve; The output prediction values include: average remaining life (unit: years); the earliest expected time to reach the dangerous critical value (unit: days); long-term maintenance and replacement suggestion points.

[0065] Further, the formula of the Monte Carlo algorithm is as follows:

[0066] Wherein, tk represents the predicted component rust expansion or crack damage time in the kth Monte Carlo simulation; q represents the current time step; n represents the number of disturbance factors; S represents the number of corrosion evolution stages; wi,s (q) represents the time-varying disturbance weight of disturbance factor i at stage s; qi,s (q) represents the risk driving strength of disturbance factor i to stage s in the kth sampling; Di,s represents the cumulative damage amount of disturbance factor i at stage s; β(q) represents the dynamic environmental tolerance factor under the current environmental combination; αs represents the risk penalty coefficient of stage s; βs represents the total risk tolerance threshold of structural durability.

[0067] In the above formula, The calculation formula of is as follows:

[0068] wherein, denotes the time-varying weight of the disturbance factor i at stage s; denotes the normalization factor, making the sum of all factor weights equal to 1; denotes the rate of change adjustment coefficient; denotes the principal component evolution value of the disturbance factor i at time q; denotes the first-order difference of the disturbance factor principal component, reflecting the mutation trend; denotes the entropy sensitivity adjustment coefficient; denotes the sample entropy value of the disturbance factor i.

[0069] In the above formula, The calculation formula of is as follows:

[0070] wherein, denotes the cumulative damage amount of the disturbance factor i at stage s; denotes the disturbance intensity of the disturbance factor i at time ; denotes the stage indicator function, which has a value of 1 at stage s and 0 otherwise; denotes the risk gain function of the disturbance factor i; denotes the integral variable representing each time in the past.

[0071] S5: Based on the multi-level early warning index set, a fuzzy logic early warning decision mechanism is established, which triggers a graded alarm and generates a protection decision scheme when the prediction index exceeds the safety threshold; the protection decision scheme includes concrete member corrosion protection suggestions, maintenance priority ranking and sensor layout optimization.

[0072] wherein, the determination condition of the prediction index exceeding the safety threshold is determined based on a multi-dimensional index fuzzy fusion mechanism, including the following steps: Set the fuzzy interval set corresponding to the short-term early warning level, medium-term risk index and long-term life prediction value, respectively mapped to the low risk, alert, danger and limit four types of safety levels; Based on the fuzzy membership function, calculate the fuzzy deviation degree of each index relative to the corresponding safety threshold; Construct a fuzzy rule base, fuse the membership strength and weight coefficient of each type of index, and perform fuzzy reasoning to output the comprehensive risk level; When the fuzzy reasoning result level exceeds the specified risk level threshold, the alarm mechanism is triggered.

[0073] Specifically, after the alarm is triggered, the system automatically generates a customized protection scheme, which includes the following three parts: (1) Corrosion protection suggestions: If the risk level is "extreme", surface encapsulation, carbon fiber reinforcement or electrochemical corrosion protection are recommended; If the risk level is "dangerous", regular injection of sealant, increasing the thickness of the corrosion protection layer or using a migrating corrosion inhibitor are recommended; For mild components, it is recommended to strengthen ventilation, moisture removal and local maintenance management.

[0074] (2) Maintenance priority ranking: The system ranks according to the future deterioration rate of each component in the corrosion trajectory model, and schedules high-risk areas first; Automatic generation of maintenance task queue and recommended maintenance period (such as 15 days to complete reinforcement).

[0075] (3) Sensor layout optimization recommendations: According to the influence degree of each disturbance factor in the heterogeneous graph and the response intensity of the region, the current sensor layout is re-evaluated; For key high-risk cross-regions, it is recommended to increase the density of redundant nodes; At the same time, some weak response and information redundant points are marked for reduction and optimization.

[0076] In summary, the present application realizes in-situ monitoring of the real corrosion interface of concrete components by laying out multi-dimensional and multi-type sensors in high-tide, mid-tide and low-tide zones, ensuring the representativeness and real-time of the data. The environmental parameters cover the main corrosion driving factors in the marine environment, and an environmental corrosion data matrix with strong spatio-temporal continuity and comprehensive parameter coverage is constructed, providing a solid data foundation for subsequent analysis.

[0077] The present application uses tensor column block decomposition algorithm to process multi-modal corrosion data, and realizes regional compression of environmental information and extraction of main disturbance factors through spatial partitioning and column block modeling of component functional areas. This method not only retains key disturbance trends (such as chloride ion flux disturbance, pH fluctuation frequency, etc.), but also weakens irrelevant noise, improving the sensitivity and robustness of corrosion factor identification. Through mutual information and spectral analysis, the collaborative relationship between disturbance factors is established, and for the first time, multi-type corrosion environmental parameters are converted into heterogeneous graph structure representation, revealing the nonlinear correlation mechanism between wind erosion, hydrochemistry and periodic fluctuation, etc. and providing interpretable graph support for the identification of high-risk corrosion factor combinations.

[0078] The application is based on the change rate and mutation strength of the disturbance factor, constructs a joint identification criterion, and realizes accurate screening and regional labeling of high-risk disturbance factors through sample entropy growth rate and frequency domain mutation identification technology, significantly improving the focusing efficiency of corrosion modeling and the forward-looking of risk discrimination. A heterogeneous graph structure is constructed with the type of disturbance factor as the node, and HGNN is used to model the nonlinear trajectory of corrosion evolution state, realizing deep interaction modeling of different factor driven paths. The corrosion evolution sequence with clear stage division and clear state transition can be output, meeting the needs of visualization, interpretation and prediction of corrosion trend.

[0079] The application models the probability relationship of the disturbance factor to the corrosion stage transition through dynamic Bayesian network, and combines Monte Carlo simulation to simulate the corrosion evolution path under various environmental combinations, effectively predicting the rust expansion cracking time and bearing capacity decay process of the structure, providing scientific support for formulating long-term maintenance and reinforcement plan. A three-dimensional index set of short-term warning level, medium-term risk index and long-term life prediction is constructed, and various indexes are fused through fuzzy logic mechanism to realize intelligent grading judgment of corrosion risk. The system can automatically generate protection strategies and maintenance plans according to different risk levels, promoting the change of operation and maintenance from "passive response" to "active pre-control". Through the corrosion trajectory map and high-risk disturbance factor space mapping, the priority of the key maintenance area is realized. At the same time, the sensor network is optimized and reconstructed in redundancy combined with the results of heterogeneous graph analysis, improving the resource utilization and key area perception ability of the monitoring network.

[0080] The above embodiments only describe the preferred embodiments of the application, and do not limit the scope of the application. Without departing from the design spirit of the application, various modifications and improvements of the technical solutions of the application made by ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the application.

Claims

1. A method for early warning of the corrosion damage of concrete structures in brackish water areas under the influence of multiple factors, characterized in that, The method comprises the following steps: S1: Real-time collection of concrete member environmental parameter data through a sensing network in a saltwater-freshwater intersection area, and construction of an environmental corrosion original data matrix; S2: Based on the environmental corrosion original data matrix, a multi-modal extraction algorithm based on tensor column block decomposition is used for dimension compression and latent coupling factor extraction, and a disturbance factor feature map is constructed; the disturbance factor feature map includes chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity, and cross-correlation degree of sea wind corrosion index; S3: Based on the disturbance factor feature map, a heterogeneous graph neural network algorithm is used to construct a concrete member corrosion evolution trajectory model, different types of factors in the disturbance factor feature map are used as heterogeneous nodes, coupling relationship edges between factors are constructed, and a risk situation map is output; S4: Based on the risk situation map, a multi-stage corrosion evolution prediction is performed using a dynamic Bayesian network, and a Monte Carlo simulation is used to calculate the concrete cover rust expansion cracking time and bearing capacity degradation path under different environmental scenarios, and a multi-level early warning index set including short-term early warning level, medium-term risk index, and long-term service life prediction is generated; S5: Based on the multi-level early warning index set, a fuzzy logic early warning decision mechanism is established, and when the prediction index exceeds the safety threshold, a graded alarm is triggered and a protection decision scheme is generated.

2. The method of claim 1, wherein the method is characterized by: The concrete member environmental parameter data includes chloride ion concentration, sulfate concentration, water temperature, pH value, flow rate, dry-wet cycle frequency, atmospheric humidity, and sea wind corrosion index in the area.

3. The method of claim 1, wherein the method is characterized by: The step S2 comprises the following steps: The environmental corrosion original data matrix is modeled as a third-order tensor, and a three-dimensional tensor representation structure is constructed based on time series, spatial monitoring points, and corrosion-related parameters, serving as an input expression form of multi-modal corrosion environmental data; The three-dimensional tensor is divided into blocks according to column dimensions, and the tensor is decomposed into a plurality of column sub-tensor blocks along the spatial dimension based on the spatial clustering results of the monitoring area, and the low-rank reconstruction and principal component extraction of each sub-tensor block are performed through the tensor column block decomposition method, and the disturbance principal component factors of the area under different time scales are extracted; Based on the extracted disturbance principal component factors, the high-order mutual information and frequency domain cross-spectral density between each disturbance factor are calculated, and a disturbance factor feature map is constructed; the nodes in the disturbance factor feature map include chloride ion flux disturbance, pH value fluctuation frequency, dry-wet cycle linkage intensity, and cross-correlation degree of sea wind corrosion index, and the edge weight of the graph represents the degree of cooperative change between different factors; A joint identification criterion of disturbance factor change rate and mutation intensity is used to perform entropy analysis on the node disturbance values in the feature map, identify potential high-risk factor combination areas, and use the results as input basis for constructing a corrosion evolution trajectory model.

4. The method for early warning of concrete member erosion damage in brackish water area influenced by multiple factors according to claim 3, characterized in that, The joint identification criterion is specifically determined by comprehensively considering the disturbance factor change rate and the mutation intensity; the disturbance factor change rate is calculated based on the normalized first-order difference of the disturbance principal component in adjacent time periods; and the mutation intensity is extracted based on the joint use of the sample entropy change trend and the frequency domain mutation point identification algorithm of the disturbance factor sequence.

5. The method for early warning of concrete structure damage in brackish water area caused by multi-factor effect according to claim 1, characterized in that, The step S3 comprises the following steps: Based on each type of disturbance factor in the disturbance factor feature map, a corresponding type of heterogeneous node set is constructed, and the high-order mutual information and frequency domain cross spectral density calculated between the disturbance factors are taken as the edge weight parameters in the heterogeneous graph to form an initial heterogeneous factor graph; Based on the heterogeneous factor graph, a concrete member corrosion evolution trajectory model is constructed using a heterogeneous graph neural network to nonlinearly model the node evolution path of each disturbance factor and extract key corrosion state features in the multi-stage evolution process; Through the feature dynamic embedding and edge coupling weight change of the heterogeneous factor in different time windows, a member-level corrosion state evolution sequence is outputted; The corrosion state evolution sequence is visualized as a corrosion evolution trajectory map, with the nodes of the map representing each evolution stage and the paths of the map representing the corrosion trend change trajectory driven by the factors.

6. The method for early warning of concrete member erosion damage in brackish water area influenced by multiple factors according to claim 5, characterized in that, The formula of the concrete member corrosion evolution trajectory model is as follows: ； wherein, denotes the evolution state vector of the perturbation factor node v at the t+1th layer of the heterogeneous graph neural network; denotes the state vector of the perturbation factor node u which has a coupling relationship with the node v at the tth layer; denotes the set of all adjacent nodes which have a relationship with the node v under the edge type r; R denotes the set of all edge types contained in the heterogeneous graph, including the high-order mutual information coupling relationship between the perturbation factor pairs; and denote the degree of the nodes v and u under the edge type r, respectively; denotes the feature transformation and representation enhancement on the relationship between the perturbation factor nodes; denotes the bias vector corresponding to the edge type r; denotes the activation function.

7. The method for early warning of concrete structure damage in brackish water area caused by multi-factors according to claim 1, characterized in that, The step S4 includes the following steps: Based on the corrosion evolution trajectory map, the state evolution sequence of each disturbance factor in the time dimension is extracted, and a corrosion state stage set is constructed based on the node state transition path to divide the evolution stages into mild erosion, crack initiation, protective layer damage, and structural performance degradation; A dynamic Bayesian network model is constructed with the corrosion stage evolution relationship driven by the disturbance factors as the nodes, and the node embedding vector evolution trend of the disturbance factor is used as the input feature to conditionally model the transition probability between stages; The Monte Carlo algorithm is used to simulate and calculate the diversified evolution paths of the disturbance factors under different environmental condition combinations, and the dynamic Bayesian network model is sampled for multiple rounds to obtain the probability density distribution of the rust expansion and cracking time and the member bearing capacity degradation curve; According to the multiple rounds of Monte Carlo simulation results, the quantile interval is extracted to construct a multi-level early warning index set, wherein the short-term early warning level is based on the disturbance factor mutation probability and the Bayesian node edge weight change degree, the medium-term risk index is calculated by weighting the cumulative transition probability in the corrosion state transition chain, and the long-term service life prediction value is obtained by fitting the structural bearing capacity decay function with time.

8. The method for early warning of concrete member erosion damage in brackish water area influenced by multiple factors according to claim 7, characterized in that, The formula of the Monte Carlo algorithm is as follows: where, represents the predicted time to corrosion-induced distress or failure of the component in the kth Monte Carlo simulation; q represents the current time step; n represents the number of disturbance factors; S represents the number of corrosion evolution stages; represents the time-varying disturbance weight of disturbance factor i at stage s; represents the risk-driven intensity of disturbance factor i to stage s in the kth sample; represents the cumulative damage of disturbance factor i at stage s; represents the dynamic environmental tolerance factor under the current environmental combination; represents the risk penalty coefficient of stage s; represents the total risk tolerance threshold of structural durability.

9. The method for early warning of concrete structure damage in brackish water area caused by multi-factors according to claim 1, characterized in that, The determination condition of the prediction index exceeding the safety threshold is determined based on a multi-dimensional index fuzzy fusion mechanism, including the following steps: Set the fuzzy interval set corresponding to the short-term early warning level, the medium-term risk index, and the long-term life prediction value, respectively, to map to the low risk, warning, danger, and limit four safety levels; Based on the fuzzy membership function, the fuzzy deviation degree of each index with respect to the corresponding safety threshold is calculated; A fuzzy rule base is constructed to fuse the membership strength and weight coefficient of each index, perform fuzzy reasoning, and output the comprehensive risk level; When the fuzzy reasoning result level exceeds the specified risk level threshold, the alarm mechanism is triggered.

10. The method for early warning of concrete structure damage in brackish water area caused by multi-factors according to claim 1, characterized in that, The protection decision scheme includes concrete member corrosion protection suggestions, maintenance priority ranking, and sensor layout optimization.

Images