Method, device and medium for monitoring a salt corrosion self-aware airport pavement based on conductive concrete

Abstract

The application relates to a salt corrosion self-sensing airport runway pavement monitoring method, device and medium based on conductive concrete, which comprises the following steps: embedding an electrode grid in a matrix mode in a conductive concrete airport runway pavement layer, collecting three-dimensional space complex impedance of the pavement, extracting an ionic conductive component through spectrum decomposition, correcting output corrected conductivity through multi-physical field coupling, mapping and converting the corrected conductivity into salt concentration to obtain a discrete salt concentration node set, reconstructing the conductivity in the whole pavement volume through three-dimensional resistivity inversion based on a physical information neural network, converting to obtain a reconstructed three-dimensional salt concentration field, modeling spatial correlation and time sequence dependence between each electrode node through a space-time graph neural network, outputting salt concentration prediction values of each electrode node in the future for a set number of days, and calculating spatial gradients and time changes based on the salt concentration prediction values and historical time sequence data to construct a salt corrosion risk index for monitoring. Compared with the prior art, the application has the advantages of high pavement monitoring precision and high reliability.

Description

Technical Field

[0001] This invention relates to the field of health monitoring of transportation infrastructure, and in particular to a method, equipment and medium for monitoring airport pavement salt corrosion based on conductive concrete. Background Technology

[0002] As crucial strategic infrastructure, offshore airports are subjected to long-term exposure to a complex, corrosive environment characterized by high salinity and humidity. Chloride ions (Cl-) in seawater... - ), sulfate ions (SO4) 2- Salt corrosion, through infiltration, diffusion, and crystallization, penetrates the rigid pavement, causing typical salt corrosion defects in cement concrete such as peeling and particle loss. These defects not only damage the pavement structure but can also form foreign object debris (FOD), posing a serious threat to aircraft taxiing safety. They are characterized by their high latency and rapid evolution. Traditional methods for detecting and assessing pavement defects mainly rely on manual inspections and experience-based judgment. However, these methods are not only inefficient and inaccurate, making it difficult to detect and address defects in their early stages, but also necessitate runway closures, significantly impacting airport operations and management.

[0003] To improve monitoring efficiency, researchers have introduced various sensor technologies. However, existing sensor technologies still have many shortcomings. For example, traditional point sensors such as settlement plates and monitoring piles have limited sensing range and cannot achieve continuous monitoring of salinity distribution across the entire pavement. Distributed fiber optic sensing technology can expand the monitoring range, but it lacks an effective strain-displacement analysis method when buried laterally, and its installation and maintenance costs are high, and it is also quite sensitive to temperature changes.

[0004] In recent years, smart concrete with self-sensing capabilities has provided a new approach to solving pavement structure problems. By incorporating conductive materials (such as carbon fiber and carbon nanotubes) into traditional concrete, a continuous conductive network can be formed. This network can then be used to sense internal stress, strain, and cracks by monitoring changes in the concrete's resistivity or conductivity. However, in practical engineering applications, especially for large-scale, wide, and thick pavement structures like airport pavements, smart concrete technology still faces multiple bottlenecks. Regarding data sensing, the electrical signals of conductive concrete are not only affected by internal stress and salt erosion, but are also highly susceptible to significant interference from environmental temperature and humidity. Temperature changes cause thermal expansion and contraction in concrete, altering the contact distance between conductive particles and thus affecting resistivity; humidity changes alter the water content and ion polarization effect in the pores, further impacting electrical properties. Furthermore, current research largely focuses on qualitative analysis or the effects of single environmental variables, lacking dynamic correction models under multi-physics coupling effects. This makes accurately mapping electrical signals to the actual salinity field extremely complex. Summary of the Invention

[0005] The purpose of this invention is to overcome the defects of the prior art by providing a method, equipment and medium for monitoring airport pavement salt corrosion based on conductive concrete.

[0006] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a method for monitoring salt corrosion of airport pavement based on conductive concrete is provided, comprising: S1. An electrode grid is embedded in the conductive concrete airport pavement layer in a matrix manner, and the electrode grid is subjected to multi-frequency current excitation and complex impedance acquisition to obtain the three-dimensional spatial complex impedance of the pavement. S2. Perform spectral decomposition on the three-dimensional spatial complex impedance of the pavement, extract the ionic conductivity component, perform multi-physics field coupling correction on the ionic conductivity component, and output the corrected conductivity. S3. Based on the mapping mechanism, the corrected conductivity is converted into salt concentration to obtain a discrete salt concentration node set; S4. Based on the discrete salt concentration node set, and taking the steady-state electric field partial differential equation as the forward modeling basis, the conductivity is reconstructed in the entire surface volume through the three-dimensional resistivity inversion based on the physical information neural network, and the reconstructed three-dimensional salt concentration field is obtained by transformation based on the mapping mechanism. S5. Using the historical time series data of the reconstructed three-dimensional salt concentration field as input, a spatiotemporal graph neural network is used to model the spatial correlation and temporal dependence between each electrode node, and output the predicted salt concentration at each electrode node for a set number of days in the future. S6. Calculate the spatial gradient and temporal variation based on the predicted salt concentration and the historical time series data of the reconstructed three-dimensional salt concentration field. Construct a salt corrosion risk index based on the spatial gradient and temporal variation. Determine the warning level based on the salt corrosion risk index threshold to obtain the salt corrosion self-sensing airport pavement monitoring results.

[0007] Preferably, in step S1, the electrode grid is embedded in the conductive concrete airport pavement layer in a matrix manner, specifically including: embedding electrodes in a matrix manner along the length and width directions of the conductive concrete airport pavement layer, the electrode embedding depth being adapted to the pavement layer thickness, and each electrode being connected by an insulated wire to form a matrix electrode network covering the entire area.

[0008] Preferably, in step S2, temperature T, relative humidity RH, stress state ε, and carbonization depth d are comprehensively considered. c and porosity The influence on ionic conductivity components Perform multiphysics coupling correction and output corrected conductivity , wherein the corrected conductivity The calculation expression is: , , , , , , In the formula: This is a temperature correction factor; This is the relative humidity correction factor; This is a stress state correction factor; This is a carbonization depth correction factor; This is a porosity correction factor; It is the activation energy; It is the gas constant; and These are the reference temperature and reference relative humidity, respectively. Humidity coefficient; In response to the situation; The coefficients of the first-order strain term; These are the coefficients of the quadratic strain term; The carbonization coefficient; This represents the critical carbonization depth. Reference porosity; Porosity index.

[0009] Preferably, a power-law based mapping mechanism is used to convert the corrected conductivity into the corresponding salt concentration, where the salt concentration S is related to the corrected conductivity. The mapping relationship between them is expressed as follows: , In the formula: coefficients a, b, and c are obtained by measuring several specimens with known salt concentrations under controlled temperature and humidity conditions. The data was obtained by fitting after temperature and humidity correction.

[0010] Preferably, in step S4, based on the discrete salt concentration node set and using the steady-state electric field partial differential equation as the forward modeling basis, conductivity is reconstructed across the entire channel volume through three-dimensional resistivity inversion based on a physical information neural network, and the reconstructed three-dimensional salt concentration field is obtained based on a mapping mechanism. Specifically, this includes: 1) Based on the discrete salt concentration node set N represents the number of nodes, and the measured complex impedances of each electrode node are used to construct an observation data vector. ; 2) Establish forward modeling operators based on the steady-state electric field partial differential equations. Specifically, it includes: Construct the steady-state electric field partial differential equation, the expression of which is: , In the formula: It is a three-dimensional spatial coordinate vector. , The solution domain is the pavement volume; Resistivity; Potential; A forward modeling operator is established based on the steady-state electric field partial differential equation. To be able to use given model parameters Under the given conditions, solving the steady-state electric field partial differential equation yields the model-predicted complex impedance response vector between each electrode pair, where the forward modeling operator... The expression is: , In the formula: The s-th excitation frequency in the number of data samples The model predicts the complex impedance by applying the corresponding current excitation boundary conditions to the steady-state electric field partial differential equation and solving it numerically. 3) Vectorize the observation data and forward operator Substituting the regularized weighted least squares objective function into the solution yields the model-predicted complex impedance response vectors between each electrode pair. The regularized weighted least squares objective function is expressed as follows: , In the formula: and These are the data weight matrix and the model regularization matrix, respectively. For reference model parameters; For regularization parameters, use the forward operator; 4) Conductivity is reconstructed across the entire pavement volume through three-dimensional resistivity inversion based on physical information neural networks; In three-dimensional spatial coordinates and excitation frequency Using resistivity σ and potential φ as inputs, and the collected multi-frequency complex impedance observations of each electrode pair and the output three-dimensional spatial coordinates of each electrode node as supervisory data, a physical information neural network is constructed. The physical constraint loss function is: , , , , In the formula: The s-th excitation frequency in the data sample The measured complex impedance is as follows; Let be the three-dimensional spatial coordinates of the j-th physical constraint sampling point; M is the number of sampling points inside the physical constraint. For gradient operators; Let be the resistivity value at the j-th physical constraint sampling point; Let J be the potential value at the j-th physical constraint sampling point; The degree of boundary condition violation at the k-th boundary condition sampling point; K is the number of boundary condition sampling points. , , These are the weight coefficients for the corresponding items.

[0011] Preferably, in step S5, using the historical time-series data of the reconstructed three-dimensional salt concentration field as input, a spatiotemporal graph neural network is used to model the spatial correlation and temporal dependency between each electrode node, and the predicted salt concentration values ​​at each electrode node for a set number of days in the future are output, specifically including: Construct a spatial graph G=(V,E) based on the electrode array coordinates, and calculate the edge weight matrix. W The formula for edge weights in the spatial graph is: In the formula: Let be the Euclidean distance between the p-th electrode and the q-th electrode; For spatially relevant scale parameters; Constructing a graph convolutional network to extract spatial features, the message passing formula is: , In the formula: Let p be the feature representation of the k-th layer node; Let q be the feature representation of the neighbor node at layer k. Let p be the updated feature of the (k+1)th layer node; For activation functions; This is the transformation matrix representing the node's own characteristics. The transformation matrix for adjacent features; Let p be the set of adjacent nodes; Historical time series data is encoded using bidirectional LSTM. , For the time window, the salt concentration prediction value Ŝ (t+Δt) at each electrode node is decoded by combining the time fusion Transformer for the future set number of days, where Δt is the time interval.

[0012] Preferably, the salt corrosion risk index in S6 is calculated using the following expression: , , , In the formula: Salt corrosion risk index; G The spatial gradient is characterized by the block-scale gradient obtained by statistical aggregation of the spatial gradient magnitude calculated based on the three-dimensional salt concentration field S and then applied to the grid blocks. T The time-series change rate is calculated based on time-series forecast data. These are standardized quantities based on the background distribution. These are the mean and standard deviation of the background statistics, respectively. γ is the weight; γ is the confidence correction factor, calculated based on the inversion confidence interval and observation uncertainty.

[0013] Preferably, the method further includes: By continuously monitoring the signal consistency of each sensor node through graph attention weight, when the attention weight of any node suddenly drops below the first set value, an alarm is automatically triggered. At this time, the corresponding node is marked as an untrusted node, and the data of the untrusted node is downweighted in the current round of inversion and salt corrosion risk index calculation. Specifically, for data marked as untrusted nodes, their corresponding weight coefficients in the data weight matrix are updated using the following expression: , In the formula: The updated weight coefficients for node p; Let be the initial weight coefficients for node p; The attention weight of node p relative to the normal baseline value The sudden drop; This is the weighting factor, and its value range is [0, 1). When the node attention weight recovers to more than M% of the normal baseline value, the weight reduction flag is removed, and the node is restored to its initial weight coefficient. Participate in subsequent calculations.

[0014] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement any of the methods described above.

[0015] According to a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0016] Compared with the prior art, the present invention has the following advantages: (1) This invention is the first to couple the whole-domain embedded electrode grid of conductive concrete, the spectrum separation technology of multi-frequency impedance, and the three-dimensional resistivity inversion based on physical information neural network into an integrated monitoring system. It introduces multi-physics field coupling correction, realizes the quantitative reconstruction of the salt concentration field in the pavement volume and outputs the confidence interval and uncertainty quantification results. It adopts spatiotemporal graph neural network to realize multi-point joint prediction and sensor health monitoring, and proposes a salt corrosion risk index for early warning and graded decision-making. Thus, it forms a systematic breakthrough in the aspects of full-domain continuous coverage, salt corrosion identification accuracy and engineering availability, and provides a more direct quantitative basis and engineering implementation path for the health management of airport pavement in marine areas.

[0017] (2) This invention achieves a leap from traditional point sampling to full-area three-dimensional continuous monitoring by deploying a multi-layer electrode grid array in the conductive concrete pavement layer and combining it with multi-frequency complex impedance measurement. At the same time, it separates the ionic conductivity component and electronic conductivity component in the complex impedance signal through spectrum decomposition technology and corrects it by combining multi-physics field environmental factors, which greatly improves the extraction accuracy of salt ion concentration data.

[0018] (3) The inversion method based on physical information neural network embeds the partial differential equation of steady electric field as a physical constraint into the loss function to ensure that the inversion result conforms to the basic law of electromagnetic field and the inversion reliability is higher. Then, the spatial correlation graph between electrode nodes is constructed through spatiotemporal graph neural network, and the spatial diffusion mode of salt corrosion is captured by graph convolution. Combined with time series modeling, the future salt concentration distribution evolution is predicted, which improves the accuracy of salt concentration prediction.

[0019] (4) The present invention continuously monitors the signal consistency of each sensor node through graph attention weight. When the attention weight of any node suddenly drops to a set value, an alarm is automatically triggered. At the same time, the data of the trusted node is downweighted in the current round of inversion and salt corrosion risk index calculation, thus ensuring the accuracy and reliability of sensor monitoring data.

[0020] (5) The salt corrosion risk index designed in this invention integrates two key dimensions: spatial gradient and time change rate. It eliminates dimensional differences through standardization and establishes a continuous-scale risk quantification index by combining confidence factor weighting. This realizes the transformation from subjective judgment to objective quantification and from passive response to proactive prevention. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] Example This embodiment proposes a self-sensing airport pavement monitoring method based on conductive concrete. This method utilizes conductive concrete and matrix-embedded electrodes to form a global sensing grid. Multi-frequency complex impedance measurement is used to achieve spectral decomposition of the pavement's conductivity within the pavement volume, separating the ionic conductivity component dominated by pore liquid ion conduction from the electronic conductivity component contributed by the carbonaceous conductive network. The measured low-frequency ionic conductivity component is corrected by multi-physics coupling of temperature, relative humidity, stress state, carbonization depth, and porosity. Combined with three-dimensional resistivity inversion based on a physical information neural network (PINN), a conductivity field with a confidence interval is reconstructed spatially. Then, the conductivity field is mapped to a quantitative salt concentration field through the calibrated conductivity-salt concentration coupling relationship. Furthermore, a spatiotemporal graph neural network (ST-GNN) is used to model the spatial correlation and temporal dependence of distributed electrodes, enabling multi-point joint prediction across the entire field. A graph attention mechanism is used to monitor the sensor's health status. Based on spatial gradient, temporal rate of change, and statistical significance threshold, a comprehensive salt corrosion risk index is constructed to achieve early identification and graded warning.

[0024] Next, the method of this embodiment will be described in detail, such as Figure 1 As shown, the specific steps include: S1. An electrode grid is embedded in the conductive concrete airport pavement layer in a matrix manner. The electrode grid is subjected to multi-frequency current excitation and complex impedance acquisition to obtain the three-dimensional spatial complex impedance Z(ω) of the pavement.

[0025] S101. An electrode network is embedded in a matrix along the length and width of the conductive concrete airport pavement layer. The electrodes of the electrode grid are copper electrodes. The electrode embedding depth is adapted to the pavement layer thickness. In this embodiment, the row spacing and column spacing are both 0.5m-2m. Each electrode is connected by an insulated wire to form a matrix electrode network that covers the entire area.

[0026] Electrode to three-dimensional complex impedance With equivalent complex conductivity The calibration relationship between them satisfies an approximate expression: , In the formula: K is a scaling factor related to electrode geometry and boundary conditions, which is determined by reference specimen calibration or numerical forward model (unit: m). Let be the equivalent complex conductivity, and let be the conductivity with respect to the excitation frequency. The function.

[0027] S102. Perform multi-frequency current excitation and complex impedance acquisition on the electrode grid. In this embodiment, the measured excitation frequency is... The range is 20Hz-50kHz, and the three-dimensional spatial complex impedance of the pavement is obtained. .

[0028] S2, Three-dimensional complex impedance of the pavement Spectral decomposition was performed to extract the ionic conductivity component. and the ionic conductivity component Perform multiphysics coupling correction and output corrected conductivity .

[0029] S201. The collected three-dimensional spatial complex impedance of the pavement. Perform spectral decomposition to extract the ionic conductivity component. .

[0030] Specifically, the spectral decomposition is based on the complex conductivity model, and its mathematical expression is: , In the formula: Equivalent complex conductivity (unit: S·m-1); This is a low-frequency ionic conductivity component; The electronic conductivity component has an approximately constant frequency; The effective dielectric constant; It is the imaginary unit.

[0031] S202, ionic conductivity component Perform multiphysics coupling correction and output corrected conductivity .

[0032] Taking into account temperature T, relative humidity RH, stress state ε, and carbonization depth d c and porosity The influence on ionic conductivity components Perform multiphysics coupling correction and output corrected conductivity Among them, corrected conductivity The calculation expression is: , , , , , , In the formula: This is a temperature correction factor; This is the relative humidity correction factor; This is a stress state correction factor; This is a carbonization depth correction factor; This is a porosity correction factor; In this embodiment, the activation energy is set to 40-50 kJ / mol. The gas constant is approximately 8.314 J / (mol・K); and These are the reference temperature and reference relative humidity, respectively. The humidity coefficient is set to 2.5-3.5 in this embodiment. In response to the situation; The coefficients of the first-order strain term; The coefficient for the quadratic strain term is 0.08 in this embodiment. The carbonization coefficient; The critical carbonization depth is set to 10 mm in this embodiment; Reference porosity; The porosity index is 1.5-2.0 in this embodiment.

[0033] S3. Based on the mapping mechanism, the conductivity will be corrected. Converting to the corresponding salt concentration S yields a discrete salt concentration node set. N is the number of nodes.

[0034] The mapping mechanism can be characterized using empirical or mechanistic coupling methods. In this embodiment, a power-law based mapping mechanism is preferred, where the salt concentration S is related to the corrected conductivity. The mapping relationship between them is expressed as follows: , In the formula: coefficients a, b, and c are obtained by measuring several specimens with known salt concentrations under controlled temperature and humidity conditions. The data was obtained by fitting after temperature and humidity correction.

[0035] After converting to the corresponding salt concentration S, the conversion is performed point-by-point for all electrode nodes to obtain the discrete salt concentration values ​​at each node, forming a set of discrete salt concentration nodes with spatial coordinates. N represents the number of nodes, which are used as observation data.

[0036] S4. Based on the discrete salt concentration node set Based on the steady-state electric field partial differential equation as the forward modeling foundation, conductivity is reconstructed across the entire surface volume through three-dimensional resistivity inversion using a physical information neural network. The reconstructed three-dimensional salt concentration field is then obtained through a mapping mechanism. Specifically, this includes: 1) Based on the discrete salt concentration node set The measured complex impedances of each corresponding electrode node are used to construct an observation data vector. .

[0037] 2) Establish forward modeling operators based on the steady-state electric field partial differential equations. Specifically, it includes: Construct the steady-state electric field partial differential equation, the expression of which is: , In the formula: It is a three-dimensional spatial coordinate vector. , The solution domain is the pavement volume; Resistivity; Potential; A forward modeling operator is established based on the steady-state electric field partial differential equation. To be able to use given model parameters Under the given conditions, solving the steady-state electric field partial differential equation yields the model-predicted complex impedance response vector between each electrode pair, where the forward modeling operator... The expression is: , In the formula: The s-th excitation frequency in the number of data samples The model predicts the complex impedance by applying the corresponding current excitation boundary conditions to the steady-state electric field partial differential equation and solving it numerically.

[0038] 3) Vectorize the observation data and forward operator Substituting the regularized weighted least squares objective function into the solution yields the model-predicted complex impedance response vectors between each electrode pair. The regularized weighted least squares objective function is expressed as follows: , In the formula: and These are the data weight matrix and the model regularization matrix, respectively. For reference model parameters; For regularization parameters, the forward operator is used.

[0039] 4) Conductivity is reconstructed across the entire pavement volume through three-dimensional resistivity inversion based on physical information neural networks; In three-dimensional spatial coordinates and excitation frequency As input, resistivity σ and potential φ are used as outputs, and the measured complex impedance of each electrode pair is collected. The three-dimensional spatial coordinates of each output electrode node are used as supervision data to construct a physical information neural network; The physical constraint loss function is: , , , , In the formula: The s-th excitation frequency in the data sample The measured complex impedance is as follows; Let be the three-dimensional spatial coordinates of the j-th physical constraint sampling point; M is the number of sampling points inside the physical constraint. For gradient operators; Let be the resistivity value at the j-th physical constraint sampling point; Let J be the potential value at the j-th physical constraint sampling point; The degree of boundary condition violation at the k-th boundary condition sampling point; K is the number of boundary condition sampling points. , , These are the weight coefficients for the corresponding items.

[0040] In this embodiment, the physical information neural network structure consists of 5 fully connected layers, with the number of hidden layer nodes being [128, 256, 512, 256, 128], and the activation function is tanh.

[0041] In this embodiment, the Adam optimizer is used to train the network with an initial learning rate of 1×10⁻⁶. -3 The cosine annealing strategy was used to reduce the temperature to 1×10⁻⁶. -5 The process is iterated 10,000 times. Fifty forward propagations are performed using Monte Carlo Dropout (holding ratio p=0.8), and the mean resistivity distribution is output. and standard deviation Finally, calculate the 95% confidence interval. Used for quantitative assessment of uncertainty.

[0042] 5) Output the mean distribution σ̄(x,y,z) and standard deviation σ of the currently reconstructed conductivity field using the Monte Carlo Dropout method. std The reconstructed three-dimensional salt concentration field S(x,y,z) is obtained by mapping and transforming the reconstructed conductivity field using (x,y,z) and the 95% confidence interval.

[0043] S5. Using the historical time-series data of the reconstructed three-dimensional salt concentration field as input, a spatiotemporal graph neural network is used to model the spatial correlation and temporal dependence between each electrode node, and output the predicted salt concentration at each electrode node for a set number of days in the future.

[0044] In this embodiment, the multi-point joint prediction based on the spatiotemporal graph neural network ST-GNN specifically includes: Based on electrode array coordinates Construct a spatial graph G=(V,E) and compute the edge weight matrix. W The formula for edge weights in the spatial graph is: ,in, Let be the Euclidean distance between the p-th electrode and the q-th electrode; The spatially relevant scale parameters (in this embodiment, 1-3 meters) are used.

[0045] A graph convolutional network is constructed to extract spatial features, and its message passing formula is as follows: , In the formula: Let p be the feature representation of the k-th layer node, where p = 1 ~ N; Let q be the feature representation of the neighbor node at layer k. The neighbor set of node p Let p be the updated feature of the (k+1)th layer node; For activation functions; This is the transformation matrix representing the node's own characteristics. is the transformation matrix of adjacent features.

[0046] Historical time-series data is encoded using a bidirectional LSTM (256 hidden units). By combining a time-fusion Transformer (6 attention heads) to decode the predicted salt concentration Ŝs(t+Δt) and its confidence interval at each electrode node for the next 7-30 days, the development trend of salt corrosion can be predicted in advance.

[0047] As another preferred embodiment, the graph attention weight α is used. pq Continuously monitor the signal consistency of each sensor node, when the attention weight α of any node... pq When the sudden drop exceeds a first preset value, an alarm is automatically triggered. At this time, the corresponding node is marked as an untrusted node, and the data of the untrusted node is downweighted in the current round of inversion and salt corrosion risk index calculation. Specifically, in this embodiment, when the attention weight α of any node... ij When the sudden drop exceeds 50%, it indicates that the sensor output data is abnormal (such as wiring corrosion, electrode failure, etc.), and the sensor fault alarm will be automatically triggered at this time.

[0048] As another preferred embodiment, the data from untrusted nodes are weighted less heavily during the current round of inversion and salt erosion risk index calculation, specifically as follows: For data marked as untrusted nodes, update their corresponding weight coefficients in the data weight matrix. The update expression is: , In the formula: The updated weight coefficients for node p; Let be the initial weight coefficients for node p; The attention weight of node p relative to the normal baseline value The sudden drop; This is the weighting factor, and its value ranges from [0, 1].

[0049] When the node attention weight recovers to M% (e.g., 90%) or more of the normal baseline value, the weight reduction flag is removed, and the node is restored to its initial weight coefficient. Participate in subsequent calculations.

[0050] S6. Calculate the spatial gradient and temporal variation based on the predicted salt concentration and the historical time-series data of the reconstructed three-dimensional salt concentration field S(x,y,z), and construct a salt corrosion risk index based on the spatial gradient and temporal variation. The warning level is determined based on the salt corrosion risk index threshold, and the monitoring results of the airport pavement with salt corrosion self-sensing are obtained.

[0051] In this embodiment, the salt corrosion risk index The calculation expression is: , , γ=1-(σ σ / μ σ ), In the formula: G The spatial gradient is characterized by the block-scale gradient (using interval statistical aggregation, such as the maximum or mean value) obtained by statistically aggregating the spatial gradient magnitude calculated from the three-dimensional salt concentration field S(x,y,z) through grid blocks. ; T The time-series change rate is calculated based on time-series forecast data. ; These are standardized quantities based on the background distribution. These are the mean and standard deviation of the background statistics, respectively. γ is the weight; γ is the confidence correction factor based on the inversion confidence interval and observation uncertainty, 0 < γ ≤ 1.

[0052] Due to differences in salt corrosion risk index under different road sections, structures, or time periods The historical distributions (mean and variance) differ, therefore, several data points (n≥30) from disease-free and typical operating periods were collected first. Data, calculate the mean with standard deviation After standardization, it is mapped to the standardized 0–100 exponent or percentile: , , If the historical distribution is significantly non-normal, use the empirical quantile mapping instead of the normal quantile mapping: , Based on the comprehensive salt corrosion risk index The standard is used to classify the risk levels, and the comprehensive salt corrosion risk index is determined based on preset thresholds. Perform multi-level judgments and output the corresponding multi-level warning levels, as determined in Table 1 below.

[0053] Table 1

[0054] In this embodiment, the project site is an airport located in a sea area of ​​a city in a certain province. The project involves the intelligent pavement renovation of the airport's main runway, which is 3200m long, 45m wide, and 350mm thick. The region has a tropical maritime climate with an average annual temperature of 26.8℃, an average annual humidity of 82%, 200 days of salt fog annually, and a peak air chloride ion concentration of 2.1 mg / m³. 3 The saline-erosion environment is extremely harsh. The project commenced in March 2023, selecting the section from K1+400 to K2+000 as a demonstration section (600m long, 27,000m²). 2 The project adopted a technical approach combining in-situ monitoring and preventative maintenance. The project lasted 20 months and was accepted in October 2024.

[0055] Table 2. Mix Proportion Design of Conductive Concrete

[0056] Table 3 Test results of conductive concrete performance

[0057] While maintaining basic mechanical properties (strength decrease <5%), conductive concrete has increased electrical conductivity by 435 times and chloride ion penetration resistance by 79.3%, fully meeting the application requirements of marine environments.

[0058] Table 4 Electrode Grid Layout Parameters

[0059] An array of embedded electrodes is constructed within the conductive concrete pavement layer, followed by surface passivation treatment. The electrodes are connected via a dual-core shielded cable (0.75mm²). 2 Connect to the edge hub box, with a single cable length ≤ 50m.

[0060] Multi-frequency current excitation and complex impedance acquisition were performed on the electrode grid. The excitation frequency range during measurement was 20Hz-50kHz. The three-dimensional spatial complex impedance Z(f) of the pavement was obtained. Table 5 below shows the measured data of a typical measuring point at K1+650 meters in March 2024.

[0061] Table 5. Results of Multi-Frequency Complex Impedance Measurement

[0062] The scaling factor K is determined through reference specimen calibration or numerical forward modeling. The scaling factor K for this project is 1.85m.

[0063] Using complex conductivity model Spectral decomposition is performed on the acquired complex impedance data.

[0064] Spectral decomposition at measuring point K1+650 meters, and calculation of complex conductivity from complex impedance. For example, with f=20Hz: Given: ; Real part calculation: ; Imaginary part calculation: .

[0065] Table 6. Calculation results of complex conductivity at different excitation frequencies

[0066] Levenberg-Marquardt nonlinear fitting objective function: , The model It converged after 18 iterations.

[0067] Table 7 Spectrum Decomposition Fitting Parameters

[0068] For ionic conductivity components Multiphysics coupling correction is performed, taking into account temperature T, relative humidity RH, stress state ε, and carbonization depth d. c and porosity The influence of output correction conductivity Correction formula: , , , , , , Measured environmental parameters at measuring point K1+650 meters: Temperature T=31.2℃ (304.35K), Relative humidity RH=85%, Stress state ε=280με, Carbonization depth d c =6.8mm (pH=12.6), porosity =0.142.

[0069] Perform corrective calculations: Temperature correction factor calculation: known , , , ,but: ; Humidity correction factor calculation: Given RH=85%, RH ref =80%, β=3.2, then: ; Stress correction factor calculation: Given ε = 280 × 10 -6 α ε =-0.15, β ε =0.08, ; Carbonization correction factor calculation: Given d c =6.8mm,d critical =10mm, γ=0.3, ; Pore ​​correction factor calculation: known =0.142, =0.125, =1.9, then: ; ; Corrected conductivity: .

[0070] Table 8 Summary of Multiphysics Correction Factors

[0071] Using a mapping module, and employing empirically or mechanistically coupled veil functions, the corrected conductivity is... Convert to the corresponding salt concentration S. Mapping relationship: .

[0072] By preparing different salt concentrations (0.5, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 8.0 kg / m³), 3 The resistivity of conductive concrete specimens was measured and converted to conductivity under standard conditions (T=25℃, RH=80%), and the mapping parameters were obtained by nonlinear regression fitting.

[0073] Table 9 Conductivity-Salt Concentration Calibration Data

[0074] Fitting results: ; Goodness of fit: ; Calculation of salt concentration at measuring point K1+650 meters: (Given information) =0.105 S / m, then: .

[0075] Table 10 Mapping Transformation Calculation Results

[0076] Based on the steady-state electric field partial differential equation as the forward modeling basis, the three-dimensional resistivity field is inverted using the Physical Information Neural Network (PINN-ERT), the conductivity field is reconstructed, and the confidence interval is calculated.

[0077] Governing equations: .

[0078] A deep neural network is constructed using an inversion method, with three-dimensional spatial coordinates as input. and excitation frequency The outputs are resistivity σ(x,y,z) and potential φ(x,y,z).

[0079] Table 11 PINN Network Structure Parameters

[0080] loss function The following parts are included: Data fitting terms: , Physical constraints: , Boundary condition terms: , Total loss function: , PINN inversion calculation for segment K1+400~K2+000: The input data includes 1026 electrode position coordinates, 6 excitation frequency points (20Hz, 100Hz, 500Hz, 1kHz, 10kHz, 50kHz), a total of 6156 complex impedance measurements, 10000 randomly sampled internal points in the solution domain for physical constraints, and 2000 boundary sampling points.

[0081] Table 12 PINN Training Hyperparameters

[0082] The training convergence record is as follows, with a training time of approximately 68 minutes (Dell R740xd server, 4×NVIDIA V100 GPUs).

[0083] Table 13 PINN Training Convergence Record

[0084] Table 14. Inversion results of three-dimensional salt concentration distribution in the K1+400~K2+000 segment of PINN.

[0085] Spatial gradient statistics are as follows: 1) Maximum spatial gradient: 0.42 kg / m 4 (Located at K1+650 meters above the surface); 2) Longitudinal gradient (along the runway direction): 0.15~0.38 kg / m 4 ; 3) Lateral gradient (perpendicular to the track direction): 0.08~0.22 kg / m 4 ; 4) Depth gradient (vertically downward): 0.25~0.42 kg / m 4 .

[0086] Uncertainty quantification was performed by selecting the K1+650 meter measurement point (x=1650m, y=0m, z=0.06m) for 50 forward propagation samplings, with a Dropout retention rate of p=0.8.

[0087] Table 15 Quantification Results of Uncertainty at Measurement Point K1+650m

[0088] The confidence level correction factor is used as a quality weight in the subsequent risk index calculation: .

[0089] Multi-point joint prediction of salt concentration distribution evolution over the next 7-30 days is performed based on a spatiotemporal graphical neural network (ST-GNN), according to the electrode array coordinates (x... i ,yi ,z i Construct a spatial graph G=(V,E), with edge weights.

[0090] Taking the neighboring nodes of the K1+650 meter measuring point as an example, the parameter settings are as follows: Number of nodes: |V|=1026 (corresponding to 1026 electrodes); Spatial correlation scale Γ=2.5m; Sparsity threshold: w ij Edges with an edge value less than 0.08 are removed; the adjacency matrix contains approximately 15,400 non-zero elements.

[0091] Table 16 Quantification Results of Uncertainty at Measurement Point K1+650m

[0092] Table 17 ST-GNN Model Structure

[0093] Message passing formula: .

[0094] The training data is as follows: 1) Historical data: September 2023 - February 2024 (18 months); 2) Sample size: 1026 nodes × 540 days = 553980 time series samples; 3) Training set: 70% (387,786 samples); 4) Validation set: 15% (83,097 samples); 5) Test set: 15% (83,097 samples).

[0095] Table 18 ST-GNN Training Configuration

[0096] Table 19 Performance Indicators for Different Prediction Spans

[0097] Table 20 Performance Comparison of Forecasting Methods (30-day span)

[0098] Table 21 Prediction of Salt Concentration at K1+650m Meter for the Next 30 Days

[0099] Therefore, the trend over the next 30 days can be predicted, with the concentration increasing from 4.82 kg / m³. 3 Increased to 5.18 kg / m 3 The average daily growth rate is approximately 0.25%, and it may exceed 6.0 kg / m³ in early June. 3Severity threshold.

[0100] An anomaly was detected in the graph attention mechanism on April 12, 2024: Table 22 Attention Weight Anomaly Detection Record for Node #487

[0101] After on-site inspection, it was found that the electrode terminals were corroded due to moisture. After replacement, the weight returned. This verified the system's self-monitoring capability.

[0102] Finally, based on the calculation of the spatial gradient G and the time rate of change T of the conductivity / salt concentration field, a unified comprehensive salt corrosion risk index (CRI) is constructed. index and warning levels.

[0103] Spatial gradient: ; Rate of change over time: ; Taking the K1+650 meter measuring point as an example, the process of calculating the risk index is as follows: First, using the K1+650 meter measuring point as the center, extract the salt concentration data of adjacent nodes: Table 23 Salt concentration distribution around the K1+650 meter measuring point (kg / m³) 3 )

[0104] Calculate the partial derivatives: .

[0105] Calculate the spatial gradient magnitude: ; Since depth-direction gradients dominate, the statistical analysis focuses on the horizontal and vertical gradients within a 6cm range of the surface layer. ; However, considering the existence of localized concentration anomalies in this area, the actual statistical maximum value is: .

[0106] Next, we calculate the rate of change over time, T.

[0107] Table 24 Historical Salt Concentration Records at K1+650m Measuring Point

[0108] Calculate the rate of change over the past 30 days: .

[0109] Then, based on the background statistics for the entire 18 months, and after standardization: Table 25 Background Statistical Parameters (K1+400~K2+000 range, 18 months)

[0110] standardization: , , Calculate the raw risk index CRI raw The formula is as follows: , Substitute the values: , Standardized Z-score, based on historical CRI raw Distribution parameter: μ CR =0.512, σ CR =0.338 , Mapped to an exponent of 0-100, the formula is as follows: , Substitute the values: , , Table 26 Comprehensive Salt Corrosion Risk Index Early Warning Classification Standards

[0111] Table 27 Summary of CRI calculations at K1+650m measuring point

[0112] The determination result indicates that the measuring point is at K1+650 meters. =95.7, which is a dangerous level and requires immediate intervention.

[0113] Table 28. Distribution of Risk Levels Across the Region from K1+400 to K2+000

[0114] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0115] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0116] The processing unit executes the various methods and processes described above, such as methods S1 to S6. For example, in some embodiments, methods S1 to S6 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S6 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S6 by any other suitable means (e.g., by means of firmware).

[0117] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload programmable logic devices (CPLDs), and so on.

[0118] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0119] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring salt corrosion on airport pavement based on conductive concrete, characterized in that, include: S1. An electrode grid is embedded in the conductive concrete airport pavement layer in a matrix manner, and the electrode grid is subjected to multi-frequency current excitation and complex impedance acquisition to obtain the three-dimensional spatial complex impedance of the pavement. S2. Perform spectral decomposition on the three-dimensional spatial complex impedance of the pavement, extract the ionic conductivity component, perform multi-physics field coupling correction on the ionic conductivity component, and output the corrected conductivity. S3. Based on the mapping mechanism, the corrected conductivity is converted into salt concentration to obtain a discrete salt concentration node set; S4. Based on the discrete salt concentration node set, and taking the steady-state electric field partial differential equation as the forward modeling basis, the conductivity is reconstructed in the entire surface volume through the three-dimensional resistivity inversion based on the physical information neural network, and the reconstructed three-dimensional salt concentration field is obtained by transformation based on the mapping mechanism. S5. Using the historical time series data of the reconstructed three-dimensional salt concentration field as input, a spatiotemporal graph neural network is used to model the spatial correlation and temporal dependence between each electrode node, and output the predicted salt concentration at each electrode node for a set number of days in the future. S6. Calculate the spatial gradient and temporal variation based on the predicted salt concentration and the historical time series data of the reconstructed three-dimensional salt concentration field. Construct a salt corrosion risk index based on the spatial gradient and temporal variation. Determine the warning level based on the salt corrosion risk index threshold to obtain the salt corrosion self-sensing airport pavement monitoring results.

2. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, In step S1, an electrode grid is embedded in the conductive concrete airport pavement layer in a matrix manner. Specifically, this includes embedding electrodes in a matrix manner along the length and width directions of the conductive concrete airport pavement layer. The electrode embedding depth is adapted to the thickness of the pavement layer. Each electrode is connected by an insulated wire to form a matrix electrode network that covers the entire area.

3. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, In S2, temperature T, relative humidity RH, stress state ε, and carbonization depth d are comprehensively considered. c and porosity The influence on ionic conductivity components Perform multiphysics coupling correction and output corrected conductivity , wherein the corrected conductivity The calculation expression is: ， ， ， ， ， ， In the formula: This is a temperature correction factor; This is the relative humidity correction factor; This is a stress state correction factor; This is a carbonization depth correction factor; This is a porosity correction factor; It is the activation energy; It is the gas constant; and These are the reference temperature and reference relative humidity, respectively. Humidity coefficient; In response to the situation; The coefficients of the first-order strain term; These are the coefficients of the quadratic strain term; The carbonization coefficient; This represents the critical carbonization depth. Reference porosity; Porosity index.

4. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, A power-law-based mapping mechanism is used to convert the corrected conductivity into the corresponding salt concentration. The salt concentration S is related to the corrected conductivity. The mapping relationship between them is expressed as: ， In the formula: the coefficients a, b, and c are obtained by fitting the conductivity measured with several specimens of known salt concentration under controlled temperature and humidity conditions after temperature and humidity correction.

5. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, In step S4, based on the discrete salt concentration node set and using the steady-state electric field partial differential equation as the forward modeling basis, conductivity is reconstructed across the entire channel volume through three-dimensional resistivity inversion based on a physical information neural network. The reconstructed three-dimensional salt concentration field is then obtained based on a mapping mechanism, specifically including: 1) Based on the discrete salt concentration node set N represents the number of nodes, and the measured complex impedances of each electrode node are used to construct an observation data vector. ; 2) Establish forward modeling operators based on the steady-state electric field partial differential equations. Specifically, it includes: Construct the steady-state electric field partial differential equation, the expression of which is: ， In the formula: It is a three-dimensional spatial coordinate vector. , The solution domain is the pavement volume. Resistivity; Potential; A forward modeling operator is established based on the steady-state electric field partial differential equation. To be able to use given model parameters Under the given conditions, solving the steady-state electric field partial differential equation yields the model-predicted complex impedance response vector between each electrode pair, where the forward modeling operator... The expression is: ， In the formula: The s-th excitation frequency in the number of data samples The model predicts the complex impedance by applying the corresponding current excitation boundary conditions to the steady-state electric field partial differential equation and solving it numerically. 3) Vectorize the observation data and forward operator Substituting the regularized weighted least squares objective function into the solution yields the model-predicted complex impedance response vectors between each electrode pair. The regularized weighted least squares objective function is expressed as follows: ， In the formula: and These are the data weight matrix and the model regularization matrix, respectively. For reference model parameters; For regularization parameters, use the forward operator; 4) Conductivity is reconstructed across the entire pavement volume through three-dimensional resistivity inversion based on physical information neural networks; In three-dimensional spatial coordinates and excitation frequency Using resistivity σ and potential φ as inputs, and the collected multi-frequency complex impedance observations of each electrode pair and the three-dimensional spatial coordinates of each electrode node as supervisory data, a physical information neural network is constructed. The physical constraint loss function is: ， ， ， ， In the formula: The s-th excitation frequency in the data sample The measured complex impedance is as follows; Let be the three-dimensional spatial coordinates of the j-th physical constraint sampling point; M is the number of sampling points inside the physical constraint. For gradient operators; Let be the resistivity value at the j-th physical constraint sampling point; Let J be the potential value at the j-th physical constraint sampling point; The degree of boundary condition violation at the k-th boundary condition sampling point; K is the number of boundary condition sampling points. , , These are the weight coefficients for the corresponding items.

6. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, In step S5, using historical time-series data of the reconstructed three-dimensional salt concentration field as input, a spatiotemporal graph neural network is employed to model the spatial correlation and temporal dependency between electrode nodes, outputting predicted salt concentration values ​​at each electrode node for a set number of days in the future, specifically including: Construct a spatial graph G=(V,E) based on the electrode array coordinates, and calculate the edge weight matrix. W The formula for edge weights in the spatial graph is: In the formula: Let be the Euclidean distance between the p-th electrode and the q-th electrode; For spatially relevant scale parameters; Constructing a graph convolutional network to extract spatial features, the message passing formula is: ， In the formula: Let p be the feature representation of the k-th layer node; Let q be the feature representation of the neighbor node at layer k. Let p be the updated feature of the (k+1)th layer node; For activation functions; This is the transformation matrix representing the node's own characteristics. The transformation matrix for adjacent features; Let p be the set of adjacent nodes; Historical time series data is encoded using bidirectional LSTM. , For the time window, the salt concentration prediction value Ŝ (t+Δt) at each electrode node is decoded by combining the time fusion Transformer for a set number of days in the future, where Δt is the time interval.

7. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, The salt corrosion risk index in S6 is calculated using the following expression: ， ， ， In the formula: Salt corrosion risk index; G The spatial gradient is characterized by the block-scale gradient obtained by statistical aggregation of the spatial gradient magnitude calculated based on the three-dimensional salt concentration field S and then applied to the grid blocks. T The time-series change rate is calculated based on time-series forecast data. These are standardized quantities based on the background distribution. These are the mean and standard deviation of the background statistics, respectively. γ is the weight; γ is the confidence correction factor, calculated based on the inversion confidence interval and observation uncertainty.

8. The method for monitoring salt corrosion of airport pavement based on conductive concrete according to claim 1, characterized in that, The method further includes: By continuously monitoring the signal consistency of each sensor node through graph attention weight, when the attention weight of any node suddenly drops below the first set value, an alarm is automatically triggered. At this time, the corresponding node is marked as an untrusted node, and the data of the untrusted node is downweighted in the current round of inversion and salt corrosion risk index calculation. Specifically, for data marked as untrusted nodes, their corresponding weight coefficients in the data weight matrix are updated using the following expression: ， In the formula: The updated weight coefficients for node p; Let be the initial weight coefficients for node p; The attention weight of node p relative to the normal baseline value The sudden drop; This is the weighting factor, and its value range is [0, 1). When the node attention weight recovers to more than M% of the normal baseline value, the weight reduction flag is removed, and the node is restored to its initial weight coefficient. Participate in subsequent calculations.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images