A structure performance judgment and target re-measurement method and system based on GPR

Abstract

The application provides a structure performance determination and target retest method and system based on GPR, and belongs to the technical field of structure mechanical performance evaluation and detection based on computer data processing; initial GPR data is preprocessed, and data compatibility states are screened according to a calibration error domain, are mapped into structure model parameters, and structure performance categories are obtained; when the compatibility states cross performance boundaries, candidate retest actions are enumerated, new GPR responses of each state are predicted, response units and possible reserved state subsets are formed; actions of each subset belonging to a single performance category are determined as sufficient retest actions, the minimum one is selected and executed when the actions exist, otherwise, target retest actions or limited stop results are output according to the worst case performance range. The retest data is used for updating states and performance after quality inspection, and the output result is obtained. The application reserves multiple compatible states in the error domain, and propagates them to the structure performance space one by one, so that states crossing the performance decision boundary are avoided to be ignored due to interpretation truncation.

Description

Technical Field

[0001] This invention belongs to the field of structural mechanical performance evaluation and testing technology based on computer data processing, and particularly relates to a structural performance determination and target retesting method and system based on GPR. Background Technology

[0002] Ground penetrating radar (GPR) can provide the electromagnetic response of the internal medium, interfaces, and local anomalies of an engineering structure. Taking tunnel lining as an example, local separation, voids, contact anomalies, reinforcement obstruction, water content changes, and coupling differences may produce similar initial waveforms or image features. Selecting only the single interpretation state with the smallest mismatch and directly inputting it into the structural model may miss states that are still compatible with measured data but correspond to different structural performance categories.

[0003] Existing retesting methods typically employ fixed-interval encryption, expanded scanning range, or sorting by general uncertainty indicators. While these methods increase the amount of data, they may not necessarily generate new, distinguishable electromagnetic responses for differences in states affecting safety, criticality, or risk categories, and they lack verification of whether all possible results after retesting support a single performance category. Therefore, there is an urgent need for a method that organizes compatible state screening, structural performance propagation, candidate configuration response prediction, state subset formation, and actual retesting updates into a continuous technical chain. Summary of the Invention

[0004] The problem this invention aims to solve is: when the initial GPR data allows for multiple interpretations of anomalous states, how to map still indistinguishable responses to a subset of potentially retained states through implementable new GPR observations, determine whether the retest action is sufficient to eliminate state ambiguity across performance categories, and output a result that can be determined, retesting can continue, or the test can be stopped with restrictions.

[0005] The first aspect of this invention proposes a structural performance determination and target retesting method based on GPR, comprising the following steps: S101, preprocess the initial GPR data and extract GPR features; S102, input the candidate state into the electromagnetic model to generate prediction features, calculate the mismatch between the prediction features and the GPR features according to the calibration error domain, and screen compatible states; S103 transforms the compatibility state into structural model parameters and calculates the structural performance in combination with mechanical conditions; S104, aggregate structural performance according to performance boundaries; output a determinable result when the same type is used, and form evidence of performance to be retested when the types are cross-class. S105, based on the performance evidence to be retested and the on-site constraints, enumerate and screen candidate retesting actions; S106, the new GPR response of each compatible state in forward modeling under candidate actions, forming response units and possible retained state subsets according to the action error domain; S107, add each candidate action that belongs to a single performance category in the sufficient action set; sort by execution cost when the sufficient action set is not empty, and sort by worst-case performance range when the sufficient action set is empty, and output the target retest action or restricted stop result; S108, Perform the target retest action, check the quality of the retest data, and update the compatibility status and structural performance by combining the qualified data with the existing GPR data; S109 aggregates and updates the structural performance according to the performance boundary, outputs a determineable result and returns to S105 to continue retesting, or outputs a limited stop result.

[0006] Preferably, the S101 process includes: sequentially performing time-zero correction, DC drift removal, bandpass filtering, and position registration on the initial GPR data, and extracting at least one GPR feature from interface travel time, reflection amplitude, frequency band attenuation, and interchannel continuity.

[0007] Preferably, the S102 process includes: establishing and discretizing the candidate state space based on design data, initial anomalous region, structural geometric boundary and physical feasible interval; inputting the candidate states one by one into the electromagnetic forward model to generate predictive features; calculating the mismatch between the predictive features and GPR features using the least squares method with calibration error domain weighting; and forming a compatible state set of candidate states whose mismatch does not exceed the error threshold. The compatible state represents an anomalous state that cannot be excluded by the current GPR data under the calibrated error level.

[0008] Preferably, the S103 process includes: mapping the abnormal location and abnormal spatial range in the candidate state to the support weakening zone, mapping the effective gap to the contact closure gap, mapping the filling category to the equivalent contact stiffness, support stiffness interval or water pressure boundary, and inputting them together with the load, material parameters and boundary conditions into the structural response model; and solving the structural performance of each compatible state one by one. The structural response model employs at least one of the following: a beam-spring model, a ring beam model, or a shell finite element model. The abnormal range is set as the support weakening zone, the effective gap is set as the contact closure gap, and the local cracking safety margin is calculated based on the difference between the material tensile strength design value and the calculated maximum tensile stress.

[0009] Preferably, the S105 process includes: enumerating at least one observation configuration among the following for the performance evidence to be retested: new survey line location, survey line direction, local survey line densification, scanning range, sampling density, antenna frequency band, and number of superpositions; eliminating unexecutable configurations based on equipment capability, on-site accessibility, and safety constraints; retrieving the action error domain for the retained observation configurations; and calculating the comprehensive execution cost based on scanning length, operation time, cost, accessibility, and data quality risk to obtain a candidate retest action set.

[0010] Preferably, the process of S106 includes: performing electromagnetic forward modeling on each compatible state under the observation configuration corresponding to each candidate retest action to obtain the predicted new GPR response; constructing a response confusion graph with the predicted new GPR response as a node and the normalized response distance not greater than the action error threshold as the connecting edge; obtaining the connected components of the response confusion graph as response units, and forming a subset of possible retained states with the compatible states corresponding to each response unit.

[0011] Preferably, the process of S107 includes: classifying the structural performance and performance boundary of each possible retained state subset one by one; adding the candidate retest actions corresponding to only one performance category for each possible retained state subset to the sufficient action set; when the sufficient action set is not empty, outputting the target retest actions in ascending order of comprehensive execution cost; when the sufficient action set is empty, calculating the worst-case performance range of each candidate retest action and outputting the target retest actions in ascending order that do not exceed the preset allowable upper limit; when all exceed the preset allowable upper limit, outputting the GPR observation capability limitation stop result.

[0012] Preferably, the process of S108 includes: checking the survey line position, antenna attitude and coupling, scanning speed, signal-to-noise ratio, repeatability and position coding continuity of the target retest data; when the quality is qualified, joint weighted least squares matching is used to re-screen the compatibility state and calculate the structural performance.

[0013] Preferably, the process of S109 includes: comparing and classifying the updated structural performance with the performance boundary; outputting a decisionable result when they belong to the same performance category; returning to S105 when they still cross categories and there are executable candidate actions; and outputting a corresponding stop result when the data is invalid, the model is not applicable, the operation is limited, or the GPR observation capability is limited.

[0014] The second aspect of this invention proposes a structural performance assessment and target retesting system based on GPR, comprising: The data preprocessing module is used to preprocess the initial GPR data and extract GPR features; The compatible state screening module is used to input candidate states into the electromagnetic model to generate prediction features, calculate the mismatch between the prediction features and GPR features according to the calibration error domain, and screen compatible states. The performance propagation module is used to map compatible states to structural model parameters and calculate structural performance results in combination with mechanical conditions. The performance aggregation module is used to aggregate structural performance according to performance boundaries; when the same type is used, it outputs a judgment result, and when the types are cross-type, it forms evidence of performance to be retested. The candidate action generation module is used to enumerate and filter candidate retest actions based on the performance evidence to be retested and the on-site constraints; The response subset formation module is used to add new GPR responses for each compatible state under candidate actions in forward modeling, and to form response units and possible retained state subsets according to the action error domain. The target action output module is used to include candidate actions belonging to a single performance category in each possible subset of retained states into the sufficient action set; when the sufficient action set is not empty, it is sorted by execution cost; when the sufficient action set is empty, it is sorted by worst-case performance range, and the target retest action or restricted stop result is output. The retest update module is used to perform target retest actions, check the quality of retest data, and jointly update the compatibility status and structural performance with qualified data and existing GPR data. The loop output module is used to aggregate the updated structural performance according to the performance boundary, output the verifiable result and return to S105 to continue the retest, or output the limited stop result.

[0015] Compared with the prior art, the present invention has the following innovative features and beneficial effects: Compared with directly inputting a single optimal GPR interpretation into the structural model, this invention preserves multiple data-compatible states within the error domain and propagates them one by one to the structural performance space, avoiding the neglect of states across performance decision boundaries due to interpretation truncation. Compared with retesting methods that use fixed-rule encryption or sorting by general information indicators, this invention verifies the sufficiency of retesting based on the newly added GPR response, action error domain, and possible retained state subset under candidate configurations, prioritizing the elimination of states that cross performance decision boundaries rather than simply increasing the amount of data. Compared with the combined retesting method of directly executing multi-band, orthogonal measurement lines and extended range, the present invention first determines whether the state subset corresponding to each possible retesting result belongs to a single performance category; when there is sufficient action, the one with the lowest execution cost is selected; when there is no sufficient action, the target action is selected according to the worst-case performance range, or when it is greater than the allowable upper limit, the GPR observation capability is explicitly limited. Attached Figure Description

[0016] Figure 1 This is a sequential overall flowchart of the present invention.

[0017] Figure 2 This is a diagram showing the propagation relationship between initial data, compatibility states, and structural performance.

[0018] Figure 3 A diagram illustrating the generation of new GPR responses for candidate retest actions and the narrowing of the scope of performance evidence.

[0019] Figure 4 Supplementary processing map for full remeasurement and limited GPR observation capabilities.

[0020] Figure 5 This is a diagram showing the quality check and status update of the actual retest data.

[0021] Figure 6 This is a schematic diagram of an embodiment of a partial separation scenario of tunnel lining.

[0022] Figure 7 This is a timing diagram of the modules in the system of the present invention. Detailed Implementation

[0023] The overall process of this invention is as follows Figure 1 As shown: S101, preprocess the initial GPR data and extract GPR features; S102, input the candidate state into the electromagnetic model to generate prediction features, calculate the mismatch between the prediction features and the GPR features according to the calibration error domain, and screen compatible states; S103 transforms the compatibility state into structural model parameters and calculates the structural performance in combination with mechanical conditions; S104, aggregate structural performance according to performance boundaries; output a determinable result when the same type is used, and form evidence of performance to be retested when the types are cross-class. S105, based on the performance evidence to be retested and the on-site constraints, enumerate and screen candidate retesting actions; S106, the new GPR response of each compatible state in forward modeling under candidate actions, forming response units and possible retained state subsets according to the action error domain; S107, add each candidate action that belongs to a single performance category in the sufficient action set; sort by execution cost when the sufficient action set is not empty, and sort by worst-case performance range when the sufficient action set is empty, and output the target retest action or restricted stop result; S108, Perform the target retest action, check the quality of the retest data, and update the compatibility status and structural performance by combining the qualified data with the existing GPR data; S109 aggregates and updates the structural performance according to the performance boundary, outputs a determineable result and returns to S105 to continue retesting, or outputs a limited stop result.

[0024] The following embodiments illustrate the present invention according to the sequence of steps S101 to S109. Each term is explained in the step in which it first participates in data processing. The thresholds and weights in the formulas are set before the start of the corresponding evaluation round. The specific values ​​in the embodiments are used to illustrate the data processing process and do not constitute a fixed limitation on other projects.

[0025] Example 1: S101: Initial GPR Data Preprocessing and Feature Extraction This step involves inputting the initial GPR data and initial observation configuration. Initial GPR data refers to the A-scan, B-scan, or image data generated from them acquired before the target re-survey is performed; the initial observation configuration includes the survey line coordinates, survey line direction, antenna frequency band, channel spacing, sampling interval, number of stacks, scan speed, and coupling mode.

[0026] During processing, the time zero of each channel is first corrected using metal plate reflection or direct peak wave, and then DC drift and low-frequency background are removed using moving average or polynomial fitting. Subsequently, bandpass filtering is performed according to the effective frequency band of the antenna, and channel resampling and position registration are completed based on mileage coding and structural coordinates. The interface travel time, reflection amplitude, band attenuation, and inter-channel continuity are extracted window by window from the corrected data, and observation feature vectors are constructed according to their characteristic order. , as the input of S102.

[0027] S102: Candidate State Forward Modeling and Compatible State Selection This step inputs the GPR feature vector, candidate state space, electromagnetic response model, and calibration error domain. The candidate state is a set of parameterized descriptions of possible anomalies within the structure; taking tunnel lining as an example, the state vector... From abnormal longitudinal position Anomaly Circular Center Longitudinal length Circumferential width Effective gap and fill category composition: ; In the formula, The candidate state vector for GPR; The abnormal longitudinal center location; This is an abnormal circumferential central angle; This is an abnormal longitudinal length; This refers to the abnormal circumferential width. The effective gap for the abnormal region; The value represents the type of filling medium, and its value corresponds one-to-one with the corresponding set of dielectric constant, conductivity, and magnetic permeability parameters.

[0028] Based on the design data and the initial anomaly region, the physical feasible ranges of each parameter are given, and a finite set of candidate states is obtained by discretizing them according to a preset grid interval. Each candidate state and the initial observation configuration are then input into a finite-difference time-domain model, a ray model, or a surrogate model verified by leaving out samples to obtain the corresponding predicted features.

[0029] Taking the FDTD model as an example, a three-dimensional computational domain is first established based on the structural geometry, lining material, and candidate states. The position, range, effective gap, and filling category of the candidate states are written into the geometric range, dielectric constant, conductivity, and permeability of the corresponding grid cells, respectively. The grid size is determined by the minimum wavelength corresponding to the highest simulation frequency, ensuring that the cell side length is no greater than one-tenth of the minimum wavelength, and the time step is determined according to the Courant stability condition. A pulse source with the same center frequency and polarization direction as the initial observation antenna is used, and a perfectly matched layer absorption boundary is set at the outer boundary of the computational domain. The time-domain waveforms of each receiving point are recorded according to the actual survey line coordinates. The same time-zero correction, filtering, position registration, and feature extraction as S101 are performed on the simulated waveforms to form predicted features. When using a surrogate model, the state vector and observation configuration of the aforementioned FDTD samples are used as inputs, and the predicted features are used as outputs to divide the training set and the reserved check set. The reserved check set is only used when the residual does not exceed the corresponding action error domain, and the residual of the surrogate model is included in the calibration standard deviation.

[0030] The calibration error domain is the set of permissible deviations corresponding to the equipment model, antenna frequency band, structural type, coupling method, and preprocessing version. The standard deviation of each feature is pre-calibrated using at least one of the following: repeated data acquisition, reserved test specimens or historical samples, electromagnetic forward modeling residuals, surrogate model residuals, and positioning errors; then, the mismatch is calculated using the weighted least squares method according to the preset feature weights. ; In the formula, Candidate state Weighted mismatch under initial observations; This represents the initial number of observed features; This refers to the initial observation feature number; For the first Preset weights for initial observed features; For the first Measured features; For state The corresponding number Predictive features; For the first The standard deviation of each feature is used for calibration. Candidate states with mismatches not exceeding the initial error threshold are retained to form a set of compatible states. ; In the formula, For the first The set of compatible states for round evaluation; For evaluation round sequence number; Candidate state; The discrete candidate state space of GPR; This represents the upper limit of the initial observation mismatch. The compatible states here refer to anomalous states that cannot be excluded by the current GPR data at the calibrated error level; this step retains all compatible states and does not replace the set with the single state with the smallest mismatch.

[0031] S103: Propagation of compatibility state to structural performance This step involves inputting the set of compatible states, loads, material parameters, support conditions, and structural boundaries. For each compatible state, parameter mapping is performed: abnormal locations and spatial extents are written into the support weakening region, effective gaps are written into the closed gaps of contact elements, and the filling category is converted into equivalent contact stiffness, support stiffness intervals, or hydraulic pressure boundaries. Then, the finite element structural response model of the ring beam, beam-spring, or shell is input to obtain the performance indices. : ; In the formula, For state Corresponding structural performance indicators; Solving operators for structural response; A mapping operator that converts GPR candidate states into structural model parameters; It is a vector of mechanical conditions consisting of load, material, support, and boundary conditions; This refers to the structural response model and its version used in this round. Structural performance can be expressed using cracking safety margin, load-bearing safety margin, deformation limit margin, internal forces, or contact pressure; when using cracking safety margin, the calculation formula is: ; In the formula, For localized cracking safety margin; This is the design value for the tensile strength of the material; For state The maximum tensile stress is calculated under the same conditions. All compatible states are solved one by one according to the same model version and mechanical conditions to obtain the state-performance correspondence table. Figure 2 As shown.

[0032] S104: Performance Classification and Evidence Formation for Performance to be Retested This step involves inputting the performance indicators and preset performance boundaries for each compatibility state. Taking cracking safety margin as an example, upper and lower boundaries are set: margins greater than the upper boundary are classified as safe, margins between the two boundaries are classified as critical, and margins less than the lower boundary are classified as risky. The performance boundaries are fixed before the start of the evaluation round and are not adjusted in reverse according to the classification results.

[0033] After classifying the compatible states one by one, if the category set contains only one category, a performance determination record is directly formed; if the category set contains at least two categories, the compatible states that cause cross-category issues, the corresponding performance values, and the categories are extracted to form performance evidence to be retested. The performance evidence to be retested is not new detection data, but rather the correspondence between the states retained in S102 and the performance obtained in S103.

[0034] When verifying the stability of cross-category evidence, observation-compatible branches are formed based on state adjacency relationships. State adjacency relationship refers to two discrete states having only one anomalous parameter in adjacent ranges, with the remaining parameters being the same. Then, at least two of the following checks are performed to verify whether the branches persist, in order to eliminate spurious branches caused by a single threshold or single model perturbation: error threshold disturbance, survey line position deviation, antenna coupling change, and electromagnetic model verification.

[0035] S105: Generation and Execution Cost Calculation of Candidate Retest Actions This step involves inputting the performance evidence to be retested, GPR equipment capabilities, site accessibility, and safety constraints. Candidate retest actions are a set of parameters that can change the configuration of newly added GPR physical observations, and may include the location and direction of new survey lines, orthogonal survey lines, local densification, scanning range, sampling density, antenna frequency band, and number of superpositions. Configurations are first enumerated around cross-category states, and then configurations that exceed the equipment frequency band, are inaccessible, or do not meet operational safety requirements are eliminated.

[0036] For each executable configuration, the motion error domain is retrieved based on device, frequency band, survey line direction, sampling density, and coupling method. The motion error domain includes the calibration standard deviation and motion error threshold for each newly added response feature under that configuration. Scan length, operation time, cost, accessibility risk, and data quality risk are normalized separately, and the overall execution cost is calculated using preset weights. ; In the formula, For candidate retesting actions The overall execution cost; For candidate retest actions; The cost items are numbered sequentially, corresponding to scan length, operation time, cost, accessibility risk, and data quality risk, respectively. For the first Preset weights for item costs; For action The The normalized value of each item is calculated. The weights of each item are pre-set based on on-site schedule, safety level, and data quality requirements, and remain unchanged throughout the same round of candidate action comparison. The output includes the candidate action number, observation configuration, action error domain, and overall execution cost, for S106 to evaluate individually.

[0037] S106: The new response distinguishes between relations and the possible subset of retained states. This step inputs the set of compatible states, candidate retest actions, and corresponding action error domains. For each candidate action, keeping its survey line, frequency band, and sampling parameters unchanged, electromagnetic forward modeling is performed on each compatible state to obtain the predicted new response. For any two compatible states, the normalized response distance is calculated using the following formula. : ; In the formula, For action Next state With state The normalized new response distance; and This is the compatibility state number; For action The corresponding number of newly added response features; Add a new response feature sequence number; For the first Preset weights for newly added response features; For state In action The next The item predicts new response features; For this feature in action The calibration standard deviation is below; For action The confusing distance threshold is determined. For each newly predicted response as a node, a confusing edge is connected when the normalized response distance is not greater than the action error threshold, resulting in a confusing response graph. The connected components of this graph are then determined; each connected component is called a response unit. Connectivity rules can eliminate ambiguities caused by the non-transitive nature of pairwise distance relationships.

[0038] The response unit is inversely mapped to the compatible state that produces the response, forming a subset of possible retained states. : ; In the formula, For action The A subset of possible retained states; The response unit number; For state In action The predicted new response vector is shown below; For action The Each response unit. A possible retained state subset represents the set of states that might still be retained based on the observation capabilities of that action when a new response falls into a given response unit after the corresponding action is executed. S106 outputs all possible retained state subsets for each candidate action, along with the corresponding performance value and performance category for each state in S103. The process of generating new GPR responses, forming response units, and narrowing the scope of performance evidence in the candidate retest configuration is as follows: Figure 3 As shown.

[0039] S107: Thorough retesting of motion selection and determination of target motion This step inputs the possible retained state subsets, performance categories, and execution costs for each candidate action. If each possible retained state subset for the same action contains only one performance category, then regardless of which response unit the actual retest response falls into, cross-category states will not be retained, and the action will be added to the sufficient action set. : ; In the formula, To provide a full set of actions; This is the set of all executable candidate retest actions; An operator that maps a set of states to the set of performance categories they contain; The cardinality of the set; Indicates the action All response units satisfy the conditions. When the sufficient action set is not empty, sort by comprehensive execution cost from smallest to largest and select the one with the smallest cost. When the sufficient action set is empty, calculate the worst-case performance range for each action: first, find the difference between the maximum and minimum values ​​of the same performance index within each possible subset of retained states, then take the maximum value from all subsets. ; In the formula, For action Worst-case performance range; and These are the maximum and minimum values ​​of the performance index within the corresponding set, respectively; For state The corresponding structural performance indicators. The preset allowable upper limit is given before action evaluation begins, based on structural performance decision tolerance, risk level, and acceptable subsequent verification resources. When the sufficient action set is empty, actions not exceeding the allowable upper limit are selected in ascending order of worst-case performance range; when all candidate actions have been evaluated and the minimum value still exceeds the allowable upper limit, the output GPR observation capability is limited and stops, and unresolved states and evaluated actions are recorded. The target action calculation formula when sufficient actions exist is: ; In the formula, Retest the action for the target; This means selecting the action from the specified set of actions that minimizes the objective function. For action The overall execution cost.

[0040] The process of thorough action screening, worst-case performance range comparison, and handling limitations in GPR observation capabilities is as follows: Figure 4 As shown.

[0041] S108: Target Retesting Implementation, Quality Inspection, and Joint Updates Perform target re-measurement according to the observation configuration output by S107, and collect new GPR data. First, check the positional deviation between the actual survey line and the target survey line, antenna attitude and coupling, scanning speed, signal-to-noise ratio, repeated acquisition differences, and position coding continuity; if all indicators do not exceed the preset quality limit, proceed to joint update; otherwise, perform re-measurement or substitution actions, and stop when the preset number of re-measurements is reached and the output data becomes invalid.

[0042] When the quality is acceptable, calculate the initial observation mismatch and the re-measurement observation mismatch for the same candidate state, and form a joint mismatch according to their respective error domains. : ; In the formula, For state The combined mismatch; This represents the initial observation mismatch. Retesting the action for the target The mismatch was observed during the retest. This represents the upper limit of the joint mismatch. The state where the joint mismatch does not exceed the joint threshold is retained, and the structural performance is recalculated according to S103. Simultaneously, the actual newly added response is mapped to the response unit established in S106; if the normalized deviation between the actual response and all predicted responses exceeds the action error threshold, the state set is not forcibly updated, and the output model is deemed inapplicable or the observation configuration deviation is stopped. The actual retest data quality check, joint update, and result aggregation process is as follows: Figure 5 As shown.

[0043] S109: Update Result Aggregation and Loop Termination The structural performance of the updated compatible state is reclassified according to the same performance boundary used in S104. When the category set contains only one category, the output performance can be determined and recording stops; when the category set still spans multiple categories and there are still unevaluated or regenerable candidate actions, the updated compatible state, performance range, and field constraints are returned to S105 to proceed to the next round; when the data is invalid, the model is not applicable, field operations are limited, or GPR observation capabilities are limited, the corresponding recording stops.

[0044] Each output should at least save the evaluation object number, initial and retest observation configuration, error domain version, compatible state set, performance category, candidate actions and costs, actual quality inspection results, number of loops and reason for stopping, so that engineering reviewers can verify the complete sequence.

[0045] Example 2: This embodiment uses the partial separation of a reinforced concrete tunnel lining as a construction scenario, and provides the data results from S101 to S109 step by step, comparing them with existing fixed-rule re-measurement schemes. The following values ​​are used to illustrate algorithmic relationships and calculation order, and do not represent specific engineering inspection statistics or performance guarantees. The relationship between the partial separation of the tunnel lining, the initial GPR survey line, and the target re-measurement location in this embodiment is as follows: Figure 6 As shown.

[0046] Example data for S101 and S102: Initial GPR data were acquired circumferentially along the inner surface of the lining using a 400 MHz antenna, with a survey line length of 6.0 m, a track spacing of 0.02 m, and 8 stacking operations. After zero-correction, 5% sliding window background removal, 150–650 MHz bandpass filtering, and odometer registration, four features were extracted and normalized according to their respective calibration standard deviations to obtain the results. .

[0047] Candidate states are discretized within the design allowable range, and electromagnetic forward modeling is performed. Each feature weight is set to 1, and the initial error threshold is 0.20. The weighted mismatch of the four candidate states is calculated. , , , They are respectively: ; In the formula, , , and These represent states 1 through 4, respectively. to These are the initial observed mismatches for the four states. Therefore, states 1, 2, and 3 enter the compatible state set, while state 4 is excluded because its mismatch of 0.360 is greater than 0.20.

[0048] Example data for S103 and S104 The three compatible states were mapped to the weakened zone of the ring beam support and the contact closure gap, respectively. The solution was obtained under the same surrounding rock pressure, material parameters, and boundary conditions, and verified using a shell finite element model at a state close to the performance boundary. The boundary between the safe and critical classes was set to 0.08 MPa, and the boundary between the critical and risk classes was set to -0.02 MPa. The results are shown in Table 1.

[0049] Table 1 Initial mismatch, cracking safety margin, and performance category of compatibility condition

[0050] The compatibility status covers the safety, critical, and risk categories simultaneously. Initial observations cannot support a single performance category. Therefore, S104 forms the performance evidence to be retested, consisting of status 1, status 2, status 3, and their cracking safety margins.

[0051] Example data from S105 to S107 Four executable actions are enumerated around the anomaly center: Action 1 is to densify the original survey line longitudinally, Action 2 is to add a new orthogonal survey line, Action 3 is to switch frequency bands, and Action 4 is to switch frequency bands and densify locally. After weighting by scan length, operation time, cost, accessibility, and data quality risk, the relative comprehensive execution costs of the four are 1.00, 1.00, 1.20, and 1.65, respectively. Each action uses an independently calibrated action error domain, and this example uses a uniform confusion threshold of 0.25.

[0052] Forward modeling is performed on the three compatible states under each of the four actions, and the pairwise normalized response distances are calculated according to S106. Taking action 1 as an example, the distance between state 1 and state 2 is 0.18, which is no greater than 0.25, so a confusing edge is established; the remaining distances are greater than 0.25, and the connected components are {state 1, state 2} and {state 3}. The processing results for all actions are shown in Table 2.

[0053] Table 2. Response distance of candidate retest actions, possible subset of retained states, and S107 processing results.

[0054] Each possible subset of retained states for actions 3 and 4 contains only one performance category, and therefore is added to the sufficient action set; in ascending order of comprehensive execution cost, S107 outputs action 3, i.e., the band switching action. This process first verifies the sufficiency of the retest, and then compares the costs among the sufficient actions.

[0055] Example data for S108 and S109 According to action 3, the target retest data was collected. The measurement line position deviation was 8 mm, which is less than the 20 mm limit; the correlation coefficient of repeated acquisition was 0.94, which is greater than the 0.90 limit; the signal-to-noise ratio, scanning speed, and position coding continuity were all qualified. The actual newly added response fell into the response unit corresponding to state 2. The retest mismatches for states 1, 2, and 3 were calculated to be 0.34, 0.05, and 0.30, respectively. Substituting these values ​​into the joint mismatch formula, we get: ; In the formula, , and These represent states 1 through 3, respectively. to These are the joint mismatch values ​​for the three states. A joint threshold of 0.20 is set, and only state 2 is retained after the update; its cracking safety margin is 0.04 MPa, belonging to the critical category. Therefore, S109 outputs "Performance Determinable—Critical Category" and stops, without generating another round of retesting.

[0056] A constructive comparison with existing fixed-rule retesting schemes: Existing fixed-rule schemes do not predict the possible subset of retained states under candidate configurations. Instead, they first perform vertical encryption; if the states still cross categories, they then perform orthogonal testing. In this example, we assume that all actual responses correspond to state 2: the first round of vertical encryption retains {state 1, state 2}, still crossing security and critical classes; the second round of orthogonal testing retains {state 2, state 3}, and {state 2} is obtained only after intersecting with the first round set. This invention, however, compares four actions before execution, selects action 3 at once, and obtains {state 2}.

[0057] Table 3 Comparison of the target retesting scheme of this invention with existing fixed rule schemes

[0058] As shown in Table 3, the relative reduction in execution cost is... Furthermore, the number of retest rounds to achieve performance category convergence is reduced from 2 rounds to 1 round. This comparison illustrates that the effectiveness of this invention comes from the data processing chain of "predicting new GPR responses—forming a subset of possible retained states—verifying sufficiency—selecting target actions," rather than simply increasing the amount of GPR data.

[0059] like Figure 7 As shown, the system sequentially includes a data preprocessing module, a compatible state filtering module, a performance propagation module, a performance aggregation module, a candidate action generation module, a response subset formation module, a target action output module, a retest update module, and a loop output module. Each module performs the data processing corresponding to S101 to S109, and transmits the evaluation object number, observation configuration version, error domain version, state set, response unit, performance category, action number, and stop reason.

[0060] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0061] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A structural performance assessment and target retesting method based on GPR, characterized in that, Includes the following steps: S101, preprocess the initial GPR data and extract GPR features; S102, input the candidate state into the electromagnetic model to generate prediction features, calculate the mismatch between the prediction features and the GPR features according to the calibration error domain, and screen compatible states; The specific process includes: establishing and discretizing the candidate state space based on design data, initial abnormal regions, structural geometric boundaries and physically feasible intervals; inputting each candidate state into the electromagnetic forward model to generate predictive features; using the least squares method with calibration error domain weighting to calculate the mismatch between the predictive features and the GPR features; and forming a compatible state set by candidate states whose mismatch does not exceed the error threshold. The compatible state represents an abnormal state that cannot be excluded by the current GPR data under the calibrated error level. S103 transforms the compatibility state into structural model parameters and calculates the structural performance in combination with mechanical conditions; S104, aggregate structural performance according to performance boundaries; output a determinable result when the same type is used, and form evidence of performance to be retested when the types are cross-class. S105, based on the performance evidence to be retested and the on-site constraints, enumerate and screen candidate retesting actions; S106, perform forward modeling of the new GPR responses of each compatible state under candidate actions, and form response units and possible retained state subsets according to the action error domain; the specific process includes: performing electromagnetic forward modeling on each compatible state under the observation configuration corresponding to each candidate retest action to obtain the predicted new GPR responses; constructing a response confusion graph with the predicted new GPR responses as nodes and the normalized response distance not greater than the action error threshold as connecting edges; obtaining the connected components of the response confusion graph as response units, and forming a possible retained state subset with the compatible states corresponding to each response unit; S107, add candidate actions belonging to a single performance category to the sufficient action set for each possible subset of retained states; sort the sufficient action set by execution cost when it is not empty, and sort it by worst-case performance range when it is empty, and output the target retest action or restricted stop result; the specific process includes: classifying the structural performance and performance boundary of each possible subset of retained states one by one, adding the candidate retest actions corresponding to only one performance category for each possible subset of retained states to the sufficient action set; when the sufficient action set is not empty, output the target retest actions in ascending order of comprehensive execution cost; when the sufficient action set is empty, calculate the worst-case performance range of each candidate retest action and output the target retest actions not exceeding the preset allowable upper limit in ascending order, and output the restricted stop result of GPR observation capability when all exceed the preset allowable upper limit; S108, Perform the target retest action, check the quality of the retest data, and update the compatibility status and structural performance by combining the qualified data with the existing GPR data; S109 aggregates and updates the structural performance according to the performance boundary, outputs a determineable result and returns to S105 to continue retesting, or outputs a limited stop result.

2. The structural performance determination and target retesting method based on GPR as described in claim 1, characterized in that, The S101 process includes: sequentially performing time-zero correction, DC drift removal, bandpass filtering, and position registration on the initial GPR data, and extracting at least one GPR feature from interface travel time, reflection amplitude, frequency band attenuation, and interchannel continuity.

3. The structural performance determination and target retesting method based on GPR as described in claim 1, characterized in that: The S103 process includes: mapping the abnormal locations and abnormal spatial ranges in the candidate states to support weakening zones, mapping the effective gaps to contact closure gaps, and mapping the filling categories to equivalent contact stiffness, support stiffness intervals, or water pressure boundaries, and inputting them together with loads, material parameters, and boundary conditions into the structural response model; and solving the structural performance of each compatible state one by one. The structural response model employs at least one of the following: a beam-spring model, a ring beam model, or a shell finite element model. The abnormal range is set as the support weakening zone, the effective gap is set as the contact closure gap, and the local cracking safety margin is calculated based on the difference between the material tensile strength design value and the calculated maximum tensile stress.

4. The structural performance determination and target retesting method based on GPR as described in claim 1, characterized in that: The S105 process includes: enumerating at least one observation configuration among the following based on the performance evidence to be retested: new survey line location, survey line direction, local survey line densification, scanning range, sampling density, antenna frequency band, and number of superpositions; eliminating unexecutable configurations based on equipment capability, on-site accessibility, and safety constraints; retrieving the action error domain for the retained observation configurations; and calculating the comprehensive execution cost based on scanning length, operation time, cost, accessibility, and data quality risk to obtain a candidate retest action set.

5. The structural performance determination and target retesting method based on GPR as described in claim 1, characterized in that: The process of S108 includes: checking the survey line position, antenna attitude and coupling, scanning speed, signal-to-noise ratio, repeatability and position coding continuity of the target retest data; when the quality is qualified, joint weighted least squares matching is used to re-screen the compatibility state and calculate the structural performance.

6. The structural performance determination and target retesting method based on GPR as described in claim 1, characterized in that: The process of S109 includes: comparing and classifying the updated structural performance with the performance boundary; outputting a decisionable result when they belong to the same performance category; returning to S105 when they still cross categories and there are executable candidate actions; and outputting a corresponding stop result when the data is invalid, the model is not applicable, the operation is limited, or the GPR observation capability is limited.

7. A structural performance assessment and target retesting system based on GPR, characterized in that: The method applied to any one of claims 1 to 6 includes: The data preprocessing module is used to preprocess the initial GPR data and extract GPR features; The compatible state screening module is used to input candidate states into the electromagnetic model to generate prediction features, calculate the mismatch between the prediction features and GPR features according to the calibration error domain, and screen compatible states. The performance propagation module is used to map compatible states to structural model parameters and calculate structural performance results in combination with mechanical conditions. The performance aggregation module is used to aggregate structural performance according to performance boundaries; when the same type is used, it outputs a judgment result, and when the types are cross-type, it forms evidence of performance to be retested. The candidate action generation module is used to enumerate and filter candidate retest actions based on the performance evidence to be retested and the on-site constraints; The response subset formation module is used to add new GPR responses for each compatible state under candidate actions in forward modeling, and to form response units and possible retained state subsets according to the action error domain. The target action output module is used to include candidate actions belonging to a single performance category in each possible subset of retained states into the sufficient action set; when the sufficient action set is not empty, it is sorted by execution cost; when the sufficient action set is empty, it is sorted by worst-case performance range, and the target retest action or restricted stop result is output. The retest update module is used to perform target retest actions, check the quality of retest data, and jointly update the compatibility status and structural performance with qualified data and existing GPR data. The loop output module is used to aggregate the updated structural performance according to the performance boundary, output the verifiable result and return to S105 to continue the retest, or output the limited stop result.

Images