A tunnel lining service safety intelligent diagnosis method and system

Abstract

The application relates to a tunnel lining service safety intelligent diagnosis method and system, and relates to the technical field of tunnel health assessment. The method first acquires lining resistivity monitoring data, multivariate physical and chemical parameters containing temperature parameters and inspection information; resistivity data is corrected by using a temperature compensation model and a benchmark database, a time series data set is generated, and a structure health state evolution cloud map is obtained by inversion; a cloud map and multi-source information fusion analysis are used to determine disease parameters; the disease parameters are mapped and updated to a parameterized finite element benchmark model, a structure safety coefficient is calculated to realize quantitative evaluation of service safety, and finally, a visual report containing disease positions, a safety cloud map and maintenance suggestions is output according to preset threshold grading early warning. The application realizes full-process automation from medium characteristic monitoring to structure safety quantitative evaluation, significantly improves disease identification precision and early warning capability, and provides efficient and reliable support for tunnel maintenance decision-making.

Description

Technical Field

[0001] This application relates to the field of tunnel health assessment technology, and in particular to an intelligent diagnostic method and system for the service safety of tunnel linings. Background Technology

[0002] Tunnel linings are used in complex geological environments for extended periods. Under the influence of groundwater erosion, freeze-thaw cycles, and changes in surrounding rock pressure, they are prone to defects such as leakage, cracking, concrete deterioration, and backfilling. Existing monitoring methods, such as manual inspections, point sensors, and ground-penetrating radar, are subjective, have limited coverage, or are difficult to monitor continuously, making it difficult to meet the needs of full life-cycle safety management.

[0003] Because of its sensitivity to changes in concrete moisture content and cracks, the resistivity method has been applied to long-term monitoring of tunnel linings, obtaining the distribution of dielectric electrical properties through electrode array inversion imaging. Finite element numerical simulation is also widely used for stress analysis and safety assessment of tunnel structures. Environmental parameters such as temperature and humidity, as well as inspection information, are also used to assist in the evaluation of structural condition.

[0004] In existing technologies, resistivity monitoring and structural mechanics analysis are usually separate professional processes. Resistivity inversion obtains the distribution of electrical properties of the medium, while finite element analysis calculates the structural response based on mechanical parameters. These two belong to different interpretation systems, and a unified path from the state of the medium to the assessment of structural safety has not yet been formed. Summary of the Invention

[0005] To achieve fully automated quantitative diagnosis and early warning of the safety status of tunnel lining structures, this application provides a method and system for intelligent diagnosis of tunnel lining service safety.

[0006] Firstly, this application provides a method for intelligent diagnosis of tunnel lining service safety, including:

[0007] Acquire resistivity monitoring data, multivariate physicochemical parameters and inspection information of the lining structure. The multivariate physicochemical parameters shall include at least the temperature parameters used for temperature compensation.

[0008] Based on the preset benchmark database and preset temperature compensation model, the resistivity monitoring data is corrected using temperature parameters to generate a time series dataset that quantitatively characterizes the relative change in resistivity. The time series dataset is then inverted and imaged to generate a structural health state evolution cloud map that reflects the evolution of the lining medium characteristics.

[0009] By integrating and analyzing the structural health status evolution cloud map with multivariate physicochemical parameters and inspection information, the disease parameters can be determined.

[0010] Based on the preset parametric finite element reference model, the disease parameters are mapped to the mechanical model input parameters. The parametric finite element reference model is updated based on the mechanical model input parameters, and the structural safety factor is calculated based on the updated parametric finite element reference model to obtain a quantitative assessment result of the structural service safety.

[0011] The quantitative assessment results are classified and warned according to the preset safety threshold. When the warning is triggered, a visual report containing the location of the disease, safety cloud map and maintenance suggestions is generated and output based on the preset maintenance strategy, quantitative assessment results and disease parameters.

[0012] Secondly, this application provides an intelligent diagnostic system for the service safety of tunnel linings, comprising:

[0013] The data acquisition module is used to acquire resistivity monitoring data, multivariate physicochemical parameters and inspection information of the lining structure. The multivariate physicochemical parameters include at least the temperature parameters used for temperature compensation.

[0014] The data generation module is used to correct resistivity monitoring data using temperature parameters based on a preset benchmark database and a preset temperature compensation model, generate a time series dataset that quantitatively characterizes the relative change in resistivity, and perform inversion imaging on the time series dataset to generate a structural health state evolution cloud map that reflects the evolution of lining medium characteristics.

[0015] The data fusion module is used to integrate and analyze the structural health status evolution cloud map with multivariate physicochemical parameters and inspection information to determine the disease parameters;

[0016] The data processing module is used to map the fault parameters to mechanical model input parameters according to the preset parametric finite element reference model, update the parametric finite element reference model based on the mechanical model input parameters, and calculate the structural safety factor based on the updated parametric finite element reference model to obtain a quantitative assessment result of the structural service safety.

[0017] The data output module is used to provide graded early warnings for quantitative assessment results based on preset safety thresholds. When an early warning is triggered, it generates and outputs a visual report containing the location of the disease, a safety cloud map, and maintenance suggestions based on preset maintenance strategies, quantitative assessment results, and disease parameters.

[0018] In summary, the beneficial technical effects of this application are as follows: This intelligent diagnostic method accurately acquires resistivity monitoring data, multivariate physicochemical parameters, and inspection information; uses a temperature compensation model to correct resistivity data and perform inversion imaging; combines multivariate information fusion analysis to determine disease parameters; relies on a parametric finite element benchmark model to achieve quantitative calculation of structural safety factors; and combines graded early warning and visual report output to ensure the entire process of tunnel lining diagnosis, from "full monitoring coverage to early damage identification, quantitative safety assessment, and precise early warning." It endows the lining structure with comprehensive capabilities for continuous monitoring across the entire area, early warning of local damage, and quantitative assessment of safety status, overcoming the shortcomings of existing monitoring methods, such as strong subjectivity, limited coverage, and lack of correlation between resistivity monitoring and mechanical assessment. This improves the accuracy and timeliness of tunnel lining service safety diagnosis and provides technical support for lining safety monitoring, assessment, and early warning. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an intelligent diagnostic method for the service safety of tunnel lining according to an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of the parametric finite element reference model provided in the embodiments of this application; the left figure is the initial health state model, and the right figure is the model of the safe region and the damaged region. Detailed Implementation

[0021] The present application will be further described in detail below with reference to the accompanying drawings.

[0022] Reference Figure 1 This application discloses a method for intelligent diagnosis of tunnel lining service safety, comprising:

[0023] S1: Obtain resistivity monitoring data, multivariate physicochemical parameters, and inspection information of the lining structure. The multivariate physicochemical parameters shall include at least the temperature parameters used for temperature compensation.

[0024] Among them, the lining structure is a permanent load-bearing structure in tunnel engineering used to support the surrounding rock, isolate soil pressure and groundwater, and ensure the safety of the internal operating space of the tunnel. It is usually made of reinforced concrete and is arched or curved to fit the shape of the tunnel cross section. Typical cross sections include key load-bearing parts such as the arch crown, arch waist, sidewalls, and invert arch. It directly bears external loads such as soil pressure, water pressure, and train vibration, and is the core structure to ensure the safe operation of the tunnel.

[0025] In this embodiment, the resistivity monitoring data is a global distribution of electrical data reflecting the internal density, moisture content changes, crack development, voids, and steel reinforcement corrosion of the lining concrete. This data is collected by a resistivity monitoring network deployed inside and on the surface of the lining structure. At key sections of the lining structure, stainless steel or graphene composite electrodes are pre-embedded in a two-dimensional grid or three-dimensional array. These electrodes are connected to a multi-channel high-density resistivity acquisition instrument, which automatically collects data at fixed time intervals, thus obtaining the original resistivity distribution data within the lining structure's cross-sectional area. Under normal operating conditions, during the rainy season, temperature changes, or structural deterioration, the resistivity values ​​will exhibit regular or abnormal changes.

[0026] In this embodiment, the multi-dimensional physicochemical parameters are multi-point measured physical and chemical indicators used to assist in verifying, correcting, and supplementing resistivity monitoring results. These parameters include at least temperature parameters, and may also include strain parameters, humidity parameters, corrosion potential parameters, etc. Specifically: temperature parameters are acquired by digital temperature sensors deployed inside the lining structure; strain parameters are acquired by fiber optic strain sensors deployed at key stress points of the lining structure; humidity parameters are acquired by humidity sensors deployed at construction joints and high-risk leakage areas; and corrosion potential parameters are acquired by steel corrosion sensors deployed near the reinforcing bars. These various sensors collectively form a multi-dimensional verification sensor array, arranged on the same monitoring cross-section as the resistivity electrodes, achieving synchronous and same-point acquisition with resistivity data. The sensors and electrodes are deployed at key locations on the same monitoring cross-section, such as the arch crown, arch waist, sidewalls, construction joints, and high-risk leakage areas, to achieve synchronous acquisition with resistivity data.

[0027] In this embodiment, the inspection information is the lining appearance and internal condition information obtained by manual on-site inspection, automated inspection equipment or periodic detection by ground-penetrating radar. It includes text records, on-site photos, videos, defect markings, radar profile results, etc. It mainly records whether there are cracks, water seepage, spalling, flaking, efflorescence, construction joint deformation, etc. on the surface of the lining structure. It is used to corroborate the resistivity monitoring results and improve the accuracy of defect judgment.

[0028] In this embodiment, the temperature parameter is a key parameter used for temperature compensation and elimination of environmental interference in resistivity monitoring data, and is directly measured by a digital temperature sensor placed near the electrodes. Temperature significantly affects the resistivity of concrete, and resistivity values ​​measured in the same area at different temperatures show significant deviations. Therefore, it is necessary to collect temperature parameters simultaneously to provide a basis for subsequent resistivity correction.

[0029] Understandably, by simultaneously acquiring three types of data—resistivity monitoring data, multivariate physicochemical parameters, and inspection information—a multi-dimensional sensing system covering the interior, surface, entire area, and multiple points of the lining structure can be constructed. This ensures large-scale monitoring of internal deterioration while improving the reliability of judgments through point sensors and inspection information.

[0030] S2: Based on the preset benchmark database and preset temperature compensation model, the resistivity monitoring data is corrected using temperature parameters to generate a time series dataset that quantitatively characterizes the relative change in resistivity. The time series dataset is then inverted and imaged to generate a structural health state evolution cloud map that reflects the evolution of the lining medium characteristics.

[0031] The specific method for generating a time-series dataset that quantitatively characterizes the relative change in resistivity by correcting resistivity monitoring data using temperature parameters based on a preset benchmark database and a preset temperature compensation model includes:

[0032] S201: Obtain temperature parameters from multivariate physicochemical parameters.

[0033] The multivariate physicochemical parameters are acquired in real time by a multivariate verification sensor array in the multidimensional sensing layer. Temperature parameters are directly measured by digital temperature sensors positioned near the resistivity electrodes and embedded within the lining concrete, continuously and stably reflecting the true internal temperature of the lining structure at the corresponding location. The temperature parameters are output in degrees Celsius, serving as the direct input for subsequent resistivity-temperature compensation. It is understandable that temperature is the most significant environmental factor affecting concrete resistivity measurements; therefore, acquiring accurate and synchronized temperature parameters is a prerequisite for ensuring the reliability of subsequent resistivity correction results.

[0034] S202: Using a preset temperature compensation model, the resistivity monitoring data is corrected based on the acquired temperature parameters to eliminate the interference of temperature changes on the resistivity measurement value and obtain a corrected resistivity value uniform to the preset reference temperature.

[0035] The preset temperature compensation model adopts a concrete resistivity temperature correction model based on the Arrhenius equation. This concrete resistivity temperature correction model is calibrated through indoor isothermal resistivity tests and is suitable for correcting the temperature influence on tunnel lining concrete materials. The specific expression is as follows:

[0036] ;

[0037] In the formula: ρ c The resistivity is the temperature-compensated corrected value, in units of: ;ρ m These are the raw measured resistivity values ​​collected on-site by a multi-channel resistivity acquisition instrument. Unit: ;T m The absolute internal temperature of the concrete as measured at the current moment, in Kelvin (K); T b The preset reference temperature is taken from the average temperature of the tunnel lining when it is in its initial healthy state, and the unit is Kelvin (K). The temperature characteristic coefficient related to concrete materials is, in essence, the ratio of the activation energy of ion migration in the pore fluid of concrete to the gas constant. It is pre-calibrated through standard laboratory tests, and its unit is Kelvin (K). Preset reference temperature. The average temperature of the tunnel lining in its initial healthy state should be used as a priority, or a uniform reference temperature can be used. e represents the natural constant. Calculations using this model can uniformly convert resistivity values ​​measured at different temperatures to a reference temperature, completely eliminating resistivity value deviations caused by diurnal temperature variations, seasonal changes, and environmental temperature fluctuations, thus ensuring the comparability of monitoring data from different times and locations.

[0038] Understandably, standardizing the original resistivity using a temperature compensation model can eliminate environmental interference and retain effective electrical signals that truly reflect the internal density, moisture content, and damage level of the lining.

[0039] S203: Locate the reference resistivity value under the health condition corresponding to the current monitoring location from the preset reference database.

[0040] The preset benchmark database is an original database established during the initial stage of tunnel completion, when the lining structure is in a healthy state—in good condition, without cracks, leaks, or deterioration—using the same electrodes and the same data acquisition methods. The benchmark database stores the corresponding benchmark resistivity value for each electrode location and each spatial coordinate point. During data processing, the system automatically matches and retrieves the benchmark resistivity value of the same location and monitoring point in a healthy state based on the current monitoring electrode number or spatial coordinates, using it as a reference for subsequent relative change calculations.

[0041] Understandably, using the reference resistivity under healthy conditions as a comparison benchmark can accurately reflect the degree of deterioration of the structure from its initial state to its current state, providing a unified and reliable basis for subsequent disease identification.

[0042] S204: Compare the corrected resistivity value with the reference resistivity value and calculate the relative rate of change between the two.

[0043] After temperature compensation is completed, the corrected resistivity value at the same monitoring point is compared with the reference resistivity value point by point, and the relative rate of change of resistivity is calculated. The calculation formula is as follows:

[0044] ;

[0045] In the formula: η is the relative rate of change of resistivity, expressed as a percentage ( ) Measurement; ρ c The current resistivity value after temperature compensation, in units of: ;ρ b This is the health baseline resistivity value for this monitoring point, in units of: When the relative rate of change is positive, it indicates an increase in resistivity, which usually corresponds to a dry area, reduced density, and voids behind the concrete. When the relative rate of change is negative, it indicates a decrease in resistivity, which usually corresponds to a water-containing area, leakage, crack development, and concrete deterioration.

[0046] Understandably, using relative rate of change rather than absolute resistivity values ​​for analysis can further eliminate systematic errors caused by initial material inhomogeneity and differences in electrode installation, making the characterization of structural state changes more sensitive and accurate.

[0047] S205: Based on the relative rate of change, generate a time-series dataset that quantitatively characterizes the relative change in resistivity, using a benchmark database as a reference.

[0048] The resistivity relative change rate calculated for each monitoring time and each monitoring point is arranged sequentially according to the monitoring time. Combined with the corresponding electrode location, spatial coordinates, and monitoring time, a complete time-series dataset is formed. This dataset uses a benchmark database as a unified reference, time as the axis, and relative change rate as the quantitative indicator. It can continuously and intuitively reflect the resistivity evolution law of each location of the lining structure during long-term service. It is the core input data for subsequent resistivity time-series inversion imaging and generation of structural health status evolution cloud maps.

[0049] Understandably, time-series datasets transform discrete single-point monitoring data into continuous, spatiotemporally corresponding quantized sequences, providing a standardized, high-quality data foundation for subsequent inversion imaging, dynamic feature extraction, and multi-source fusion diagnosis.

[0050] Furthermore, specific methods for generating structural health state evolution cloud maps reflecting the evolution of lining medium characteristics by performing inversion imaging on time-series datasets include:

[0051] S211: The corrected resistivity values ​​and their corresponding spatial coordinates at each monitoring time in the time series dataset are preprocessed, outlier data are removed, and spatial interpolation is performed to form the apparent resistivity distribution data on the regular grid corresponding to each monitoring time.

[0052] In the data preprocessing stage, quality control is first carried out on the resistivity data at each monitoring time in the time series dataset: abnormal resistivity values ​​caused by poor electrode contact, signal interference, instrument failure, etc., are eliminated based on the difference method of adjacent electrode data to ensure data validity; then, Kriging interpolation, inverse distance weighted interpolation, or spline interpolation methods are used to spatially interpolate the resistivity data after outlier elimination, transforming discrete electrode measuring point data into continuous lining cross-sectional distribution data, and finally forming apparent resistivity distribution data on a regular grid corresponding to each monitoring time. The resolution of the regular grid is determined according to the tunnel cross-sectional size and monitoring accuracy to ensure that the grid size is adapted to the electrode layout density and lining structural characteristics.

[0053] S212: Input the apparent resistivity distribution data at each monitoring time into the preset resistivity tomography inversion model, and obtain the true resistivity distribution inside the lining after inversion at the corresponding time through iterative solution.

[0054] Each monitoring moment refers to a periodic resistivity data collection point conducted at fixed time intervals. The monitoring interval is determined based on the tunnel's service status: during the stable operation period, a 24-hour interval is used as the standard collection interval; during the rainy season, freeze-thaw period, periods of active geological activity, or in areas prone to structural defects, the monitoring interval is shortened to 6 to 12 hours to improve the sensitivity of detecting structural deterioration. All monitoring moments are marked with absolute timestamps and are strictly synchronized with the collection moments of multi-dimensional physicochemical parameters such as temperature, humidity, and strain to ensure temporal consistency.

[0055] It is worth noting that the preset resistivity tomography inversion model is a time-lapse resistivity inversion model based on finite element forward modeling and least squares smoothing constraints. This model is pre-built during the system initialization phase, and the specific construction process is as follows:

[0056] (1) Model geometry and mesh construction

[0057] Based on the actual cross-sectional dimensions of the tunnel lining, a two-dimensional plane strain calculation geometric model was established. The cross-sectional area includes the lining structure, the surrounding rock behind it, and a certain range of surrounding media. A structured quadrilateral mesh was used to discretize the computational domain, with the mesh element size matched to the electrode spacing. The mesh resolution inside the lining was higher than that of the external surrounding rock, ensuring high-precision characterization of the internal region of the lining. The electrode positions served as potential observation points and current injection points, directly mapped to the mesh nodes to form a fixed observation configuration file.

[0058] (2) Construction of forward model layers

[0059] The model includes a forward modeling layer, which solves the DC electric field control equations based on the control volume method or the finite element method. The core control equations are:

[0060] ;

[0061] Where σ is the conductivity, in Siemens units per meter (S / m), and σ = 1 / ρ, where ρ is the resistivity, in units of... φ represents electric potential, measured in volts (V). The supply current is treated as the equivalent linear current density per unit longitudinal length of the tunnel under the simplified assumption of two-dimensional plane strain, in amperes per meter (m). ); The Dirac function is a two-dimensional function, with units of: ; The spatial position vector of the field point, unit: meters ( ); Location of the power supply electrode, unit: meter (m) The forward modeling layer is used to quickly calculate the surface potential response under any power supply and measurement electrode combination, given a known resistivity distribution, and outputs the theoretical value of apparent resistivity.

[0062] (3) Sensitivity matrix calculation layer

[0063] The model includes a sensitivity matrix calculation layer to calculate the contribution of resistivity change in each grid cell to the observed apparent resistivity. The sensitivity matrix J has dimensions of the number of observation points × the number of grid cells, and the matrix elements... The rate of change of the apparent resistivity at the i-th observation point caused by the change in resistivity of the j-th element is obtained by solving the problem using the finite element perturbation method or the adjoint state method.

[0064] (4) Smoothing constraint layer

[0065] The model includes a smoothing constraint layer to suppress noise oscillations and artifacts during the inversion process. The smoothing constraint layer uses a second-order Laplace smoothing operator to construct a spatial smoothing matrix C, ensuring continuous resistivity changes between adjacent units and avoiding non-physical abrupt distributions.

[0066] (5) Model input-output layer

[0067] The model includes an input / output interface layer, which reads the apparent resistivity distribution data, electrode coordinate files, and mesh files at various time points, and outputs the inverted true resistivity distribution results. The output format is a resistivity numerical matrix corresponding to the mesh cells, which can be directly used for subsequent visualization and changing field calculations.

[0068] The aforementioned forward modeling layer, sensitivity matrix calculation layer, smoothing constraint layer, and input / output interface layer are connected sequentially to form a complete resistivity tomography inversion model.

[0069] In the inversion calculation, the objective function is composed of a data fitting term and a smoothing constraint term, specifically in the form of:

[0070] ;

[0071] In the formula, Φ is the overall objective function; Φ d Φ is the data fitting term; λ is the regularization coefficient; r These are model constraint terms. Among them, the data fitting term Φ... d The expression is:

[0072] ;

[0073] In the formula, d o This is the measured apparent resistivity vector, in units of: ;d c For forward modeling, calculate the apparent resistivity vector, unit: W d This is the data weight matrix;

[0074] The expression for the model constraint term Φr is:

[0075] .

[0076] In the formula, m is the resistivity distribution vector, with units of: ;m T C is the transpose of the resistivity distribution vector; C is the smoothing constraint matrix; C T This is the transpose of the smooth constraint matrix.

[0077] The physical meaning of the objective function is: to make the spatial distribution of resistivity as smooth and continuous as possible, while ensuring that the calculated apparent resistivity is as close as possible to the measured apparent resistivity.

[0078] In the iterative solution process, the conjugate gradient method is used to minimize the objective function. The specific steps are as follows:

[0079] A11: Using the resistivity distribution of the baseline healthy state as the initial model m (0) ;

[0080] A12: Calculate the forward response based on the current model. The forward response is the apparent resistivity vector calculated using the forward model. Define the residual vector ,in The difference between the measured apparent resistivity vector and the forward-modeled apparent resistivity vector, in units of: ;

[0081] A13: Calculate the gradient vector of the objective function. ;

[0082] A14: Construct a conjugate search direction from the gradient vector, perform a one-dimensional search along this direction, and update the model parameters. ,in, The optimal iteration step size coefficient for the k-th iteration is determined through a one-dimensional search. This represents the conjugate search direction vector for the k-th iteration, in units of: It is constructed by the gradient of the objective function and the direction of the preceding iterations;

[0083] A15: Repeat steps A12–A14 until the preset convergence condition is met.

[0084] It is worth noting that the preset convergence condition satisfies both of the following:

[0085] The relative residual of the data is less than the set threshold:

[0086] ;

[0087] The model parameter update amount is less than the set threshold:

[0088] ;

[0089] In the formula, The vector's 2-norm; k is the iteration number; m (k) For the k-th iteration resistivity model, the unit is: , The relative residual convergence threshold is set independently based on the noise level of the observed data. The convergence threshold for the relative update amount of model parameters is set independently based on the sensitivity of the model parameters, and the two can take different values.

[0090] When both conditions are met, the inversion is considered to have converged, and the iteration is stopped. The resistivity distribution of the model at this time is the true resistivity distribution inside the lining at that moment.

[0091] S213: Compare the true resistivity distribution obtained from the inversion at each monitoring time with the health state benchmark resistivity distribution at the corresponding location in the benchmark database, calculate the relative rate of change of resistivity, and generate the resistivity change field at each monitoring time.

[0092] After obtaining the true resistivity distribution at each time point, the system automatically uses the health status reference resistivity distribution data at the corresponding location in the spatial coordinate positioning reference database to calculate the relative rate of change between the true resistivity and the reference resistivity unit by unit or node by node. The calculation formula is as follows:

[0093] ;

[0094] In the formula, η i Let be the relative rate of change of resistivity of the i-th computational unit / node; This is the inverted true resistivity value for this element / node, in units of: ; The reference resistivity value for the health status at the corresponding location, in units of: The relative rate of change of each unit / node is visualized and mapped according to its spatial distribution to generate a resistivity change field at each monitoring time. The positive and negative values ​​and spatial distribution of the resistivity change field can directly characterize the change in medium properties of the lining structure compared to the healthy state. Positive rate of change corresponds to the state of dryness, decreased density, and voids behind the lining, while negative rate of change corresponds to the state of water content, crack development, and water leakage.

[0095] S214: Combine the resistivity change fields generated at different monitoring times according to the time series, use time series analysis to extract the dynamic characteristics of resistivity change, and generate a structural health state evolution cloud map based on the dynamic characteristics that can intuitively show the spatiotemporal evolution law of the lining medium characteristics.

[0096] First, the resistivity change fields at each monitoring time are spliced ​​together in chronological order to construct a three-dimensional spatiotemporal evolution data volume containing spatial dimensions (mileage, circumferential) and temporal dimensions. Then, time-series analysis methods are used to extract the dynamic characteristics of resistivity changes.

[0097] It is worth noting that specific methods for extracting the dynamic characteristics of resistivity changes include:

[0098] S2141: Arrange the resistivity change fields at each monitoring time in chronological order to construct a three-dimensional spatiotemporal data volume. The three-dimensional spatiotemporal data volume contains multiple spatial units, and each spatial unit corresponds to a time-series curve reflecting the evolution of resistivity at that location over time.

[0099] The two-dimensional resistivity change fields corresponding to each moment are stacked and combined sequentially along the time axis to form a three-dimensional spatiotemporal data volume containing circumferential coordinates, longitudinal coordinates, and time coordinates. The three-dimensional spatiotemporal data volume covers the entire computational grid of the entire monitoring section. Each grid node corresponds to an independent spatial cell. Each spatial cell stores the relative rate of change of resistivity at that location at all monitoring moments, thus forming a complete time-series curve that reflects the continuous evolution of resistivity at that location over service time. The time-series curve uses the monitoring moment as the abscissa and the relative rate of change of resistivity as the ordinate, completely recording the electrical evolution of the structure from a healthy state to the current moment.

[0100] S2142: Input the three-dimensional spatiotemporal data volume into the preset physical information constrained temporal decomposition model. The physical information constrained temporal decomposition model has built-in physical evolution prior knowledge. Based on the physical evolution prior knowledge, the temporal curve of each spatial unit is decomposed into trend component, seasonal component and anomaly component.

[0101] The preset physical information constraint time series decomposition model is a time series decomposition model based on the physical laws of concrete deterioration. The physical evolution prior knowledge built into the model includes: the long-term change of tunnel lining resistivity is dominated by structural deterioration and exhibits monotonous slow change characteristics; the periodic changes in ambient temperature and humidity will cause the resistivity to show annual or seasonal fluctuations; sudden cracks, leakage, void expansion and other defects will cause instantaneous changes in resistivity.

[0102] The physical information-constrained time-series decomposition model employs a calculation method combining weighted least squares fitting and physical constraint regularization. Each time-series curve is decomposed point-by-point, yielding three independent components: the trend component represents long-term resistivity changes caused by irreversible structural damage such as material deterioration, crack propagation, and steel corrosion within the lining; this component exhibits a stable rate of resistivity change and a unidirectional trend. The seasonal component represents periodic resistivity fluctuations caused by periodic changes in environmental factors such as temperature and humidity; this component has a fixed period and stable amplitude. The anomaly component represents instantaneous, large-scale, and irregular resistivity mutations caused by events such as sudden loads, localized leakage, crack penetration, and void propagation; these mutations have significant amplitudes and short durations. The sum of the values ​​of these three components at the same moment equals the corresponding value of the original time-series curve.

[0103] It should be noted that the physical information-constrained time-series decomposition model is pre-constructed during the system initialization phase. With the physical mechanism of concrete resistivity degradation as the core constraint, the model is composed of a time-series input layer, a physical constraint layer, a component decomposition layer, and a component output layer connected sequentially. The time-series input layer receives the three-dimensional spatiotemporal data volume, performs data format unification and outlier preprocessing, and inputs the standardized time-series curves of each spatial unit into subsequent calculation layers. The physical constraint layer embeds three physical characteristics—long-term monotonic gradual change, periodic stable fluctuation, and instantaneous abrupt change—as hard constraints into the decomposition process, limiting the variation patterns and amplitude ranges of trend components, seasonal components, and outlier components to ensure that the decomposition results conform to engineering physical laws. The component decomposition layer uses an improved weighted least squares fitting and regularization optimization algorithm to optimally separate the time-series curves under physical constraints, simultaneously outputting trend components, seasonal components, and outlier components to ensure that each component is independent and has a clear physical meaning. The component output layer outputs the three types of components obtained from the decomposition according to spatial coordinates and timestamps, providing standardized data for subsequent environmental interference removal and dynamic feature extraction. This model combines physical priors with numerical optimization to achieve time-series decomposition without physical distortion, effectively improving the accuracy and reliability of deterioration feature extraction.

[0104] S2143: Based on seasonal components, identify the amplitude and phase of periodic environmental disturbances, generate an environmental baseline correction field, and use the environmental baseline correction field to strip the trend components and anomalous components from environmental disturbances, thereby obtaining the environmentally corrected trend components and anomalous components.

[0105] The seasonal component reflects the complete spatiotemporal distribution of periodic environmental disturbances. Global statistics are performed on the seasonal components of all spatial units to extract the uniform period, peak amplitude, and phase characteristics of environmental disturbances, constructing an environmental baseline correction field covering the entire monitoring section. The environmental baseline correction field is a two-dimensional spatial field, completely corresponding to the spatial coordinates of the resistivity change field. Each spatial unit corresponds to an environmental disturbance correction coefficient, used to characterize the intensity of the influence of environmental fluctuations at that location. The corresponding values ​​of the environmental baseline correction field are subtracted from the original trend component and the original anomaly component, respectively, to completely remove periodic environmental disturbances. This yields an environmentally corrected trend component that reflects only the actual structural deterioration, and an environmentally corrected anomaly component that reflects only sudden damage. This eliminates the interference of environmental factors such as temperature and humidity on the judgment of structural deterioration, ensuring the authenticity and reliability of dynamic feature extraction.

[0106] S2144: Based on the trend components after environmental correction, calculate the resistivity change rate field and change acceleration field of each spatial unit; identify the deterioration acceleration region based on the spatial distribution of the change acceleration field, and track the displacement of the spatial boundary of the deterioration acceleration region between adjacent monitoring times to generate the boundary expansion trajectory.

[0107] Specifically, the first and second-order time-differences of the environmentally corrected trend components are used to calculate the resistivity change rate and acceleration for each spatial cell. The combination of the rates of all spatial cells forms the resistivity change rate field (unit: % / d), and the combination of the accelerations of all spatial cells forms the resistivity change acceleration field (unit: % / d²). The resistivity change rate field characterizes the speed of degradation development; a larger value indicates a faster degradation rate. The change acceleration field characterizes the trend of degradation rate changes; a positive value indicates that degradation continues to accelerate, while a negative value indicates that degradation tends to level off.

[0108] By setting an acceleration threshold, regions with varying accelerations greater than the threshold are marked as deterioration acceleration regions. The selectable range for the acceleration threshold is... The preferred value is 0.3% / d². A spatial connectivity identification algorithm is used to extract the outer contour boundary of the accelerated deterioration area. The boundary coordinate changes are compared between continuous monitoring times, and the direction and distance of boundary movement are recorded. These are then connected in chronological order to form the deterioration area boundary expansion trajectory. The trajectory is stored as a sequence of coordinate points, intuitively reflecting the expansion direction, speed, and extent of the diseased area.

[0109] It is worth noting that the spatial connectivity identification algorithm is a spatial region extraction algorithm based on the binarization segmentation and region growing method of the resistivity change acceleration field. This algorithm is pre-built during the system initialization phase. The algorithm first marks the cells in the changing acceleration field that are greater than the acceleration threshold as target cells. Based on the 8-neighbor or 4-neighbor connectivity rule, it performs region growing starting from the initial seed point, aggregating spatially adjacent grid cells that belong to the same deterioration acceleration into independent connected regions. Then, it performs boundary tracing for each connected region, extracting the coordinates of the geometric center point of the outermost grid cell in a clockwise or counterclockwise order to form a closed outer contour boundary point set. The boundary coordinates output by the algorithm are completely aligned with the spatial coordinates of the lining section, ensuring the accuracy of the boundary position.

[0110] S2145: Based on the abnormal components after environmental correction, identify the resistivity change time and magnitude of each spatial unit, mark the spatial units that meet the preset change conditions as disease event points, and aggregate to generate a spatiotemporal distribution dataset of disease events.

[0111] The preset mutation conditions include: the absolute value of the mutation amplitude of the relative rate of change of resistivity is greater than or equal to 5%, and the duration of the mutation does not exceed two monitoring times, with an interval of 6h to 24h between the two monitoring times.

[0112] It is worth noting that a full-domain point-by-point scan is performed on the environmentally corrected anomalous components to determine whether each spatial unit meets the abrupt change conditions at each monitoring moment. The moment when the abrupt change conditions are met is marked as the abrupt change moment, and the corresponding amplitude is marked as the abrupt change amplitude. Spatial units that meet the abrupt change conditions are marked as disease event points, and their spatial coordinates, abrupt change moment, abrupt change amplitude, and abrupt change direction are recorded. The abrupt change direction includes abrupt changes in resistivity increase and decrease. All disease event points identified within the monitoring period are aggregated and organized according to time and spatial location to form a spatiotemporal distribution dataset of disease events. This dataset contains the location, occurrence time, change amplitude, and change type of all sudden diseases, providing direct evidence for local damage early warning.

[0113] S2146: Based on the spatial coupling relationship between the trend component and the anomaly component after environmental correction, calculate the comprehensive index of the frequency of anomalies and the degree of trend deterioration in each spatial region, and identify the region whose comprehensive index exceeds the preset risk threshold as a deterioration-sensitive region.

[0114] In this process, for each spatial region, the number of abrupt changes in the anomalous component within a set time period is counted to obtain the anomaly frequency. Simultaneously, the trend component is normalized globally, and the relative rate of change of resistivity at the current moment is calculated compared to the total change at the initial moment to obtain the degree of trend deterioration. A weighted summation method is used to construct the comprehensive index ω, calculated using the following formula:

[0115] ;

[0116] in, For frequency weighting coefficients; The weighting coefficient represents the degree of deterioration. The normalized anomaly occurrence frequency is represented by the anomaly occurrence frequency. Divide by the preset reference frequency get; The degree of trend deterioration after normalization is represented by the total change of the trend component at the current time relative to the initial time. Divided by the preset benchmark change range The preset risk threshold is a critical value for judging the comprehensive index, determined by engineering experience and experimental calibration. Areas with a comprehensive index greater than the preset risk threshold are identified as deterioration-sensitive areas. These areas have poor structural stability, rapid deterioration, and are prone to sudden damage, making them key areas for monitoring and reinforcement. The range of the preset risk threshold is as follows: .

[0117] S2147: The resistivity change rate field, resistivity change acceleration field, boundary expansion trajectory, spatiotemporal distribution dataset of disease events, and deterioration-sensitive areas are used together as dynamic features of resistivity change.

[0118] The five types of dynamic characteristics described above comprehensively characterize the spatiotemporal evolution of lining resistivity from different dimensions: the resistivity change rate field and the change acceleration field reflect the dynamic characteristics of deterioration development; the boundary expansion trajectory reflects the spatiotemporal expansion pattern of the deteriorated area; the spatiotemporal distribution dataset of disease events reflects the occurrence pattern of local sudden diseases; and the deterioration-sensitive area reflects the distribution of weak parts in the overall structure. These five types of characteristics complement and corroborate each other, forming a complete dynamic characteristic system, providing comprehensive, accurate, and reliable quantitative basis for subsequent generation of structural health status evolution cloud maps and multi-source fusion disease diagnosis.

[0119] In summary, through steps such as constructing a three-dimensional spatiotemporal data volume, decomposing physical constraints into a temporal sequence, removing environmental interference, and extracting multi-dimensional dynamic features, it is possible to accurately separate and extract all dynamic features reflecting the true deterioration of the structure from the original time-series monitoring data, completely eliminate periodic environmental interference, and achieve quantitative description and visual representation of the lining deterioration process, providing solid data support for subsequent health status assessment and disease diagnosis.

[0120] It is worth noting that specific methods for generating structural health state evolution cloud maps based on dynamic features that can intuitively display the spatiotemporal evolution law of lining medium characteristics include:

[0121] S2151: The resistivity change rate field and the resistivity change acceleration field are vector-fused to construct a vector field characterizing the dynamics of degradation evolution; wherein, the vector direction is determined by the gradient direction of the change acceleration field, and the vector magnitude is characterized by the weighted fusion result of the resistivity change rate value and the change acceleration value.

[0122] Among them, the resistivity change rate field is used to characterize the speed of deterioration development, and the resistivity change acceleration field is used to characterize the severity of deterioration development. Both are two-dimensional scalar fields that correspond one-to-one with the grid of the lining monitoring section. During vector fusion, the spatial gradient of the acceleration field is first solved, with the gradient direction pointing towards the direction of most severe deterioration development. This direction is used as the element direction of the fused vector field. Then, a weighted summation method is used to calculate the vector amplitude, using the following formula:

[0123] .

[0124] In the formula: The magnitude of the fused vector is dimensionless. This is the rate weighting coefficient, dimensionless, with a value range of [value range missing]. ; This is the normalized value of the resistivity change rate, dimensionless, derived from the resistivity change rate. Divide by the preset reference rate get; The acceleration weighting coefficient is dimensionless and its value range is [value range missing]. ; This is the normalized value of the acceleration due to resistivity change, dimensionless, derived from the acceleration due to resistivity change. Divided by the preset reference acceleration The weighting coefficients k1 and k2 satisfy k1+k2=1, and the resulting vector field can simultaneously reflect the direction and intensity of degradation, forming a dynamic characteristic basis.

[0125] S2152: Periodically map the amplitude of the environmental baseline correction field to generate an environmental interference background layer.

[0126] Among them, the environmental baseline correction field is a two-dimensional spatial field that reflects the periodic fluctuations of temperature and humidity. The values ​​of each unit in the correction field are mapped to different colors or transparency according to the magnitude of the amplitude to form an environmental interference background layer, which is used to intuitively display the distribution of the strength of environmental interference. The environmental interference background layer corresponds completely with the grid coordinates of the lining section, without changing the expression of the core information of structural deterioration.

[0127] S2153: Map the boundary extension trajectory to the resistivity change field at each monitoring time to generate a time-stamped deterioration region boundary layer.

[0128] The trajectory of the deteriorated area boundary expansion is stored as a sequence of coordinate points at consecutive times. The boundary contour corresponding to each time moment is projected onto the resistivity change field at that time moment, and superimposed to form a deteriorated area boundary layer. Each boundary is labeled with the corresponding monitoring time to clearly show the expansion range, direction of advancement and development process of the deteriorated area over time.

[0129] S2154: Based on the spatiotemporal distribution dataset of disease events, generate a disease event labeling layer with time labels in the resistivity change field at the corresponding time.

[0130] The disease event spatiotemporal distribution dataset includes disease location, mutation time, mutation magnitude and mutation type. Each disease event is marked with spatial coordinates on the resistivity change field at the corresponding time. Different shapes or colors are used to distinguish between rising mutations and falling mutations, and the time of occurrence of the event is marked to form a disease event marker layer, which is used to intuitively locate the location and time of occurrence of sudden diseases.

[0131] S2155: Perform spatial correlation analysis between the deterioration-sensitive areas and the disease event marker layer to generate a risk level distribution layer.

[0132] The spatial extent of the deterioration-sensitive areas is overlaid and matched with the disease event marker layer. The sensitivity and frequency of events in each area are statistically analyzed. A four-level risk classification rule is used to generate a risk level distribution layer, with risk levels ranging from low to high: low risk, medium risk, high risk, and extremely high risk. The risk level distribution layer visually displays the structural safety risk distribution using color grading. The correspondence between the four risk levels and color grading is as follows: low risk corresponds to green fill, indicating stable resistivity changes, no abnormal abrupt changes, slight deterioration, and a healthy structural state; medium risk corresponds to yellow fill, indicating a slow deterioration trend, occasional abnormal fluctuations, and a generally stable structural state that requires attention; high risk corresponds to orange fill, indicating accelerated deterioration, increased frequency of abnormal events, and obvious signs of disease development; extremely high risk corresponds to red fill, indicating severe deterioration, significant resistivity abrupt changes, clear disease characteristics, and a significantly reduced structural safety reserve.

[0133] The risk level distribution layer fills each of the four color levels according to the comprehensive index value of the deterioration risk corresponding to each grid unit. The color uniformly covers the corresponding grid unit with clear boundaries, no overlap, and no omissions. It accurately corresponds to key parts of the lining section such as the arch, arch waist, sidewall, and invert arch. Maintenance personnel can directly and quickly identify the location, range, and concentration of high-risk and extremely high-risk areas through the color distribution, realizing intuitive, fast, and accurate judgment of structural safety risks.

[0134] S2156: The vector field, environmental disturbance background layer, deteriorated area boundary layer, disease event marker layer, and risk level distribution layer are spatiotemporally aligned and multi-layered to generate a time-driven structural health state evolution cloud map. As the structural health state evolution cloud map evolves in the time dimension, it simultaneously presents the expansion trajectory of the deteriorated area, the spatiotemporal distribution of periodic environmental disturbances, the occurrence time and location of disease events, and the spatial distribution of risk levels.

[0135] The multi-layer fusion uses a unified lining spatial coordinate system and timestamp as a benchmark. It sequentially overlays environmental interference background layer, resistivity change field, vector field, deterioration area boundary layer, disease event marker layer, and risk level distribution layer from bottom to top to form a multi-element integrated dynamic structural health status evolution cloud map. The structural health status evolution cloud map supports timeline dragging and playback, which can display the entire process of the evolution of the internal medium of the lining moment by moment, providing an intuitive and visual basis for disease diagnosis and safety assessment.

[0136] Understandably, the above steps combine benchmark database comparison with temperature compensation correction. First, the measured resistivity is uniformly converted to the benchmark temperature using a temperature compensation model to eliminate environmental temperature interference. Then, the relative rate of change is calculated and a time-series dataset is constructed. Subsequently, the true resistivity distribution is obtained through data preprocessing and resistivity tomography inversion, generating a time-series resistivity change field. Dynamic features are then extracted through physical information-constrained time-series decomposition, and finally, multi-layer fusion is used to generate a structural health state evolution cloud map. This step enables the removal of environmental interference, visualization of hidden defects, and dynamic tracking of degradation trajectories, effectively reducing the ambiguity of resistivity monitoring and providing accurate, intuitive, and reliable quantitative evidence for subsequent defect diagnosis and safety assessment.

[0137] S3: The structural health status evolution cloud map is integrated with multivariate physicochemical parameters and inspection information to determine the disease parameters.

[0138] Specific methods include:

[0139] S301: Extract the deterioration area boundary layer, disease event marker layer, and environmental disturbance background layer from the structural health state evolution cloud map to obtain the spatial distribution characteristics, temporal evolution characteristics, and environmental background characteristics of the disease;

[0140] Specifically, from the time-driven structural health evolution cloud map, three layers—the deteriorated area boundary layer, the disease event marker layer, and the environmental disturbance background layer—are extracted layer by layer. The deteriorated area boundary layer provides the contour coordinates, expansion range, and expansion trajectory of the diseased area, forming the spatial distribution characteristics of the disease. The disease event marker layer provides the time, magnitude, and type of disease mutations, forming the temporal evolution characteristics of the disease. The environmental disturbance background layer provides the amplitude and phase distribution of periodic fluctuations in environmental factors such as temperature and humidity, forming the environmental background characteristics of the disease. All three types of features strictly correspond to the grid coordinates of the lining monitoring section, providing basic visualization features for subsequent fusion.

[0141] S302: Extract environmental driving factors related to disease evolution from multivariate physicochemical parameters. The environmental driving factors include at least temperature and humidity parameters.

[0142] The multivariate physicochemical parameters were simultaneously acquired by a multivariate verification sensor array. Environmental driving factors that influence the occurrence and development of defects were extracted from these parameters. These environmental driving factors included at least temperature and humidity parameters. Temperature parameters were acquired by digital temperature sensors deployed inside the lining; humidity parameters were acquired by humidity sensors deployed at construction joints and high-risk leakage areas. Temperature parameters reflect the temperature change trend inside the concrete, while humidity parameters reflect the moisture content change inside the lining. These two types of parameters together constitute the external driving conditions for defect evolution, providing quantitative evidence for defect type determination.

[0143] S303: Extract manually identified disease description information from inspection information. The disease description information shall include at least the location of cracks, the area of ​​water seepage, and the extent of spalling.

[0144] The inspection information is derived from manual on-site inspection records, high-definition image recognition results, and detection data from automated inspection equipment. Defect descriptions, manually identified and verified, are extracted from the inspection information. These descriptions include at least the location of cracks, areas of seepage, and extent of spalling. Crack locations are marked with tunnel mileage and circumferential coordinates; seepage areas include the location of damp patches, the area of ​​seepage, and the degree of seepage; and spalling extent includes the location, area, and depth of concrete spalling. This defect description information is based on measured verification results, effectively validating the authenticity of electrical resistivity tomography (OTT) monitoring results and improving diagnostic reliability.

[0145] S304: Input spatial distribution characteristics, temporal evolution characteristics, environmental background characteristics, environmental driving factors and disease description information into a preset multi-source heterogeneous data fusion model. The fusion model has a built-in spatial alignment module, feature encoding module and decision fusion module.

[0146] The spatial distribution characteristics, temporal evolution characteristics, and environmental background characteristics corresponding to the structural health status evolution cloud map, the environmental driving factors corresponding to the multivariate physicochemical parameters, and the disease description information corresponding to the inspection information are simultaneously input into a pre-set multi-source heterogeneous data fusion model. This multi-source heterogeneous data fusion model is specifically designed for tunnel lining disease diagnosis and is compatible with different data formats such as images, numerical values, and text. Internally, the model sequentially includes a spatial alignment module, a feature encoding module, and a decision fusion module. These three modules are connected hierarchically and work collaboratively to complete data fusion and disease diagnosis. It should be noted that the multi-source heterogeneous data fusion model is pre-built during the system initialization phase. This model is designed for three types of heterogeneous data: structural health status evolution cloud map, multivariate physicochemical parameters, and inspection information. It is constructed based on the principles of spatiotemporal uniformity, feature compatibility, and interpretable decisions. The overall structure consists of a three-level serial structure of spatial alignment module, feature encoding module, and decision fusion module. Each module is designed independently, sequentially connected, and works collaboratively. The specific construction process is as follows:

[0147] (1) The spatial alignment module is built based on the unified spatial coordinate system of tunnel lining. The unified spatial coordinate system includes mileage coordinates, circumferential coordinates and longitudinal coordinates. The spatial alignment module has built-in coordinate transformation operators, timestamp synchronization operators and interpolation matching operators, which can realize the spatiotemporal benchmark unification of heterogeneous data.

[0148] (2) The feature encoding module is designed for the construction of multi-dimensional and high-dimensional feature data. The feature encoding module has built-in normalization unit, noise reduction unit, dimensionality reduction unit and feature fusion unit, which can transform heterogeneous features into a unified feature expression with fixed dimensions.

[0149] (3) The decision fusion module is built based on engineering cases and indoor test data. The decision fusion module has a built-in disease diagnosis knowledge base, similarity matching unit and grade judgment unit. The disease diagnosis knowledge base has pre-entered the feature templates and judgment rules of typical lining diseases, which can complete the intelligent diagnosis of diseases.

[0150] S305: Through the spatial alignment module, data from different sources are uniformly mapped to the spatial coordinate system of the lining structure to generate a spatially aligned fused feature set.

[0151] Among them, the spatial alignment module transforms the grid coordinates of the structural health status evolution cloud map, the measuring point coordinates of the multi-dimensional verification sensor array, and the defect label coordinates of the inspection information into a unified spatial coordinate system of the lining structure. This completes the position matching and timestamp synchronization of data from different sources, eliminates coordinate deviations and time misalignments, and generates a spatiotemporally aligned fusion feature set to ensure the accuracy of subsequent feature extraction and decision-making.

[0152] S306: Through the feature encoding module, the fusion feature set is reduced in dimensionality and features are extracted to obtain a comprehensive feature vector that represents the disease type, disease severity and disease evolution stage.

[0153] The feature encoding module employs a combination of principal component analysis and deep feature extraction to denoise, reduce dimensionality, and extract key features from the spatially aligned fused feature set, eliminating redundant information and retaining core features strongly correlated with the disease. The module normalizes multi-dimensional indicators such as resistivity change rate, resistivity change acceleration, abrupt change amplitude, temperature, humidity, crack length, and seepage area, fusing them to generate a fixed-dimensional comprehensive feature vector. This comprehensive feature vector fully characterizes the disease type, severity, and evolution stage, providing standardized input for the decision fusion module.

[0154] S307: Through the decision fusion module, the comprehensive feature vector is matched with the preset disease diagnosis knowledge base to output the determined disease parameters; the disease parameters include at least the disease type, disease location, disease size and disease severity level.

[0155] The decision fusion module performs similarity matching between the comprehensive feature vector and feature templates in the disease diagnosis knowledge base, determines the disease result based on the maximum matching probability, and outputs complete disease parameters. These parameters include at least the disease type, location, size, and severity level. The disease location is represented using a unified spatial coordinate system, the size is quantified as length, area, and volume, and the severity level is divided into four levels: mild, moderate, severe, and critical. This provides accurate input for subsequent mechanical model updates and safety assessments.

[0156] Understandably, this step takes structural health status evolution cloud map, multivariate physicochemical parameters, and inspection information as inputs, and completes intelligent disease diagnosis through a multi-source heterogeneous data fusion model. This step first extracts spatial distribution, temporal evolution, environmental background, environmental driving factors, and manually verified disease features from three types of data sources. Then, a spatial alignment module unifies the spatiotemporal benchmark, a feature encoding module generates a comprehensive feature vector, and finally, a decision fusion module matches it with the disease diagnosis knowledge base, outputting precise disease parameters including disease type, location, size, and severity level, providing a reliable basis for subsequent mechanical model updates and safety assessments.

[0157] S4: Based on the preset parametric finite element reference model, the fault parameters are mapped to the mechanical model input parameters. The parametric finite element reference model is updated based on the mechanical model input parameters, and the structural safety factor is calculated based on the updated parametric finite element reference model to obtain a quantitative assessment result of the structural service safety.

[0158] The specific methods for mapping disease parameters to mechanical model input parameters based on the preset parametric finite element reference model, and updating the parametric finite element reference model based on the mechanical model input parameters, include:

[0159] S401: Based on the disease type and severity level in the disease parameters, query the corresponding mechanical parameter reduction coefficient from the preset mechanical parameter mapping table; the mechanical parameter reduction coefficient includes at least the elastic modulus reduction coefficient, the permeability coefficient increment, and the surrounding rock spring stiffness reduction coefficient.

[0160] The disease parameters include disease type, location, size, and severity level. The pre-defined mechanical parameter mapping table is a quantitative correspondence table obtained through indoor testing and engineering calibration. The elastic modulus reduction factor, permeability coefficient increment, and surrounding rock spring stiffness reduction factor all correspond one-to-one with the disease type and severity level, and have been established in advance through indoor testing and engineering calibration. Specifically, for crack-type diseases, the elastic modulus reduction factor is approximately 0.95, the permeability coefficient increment is approximately 2.0, and the surrounding rock spring stiffness reduction factor is approximately 0.98 for mild cases, and approximately 0.70, 10.0, and 0.80 for severe cases, respectively. For seepage-type diseases, the elastic modulus reduction factor is approximately 0.98, the permeability coefficient increment is approximately 3.0, and the surrounding rock spring stiffness reduction factor is approximately 0.95 for mild cases, and approximately 0.75 and 12.0 for severe cases, respectively. For spalling-type defects, the elastic modulus reduction factor is approximately 0.92, the permeability coefficient increment is approximately 1.5, and the surrounding rock spring stiffness reduction factor is approximately 0.97 for mild defects, and approximately 0.65, 4.0, and 0.78 for severe defects, respectively. For void-type defects, the elastic modulus reduction factor is approximately 0.85, the permeability coefficient increment is approximately 1.0, and the surrounding rock spring stiffness reduction factor is approximately 0.70 for mild defects, and approximately 0.50, 1.0, and 0.40 for severe defects, respectively. The moderate-level parameters for various defects fall between mild and severe, and follow the same deterioration trend: the elastic modulus reduction factor and the surrounding rock spring stiffness reduction factor decrease with increasing severity, indicating a continuous weakening of structural stiffness and surrounding rock constraint; the permeability coefficient increment increases with increasing severity (remaining constant for void-type defects), indicating a continuous deterioration of the lining's seepage prevention performance.

[0161] It is worth noting that the elastic modulus reduction factor characterizes the degree of mechanical stiffness reduction in lining concrete after damage, and refers to the ratio of the elastic modulus of the lining concrete in the damaged state to that in the healthy state. The smaller the value of this factor, the more severe the deterioration of the concrete, such as cracking, spalling, loosening, and freeze-thaw damage, and the weaker the structural bearing and deformation capacity. The permeability coefficient increment is a parameter characterizing the degree of deterioration of the seepage prevention performance of the lining concrete, and refers to the increase in the permeability coefficient of the concrete in the damaged area compared to the healthy state. The more severe the damage (crack development, expansion of leakage channels, loosening of the structure), the larger the permeability coefficient increment, reflecting a more significant decrease in the lining's seepage resistance and a more intense groundwater intrusion. The surrounding rock spring stiffness reduction factor characterizes the degree of deterioration of the contact state between the lining and the surrounding rock, and refers to the ratio of the surrounding rock support spring stiffness after damage to that in the healthy state. The smaller the factor, the more obvious the voids behind the lining, the loosening of the surrounding rock, and the failure of the bond, the weaker the constraint and support of the surrounding rock on the lining, and the more unfavorable the structural stress state.

[0162] S402: Based on the location and size of the disease in the disease parameters, locate the set of elements affected by the disease in the parametric finite element reference model, and determine the damage degree weight distribution of each affected element in the disease area.

[0163] Among them, the parametric finite element benchmark model adopts the structural health state evolution cloud. Figure 1 A unified spatial coordinate system for the lining is established. Based on the mileage coordinates, circumferential coordinates, and longitudinal coordinates of the defect location, as well as the length, area, and volume information corresponding to the defect size, all mesh elements affected by the defect are accurately matched and extracted in the parametric finite element reference model, forming a set of elements affected by the defect. Using the geometric center of the defect area as the origin, the Euclidean distance h from the centroid of the element to the center of the defect is calculated for each element within the set of elements affected by the defect. A Gaussian weighted function is used to calculate the element damage degree weight ξ, with the specific calculation formula as follows:

[0164] ;

[0165] In the formula, ξ is the damage degree weight of the unit, ranging from 0 to 1; h is the Euclidean distance from the centroid of the unit to the center of the disease, in meters (m); q is the standard deviation of the Gaussian distribution of the disease's influence range, in meters (m). The standard deviation q is determined by the equivalent radius R of the disease, satisfying q = R / 3. This relationship is based on the statistical principle that approximately 99.7% of the probability mass in a Gaussian distribution falls within the mean ± 3 times the standard deviation. The damage degree weight ξ is inversely proportional to the distance of the unit from the disease center. The closer to the disease center, the larger the damage degree weight ξ, and the more significant the reduction in the unit's mechanical parameters. The farther from the disease center, the smaller the damage degree weight ξ, and the smoother the reduction in mechanical parameters, thus achieving a continuous and smooth spatial distribution of damage degree from the inside out in the disease area.

[0166] like Figure 2 As shown, the left figure is the initial healthy state model, i.e., the parameterized finite element reference model in the healthy state, where all mesh elements retain their original mechanical parameters; the right figure is the safe region and damaged region model, i.e., the parameterized finite element reference model updated according to the disease parameters, where the damaged region corresponds to the set of elements affected by the disease, and each element is weighted according to its actual calculated degree of damage. Values ​​are used for level fill. The higher the value, the closer the color level is to the indicator color of the disease center. The smaller the value, the closer the color gradation is to the color of the safe zone; the safe zone corresponds to a set of units unaffected by disease, maintaining their original healthy mechanical parameters. This color gradation distribution is related to the aforementioned Gaussian weighting function. The numerical distribution is consistent, that is, at the center of the disease. , The highest color level; following Increase Decays exponentially, with corresponding color gradations; when... hour The color gradation transitions to the safe zone base color.

[0167] It is worth noting that, during the initial healthy operation phase of the tunnel, an initial benchmark database of multi-source data, including resistivity and temperature, was collected and established, along with a parametric finite element benchmark model. When the tunnel lining structure was in its initial healthy state during operation, initial resistivity, temperature, humidity, and strain data for the entire lining section were collected to form a multi-source data initial benchmark database, providing a unified benchmark for time-series comparisons and change calculations. The parametric finite element benchmark model was simultaneously constructed during the system initialization phase, using Abaqus or COMSOL Multiphysics software to establish a two-dimensional plane strain model or a three-dimensional solid model that perfectly matches the actual monitoring section. The parametric finite element reference model adopts structured mesh generation, and the mesh elements maintain coordinate correspondence with the spatial elements of the resistivity monitoring section. The material constitutive relations of the lining concrete, initial support, and surrounding rock are defined in sequence, and the original mechanical parameters such as elastic modulus, Poisson's ratio, compressive strength, and permeability coefficient under healthy conditions are input. The boundary conditions use surrounding rock spring constraints to simulate the stratum resistance, and design earth pressure, water pressure, and operational loads are applied to form a parametric finite element reference model that can reflect the healthy service state of the lining structure. The material properties, boundary conditions, and load conditions of the model are all set as editable variables, providing a standardized interface for subsequent dynamic updates based on disease parameters. Among them, the elastic modulus is a mechanical parameter characterizing the ability of concrete to resist deformation; Poisson's ratio is a dimensionless parameter reflecting the ratio of lateral to longitudinal deformation of the material; compressive strength is the maximum compressive stress that concrete can withstand; permeability coefficient is used to reflect the seepage prevention performance of concrete; earth pressure is the pressure exerted by the surrounding rock on the lining structure; water pressure is the hydrostatic pressure exerted by groundwater on the lining; operational load includes the loads applied to the structure during operation, such as train passage; material properties are a set of parameters describing the material properties of the structure itself; boundary conditions are used to define the boundary constraints and contact states of the model; and load conditions are the sum of various external actions applied to the finite element model.

[0168] S403: Based on the mechanical parameter reduction coefficient and damage degree weight distribution, calculate the updated mechanical parameters of each affected element, which are used as input parameters for the mechanical model.

[0169] Specifically, the elastic modulus reduction factor, permeability coefficient increment, and surrounding rock spring stiffness reduction factor corresponding to each element are multiplied by the damage degree weight of the corresponding element to obtain the updated mechanical parameters of each element. The updated mechanical parameters include the updated elastic modulus, updated permeability coefficient, and updated surrounding rock spring stiffness. These parameters are the core mechanical model input parameters of the parameterized finite element reference model.

[0170] S404: Through the preset model update interface, the updated mechanical parameters are automatically assigned to the corresponding elements in the parametric finite element reference model to generate the updated parametric finite element reference model.

[0171] The preset model update interface is an automated read / write interface developed based on a scripting language. The model update interface can directly read the updated mechanical parameters and batch assign them to the corresponding mesh elements of the parametric finite element reference model, automatically overwrite the original healthy state mechanical parameters, complete the dynamic correction of material properties and boundary conditions, and generate an updated parametric finite element reference model that can truly reflect the current deterioration state and disease distribution of the lining.

[0172] In this embodiment of the application, after generating the updated parametric finite element baseline model, the specific method further includes:

[0173] S411: Determine whether the updated parametric finite element reference model meets the preset damage characterization accuracy requirements;

[0174] S412: If yes, then the updated parametric finite element datum model will be used as the refined parametric finite element datum model.

[0175] S413: If not, then based on the resistivity change rate field, resistivity change acceleration field, boundary expansion trajectory and structural health state evolution cloud map, a refined correction is performed to generate a refined parameterized finite element reference model.

[0176] Among them, the resistivity change rate field represents the spatial distribution field of the relative change in resistivity per unit time, reflecting the speed of deterioration; the resistivity change acceleration field represents the spatial distribution field of the rate of change of resistivity over time, reflecting the intensification or deceleration of the deterioration trend; the boundary expansion trajectory represents the path of positional change of the boundary layer of the deteriorated area over time, characterizing the direction and extent of disease development; the continuous damage field model represents a model based on continuum mechanics, using continuous and smooth field variables to describe the spatial distribution of the degree of damage inside the lining; the data assimilation algorithm represents an algorithm that optimally fuses the calculated values ​​of the model with the measured values, reducing model errors; the spatial heterogeneous mechanical parameter field represents the non-uniform distribution field of the mechanical parameters of each finite element unit that continuously change with spatial position; and the damage mechanics constitutive relation is a mathematical expression describing the quantitative relationship between material damage variables and stress, strain, and elastic modulus decay.

[0177] Specifically, the preset damage characterization accuracy requirements are based on pre-calibrated quantitative judgment standards, including: a spatial correlation coefficient between the distribution of mechanical parameters of finite element elements and the structural health state evolution cloud map greater than or equal to 0.90; a defect boundary position deviation less than or equal to 5 cm; and a relative error between the calculated mechanical response value and the measured value of multivariate physicochemical parameters less than or equal to 5%. The calculation results of the updated parametric finite element baseline model are compared item by item with the structural health state evolution cloud map, multivariate physicochemical parameters, and inspection defect description information to determine whether all the above accuracy indicators are simultaneously met. If the updated parametric finite element baseline model fully meets the preset damage characterization accuracy requirements, it indicates that the model can accurately reflect the current deterioration state and defect distribution characteristics of the lining structure, requiring no further correction. The updated parametric finite element baseline model is then directly determined as the refined parametric finite element baseline model for subsequent calculations of structural safety factors and failure probabilities. If the updated parametric finite element baseline model does not meet the preset damage characterization accuracy requirements, it indicates a deviation between the model's damage distribution, mechanical parameter reduction, and actual structural deterioration. Fine-tuning is required by combining resistivity temporal variation characteristics and the disease evolution trajectory. Correction data sources include the resistivity change rate field, resistivity change acceleration field, the deterioration region boundary expansion trajectory, and structural health state evolution cloud map. Through multi-dimensional information constraints, a precise match between the mechanical model and the actual disease state is achieved.

[0178] It is worth noting that the refined corrections include:

[0179] S414: Extract the boundary geometric information of the diseased area from the structural health state evolution cloud map, and combine it with the spatial distribution of the resistivity change rate field and resistivity change acceleration field to construct a continuous damage field model that characterizes the continuous spatial distribution of damage.

[0180] Based on the structural health state evolution cloud map, the geometric information of the disease boundary in the deterioration region boundary layer is extracted, including the disease contour coordinates, disease center location, disease equivalent radius, and disease major and minor axes. This boundary geometric information is spatially superimposed with the resistivity change rate field and resistivity change acceleration field. Using the absolute value of resistivity change, resistivity change rate, and change acceleration as weighting factors for the damage field, a continuous damage field model is constructed. The continuous damage field model uses the damage variable D... m Characterizes the degree of degradation of a unit, with a value ranging from 0 to 1, D m =0 represents no damage, D m =1 represents complete damage. The model expression is as follows:

[0181] ;

[0182] In the formula: D m (x,y,z) represents the damage variable at spatial coordinates (x,y,z); ξc The normalized numerical value of the structural health state evolution contour map; ξ r ξ represents the normalized value of the resistivity change rate field; a This represents the normalized value of the resistivity change acceleration field. For nonlinear mapping functions, which are dimensionless, at least a weighted average model is used, and the expression is: ,in, These are the weighting coefficients. ,and This model achieves a continuous and smooth transition of damage intensity from the center of the lesion outwards, replacing the traditional method of uniform reduction by partitioning, and more closely reflects the actual deterioration distribution.

[0183] S415: Based on the strain and humidity parameters in the multivariate physicochemical parameters, a preset data assimilation algorithm is used to reverse the continuous damage field model to obtain the assimilated damage field distribution.

[0184] The Kalman filter data assimilation algorithm is used as the preset data assimilation algorithm. The strain parameter and humidity parameter in the multivariate physicochemical parameters are used as observed values ​​to perform reverse correction and constraint on the continuous damage field model. A cost function between the model prediction value and the measured value is constructed. By minimizing the cost function, the damage variable is continuously updated iteratively to eliminate the initial bias of the model and the uncertainty of resistivity inversion. The assimilated damage field distribution is obtained with high consistency with the measured mechanical response and water content distribution, so that the damage field simultaneously satisfies the resistivity evolution characteristics and the measured structural state.

[0185] S416: Based on the assimilated damage field distribution and the preset damage mechanical constitutive relation, calculate the spatial heterogeneous mechanical parameter field of each finite element element.

[0186] The preset damage mechanical constitutive relation adopts the uniaxial stress damage constitutive relation of concrete, and the expression is as follows:

[0187] ;

[0188] In the formula: This represents the elastic modulus of the element after damage, i.e., the mechanical parameter of a spatially heterogeneous structure, in Pascals (Pa). The modulus of elasticity in the healthy state is expressed in Pascals (Pa); Dn represents the assimilated damage variable. Similarly, damage correlation calculations are performed on the permeability coefficient and the stiffness of the surrounding rock spring.

[0189] ;

[0190] ;

[0191] In the formula, k d Indicates the permeability coefficient after damage, in meters per second (m / s). k0 is the permeability coefficient in the healthy state, in meters per second (m / s). Kd represents the spring stiffness of the surrounding rock after damage, in units of: K0 represents the spring stiffness of the surrounding rock under healthy conditions, in units of: ;α k This represents the permeability coefficient growth factor, determined experimentally. Based on the above relationships, the elastic modulus, permeability coefficient, and surrounding rock spring stiffness are calculated element by element, forming a spatially heterogeneous mechanical parameter field, thus realizing the continuous variation of mechanical parameters with the spatial distribution of damage.

[0192] S417: Through the preset model update interface, spatial heterogeneous mechanical parameter fields are assigned to the corresponding elements to generate a refined parametric finite element reference model.

[0193] The system utilizes a pre-defined model update interface developed using Python scripts to read the spatially heterogeneous mechanical parameter field. It then precisely assigns the mechanical parameters of each element to the corresponding mesh elements of the updated parametric finite element reference model according to the unified spatial coordinate system of the lining. This automatically overwrites the original uniformly reduced parameters, completing the model refinement and ultimately generating a refined parametric finite element reference model. This model accurately reflects the spatial continuity and heterogeneity of lining structure deterioration, significantly improving the accuracy of subsequent mechanical calculations, safety factor assessments, and failure probability analyses.

[0194] Understandably, after updating the parametric finite element baseline model, the above steps further refine and improve the model's accuracy. First, it's determined whether the updated model meets the preset damage characterization accuracy requirements. If it does, it's directly used as the refined model; otherwise, corrections are made by combining the resistivity change rate field, resistivity change acceleration field, boundary expansion trajectory, and structural health state evolution cloud map. A continuous damage field model is constructed by extracting the geometric information of the damage boundary. Then, using strain and humidity parameters as observations, a Kalman filter data assimilation algorithm is used to reverse-correct the damage field. Based on the damage mechanics constitutive relation, the spatial heterogeneous mechanical parameter field is calculated. Finally, parameter assignment is completed through the model interface, generating a refined parametric finite element baseline model that accurately reflects the continuous distribution and spatial heterogeneity of lining damage, providing a more accurate computational foundation for subsequent structural safety assessments.

[0195] Furthermore, in this embodiment of the application, the specific method for calculating the structural safety factor based on the updated parametric finite element benchmark model to obtain a quantitative assessment result of the structural service safety includes:

[0196] S421: Apply a preset design load condition to the refined parametric finite element reference model, perform finite element solution, and obtain the displacement field, stress field, and internal force distribution.

[0197] The process involves applying preset design load cases to the refined parametric finite element reference model. These load cases include the structure's self-weight, surrounding rock pressure, groundwater pressure, train operation load, and additional temperature load. The loads are applied according to the basic combinations specified in current tunnel design codes. Static analysis is performed using a finite element solver. After convergence, the output shows the global displacement field of the structure, the stress field of concrete and steel reinforcement, and the axial force, bending moment, and shear force distribution at each section of the lining, thus fully characterizing the mechanical response features of the lining structure under the current distressed state.

[0198] S422: Extract key mechanical indicators from the solution results. Key mechanical indicators include at least the crown settlement, sidewall horizontal convergence, maximum compressive stress in concrete, maximum tensile stress in steel reinforcement, and bending moment and axial force of the control section.

[0199] Specifically, key mechanical indicators controlling structural safety are extracted from the finite element method (FEM) results. These include: crown settlement (vertical displacement of the lining crown node); sidewall horizontal convergence (relative horizontal displacement of corresponding measuring points on both sidewalls); maximum compressive stress in concrete (absolute value of the minimum principal stress in the concrete element); and maximum tensile stress in reinforcing steel (maximum value of the tensile stress in the reinforcing steel element). Simultaneously, bending moments and axial forces at control sections such as the crown, arch waist, sidewalls, and invert are extracted as the basis for internal force safety evaluation.

[0200] It is worth noting that: structural self-weight is the vertical load generated by the weight of the tunnel lining material itself; surrounding rock pressure is the passive pressure exerted on the outer wall of the lining by the deformation and relaxation of the surrounding rock; groundwater pressure is the hydrostatic pressure and seepage pressure exerted by groundwater on the lining; train operation load is the dynamic and static action transmitted to the lining structure when a train passes; temperature-induced additional load is the additional internal force generated by the thermal expansion and contraction of the structure due to changes in ambient temperature; basic combination is the combination method of various loads according to the most unfavorable working conditions in accordance with current tunnel design specifications; static analysis is a finite element calculation method that does not consider time and velocity effects and only solves the equilibrium state of the structure; displacement field is the spatial distribution of displacement vectors of each node of the structure under load; stress field is the spatial distribution of stress of concrete and steel reinforcement elements under load; axial force is the axial internal force acting along the normal direction of the lining section; bending moment is the internal force that causes bending deformation of the lining section; shear force is the internal force that causes shear deformation of the lining section.

[0201] S423: Compare the key mechanical indicators with the preset allowable thresholds, calculate the displacement safety factor, strength safety factor and internal force safety factor, and take the minimum value as the overall structural safety factor.

[0202] Specifically, the extracted key mechanical indicators are compared with preset allowable thresholds. The allowable threshold for arch crown settlement is typically 10mm~50mm, preferably 30mm; the allowable threshold for sidewall horizontal convergence is typically 10mm~40mm, preferably 20mm; the allowable threshold for concrete compressive stress is 12.5MPa (design axial compressive strength of C30 concrete); the allowable threshold for steel reinforcement tensile stress is 270MPa (design tensile strength of HRB400 steel reinforcement); and the allowable thresholds for bending moment and axial force at the lining control section are determined according to the bearing capacity calculation formula in the standard.

[0203] It is worth noting that the displacement safety factor, strength safety factor, and internal force safety factor should be calculated separately according to the standard formulas; the displacement safety factor is the ratio of the allowable displacement value to the calculated displacement value, and the calculation formula is as follows:

[0204] ;

[0205] In the formula, Indicates the displacement safety factor; This indicates the allowable displacement threshold, in millimeters (mm). ); This represents the absolute value of displacement calculated using the finite element method, in millimeters (mm). ); The preset minimum calculated displacement threshold is in millimeters. This is used to avoid an abnormally large safety factor when the calculated displacement approaches zero. The overall structural safety factor is taken as the minimum value of the displacement safety factor, strength safety factor, and internal force safety factor, representing the true safety reserve of the structure with a conservative principle.

[0206] The strength safety factor is the ratio of the material's design strength to the calculated stress, and the calculation formula is:

[0207] ;

[0208] In the formula, Indicates the strength safety factor; Indicates the permissible intensity threshold, in megapascals (MPa). This represents the stress value calculated by the finite element method, in megapascals (MPa).

[0209] The internal force safety factor is the ratio of the allowable bearing capacity of the section to the calculated internal force, and the calculation formula is:

[0210] ;

[0211] ;

[0212] In the formula, Indicates the axial force safety factor; This indicates the allowable axial force capacity of the cross section, in kilonewtons (kN). ); This represents the absolute value of axial force calculated using the finite element method, in units of: ; Indicates the bending moment safety factor; This represents the allowable bending moment capacity of the cross section, in units of: ; This represents the absolute value of the bending moment calculated using the finite element method, in units of: The internal force safety factor is taken as the smaller value between the axial force safety factor and the bending moment safety factor.

[0213] The overall structural safety factor is the minimum value of the displacement safety factor, strength safety factor, and internal force safety factor, which represents the true safety reserve of the structure based on the principle of conservatism.

[0214] S424: Compare and verify the calculated strain obtained by solving with the measured strain parameter in the multivariate physicochemical parameters. If the deviation exceeds the preset threshold, then perform the fine correction step again based on the current deviation information, update the fine parameterized finite element reference model and recalculate the safety factor until the deviation between the calculated strain and the measured strain converges within the preset accuracy range.

[0215] The process involves extracting the calculated strain from the finite element model at the same location as the measured strain sensor and comparing it point by point with the measured strain in the multivariate physicochemical parameters. A preset deviation threshold of 5% is set. If the relative deviation exceeds this threshold, the deviation is used as a correction constraint, and the process returns to steps S414–S417 to re-execute the continuous damage field construction, data assimilation correction, and spatial heterogeneous mechanical parameter field update. This generates a corrected, refined parameterized finite element benchmark model, which is then re-solved until the deviation between the calculated strain and the measured strain converges to within 5%, ensuring the model's accuracy and reliability.

[0216] S425: The overall structural safety factor that finally converges is used as the quantitative assessment result of the structural service safety.

[0217] Specifically, when the absolute value of the relative error between the calculated strain and the measured strain is no greater than 5%, the preset convergence accuracy is met, and the overall structural safety factor obtained from the current iteration is used as the final quantitative assessment result. This result objectively reflects the actual safety reserve of the lining structure under the current distribution of defects, material deterioration, and stress state, providing the sole quantitative basis for subsequent graded early warning.

[0218] S5: Based on the preset safety threshold, the quantitative assessment results are given graded warnings. When the warning is triggered, a visual report containing the location of the disease, the safety cloud map and maintenance suggestions is generated and output based on the preset maintenance strategy, quantitative assessment results and disease parameters.

[0219] Specifically, the system compares the final converged overall structural safety factor with a preset safety threshold. Based on the overall structural safety factor, the safety threshold is divided into three levels: a safety factor ≥ 2.0 indicates a safe state, 1.5 ≤ safety factor < 2.0 indicates a state of attention, and a safety factor < 1.5 indicates a state of alert; these correspond to three levels of warnings: blue (normal), yellow (attention), and red (alert). A safe state does not trigger a warning, only generating a periodic health report; a state of attention triggers a warning; and a state of alert triggers an emergency warning. When a warning is triggered, the system, based on a preset maintenance strategy and considering the type, location, size, severity level, and structural safety factor of the defect, automatically generates a visual report including spatial location markings of the defect, a structural safety cloud map, a comparison of mechanical indicators, damage distribution, and targeted maintenance recommendations. This report is output in the form of charts, vector cloud maps, and text descriptions, providing an intuitive basis for operational decisions. The above three-level warning system, based on the overall structural safety factor, is a different assessment dimension from the four-level risk classification based on the comprehensive index of resistivity dynamic characteristics. The former is used to determine the structural bearing capacity safety, while the latter is used to identify the spatial distribution of deterioration trends.

[0220] In summary, this application addresses the technical problems of existing tunnel lining monitoring, such as the strong subjectivity of manual inspections, insufficient coverage of point-based monitoring, significant ambiguity in resistivity interpretation, and the inability to convert monitoring indicators into structural safety factors. It proposes a closed-loop intelligent diagnostic method through steps S1 to S5: During the initial healthy phase of the structure, a multi-source benchmark database and a parameterized finite element benchmark model are established. Resistivity and multivariate physicochemical parameters are collected synchronously over a long period. Temperature compensation and time-series inversion are used to generate a structural health state evolution cloud map. This cloud map, sensor data, and inspection information are then integrated to achieve accurate diagnosis of defects. Defect parameters are mapped to mechanical parameters and the finite element model is dynamically updated. After accuracy assessment and refinement, the overall structural safety factor is calculated. Finally, early warnings are issued according to safety threshold levels, and a visual report is output. This method effectively eliminates the ambiguity of resistivity, achieving a complete transformation from resistivity field changes to defect identification and then to quantitative assessment of structural safety. It significantly improves the accuracy, automation level, and early warning capability of tunnel lining safety assessment, providing reliable support for operation and maintenance decisions.

[0221] This application provides an intelligent diagnostic system for the service safety of tunnel lining. The system is based on resistivity imaging and multi-source information fusion to realize the whole process of intelligent diagnosis of tunnel lining structure from medium characteristic perception to structural safety quantitative assessment. Specifically, it includes a data acquisition module, a data generation module, a data fusion module, a data processing module, and a data output module. The modules work together and the data is transmitted step by step to form a closed-loop intelligent diagnostic system.

[0222] It is worth noting that the data acquisition module is used to acquire resistivity monitoring data, multivariate physicochemical parameters, and inspection information of the lining structure. The resistivity monitoring data is acquired synchronously by an electrode array embedded in the lining and a multi-channel acquisition instrument. The multivariate physicochemical parameters include at least temperature parameters, humidity parameters, strain parameters, and corrosion potential parameters. The temperature parameters are used for subsequent resistivity temperature compensation calculations. The inspection information includes on-site manual records, high-definition image recognition results, and detection data from automated inspection equipment.

[0223] The data generation module is connected to the data acquisition module. It is used to correct resistivity monitoring data using temperature parameters based on a preset benchmark database and a preset temperature compensation model. It generates a time series dataset that quantitatively represents the relative change in resistivity, and performs time-lapse inversion imaging on the time series dataset to generate a structural health status evolution cloud map that reflects the humidity distribution, crack development and material deterioration process inside the lining, providing basic data for disease visualization analysis.

[0224] The data fusion module is connected to the data generation module and is used to perform spatiotemporal alignment and fusion analysis of the structural health status evolution cloud map, multivariate physicochemical parameters and inspection information. It extracts the spatial distribution characteristics, temporal evolution characteristics, environmental background characteristics, environmental driving factors and manually verified disease description information of the disease. Through a multi-source heterogeneous data fusion model, it completes intelligent matching and judgment and outputs disease parameters including disease type, disease location, disease size and disease severity level.

[0225] The data processing module is connected to the data fusion module. It is used to map the disease parameters to the mechanical model identifiable input parameters according to the preset parametric finite element reference model. Based on the mechanical parameter reduction coefficient and damage weight distribution, it updates the material properties and boundary conditions of the parametric finite element reference model. It performs accuracy judgment and fine correction on the updated model. Based on the finally converged fine parametric finite element reference model, it applies the design load condition to perform finite element solution, extracts key mechanical indicators and calculates the overall structural safety factor, and obtains the quantitative assessment results of the structural service safety.

[0226] The data output module is connected to the data processing module and is used to provide graded early warnings for quantitative assessment results based on preset safety thresholds. When an early warning is triggered, it automatically generates and outputs a visual report containing spatial location markings of defects, structural safety cloud maps, mechanical index comparisons, and targeted maintenance suggestions based on preset maintenance strategies, structural safety coefficients, and defect parameters. This provides intuitive, quantitative, and actionable decision-making basis for tunnel operation and maintenance management.

[0227] This system integrates resistivity monitoring, multi-sensor verification, multi-source information fusion, finite element model updating, and structural safety assessment through inter-module collaboration. It effectively solves the technical problems of traditional monitoring, such as strong subjectivity, limited coverage, multiple interpretations of resistivity, and inability to quantitatively assess bearing capacity. It enables early detection of deterioration of tunnel lining structures, accurate identification of defects, and quantitative early warning of safety status, thereby improving the intelligence and safety of tunnel operation and maintenance.

[0228] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent diagnosis of tunnel lining service safety, characterized in that, include: Acquire resistivity monitoring data, multivariate physicochemical parameters and inspection information of the lining structure. The multivariate physicochemical parameters include at least the temperature parameter used for temperature compensation, and also the humidity parameter. Based on the preset benchmark database and preset temperature compensation model, the resistivity monitoring data is corrected using temperature parameters to generate a time series dataset that quantitatively characterizes the relative change in resistivity. The time series dataset is then inverted and imaged to generate a structural health state evolution cloud map that reflects the evolution of the lining medium characteristics. By integrating and analyzing the structural health status evolution cloud map with the aforementioned multivariate physicochemical parameters and inspection information, the disease parameters are determined. Based on the preset parametric finite element reference model, the disease parameters are mapped to the mechanical model input parameters. The parametric finite element reference model is updated based on the mechanical model input parameters, and the structural safety factor is calculated based on the updated parametric finite element reference model to obtain a quantitative assessment result of the structural service safety. The quantitative assessment results are classified and warned according to the preset safety threshold. When the warning is triggered, a visual report containing the location of the disease, safety cloud map and maintenance suggestions is generated and output based on the preset maintenance strategy, quantitative assessment results and disease parameters.

2. The intelligent diagnostic method for tunnel lining service safety according to claim 1, characterized in that, Based on a pre-set benchmark database and a pre-set temperature compensation model, resistivity monitoring data is corrected using temperature parameters to generate a time-series dataset that quantitatively characterizes the relative changes in resistivity, including: Temperature parameters are obtained from multivariate physicochemical parameters; A preset temperature compensation model is used to correct the resistivity monitoring data based on the acquired temperature parameters, eliminating the interference of temperature changes on the resistivity measurement value and obtaining a corrected resistivity value uniform to the preset reference temperature. From the preset benchmark database, locate the benchmark resistivity value under the health condition corresponding to the current monitoring location; The corrected resistivity value is compared with the reference resistivity value, and the relative rate of change between the two is calculated. Based on the relative rate of change, a time-series dataset is generated that quantifies the relative change in resistivity, using a benchmark database as a reference.

3. The intelligent diagnostic method for tunnel lining service safety according to claim 2, characterized in that, Inversion imaging of the time-series dataset generates structural health state evolution cloud maps reflecting the evolution of lining medium properties, including: The corrected resistivity values ​​and their corresponding spatial coordinates at each monitoring time in the time series dataset are preprocessed, outlier data are removed and spatial interpolation is performed to form the apparent resistivity distribution data on the regular grid corresponding to each monitoring time. The apparent resistivity distribution data at each monitoring time is input into the preset resistivity tomography inversion model, and the true resistivity distribution inside the lining after inversion at the corresponding time is obtained through iterative solution. The true resistivity distribution obtained from the inversion at each monitoring time is compared with the health state benchmark resistivity distribution at the corresponding location in the benchmark database. The relative rate of change of resistivity is calculated to generate the resistivity change field at each monitoring time. The resistivity change fields generated at different monitoring times are combined according to the time series, and the dynamic characteristics of resistivity change are extracted by time series analysis. Based on the dynamic characteristics, a structural health state evolution cloud map that can intuitively show the spatiotemporal evolution law of the lining medium characteristics is generated.

4. The intelligent diagnostic method for tunnel lining service safety according to claim 3, characterized in that, Extracting dynamic characteristics of resistivity changes includes: Arrange the resistivity change fields at each monitoring time in chronological order to construct a three-dimensional spatiotemporal data volume. The three-dimensional spatiotemporal data volume contains multiple spatial units, and each spatial unit corresponds to a time series curve reflecting the evolution of resistivity at that location over time. The three-dimensional spatiotemporal data volume is input into the preset physical information constrained time series decomposition model. The physical information constrained time series decomposition model has built-in physical evolution prior knowledge. Based on the physical evolution prior knowledge, the time series curve of each spatial unit is decomposed into trend component, seasonal component and anomaly component. Based on seasonal components, the amplitude and phase of periodic environmental disturbances are identified, an environmental baseline correction field is generated, and the environmental baseline correction field is used to remove environmental disturbances from the trend components and anomalous components to obtain the environmentally corrected trend components and anomalous components. Based on the trend components after environmental correction, the resistivity change rate field and change acceleration field of each spatial unit are calculated; based on the spatial distribution of the change acceleration field, the degradation acceleration region is identified, and the displacement of the spatial boundary of the degradation acceleration region between adjacent monitoring times is tracked to generate the boundary expansion trajectory. Based on the abnormal components after environmental correction, the resistivity change time and change magnitude of each spatial unit are identified. Spatial units that meet the preset change conditions are marked as disease event points, and the data are aggregated to generate a spatiotemporal distribution dataset of disease events. Based on the spatial coupling relationship between the trend component and the anomaly component after environmental correction, a comprehensive index of the frequency of anomalies and the degree of trend deterioration in each spatial region is calculated, and regions whose comprehensive index exceeds the preset risk threshold are identified as deterioration-sensitive regions. The resistivity change rate field, resistivity change acceleration field, boundary expansion trajectory, spatiotemporal distribution dataset of disease events, and deterioration-sensitive areas are collectively used as dynamic features of resistivity change.

5. The intelligent diagnostic method for tunnel lining service safety according to claim 4, characterized in that, The structural health state evolution cloud map, generated based on dynamic features, can intuitively display the spatiotemporal evolution law of the lining medium properties, including: The resistivity change rate field and the resistivity change acceleration field are vector-fused to construct a vector field characterizing the dynamics of degradation evolution; wherein, the vector direction is determined by the gradient direction of the change acceleration field, and the vector magnitude is characterized by the weighted fusion result of the resistivity change rate value and the change acceleration value. The environmental baseline correction field is periodically mapped to generate an environmental interference background layer. The boundary extension trajectory is mapped onto the resistivity change field at each monitoring time to generate a time-labeled deterioration region boundary layer. Based on the spatiotemporal distribution dataset of disease events, a disease event labeling layer with time labels is generated in the resistivity change field at the corresponding time. Spatial correlation analysis was performed between the deterioration-sensitive areas and the disease event marker layer to generate a risk level distribution layer. The vector field, environmental disturbance background layer, deteriorated area boundary layer, disease event marker layer, and risk level distribution layer are spatiotemporally aligned and multi-layered to generate a time-driven structural health state evolution cloud map. As the structural health state evolution cloud map evolves in the time dimension, it simultaneously presents the expansion trajectory of the deteriorated area, the spatiotemporal distribution of periodic environmental disturbances, the occurrence time and location of disease events, and the spatial distribution of risk levels.

6. The intelligent diagnostic method for tunnel lining service safety according to claim 5, characterized in that, By integrating and analyzing the structural health status evolution cloud map with multivariate physicochemical parameters and inspection information, the following disease parameters were determined: The deterioration zone boundary layer, disease event marker layer, and environmental disturbance background layer are extracted from the structural health state evolution cloud map to obtain the spatial distribution characteristics, temporal evolution characteristics, and environmental background characteristics of the disease. Environmental driving factors related to disease evolution are extracted from multiple physicochemical parameters, including at least temperature and humidity parameters. Extract manually identified disease description information from inspection information. The disease description information should include at least the location of cracks, the area of ​​water seepage, and the extent of peeling. Spatial distribution characteristics, temporal evolution characteristics, environmental background characteristics, environmental driving factors and disease description information are input into a preset multi-source heterogeneous data fusion model. The fusion model has a built-in spatial alignment module, feature encoding module and decision fusion module. The spatial alignment module maps data from different sources to the spatial coordinate system of the lining structure, generating a spatially aligned fusion feature set. The feature encoding module performs dimensionality reduction and feature extraction on the fused feature set to obtain a comprehensive feature vector representing the disease type, disease severity, and disease evolution stage. The decision fusion module matches the comprehensive feature vector with the preset disease diagnosis knowledge base and outputs the determined disease parameters. The disease parameters include at least the disease type, disease location, disease size, and disease severity level.

7. The intelligent diagnostic method for tunnel lining service safety according to claim 5, characterized in that, Based on the pre-defined parametric finite element reference model, the disease parameters are mapped to the mechanical model input parameters. Updating the parametric finite element reference model based on the mechanical model input parameters includes: Based on the disease type and severity level in the disease parameters, the corresponding mechanical parameter reduction coefficient is retrieved from the preset mechanical parameter mapping table; the mechanical parameter reduction coefficient includes at least the elastic modulus reduction coefficient, the permeability coefficient increment, and the surrounding rock spring stiffness reduction coefficient. Based on the location and size of the disease in the disease parameters, locate the set of elements affected by the disease in the parametric finite element reference model, and determine the weight distribution of the damage degree of each affected element in the disease area. Based on the mechanical parameter reduction coefficient and damage degree weight distribution, the updated mechanical parameters of each affected unit are calculated and used as input parameters for the mechanical model. The updated mechanical parameters are automatically assigned to the corresponding elements in the parametric finite element reference model through the preset model update interface, thereby generating the updated parametric finite element reference model.

8. The intelligent diagnostic method for tunnel lining service safety according to claim 7, characterized in that, After generating the updated parametric finite element baseline model, the following steps are also included: Determine whether the updated parametric finite element reference model meets the preset damage characterization accuracy requirements; If so, the updated parametric finite element reference model will be used as the refined parametric finite element reference model. If not, then based on the resistivity change rate field, resistivity change acceleration field, boundary extension trajectory, and structural health state evolution cloud map, a refined correction is performed to generate a refined parametric finite element baseline model; the refined correction includes: The boundary geometric information of the diseased area is extracted from the structural health state evolution cloud map, and combined with the spatial distribution of the resistivity change rate field and resistivity change acceleration field, a continuous damage field model representing the continuous spatial distribution of damage is constructed. Based on the strain and humidity parameters in the multivariate physicochemical parameters, a preset data assimilation algorithm is used to reverse the continuous damage field model to obtain the assimilated damage field distribution. Based on the assimilated damage field distribution and the preset damage mechanical constitutive relation, the spatial heterogeneous mechanical parameter field of each finite element is calculated. By using a pre-defined model update interface, spatial heterogeneous mechanical parameter fields are assigned to the corresponding elements to generate a refined parametric finite element reference model.

9. The intelligent diagnostic method for tunnel lining service safety according to claim 8, characterized in that, The structural safety factor is calculated based on the updated parametric finite element baseline model, and the quantitative assessment results of the structural service safety are obtained, including: Apply a preset design load condition to the refined parametric finite element reference model, perform finite element solution, and obtain the displacement field, stress field and internal force distribution; Extract key mechanical indicators from the solution results. These key mechanical indicators include at least the crown settlement, sidewall horizontal convergence, maximum compressive stress in concrete, maximum tensile stress in steel reinforcement, and bending moment and axial force at the control section. By comparing the key mechanical indicators with the preset allowable thresholds, the displacement safety factor, strength safety factor and internal force safety factor are calculated, and the minimum value is taken as the overall structural safety factor. The calculated strain obtained by the solution is compared and verified with the measured strain parameter in the multivariate physicochemical parameters. If the deviation exceeds the preset threshold, the fine correction step is carried out again based on the current deviation information, the fine parameterized finite element reference model is updated and the safety factor is recalculated until the deviation between the calculated strain and the measured strain converges within the preset accuracy range. The final convergent overall structural safety factor is used as the quantitative assessment result of the structural service safety.

10. A smart diagnostic system for the service safety of tunnel lining, characterized in that, The method described in any one of claims 1 to 9 includes: The data acquisition module is used to acquire resistivity monitoring data, multivariate physicochemical parameters and inspection information of the lining structure. The multivariate physicochemical parameters include at least the temperature parameters used for temperature compensation. The data generation module is used to correct resistivity monitoring data using temperature parameters based on a preset benchmark database and a preset temperature compensation model, generate a time series dataset that quantitatively characterizes the relative change in resistivity, and perform inversion imaging on the time series dataset to generate a structural health state evolution cloud map that reflects the evolution of lining medium characteristics. The data fusion module is used to integrate and analyze the structural health status evolution cloud map with multivariate physicochemical parameters and inspection information to determine the disease parameters; The data processing module is used to map the fault parameters to mechanical model input parameters according to the preset parametric finite element reference model, update the parametric finite element reference model based on the mechanical model input parameters, and calculate the structural safety factor based on the updated parametric finite element reference model to obtain a quantitative assessment result of the structural service safety. The data output module is used to provide graded early warnings for quantitative assessment results based on preset safety thresholds. When an early warning is triggered, it generates and outputs a visual report containing the location of the disease, a safety cloud map, and maintenance suggestions based on preset maintenance strategies, quantitative assessment results, and disease parameters.

Images