Station structure health monitoring and maintenance method and system based on digital twinning

Abstract

The application discloses a station structure health monitoring and maintenance method and system based on digital twinning, which comprises the following steps: firstly, a three-dimensional geometric model containing a station main structure and surrounding soil is constructed, a virtual-actual data mapping channel of a physical sensor network and a virtual model node is built, and an initial damage field is given to obtain a digital twinning reference model; secondly, structure deformation and pore water pressure monitoring data flow are accessed, soil permeability coefficients and boundary water heads are dynamically inverted and updated through a set Kalman filtering algorithm, and a parameter calibration model synchronized with a physical station in real time is obtained; thirdly, multi-physical field coupling response calculation displacement and anti-floating safety factors are solved, and when indexes exceed the standard, dominant risk factors and spatial positions are identified relying on partial derivative matrices; finally, a virtual pre-play maintenance scheme is made for the risk factors, a safety recovery efficiency and a maintenance cost are weighted and optimized, and on-site execution instructions are generated, thereby providing reliable technical support for station safe operation.

Description

Technical Field

[0001] This invention belongs to the field of station structure monitoring technology, and more specifically, relates to a method and system for station structure health monitoring and maintenance based on digital twins. Background Technology

[0002] Urban rail transit has become the core carrier of modern urban public transportation. Underground stations, as key nodes in the rail transit network, are constantly subjected to multiple factors, including soil pressure, groundwater seepage, surface dynamic loads, and disturbances from surrounding construction, all within a complex rock and soil environment. As the years of operation increase, station structures are prone to uneven deformation, wall cracking, water leakage, and reduced buoyancy resistance, directly threatening structural safety and stable line operation. This places stringent requirements on structural health monitoring and meticulous maintenance.

[0003] Current underground railway station structural health monitoring and operation and maintenance management still have many shortcomings. Traditional monitoring systems mainly rely on single-point sensor data collection, resulting in fragmented data and a lack of spatial correlation mapping between the structure and the surrounding soil, making it difficult to achieve global real-time perception of the structural status. Manual on-site inspections depend on manual reconnaissance and experience-based judgment, leading to low work efficiency, long inspection cycles, and an inability to capture the dynamic evolution of structural defects in a timely manner, resulting in significant lags in damage identification and safety early warning.

[0004] Existing structural analysis models mostly employ static modeling methods, relying on pre-survey data for core parameters. This lack of dynamic correction based on on-site measured data results in significant deviations from actual working conditions for key parameters such as soil permeability coefficient and boundary water head. Furthermore, the accuracy of multi-physics coupled response calculations is insufficient, making it difficult to accurately represent the real-time stress and deformation state of the structure. Structural safety management often depends on single indicator thresholds for early warning, failing to pinpoint the source of risk and dominant influencing factors, resulting in weak risk tracing and targeted control capabilities.

[0005] Maintenance plans often rely on subjective decisions based on engineering experience, lacking digital virtual simulation tools. The combination and optimization of intervention measures such as emergency precipitation, weighting, and grouting are lacking, which can easily lead to waste of maintenance resources or inadequate handling. There is also a lack of a scientific decision-making mechanism that balances structural safety restoration effectiveness with maintenance costs.

[0006] Digital twin technology offers a new approach to the integrated virtual and physical operation and maintenance of engineering structures. However, its application in underground railway stations has not yet formed a complete system, failing to cover the entire process of integrated modeling, dynamic parameter calibration, accurate risk identification, and intelligent solution optimization. To address the pain points of traditional operation and maintenance models and ensure the long-term safe and stable operation of underground railway stations, developing structural health monitoring and maintenance technologies adapted to the actual engineering situation has significant engineering application value and practical significance. Summary of the Invention

[0007] This invention aims to solve the problems of lagging health monitoring of underground station structures, distorted modeling parameters, difficulty in locating risk sources, and reliance on experience for maintenance decisions. By using digital twins to achieve the fusion and dynamic synchronization of virtual and real data, it can complete structural status assessment, rapid identification of dominant risks, and intelligent optimization of maintenance solutions, thereby improving the efficiency and safety of station operation and maintenance and ensuring the long-term stable operation of underground rail transit structures.

[0008] To address the aforementioned deficiencies or improvement needs of existing technologies, as a first aspect of this invention, the present invention provides a method for monitoring and maintaining the structural health of a railway station based on digital twins, comprising: S1. Construct a three-dimensional geometric model including the main structure and the surrounding soil, and at the same time establish a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assign the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain a digital twin benchmark model. S2. The time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array is accessed in real time through the virtual and real data mapping channel. The ensemble Kalman filter algorithm is used to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. S3. In the real-time synchronous evolution digital twin model, the multi-physics field coupled response is solved synchronously to calculate the structural displacement and the anti-buoyancy safety factor; when the structural displacement or the anti-buoyancy safety factor exceeds the preset control standard, the partial derivative matrix of each inversion update parameter with respect to the structural displacement and the anti-buoyancy safety factor is calculated, and the dominant risk factor and its spatial location are identified based on the element with the largest absolute value in the partial derivative matrix. S4. For the dominant risk factors, conduct parallel virtual simulations in the digital twin model of maintenance intervention combination schemes including emergency precipitation, ballast configuration and grouting for leak sealing; use the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function to solve the objective function value of each maintenance intervention combination scheme, and select the scheme with the optimal objective function value to generate on-site execution instructions.

[0009] Furthermore, in S1, a virtual-real data mapping channel is established between the physical station sensor network and the virtual model nodes, and the initial damage field identified by the current status detection is assigned to the three-dimensional geometric model, specifically including: The three-dimensional spatial coordinates of the physical station sensors and the grid coordinates of the virtual model nodes are obtained. A spatial mapping matrix is ​​constructed using the radial basis function interpolation method, and the sensor measured data are distributed to the corresponding virtual model nodes according to the distance weight. The width of structural cracks and the location of water leakage identified by the current status detection are converted into equivalent stiffness reduction coefficients and local permeability amplification coefficients. By modifying the elastic matrix and permeability tensor of the corresponding mesh elements in the three-dimensional geometric model, the spatial discretization of the initial damage field is completed.

[0010] Furthermore, in S2, an ensemble Kalman filter algorithm is used to drive the dynamic joint inversion and state update of the soil permeability coefficient and boundary head conditions in the benchmark model, specifically including: Based on the prior mean and prior covariance matrix of soil permeability coefficient and boundary head conditions, Latin hypercube sampling is used to generate an initial parameter set containing multiple independent samples, and the initial parameter set is concatenated with the model state variables to form an augmented state vector to obtain the initial state set. In each assimilation cycle, the innovation vector between the measured time series data stream and the model predicted observation is calculated. The Mahalanobis distance of the innovation vector is calculated using the preset observation error covariance matrix. When the Mahalanobis distance is greater than the confidence threshold set based on the chi-square distribution, the corresponding sensor data is determined to be an outlier and is removed. The cross-covariance matrix of state and observation and the Kalman gain matrix are calculated using the effective observation data after outlier removal. The ratio of the actual sample variance to the theoretical variance of the innovation vector is also calculated and used as the adaptive covariance inflation factor. The effective observation information is weighted and superimposed onto the prior state set using the Kalman gain matrix to obtain the posterior state set. The dispersion of the parameter samples in the posterior state set is amplified using the adaptive covariance expansion factor to complete the joint update of the soil permeability coefficient and boundary head conditions.

[0011] Furthermore, the construction process of the adaptive covariance inflation factor specifically includes: The difference between the effective observation data after removing outliers within the current assimilation period and the mean of the model-predicted observation set is calculated to obtain the current innovation vector; Based on a sliding time window of a preset time length, the mean of the outer product of all historical innovation vectors and the current innovation vector within the sliding time window is calculated to obtain the actual covariance matrix of the innovation, and the trace value of the actual covariance matrix of the innovation is extracted as the actual sample variance. The observation error covariance matrix of the current assimilation period is added to the prior set covariance matrix of the model predicted observations to obtain the innovation theory covariance matrix, and the trace value of the innovation theory covariance matrix is ​​extracted as the theoretical variance. Calculate the ratio of the actual sample variance to the theoretical variance, and then sum the ratio with the historical adaptive covariance inflation factor of the previous assimilation period to obtain the smooth adaptive covariance inflation factor of the current assimilation period. The smoothing adaptive covariance inflation factor is multiplied by the deviation vector between each sample in the prior state set and the set mean to amplify the dispersion of the prior state set.

[0012] Furthermore, the calculation of the partial derivative matrix of each inversion update parameter with respect to structural displacement and anti-buoyancy safety factor in S3 specifically includes: The sum of squared errors between the calculated and measured values ​​of structural displacement and the preset evaluation function of the anti-buoyancy safety factor are constructed as the objective functional. The accompanying water pressure variable corresponding to the soil seepage control equation and the accompanying displacement variable corresponding to the structural mechanical equilibrium control equation are introduced as Lagrange multipliers. The objective functional is linearly combined with the weak form of the soil seepage control equation and the structural mechanical equilibrium control equation to construct an augmented functional. The first variational function of the augmented generalized function with respect to the forward water pressure variable and the forward displacement variable is obtained and set to zero. The accompanying water pressure control equation and the accompanying displacement control equation are derived. The accompanying water pressure control equation and the accompanying displacement control equation are solved simultaneously with the accompanying boundary conditions to obtain the accompanying water pressure field and the accompanying displacement field. The partial derivatives of the augmented functional with respect to the soil permeability coefficient and the boundary head condition are calculated to obtain the gradient analytical expression containing the inner product of the spatial gradients of the forward-modeled state variables and the associated variables. The associated hydraulic pressure field, the associated displacement field, and the forward-modeled hydraulic pressure field and displacement field obtained by forward modeling are substituted into the gradient analytical expression, and spatial integration is performed in the computational domain of the entire station structure and the surrounding soil to analytically obtain the gradient vectors of the soil permeability coefficient and the boundary head condition with respect to the target functional. The gradient vectors corresponding to all inversion update parameters are assembled according to the parameter dimensions to generate the partial derivative matrix.

[0013] Furthermore, in step S3, identifying the dominant risk factor and its spatial location based on the element with the largest absolute value in the partial derivative matrix specifically includes: The gradient vector corresponding to the structural displacement target functional in the partial derivative matrix is ​​extracted as the first partial derivative row vector, and the gradient vector corresponding to the anti-buoyancy safety factor target functional is extracted as the second partial derivative row vector. The elements with the largest absolute values ​​are searched by traversing the first and second partial derivative row vectors respectively, and the first and second largest absolute value elements are extracted. Obtain the column indices of the first and second maximum absolute value elements in the partial derivative matrix, match the corresponding parameter types in the preset inversion update parameter sequence according to the column indices, and identify the matched parameter types as the dominant risk factors affecting the corresponding objective functional. The parameter types include soil permeability coefficient or boundary head conditions. The finite element mesh mapping table of the station structure and surrounding soil is queried according to the column index to obtain the grid cell number or node number where the dominant risk factor is located, and the grid cell number or node number is mapped to the digital twin three-dimensional geometric model to determine the three-dimensional spatial location of the dominant risk factor in the physical station.

[0014] Furthermore, S4 uses a weighted sum of structural safety restoration effectiveness and maintenance costs as the objective function, specifically including: The structural safety recovery efficiency is defined as the weighted sum of the maximum displacement reduction rate and the anti-buoyancy safety factor improvement rate of the structure after virtual pre-simulation. The maintenance cost is defined as the sum of the products of the pumping volume, the volume of counterweight material, and the grouting volume involved in each maintenance intervention combination scheme and the corresponding unit price; The entropy weight method is used to calculate the weights of displacement reduction rate and anti-buoyancy improvement rate in the structural safety restoration effectiveness, as well as the weights of safety effectiveness and maintenance cost in the overall objective function, based on the data dispersion of various indicators in the historical maintenance database.

[0015] Furthermore, the parallel virtual rehearsal in the digital twin model in S4 includes a maintenance intervention combination scheme of emergency precipitation, weight configuration, and grouting for leak sealing, specifically including: For emergency precipitation, the boundary conditions of the precipitation wells are activated in the model and the precipitation depth time curve is set; for the counterweight configuration, the equivalent surface load is applied to the corresponding grid nodes of the station floor and roof; for grouting and plugging, the soil permeability coefficient within the preset range around the leakage point is reduced to the design permeability of the grouting body, and the grouting expansion volume strain is applied. During the rehearsal, an explicit-implicit alternating time integration strategy was adopted, using explicit solutions for the transient fluid impact in the early stage of intervention and implicit solutions for the consolidation settlement in the later stage of intervention.

[0016] As a second aspect of the present invention, a station structure health monitoring and maintenance system based on digital twins is also provided, comprising: The twin benchmark model construction unit is used to construct a three-dimensional geometric model including the main structure and the surrounding soil, while establishing a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assigning the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain the digital twin benchmark model. The model parameter calibration unit is used to access the time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array in real time through the virtual and real data mapping channel. It uses an ensemble Kalman filter algorithm to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. The dominant risk factor identification unit is used to simultaneously solve the multi-physics coupling response in a real-time synchronously evolving digital twin model to calculate the structural displacement and anti-buoyancy safety factor. When the structural displacement or anti-buoyancy safety factor exceeds the preset control standard, the unit calculates the partial derivative matrix of each inversion update parameter with respect to the structural displacement and anti-buoyancy safety factor, and identifies the dominant risk factor and its spatial location based on the element with the largest absolute value in the partial derivative matrix. The intervention scheme optimization unit is used to conduct parallel virtual simulations of maintenance intervention combinations, including emergency precipitation, ballast configuration, and grouting and leak sealing, in a digital twin model for the dominant risk factors. Using the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function, it solves for the objective function value of each maintenance intervention combination scheme, selects the scheme with the optimal objective function value, and generates on-site execution instructions.

[0017] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor as described in any one of the claims: a method for monitoring and maintaining the health of a station structure based on a digital twin.

[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The present invention provides a method for monitoring and maintaining the health of railway station structures based on digital twins. This method constructs a three-dimensional geometric model containing the main structure and surrounding soil, establishes a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, assigns the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain a digital twin baseline model, and then accesses the time-series data stream of structural deformation and pore water pressure in real time through the virtual-real data mapping channel. An ensemble Kalman filter algorithm is used to drive the dynamic joint inversion and state update of soil permeability coefficient and boundary water head conditions, and anomaly data removal and adaptive covariance expansion are completed simultaneously. This allows the digital twin model to evolve synchronously with the physical station in real time, corrects the deviation of key model parameters, improves the fit between the model and actual working conditions, and provides a digital model foundation for structural safety analysis.

[0019] 2. The station structure health monitoring and maintenance method based on digital twin of the present invention solves the multi-physics field coupling response synchronously in a real-time synchronously evolving digital twin model to calculate the structural displacement and anti-buoyancy safety factor. When the structural displacement or anti-buoyancy safety factor exceeds the preset control standard, the partial derivative matrix of each inversion update parameter with respect to the structural displacement and anti-buoyancy safety factor is calculated. Based on the element with the largest absolute value in the partial derivative matrix, the dominant risk factor and its spatial location are identified. This method can quantify the structural safety status, quickly lock the core factors affecting structural safety, clarify the three-dimensional spatial location of the risk factor, realize the source and location of structural risks, and provide a basis for subsequent targeted maintenance.

[0020] 3. The digital twin-based station structure health monitoring and maintenance method of the present invention performs parallel virtual simulations of maintenance intervention combinations for emergency precipitation, ballast configuration, and grouting and leak sealing in a digital twin model targeting the dominant risk factors. It employs an explicit-implicit alternating time integration strategy to complete the simulation calculations, using the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function to solve for the optimal solution and generate on-site execution instructions. This allows for the simulation and verification of multiple maintenance schemes in a virtual environment. By relying on scientific weight allocation to balance safety restoration effects and maintenance costs, the optimal treatment scheme is selected, avoiding the drawbacks of experience-based decision-making and achieving intelligent optimization of maintenance schemes and on-site execution guidance. Attached Figure Description

[0021] Figure 1 This is a flowchart of a station structure health monitoring and maintenance method based on digital twin according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system units in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Please refer to Figure 1 This embodiment 1 provides a method for monitoring and maintaining the structural health of a station based on digital twins, including: S1. Construct a three-dimensional geometric model including the main structure and the surrounding soil, and at the same time establish a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assign the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain a digital twin benchmark model. S2. The time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array is accessed in real time through the virtual and real data mapping channel. The ensemble Kalman filter algorithm is used to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. S3. In the real-time synchronous evolution digital twin model, the multi-physics field coupled response is solved synchronously to calculate the structural displacement and the anti-buoyancy safety factor; when the structural displacement or the anti-buoyancy safety factor exceeds the preset control standard, the partial derivative matrix of each inversion update parameter with respect to the structural displacement and the anti-buoyancy safety factor is calculated, and the dominant risk factor and its spatial location are identified based on the element with the largest absolute value in the partial derivative matrix. S4. For the dominant risk factors, conduct parallel virtual simulations in the digital twin model of maintenance intervention combination schemes including emergency precipitation, ballast configuration and grouting for leak sealing; use the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function to solve the objective function value of each maintenance intervention combination scheme, and select the scheme with the optimal objective function value to generate on-site execution instructions.

[0024] This embodiment 1 further elaborates on the above steps.

[0025] (1) Construction of twin baseline model Underground stations operate in complex engineering environments where soil and groundwater interact. Digital reconstruction of the initial structural state is a prerequisite for health monitoring. Traditional modeling can only present the structural geometry and cannot achieve collaborative association between the physical entity and the virtual model. To address this issue, a three-dimensional geometric model covering the station's main structure and surrounding soil was first constructed.

[0026] In other preferred embodiments, the specific process of constructing a three-dimensional geometric model covering the main structure of the station and the surrounding soil is as follows: First, based on the station's as-built drawings, geological survey report, and laser point cloud scanning data, a geometric model of the station's main structure was established using BIM modeling software, including components such as the base slab, side walls, middle slab, roof slab, and columns. Simultaneously, based on the soil layer distribution information in the geological survey report, a three-dimensional geological model of the surrounding soil was established, extending outwards from the structure. The boundary range of the soil model was typically three to five times the maximum span of the station structure to eliminate boundary effects. Subsequently, the geometric model was imported into a finite element analysis platform for mesh generation. For the station's main structure, eight-node hexahedral reduced integral solid elements or shell elements were used for discretization; for the surrounding soil, ten-node quadratic tetrahedral elements or eight-node hexahedral solid elements were used. Regarding material and constitutive settings, the station's concrete structure adopted a concrete damage-plastic model considering plastic damage, while the soil adopted a modified Cambridge model or a Mohr-Coulomb elastoplastic constitutive model. Between the outer surface of the structure and the contact surface of the soil, a face-to-face contact pair based on the penalty function is set to simulate the friction and slip behavior between the soil and the structure, and finally a finite element reference mesh model containing node coordinates, element topology relationships and initial material properties is generated.

[0027] Simultaneously, a virtual-real data mapping channel is established between the physical station sensor network and the virtual model nodes, and the initial damage field identified by the current status detection is assigned to the three-dimensional geometric model to obtain a digital twin baseline model. The specific process of assigning the initial damage field identified by the current status detection to the three-dimensional geometric model is as follows: First, the measured 3D spatial coordinates of sensors such as strain gauges, earth pressure cells, water level gauges, and inclinometers installed in the physical station are extracted, along with the 3D coordinates of the corresponding grid nodes in the virtual model. Next, a Gaussian radial basis function (RBF) is used as the kernel function to construct a spatial mapping matrix. For any virtual model node, the Euclidean distance from it to all physical sensors within a defined radius of influence is calculated, and these distances are substituted into the Gaussian RBF to calculate the shape parameters. By solving a system of linear equations containing the coordinates of all sensors, the weighting coefficients of each sensor's contribution to the virtual node's data are obtained, thus constructing a sparse spatial mapping matrix with dimensions equal to the number of nodes multiplied by the number of sensors. Subsequently, the measured physical quantity vectors collected by the sensors are multiplied left by this spatial mapping matrix, allowing for a smooth and continuous distribution of discrete measured data to the corresponding virtual model nodes according to distance weights.

[0028] Then, the width of structural cracks and the location of water leakage identified by the current status detection are converted into equivalent stiffness reduction factor and local permeability amplification factor.

[0029] Specifically, for crack damage in the station structure, a stiffness reduction method based on fracture mechanics and the equivalent strain principle is adopted. First, a crack distribution map of the structural surface is obtained using a 3D laser scanner or machine vision equipment, extracting the spatial orientation, length, and average width of the cracks. For concrete mesh elements containing cracks, they are treated as orthotropically damaged materials. Second, based on the relationship between crack width and concrete fracture energy, the equivalent tensile damage factor of the element in the direction perpendicular to the crack is calculated. Subsequently, this damage factor is used to correct the original isotropic elastic stiffness matrix of the element. Specifically, the normal stiffness component and related shear stiffness component perpendicular to the crack direction in the elastic matrix are multiplied by a reduction factor, while the stiffness parallel to the crack direction remains unchanged. This allows the finite element model to accurately reflect the local stiffness degradation and stress redistribution caused by cracks.

[0030] To address water leakage points in the station structure, an equivalent simulation was conducted using the permeability tensor amplification method based on porous media seepage theory. First, the three-dimensional coordinates and severity levels of leakage points were determined using infrared thermal imaging or manual inspection. The soil or structural mesh elements containing these leakage points were then located in the finite element model. Second, for leaking elements, internal microcracks or pore channels lead to a sharp increase in local permeability. A mapping relationship between leakage level and equivalent porosity increment was established through field pumping tests or empirical formulas. Then, using the microscopic derivation of Darcy's law, the local permeability amplification factor of the element in the main seepage direction was calculated. Finally, in the finite element solver, a user-defined material subroutine was directly invoked to read the numbers of these specific elements and replace their original isotropic permeability coefficient matrices with an anisotropic permeability tensor matrix containing the amplification factor.

[0031] Finally, by modifying the underlying data of the elastic matrix and permeation tensor of the corresponding mesh elements in the three-dimensional geometric model as described above, the spatial discretization of the initial damage field in the three-dimensional geometric model was completed, and a digital twin benchmark model was finally formed.

[0032] (2) Model parameter calibration After completing the construction of the twin baseline model, relying on the established virtual-real data mapping channel, and through real-time access to the time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array, the ensemble Kalman filter algorithm is used to drive the dynamic joint inversion and state update of the soil permeability coefficient and boundary water head conditions in the baseline model, resulting in a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. The specific process of dynamic joint inversion and state update is as follows: First, construct the initial state set. Obtain the prior mean vector of soil permeability coefficient and boundary head conditions. and the prior covariance matrix Based on the aforementioned prior mean vector and the prior covariance matrix The Latin hypercube sampling method is used to generate a sample containing... Initial parameter set for each independent sample The initial parameter set The set of model state variables of the digital twin benchmark model By concatenating the vectors along their dimensions, an augmented state vector is constructed, thus obtaining the initial state set. The initial state set It covers the uncertainty distribution of all parameters and state variables to be inverted at the initial moment of the model.

[0033] Secondly, in each assimilation cycle Perform anomaly data identification and removal. Obtain the current assimilation cycle. Measured time series data stream and the set of observations predicted by the model ,in For the set of prior states, This is the observation operator matrix, used to map state variables to the observation space. The measured time-series data stream is then calculated. With the model predicting the set of observations The difference between corresponding samples is used to obtain the set of innovation vectors. For each innovation vector in the set of innovation vectors... Using a pre-set observation error covariance matrix Calculate its Mahalanobis distance The Mahalanobis distance With confidence thresholds set based on chi-square distribution Comparison, when the Mahalanobis distance Greater than the confidence threshold When an outlier occurs, the corresponding sensor measurement data is identified as an outlier and removed, leaving only the valid observation data set after outlier removal. and its corresponding effective information vector.

[0034] Next, an adaptive covariance inflation factor is constructed to suppress filter divergence. The current assimilation period is calculated. Valid observation data after outlier removal Mean of the set of observations predicted by the model The difference between them yields the current innovation vector. Set a sliding time window with a preset duration. In the sliding time window Within this context, calculate the sum of all historical information vectors and the current information vector. The mean of the outer product is used to obtain the actual covariance matrix of the new information. Extract the actual covariance matrix of the new information. The trace value is used as the actual sample variance. The current assimilation period observation error covariance matrix Covariance matrix of the prior set of observations predicted by the model Adding them together yields the covariance matrix of the innovation theory. And extract the covariance matrix of the novel theory. The trace value is used as the theoretical variance Calculate the variance of the actual sample. With respect to the theoretical variance The ratio, and then the ratio is compared with the previous assimilation period. Historical adaptive covariance inflation factor We perform a weighted summation to obtain the current assimilation period. Smoothing adaptive covariance inflation factor .

[0035] Subsequently, the smoothing adaptive covariance inflation factor was used. The prior state set is subjected to inflation correction. The smooth adaptive covariance inflation factor is then adjusted. Multiply by the set of prior states The deviation vector of each sample from the set mean, used to represent the prior state set. The dispersion is amplified to obtain the expanded prior state set. Based on the expanded prior state set, the cross-covariance matrix between the state and the observation is recalculated. and the prior set covariance matrix of the model's predicted observations Using the aforementioned cross-covariance matrix The prior set covariance matrix and the observation error covariance matrix Calculate the Kalman gain matrix for the current assimilation period. .

[0036] Finally, the execution state and parameters are jointly updated. This is done using the Kalman gain matrix. The effective observations are weighted and superimposed onto the prior state set. , thus obtaining the posterior state set The set of posterior states This is the current assimilation period. The updated optimal estimates of soil permeability coefficient, boundary head conditions, and model state variables are obtained. Through iterative processes described above, the soil permeability coefficient and boundary head conditions in the digital twin baseline model evolve synchronously with the measured data from the physical station in real time, completing dynamic joint inversion and state updates.

[0037] (3) Identification of dominant risk factors After completing the dynamic joint inversion and state update of the digital twin model, the multiphysics coupled response is solved synchronously in the real-time synchronously evolving digital twin model to calculate the structural displacement and buoyancy safety factor. First, the dynamically calibrated soil permeability coefficient, boundary head conditions, and the initially assigned equivalent stiffness reduction coefficient and local permeability amplification coefficient are used as input parameters. Second, the coupled solution of the seepage field and stress field is carried out: in the seepage field, the pore water pressure distribution in the surrounding soil and inside the structure is calculated by combining real-time pore water pressure time series data; in the stress field, the pore water pressure is converted into effective stress and buoyancy load, coupled to the structure-soil interaction model including the bottom plate, side walls, middle plate, top plate and columns, to solve the overall stress and strain state.

[0038] Subsequently, based on the solution results, the maximum deformation of key structural nodes is extracted as the structural displacement. Simultaneously, the total upward water pressure (buoyancy) on the bottom surface of the base plate is calculated by integration, and its ratio is calculated with the station structure's self-weight, the weight of the overburden, and the sidewall friction resistance to obtain the buoyancy safety factor. Finally, the calculated structural displacement and buoyancy safety factor are compared with preset control standards. When both indicators approach or exceed the standard limits, the partial derivative matrices of each inversion update parameter with respect to the structural displacement and the buoyancy safety factor are calculated. Based on the element with the largest absolute value in the partial derivative matrix, the dominant risk factor and its spatial location are identified. The specific process for identifying the dominant risk factor and its spatial location is as follows: First, construct the objective functional. The mathematical expression for . Define the vector of calculated structural displacements as . The measured vector of structural displacement is The displacement weight matrix is The calculated value of the anti-buoyancy safety factor is defined as follows: The target threshold for the anti-buoyancy safety factor is The anti-buoyancy weight coefficient is The objective functional is constructed by linearly combining the weighted sum of squares of structural displacement errors with the sum of squares of deviations from the anti-buoyancy safety factor. The formula is: Among them, superscript This represents the matrix transpose operation.

[0039] Secondly, construct augmented functionals The mathematical expression for this is introduced. The accompanying water pressure variable corresponding to the soil seepage control equation is also introduced. and the accompanying displacement variable vector corresponding to the structural mechanics equilibrium governing equations As a Lagrange multiplier, the weak form of the soil seepage governing equation is defined as follows: The weak form of the structural mechanics equilibrium governing equations is: The target functional With the weak form and Perform linear combinations to construct an augmented generalized function. The formula is: .in, This represents the entire computational domain of the station structure and surrounding soil. Includes forward water pressure variables and soil permeability coefficient Spatial gradient term, Includes forward displacement variable vector The strain term and the forward water pressure variable The resulting pore water pressure coupled body force term.

[0040] Next, the adjoint governing equations are derived. For the augmented generalized functional... Regarding the forward water pressure variable respectively and forward displacement variable vector Find the first-order variation. Let the first-order variation... Using integration by parts to process The spatial gradient term in the equation is obtained by integrating the boundary integral with the pre-defined accompanying hydraulic pressure boundary condition. By combining these methods, the accompanying water pressure control equations are derived. The source terms of this equation are derived from the objective functional. Zhongzheng water pressure variable The implicit partial derivatives with respect to structural displacement and the safety factor against buoyancy are formed. Similarly, let the first variational... Using the principle of virtual work to process The strain term in the equation combines the boundary integral with the preset accompanying displacement boundary conditions. By combining these methods, the accompanying displacement control equations are derived. The source terms of this equation are directly derived from the objective functional. Forward displacement variable vector The explicit partial derivatives constitute the equation. Solve the accompanying hydraulic pressure control equations simultaneously. and the accompanying displacement control equation The accompanying water pressure field can then be obtained. and accompanying displacement field .

[0041] Subsequently, the gradient analytical expression is performed. The derivation. Due to the augmented generalized functional When satisfying the forward control equations and the adjoint control equations, its total derivative with respect to the inversion parameters is equal to its partial derivatives. Therefore, for the augmented generalized functional... Regarding soil permeability coefficient and boundary head conditions Directly calculate the partial derivative. During the differentiation process, consider the soil permeability coefficient. Weak seepage form Differentiating, we obtain the forward hydraulic pressure gradient. With accompanying water pressure gradient The dot product inner product term; for terms including boundary head conditions. Differentiating the boundary integral term yields the boundary inner product term of the boundary head variable and the associated variable. Combining these inner product terms, we obtain the analytical expression for the gradient, which includes the inner product of the spatial gradients of the forward state variable and the associated variable. .

[0042] Finally, the gradient vector is calculated and the partial derivative matrix is ​​assembled. The forward-modeled hydraulic pressure field is then used. Forward displacement field and the associated water pressure field obtained by the solution Accompanying displacement field Substitute into the gradient analytical expression In the calculation domain of the entire station structure and surrounding soil. and its boundaries The soil permeability coefficient can be directly obtained analytically by performing spatial integration within the soil. and boundary head conditions For the target functional gradient vector Extracting the gradient vector The part corresponding to the structural displacement objective functional is used as the first partial derivative row vector. Extract the portion of the objective functional corresponding to the anti-buoyancy safety factor as the second partial derivative row vector. The first partial derivative row vector and the second partial derivative row vector The partial derivative matrix is ​​assembled according to the parameter dimensions to generate a complete matrix. .

[0043] Subsequently, the inversion update parameter sequence is constructed and the partial derivative matrix is ​​determined. The topological structure. The computational domain of the station structure and surrounding soil. Discretize the soil into multiple finite element meshes and extract the soil permeability coefficient at the center of each mesh element. and the boundary of the computational domain Boundary head conditions for each boundary node All soil permeability coefficients to be inverted and boundary head conditions Arrange the parameters in a preset numbering order to construct a one-dimensional inversion update parameter sequence vector. Define the inversion update parameter sequence vector. The total dimension is That is, including The parameters to be inverted. Based on the previous gradient calculation process, the partial derivative matrix generated. The dimension is The first row corresponds to the gradient of the structural displacement objective functional with respect to each parameter, the second row corresponds to the gradient of the anti-buoyancy safety factor objective functional with respect to each parameter, and the third row corresponds to the gradient of the anti-buoyancy safety factor objective functional with respect to each parameter. The column corresponds to the inversion update parameter sequence vector. The Middle The gradient values ​​of the parameters, where From arrive Positive integers.

[0044] Next, extract the partial derivative row vectors and search for the element with the largest absolute value and its column index. Extract the partial derivative matrix. The first row of data is defined as the row vector of the first partial derivative. Extract the partial derivative matrix The second row of data is defined as the row vector of the second partial derivative. The row vector with respect to the first partial derivative In Calculate the absolute value of each element one by one, compare the magnitudes of all absolute values, and extract the element with the largest absolute value as the first element with the largest absolute value. and record In the first partial derivative row vector The column index in the table is defined as the first column index. Similarly, with respect to the second partial derivative row vector In Calculate the absolute value of each element one by one, and extract the element with the largest absolute value as the second largest absolute value element. and record In the second partial derivative row vector The column index in the code is defined as the second column index. .

[0045] Next, the parameter types of the dominant risk factor are identified based on the column index. This is based on the first column index. and the index of the second column In the inversion update parameter sequence vector The middle is located to the first The parameter and the first There are several parameters. A preset parameter attribute mapping table is queried, which records the inversion update parameter sequence vector. The physical type of the parameter corresponding to each index position. If the... If a parameter is marked as a permeability attribute in the parameter attribute mapping table, then the dominant risk factor affecting the objective functional of structural displacement is identified as the soil permeability coefficient. If it is marked as a head attribute, it is identified as a boundary head condition. Similarly, according to the first... The labels of the parameters in the parameter attribute mapping table identify the dominant risk factor affecting the objective functional of the anti-buoyancy safety factor as the soil permeability coefficient. or boundary head conditions .

[0046] Then, the mesh number is obtained by querying the finite element mesh mapping table based on the column index. During the finite element preprocessing stage, a finite element mesh mapping table is pre-constructed, which records the inversion update parameter sequence vector. The index position of each parameter in the table corresponds one-to-one with the finite element mesh element number or boundary node number to which it belongs. This is based on the first column of indexes. Query the finite element mesh mapping table to obtain the first mesh element number or first boundary node number corresponding to the dominant risk factor affecting structural displacement; according to the second column index... Query the finite element mesh mapping table to obtain the second mesh element number or second boundary node number corresponding to the dominant risk factor affecting the anti-buoyancy safety factor.

[0047] Finally, the grid numbers are mapped to the digital twin 3D geometric model to determine their 3D spatial locations. A spatial coordinate transformation matrix and a primitive mapping dictionary are obtained between the digital twin 3D geometric model and the finite element calculation mesh. The first grid cell number or first boundary node number, and the second grid cell number or second boundary node number are input into the primitive mapping dictionary to match the corresponding first and second 3D visualization primitives in the digital twin 3D geometric model. Using the spatial coordinate transformation matrix, the 3D spatial coordinate set of the first and second 3D visualization primitives in the global coordinate system of the physical station is calculated. In the rendering engine of the digital twin system, the first and second 3D visualization primitives are highlighted or have spatial markers added based on the 3D spatial coordinate set, thereby intuitively displaying the 3D spatial location of the dominant risk factor in the 3D visualization scene of the physical station.

[0048] (4) Optimization of intervention program After determining the dominant risk factors of the station structure and their three-dimensional spatial locations, a set of maintenance intervention schemes, including emergency precipitation, ballast configuration, and grouting for leak sealing, was first constructed in the digital twin model. ,in Indicates the first A combination of maintenance interventions, This represents the total number of scenarios. For emergency precipitation measures, the boundary conditions of the precipitation wells are activated in the digital twin model, and a time-series curve of the drawdown depth is set. ,in To simulate time variables, equivalent surface loads are applied to the grid nodes corresponding to the station's floor and roof slabs for counterweight configuration measures. For grouting and sealing measures, the permeability coefficient of the soil within a predetermined range around the leakage point is reduced to the design permeability of the grout. And apply grouting expansion volumetric strain in this area. .

[0049] In the numerical solution process of the virtual simulation, an explicit-implicit alternating time integration strategy is adopted to ensure computational accuracy and convergence. The total simulation time is set to... The critical time for alternation is In the initial transient fluid shock phase of the intervention, i.e., the time interval... Within this phase, an explicit solution algorithm is employed to process the fluid dynamics equations in order to capture the transient high-frequency response of pore water pressure; during the consolidation and settling stage in the later stages of intervention, i.e., the time interval... Internally, the implicit solution algorithm is switched to handle the soil consolidation equation, thereby improving solution efficiency while ensuring computational stability, and finally outputting various solutions. The structural response state at the end of the rehearsal.

[0050] Based on the structural response state output from the pre-simulation, the structural safety restoration effectiveness of each scheme is quantitatively calculated. First, calculate the maximum displacement reduction rate of the structure. ,in To determine the maximum structural displacement in the pre-intervention digital twin model, For the plan The maximum structural displacement after the simulation; then, the improvement rate of the anti-buoyancy safety factor. ,in The anti-buoyancy safety factor before intervention. For the plan The anti-buoyancy safety factor after the pre-test. Subsequently, the formula for structural safety recovery efficiency is constructed. ,in As the displacement reduction rate weight, Weighting for anti-buoyancy lift rate.

[0051] Simultaneously, the system quantitatively calculates the maintenance costs of each solution. Constructing a maintenance cost formula ,in , , These are the unit prices for pumping, counterweight materials, and grouting, respectively. , , The respective schemes The pumping volume, ballast material volume, and grouting volume involved in the calculation are all automatically extracted through boundary integrals and volume integrals of the digital twin model during the pre-simulation process, ensuring the objectivity of the cost calculation.

[0052] To determine the weighting coefficients in the above formula, a historical maintenance database was accessed, and the entropy weighting method was used for objective assignment. First, the data matrices of various indicators in the historical maintenance database were extracted, and the information entropy of displacement reduction rate, anti-buoyancy improvement rate, safety performance, and maintenance cost was calculated. Second, the data dispersion was calculated based on the information entropy; indicators with greater data dispersion were assigned higher weights. Thus, the weight of the displacement reduction rate within the structural safety restoration performance was determined. Weight of anti-buoyancy improvement rate and the security effectiveness weights in the overall objective function. Weighted by maintenance costs .

[0053] Finally, construct the overall objective function. and iterate through the set of solutions. Solve for the objective function value of each scheme The system selects values ​​that maximize the objective function value. The largest possible solution is considered the optimal maintenance intervention combination. After determining the optimal solution, the corresponding descent time series curve is automatically extracted. Equivalent surface load Grouting body design permeability and grouting expansion volume strain These core control parameters are used to generate on-site execution instructions that include equipment start-up and shutdown sequences, material formulation quantities, and construction process parameters. These instructions are then sent to the construction terminal to guide on-site maintenance operations.

[0054] Please refer to Figure 2 This embodiment 2 provides a station structure health monitoring and maintenance system based on digital twins, including: The twin benchmark model construction unit is used to construct a three-dimensional geometric model including the main structure and the surrounding soil, while establishing a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assigning the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain the digital twin benchmark model. The model parameter calibration unit is used to access the time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array in real time through the virtual and real data mapping channel. It uses an ensemble Kalman filter algorithm to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. The dominant risk factor identification unit is used to simultaneously solve the multi-physics coupling response in a real-time synchronously evolving digital twin model to calculate the structural displacement and anti-buoyancy safety factor. When the structural displacement or anti-buoyancy safety factor exceeds the preset control standard, the unit calculates the partial derivative matrix of each inversion update parameter with respect to the structural displacement and anti-buoyancy safety factor, and identifies the dominant risk factor and its spatial location based on the element with the largest absolute value in the partial derivative matrix. The intervention scheme optimization unit is used to conduct parallel virtual simulations of maintenance intervention combinations, including emergency precipitation, ballast configuration, and grouting and leak sealing, in a digital twin model for the dominant risk factors. Using the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function, it solves for the objective function value of each maintenance intervention combination scheme, selects the scheme with the optimal objective function value, and generates on-site execution instructions.

[0055] This embodiment 3 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement any step of a digital twin-based method for monitoring and maintaining the health of a station structure.

[0056] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0058] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring and maintaining the structural health of a station based on digital twins, characterized in that, include: S1. Construct a three-dimensional geometric model including the main structure and the surrounding soil, and at the same time establish a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assign the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain a digital twin benchmark model. S2. The time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array is accessed in real time through the virtual and real data mapping channel. The ensemble Kalman filter algorithm is used to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. S3. In the real-time synchronous evolution digital twin model, the multi-physics field coupled response is solved synchronously to calculate the structural displacement and anti-buoyancy safety factor; When the structural displacement or anti-buoyancy safety factor exceeds the preset control standard, calculate the partial derivative matrix of each inversion update parameter with respect to the structural displacement and anti-buoyancy safety factor, and identify the dominant risk factor and its spatial location based on the element with the largest absolute value in the partial derivative matrix. S4. For the dominant risk factors, conduct parallel virtual simulations in the digital twin model of maintenance intervention combination schemes including emergency precipitation, ballast configuration and grouting for leak sealing; use the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function to solve the objective function value of each maintenance intervention combination scheme, and select the scheme with the optimal objective function value to generate on-site execution instructions.

2. The method for monitoring and maintaining the health of station structures based on digital twins according to claim 1 is characterized in that, In step S1, a virtual-real data mapping channel is established between the physical station sensor network and the virtual model nodes, and the initial damage field identified by the current status detection is assigned to the three-dimensional geometric model. Specifically, this includes: The three-dimensional spatial coordinates of the physical station sensors and the grid coordinates of the virtual model nodes are obtained. A spatial mapping matrix is ​​constructed using the radial basis function interpolation method, and the sensor measured data are distributed to the corresponding virtual model nodes according to the distance weight. The width of structural cracks and the location of water leakage identified by the current status detection are converted into equivalent stiffness reduction coefficients and local permeability amplification coefficients. By modifying the elastic matrix and permeability tensor of the corresponding mesh elements in the three-dimensional geometric model, the spatial discretization of the initial damage field is completed.

3. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 1, characterized in that, In S2, an ensemble Kalman filter algorithm is used to drive the dynamic joint inversion and state update of the soil permeability coefficient and boundary head conditions in the benchmark model, specifically including: Based on the prior mean and prior covariance matrix of soil permeability coefficient and boundary head conditions, Latin hypercube sampling is used to generate an initial parameter set containing multiple independent samples, and the initial parameter set is concatenated with the model state variables to form an augmented state vector to obtain the initial state set. In each assimilation cycle, the innovation vector between the measured time series data stream and the model predicted observation is calculated. The Mahalanobis distance of the innovation vector is calculated using the preset observation error covariance matrix. When the Mahalanobis distance is greater than the confidence threshold set based on the chi-square distribution, the corresponding sensor data is determined to be an outlier and is removed. The cross-covariance matrix of state and observation and the Kalman gain matrix are calculated using the effective observation data after outlier removal. The ratio of the actual sample variance to the theoretical variance of the innovation vector is also calculated and used as the adaptive covariance inflation factor. The effective observation information is weighted and superimposed onto the prior state set using the Kalman gain matrix to obtain the posterior state set. The dispersion of the parameter samples in the posterior state set is amplified using the adaptive covariance expansion factor to complete the joint update of the soil permeability coefficient and boundary head conditions.

4. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 3, characterized in that, The process of constructing the adaptive covariance inflation factor specifically includes: The difference between the effective observation data after removing outliers within the current assimilation period and the mean of the model-predicted observation set is calculated to obtain the current innovation vector; Based on a sliding time window of a preset time length, the mean of the outer product of all historical innovation vectors and the current innovation vector within the sliding time window is calculated to obtain the actual covariance matrix of the innovation, and the trace value of the actual covariance matrix of the innovation is extracted as the actual sample variance. The observation error covariance matrix of the current assimilation period is added to the prior set covariance matrix of the model predicted observations to obtain the innovation theory covariance matrix, and the trace value of the innovation theory covariance matrix is ​​extracted as the theoretical variance. Calculate the ratio of the actual sample variance to the theoretical variance, and then sum the ratio with the historical adaptive covariance inflation factor of the previous assimilation period to obtain the smooth adaptive covariance inflation factor of the current assimilation period. The smoothing adaptive covariance inflation factor is multiplied by the deviation vector between each sample in the prior state set and the set mean to amplify the dispersion of the prior state set.

5. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 1, characterized in that, The calculation of the partial derivative matrices of each inversion update parameter with respect to structural displacement and anti-buoyancy safety factor in S3 specifically includes: The sum of squared errors between the calculated and measured values ​​of structural displacement and the preset evaluation function of the anti-buoyancy safety factor are constructed as the objective functional. The accompanying water pressure variable corresponding to the soil seepage control equation and the accompanying displacement variable corresponding to the structural mechanical equilibrium control equation are introduced as Lagrange multipliers. The objective functional is linearly combined with the weak form of the soil seepage control equation and the structural mechanical equilibrium control equation to construct an augmented functional. The first variational function of the augmented generalized function with respect to the forward water pressure variable and the forward displacement variable is obtained and set to zero. The accompanying water pressure control equation and the accompanying displacement control equation are derived. The accompanying water pressure control equation and the accompanying displacement control equation are solved simultaneously with the accompanying boundary conditions to obtain the accompanying water pressure field and the accompanying displacement field. The partial derivatives of the augmented functional with respect to the soil permeability coefficient and the boundary head condition are calculated to obtain the gradient analytical expression containing the inner product of the spatial gradients of the forward-modeled state variables and the associated variables. The associated hydraulic pressure field, the associated displacement field, and the forward-modeled hydraulic pressure field and displacement field obtained by forward modeling are substituted into the gradient analytical expression, and spatial integration is performed in the computational domain of the entire station structure and the surrounding soil to analytically obtain the gradient vectors of the soil permeability coefficient and the boundary head condition with respect to the target functional. The gradient vectors corresponding to all inversion update parameters are assembled according to the parameter dimensions to generate the partial derivative matrix.

6. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 5, characterized in that, The identification of dominant risk factors and their spatial locations based on the element with the largest absolute value in the partial derivative matrix in step S3 specifically includes: The gradient vector corresponding to the structural displacement target functional in the partial derivative matrix is ​​extracted as the first partial derivative row vector, and the gradient vector corresponding to the anti-buoyancy safety factor target functional is extracted as the second partial derivative row vector. The elements with the largest absolute values ​​are searched by traversing the first and second partial derivative row vectors respectively, and the first and second largest absolute value elements are extracted. Obtain the column indices of the first and second maximum absolute value elements in the partial derivative matrix, match the corresponding parameter types in the preset inversion update parameter sequence according to the column indices, and identify the matched parameter types as the dominant risk factors affecting the corresponding objective functional. The parameter types include soil permeability coefficient or boundary head conditions. The finite element mesh mapping table of the station structure and surrounding soil is queried according to the column index to obtain the grid cell number or node number where the dominant risk factor is located, and the grid cell number or node number is mapped to the digital twin three-dimensional geometric model to determine the three-dimensional spatial location of the dominant risk factor in the physical station.

7. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 1, characterized in that, The objective function in S4 is a weighted sum of structural safety restoration effectiveness and maintenance cost, specifically including: The structural safety recovery efficiency is defined as the weighted sum of the maximum displacement reduction rate and the anti-buoyancy safety factor improvement rate of the structure after virtual pre-simulation. The maintenance cost is defined as the sum of the products of the pumping volume, the volume of counterweight material, and the grouting volume involved in each maintenance intervention combination scheme and the corresponding unit price; The entropy weight method is used to calculate the weights of displacement reduction rate and anti-buoyancy improvement rate in the structural safety restoration effectiveness, as well as the weights of safety effectiveness and maintenance cost in the overall objective function, based on the data dispersion of various indicators in the historical maintenance database.

8. The method for monitoring and maintaining the structural health of a station based on digital twins according to claim 1, characterized in that, The parallel virtual simulation in the digital twin model in S4 includes a maintenance intervention combination scheme of emergency precipitation, weight configuration, and grouting for leak sealing, specifically including: For emergency precipitation, the boundary conditions of the precipitation wells are activated in the model and the precipitation depth time curve is set; for the counterweight configuration, the equivalent surface load is applied to the corresponding grid nodes of the station floor and roof; for grouting and plugging, the soil permeability coefficient within the preset range around the leakage point is reduced to the design permeability of the grouting body, and the grouting expansion volume strain is applied. During the rehearsal, an explicit-implicit alternating time integration strategy was adopted, using explicit solutions for the transient fluid impact in the early stage of intervention and implicit solutions for the consolidation settlement in the later stage of intervention.

9. A station structure health monitoring and maintenance system based on digital twins, characterized in that, include: The twin benchmark model construction unit is used to construct a three-dimensional geometric model including the main structure and the surrounding soil, while establishing a virtual-real data mapping channel between the physical station sensor network and the virtual model nodes, and assigning the initial damage field identified by the current status detection to the three-dimensional geometric model to obtain the digital twin benchmark model. The model parameter calibration unit is used to access the time-series data stream of structural deformation and pore water pressure collected by the automated monitoring array in real time through the virtual and real data mapping channel. It uses an ensemble Kalman filter algorithm to drive the soil permeability coefficient and boundary water head conditions in the benchmark model to perform dynamic joint inversion and state update, so as to obtain a parameter-calibrated digital twin model that evolves synchronously with the physical station in real time. The dominant risk factor identification unit is used to simultaneously solve the multi-physics coupling response in a real-time synchronously evolving digital twin model to calculate the structural displacement and anti-buoyancy safety factor. When the structural displacement or anti-buoyancy safety factor exceeds the preset control standard, calculate the partial derivative matrix of each inversion update parameter with respect to the structural displacement and anti-buoyancy safety factor, and identify the dominant risk factor and its spatial location based on the element with the largest absolute value in the partial derivative matrix. The intervention scheme optimization unit is used to conduct parallel virtual simulations of maintenance intervention combinations, including emergency precipitation, ballast configuration, and grouting and leak sealing, in a digital twin model for the dominant risk factors. Using the weighted sum of structural safety restoration effectiveness and maintenance cost as the objective function, it solves for the objective function value of each maintenance intervention combination scheme, selects the scheme with the optimal objective function value, and generates on-site execution instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor according to any one of claims 1-8, a method for monitoring and maintaining the health of a station structure based on digital twins.

Images