Expressway waterproofing and drainage integrated intelligent structural health monitoring method and system

Abstract

The application provides a highway drainage and waterproof integrated intelligent structural health monitoring method and system, and relates to the technical field of road health monitoring.The method comprises the following steps: collecting deformation and seepage parameters of a drainage and waterproof structure, performing spatial gradient calculation and seepage time series analysis to obtain structural state characteristics;obtaining drainage connectivity characteristics based on a drainage path topological relationship, obtaining waterproof degradation characteristics based on a waterproof level relationship, and coupling and mapping to obtain a cooperative state vector; calculating a drainage component and a waterproof component interaction coefficient based on the cooperative state vector, judging a conduction influence and a reverse restraint degree, and correcting the state vector; identifying associated nodes in a dependency relationship graph, backtracking seepage characteristics to obtain a damage propagation path, calculating a propagation rate, and generating a health monitoring report.

Description

Technical Field

[0001] This invention relates to the field of road health monitoring technology, and in particular to a method and system for integrated intelligent structure health monitoring of drainage and flood control on highways. Background Technology

[0002] In the field of highway operation and maintenance, health monitoring of drainage and waterproofing systems is a crucial aspect of ensuring the long-term stability of pavement structures. Current technologies primarily rely on independently deployed sensor networks to monitor drainage facilities and waterproofing layers separately. For example, water level and flow sensors are installed in drainage ditches and catch basins to assess drainage efficiency, while strain gauges and crack gauges are placed on or inside the waterproofing layer to determine the degree of damage. Monitoring data is typically recorded and stored in isolation, with anomaly alerts triggered through manual inspections or simple threshold comparisons.

[0003] However, in practical applications, drainage systems and waterproofing layers are strongly coupled during actual service. Existing technologies cannot reveal the bidirectional transmission effects, making it difficult to accurately assess the true extent of the synergistic degradation of the drainage and waterproofing systems. Furthermore, monitoring data lacks topological support, making it impossible to trace and predict the propagation path of defects. This results in maintenance decisions often lagging behind the development of defects and the inability to achieve targeted local intervention, leading to a waste of maintenance resources. Summary of the Invention

[0004] This invention provides a method and system for integrated intelligent structure health monitoring of drainage and waterproofing in highways, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a method for health monitoring of an integrated intelligent structure for drainage and flood control in highways, comprising:

[0006] Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Temporal cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics.

[0007] Based on the pre-constructed drainage path topology, graph propagation operation is performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operation is performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector.

[0008] The interaction coefficient is calculated based on the drainage component and the waterproof component in the cooperative state vector. The degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity are determined according to the interaction coefficient. The cooperative state vector is then corrected according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector.

[0009] Based on the structural position corresponding to the modified state vector, the associated nodes are identified in the pre-established structural dependency graph. The seepage characteristics corresponding to the associated nodes are back-analyzed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path and a health monitoring report is generated.

[0010] In one alternative implementation,

[0011] Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Time-series cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics, including:

[0012] Deformation and seepage parameters of the highway drainage and waterproofing structure at multiple monitoring points are collected by a distributed sensing unit. A spatial grid is constructed according to the spatial location of the monitoring points for the deformation parameters, and the deformation difference between adjacent grid units is calculated to obtain a spatial gradient field. The spatial gradient field is decomposed into radial gradient components. Based on the radial gradient components, local abrupt regions are identified, and the gradient amplitude of the abrupt regions is extracted as gradient features.

[0013] The seepage parameters are accumulated and summed in segments according to the time series to obtain the cumulative seepage flow rate for each time period. The rate of change of the cumulative seepage flow rate between adjacent time periods is calculated, and time periods with a rate of change exceeding a preset threshold are identified as abnormal time periods. The seepage parameters corresponding to the abnormal time periods are extracted, and frequency domain features are obtained through spectrum analysis. The seepage characteristics are then determined in combination with the cumulative seepage flow rate.

[0014] Based on the local mutation region, the target monitoring point is located in the monitoring point grid. The seepage feature corresponding to the target monitoring point is extracted to obtain the target seepage feature. The correlation between the gradient feature and the target seepage feature is calculated and a coupling matrix is ​​constructed based on the correlation. The gradient feature and the seepage feature are then subjected to tensor operation through the coupling matrix to obtain the structural state feature.

[0015] In one alternative implementation,

[0016] Based on the pre-constructed drainage path topology, graph propagation operations are performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operations are performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector, including:

[0017] The structural state features are mapped to each node of the drainage path topology to obtain node state features. The drainage path topology is spectrally decomposed to obtain a Laplace feature vector set. The node state features are projected onto the Laplace feature vector set to obtain spectral coefficients and spectral propagation is performed to obtain propagated node features. The gradient abrupt change positions of the propagated node features are identified and the corresponding spectral coefficients are extracted to form drainage connectivity features.

[0018] The structural state features are assigned to each level of the waterproofing hierarchy to obtain the hierarchical state features. A hypergraph representation corresponding to the waterproofing hierarchy is constructed, and the hierarchical state features are propagated by hyperedges to obtain hyperedge enhancement features. The hyperedge enhancement features are decomposed by tensors to obtain core tensors, and the residual between the core tensors and the hyperedge enhancement features is calculated. Based on the residual, core tensor slices are extracted to form waterproofing degradation features.

[0019] Tensor contraction is performed on the drainage connectivity feature and the waterproofing degradation feature to obtain a contraction tensor. An attention distribution is calculated on the contraction tensor and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is performed on the coupled feature tensor to obtain drainage components and waterproofing components, and a cooperative state vector is obtained by solving the problem.

[0020] In one alternative implementation,

[0021] The attention distribution is calculated for the contraction tensor, and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is then performed on the coupled feature tensor to obtain drainage and waterproofing components, and the cooperative state vector is obtained by solving for these components.

[0022] The contraction tensor is sliced ​​along the drainage path dimension and the waterproofing layer dimension to obtain drainage dimension slice group and waterproofing dimension slice group respectively. Self-attention calculation is performed on the drainage dimension slice group and the drainage connectivity feature to obtain drainage self-attention feature. Self-attention calculation is performed on the waterproofing dimension slice group and the waterproofing degradation feature to obtain waterproofing self-attention feature. Cross-attention calculation is performed on the drainage self-attention feature and the waterproofing self-attention feature, and sparse attention distribution is obtained after gated linear unit transformation and normalization.

[0023] The drainage connectivity features and the waterproofing degradation features are projected into a high-dimensional feature space and weighted and aggregated based on the sparse attention distribution to obtain attention enhancement features. The attention enhancement features are then propagated through a graph neural network to obtain topology enhancement features and mapped through a bilinear pooling layer to obtain a coupled feature tensor.

[0024] The coupled feature tensor is decomposed by Tucker to obtain a core tensor and multiple factor matrices. Drainage-related modal slices are extracted from the core tensor and reconstructed with the corresponding factor matrices to obtain drainage components. Waterproofing-related modal slices are extracted and reconstructed with the corresponding factor matrices to obtain waterproofing components. The drainage components and waterproofing components are fused through residual connection and then processed by layer normalization to obtain the cooperative state vector.

[0025] In one alternative implementation,

[0026] The interaction coefficient is calculated based on the drainage component and the waterproofing component in the cooperative state vector. The degree of transmission influence of drainage failure on waterproofing performance and the degree of reverse constraint of waterproofing damage on drainage capacity are determined based on the interaction coefficient, including:

[0027] The drainage component and the waterproofing component are separated from the cooperative state vector and expanded into drainage state sequence and waterproofing state sequence, respectively. The drainage state sequence and the waterproofing state sequence are time-aligned and the covariance is calculated under different lag time differences to obtain the covariance matrix. The positive conduction intensity of drainage to waterproofing and the reverse conduction intensity of waterproofing to drainage are extracted from the covariance matrix and combined to obtain the bidirectional interaction coefficient.

[0028] The drainage component is mapped to the nodes of the drainage path topology and the failure node is identified to obtain the failure time. The waterproof component is mapped to each waterproof layer and the amount of degradation change after the failure time is calculated to obtain the degradation increment sequence. The degradation increment sequence is multiplied element by element based on the positive conduction intensity in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the degree of conduction influence.

[0029] The failure time is obtained by identifying the damaged layer from the waterproof layer that has deteriorated beyond the preset deterioration threshold. The change in drainage capacity of the node after the failure time is calculated as a capacity decay sequence. The capacity decay sequence is multiplied element by element based on the reverse conduction strength in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the reverse constraint degree.

[0030] In one alternative implementation,

[0031] The modified state vector is obtained by modifying the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint, including:

[0032] The conduction intensity value is extracted from the degree of conduction influence to construct a conduction intensity matrix, and the constraint intensity value is extracted from the degree of reverse constraint to construct a constraint intensity matrix. The conduction intensity matrix and the constraint intensity matrix are subjected to asymmetric coupling analysis to obtain a coupling intensity field. The coupling intensity field is decomposed to obtain the conduction vector and the constraint vector, and the vector angle is calculated to identify the conduction-dominant cooperative region and the constraint-dominant antagonistic region.

[0033] The pre-obtained connectivity failure nodes are taken as drainage failure nodes, and the pre-obtained damage levels are taken as damage waterproofing levels. In the dominant conduction cooperative region, a dominant conduction path is constructed from the drainage failure nodes to the corresponding waterproofing levels along the conduction vector, and the path attenuation coefficient is calculated. In the dominant constraint antagonistic region, a reverse blocking path is constructed from the damage waterproofing levels to the corresponding drainage nodes along the constraint vector, and the path damping coefficient is calculated.

[0034] The failure state change of the drainage failure node is extracted and accumulated along the dominant conduction path according to the path attenuation coefficient to the corresponding waterproofing level to obtain the corrected waterproofing component. The damage and deterioration change of the damaged waterproofing level is extracted and accumulated along the reverse blocking path according to the path damping coefficient to the corresponding drainage node to obtain the corrected drainage component. The corrected waterproofing component and the corrected drainage component are spliced ​​together to obtain the corrected state vector.

[0035] In one alternative implementation,

[0036] Based on the structural location corresponding to the corrected state vector, associated nodes are identified in a pre-established structural dependency graph. Backtracking analysis of the seepage characteristics corresponding to the associated nodes is performed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path, and a health monitoring report is generated, including:

[0037] The corrected drainage component and the corrected waterproof component are extracted from the corrected state vector and the corresponding structural positions are obtained by parsing. The structural positions are mapped to the pre-established structural dependency graph to obtain the position nodes. The structural dependency graph is topologically sorted to obtain the dependency level. Based on the dependency level, the predecessor nodes are traced upward from the position nodes. The predecessor nodes are connected to the nodes and nodes that form a strong connected component with the position nodes are identified as associated nodes.

[0038] Extract the seepage features corresponding to the associated nodes, perform reverse traversal over time, and calculate the feature change gradient of adjacent time points. Identify the time when the feature change gradient exceeds a preset gradient threshold as the damage initiation time and record the corresponding node as the damage initiation node. Search for a path from the damage initiation node along the structural dependency graph to the location node and determine the damage propagation path.

[0039] The seepage characteristics of adjacent nodes in the damage propagation path are extracted and the time difference between adjacent nodes at the damage initiation time is calculated to obtain the propagation delay. The structural distance between adjacent nodes is extracted from the structural dependency graph and the propagation rate is calculated based on the propagation delay. The damage propagation path and the propagation rate are combined with the corrected state vector to generate a health monitoring report.

[0040] A second aspect of the present invention provides an integrated intelligent structural health monitoring system for drainage and flood control in highways, comprising:

[0041] The state feature construction unit is used to collect deformation parameters and seepage parameters of the drainage and waterproofing structure of the highway, perform spatial gradient calculation on the deformation parameters to obtain gradient features, perform time-series cumulative analysis on the seepage parameters to obtain seepage features, and fuse them with the gradient features to obtain structural state features.

[0042] The collaborative state coupling unit is used to perform graph propagation operation on the structural state features based on the pre-constructed drainage path topology relationship to obtain drainage connectivity features, perform hierarchical aggregation operation on the structural state features based on the pre-constructed waterproofing hierarchy relationship to obtain waterproofing degradation features, and couple and map the drainage connectivity features and the waterproofing degradation features to obtain a collaborative state vector.

[0043] The interaction relationship correction unit is used to calculate the interaction coefficient based on the drainage component and the waterproof component in the cooperative state vector, determine the degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity according to the interaction coefficient, and correct the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector.

[0044] The damage propagation analysis unit is used to identify associated nodes in a pre-established structural dependency graph based on the structural position corresponding to the modified state vector, perform back-tracking analysis on the seepage characteristics corresponding to the associated nodes to obtain the damage propagation path, calculate the propagation rate along the damage propagation path, and generate a health monitoring report.

[0045] A third aspect of the present invention provides an electronic device, comprising:

[0046] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0047] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0048] In this invention, the joint acquisition and feature fusion of deformation and seepage parameters comprehensively reflects the multidimensional response state of drainage and waterproofing structures. Spatial gradient calculation of deformation parameters accurately captures local deformation anomalies, while time-series cumulative analysis of seepage parameters effectively identifies changes in seepage trends. The structural state features formed by the fusion of these two methods avoid the one-sidedness of single-parameter criteria, significantly improving the sensitivity and accuracy of structural anomaly detection. Graph propagation operations based on drainage path topology and hierarchical aggregation operations based on waterproofing layer relationships extract key features from the dimensions of connectivity and degradation, respectively. These features are then coupled and mapped to form a collaborative state vector, achieving... The quantitative characterization of the interaction between the drainage system and the waterproofing system, and the calculation of the interaction coefficient, can clearly distinguish the degree of transmission of the impact of drainage failure on waterproofing performance and the degree of reverse constraint of waterproofing damage on drainage capacity. Based on this, the cooperative state vector is corrected to eliminate information bias and make the assessment results closer to the actual health status of the structure. Based on the corrected state vector, the associated nodes are identified in the structural dependency graph, and the seepage characteristics are back-analyzed to accurately locate the starting position and propagation path of damage propagation. The propagation rate is calculated along the propagation path, providing data support for predicting structural deterioration trends and prioritizing maintenance, and effectively reducing the operation and maintenance costs of highways throughout their entire life cycle. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the intelligent structural health monitoring method for integrated drainage and waterproofing in highways, as described in an embodiment of the present invention.

[0050] Figure 2 This is a flowchart illustrating the modified state vector generation process of the intelligent structure health monitoring method for integrated drainage and waterproofing of highways, as described in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the integrated intelligent structure health monitoring method for drainage and waterproofing in highways, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Temporal cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics.

[0055] Based on the pre-constructed drainage path topology, graph propagation operation is performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operation is performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector.

[0056] The interaction coefficient is calculated based on the drainage component and the waterproof component in the cooperative state vector. The degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity are determined according to the interaction coefficient. The cooperative state vector is then corrected according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector.

[0057] Based on the structural position corresponding to the modified state vector, the associated nodes are identified in the pre-established structural dependency graph. The seepage characteristics corresponding to the associated nodes are back-analyzed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path and a health monitoring report is generated.

[0058] In one alternative implementation,

[0059] Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Time-series cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics, including:

[0060] Deformation and seepage parameters of the highway drainage and waterproofing structure at multiple monitoring points are collected by a distributed sensing unit. A spatial grid is constructed according to the spatial location of the monitoring points for the deformation parameters, and the deformation difference between adjacent grid units is calculated to obtain a spatial gradient field. The spatial gradient field is decomposed into radial gradient components. Based on the radial gradient components, local abrupt regions are identified, and the gradient amplitude of the abrupt regions is extracted as gradient features.

[0061] The seepage parameters are accumulated and summed in segments according to the time series to obtain the cumulative seepage flow rate for each time period. The rate of change of the cumulative seepage flow rate between adjacent time periods is calculated, and time periods with a rate of change exceeding a preset threshold are identified as abnormal time periods. The seepage parameters corresponding to the abnormal time periods are extracted, and frequency domain features are obtained through spectrum analysis. The seepage characteristics are then determined in combination with the cumulative seepage flow rate.

[0062] Based on the local mutation region, the target monitoring point is located in the monitoring point grid. The seepage feature corresponding to the target monitoring point is extracted to obtain the target seepage feature. The correlation between the gradient feature and the target seepage feature is calculated and a coupling matrix is ​​constructed based on the correlation. The gradient feature and the seepage feature are then subjected to tensor operation through the coupling matrix to obtain the structural state feature.

[0063] Multi-point synchronous data acquisition is performed on the highway drainage and waterproofing structure using distributed sensing units. Sensing nodes are deployed at key cross-sections, expansion joints, drainage ditch walls, and waterproofing layer locations. Each monitoring point synchronously outputs deformation and seepage parameters. Deformation parameters include displacement, strain, and settlement, while seepage parameters include pore water pressure, seepage velocity, and moisture content. The deployment density of the distributed sensing units should satisfy the spatial sampling theorem to ensure sufficient spacing between adjacent monitoring points to capture the spatial variations in structural deformation, avoiding the loss of information on local abrupt changes due to sparse sampling.

[0064] Based on the actual spatial coordinates of the monitoring points, all monitoring points are mapped onto a three-dimensional spatial grid. The grid cell division is determined jointly based on the structural geometric characteristics and the distribution density of the sensing nodes. The average value of the deformation parameter within each grid cell is taken as the representative value. Then, the deformation difference between adjacent grid cells is calculated. The absolute value of the difference reflects the degree of deformation non-uniformity in adjacent areas. The differences in each direction are summarized to construct a spatial gradient field. The spatial gradient field stores the gradient direction and magnitude information at each grid node in vector form. The spatial gradient field is decomposed directionally to extract the gradient component along the radial direction of the structure, i.e., the radial gradient component. This component is most sensitive to the deformation response perpendicular to the structural surface and can effectively reflect the normal deformation trend of the drainage and waterproofing structure under stress or seepage.

[0065] After obtaining the radial gradient components, abrupt change regions are identified by setting local abrupt change criteria. Specifically, the deviation between the radial gradient component of each grid node and the mean of its neighborhood is calculated. When the deviation exceeds a certain multiple of the neighborhood standard deviation, the region where the node is located is determined to be a local abrupt change region. For all identified local abrupt change regions, the amplitude of the radial gradient component at the corresponding grid node is extracted and summarized to form a gradient feature vector. Each dimension of this vector corresponds to the gradient amplitude of a abrupt change region, directly representing the degree of deformation at that location.

[0066] The seepage parameters are segmented along the time axis, dividing the complete monitoring period into several equal-length sub-time periods. The seepage parameters within each sub-time period are accumulated and summed to obtain the cumulative seepage flow rate for that time period. The ratio of the difference in cumulative seepage flow rate between adjacent time periods to the length of the time period is the cumulative seepage flow rate change rate. When the change rate of a certain time period exceeds a preset threshold, that time period is marked as an abnormal period. The threshold can be determined based on the statistical distribution of historical monitoring data, taking the high percentile of the change rate distribution under normal operating conditions as a reference upper limit. A fast Fourier transform is performed on the seepage parameter sequence within the abnormal period to obtain frequency domain characteristics. The dominant frequency component and its energy distribution in the frequency domain characteristics can reveal the periodic pattern of seepage fluctuations, distinguishing between temporary seepage increases caused by external rainfall excitation and persistent seepage anomalies caused by structural deterioration. The frequency domain characteristics are combined with the cumulative seepage flow rate of each time period to form seepage characteristics. The seepage characteristics contain both cumulative information in the time domain and dynamic response information in the frequency domain, comprehensively describing the time-frequency characteristics of seepage behavior.

[0067] After independently extracting gradient and seepage features, target monitoring points are located in the monitoring point grid based on local abrupt change regions. Target monitoring points are the nodes with the largest gradient amplitudes within the abrupt change regions; these nodes are considered to be the locations where structural state changes are most significant. Seepage features corresponding to the target monitoring points are extracted, resulting in a subset of target seepage features. The correlation between the gradient feature vector and the target seepage feature vector is calculated. The correlation is measured using a normalized cross-correlation method. Let the gradient feature vector be... The target seepage feature vector is Then the normalized correlation between the two is for: ,in express transpose, and These are the Euclidean norms of the two vectors, respectively. Correlation degree. The range of values ​​is The closer the value is to 1, the higher the degree of coordination between deformation abrupt change and seepage anomaly in space and time; the closer it is to 0, the more independent the two are.

[0068] Based on correlation Constructing the coupling matrix The coupling matrix is ​​constructed by performing an outer product operation between each component of the gradient eigenvector and each component of the target seepage eigenvector, and then multiplying by the correlation degree. As a weighting coefficient, i.e. ,in This represents the outer product operation. (Coupling matrix) Each element reflects the coupling strength between the corresponding gradient component and the seepage component. When the correlation is low, the overall magnitude of the coupling matrix is ​​compressed to avoid excessive interference from weak related information in subsequent feature fusion.

[0069] The gradient features and seepage features are coupled with a tensor operation through a coupling matrix to obtain the structural state features. The specific form of the tensor operation is to convert the gradient feature vector... With seepage eigenvector These are respectively used as the two modal inputs of the tensor, where, To obtain complete seepage characteristics, time-frequency information from all monitoring points is included, using a coupling matrix. As an intermodal interaction kernel, it performs multilinear mapping operations and outputs a structural state feature tensor. Structural state characteristic tensor In the spatial dimension, the location information of each monitoring point is preserved, and in the feature dimension, the joint distribution of deformation gradient and seepage time-frequency response is encoded simultaneously, providing a high-dimensional and physically meaningful input representation for subsequent graph propagation operations based on drainage path topology and hierarchical aggregation operations based on waterproofing layer relationships.

[0070] In practical engineering applications, the monitoring environment of highway drainage and waterproofing structures is complex. Sensing units may be affected by vibration interference or signal transmission attenuation, leading to missing or abnormally fluctuating deformation or seepage parameters at some monitoring points. To address this, the raw data must be preprocessed before constructing the spatial grid. This includes missing value imputation based on neighborhood interpolation and outlier removal based on the 3σ criterion. This ensures the data used for spatial gradient field calculation and time-series cumulative analysis has sufficient integrity and reliability, thereby guaranteeing the quality of gradient and seepage feature extraction. Ultimately, this ensures that the final structural state characteristics accurately reflect the actual health status of the highway drainage and waterproofing structure.

[0071] In one alternative implementation,

[0072] Based on the pre-constructed drainage path topology, graph propagation operations are performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operations are performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector, including:

[0073] The structural state features are mapped to each node of the drainage path topology to obtain node state features. The drainage path topology is spectrally decomposed to obtain a Laplace feature vector set. The node state features are projected onto the Laplace feature vector set to obtain spectral coefficients and spectral propagation is performed to obtain propagated node features. The gradient abrupt change positions of the propagated node features are identified and the corresponding spectral coefficients are extracted to form drainage connectivity features.

[0074] The structural state features are assigned to each level of the waterproofing hierarchy to obtain the hierarchical state features. A hypergraph representation corresponding to the waterproofing hierarchy is constructed, and the hierarchical state features are propagated by hyperedges to obtain hyperedge enhancement features. The hyperedge enhancement features are decomposed by tensors to obtain core tensors, and the residual between the core tensors and the hyperedge enhancement features is calculated. Based on the residual, core tensor slices are extracted to form waterproofing degradation features.

[0075] Tensor contraction is performed on the drainage connectivity feature and the waterproofing degradation feature to obtain a contraction tensor. An attention distribution is calculated on the contraction tensor and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is performed on the coupled feature tensor to obtain drainage components and waterproofing components, and a cooperative state vector is obtained by solving the problem.

[0076] Mapping structural state features to nodes in the drainage path topology requires assigning structural state features to the corresponding nodes in the topology graph based on the spatial relationships of physical drainage elements such as drainage pipe networks, drainage ditches, and sump pits, forming a set of node state features. The drainage path topology is represented by a graph structure, where nodes represent key drainage locations, edges represent drainage channel connections, and node state features are the mapping results of deformation and seepage fusion information at that location onto the graph structure.

[0077] Spectral decomposition of the drainage path topology is performed to construct the normalized Laplace matrix of the topology graph. Then, eigenvalue decomposition is performed to obtain a set of Laplace eigenvectors composed of eigenvectors. The corresponding set of eigenvalues ​​is denoted as The node state feature matrix Project to The spanned spectral space yields the spectral coefficient matrix. ,satisfy The spectral propagation process is achieved by applying spectral filtering to the spectral coefficients; the filtering function is based on eigenvalues. The design enables selective enhancement of low-frequency smooth components and high-frequency abrupt components in the frequency domain. Post-propagation node characteristics. Depend on The calculation yielded, where This is the spectral domain filtering matrix.

[0078] Identify post-propagation node features For each node at a gradient abrupt change location, the first-order difference of its eigenvalues ​​is calculated along the topological path. When the difference magnitude exceeds a preset threshold, a gradient abrupt change is determined to exist at that location. The spectral coefficients corresponding to that location are extracted, and the spectral coefficients of all abrupt change locations are aggregated to form a drainage connectivity feature vector. The locations of gradient abrupt changes physically correspond to sections in the drainage path where there are blockages, breaks, or connectivity anomalies. The high-frequency components in the spectral coefficients reflect the degree of local anomalies in drainage connectivity, while the low-frequency components characterize the overall decline trend of global drainage capacity.

[0079] Structural state features are assigned to each level of the waterproofing hierarchy. Based on the physical hierarchy of the highway waterproofing structure, including surface waterproofing layers, base waterproofing layers, and structural waterproofing layers, structural state features are assigned values ​​according to their spatial location and hierarchical affiliation, resulting in a set of hierarchical state features for each level. The waterproofing hierarchy is represented in the form of a hypergraph, where hyperedges can simultaneously connect nodes from multiple different levels, reflecting the cross-level coupling effect between waterproofing layers. The construction of the hypergraph representation is based on the physical dependencies between waterproofing layers. When a waterproofing layer deteriorates, the waterproofing performance of adjacent layers is also affected; this multi-body relationship is modeled using hyperedges.

[0080] Hyperedge propagation is performed on the hierarchical state features, and the features of the nodes associated with each hyperedge are weighted and aggregated. The aggregation weight is determined based on the association strength between the node and the hyperedge. After hyperedge propagation, each node obtains a hyperedge-enhanced feature that incorporates cross-level information. The super-edge propagation operation can effectively capture the cooperative degradation patterns between waterproofing layers. For example, local damage to the surface waterproofing layer often accelerates the seepage of the base waterproofing layer. This transmission relationship is explicitly expressed in the super-edge enhancement feature.

[0081] Hyperedge enhancement features Perform tensor decomposition, and The organization is in the form of a third-order tensor, expanded along the hierarchical, positional, and feature dimensions respectively, and the core tensor is obtained using Tucker decomposition. and modal factor matrices. Core tensor The main structural information of the superedge enhancement features is compressed, reflecting the dominant mode of waterproofing degradation. Residual tensor It is calculated from the difference between the hyperedge enhancement features and the core tensor reconstruction results. ,in , , These are the factor matrices for the three modes, Indicates along the first Modal tensor product operation. The residual tensor contains local degradation details not captured by the core tensor. Core tensor slices are extracted based on the residuals, and the core tensor slices corresponding to the slices with larger residual amplitudes are selected. These slices are then concatenated to form a waterproof degradation feature vector. This allows the waterproofing degradation characteristics to simultaneously include both the globally dominant degradation mode and local abnormal degradation information.

[0082] drainage connectivity characteristics Characteristics of waterproofing degradation Perform tensor shrinkage operations, expanding both into higher-order tensors, and then shrinking them along the shared spatial dimension. The joint distribution information of drainage connectivity and waterproofing degradation at various spatial locations is preserved. The physical significance of the tensor contraction operation lies in associating the drainage state and the waterproofing state in the same spatial coordinate system, eliminating the dimensionality mismatch between the two in the feature space, and providing a unified tensor representation basis for subsequent interactive mapping.

[0083] For the contraction tensor Calculate the attention distribution, After flattening along the feature dimension, the input to the attention scoring function yields the attention weight matrix at each position. Attention weights reflect the importance of the combined state of drainage and waterproofing at different spatial locations to the overall structural health. When performing bilinear interaction mapping based on attention distribution, attention weights are used... The contraction tensor is weighted and then subjected to a bilinear transformation matrix. By cross-mapping the drainage component and the waterproofing component, a coupled feature tensor is obtained. ,satisfy ,in This represents an element-wise weighted operation. Bilinear interaction mapping can explicitly model the nonlinear coupling relationship between drainage and waterproofing states, avoiding the loss of interaction information caused by simple splicing methods.

[0084] For coupled feature tensors Perform mode decomposition using nonnegative matrix decomposition or singular value decomposition. It is decomposed into several modal components, among which the set of components corresponding to the physical meaning of drainage constitutes the drainage component. The set of components corresponding to the physical meaning of waterproofing constitutes the waterproofing component. The physical assignment of modal components is based on the connectivity characteristics of each component with the original drainage system. and waterproofing degradation characteristics The correlation is determined, and components with a correlation higher than a preset threshold are classified into their corresponding categories. Drainage components are then... With waterproof component After concatenation and linear transformation, the cooperative state vector is obtained. This vector simultaneously encodes the current health status of the drainage system and the waterproofing system, as well as their interaction information, providing the basic input for subsequent calculation of the interaction coefficients and the corrected state vector.

[0085] In one alternative implementation,

[0086] The attention distribution is calculated for the contraction tensor, and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is then performed on the coupled feature tensor to obtain drainage and waterproofing components, and the cooperative state vector is obtained by solving for these components.

[0087] The contraction tensor is sliced ​​along the drainage path dimension and the waterproofing layer dimension to obtain drainage dimension slice group and waterproofing dimension slice group respectively. Self-attention calculation is performed on the drainage dimension slice group and the drainage connectivity feature to obtain drainage self-attention feature. Self-attention calculation is performed on the waterproofing dimension slice group and the waterproofing degradation feature to obtain waterproofing self-attention feature. Cross-attention calculation is performed on the drainage self-attention feature and the waterproofing self-attention feature, and sparse attention distribution is obtained after gated linear unit transformation and normalization.

[0088] The drainage connectivity features and the waterproofing degradation features are projected into a high-dimensional feature space and weighted and aggregated based on the sparse attention distribution to obtain attention enhancement features. The attention enhancement features are then propagated through a graph neural network to obtain topology enhancement features and mapped through a bilinear pooling layer to obtain a coupled feature tensor.

[0089] The coupled feature tensor is decomposed by Tucker to obtain a core tensor and multiple factor matrices. Drainage-related modal slices are extracted from the core tensor and reconstructed with the corresponding factor matrices to obtain drainage components. Waterproofing-related modal slices are extracted and reconstructed with the corresponding factor matrices to obtain waterproofing components. The drainage components and waterproofing components are fused through residual connection and then processed by layer normalization to obtain the cooperative state vector.

[0090] The shrinkage tensor is sliced ​​along the drainage path dimension and the waterproofing layer dimension, splitting the 3D tensor into two structured subsets according to physical semantics. When slicing along the drainage path dimension, the indices of the waterproofing layer dimension and the sensor state dimension are fixed, and 2D slices corresponding to each drainage path are extracted sequentially, forming a drainage dimension slice group. When slicing along the waterproofing layer dimension, the indices of the drainage path dimension and the sensor state dimension are fixed, and 2D slices corresponding to each waterproofing layer are extracted sequentially, forming a waterproofing dimension slice group. Both sets of slices retain the local structural information of the shrinkage tensor in their respective physical dimensions, providing a physically meaningful input for subsequent self-attention calculations.

[0091] Drainage dimension slice group and drainage connectivity feature vector They jointly participate in the self-attention calculation. Each slice in the drainage dimension slice group is flattened and then stitched together. As the input source for the query matrix, key matrix, and value matrix, three sets of representations (query, key, and value) are generated through linear projection. After calculating the dot product attention score and normalizing it using softmax, the weighted aggregated value matrix yields the drainage self-attention feature. This allows the state information at each location along the drainage path to be mutually perceptive, capturing long-range dependencies between path nodes. Waterproofing dimension slice group and waterproofing degradation feature vector. By participating in self-attention calculation in the same way, waterproof self-attention features are obtained, which allows the degradation transmission relationship between waterproof layers to be explicitly expressed in the feature space.

[0092] Cross-attention calculation is performed on drainage self-attention features and waterproofing self-attention features. Drainage self-attention features are used as the query source, and waterproofing self-attention features are used as the key and value source to calculate cross-modal attention scores, capturing the dependency between drainage and waterproofing states. The cross-attention output is then subjected to non-linear filtering via gated linear unit transformation. The gating mechanism uses a sigmoid activation function to control the opening and closing ratio of information channels, suppressing redundant feature responses unrelated to structural health. Normalization is performed using layer normalization, standardizing the mean and variance of the feature vector for each sample to eliminate numerical distribution shifts caused by differences in the dimensions of different physical quantities, resulting in a sparse attention distribution. In this process, a large number of attention weights are compressed to near zero by the gating mechanism, retaining only the interaction terms that make a significant contribution to the structural state judgment.

[0093] Drainage connectivity feature vector With waterproof degradation feature vector Each feature is projected onto a high-dimensional feature space through a linear transformation to expand the feature dimension and enhance expressive power. The projected high-dimensional features are then distributed according to a sparse attention distribution. Weighted aggregation is performed, linearly combining each feature dimension according to its attention weight. This strengthens high-weighted interaction patterns in the aggregation result while effectively suppressing low-weighted noise patterns, thus obtaining attention-enhanced features. .

[0094] Attention enhancement features The input graph neural network performs topology propagation. The graph neural network uses the structural dependency graph as the propagation topology, and... As initial node features, feature information from neighboring nodes is aggregated through multiple rounds of message passing, integrating the feature vector of each structural location into its neighborhood context within the drainage topology, thus obtaining topology-enhanced features. The topology-enhanced features are then mapped through a bilinear pooling layer, which... The two sub-vectors are used as the left and right inputs, respectively, and are transformed by the bilinear transformation matrix. The outer product is calculated and flattened to capture the second-order interaction information between feature dimensions. The mapping result is then subjected to dimensionality compression and nonlinear activation to obtain the coupled feature tensor. Coupled feature tensor It also encodes the high-order cooperative relationship between drainage connectivity and waterproofing degradation, providing complete input for subsequent modal decomposition.

[0095] For coupled feature tensors Perform Tucker decomposition, decomposing it into the core tensor. With three factor matrices , , and residual tensor ,satisfy ,in Indicates along the first Modal tensor multiplication. Core tensor. Compressed The main change patterns are represented by factor matrices corresponding to the basis spaces of the drainage path dimension, waterproofing layer dimension, and sensor state dimension, respectively. Tucker decomposition is solved iteratively using alternating least squares. In each iteration, the remaining factor matrices are fixed, and the principal components are extracted by singular value decomposition of the expanded matrix of the current mode. This process is repeated until the reconstruction error is lower than a preset threshold.

[0096] From the core tensor When extracting drainage-related modal slices, the corresponding slice subset is taken along the first mode (drainage path dimension), and compared with the factor matrix. Row vectors related to drainage path and Tensor reconstruction is performed on the column vectors related to the sensor state to obtain the drainage component. Its dimension is the same as the drainage connectivity feature vector. To maintain consistency, the physical meaning is the pure drainage function state representation decoupled from the coupling tensor. When extracting waterproof-related modal slices, the corresponding slice subset is taken along the second mode (waterproofing layer dimension), and compared with the factor matrix. Row vectors related to intermediate waterproofing layers and Tensor reconstruction is performed on the column vectors related to the sensor state to obtain the waterproof component. The physical meaning is a pure waterproof functional state representation decoupled from the coupling tensor. The extraction process of both components retains the intermodal interaction information captured by Tucker decomposition, avoiding the information aliasing problem caused by simple feature splicing.

[0097] Drainage component With waterproof component The fusion is performed using residual connections, directly adding the two features and then superimposing the residuals onto the flattened vector of the input shrinkage tensor. This preserves details in the original features that were not fully captured by Tucker decomposition, preventing gradient vanishing or information loss during deep feature extraction. The residual fusion result is then normalized, and the mean is calculated for each element of the fusion vector. with standard deviation ,according to After standardization, an affine transformation is performed using learnable scaling and offset parameters, where... The numerical stability constant is used to finally obtain the cooperative state vector. Cooperative state vector It maintains stability in numerical distribution and simultaneously carries the decoupled representation of drainage function state and waterproof function state and their coupling interaction information in semantics, providing a high-quality input foundation for subsequent interaction coefficient calculation and modified state vector generation.

[0098] In one alternative implementation,

[0099] The interaction coefficient is calculated based on the drainage component and the waterproofing component in the cooperative state vector. The degree of transmission influence of drainage failure on waterproofing performance and the degree of reverse constraint of waterproofing damage on drainage capacity are determined based on the interaction coefficient, including:

[0100] The drainage component and the waterproofing component are separated from the cooperative state vector and expanded into drainage state sequence and waterproofing state sequence, respectively. The drainage state sequence and the waterproofing state sequence are time-aligned and the covariance is calculated under different lag time differences to obtain the covariance matrix. The positive conduction intensity of drainage to waterproofing and the reverse conduction intensity of waterproofing to drainage are extracted from the covariance matrix and combined to obtain the bidirectional interaction coefficient.

[0101] The drainage component is mapped to the nodes of the drainage path topology and the failure node is identified to obtain the failure time. The waterproof component is mapped to each waterproof layer and the amount of degradation change after the failure time is calculated to obtain the degradation increment sequence. The degradation increment sequence is multiplied element by element based on the positive conduction intensity in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the degree of conduction influence.

[0102] The failure time is obtained by identifying the damaged layer from the waterproof layer that has deteriorated beyond the preset deterioration threshold. The change in drainage capacity of the node after the failure time is calculated as a capacity decay sequence. The capacity decay sequence is multiplied element by element based on the reverse conduction strength in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the reverse constraint degree.

[0103] Separating the drainage and waterproofing components from the cooperative state vector is the starting point for calculating the interaction coefficients. After coupling mapping, the cooperative state vector carries both drainage connectivity information and waterproofing degradation information. By applying the corresponding projection matrix to the cooperative state vector, the drainage and waterproofing components are extracted separately. The drainage component is expanded into a drainage state sequence in chronological order, and the waterproofing component is expanded into a waterproofing state sequence in chronological order. Each sequence records the dynamic evolution of the drainage connectivity state and the waterproofing layer state at different acquisition times. Since sensor acquisition frequency or data transmission delay may cause slight deviations in the timestamps of the two sequences, time alignment processing is required for the drainage and waterproofing state sequences. Linear interpolation is used to unify the two sequences to the same time resolution, ensuring that subsequent covariance calculations are performed under the same time reference.

[0104] After time alignment is completed, at different lag time differences The covariance between the drainage state sequence and the waterproofing state sequence is calculated separately to form a covariance matrix. Let the drainage state sequence be... The waterproof status sequence is ,in Given the sequence length, then the time difference is... The covariance is calculated as follows: ,in and These are the means of the corresponding sequences. When When the covariance reflects the strength of the correlation between changes in drainage and changes in waterproofing, it corresponds to a positive transmission relationship between drainage and waterproofing; when In this context, covariance reflects the strength of the correlation between changes in waterproofing and changes in drainage, corresponding to the reverse transmission relationship between waterproofing and drainage. From the covariance matrix... The maximum covariance value of forward conduction is extracted as the forward conduction intensity. The maximum covariance value of the reverse propagation is extracted as the reverse propagation strength. The two are combined to form a two-way interaction coefficient. This fully depicts the two-way coupling relationship between drainage and waterproofing.

[0105] After obtaining the bidirectional interaction coefficient, the degree of transmission impact of drainage failure on waterproofing performance is calculated. Drainage components are mapped to nodes in the drainage path topology. By comparing the drainage connectivity state value of each node with a preset connectivity threshold, the set of nodes where drainage connectivity has failed is identified, and the earliest failure time corresponding to each failed node is recorded. For the waterproofing component, it is mapped to each waterproofing layer, and the failure time of each waterproofing layer is extracted. The waterproofing degradation values ​​at various subsequent times, and The degradation increment sequence is obtained by subtracting the degradation baseline value at time 1. This sequence reflects the incremental change in the degree of degradation of the waterproofing layer relative to the time of failure after drainage failure. Based on positive conduction strength... Perform element-wise multiplication on the degraded increment sequence to obtain the weighted degraded increment sequence. Then, the weighted degradation increment sequence is accumulated and summed along the time dimension to obtain the degree of transmission impact. Degree of transmission influence The value of directly reflects the cumulative damage contribution of drainage failure to the waterproofing layer through mechanisms such as water penetration and stress redistribution. The larger the value, the more significant the negative transmission of drainage failure to waterproofing performance.

[0106] Calculate the degree to which waterproofing damage negatively impacts drainage capacity, and identify degradation values ​​exceeding preset degradation thresholds from the degradation states of each waterproofing layer. The damage level is recorded, and the earliest damage time corresponding to each damage level is recorded. The method for determining the moment of failure is to iterate through the time-series degradation curves mapped from the waterproof components to each layer, and find the point where the degradation value first exceeds... The time of failure is taken as the failure time of that level. The failure time of each node in the drainage path topology is then used as the failure time. The subsequent drainage capacity value is subtracted from the baseline drainage capacity value at the time of failure to obtain the capacity decay sequence. Negative values ​​in the capacity decay sequence indicate a decrease in drainage capacity, while positive values ​​indicate abnormal fluctuations in drainage capacity. This is based on the reverse conduction strength. Perform element-wise multiplication on the capacity decay sequence to obtain the weighted decay sequence. Then, the weighted decay sequence is accumulated and summed along the time dimension to obtain the degree of reverse constraint. Degree of reverse constraint The study quantifies the sustained limiting effect of waterproofing layer damage on drainage capacity by altering the internal seepage field of the structure and increasing the head pressure of the drainage system. The larger the value, the more severe the reverse inhibition of drainage capacity caused by waterproofing damage.

[0107] Degree of transmission influence With reverse constraint degree Together, these factors form the basis for a quantitative description of the bidirectional coupled failure of drainage and waterproofing structures. In practical engineering applications, under long-term loads and hydrological conditions, drainage failure and waterproofing damage in highway drainage and waterproofing structures often do not occur in isolation, but rather are mutually induced and cyclically exacerbated. By introducing a bidirectional interaction coefficient and calculating the transmission quantification index in both directions, it is possible to accurately determine which failure mode dominates in the current structural state, thus providing a basis for subsequently correcting the direction and magnitude of the state vector. Significantly greater than When this occurs, it indicates that drainage failure is the main driving factor in the current damage evolution, and the weight of drainage connectivity-related components should be increased during correction; when Significantly greater than When the values ​​are close, it indicates that the constraint of waterproofing damage on drainage capacity has become the main contradiction. When correcting, the contribution ratio of waterproofing degradation-related components should be adjusted first. When the values ​​are close, it indicates that the failure of the two-way coupling between drainage and waterproofing is in the stage of equilibrium evolution. Symmetry correction of the cooperative state vector is required to preserve the integrity of the two-way coupling information.

[0108] At the engineering implementation level, a pre-set degradation threshold is implemented. The value needs to be determined comprehensively based on the performance degradation characteristics of the specific waterproofing material and structural design specifications. Different thresholds can be set for different waterproofing layers to accommodate the failure modes of different materials. (Lapse time difference) The search range should be determined based on the typical response delay intervals of drainage and waterproofing events in historical monitoring data to avoid introducing invalid covariance values ​​due to an overly broad search range, which could interfere with the accuracy of the two-way interaction coefficient. Furthermore, for monitoring data with a short acquisition time span, the sequence length... When the time covariance is small, statistical instability exists. The robustness of the two-way interaction coefficient can be improved by averaging the covariance results of multiple time periods using a sliding window method.

[0109] Figure 2 This is a flowchart illustrating the modified state vector generation process of the intelligent structure health monitoring method for integrated drainage and waterproofing of highways, as described in an embodiment of the present invention.

[0110] In one alternative implementation,

[0111] The modified state vector is obtained by modifying the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint, including:

[0112] The conduction intensity value is extracted from the degree of conduction influence to construct a conduction intensity matrix, and the constraint intensity value is extracted from the degree of reverse constraint to construct a constraint intensity matrix. The conduction intensity matrix and the constraint intensity matrix are subjected to asymmetric coupling analysis to obtain a coupling intensity field. The coupling intensity field is decomposed to obtain the conduction vector and the constraint vector, and the vector angle is calculated to identify the conduction-dominant cooperative region and the constraint-dominant antagonistic region.

[0113] The pre-obtained connectivity failure nodes are taken as drainage failure nodes, and the pre-obtained damage levels are taken as damage waterproofing levels. In the dominant conduction cooperative region, a dominant conduction path is constructed from the drainage failure nodes to the corresponding waterproofing levels along the conduction vector, and the path attenuation coefficient is calculated. In the dominant constraint antagonistic region, a reverse blocking path is constructed from the damage waterproofing levels to the corresponding drainage nodes along the constraint vector, and the path damping coefficient is calculated.

[0114] The failure state change of the drainage failure node is extracted and accumulated along the dominant conduction path according to the path attenuation coefficient to the corresponding waterproofing level to obtain the corrected waterproofing component. The damage and deterioration change of the damaged waterproofing level is extracted and accumulated along the reverse blocking path according to the path damping coefficient to the corresponding drainage node to obtain the corrected drainage component. The corrected waterproofing component and the corrected drainage component are spliced ​​together to obtain the corrected state vector.

[0115] The conduction intensity value is extracted from the degree of conduction influence. For each pair of drainage failure nodes and waterproofing layers, the corresponding degree of conduction influence is quantified into a scalar intensity value, and arranged in matrix form according to the node number and layer number to obtain the conduction intensity matrix. ,in The number of drainage failure nodes. This represents the number of waterproofing layers. Similarly, the constraint strength values ​​are extracted from the degree of reverse constraint and arranged in matrix form according to the correspondence between the waterproofing layer number and the drainage node number to obtain the constraint strength matrix. The dimensions of the two matrices are transposes of each other, but they are not numerically symmetrical. This is because the transmission effect of drainage failure on waterproofing performance and the reverse constraint effect of waterproofing damage on drainage capacity have different directions in physical mechanisms, and the two cannot be simply regarded as symmetrical coupling.

[0116] An asymmetric coupling analysis is performed on the conduction strength matrix and the constraint strength matrix, and the two matrices are jointly represented in a unified node-hierarchy space to construct an asymmetric coupling matrix. Its upper left block corresponds to the internal association between drainage nodes, and the upper right block is filled with... Fill in the bottom left block The bottom right block corresponds to the internal relationships between waterproofing layers. Perform singular value decomposition to obtain the left singular vector matrix. Singular value diagonal matrix and right singular vector matrix The magnitude of singular values ​​measures the coupling strength in each decomposition direction, and the whole constitutes a coupling strength field. The coupling strength field spatially describes the distribution of the degree of mutual influence between the drainage and waterproofing subsystems, with the direction with larger singular values ​​corresponding to the propagation mode of the strongest coupling between the systems.

[0117] The method for extracting the conduction vector and constraint vector from the coupling strength field is as follows: The column vector corresponding to the largest singular value is used as the propagation vector. This indicates the dominant direction of the transmission of drainage failure to the waterproofing layer; with The column vector corresponding to the largest singular value is used as the constraint vector. This indicates the dominant direction of the reverse constraint of waterproofing damage on drainage nodes. Calculation and The angle between The included angle is obtained by taking the inverse cosine function of the ratio of the inner product of the two elements to the product of their moduli, i.e. .when When the values ​​are small, the directions of the conduction vector and the constraint vector tend to be consistent, indicating that drainage failure and waterproofing damage exacerbate each other in the same area, corresponding to the conduction-dominated synergistic region; when... When the value is large, the two vectors are oriented in opposite directions, indicating that the restraining effect of waterproofing damage and the conductive effect of drainage failure are mutually antagonistic in this area, corresponding to the dominant antagonistic area. A preset included angle threshold is used. As a boundary, the coupling intensity field is divided into two types of regions: the conduction-dominant cooperative region and the constraint-dominant antagonistic region.

[0118] In the conduction-dominant coordinated region, the connectivity failure node is taken as the drainage failure node, and the damaged level is taken as the damaged waterproofing level, along the conduction vector. Starting from each drainage failure node, the search proceeds hop-by-hop according to the edge connections in the structural dependency graph, prioritizing nodes with... The adjacent node or level with the largest directional projection is used as the next hop until the corresponding waterproofing level is reached, thus constructing a dominant conduction path. For each dominant conduction path, the path attenuation coefficient is calculated. Its value is obtained by multiplying the conduction strength values ​​of each hop along the path and then taking the logarithm and normalizing it. It reflects the degree of energy attenuation when the drainage failure state is conducted to the waterproofing layer along the path. The longer the path and the smaller the conduction strength at intermediate nodes, the better. The smaller.

[0119] In the region of constraint-dominant confrontation, along the constraint vector Starting from each damaged waterproofing layer, the search proceeds hop-by-hop according to the edge connections in the structural dependency graph, prioritizing layers that are compatible with the existing waterproofing layers. The neighboring node with the largest directional projection is used as the next hop until the corresponding drainage node is reached, thus constructing a reverse blocking path. For each reverse blocking path, the path damping coefficient is calculated. Its value is obtained by multiplying the constraint strength values ​​of each jump along the path and then taking the logarithm and normalizing it. It reflects the degree of attenuation when the waterproofing failure state applies reverse blocking to the drainage node along the path. The longer the path and the weaker the constraint strength, the better. The smaller.

[0120] Extract the change in failure state of drainage failure nodes At that time, the difference between the drainage component of the node at the current moment and the drainage component under the baseline normal state is used to obtain the state deviation vector caused by the failure. This vector is then propagated along the corresponding dominant transmission path. According to path attenuation coefficient The weighted and accumulated waterproofing component is added to the waterproofing layer corresponding to the end point of the path, which is the corrected waterproofing component. satisfy: ,in This represents the waterproofing component before correction. When multiple dominant conduction paths converge at the same waterproofing layer, the contribution of each path is weighted and summed. The weight is the normalized value of the attenuation coefficient of each path, avoiding excessive correction due to repeated superposition.

[0121] Extract the change in damage and deterioration of the damaged waterproofing layer. The difference between the waterproof component of this layer at the current moment and the waterproof component under normal conditions is used to obtain the state deviation vector caused by the damage. Along the corresponding reverse blocking path, Based on path damping coefficient The weighted average is then added to the drainage component corresponding to the drainage node at the end of the path, which is the corrected drainage component. satisfy ,in This represents the drainage component before correction. When multiple reverse blocking paths converge at the same drainage node, the contributions of each path are also normalized according to the path damping coefficient and then weighted and summed.

[0122] Complete the correction of waterproofing components With modified drainage component After calculation, the two are concatenated along the feature dimension to obtain the corrected state vector. The splicing operation maintains the semantic independence of the drainage and waterproofing subsystems, while reflecting the state coupling correction achieved between them through dominant transmission paths and reverse blocking paths at the vector level. (Corrected state vector) Compared to the original cooperative state vector, it more accurately reflects the true health status of the drainage and waterproofing structure under bidirectional coupling, providing a more reliable input for subsequent damage propagation path identification and health monitoring report generation.

[0123] In one alternative implementation,

[0124] Based on the structural location corresponding to the corrected state vector, associated nodes are identified in a pre-established structural dependency graph. Backtracking analysis of the seepage characteristics corresponding to the associated nodes is performed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path, and a health monitoring report is generated, including:

[0125] The corrected drainage component and the corrected waterproof component are extracted from the corrected state vector and the corresponding structural positions are obtained by parsing. The structural positions are mapped to the pre-established structural dependency graph to obtain the position nodes. The structural dependency graph is topologically sorted to obtain the dependency level. Based on the dependency level, the predecessor nodes are traced upward from the position nodes. The predecessor nodes are connected to the nodes and nodes that form a strong connected component with the position nodes are identified as associated nodes.

[0126] Extract the seepage features corresponding to the associated nodes, perform reverse traversal over time, and calculate the feature change gradient of adjacent time points. Identify the time when the feature change gradient exceeds a preset gradient threshold as the damage initiation time and record the corresponding node as the damage initiation node. Search for a path from the damage initiation node along the structural dependency graph to the location node and determine the damage propagation path.

[0127] The seepage characteristics of adjacent nodes in the damage propagation path are extracted and the time difference between adjacent nodes at the damage initiation time is calculated to obtain the propagation delay. The structural distance between adjacent nodes is extracted from the structural dependency graph and the propagation rate is calculated based on the propagation delay. The damage propagation path and the propagation rate are combined with the corrected state vector to generate a health monitoring report.

[0128] The corrected drainage and corrected waterproofing components are extracted from the corrected state vector. The corresponding structural location coordinates are obtained by parsing the spatial coding information of each component. Each component has a pre-embedded structural location identifier during generation; the parsing process uses a location index table to restore the location code in the component to specific structural coordinates or component numbers. The parsed structural locations are mapped to a pre-established structural dependency graph, and the corresponding nodes are located in the graph, called location nodes. The structural dependency graph describes the physical and functional dependencies between structural components in the form of a directed graph. Nodes represent independent structural units (such as drainage pipe sections, waterproofing layer zones, expansion joint sections, etc.), and directed edges represent the dependency direction between components, i.e., the downstream component depends on the normal working state of the upstream component.

[0129] Perform a topological sort on the structural dependency graph to obtain the dependency level of each node. The topological sort can be implemented using Kahn's algorithm or a depth-first search, placing nodes with an in-degree of zero at the highest level and assigning level numbers downwards sequentially, ultimately obtaining the level value of each node. ,in The variable is a non-negative integer; smaller values ​​indicate higher levels (closer to the source of the dependency chain). Based on the dependency hierarchy, the system traces upwards from the location node along the reverse direction of the directed edges, visiting its predecessor nodes layer by layer. This means that all nodes in the directed graph that point to the location node are considered as points. The tracing process uses a breadth-first search strategy, with a maximum tracing depth. To avoid excessive computational overhead in large-scale graphs, only retain hierarchical differences not exceeding a certain threshold. The set of predecessor nodes.

[0130] Connectivity testing is performed on the set of predecessor nodes obtained through tracing, and the Tarjan algorithm is used to identify strongly connected components in the structural dependency graph. A strongly connected component is a maximum subgraph in which a directed path exists between any two nodes. Location nodes are included in the detection scope, and predecessor nodes belonging to the same strongly connected component as location nodes are identified and marked as associated nodes. The physical meaning of associated nodes is that there is a mutual reachability dependency between the structural component and the location where the anomaly occurred, and its state change has a direct or indirect impact on the location node; therefore, it needs to be included in the scope of damage backtracking analysis.

[0131] Extract the time-series sequence of seepage characteristics corresponding to the associated nodes, and traverse it backwards along the time axis, i.e., gradually traverse back from the current monitoring time to the historical time. For each associated node, calculate the gradient of seepage characteristics changes between adjacent time points. ,in Indicates the first The time step and the first The difference norm of the seepage eigenvectors between time steps reflects the degree of change in the seepage state at that moment. With preset gradient threshold When comparing, When the time is reached, this time is marked as a candidate time for damage initiation, and the corresponding node is recorded as a candidate node for damage initiation. Among all candidate times, the time with the earliest time and the most significant gradient change is selected as the final damage initiation time. The corresponding node is used as the damage initiation node. The determination of the damage initiation time adopts a dual criterion: on the one hand, the gradient must exceed the threshold, and on the other hand, the seepage characteristics must remain relatively stable for several consecutive time steps before that time, thereby eliminating misjudgments caused by occasional disturbances.

[0132] Starting from the damage initiation node, a path is searched towards the location node along the directed edges of the structural dependency graph. A depth-first search algorithm is used to enumerate all directed paths from the damage initiation node to the location node. The paths are then sorted according to the sum of the gradients of the seepage characteristic changes at each node along the path, and the path with the largest cumulative gradient value is selected as the damage propagation path. This path represents the channel where the damage signal propagates most significantly in the structure and reflects the main direction of actual damage expansion. The damage propagation path is recorded in the form of an ordered sequence of nodes, including the identifier, level, and corresponding damage occurrence time of each node on the path.

[0133] Seepage characteristics of adjacent node pairs are extracted along the damage propagation path. The damage initiation time for each node is determined, and the time difference between the damage initiation times of adjacent nodes is calculated to obtain the propagation delay. ,in Indicates damage from node propagation to nodes The time intervals are consistent with the data collection time step. Simultaneously, adjacent nodes are extracted from the edge attributes of the structural dependency graph. and Structural distance between This distance is the actual path length between the two components in physical space, measured in meters. The propagation rate is calculated based on the propagation delay and the structural distance. ,satisfy This refers to the distance the damage signal travels through the structure per unit time. When the path contains multiple pairs of adjacent nodes, the propagation rate of each segment is calculated separately, and the weighted average is taken as the overall propagation rate. The weight is the proportion of the distance of each structural segment to the total path length, in order to reflect the difference in contribution of different structural segments to the overall propagation characteristics.

[0134] Generate a health monitoring report, which includes the node sequence of the damage propagation path, the damage initiation time at each node, and the propagation rate of each path segment. and overall propagation rate This is integrated with the corrected drainage and corrected waterproofing components in the corrected state vector. The report includes the following core information: the location and time of damage initiation, a complete description of the damage propagation path, the propagation rate distribution of each structural segment, the overall health status of drainage and waterproofing reflected in the current corrected state vector, and the estimated time when the damage reaches the critical node based on the propagation rate. When the propagation rate... Exceeding the preset rate alarm threshold The report automatically triggers a high-risk warning flag, prompting relevant maintenance personnel to prioritize inspection and intervention at key nodes along the damage propagation path. The report is output in a structured data format and supports interface integration with highway operation and maintenance management platforms, enabling real-time push and historical archiving of monitoring results, providing data support for subsequent maintenance decisions and planning.

[0135] A second aspect of the present invention provides an integrated intelligent structural health monitoring system for drainage and flood control in highways, comprising:

[0136] The state feature construction unit is used to collect deformation parameters and seepage parameters of the drainage and waterproofing structure of the highway, perform spatial gradient calculation on the deformation parameters to obtain gradient features, perform time-series cumulative analysis on the seepage parameters to obtain seepage features, and fuse them with the gradient features to obtain structural state features.

[0137] The collaborative state coupling unit is used to perform graph propagation operation on the structural state features based on the pre-constructed drainage path topology relationship to obtain drainage connectivity features, perform hierarchical aggregation operation on the structural state features based on the pre-constructed waterproofing hierarchy relationship to obtain waterproofing degradation features, and couple and map the drainage connectivity features and the waterproofing degradation features to obtain a collaborative state vector.

[0138] The interaction relationship correction unit is used to calculate the interaction coefficient based on the drainage component and the waterproof component in the cooperative state vector, determine the degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity according to the interaction coefficient, and correct the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector.

[0139] The damage propagation analysis unit is used to identify associated nodes in a pre-established structural dependency graph based on the structural position corresponding to the modified state vector, perform back-tracking analysis on the seepage characteristics corresponding to the associated nodes to obtain the damage propagation path, calculate the propagation rate along the damage propagation path, and generate a health monitoring report.

[0140] A third aspect of the present invention provides an electronic device, comprising:

[0141] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0142] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0143] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for integrated intelligent structural health monitoring of drainage and flood control in highways, characterized in that, include: Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Temporal cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics. Based on the pre-constructed drainage path topology, graph propagation operation is performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operation is performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector. The interaction coefficient is calculated based on the drainage component and the waterproof component in the cooperative state vector. The degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity are determined according to the interaction coefficient. The cooperative state vector is then corrected according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector. Based on the structural position corresponding to the modified state vector, the associated nodes are identified in the pre-established structural dependency graph. The seepage characteristics corresponding to the associated nodes are back-analyzed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path and a health monitoring report is generated. Based on the structural location corresponding to the corrected state vector, associated nodes are identified in a pre-established structural dependency graph. Backtracking analysis of the seepage characteristics corresponding to the associated nodes is performed to obtain the damage propagation path. The propagation rate is calculated along the damage propagation path, and a health monitoring report is generated, including: The corrected drainage component and the corrected waterproof component are extracted from the corrected state vector and the corresponding structural positions are obtained by parsing. The structural positions are mapped to the pre-established structural dependency graph to obtain the position nodes. The structural dependency graph is topologically sorted to obtain the dependency level. Based on the dependency level, the predecessor nodes are traced upward from the position nodes. The predecessor nodes are connected to the nodes and nodes that form a strong connected component with the position nodes are identified as associated nodes. Extract the seepage features corresponding to the associated nodes, perform reverse traversal over time, and calculate the feature change gradient of adjacent time points. Identify the time when the feature change gradient exceeds a preset gradient threshold as the damage initiation time and record the corresponding node as the damage initiation node. Search for a path from the damage initiation node along the structural dependency graph to the location node and determine the damage propagation path. The seepage characteristics of adjacent nodes in the damage propagation path are extracted and the time difference between adjacent nodes at the damage initiation time is calculated to obtain the propagation delay. The structural distance between adjacent nodes is extracted from the structural dependency graph and the propagation rate is calculated based on the propagation delay. The damage propagation path and the propagation rate are combined with the corrected state vector to generate a health monitoring report.

2. The method according to claim 1, characterized in that, Deformation and seepage parameters of the drainage and waterproofing structure of the highway are collected. Spatial gradient calculation is performed on the deformation parameters to obtain gradient characteristics. Time-series cumulative analysis is performed on the seepage parameters to obtain seepage characteristics. These characteristics are then fused with the gradient features to obtain structural state characteristics, including: Deformation and seepage parameters of the highway drainage and waterproofing structure at multiple monitoring points are collected by a distributed sensing unit. A spatial grid is constructed according to the spatial location of the monitoring points for the deformation parameters, and the deformation difference between adjacent grid units is calculated to obtain a spatial gradient field. The spatial gradient field is decomposed into radial gradient components. Based on the radial gradient components, local abrupt regions are identified, and the gradient amplitude of the abrupt regions is extracted as gradient features. The seepage parameters are accumulated and summed in segments according to the time series to obtain the cumulative seepage flow rate for each time period. The rate of change of the cumulative seepage flow rate between adjacent time periods is calculated, and time periods with a rate of change exceeding a preset threshold are identified as abnormal time periods. The seepage parameters corresponding to the abnormal time periods are extracted, and frequency domain features are obtained through spectrum analysis. The seepage characteristics are then determined in combination with the cumulative seepage flow rate. Based on the local mutation region, the target monitoring point is located in the monitoring point grid. The seepage feature corresponding to the target monitoring point is extracted to obtain the target seepage feature. The correlation between the gradient feature and the target seepage feature is calculated and a coupling matrix is ​​constructed based on the correlation. The gradient feature and the seepage feature are then subjected to tensor operation through the coupling matrix to obtain the structural state feature.

3. The method according to claim 1, characterized in that, Based on the pre-constructed drainage path topology, graph propagation operations are performed on the structural state features to obtain drainage connectivity features. Based on the pre-constructed waterproofing hierarchy, hierarchical aggregation operations are performed on the structural state features to obtain waterproofing degradation features. The drainage connectivity features and the waterproofing degradation features are coupled and mapped to obtain a cooperative state vector, including: The structural state features are mapped to each node of the drainage path topology to obtain node state features. The drainage path topology is spectrally decomposed to obtain a Laplace feature vector set. The node state features are projected onto the Laplace feature vector set to obtain spectral coefficients and spectral propagation is performed to obtain propagated node features. The gradient abrupt change positions of the propagated node features are identified and the corresponding spectral coefficients are extracted to form drainage connectivity features. The structural state features are assigned to each level of the waterproofing hierarchy to obtain the hierarchical state features. A hypergraph representation corresponding to the waterproofing hierarchy is constructed, and the hierarchical state features are propagated by hyperedges to obtain hyperedge enhancement features. The hyperedge enhancement features are decomposed by tensors to obtain core tensors, and the residual between the core tensors and the hyperedge enhancement features is calculated. Based on the residual, core tensor slices are extracted to form waterproofing degradation features. Tensor contraction is performed on the drainage connectivity feature and the waterproofing degradation feature to obtain a contraction tensor. An attention distribution is calculated on the contraction tensor and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is performed on the coupled feature tensor to obtain drainage components and waterproofing components, and a cooperative state vector is obtained by solving the problem.

4. The method according to claim 3, characterized in that, The attention distribution is calculated for the contraction tensor, and a bilinear interactive mapping is performed based on the attention distribution to obtain a coupled feature tensor. Mode decomposition is then performed on the coupled feature tensor to obtain drainage and waterproofing components, and the cooperative state vector is obtained by solving for these components. The contraction tensor is sliced ​​along the drainage path dimension and the waterproofing layer dimension to obtain drainage dimension slice group and waterproofing dimension slice group respectively. Self-attention calculation is performed on the drainage dimension slice group and the drainage connectivity feature to obtain drainage self-attention feature. Self-attention calculation is performed on the waterproofing dimension slice group and the waterproofing degradation feature to obtain waterproofing self-attention feature. Cross-attention calculation is performed on the drainage self-attention feature and the waterproofing self-attention feature, and sparse attention distribution is obtained after gated linear unit transformation and normalization. The drainage connectivity features and the waterproofing degradation features are projected into a high-dimensional feature space and weighted and aggregated based on the sparse attention distribution to obtain attention enhancement features. The attention enhancement features are then propagated through a graph neural network to obtain topology enhancement features and mapped through a bilinear pooling layer to obtain a coupled feature tensor. The coupled feature tensor is decomposed by Tucker to obtain a core tensor and multiple factor matrices. Drainage-related modal slices are extracted from the core tensor and reconstructed with the corresponding factor matrices to obtain drainage components. Waterproofing-related modal slices are extracted from the core tensor and reconstructed with the corresponding factor matrices to obtain waterproofing components. The drainage components and waterproofing components are fused through residual connection and then subjected to layer normalization to obtain the cooperative state vector.

5. The method according to claim 1, characterized in that, The interaction coefficient is calculated based on the drainage and waterproofing components in the cooperative state vector. The degree of transmission of drainage failure to waterproofing performance and the degree of reverse constraint of waterproofing damage on drainage capacity are determined based on the interaction coefficient, including: The drainage component and the waterproofing component are separated from the cooperative state vector and expanded into drainage state sequence and waterproofing state sequence, respectively. The drainage state sequence and the waterproofing state sequence are time-aligned and the covariance is calculated under different lag time differences to obtain the covariance matrix. The positive conduction intensity of drainage to waterproofing and the reverse conduction intensity of waterproofing to drainage are extracted from the covariance matrix and combined to obtain the bidirectional interaction coefficient. The drainage component is mapped to the nodes of the drainage path topology and the failure node is identified to obtain the failure time. The waterproof component is mapped to each waterproof layer and the amount of degradation change after the failure time is calculated to obtain the degradation increment sequence. The degradation increment sequence is multiplied element by element based on the positive conduction intensity in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the degree of conduction influence. The failure time is obtained by identifying the damaged layer from the waterproof layer that has deteriorated beyond the preset deterioration threshold. The change in drainage capacity of the node after the failure time is calculated as a capacity decay sequence. The capacity decay sequence is multiplied element by element based on the reverse conduction strength in the bidirectional interaction coefficient and accumulated along the time dimension to obtain the reverse constraint degree.

6. The method according to claim 1, characterized in that, The modified state vector is obtained by modifying the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint, including: The conduction intensity value is extracted from the degree of conduction influence to construct a conduction intensity matrix, and the constraint intensity value is extracted from the degree of reverse constraint to construct a constraint intensity matrix. The conduction intensity matrix and the constraint intensity matrix are subjected to asymmetric coupling analysis to obtain a coupling intensity field. The coupling intensity field is decomposed to obtain a conduction vector and a constraint vector. The vector angle between the conduction vector and the constraint vector is calculated to identify the conduction-dominant synergistic region and the constraint-dominant antagonistic region. The pre-obtained connectivity failure nodes are taken as drainage failure nodes, and the pre-obtained damage levels are taken as damage waterproofing levels. In the dominant conduction cooperative region, a dominant conduction path is constructed from the drainage failure nodes to the corresponding waterproofing levels along the conduction vector, and the path attenuation coefficient is calculated. In the dominant constraint antagonistic region, a reverse blocking path is constructed from the damage waterproofing levels to the corresponding drainage nodes along the constraint vector, and the path damping coefficient is calculated. The failure state change of the drainage failure node is extracted and accumulated along the dominant conduction path according to the path attenuation coefficient to the corresponding waterproofing level to obtain the corrected waterproofing component. The damage and deterioration change of the damaged waterproofing level is extracted and accumulated along the reverse blocking path according to the path damping coefficient to the corresponding drainage node to obtain the corrected drainage component. The corrected waterproofing component and the corrected drainage component are spliced ​​together to obtain the corrected state vector.

7. An integrated intelligent structural health monitoring system for drainage and flood control on highways, used to implement the method described in any one of claims 1-6, characterized in that, include: The state feature construction unit is used to collect deformation parameters and seepage parameters of the drainage and waterproofing structure of the highway, perform spatial gradient calculation on the deformation parameters to obtain gradient features, perform time-series cumulative analysis on the seepage parameters to obtain seepage features, and fuse them with the gradient features to obtain structural state features. The collaborative state coupling unit is used to perform graph propagation operation on the structural state features based on the pre-constructed drainage path topology relationship to obtain drainage connectivity features, perform hierarchical aggregation operation on the structural state features based on the pre-constructed waterproofing hierarchy relationship to obtain waterproofing degradation features, and couple and map the drainage connectivity features and the waterproofing degradation features to obtain a collaborative state vector. The interaction relationship correction unit is used to calculate the interaction coefficient based on the drainage component and the waterproof component in the cooperative state vector, determine the degree of transmission influence of drainage failure on waterproof performance and the degree of reverse constraint of waterproof damage on drainage capacity according to the interaction coefficient, and correct the cooperative state vector according to the degree of transmission influence and the degree of reverse constraint to obtain the corrected state vector. The damage propagation analysis unit is used to identify associated nodes in a pre-established structural dependency graph based on the structural position corresponding to the modified state vector, perform back-tracking analysis on the seepage characteristics corresponding to the associated nodes to obtain the damage propagation path, calculate the propagation rate along the damage propagation path, and generate a health monitoring report.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images